diff --git "a/ParakeetEncoder_4bit_par.mlmodelc/model.mil" "b/ParakeetEncoder_4bit_par.mlmodelc/model.mil"
new file mode 100644--- /dev/null
+++ "b/ParakeetEncoder_4bit_par.mlmodelc/model.mil"
@@ -0,0 +1,3402 @@
+program(1.3)
+[buildInfo = dict<string, string>({{"coremlc-component-MIL", "3405.2.1"}, {"coremlc-version", "3404.23.1"}})]
+{
+    func main<ios18>(tensor<fp32, [1, 128, 1001]> audio_signal, tensor<int32, [1]> length) {
+            int32 var_25 = const()[name = string("op_25"), val = int32(-1)];
+            tensor<int32, [3]> x_1_perm_0 = const()[name = string("x_1_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string audio_signal_to_fp16_dtype_0 = const()[name = string("audio_signal_to_fp16_dtype_0"), val = string("fp16")];
+            string cast_0_to_fp16_dtype_0 = const()[name = string("cast_0_to_fp16_dtype_0"), val = string("fp16")];
+            fp16 var_82_promoted_to_fp16 = const()[name = string("op_82_promoted_to_fp16"), val = fp16(-0x1p+0)];
+            tensor<fp16, [1]> length_to_fp16 = cast(dtype = cast_0_to_fp16_dtype_0, x = length)[name = string("cast_2")];
+            tensor<fp16, [1]> var_83_cast_fp16 = add(x = length_to_fp16, y = var_82_promoted_to_fp16)[name = string("op_83_cast_fp16")];
+            fp16 _inversed_85_y_0_to_fp16 = const()[name = string("_inversed_85_y_0_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1]> _inversed_85_cast_fp16 = mul(x = var_83_cast_fp16, y = _inversed_85_y_0_to_fp16)[name = string("_inversed_85_cast_fp16")];
+            fp16 var_86_to_fp16 = const()[name = string("op_86_to_fp16"), val = fp16(0x1p+0)];
+            tensor<fp16, [1]> lengths_1_cast_fp16 = add(x = _inversed_85_cast_fp16, y = var_86_to_fp16)[name = string("lengths_1_cast_fp16")];
+            tensor<fp16, [1]> lengths_3_cast_fp16 = floor(x = lengths_1_cast_fp16)[name = string("lengths_3_cast_fp16")];
+            fp16 var_90_promoted_to_fp16 = const()[name = string("op_90_promoted_to_fp16"), val = fp16(-0x1p+0)];
+            tensor<fp16, [1]> var_91_cast_fp16 = add(x = lengths_3_cast_fp16, y = var_90_promoted_to_fp16)[name = string("op_91_cast_fp16")];
+            fp16 _inversed_93_y_0_to_fp16 = const()[name = string("_inversed_93_y_0_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1]> _inversed_93_cast_fp16 = mul(x = var_91_cast_fp16, y = _inversed_93_y_0_to_fp16)[name = string("_inversed_93_cast_fp16")];
+            fp16 var_94_to_fp16 = const()[name = string("op_94_to_fp16"), val = fp16(0x1p+0)];
+            tensor<fp16, [1]> lengths_7_cast_fp16 = add(x = _inversed_93_cast_fp16, y = var_94_to_fp16)[name = string("lengths_7_cast_fp16")];
+            tensor<fp16, [1]> lengths_9_cast_fp16 = floor(x = lengths_7_cast_fp16)[name = string("lengths_9_cast_fp16")];
+            fp16 var_98_promoted_to_fp16 = const()[name = string("op_98_promoted_to_fp16"), val = fp16(-0x1p+0)];
+            tensor<fp16, [1]> var_99_cast_fp16 = add(x = lengths_9_cast_fp16, y = var_98_promoted_to_fp16)[name = string("op_99_cast_fp16")];
+            fp16 _inversed_101_y_0_to_fp16 = const()[name = string("_inversed_101_y_0_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1]> _inversed_101_cast_fp16 = mul(x = var_99_cast_fp16, y = _inversed_101_y_0_to_fp16)[name = string("_inversed_101_cast_fp16")];
+            fp16 var_102_to_fp16 = const()[name = string("op_102_to_fp16"), val = fp16(0x1p+0)];
+            tensor<fp16, [1]> lengths_13_cast_fp16 = add(x = _inversed_101_cast_fp16, y = var_102_to_fp16)[name = string("lengths_13_cast_fp16")];
+            tensor<fp16, [1]> lengths_cast_fp16 = floor(x = lengths_13_cast_fp16)[name = string("lengths_cast_fp16")];
+            tensor<int32, [1]> input_1_axes_0 = const()[name = string("input_1_axes_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1, 128, 1001]> audio_signal_to_fp16 = cast(dtype = audio_signal_to_fp16_dtype_0, x = audio_signal)[name = string("cast_1")];
+            tensor<fp16, [1, 1001, 128]> x_1_cast_fp16 = transpose(perm = x_1_perm_0, x = audio_signal_to_fp16)[name = string("transpose_315")];
+            tensor<fp16, [1, 1, 1001, 128]> input_1_cast_fp16 = expand_dims(axes = input_1_axes_0, x = x_1_cast_fp16)[name = string("input_1_cast_fp16")];
+            string input_3_pad_type_0 = const()[name = string("input_3_pad_type_0"), val = string("custom")];
+            tensor<int32, [4]> input_3_pad_0 = const()[name = string("input_3_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
+            tensor<int32, [2]> input_3_strides_0 = const()[name = string("input_3_strides_0"), val = tensor<int32, [2]>([2, 2])];
+            tensor<int32, [2]> input_3_dilations_0 = const()[name = string("input_3_dilations_0"), val = tensor<int32, [2]>([1, 1])];
+            int32 input_3_groups_0 = const()[name = string("input_3_groups_0"), val = int32(1)];
+            tensor<fp16, [256, 1, 3, 3]> model_pre_encode_conv_0_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [256, 1, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64))), lut = tensor<fp16, [8, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1280))))[name = string("model_pre_encode_conv_0_weight_to_fp16_palettized")];
+            tensor<fp16, [256]> model_pre_encode_conv_0_bias_to_fp16 = const()[name = string("model_pre_encode_conv_0_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1600)))];
+            tensor<fp16, [1, 256, 501, 64]> input_3_cast_fp16 = conv(bias = model_pre_encode_conv_0_bias_to_fp16, dilations = input_3_dilations_0, groups = input_3_groups_0, pad = input_3_pad_0, pad_type = input_3_pad_type_0, strides = input_3_strides_0, weight = model_pre_encode_conv_0_weight_to_fp16_palettized, x = input_1_cast_fp16)[name = string("input_3_cast_fp16")];
+            tensor<fp16, [1, 256, 501, 64]> input_5_cast_fp16 = relu(x = input_3_cast_fp16)[name = string("input_5_cast_fp16")];
+            string input_7_pad_type_0 = const()[name = string("input_7_pad_type_0"), val = string("custom")];
+            tensor<int32, [4]> input_7_pad_0 = const()[name = string("input_7_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
+            tensor<int32, [2]> input_7_strides_0 = const()[name = string("input_7_strides_0"), val = tensor<int32, [2]>([2, 2])];
+            int32 input_7_groups_0 = const()[name = string("input_7_groups_0"), val = int32(256)];
+            tensor<int32, [2]> input_7_dilations_0 = const()[name = string("input_7_dilations_0"), val = tensor<int32, [2]>([1, 1])];
+            tensor<fp16, [256, 1, 3, 3]> model_pre_encode_conv_2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [256, 1, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2176))), lut = tensor<fp16, [8, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3392))))[name = string("model_pre_encode_conv_2_weight_to_fp16_palettized")];
+            tensor<fp16, [256]> model_pre_encode_conv_2_bias_to_fp16 = const()[name = string("model_pre_encode_conv_2_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3712)))];
+            tensor<fp16, [1, 256, 251, 32]> input_7_cast_fp16 = conv(bias = model_pre_encode_conv_2_bias_to_fp16, dilations = input_7_dilations_0, groups = input_7_groups_0, pad = input_7_pad_0, pad_type = input_7_pad_type_0, strides = input_7_strides_0, weight = model_pre_encode_conv_2_weight_to_fp16_palettized, x = input_5_cast_fp16)[name = string("input_7_cast_fp16")];
+            string input_9_pad_type_0 = const()[name = string("input_9_pad_type_0"), val = string("valid")];
+            tensor<int32, [2]> input_9_strides_0 = const()[name = string("input_9_strides_0"), val = tensor<int32, [2]>([1, 1])];
+            tensor<int32, [4]> input_9_pad_0 = const()[name = string("input_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [2]> input_9_dilations_0 = const()[name = string("input_9_dilations_0"), val = tensor<int32, [2]>([1, 1])];
+            int32 input_9_groups_0 = const()[name = string("input_9_groups_0"), val = int32(1)];
+            tensor<fp16, [256, 256, 1, 1]> model_pre_encode_conv_3_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [256, 256, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4288))), lut = tensor<fp16, [8, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37120))))[name = string("model_pre_encode_conv_3_weight_to_fp16_palettized")];
+            tensor<fp16, [256]> model_pre_encode_conv_3_bias_to_fp16 = const()[name = string("model_pre_encode_conv_3_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37440)))];
+            tensor<fp16, [1, 256, 251, 32]> input_9_cast_fp16 = conv(bias = model_pre_encode_conv_3_bias_to_fp16, dilations = input_9_dilations_0, groups = input_9_groups_0, pad = input_9_pad_0, pad_type = input_9_pad_type_0, strides = input_9_strides_0, weight = model_pre_encode_conv_3_weight_to_fp16_palettized, x = input_7_cast_fp16)[name = string("input_9_cast_fp16")];
+            tensor<fp16, [1, 256, 251, 32]> input_11_cast_fp16 = relu(x = input_9_cast_fp16)[name = string("input_11_cast_fp16")];
+            string input_13_pad_type_0 = const()[name = string("input_13_pad_type_0"), val = string("custom")];
+            tensor<int32, [4]> input_13_pad_0 = const()[name = string("input_13_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
+            tensor<int32, [2]> input_13_strides_0 = const()[name = string("input_13_strides_0"), val = tensor<int32, [2]>([2, 2])];
+            int32 input_13_groups_0 = const()[name = string("input_13_groups_0"), val = int32(256)];
+            tensor<int32, [2]> input_13_dilations_0 = const()[name = string("input_13_dilations_0"), val = tensor<int32, [2]>([1, 1])];
+            tensor<fp16, [256, 1, 3, 3]> model_pre_encode_conv_5_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [256, 1, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38016))), lut = tensor<fp16, [8, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39232))))[name = string("model_pre_encode_conv_5_weight_to_fp16_palettized")];
+            tensor<fp16, [256]> model_pre_encode_conv_5_bias_to_fp16 = const()[name = string("model_pre_encode_conv_5_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39552)))];
+            tensor<fp16, [1, 256, 126, 16]> input_13_cast_fp16 = conv(bias = model_pre_encode_conv_5_bias_to_fp16, dilations = input_13_dilations_0, groups = input_13_groups_0, pad = input_13_pad_0, pad_type = input_13_pad_type_0, strides = input_13_strides_0, weight = model_pre_encode_conv_5_weight_to_fp16_palettized, x = input_11_cast_fp16)[name = string("input_13_cast_fp16")];
+            string input_15_pad_type_0 = const()[name = string("input_15_pad_type_0"), val = string("valid")];
+            tensor<int32, [2]> input_15_strides_0 = const()[name = string("input_15_strides_0"), val = tensor<int32, [2]>([1, 1])];
+            tensor<int32, [4]> input_15_pad_0 = const()[name = string("input_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [2]> input_15_dilations_0 = const()[name = string("input_15_dilations_0"), val = tensor<int32, [2]>([1, 1])];
+            int32 input_15_groups_0 = const()[name = string("input_15_groups_0"), val = int32(1)];
+            tensor<fp16, [256, 256, 1, 1]> model_pre_encode_conv_6_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [256, 256, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40128))), lut = tensor<fp16, [8, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72960))))[name = string("model_pre_encode_conv_6_weight_to_fp16_palettized")];
+            tensor<fp16, [256]> model_pre_encode_conv_6_bias_to_fp16 = const()[name = string("model_pre_encode_conv_6_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(73280)))];
+            tensor<fp16, [1, 256, 126, 16]> input_15_cast_fp16 = conv(bias = model_pre_encode_conv_6_bias_to_fp16, dilations = input_15_dilations_0, groups = input_15_groups_0, pad = input_15_pad_0, pad_type = input_15_pad_type_0, strides = input_15_strides_0, weight = model_pre_encode_conv_6_weight_to_fp16_palettized, x = input_13_cast_fp16)[name = string("input_15_cast_fp16")];
+            tensor<fp16, [1, 256, 126, 16]> x_3_cast_fp16 = relu(x = input_15_cast_fp16)[name = string("x_3_cast_fp16")];
+            tensor<int32, [4]> var_152_perm_0 = const()[name = string("op_152_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_153 = const()[name = string("op_153"), val = tensor<int32, [3]>([1, 126, -1])];
+            tensor<fp16, [1, 126, 256, 16]> var_152_cast_fp16 = transpose(perm = var_152_perm_0, x = x_3_cast_fp16)[name = string("transpose_314")];
+            tensor<fp16, [1, 126, 4096]> input_17_cast_fp16 = reshape(shape = var_153, x = var_152_cast_fp16)[name = string("input_17_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_pre_encode_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(73856))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2171072))))[name = string("model_pre_encode_out_weight_to_fp16_palettized")];
+            tensor<fp16, [1024]> model_pre_encode_out_bias_to_fp16 = const()[name = string("model_pre_encode_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2172160)))];
+            tensor<fp16, [1, 126, 1024]> linear_0_cast_fp16 = linear(bias = model_pre_encode_out_bias_to_fp16, weight = model_pre_encode_out_weight_to_fp16_palettized, x = input_17_cast_fp16)[name = string("linear_0_cast_fp16")];
+            string cast_11_dtype_0 = const()[name = string("cast_11_dtype_0"), val = string("int32")];
+            tensor<int32, [1, 126]> expand_dims_0 = const()[name = string("expand_dims_0"), val = tensor<int32, [1, 126]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2174272)))];
+            tensor<int32, [1]> var_191_axes_0 = const()[name = string("op_191_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<int32, [1]> encoder_output_length = cast(dtype = cast_11_dtype_0, x = lengths_cast_fp16)[name = string("cast_0")];
+            tensor<int32, [1, 1]> var_191 = expand_dims(axes = var_191_axes_0, x = encoder_output_length)[name = string("op_191")];
+            tensor<bool, [1, 126]> pad_mask_1 = less(x = expand_dims_0, y = var_191)[name = string("pad_mask_1")];
+            tensor<int32, [1]> var_193_axes_0 = const()[name = string("op_193_axes_0"), val = tensor<int32, [1]>([1])];
+            tensor<bool, [1, 1, 126]> var_193 = expand_dims(axes = var_193_axes_0, x = pad_mask_1)[name = string("op_193")];
+            tensor<int32, [3]> var_194 = const()[name = string("op_194"), val = tensor<int32, [3]>([1, 126, 1])];
+            tensor<bool, [1, 126, 126]> pad_mask_for_att_mask_1 = tile(reps = var_194, x = var_193)[name = string("pad_mask_for_att_mask_1")];
+            tensor<int32, [3]> var_196_perm_0 = const()[name = string("op_196_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<bool, [1, 126, 126]> var_196 = transpose(perm = var_196_perm_0, x = pad_mask_for_att_mask_1)[name = string("transpose_313")];
+            tensor<bool, [1, 126, 126]> pad_mask_for_att_mask = logical_and(x = pad_mask_for_att_mask_1, y = var_196)[name = string("pad_mask_for_att_mask")];
+            tensor<bool, [1, 126, 126]> const_7 = const()[name = string("const_7"), val = tensor<bool, [1, 126, 126]>([[[true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true]]])];
+            tensor<bool, [1, 126, 126]> att_mask = logical_and(x = pad_mask_for_att_mask, y = const_7)[name = string("att_mask")];
+            tensor<bool, [1, 126, 126]> mask_1 = logical_not(x = att_mask)[name = string("mask_1")];
+            tensor<bool, [1, 126]> pad_mask = logical_not(x = pad_mask_1)[name = string("pad_mask")];
+            tensor<int32, [1]> input_21_axes_0 = const()[name = string("input_21_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_0_norm_feed_forward1_weight_to_fp16 = const()[name = string("model_layers_0_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2174848)))];
+            tensor<fp16, [1024]> model_layers_0_norm_feed_forward1_bias_to_fp16 = const()[name = string("model_layers_0_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2176960)))];
+            fp16 var_4_to_fp16 = const()[name = string("op_4_to_fp16"), val = fp16(0x1.5p-17)];
+            tensor<fp16, [1, 126, 1024]> input_21_cast_fp16 = layer_norm(axes = input_21_axes_0, beta = model_layers_0_norm_feed_forward1_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_0_norm_feed_forward1_weight_to_fp16, x = linear_0_cast_fp16)[name = string("input_21_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_0_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2179072))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4276288))))[name = string("model_layers_0_feed_forward1_linear1_weight_to_fp16_palettized")];
+            tensor<fp16, [4096]> linear_1_bias_0_to_fp16 = const()[name = string("linear_1_bias_0_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4280448)))];
+            tensor<fp16, [1, 126, 4096]> linear_1_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_0_feed_forward1_linear1_weight_to_fp16_palettized, x = input_21_cast_fp16)[name = string("linear_1_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_25_cast_fp16 = silu(x = linear_1_cast_fp16)[name = string("input_25_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_0_feed_forward1_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4288704))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6385920))))[name = string("model_layers_0_feed_forward1_linear2_weight_to_fp16_palettized")];
+            tensor<fp16, [1024]> linear_2_bias_0_to_fp16 = const()[name = string("linear_2_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6387008)))];
+            tensor<fp16, [1, 126, 1024]> linear_2_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_0_feed_forward1_linear2_weight_to_fp16_palettized, x = input_25_cast_fp16)[name = string("linear_2_cast_fp16")];
+            fp16 var_227_to_fp16 = const()[name = string("op_227_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_228_cast_fp16 = mul(x = linear_2_cast_fp16, y = var_227_to_fp16)[name = string("op_228_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_31_cast_fp16 = add(x = linear_0_cast_fp16, y = var_228_cast_fp16)[name = string("input_31_cast_fp16")];
+            tensor<int32, [1]> query_1_axes_0 = const()[name = string("query_1_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_0_norm_self_att_weight_to_fp16 = const()[name = string("model_layers_0_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6389120)))];
+            tensor<fp16, [1024]> model_layers_0_norm_self_att_bias_to_fp16 = const()[name = string("model_layers_0_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6391232)))];
+            tensor<fp16, [1, 126, 1024]> query_1_cast_fp16 = layer_norm(axes = query_1_axes_0, beta = model_layers_0_norm_self_att_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_0_norm_self_att_weight_to_fp16, x = input_31_cast_fp16)[name = string("query_1_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_0_self_attn_linear_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6393344))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6917696))))[name = string("model_layers_0_self_attn_linear_q_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_3_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_0_self_attn_linear_q_weight_to_fp16_palettized, x = query_1_cast_fp16)[name = string("linear_3_cast_fp16")];
+            tensor<int32, [4]> var_244 = const()[name = string("op_244"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> q_1_cast_fp16 = reshape(shape = var_244, x = linear_3_cast_fp16)[name = string("q_1_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_0_self_attn_linear_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6918784))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7443136))))[name = string("model_layers_0_self_attn_linear_k_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_4_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_0_self_attn_linear_k_weight_to_fp16_palettized, x = query_1_cast_fp16)[name = string("linear_4_cast_fp16")];
+            tensor<int32, [4]> var_248 = const()[name = string("op_248"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> k_1_cast_fp16 = reshape(shape = var_248, x = linear_4_cast_fp16)[name = string("k_1_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_0_self_attn_linear_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7444224))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7968576))))[name = string("model_layers_0_self_attn_linear_v_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_5_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_0_self_attn_linear_v_weight_to_fp16_palettized, x = query_1_cast_fp16)[name = string("linear_5_cast_fp16")];
+            tensor<int32, [4]> var_252 = const()[name = string("op_252"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> v_1_cast_fp16 = reshape(shape = var_252, x = linear_5_cast_fp16)[name = string("v_1_cast_fp16")];
+            tensor<int32, [4]> value_1_perm_0 = const()[name = string("value_1_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> model_layers_0_self_attn_pos_bias_u_to_fp16 = const()[name = string("model_layers_0_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7969664)))];
+            tensor<fp16, [1, 126, 8, 128]> var_264_cast_fp16 = add(x = q_1_cast_fp16, y = model_layers_0_self_attn_pos_bias_u_to_fp16)[name = string("op_264_cast_fp16")];
+            tensor<fp16, [8, 128]> model_layers_0_self_attn_pos_bias_v_to_fp16 = const()[name = string("model_layers_0_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7971776)))];
+            tensor<fp16, [1, 126, 8, 128]> var_266_cast_fp16 = add(x = q_1_cast_fp16, y = model_layers_0_self_attn_pos_bias_v_to_fp16)[name = string("op_266_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_1_perm_0 = const()[name = string("q_with_bias_v_1_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_7_transpose_x_0 = const()[name = string("x_7_transpose_x_0"), val = bool(false)];
+            bool x_7_transpose_y_0 = const()[name = string("x_7_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 251]> var_268_to_fp16 = const()[name = string("op_268_to_fp16"), val = tensor<fp16, [1, 8, 128, 251]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7973888)))];
+            tensor<fp16, [1, 8, 126, 128]> q_with_bias_v_1_cast_fp16 = transpose(perm = q_with_bias_v_1_perm_0, x = var_266_cast_fp16)[name = string("transpose_312")];
+            tensor<fp16, [1, 8, 126, 251]> x_7_cast_fp16 = matmul(transpose_x = x_7_transpose_x_0, transpose_y = x_7_transpose_y_0, x = q_with_bias_v_1_cast_fp16, y = var_268_to_fp16)[name = string("x_7_cast_fp16")];
+            tensor<int32, [8]> x_9_pad_0 = const()[name = string("x_9_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_9_mode_0 = const()[name = string("x_9_mode_0"), val = string("constant")];
+            fp16 const_14_to_fp16 = const()[name = string("const_14_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 252]> x_9_cast_fp16 = pad(constant_val = const_14_to_fp16, mode = x_9_mode_0, pad = x_9_pad_0, x = x_7_cast_fp16)[name = string("x_9_cast_fp16")];
+            tensor<int32, [4]> var_276 = const()[name = string("op_276"), val = tensor<int32, [4]>([1, 8, -1, 126])];
+            tensor<fp16, [1, 8, 252, 126]> x_11_cast_fp16 = reshape(shape = var_276, x = x_9_cast_fp16)[name = string("x_11_cast_fp16")];
+            tensor<int32, [4]> var_280_begin_0 = const()[name = string("op_280_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_280_end_0 = const()[name = string("op_280_end_0"), val = tensor<int32, [4]>([1, 8, 252, 126])];
+            tensor<bool, [4]> var_280_end_mask_0 = const()[name = string("op_280_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 251, 126]> var_280_cast_fp16 = slice_by_index(begin = var_280_begin_0, end = var_280_end_0, end_mask = var_280_end_mask_0, x = x_11_cast_fp16)[name = string("op_280_cast_fp16")];
+            tensor<int32, [4]> var_281 = const()[name = string("op_281"), val = tensor<int32, [4]>([1, 8, 126, 251])];
+            tensor<fp16, [1, 8, 126, 251]> matrix_bd_1_cast_fp16 = reshape(shape = var_281, x = var_280_cast_fp16)[name = string("matrix_bd_1_cast_fp16")];
+            bool matrix_ac_1_transpose_x_0 = const()[name = string("matrix_ac_1_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_1_transpose_y_0 = const()[name = string("matrix_ac_1_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_96_perm_0 = const()[name = string("transpose_96_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_97_perm_0 = const()[name = string("transpose_97_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 126]> transpose_97 = transpose(perm = transpose_97_perm_0, x = k_1_cast_fp16)[name = string("transpose_310")];
+            tensor<fp16, [1, 8, 126, 128]> transpose_96 = transpose(perm = transpose_96_perm_0, x = var_264_cast_fp16)[name = string("transpose_311")];
+            tensor<fp16, [1, 8, 126, 126]> matrix_ac_1_cast_fp16 = matmul(transpose_x = matrix_ac_1_transpose_x_0, transpose_y = matrix_ac_1_transpose_y_0, x = transpose_96, y = transpose_97)[name = string("matrix_ac_1_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_3_begin_0 = const()[name = string("matrix_bd_3_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_3_end_0 = const()[name = string("matrix_bd_3_end_0"), val = tensor<int32, [4]>([1, 8, 126, 126])];
+            tensor<bool, [4]> matrix_bd_3_end_mask_0 = const()[name = string("matrix_bd_3_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 126, 126]> matrix_bd_3_cast_fp16 = slice_by_index(begin = matrix_bd_3_begin_0, end = matrix_bd_3_end_0, end_mask = matrix_bd_3_end_mask_0, x = matrix_bd_1_cast_fp16)[name = string("matrix_bd_3_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_290_cast_fp16 = add(x = matrix_ac_1_cast_fp16, y = matrix_bd_3_cast_fp16)[name = string("op_290_cast_fp16")];
+            fp16 _inversed_scores_1_y_0_to_fp16 = const()[name = string("_inversed_scores_1_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 126, 126]> _inversed_scores_1_cast_fp16 = mul(x = var_290_cast_fp16, y = _inversed_scores_1_y_0_to_fp16)[name = string("_inversed_scores_1_cast_fp16")];
+            tensor<int32, [1]> mask_3_axes_0 = const()[name = string("mask_3_axes_0"), val = tensor<int32, [1]>([1])];
+            tensor<bool, [1, 1, 126, 126]> mask_3 = expand_dims(axes = mask_3_axes_0, x = mask_1)[name = string("mask_3")];
+            fp16 var_7_to_fp16 = const()[name = string("op_7_to_fp16"), val = fp16(-0x1.388p+13)];
+            tensor<fp16, [1, 8, 126, 126]> scores_3_cast_fp16 = select(a = var_7_to_fp16, b = _inversed_scores_1_cast_fp16, cond = mask_3)[name = string("scores_3_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_296_cast_fp16 = softmax(axis = var_25, x = scores_3_cast_fp16)[name = string("op_296_cast_fp16")];
+            fp16 var_6_to_fp16 = const()[name = string("op_6_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 126]> input_33_cast_fp16 = select(a = var_6_to_fp16, b = var_296_cast_fp16, cond = mask_3)[name = string("input_33_cast_fp16")];
+            bool x_13_transpose_x_0 = const()[name = string("x_13_transpose_x_0"), val = bool(false)];
+            bool x_13_transpose_y_0 = const()[name = string("x_13_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 126, 128]> value_1_cast_fp16 = transpose(perm = value_1_perm_0, x = v_1_cast_fp16)[name = string("transpose_309")];
+            tensor<fp16, [1, 8, 126, 128]> x_13_cast_fp16 = matmul(transpose_x = x_13_transpose_x_0, transpose_y = x_13_transpose_y_0, x = input_33_cast_fp16, y = value_1_cast_fp16)[name = string("x_13_cast_fp16")];
+            tensor<int32, [4]> var_300_perm_0 = const()[name = string("op_300_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_301 = const()[name = string("op_301"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 126, 8, 128]> var_300_cast_fp16 = transpose(perm = var_300_perm_0, x = x_13_cast_fp16)[name = string("transpose_308")];
+            tensor<fp16, [1, 126, 1024]> input_35_cast_fp16 = reshape(shape = var_301, x = var_300_cast_fp16)[name = string("input_35_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_0_self_attn_linear_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8488000))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9012352))))[name = string("model_layers_0_self_attn_linear_out_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_7_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_0_self_attn_linear_out_weight_to_fp16_palettized, x = input_35_cast_fp16)[name = string("linear_7_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_39_cast_fp16 = add(x = input_31_cast_fp16, y = linear_7_cast_fp16)[name = string("input_39_cast_fp16")];
+            tensor<int32, [1]> x_17_axes_0 = const()[name = string("x_17_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_0_norm_conv_weight_to_fp16 = const()[name = string("model_layers_0_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9013440)))];
+            tensor<fp16, [1024]> model_layers_0_norm_conv_bias_to_fp16 = const()[name = string("model_layers_0_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9015552)))];
+            tensor<fp16, [1, 126, 1024]> x_17_cast_fp16 = layer_norm(axes = x_17_axes_0, beta = model_layers_0_norm_conv_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_0_norm_conv_weight_to_fp16, x = input_39_cast_fp16)[name = string("x_17_cast_fp16")];
+            tensor<int32, [3]> input_41_perm_0 = const()[name = string("input_41_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_43_pad_type_0 = const()[name = string("input_43_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_43_strides_0 = const()[name = string("input_43_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_43_pad_0 = const()[name = string("input_43_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_43_dilations_0 = const()[name = string("input_43_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_43_groups_0 = const()[name = string("input_43_groups_0"), val = int32(1)];
+            tensor<fp16, [2048, 1024, 1]> model_layers_0_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9017664))), lut = tensor<fp16, [64, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10066304))))[name = string("model_layers_0_conv_pointwise_conv1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 1024, 126]> input_41_cast_fp16 = transpose(perm = input_41_perm_0, x = x_17_cast_fp16)[name = string("transpose_307")];
+            tensor<fp16, [1, 2048, 126]> input_43_cast_fp16 = conv(dilations = input_43_dilations_0, groups = input_43_groups_0, pad = input_43_pad_0, pad_type = input_43_pad_type_0, strides = input_43_strides_0, weight = model_layers_0_conv_pointwise_conv1_weight_to_fp16_palettized, x = input_41_cast_fp16)[name = string("input_43_cast_fp16")];
+            int32 x_19_split_num_splits_0 = const()[name = string("x_19_split_num_splits_0"), val = int32(2)];
+            int32 x_19_split_axis_0 = const()[name = string("x_19_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 126]> x_19_split_cast_fp16_0, tensor<fp16, [1, 1024, 126]> x_19_split_cast_fp16_1 = split(axis = x_19_split_axis_0, num_splits = x_19_split_num_splits_0, x = input_43_cast_fp16)[name = string("x_19_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_19_split_1_sigmoid_cast_fp16 = sigmoid(x = x_19_split_cast_fp16_1)[name = string("x_19_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_19_cast_fp16 = mul(x = x_19_split_cast_fp16_0, y = x_19_split_1_sigmoid_cast_fp16)[name = string("x_19_cast_fp16")];
+            tensor<int32, [1]> var_323_axes_0 = const()[name = string("op_323_axes_0"), val = tensor<int32, [1]>([1])];
+            tensor<bool, [1, 1, 126]> var_323 = expand_dims(axes = var_323_axes_0, x = pad_mask)[name = string("op_323")];
+            tensor<fp16, [1, 1024, 126]> input_45_cast_fp16 = select(a = var_6_to_fp16, b = x_19_cast_fp16, cond = var_323)[name = string("input_45_cast_fp16")];
+            tensor<int32, [6]> input_47_pad_0 = const()[name = string("input_47_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_47_mode_0 = const()[name = string("input_47_mode_0"), val = string("constant")];
+            fp16 const_17_to_fp16 = const()[name = string("const_17_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 134]> input_47_cast_fp16 = pad(constant_val = const_17_to_fp16, mode = input_47_mode_0, pad = input_47_pad_0, x = input_45_cast_fp16)[name = string("input_47_cast_fp16")];
+            string input_49_pad_type_0 = const()[name = string("input_49_pad_type_0"), val = string("valid")];
+            int32 input_49_groups_0 = const()[name = string("input_49_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_49_strides_0 = const()[name = string("input_49_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_49_pad_0 = const()[name = string("input_49_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_49_dilations_0 = const()[name = string("input_49_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_248_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10068416))), lut = tensor<fp16, [32, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10073088))))[name = string("const_248_to_fp16_palettized")];
+            tensor<fp16, [1024]> const_249_to_fp16 = const()[name = string("const_249_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10074176)))];
+            tensor<fp16, [1, 1024, 126]> input_51_cast_fp16 = conv(bias = const_249_to_fp16, dilations = input_49_dilations_0, groups = input_49_groups_0, pad = input_49_pad_0, pad_type = input_49_pad_type_0, strides = input_49_strides_0, weight = const_248_to_fp16_palettized, x = input_47_cast_fp16)[name = string("input_51_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_53_cast_fp16 = silu(x = input_51_cast_fp16)[name = string("input_53_cast_fp16")];
+            string x_21_pad_type_0 = const()[name = string("x_21_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_21_strides_0 = const()[name = string("x_21_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_21_pad_0 = const()[name = string("x_21_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_21_dilations_0 = const()[name = string("x_21_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_21_groups_0 = const()[name = string("x_21_groups_0"), val = int32(1)];
+            tensor<fp16, [1024, 1024, 1]> model_layers_0_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10076288))), lut = tensor<fp16, [32, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10600640))))[name = string("model_layers_0_conv_pointwise_conv2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 1024, 126]> x_21_cast_fp16 = conv(dilations = x_21_dilations_0, groups = x_21_groups_0, pad = x_21_pad_0, pad_type = x_21_pad_type_0, strides = x_21_strides_0, weight = model_layers_0_conv_pointwise_conv2_weight_to_fp16_palettized, x = input_53_cast_fp16)[name = string("x_21_cast_fp16")];
+            tensor<int32, [3]> input_55_perm_0 = const()[name = string("input_55_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 126, 1024]> input_55_cast_fp16 = transpose(perm = input_55_perm_0, x = x_21_cast_fp16)[name = string("transpose_306")];
+            tensor<fp16, [1, 126, 1024]> input_57_cast_fp16 = add(x = input_39_cast_fp16, y = input_55_cast_fp16)[name = string("input_57_cast_fp16")];
+            tensor<int32, [1]> input_59_axes_0 = const()[name = string("input_59_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_0_norm_feed_forward2_weight_to_fp16 = const()[name = string("model_layers_0_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10601728)))];
+            tensor<fp16, [1024]> model_layers_0_norm_feed_forward2_bias_to_fp16 = const()[name = string("model_layers_0_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10603840)))];
+            tensor<fp16, [1, 126, 1024]> input_59_cast_fp16 = layer_norm(axes = input_59_axes_0, beta = model_layers_0_norm_feed_forward2_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_0_norm_feed_forward2_weight_to_fp16, x = input_57_cast_fp16)[name = string("input_59_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_0_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10605952))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12703168))))[name = string("model_layers_0_feed_forward2_linear1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 4096]> linear_8_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_0_feed_forward2_linear1_weight_to_fp16_palettized, x = input_59_cast_fp16)[name = string("linear_8_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_63_cast_fp16 = silu(x = linear_8_cast_fp16)[name = string("input_63_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_0_feed_forward2_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12707328))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14804544))))[name = string("model_layers_0_feed_forward2_linear2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_9_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_0_feed_forward2_linear2_weight_to_fp16_palettized, x = input_63_cast_fp16)[name = string("linear_9_cast_fp16")];
+            fp16 var_361_to_fp16 = const()[name = string("op_361_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_362_cast_fp16 = mul(x = linear_9_cast_fp16, y = var_361_to_fp16)[name = string("op_362_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_69_cast_fp16 = add(x = input_57_cast_fp16, y = var_362_cast_fp16)[name = string("input_69_cast_fp16")];
+            tensor<int32, [1]> input_71_axes_0 = const()[name = string("input_71_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_0_norm_out_weight_to_fp16 = const()[name = string("model_layers_0_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14805632)))];
+            tensor<fp16, [1024]> model_layers_0_norm_out_bias_to_fp16 = const()[name = string("model_layers_0_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14807744)))];
+            tensor<fp16, [1, 126, 1024]> input_71_cast_fp16 = layer_norm(axes = input_71_axes_0, beta = model_layers_0_norm_out_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_0_norm_out_weight_to_fp16, x = input_69_cast_fp16)[name = string("input_71_cast_fp16")];
+            tensor<int32, [1]> input_73_axes_0 = const()[name = string("input_73_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_1_norm_feed_forward1_weight_to_fp16 = const()[name = string("model_layers_1_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14809856)))];
+            tensor<fp16, [1024]> model_layers_1_norm_feed_forward1_bias_to_fp16 = const()[name = string("model_layers_1_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14811968)))];
+            tensor<fp16, [1, 126, 1024]> input_73_cast_fp16 = layer_norm(axes = input_73_axes_0, beta = model_layers_1_norm_feed_forward1_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_1_norm_feed_forward1_weight_to_fp16, x = input_71_cast_fp16)[name = string("input_73_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_1_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14814080))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16911296))))[name = string("model_layers_1_feed_forward1_linear1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 4096]> linear_10_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_1_feed_forward1_linear1_weight_to_fp16_palettized, x = input_73_cast_fp16)[name = string("linear_10_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_77_cast_fp16 = silu(x = linear_10_cast_fp16)[name = string("input_77_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_1_feed_forward1_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16915456))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19012672))))[name = string("model_layers_1_feed_forward1_linear2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_11_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_1_feed_forward1_linear2_weight_to_fp16_palettized, x = input_77_cast_fp16)[name = string("linear_11_cast_fp16")];
+            fp16 var_390_to_fp16 = const()[name = string("op_390_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_391_cast_fp16 = mul(x = linear_11_cast_fp16, y = var_390_to_fp16)[name = string("op_391_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_83_cast_fp16 = add(x = input_71_cast_fp16, y = var_391_cast_fp16)[name = string("input_83_cast_fp16")];
+            tensor<int32, [1]> query_3_axes_0 = const()[name = string("query_3_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_1_norm_self_att_weight_to_fp16 = const()[name = string("model_layers_1_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19013760)))];
+            tensor<fp16, [1024]> model_layers_1_norm_self_att_bias_to_fp16 = const()[name = string("model_layers_1_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19015872)))];
+            tensor<fp16, [1, 126, 1024]> query_3_cast_fp16 = layer_norm(axes = query_3_axes_0, beta = model_layers_1_norm_self_att_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_1_norm_self_att_weight_to_fp16, x = input_83_cast_fp16)[name = string("query_3_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_1_self_attn_linear_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19017984))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19542336))))[name = string("model_layers_1_self_attn_linear_q_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_12_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_1_self_attn_linear_q_weight_to_fp16_palettized, x = query_3_cast_fp16)[name = string("linear_12_cast_fp16")];
+            tensor<int32, [4]> var_407 = const()[name = string("op_407"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> q_7_cast_fp16 = reshape(shape = var_407, x = linear_12_cast_fp16)[name = string("q_7_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_1_self_attn_linear_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19543424))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20067776))))[name = string("model_layers_1_self_attn_linear_k_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_13_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_1_self_attn_linear_k_weight_to_fp16_palettized, x = query_3_cast_fp16)[name = string("linear_13_cast_fp16")];
+            tensor<int32, [4]> var_411 = const()[name = string("op_411"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> k_5_cast_fp16 = reshape(shape = var_411, x = linear_13_cast_fp16)[name = string("k_5_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_1_self_attn_linear_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20068864))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20593216))))[name = string("model_layers_1_self_attn_linear_v_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_14_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_1_self_attn_linear_v_weight_to_fp16_palettized, x = query_3_cast_fp16)[name = string("linear_14_cast_fp16")];
+            tensor<int32, [4]> var_415 = const()[name = string("op_415"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> v_3_cast_fp16 = reshape(shape = var_415, x = linear_14_cast_fp16)[name = string("v_3_cast_fp16")];
+            tensor<int32, [4]> value_3_perm_0 = const()[name = string("value_3_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> model_layers_1_self_attn_pos_bias_u_to_fp16 = const()[name = string("model_layers_1_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20594304)))];
+            tensor<fp16, [1, 126, 8, 128]> var_427_cast_fp16 = add(x = q_7_cast_fp16, y = model_layers_1_self_attn_pos_bias_u_to_fp16)[name = string("op_427_cast_fp16")];
+            tensor<fp16, [8, 128]> model_layers_1_self_attn_pos_bias_v_to_fp16 = const()[name = string("model_layers_1_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20596416)))];
+            tensor<fp16, [1, 126, 8, 128]> var_429_cast_fp16 = add(x = q_7_cast_fp16, y = model_layers_1_self_attn_pos_bias_v_to_fp16)[name = string("op_429_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_3_perm_0 = const()[name = string("q_with_bias_v_3_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_29_transpose_x_0 = const()[name = string("x_29_transpose_x_0"), val = bool(false)];
+            bool x_29_transpose_y_0 = const()[name = string("x_29_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 251]> var_431_to_fp16 = const()[name = string("op_431_to_fp16"), val = tensor<fp16, [1, 8, 128, 251]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20598528)))];
+            tensor<fp16, [1, 8, 126, 128]> q_with_bias_v_3_cast_fp16 = transpose(perm = q_with_bias_v_3_perm_0, x = var_429_cast_fp16)[name = string("transpose_305")];
+            tensor<fp16, [1, 8, 126, 251]> x_29_cast_fp16 = matmul(transpose_x = x_29_transpose_x_0, transpose_y = x_29_transpose_y_0, x = q_with_bias_v_3_cast_fp16, y = var_431_to_fp16)[name = string("x_29_cast_fp16")];
+            tensor<int32, [8]> x_31_pad_0 = const()[name = string("x_31_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_31_mode_0 = const()[name = string("x_31_mode_0"), val = string("constant")];
+            fp16 const_24_to_fp16 = const()[name = string("const_24_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 252]> x_31_cast_fp16 = pad(constant_val = const_24_to_fp16, mode = x_31_mode_0, pad = x_31_pad_0, x = x_29_cast_fp16)[name = string("x_31_cast_fp16")];
+            tensor<int32, [4]> var_439 = const()[name = string("op_439"), val = tensor<int32, [4]>([1, 8, -1, 126])];
+            tensor<fp16, [1, 8, 252, 126]> x_33_cast_fp16 = reshape(shape = var_439, x = x_31_cast_fp16)[name = string("x_33_cast_fp16")];
+            tensor<int32, [4]> var_443_begin_0 = const()[name = string("op_443_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_443_end_0 = const()[name = string("op_443_end_0"), val = tensor<int32, [4]>([1, 8, 252, 126])];
+            tensor<bool, [4]> var_443_end_mask_0 = const()[name = string("op_443_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 251, 126]> var_443_cast_fp16 = slice_by_index(begin = var_443_begin_0, end = var_443_end_0, end_mask = var_443_end_mask_0, x = x_33_cast_fp16)[name = string("op_443_cast_fp16")];
+            tensor<int32, [4]> var_444 = const()[name = string("op_444"), val = tensor<int32, [4]>([1, 8, 126, 251])];
+            tensor<fp16, [1, 8, 126, 251]> matrix_bd_5_cast_fp16 = reshape(shape = var_444, x = var_443_cast_fp16)[name = string("matrix_bd_5_cast_fp16")];
+            bool matrix_ac_3_transpose_x_0 = const()[name = string("matrix_ac_3_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_3_transpose_y_0 = const()[name = string("matrix_ac_3_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_98_perm_0 = const()[name = string("transpose_98_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_99_perm_0 = const()[name = string("transpose_99_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 126]> transpose_99 = transpose(perm = transpose_99_perm_0, x = k_5_cast_fp16)[name = string("transpose_303")];
+            tensor<fp16, [1, 8, 126, 128]> transpose_98 = transpose(perm = transpose_98_perm_0, x = var_427_cast_fp16)[name = string("transpose_304")];
+            tensor<fp16, [1, 8, 126, 126]> matrix_ac_3_cast_fp16 = matmul(transpose_x = matrix_ac_3_transpose_x_0, transpose_y = matrix_ac_3_transpose_y_0, x = transpose_98, y = transpose_99)[name = string("matrix_ac_3_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_7_begin_0 = const()[name = string("matrix_bd_7_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_7_end_0 = const()[name = string("matrix_bd_7_end_0"), val = tensor<int32, [4]>([1, 8, 126, 126])];
+            tensor<bool, [4]> matrix_bd_7_end_mask_0 = const()[name = string("matrix_bd_7_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 126, 126]> matrix_bd_7_cast_fp16 = slice_by_index(begin = matrix_bd_7_begin_0, end = matrix_bd_7_end_0, end_mask = matrix_bd_7_end_mask_0, x = matrix_bd_5_cast_fp16)[name = string("matrix_bd_7_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_453_cast_fp16 = add(x = matrix_ac_3_cast_fp16, y = matrix_bd_7_cast_fp16)[name = string("op_453_cast_fp16")];
+            fp16 _inversed_scores_5_y_0_to_fp16 = const()[name = string("_inversed_scores_5_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 126, 126]> _inversed_scores_5_cast_fp16 = mul(x = var_453_cast_fp16, y = _inversed_scores_5_y_0_to_fp16)[name = string("_inversed_scores_5_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> scores_7_cast_fp16 = select(a = var_7_to_fp16, b = _inversed_scores_5_cast_fp16, cond = mask_3)[name = string("scores_7_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_459_cast_fp16 = softmax(axis = var_25, x = scores_7_cast_fp16)[name = string("op_459_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> input_85_cast_fp16 = select(a = var_6_to_fp16, b = var_459_cast_fp16, cond = mask_3)[name = string("input_85_cast_fp16")];
+            bool x_35_transpose_x_0 = const()[name = string("x_35_transpose_x_0"), val = bool(false)];
+            bool x_35_transpose_y_0 = const()[name = string("x_35_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 126, 128]> value_3_cast_fp16 = transpose(perm = value_3_perm_0, x = v_3_cast_fp16)[name = string("transpose_302")];
+            tensor<fp16, [1, 8, 126, 128]> x_35_cast_fp16 = matmul(transpose_x = x_35_transpose_x_0, transpose_y = x_35_transpose_y_0, x = input_85_cast_fp16, y = value_3_cast_fp16)[name = string("x_35_cast_fp16")];
+            tensor<int32, [4]> var_463_perm_0 = const()[name = string("op_463_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_464 = const()[name = string("op_464"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 126, 8, 128]> var_463_cast_fp16 = transpose(perm = var_463_perm_0, x = x_35_cast_fp16)[name = string("transpose_301")];
+            tensor<fp16, [1, 126, 1024]> input_87_cast_fp16 = reshape(shape = var_464, x = var_463_cast_fp16)[name = string("input_87_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_1_self_attn_linear_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21112640))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21636992))))[name = string("model_layers_1_self_attn_linear_out_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_16_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_1_self_attn_linear_out_weight_to_fp16_palettized, x = input_87_cast_fp16)[name = string("linear_16_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_91_cast_fp16 = add(x = input_83_cast_fp16, y = linear_16_cast_fp16)[name = string("input_91_cast_fp16")];
+            tensor<int32, [1]> x_39_axes_0 = const()[name = string("x_39_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_1_norm_conv_weight_to_fp16 = const()[name = string("model_layers_1_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21638080)))];
+            tensor<fp16, [1024]> model_layers_1_norm_conv_bias_to_fp16 = const()[name = string("model_layers_1_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21640192)))];
+            tensor<fp16, [1, 126, 1024]> x_39_cast_fp16 = layer_norm(axes = x_39_axes_0, beta = model_layers_1_norm_conv_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_1_norm_conv_weight_to_fp16, x = input_91_cast_fp16)[name = string("x_39_cast_fp16")];
+            tensor<int32, [3]> input_93_perm_0 = const()[name = string("input_93_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_95_pad_type_0 = const()[name = string("input_95_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_95_strides_0 = const()[name = string("input_95_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_95_pad_0 = const()[name = string("input_95_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_95_dilations_0 = const()[name = string("input_95_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_95_groups_0 = const()[name = string("input_95_groups_0"), val = int32(1)];
+            tensor<fp16, [2048, 1024, 1]> model_layers_1_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21642304))), lut = tensor<fp16, [64, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22690944))))[name = string("model_layers_1_conv_pointwise_conv1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 1024, 126]> input_93_cast_fp16 = transpose(perm = input_93_perm_0, x = x_39_cast_fp16)[name = string("transpose_300")];
+            tensor<fp16, [1, 2048, 126]> input_95_cast_fp16 = conv(dilations = input_95_dilations_0, groups = input_95_groups_0, pad = input_95_pad_0, pad_type = input_95_pad_type_0, strides = input_95_strides_0, weight = model_layers_1_conv_pointwise_conv1_weight_to_fp16_palettized, x = input_93_cast_fp16)[name = string("input_95_cast_fp16")];
+            int32 x_41_split_num_splits_0 = const()[name = string("x_41_split_num_splits_0"), val = int32(2)];
+            int32 x_41_split_axis_0 = const()[name = string("x_41_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 126]> x_41_split_cast_fp16_0, tensor<fp16, [1, 1024, 126]> x_41_split_cast_fp16_1 = split(axis = x_41_split_axis_0, num_splits = x_41_split_num_splits_0, x = input_95_cast_fp16)[name = string("x_41_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_41_split_1_sigmoid_cast_fp16 = sigmoid(x = x_41_split_cast_fp16_1)[name = string("x_41_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_41_cast_fp16 = mul(x = x_41_split_cast_fp16_0, y = x_41_split_1_sigmoid_cast_fp16)[name = string("x_41_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_97_cast_fp16 = select(a = var_6_to_fp16, b = x_41_cast_fp16, cond = var_323)[name = string("input_97_cast_fp16")];
+            tensor<int32, [6]> input_99_pad_0 = const()[name = string("input_99_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_99_mode_0 = const()[name = string("input_99_mode_0"), val = string("constant")];
+            fp16 const_27_to_fp16 = const()[name = string("const_27_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 134]> input_99_cast_fp16 = pad(constant_val = const_27_to_fp16, mode = input_99_mode_0, pad = input_99_pad_0, x = input_97_cast_fp16)[name = string("input_99_cast_fp16")];
+            string input_101_pad_type_0 = const()[name = string("input_101_pad_type_0"), val = string("valid")];
+            int32 input_101_groups_0 = const()[name = string("input_101_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_101_strides_0 = const()[name = string("input_101_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_101_pad_0 = const()[name = string("input_101_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_101_dilations_0 = const()[name = string("input_101_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_250_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22693056))), lut = tensor<fp16, [32, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22697728))))[name = string("const_250_to_fp16_palettized")];
+            tensor<fp16, [1024]> const_251_to_fp16 = const()[name = string("const_251_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22698816)))];
+            tensor<fp16, [1, 1024, 126]> input_103_cast_fp16 = conv(bias = const_251_to_fp16, dilations = input_101_dilations_0, groups = input_101_groups_0, pad = input_101_pad_0, pad_type = input_101_pad_type_0, strides = input_101_strides_0, weight = const_250_to_fp16_palettized, x = input_99_cast_fp16)[name = string("input_103_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_105_cast_fp16 = silu(x = input_103_cast_fp16)[name = string("input_105_cast_fp16")];
+            string x_43_pad_type_0 = const()[name = string("x_43_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_43_strides_0 = const()[name = string("x_43_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_43_pad_0 = const()[name = string("x_43_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_43_dilations_0 = const()[name = string("x_43_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_43_groups_0 = const()[name = string("x_43_groups_0"), val = int32(1)];
+            tensor<fp16, [1024, 1024, 1]> model_layers_1_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22700928))), lut = tensor<fp16, [32, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23225280))))[name = string("model_layers_1_conv_pointwise_conv2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 1024, 126]> x_43_cast_fp16 = conv(dilations = x_43_dilations_0, groups = x_43_groups_0, pad = x_43_pad_0, pad_type = x_43_pad_type_0, strides = x_43_strides_0, weight = model_layers_1_conv_pointwise_conv2_weight_to_fp16_palettized, x = input_105_cast_fp16)[name = string("x_43_cast_fp16")];
+            tensor<int32, [3]> input_107_perm_0 = const()[name = string("input_107_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 126, 1024]> input_107_cast_fp16 = transpose(perm = input_107_perm_0, x = x_43_cast_fp16)[name = string("transpose_299")];
+            tensor<fp16, [1, 126, 1024]> input_109_cast_fp16 = add(x = input_91_cast_fp16, y = input_107_cast_fp16)[name = string("input_109_cast_fp16")];
+            tensor<int32, [1]> input_111_axes_0 = const()[name = string("input_111_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_1_norm_feed_forward2_weight_to_fp16 = const()[name = string("model_layers_1_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23226368)))];
+            tensor<fp16, [1024]> model_layers_1_norm_feed_forward2_bias_to_fp16 = const()[name = string("model_layers_1_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23228480)))];
+            tensor<fp16, [1, 126, 1024]> input_111_cast_fp16 = layer_norm(axes = input_111_axes_0, beta = model_layers_1_norm_feed_forward2_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_1_norm_feed_forward2_weight_to_fp16, x = input_109_cast_fp16)[name = string("input_111_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_1_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23230592))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25327808))))[name = string("model_layers_1_feed_forward2_linear1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 4096]> linear_17_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_1_feed_forward2_linear1_weight_to_fp16_palettized, x = input_111_cast_fp16)[name = string("linear_17_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_115_cast_fp16 = silu(x = linear_17_cast_fp16)[name = string("input_115_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_1_feed_forward2_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25331968))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27429184))))[name = string("model_layers_1_feed_forward2_linear2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_18_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_1_feed_forward2_linear2_weight_to_fp16_palettized, x = input_115_cast_fp16)[name = string("linear_18_cast_fp16")];
+            fp16 var_524_to_fp16 = const()[name = string("op_524_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_525_cast_fp16 = mul(x = linear_18_cast_fp16, y = var_524_to_fp16)[name = string("op_525_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_121_cast_fp16 = add(x = input_109_cast_fp16, y = var_525_cast_fp16)[name = string("input_121_cast_fp16")];
+            tensor<int32, [1]> input_123_axes_0 = const()[name = string("input_123_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_1_norm_out_weight_to_fp16 = const()[name = string("model_layers_1_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27430272)))];
+            tensor<fp16, [1024]> model_layers_1_norm_out_bias_to_fp16 = const()[name = string("model_layers_1_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27432384)))];
+            tensor<fp16, [1, 126, 1024]> input_123_cast_fp16 = layer_norm(axes = input_123_axes_0, beta = model_layers_1_norm_out_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_1_norm_out_weight_to_fp16, x = input_121_cast_fp16)[name = string("input_123_cast_fp16")];
+            tensor<int32, [1]> input_125_axes_0 = const()[name = string("input_125_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_2_norm_feed_forward1_weight_to_fp16 = const()[name = string("model_layers_2_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27434496)))];
+            tensor<fp16, [1024]> model_layers_2_norm_feed_forward1_bias_to_fp16 = const()[name = string("model_layers_2_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27436608)))];
+            tensor<fp16, [1, 126, 1024]> input_125_cast_fp16 = layer_norm(axes = input_125_axes_0, beta = model_layers_2_norm_feed_forward1_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_2_norm_feed_forward1_weight_to_fp16, x = input_123_cast_fp16)[name = string("input_125_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_2_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27438720))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29535936))))[name = string("model_layers_2_feed_forward1_linear1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 4096]> linear_19_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_2_feed_forward1_linear1_weight_to_fp16_palettized, x = input_125_cast_fp16)[name = string("linear_19_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_129_cast_fp16 = silu(x = linear_19_cast_fp16)[name = string("input_129_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_2_feed_forward1_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29540096))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31637312))))[name = string("model_layers_2_feed_forward1_linear2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_20_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_2_feed_forward1_linear2_weight_to_fp16_palettized, x = input_129_cast_fp16)[name = string("linear_20_cast_fp16")];
+            fp16 var_553_to_fp16 = const()[name = string("op_553_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_554_cast_fp16 = mul(x = linear_20_cast_fp16, y = var_553_to_fp16)[name = string("op_554_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_135_cast_fp16 = add(x = input_123_cast_fp16, y = var_554_cast_fp16)[name = string("input_135_cast_fp16")];
+            tensor<int32, [1]> query_5_axes_0 = const()[name = string("query_5_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_2_norm_self_att_weight_to_fp16 = const()[name = string("model_layers_2_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31638400)))];
+            tensor<fp16, [1024]> model_layers_2_norm_self_att_bias_to_fp16 = const()[name = string("model_layers_2_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31640512)))];
+            tensor<fp16, [1, 126, 1024]> query_5_cast_fp16 = layer_norm(axes = query_5_axes_0, beta = model_layers_2_norm_self_att_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_2_norm_self_att_weight_to_fp16, x = input_135_cast_fp16)[name = string("query_5_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_2_self_attn_linear_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31642624))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32166976))))[name = string("model_layers_2_self_attn_linear_q_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_21_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_2_self_attn_linear_q_weight_to_fp16_palettized, x = query_5_cast_fp16)[name = string("linear_21_cast_fp16")];
+            tensor<int32, [4]> var_570 = const()[name = string("op_570"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> q_13_cast_fp16 = reshape(shape = var_570, x = linear_21_cast_fp16)[name = string("q_13_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_2_self_attn_linear_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32168064))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32692416))))[name = string("model_layers_2_self_attn_linear_k_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_22_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_2_self_attn_linear_k_weight_to_fp16_palettized, x = query_5_cast_fp16)[name = string("linear_22_cast_fp16")];
+            tensor<int32, [4]> var_574 = const()[name = string("op_574"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> k_9_cast_fp16 = reshape(shape = var_574, x = linear_22_cast_fp16)[name = string("k_9_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_2_self_attn_linear_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32693504))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33217856))))[name = string("model_layers_2_self_attn_linear_v_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_23_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_2_self_attn_linear_v_weight_to_fp16_palettized, x = query_5_cast_fp16)[name = string("linear_23_cast_fp16")];
+            tensor<int32, [4]> var_578 = const()[name = string("op_578"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> v_5_cast_fp16 = reshape(shape = var_578, x = linear_23_cast_fp16)[name = string("v_5_cast_fp16")];
+            tensor<int32, [4]> value_5_perm_0 = const()[name = string("value_5_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> model_layers_2_self_attn_pos_bias_u_to_fp16 = const()[name = string("model_layers_2_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33218944)))];
+            tensor<fp16, [1, 126, 8, 128]> var_590_cast_fp16 = add(x = q_13_cast_fp16, y = model_layers_2_self_attn_pos_bias_u_to_fp16)[name = string("op_590_cast_fp16")];
+            tensor<fp16, [8, 128]> model_layers_2_self_attn_pos_bias_v_to_fp16 = const()[name = string("model_layers_2_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33221056)))];
+            tensor<fp16, [1, 126, 8, 128]> var_592_cast_fp16 = add(x = q_13_cast_fp16, y = model_layers_2_self_attn_pos_bias_v_to_fp16)[name = string("op_592_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_5_perm_0 = const()[name = string("q_with_bias_v_5_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_51_transpose_x_0 = const()[name = string("x_51_transpose_x_0"), val = bool(false)];
+            bool x_51_transpose_y_0 = const()[name = string("x_51_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 251]> var_594_to_fp16 = const()[name = string("op_594_to_fp16"), val = tensor<fp16, [1, 8, 128, 251]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33223168)))];
+            tensor<fp16, [1, 8, 126, 128]> q_with_bias_v_5_cast_fp16 = transpose(perm = q_with_bias_v_5_perm_0, x = var_592_cast_fp16)[name = string("transpose_298")];
+            tensor<fp16, [1, 8, 126, 251]> x_51_cast_fp16 = matmul(transpose_x = x_51_transpose_x_0, transpose_y = x_51_transpose_y_0, x = q_with_bias_v_5_cast_fp16, y = var_594_to_fp16)[name = string("x_51_cast_fp16")];
+            tensor<int32, [8]> x_53_pad_0 = const()[name = string("x_53_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_53_mode_0 = const()[name = string("x_53_mode_0"), val = string("constant")];
+            fp16 const_34_to_fp16 = const()[name = string("const_34_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 252]> x_53_cast_fp16 = pad(constant_val = const_34_to_fp16, mode = x_53_mode_0, pad = x_53_pad_0, x = x_51_cast_fp16)[name = string("x_53_cast_fp16")];
+            tensor<int32, [4]> var_602 = const()[name = string("op_602"), val = tensor<int32, [4]>([1, 8, -1, 126])];
+            tensor<fp16, [1, 8, 252, 126]> x_55_cast_fp16 = reshape(shape = var_602, x = x_53_cast_fp16)[name = string("x_55_cast_fp16")];
+            tensor<int32, [4]> var_606_begin_0 = const()[name = string("op_606_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_606_end_0 = const()[name = string("op_606_end_0"), val = tensor<int32, [4]>([1, 8, 252, 126])];
+            tensor<bool, [4]> var_606_end_mask_0 = const()[name = string("op_606_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 251, 126]> var_606_cast_fp16 = slice_by_index(begin = var_606_begin_0, end = var_606_end_0, end_mask = var_606_end_mask_0, x = x_55_cast_fp16)[name = string("op_606_cast_fp16")];
+            tensor<int32, [4]> var_607 = const()[name = string("op_607"), val = tensor<int32, [4]>([1, 8, 126, 251])];
+            tensor<fp16, [1, 8, 126, 251]> matrix_bd_9_cast_fp16 = reshape(shape = var_607, x = var_606_cast_fp16)[name = string("matrix_bd_9_cast_fp16")];
+            bool matrix_ac_5_transpose_x_0 = const()[name = string("matrix_ac_5_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_5_transpose_y_0 = const()[name = string("matrix_ac_5_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_100_perm_0 = const()[name = string("transpose_100_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_101_perm_0 = const()[name = string("transpose_101_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 126]> transpose_101 = transpose(perm = transpose_101_perm_0, x = k_9_cast_fp16)[name = string("transpose_296")];
+            tensor<fp16, [1, 8, 126, 128]> transpose_100 = transpose(perm = transpose_100_perm_0, x = var_590_cast_fp16)[name = string("transpose_297")];
+            tensor<fp16, [1, 8, 126, 126]> matrix_ac_5_cast_fp16 = matmul(transpose_x = matrix_ac_5_transpose_x_0, transpose_y = matrix_ac_5_transpose_y_0, x = transpose_100, y = transpose_101)[name = string("matrix_ac_5_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_11_begin_0 = const()[name = string("matrix_bd_11_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_11_end_0 = const()[name = string("matrix_bd_11_end_0"), val = tensor<int32, [4]>([1, 8, 126, 126])];
+            tensor<bool, [4]> matrix_bd_11_end_mask_0 = const()[name = string("matrix_bd_11_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 126, 126]> matrix_bd_11_cast_fp16 = slice_by_index(begin = matrix_bd_11_begin_0, end = matrix_bd_11_end_0, end_mask = matrix_bd_11_end_mask_0, x = matrix_bd_9_cast_fp16)[name = string("matrix_bd_11_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_616_cast_fp16 = add(x = matrix_ac_5_cast_fp16, y = matrix_bd_11_cast_fp16)[name = string("op_616_cast_fp16")];
+            fp16 _inversed_scores_9_y_0_to_fp16 = const()[name = string("_inversed_scores_9_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 126, 126]> _inversed_scores_9_cast_fp16 = mul(x = var_616_cast_fp16, y = _inversed_scores_9_y_0_to_fp16)[name = string("_inversed_scores_9_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> scores_11_cast_fp16 = select(a = var_7_to_fp16, b = _inversed_scores_9_cast_fp16, cond = mask_3)[name = string("scores_11_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_622_cast_fp16 = softmax(axis = var_25, x = scores_11_cast_fp16)[name = string("op_622_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> input_137_cast_fp16 = select(a = var_6_to_fp16, b = var_622_cast_fp16, cond = mask_3)[name = string("input_137_cast_fp16")];
+            bool x_57_transpose_x_0 = const()[name = string("x_57_transpose_x_0"), val = bool(false)];
+            bool x_57_transpose_y_0 = const()[name = string("x_57_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 126, 128]> value_5_cast_fp16 = transpose(perm = value_5_perm_0, x = v_5_cast_fp16)[name = string("transpose_295")];
+            tensor<fp16, [1, 8, 126, 128]> x_57_cast_fp16 = matmul(transpose_x = x_57_transpose_x_0, transpose_y = x_57_transpose_y_0, x = input_137_cast_fp16, y = value_5_cast_fp16)[name = string("x_57_cast_fp16")];
+            tensor<int32, [4]> var_626_perm_0 = const()[name = string("op_626_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_627 = const()[name = string("op_627"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 126, 8, 128]> var_626_cast_fp16 = transpose(perm = var_626_perm_0, x = x_57_cast_fp16)[name = string("transpose_294")];
+            tensor<fp16, [1, 126, 1024]> input_139_cast_fp16 = reshape(shape = var_627, x = var_626_cast_fp16)[name = string("input_139_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_2_self_attn_linear_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33737280))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34261632))))[name = string("model_layers_2_self_attn_linear_out_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_25_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_2_self_attn_linear_out_weight_to_fp16_palettized, x = input_139_cast_fp16)[name = string("linear_25_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_143_cast_fp16 = add(x = input_135_cast_fp16, y = linear_25_cast_fp16)[name = string("input_143_cast_fp16")];
+            tensor<int32, [1]> x_61_axes_0 = const()[name = string("x_61_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_2_norm_conv_weight_to_fp16 = const()[name = string("model_layers_2_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34262720)))];
+            tensor<fp16, [1024]> model_layers_2_norm_conv_bias_to_fp16 = const()[name = string("model_layers_2_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34264832)))];
+            tensor<fp16, [1, 126, 1024]> x_61_cast_fp16 = layer_norm(axes = x_61_axes_0, beta = model_layers_2_norm_conv_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_2_norm_conv_weight_to_fp16, x = input_143_cast_fp16)[name = string("x_61_cast_fp16")];
+            tensor<int32, [3]> input_145_perm_0 = const()[name = string("input_145_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_147_pad_type_0 = const()[name = string("input_147_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_147_strides_0 = const()[name = string("input_147_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_147_pad_0 = const()[name = string("input_147_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_147_dilations_0 = const()[name = string("input_147_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_147_groups_0 = const()[name = string("input_147_groups_0"), val = int32(1)];
+            tensor<fp16, [2048, 1024, 1]> model_layers_2_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34266944))), lut = tensor<fp16, [64, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35315584))))[name = string("model_layers_2_conv_pointwise_conv1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 1024, 126]> input_145_cast_fp16 = transpose(perm = input_145_perm_0, x = x_61_cast_fp16)[name = string("transpose_293")];
+            tensor<fp16, [1, 2048, 126]> input_147_cast_fp16 = conv(dilations = input_147_dilations_0, groups = input_147_groups_0, pad = input_147_pad_0, pad_type = input_147_pad_type_0, strides = input_147_strides_0, weight = model_layers_2_conv_pointwise_conv1_weight_to_fp16_palettized, x = input_145_cast_fp16)[name = string("input_147_cast_fp16")];
+            int32 x_63_split_num_splits_0 = const()[name = string("x_63_split_num_splits_0"), val = int32(2)];
+            int32 x_63_split_axis_0 = const()[name = string("x_63_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 126]> x_63_split_cast_fp16_0, tensor<fp16, [1, 1024, 126]> x_63_split_cast_fp16_1 = split(axis = x_63_split_axis_0, num_splits = x_63_split_num_splits_0, x = input_147_cast_fp16)[name = string("x_63_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_63_split_1_sigmoid_cast_fp16 = sigmoid(x = x_63_split_cast_fp16_1)[name = string("x_63_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_63_cast_fp16 = mul(x = x_63_split_cast_fp16_0, y = x_63_split_1_sigmoid_cast_fp16)[name = string("x_63_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_149_cast_fp16 = select(a = var_6_to_fp16, b = x_63_cast_fp16, cond = var_323)[name = string("input_149_cast_fp16")];
+            tensor<int32, [6]> input_151_pad_0 = const()[name = string("input_151_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_151_mode_0 = const()[name = string("input_151_mode_0"), val = string("constant")];
+            fp16 const_37_to_fp16 = const()[name = string("const_37_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 134]> input_151_cast_fp16 = pad(constant_val = const_37_to_fp16, mode = input_151_mode_0, pad = input_151_pad_0, x = input_149_cast_fp16)[name = string("input_151_cast_fp16")];
+            string input_153_pad_type_0 = const()[name = string("input_153_pad_type_0"), val = string("valid")];
+            int32 input_153_groups_0 = const()[name = string("input_153_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_153_strides_0 = const()[name = string("input_153_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_153_pad_0 = const()[name = string("input_153_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_153_dilations_0 = const()[name = string("input_153_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_252_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35317696))), lut = tensor<fp16, [32, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35322368))))[name = string("const_252_to_fp16_palettized")];
+            tensor<fp16, [1024]> const_253_to_fp16 = const()[name = string("const_253_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35323456)))];
+            tensor<fp16, [1, 1024, 126]> input_155_cast_fp16 = conv(bias = const_253_to_fp16, dilations = input_153_dilations_0, groups = input_153_groups_0, pad = input_153_pad_0, pad_type = input_153_pad_type_0, strides = input_153_strides_0, weight = const_252_to_fp16_palettized, x = input_151_cast_fp16)[name = string("input_155_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_157_cast_fp16 = silu(x = input_155_cast_fp16)[name = string("input_157_cast_fp16")];
+            string x_65_pad_type_0 = const()[name = string("x_65_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_65_strides_0 = const()[name = string("x_65_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_65_pad_0 = const()[name = string("x_65_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_65_dilations_0 = const()[name = string("x_65_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_65_groups_0 = const()[name = string("x_65_groups_0"), val = int32(1)];
+            tensor<fp16, [1024, 1024, 1]> model_layers_2_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35325568))), lut = tensor<fp16, [32, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35849920))))[name = string("model_layers_2_conv_pointwise_conv2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 1024, 126]> x_65_cast_fp16 = conv(dilations = x_65_dilations_0, groups = x_65_groups_0, pad = x_65_pad_0, pad_type = x_65_pad_type_0, strides = x_65_strides_0, weight = model_layers_2_conv_pointwise_conv2_weight_to_fp16_palettized, x = input_157_cast_fp16)[name = string("x_65_cast_fp16")];
+            tensor<int32, [3]> input_159_perm_0 = const()[name = string("input_159_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 126, 1024]> input_159_cast_fp16 = transpose(perm = input_159_perm_0, x = x_65_cast_fp16)[name = string("transpose_292")];
+            tensor<fp16, [1, 126, 1024]> input_161_cast_fp16 = add(x = input_143_cast_fp16, y = input_159_cast_fp16)[name = string("input_161_cast_fp16")];
+            tensor<int32, [1]> input_163_axes_0 = const()[name = string("input_163_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_2_norm_feed_forward2_weight_to_fp16 = const()[name = string("model_layers_2_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35851008)))];
+            tensor<fp16, [1024]> model_layers_2_norm_feed_forward2_bias_to_fp16 = const()[name = string("model_layers_2_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35853120)))];
+            tensor<fp16, [1, 126, 1024]> input_163_cast_fp16 = layer_norm(axes = input_163_axes_0, beta = model_layers_2_norm_feed_forward2_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_2_norm_feed_forward2_weight_to_fp16, x = input_161_cast_fp16)[name = string("input_163_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_2_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35855232))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37952448))))[name = string("model_layers_2_feed_forward2_linear1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 4096]> linear_26_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_2_feed_forward2_linear1_weight_to_fp16_palettized, x = input_163_cast_fp16)[name = string("linear_26_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_167_cast_fp16 = silu(x = linear_26_cast_fp16)[name = string("input_167_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_2_feed_forward2_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37956608))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40053824))))[name = string("model_layers_2_feed_forward2_linear2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_27_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_2_feed_forward2_linear2_weight_to_fp16_palettized, x = input_167_cast_fp16)[name = string("linear_27_cast_fp16")];
+            fp16 var_687_to_fp16 = const()[name = string("op_687_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_688_cast_fp16 = mul(x = linear_27_cast_fp16, y = var_687_to_fp16)[name = string("op_688_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_173_cast_fp16 = add(x = input_161_cast_fp16, y = var_688_cast_fp16)[name = string("input_173_cast_fp16")];
+            tensor<int32, [1]> input_175_axes_0 = const()[name = string("input_175_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_2_norm_out_weight_to_fp16 = const()[name = string("model_layers_2_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40054912)))];
+            tensor<fp16, [1024]> model_layers_2_norm_out_bias_to_fp16 = const()[name = string("model_layers_2_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40057024)))];
+            tensor<fp16, [1, 126, 1024]> input_175_cast_fp16 = layer_norm(axes = input_175_axes_0, beta = model_layers_2_norm_out_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_2_norm_out_weight_to_fp16, x = input_173_cast_fp16)[name = string("input_175_cast_fp16")];
+            tensor<int32, [1]> input_177_axes_0 = const()[name = string("input_177_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_3_norm_feed_forward1_weight_to_fp16 = const()[name = string("model_layers_3_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40059136)))];
+            tensor<fp16, [1024]> model_layers_3_norm_feed_forward1_bias_to_fp16 = const()[name = string("model_layers_3_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40061248)))];
+            tensor<fp16, [1, 126, 1024]> input_177_cast_fp16 = layer_norm(axes = input_177_axes_0, beta = model_layers_3_norm_feed_forward1_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_3_norm_feed_forward1_weight_to_fp16, x = input_175_cast_fp16)[name = string("input_177_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_3_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40063360))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42160576))))[name = string("model_layers_3_feed_forward1_linear1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 4096]> linear_28_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_3_feed_forward1_linear1_weight_to_fp16_palettized, x = input_177_cast_fp16)[name = string("linear_28_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_181_cast_fp16 = silu(x = linear_28_cast_fp16)[name = string("input_181_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_3_feed_forward1_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42164736))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44261952))))[name = string("model_layers_3_feed_forward1_linear2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_29_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_3_feed_forward1_linear2_weight_to_fp16_palettized, x = input_181_cast_fp16)[name = string("linear_29_cast_fp16")];
+            fp16 var_716_to_fp16 = const()[name = string("op_716_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_717_cast_fp16 = mul(x = linear_29_cast_fp16, y = var_716_to_fp16)[name = string("op_717_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_187_cast_fp16 = add(x = input_175_cast_fp16, y = var_717_cast_fp16)[name = string("input_187_cast_fp16")];
+            tensor<int32, [1]> query_7_axes_0 = const()[name = string("query_7_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_3_norm_self_att_weight_to_fp16 = const()[name = string("model_layers_3_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44263040)))];
+            tensor<fp16, [1024]> model_layers_3_norm_self_att_bias_to_fp16 = const()[name = string("model_layers_3_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44265152)))];
+            tensor<fp16, [1, 126, 1024]> query_7_cast_fp16 = layer_norm(axes = query_7_axes_0, beta = model_layers_3_norm_self_att_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_3_norm_self_att_weight_to_fp16, x = input_187_cast_fp16)[name = string("query_7_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_3_self_attn_linear_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44267264))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44791616))))[name = string("model_layers_3_self_attn_linear_q_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_30_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_3_self_attn_linear_q_weight_to_fp16_palettized, x = query_7_cast_fp16)[name = string("linear_30_cast_fp16")];
+            tensor<int32, [4]> var_733 = const()[name = string("op_733"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> q_19_cast_fp16 = reshape(shape = var_733, x = linear_30_cast_fp16)[name = string("q_19_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_3_self_attn_linear_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44792704))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45317056))))[name = string("model_layers_3_self_attn_linear_k_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_31_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_3_self_attn_linear_k_weight_to_fp16_palettized, x = query_7_cast_fp16)[name = string("linear_31_cast_fp16")];
+            tensor<int32, [4]> var_737 = const()[name = string("op_737"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> k_13_cast_fp16 = reshape(shape = var_737, x = linear_31_cast_fp16)[name = string("k_13_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_3_self_attn_linear_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45318144))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45842496))))[name = string("model_layers_3_self_attn_linear_v_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_32_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_3_self_attn_linear_v_weight_to_fp16_palettized, x = query_7_cast_fp16)[name = string("linear_32_cast_fp16")];
+            tensor<int32, [4]> var_741 = const()[name = string("op_741"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> v_7_cast_fp16 = reshape(shape = var_741, x = linear_32_cast_fp16)[name = string("v_7_cast_fp16")];
+            tensor<int32, [4]> value_7_perm_0 = const()[name = string("value_7_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> model_layers_3_self_attn_pos_bias_u_to_fp16 = const()[name = string("model_layers_3_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45843584)))];
+            tensor<fp16, [1, 126, 8, 128]> var_753_cast_fp16 = add(x = q_19_cast_fp16, y = model_layers_3_self_attn_pos_bias_u_to_fp16)[name = string("op_753_cast_fp16")];
+            tensor<fp16, [8, 128]> model_layers_3_self_attn_pos_bias_v_to_fp16 = const()[name = string("model_layers_3_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45845696)))];
+            tensor<fp16, [1, 126, 8, 128]> var_755_cast_fp16 = add(x = q_19_cast_fp16, y = model_layers_3_self_attn_pos_bias_v_to_fp16)[name = string("op_755_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_7_perm_0 = const()[name = string("q_with_bias_v_7_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_73_transpose_x_0 = const()[name = string("x_73_transpose_x_0"), val = bool(false)];
+            bool x_73_transpose_y_0 = const()[name = string("x_73_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 251]> var_757_to_fp16 = const()[name = string("op_757_to_fp16"), val = tensor<fp16, [1, 8, 128, 251]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45847808)))];
+            tensor<fp16, [1, 8, 126, 128]> q_with_bias_v_7_cast_fp16 = transpose(perm = q_with_bias_v_7_perm_0, x = var_755_cast_fp16)[name = string("transpose_291")];
+            tensor<fp16, [1, 8, 126, 251]> x_73_cast_fp16 = matmul(transpose_x = x_73_transpose_x_0, transpose_y = x_73_transpose_y_0, x = q_with_bias_v_7_cast_fp16, y = var_757_to_fp16)[name = string("x_73_cast_fp16")];
+            tensor<int32, [8]> x_75_pad_0 = const()[name = string("x_75_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_75_mode_0 = const()[name = string("x_75_mode_0"), val = string("constant")];
+            fp16 const_44_to_fp16 = const()[name = string("const_44_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 252]> x_75_cast_fp16 = pad(constant_val = const_44_to_fp16, mode = x_75_mode_0, pad = x_75_pad_0, x = x_73_cast_fp16)[name = string("x_75_cast_fp16")];
+            tensor<int32, [4]> var_765 = const()[name = string("op_765"), val = tensor<int32, [4]>([1, 8, -1, 126])];
+            tensor<fp16, [1, 8, 252, 126]> x_77_cast_fp16 = reshape(shape = var_765, x = x_75_cast_fp16)[name = string("x_77_cast_fp16")];
+            tensor<int32, [4]> var_769_begin_0 = const()[name = string("op_769_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_769_end_0 = const()[name = string("op_769_end_0"), val = tensor<int32, [4]>([1, 8, 252, 126])];
+            tensor<bool, [4]> var_769_end_mask_0 = const()[name = string("op_769_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 251, 126]> var_769_cast_fp16 = slice_by_index(begin = var_769_begin_0, end = var_769_end_0, end_mask = var_769_end_mask_0, x = x_77_cast_fp16)[name = string("op_769_cast_fp16")];
+            tensor<int32, [4]> var_770 = const()[name = string("op_770"), val = tensor<int32, [4]>([1, 8, 126, 251])];
+            tensor<fp16, [1, 8, 126, 251]> matrix_bd_13_cast_fp16 = reshape(shape = var_770, x = var_769_cast_fp16)[name = string("matrix_bd_13_cast_fp16")];
+            bool matrix_ac_7_transpose_x_0 = const()[name = string("matrix_ac_7_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_7_transpose_y_0 = const()[name = string("matrix_ac_7_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_102_perm_0 = const()[name = string("transpose_102_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_103_perm_0 = const()[name = string("transpose_103_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 126]> transpose_103 = transpose(perm = transpose_103_perm_0, x = k_13_cast_fp16)[name = string("transpose_289")];
+            tensor<fp16, [1, 8, 126, 128]> transpose_102 = transpose(perm = transpose_102_perm_0, x = var_753_cast_fp16)[name = string("transpose_290")];
+            tensor<fp16, [1, 8, 126, 126]> matrix_ac_7_cast_fp16 = matmul(transpose_x = matrix_ac_7_transpose_x_0, transpose_y = matrix_ac_7_transpose_y_0, x = transpose_102, y = transpose_103)[name = string("matrix_ac_7_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_15_begin_0 = const()[name = string("matrix_bd_15_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_15_end_0 = const()[name = string("matrix_bd_15_end_0"), val = tensor<int32, [4]>([1, 8, 126, 126])];
+            tensor<bool, [4]> matrix_bd_15_end_mask_0 = const()[name = string("matrix_bd_15_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 126, 126]> matrix_bd_15_cast_fp16 = slice_by_index(begin = matrix_bd_15_begin_0, end = matrix_bd_15_end_0, end_mask = matrix_bd_15_end_mask_0, x = matrix_bd_13_cast_fp16)[name = string("matrix_bd_15_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_779_cast_fp16 = add(x = matrix_ac_7_cast_fp16, y = matrix_bd_15_cast_fp16)[name = string("op_779_cast_fp16")];
+            fp16 _inversed_scores_13_y_0_to_fp16 = const()[name = string("_inversed_scores_13_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 126, 126]> _inversed_scores_13_cast_fp16 = mul(x = var_779_cast_fp16, y = _inversed_scores_13_y_0_to_fp16)[name = string("_inversed_scores_13_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> scores_15_cast_fp16 = select(a = var_7_to_fp16, b = _inversed_scores_13_cast_fp16, cond = mask_3)[name = string("scores_15_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_785_cast_fp16 = softmax(axis = var_25, x = scores_15_cast_fp16)[name = string("op_785_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> input_189_cast_fp16 = select(a = var_6_to_fp16, b = var_785_cast_fp16, cond = mask_3)[name = string("input_189_cast_fp16")];
+            bool x_79_transpose_x_0 = const()[name = string("x_79_transpose_x_0"), val = bool(false)];
+            bool x_79_transpose_y_0 = const()[name = string("x_79_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 126, 128]> value_7_cast_fp16 = transpose(perm = value_7_perm_0, x = v_7_cast_fp16)[name = string("transpose_288")];
+            tensor<fp16, [1, 8, 126, 128]> x_79_cast_fp16 = matmul(transpose_x = x_79_transpose_x_0, transpose_y = x_79_transpose_y_0, x = input_189_cast_fp16, y = value_7_cast_fp16)[name = string("x_79_cast_fp16")];
+            tensor<int32, [4]> var_789_perm_0 = const()[name = string("op_789_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_790 = const()[name = string("op_790"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 126, 8, 128]> var_789_cast_fp16 = transpose(perm = var_789_perm_0, x = x_79_cast_fp16)[name = string("transpose_287")];
+            tensor<fp16, [1, 126, 1024]> input_191_cast_fp16 = reshape(shape = var_790, x = var_789_cast_fp16)[name = string("input_191_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_3_self_attn_linear_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46361920))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46886272))))[name = string("model_layers_3_self_attn_linear_out_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_34_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_3_self_attn_linear_out_weight_to_fp16_palettized, x = input_191_cast_fp16)[name = string("linear_34_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_195_cast_fp16 = add(x = input_187_cast_fp16, y = linear_34_cast_fp16)[name = string("input_195_cast_fp16")];
+            tensor<int32, [1]> x_83_axes_0 = const()[name = string("x_83_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_3_norm_conv_weight_to_fp16 = const()[name = string("model_layers_3_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46887360)))];
+            tensor<fp16, [1024]> model_layers_3_norm_conv_bias_to_fp16 = const()[name = string("model_layers_3_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46889472)))];
+            tensor<fp16, [1, 126, 1024]> x_83_cast_fp16 = layer_norm(axes = x_83_axes_0, beta = model_layers_3_norm_conv_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_3_norm_conv_weight_to_fp16, x = input_195_cast_fp16)[name = string("x_83_cast_fp16")];
+            tensor<int32, [3]> input_197_perm_0 = const()[name = string("input_197_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_199_pad_type_0 = const()[name = string("input_199_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_199_strides_0 = const()[name = string("input_199_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_199_pad_0 = const()[name = string("input_199_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_199_dilations_0 = const()[name = string("input_199_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_199_groups_0 = const()[name = string("input_199_groups_0"), val = int32(1)];
+            tensor<fp16, [2048, 1024, 1]> model_layers_3_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46891584))), lut = tensor<fp16, [64, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47940224))))[name = string("model_layers_3_conv_pointwise_conv1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 1024, 126]> input_197_cast_fp16 = transpose(perm = input_197_perm_0, x = x_83_cast_fp16)[name = string("transpose_286")];
+            tensor<fp16, [1, 2048, 126]> input_199_cast_fp16 = conv(dilations = input_199_dilations_0, groups = input_199_groups_0, pad = input_199_pad_0, pad_type = input_199_pad_type_0, strides = input_199_strides_0, weight = model_layers_3_conv_pointwise_conv1_weight_to_fp16_palettized, x = input_197_cast_fp16)[name = string("input_199_cast_fp16")];
+            int32 x_85_split_num_splits_0 = const()[name = string("x_85_split_num_splits_0"), val = int32(2)];
+            int32 x_85_split_axis_0 = const()[name = string("x_85_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 126]> x_85_split_cast_fp16_0, tensor<fp16, [1, 1024, 126]> x_85_split_cast_fp16_1 = split(axis = x_85_split_axis_0, num_splits = x_85_split_num_splits_0, x = input_199_cast_fp16)[name = string("x_85_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_85_split_1_sigmoid_cast_fp16 = sigmoid(x = x_85_split_cast_fp16_1)[name = string("x_85_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_85_cast_fp16 = mul(x = x_85_split_cast_fp16_0, y = x_85_split_1_sigmoid_cast_fp16)[name = string("x_85_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_201_cast_fp16 = select(a = var_6_to_fp16, b = x_85_cast_fp16, cond = var_323)[name = string("input_201_cast_fp16")];
+            tensor<int32, [6]> input_203_pad_0 = const()[name = string("input_203_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_203_mode_0 = const()[name = string("input_203_mode_0"), val = string("constant")];
+            fp16 const_47_to_fp16 = const()[name = string("const_47_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 134]> input_203_cast_fp16 = pad(constant_val = const_47_to_fp16, mode = input_203_mode_0, pad = input_203_pad_0, x = input_201_cast_fp16)[name = string("input_203_cast_fp16")];
+            string input_205_pad_type_0 = const()[name = string("input_205_pad_type_0"), val = string("valid")];
+            int32 input_205_groups_0 = const()[name = string("input_205_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_205_strides_0 = const()[name = string("input_205_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_205_pad_0 = const()[name = string("input_205_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_205_dilations_0 = const()[name = string("input_205_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_254_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47942336))), lut = tensor<fp16, [32, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47947008))))[name = string("const_254_to_fp16_palettized")];
+            tensor<fp16, [1024]> const_255_to_fp16 = const()[name = string("const_255_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47948096)))];
+            tensor<fp16, [1, 1024, 126]> input_207_cast_fp16 = conv(bias = const_255_to_fp16, dilations = input_205_dilations_0, groups = input_205_groups_0, pad = input_205_pad_0, pad_type = input_205_pad_type_0, strides = input_205_strides_0, weight = const_254_to_fp16_palettized, x = input_203_cast_fp16)[name = string("input_207_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_209_cast_fp16 = silu(x = input_207_cast_fp16)[name = string("input_209_cast_fp16")];
+            string x_87_pad_type_0 = const()[name = string("x_87_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_87_strides_0 = const()[name = string("x_87_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_87_pad_0 = const()[name = string("x_87_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_87_dilations_0 = const()[name = string("x_87_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_87_groups_0 = const()[name = string("x_87_groups_0"), val = int32(1)];
+            tensor<fp16, [1024, 1024, 1]> model_layers_3_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47950208))), lut = tensor<fp16, [32, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48474560))))[name = string("model_layers_3_conv_pointwise_conv2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 1024, 126]> x_87_cast_fp16 = conv(dilations = x_87_dilations_0, groups = x_87_groups_0, pad = x_87_pad_0, pad_type = x_87_pad_type_0, strides = x_87_strides_0, weight = model_layers_3_conv_pointwise_conv2_weight_to_fp16_palettized, x = input_209_cast_fp16)[name = string("x_87_cast_fp16")];
+            tensor<int32, [3]> input_211_perm_0 = const()[name = string("input_211_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 126, 1024]> input_211_cast_fp16 = transpose(perm = input_211_perm_0, x = x_87_cast_fp16)[name = string("transpose_285")];
+            tensor<fp16, [1, 126, 1024]> input_213_cast_fp16 = add(x = input_195_cast_fp16, y = input_211_cast_fp16)[name = string("input_213_cast_fp16")];
+            tensor<int32, [1]> input_215_axes_0 = const()[name = string("input_215_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_3_norm_feed_forward2_weight_to_fp16 = const()[name = string("model_layers_3_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48475648)))];
+            tensor<fp16, [1024]> model_layers_3_norm_feed_forward2_bias_to_fp16 = const()[name = string("model_layers_3_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48477760)))];
+            tensor<fp16, [1, 126, 1024]> input_215_cast_fp16 = layer_norm(axes = input_215_axes_0, beta = model_layers_3_norm_feed_forward2_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_3_norm_feed_forward2_weight_to_fp16, x = input_213_cast_fp16)[name = string("input_215_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_3_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48479872))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50577088))))[name = string("model_layers_3_feed_forward2_linear1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 4096]> linear_35_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_3_feed_forward2_linear1_weight_to_fp16_palettized, x = input_215_cast_fp16)[name = string("linear_35_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_219_cast_fp16 = silu(x = linear_35_cast_fp16)[name = string("input_219_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_3_feed_forward2_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50581248))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52678464))))[name = string("model_layers_3_feed_forward2_linear2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_36_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_3_feed_forward2_linear2_weight_to_fp16_palettized, x = input_219_cast_fp16)[name = string("linear_36_cast_fp16")];
+            fp16 var_850_to_fp16 = const()[name = string("op_850_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_851_cast_fp16 = mul(x = linear_36_cast_fp16, y = var_850_to_fp16)[name = string("op_851_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_225_cast_fp16 = add(x = input_213_cast_fp16, y = var_851_cast_fp16)[name = string("input_225_cast_fp16")];
+            tensor<int32, [1]> input_227_axes_0 = const()[name = string("input_227_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_3_norm_out_weight_to_fp16 = const()[name = string("model_layers_3_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52679552)))];
+            tensor<fp16, [1024]> model_layers_3_norm_out_bias_to_fp16 = const()[name = string("model_layers_3_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52681664)))];
+            tensor<fp16, [1, 126, 1024]> input_227_cast_fp16 = layer_norm(axes = input_227_axes_0, beta = model_layers_3_norm_out_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_3_norm_out_weight_to_fp16, x = input_225_cast_fp16)[name = string("input_227_cast_fp16")];
+            tensor<int32, [1]> input_229_axes_0 = const()[name = string("input_229_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_4_norm_feed_forward1_weight_to_fp16 = const()[name = string("model_layers_4_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52683776)))];
+            tensor<fp16, [1024]> model_layers_4_norm_feed_forward1_bias_to_fp16 = const()[name = string("model_layers_4_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52685888)))];
+            tensor<fp16, [1, 126, 1024]> input_229_cast_fp16 = layer_norm(axes = input_229_axes_0, beta = model_layers_4_norm_feed_forward1_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_4_norm_feed_forward1_weight_to_fp16, x = input_227_cast_fp16)[name = string("input_229_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_4_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52688000))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54785216))))[name = string("model_layers_4_feed_forward1_linear1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 4096]> linear_37_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_4_feed_forward1_linear1_weight_to_fp16_palettized, x = input_229_cast_fp16)[name = string("linear_37_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_233_cast_fp16 = silu(x = linear_37_cast_fp16)[name = string("input_233_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_4_feed_forward1_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54789376))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56886592))))[name = string("model_layers_4_feed_forward1_linear2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_38_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_4_feed_forward1_linear2_weight_to_fp16_palettized, x = input_233_cast_fp16)[name = string("linear_38_cast_fp16")];
+            fp16 var_879_to_fp16 = const()[name = string("op_879_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_880_cast_fp16 = mul(x = linear_38_cast_fp16, y = var_879_to_fp16)[name = string("op_880_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_239_cast_fp16 = add(x = input_227_cast_fp16, y = var_880_cast_fp16)[name = string("input_239_cast_fp16")];
+            tensor<int32, [1]> query_9_axes_0 = const()[name = string("query_9_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_4_norm_self_att_weight_to_fp16 = const()[name = string("model_layers_4_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56887680)))];
+            tensor<fp16, [1024]> model_layers_4_norm_self_att_bias_to_fp16 = const()[name = string("model_layers_4_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56889792)))];
+            tensor<fp16, [1, 126, 1024]> query_9_cast_fp16 = layer_norm(axes = query_9_axes_0, beta = model_layers_4_norm_self_att_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_4_norm_self_att_weight_to_fp16, x = input_239_cast_fp16)[name = string("query_9_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_4_self_attn_linear_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56891904))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57416256))))[name = string("model_layers_4_self_attn_linear_q_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_39_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_4_self_attn_linear_q_weight_to_fp16_palettized, x = query_9_cast_fp16)[name = string("linear_39_cast_fp16")];
+            tensor<int32, [4]> var_896 = const()[name = string("op_896"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> q_25_cast_fp16 = reshape(shape = var_896, x = linear_39_cast_fp16)[name = string("q_25_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_4_self_attn_linear_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57417344))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57941696))))[name = string("model_layers_4_self_attn_linear_k_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_40_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_4_self_attn_linear_k_weight_to_fp16_palettized, x = query_9_cast_fp16)[name = string("linear_40_cast_fp16")];
+            tensor<int32, [4]> var_900 = const()[name = string("op_900"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> k_17_cast_fp16 = reshape(shape = var_900, x = linear_40_cast_fp16)[name = string("k_17_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_4_self_attn_linear_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57942784))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58467136))))[name = string("model_layers_4_self_attn_linear_v_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_41_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_4_self_attn_linear_v_weight_to_fp16_palettized, x = query_9_cast_fp16)[name = string("linear_41_cast_fp16")];
+            tensor<int32, [4]> var_904 = const()[name = string("op_904"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> v_9_cast_fp16 = reshape(shape = var_904, x = linear_41_cast_fp16)[name = string("v_9_cast_fp16")];
+            tensor<int32, [4]> value_9_perm_0 = const()[name = string("value_9_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> model_layers_4_self_attn_pos_bias_u_to_fp16 = const()[name = string("model_layers_4_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58468224)))];
+            tensor<fp16, [1, 126, 8, 128]> var_916_cast_fp16 = add(x = q_25_cast_fp16, y = model_layers_4_self_attn_pos_bias_u_to_fp16)[name = string("op_916_cast_fp16")];
+            tensor<fp16, [8, 128]> model_layers_4_self_attn_pos_bias_v_to_fp16 = const()[name = string("model_layers_4_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58470336)))];
+            tensor<fp16, [1, 126, 8, 128]> var_918_cast_fp16 = add(x = q_25_cast_fp16, y = model_layers_4_self_attn_pos_bias_v_to_fp16)[name = string("op_918_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_9_perm_0 = const()[name = string("q_with_bias_v_9_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_95_transpose_x_0 = const()[name = string("x_95_transpose_x_0"), val = bool(false)];
+            bool x_95_transpose_y_0 = const()[name = string("x_95_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 251]> var_920_to_fp16 = const()[name = string("op_920_to_fp16"), val = tensor<fp16, [1, 8, 128, 251]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58472448)))];
+            tensor<fp16, [1, 8, 126, 128]> q_with_bias_v_9_cast_fp16 = transpose(perm = q_with_bias_v_9_perm_0, x = var_918_cast_fp16)[name = string("transpose_284")];
+            tensor<fp16, [1, 8, 126, 251]> x_95_cast_fp16 = matmul(transpose_x = x_95_transpose_x_0, transpose_y = x_95_transpose_y_0, x = q_with_bias_v_9_cast_fp16, y = var_920_to_fp16)[name = string("x_95_cast_fp16")];
+            tensor<int32, [8]> x_97_pad_0 = const()[name = string("x_97_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_97_mode_0 = const()[name = string("x_97_mode_0"), val = string("constant")];
+            fp16 const_54_to_fp16 = const()[name = string("const_54_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 252]> x_97_cast_fp16 = pad(constant_val = const_54_to_fp16, mode = x_97_mode_0, pad = x_97_pad_0, x = x_95_cast_fp16)[name = string("x_97_cast_fp16")];
+            tensor<int32, [4]> var_928 = const()[name = string("op_928"), val = tensor<int32, [4]>([1, 8, -1, 126])];
+            tensor<fp16, [1, 8, 252, 126]> x_99_cast_fp16 = reshape(shape = var_928, x = x_97_cast_fp16)[name = string("x_99_cast_fp16")];
+            tensor<int32, [4]> var_932_begin_0 = const()[name = string("op_932_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_932_end_0 = const()[name = string("op_932_end_0"), val = tensor<int32, [4]>([1, 8, 252, 126])];
+            tensor<bool, [4]> var_932_end_mask_0 = const()[name = string("op_932_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 251, 126]> var_932_cast_fp16 = slice_by_index(begin = var_932_begin_0, end = var_932_end_0, end_mask = var_932_end_mask_0, x = x_99_cast_fp16)[name = string("op_932_cast_fp16")];
+            tensor<int32, [4]> var_933 = const()[name = string("op_933"), val = tensor<int32, [4]>([1, 8, 126, 251])];
+            tensor<fp16, [1, 8, 126, 251]> matrix_bd_17_cast_fp16 = reshape(shape = var_933, x = var_932_cast_fp16)[name = string("matrix_bd_17_cast_fp16")];
+            bool matrix_ac_9_transpose_x_0 = const()[name = string("matrix_ac_9_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_9_transpose_y_0 = const()[name = string("matrix_ac_9_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_104_perm_0 = const()[name = string("transpose_104_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_105_perm_0 = const()[name = string("transpose_105_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 126]> transpose_105 = transpose(perm = transpose_105_perm_0, x = k_17_cast_fp16)[name = string("transpose_282")];
+            tensor<fp16, [1, 8, 126, 128]> transpose_104 = transpose(perm = transpose_104_perm_0, x = var_916_cast_fp16)[name = string("transpose_283")];
+            tensor<fp16, [1, 8, 126, 126]> matrix_ac_9_cast_fp16 = matmul(transpose_x = matrix_ac_9_transpose_x_0, transpose_y = matrix_ac_9_transpose_y_0, x = transpose_104, y = transpose_105)[name = string("matrix_ac_9_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_19_begin_0 = const()[name = string("matrix_bd_19_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_19_end_0 = const()[name = string("matrix_bd_19_end_0"), val = tensor<int32, [4]>([1, 8, 126, 126])];
+            tensor<bool, [4]> matrix_bd_19_end_mask_0 = const()[name = string("matrix_bd_19_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 126, 126]> matrix_bd_19_cast_fp16 = slice_by_index(begin = matrix_bd_19_begin_0, end = matrix_bd_19_end_0, end_mask = matrix_bd_19_end_mask_0, x = matrix_bd_17_cast_fp16)[name = string("matrix_bd_19_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_942_cast_fp16 = add(x = matrix_ac_9_cast_fp16, y = matrix_bd_19_cast_fp16)[name = string("op_942_cast_fp16")];
+            fp16 _inversed_scores_17_y_0_to_fp16 = const()[name = string("_inversed_scores_17_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 126, 126]> _inversed_scores_17_cast_fp16 = mul(x = var_942_cast_fp16, y = _inversed_scores_17_y_0_to_fp16)[name = string("_inversed_scores_17_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> scores_19_cast_fp16 = select(a = var_7_to_fp16, b = _inversed_scores_17_cast_fp16, cond = mask_3)[name = string("scores_19_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_948_cast_fp16 = softmax(axis = var_25, x = scores_19_cast_fp16)[name = string("op_948_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> input_241_cast_fp16 = select(a = var_6_to_fp16, b = var_948_cast_fp16, cond = mask_3)[name = string("input_241_cast_fp16")];
+            bool x_101_transpose_x_0 = const()[name = string("x_101_transpose_x_0"), val = bool(false)];
+            bool x_101_transpose_y_0 = const()[name = string("x_101_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 126, 128]> value_9_cast_fp16 = transpose(perm = value_9_perm_0, x = v_9_cast_fp16)[name = string("transpose_281")];
+            tensor<fp16, [1, 8, 126, 128]> x_101_cast_fp16 = matmul(transpose_x = x_101_transpose_x_0, transpose_y = x_101_transpose_y_0, x = input_241_cast_fp16, y = value_9_cast_fp16)[name = string("x_101_cast_fp16")];
+            tensor<int32, [4]> var_952_perm_0 = const()[name = string("op_952_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_953 = const()[name = string("op_953"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 126, 8, 128]> var_952_cast_fp16 = transpose(perm = var_952_perm_0, x = x_101_cast_fp16)[name = string("transpose_280")];
+            tensor<fp16, [1, 126, 1024]> input_243_cast_fp16 = reshape(shape = var_953, x = var_952_cast_fp16)[name = string("input_243_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_4_self_attn_linear_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58986560))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59510912))))[name = string("model_layers_4_self_attn_linear_out_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_43_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_4_self_attn_linear_out_weight_to_fp16_palettized, x = input_243_cast_fp16)[name = string("linear_43_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_247_cast_fp16 = add(x = input_239_cast_fp16, y = linear_43_cast_fp16)[name = string("input_247_cast_fp16")];
+            tensor<int32, [1]> x_105_axes_0 = const()[name = string("x_105_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_4_norm_conv_weight_to_fp16 = const()[name = string("model_layers_4_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59512000)))];
+            tensor<fp16, [1024]> model_layers_4_norm_conv_bias_to_fp16 = const()[name = string("model_layers_4_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59514112)))];
+            tensor<fp16, [1, 126, 1024]> x_105_cast_fp16 = layer_norm(axes = x_105_axes_0, beta = model_layers_4_norm_conv_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_4_norm_conv_weight_to_fp16, x = input_247_cast_fp16)[name = string("x_105_cast_fp16")];
+            tensor<int32, [3]> input_249_perm_0 = const()[name = string("input_249_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_251_pad_type_0 = const()[name = string("input_251_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_251_strides_0 = const()[name = string("input_251_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_251_pad_0 = const()[name = string("input_251_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_251_dilations_0 = const()[name = string("input_251_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_251_groups_0 = const()[name = string("input_251_groups_0"), val = int32(1)];
+            tensor<fp16, [2048, 1024, 1]> model_layers_4_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59516224))), lut = tensor<fp16, [64, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60564864))))[name = string("model_layers_4_conv_pointwise_conv1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 1024, 126]> input_249_cast_fp16 = transpose(perm = input_249_perm_0, x = x_105_cast_fp16)[name = string("transpose_279")];
+            tensor<fp16, [1, 2048, 126]> input_251_cast_fp16 = conv(dilations = input_251_dilations_0, groups = input_251_groups_0, pad = input_251_pad_0, pad_type = input_251_pad_type_0, strides = input_251_strides_0, weight = model_layers_4_conv_pointwise_conv1_weight_to_fp16_palettized, x = input_249_cast_fp16)[name = string("input_251_cast_fp16")];
+            int32 x_107_split_num_splits_0 = const()[name = string("x_107_split_num_splits_0"), val = int32(2)];
+            int32 x_107_split_axis_0 = const()[name = string("x_107_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 126]> x_107_split_cast_fp16_0, tensor<fp16, [1, 1024, 126]> x_107_split_cast_fp16_1 = split(axis = x_107_split_axis_0, num_splits = x_107_split_num_splits_0, x = input_251_cast_fp16)[name = string("x_107_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_107_split_1_sigmoid_cast_fp16 = sigmoid(x = x_107_split_cast_fp16_1)[name = string("x_107_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_107_cast_fp16 = mul(x = x_107_split_cast_fp16_0, y = x_107_split_1_sigmoid_cast_fp16)[name = string("x_107_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_253_cast_fp16 = select(a = var_6_to_fp16, b = x_107_cast_fp16, cond = var_323)[name = string("input_253_cast_fp16")];
+            tensor<int32, [6]> input_255_pad_0 = const()[name = string("input_255_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_255_mode_0 = const()[name = string("input_255_mode_0"), val = string("constant")];
+            fp16 const_57_to_fp16 = const()[name = string("const_57_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 134]> input_255_cast_fp16 = pad(constant_val = const_57_to_fp16, mode = input_255_mode_0, pad = input_255_pad_0, x = input_253_cast_fp16)[name = string("input_255_cast_fp16")];
+            string input_257_pad_type_0 = const()[name = string("input_257_pad_type_0"), val = string("valid")];
+            int32 input_257_groups_0 = const()[name = string("input_257_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_257_strides_0 = const()[name = string("input_257_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_257_pad_0 = const()[name = string("input_257_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_257_dilations_0 = const()[name = string("input_257_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_256_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60566976))), lut = tensor<fp16, [32, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60571648))))[name = string("const_256_to_fp16_palettized")];
+            tensor<fp16, [1024]> const_257_to_fp16 = const()[name = string("const_257_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60572736)))];
+            tensor<fp16, [1, 1024, 126]> input_259_cast_fp16 = conv(bias = const_257_to_fp16, dilations = input_257_dilations_0, groups = input_257_groups_0, pad = input_257_pad_0, pad_type = input_257_pad_type_0, strides = input_257_strides_0, weight = const_256_to_fp16_palettized, x = input_255_cast_fp16)[name = string("input_259_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_261_cast_fp16 = silu(x = input_259_cast_fp16)[name = string("input_261_cast_fp16")];
+            string x_109_pad_type_0 = const()[name = string("x_109_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_109_strides_0 = const()[name = string("x_109_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_109_pad_0 = const()[name = string("x_109_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_109_dilations_0 = const()[name = string("x_109_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_109_groups_0 = const()[name = string("x_109_groups_0"), val = int32(1)];
+            tensor<fp16, [1024, 1024, 1]> model_layers_4_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60574848))), lut = tensor<fp16, [32, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(61099200))))[name = string("model_layers_4_conv_pointwise_conv2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 1024, 126]> x_109_cast_fp16 = conv(dilations = x_109_dilations_0, groups = x_109_groups_0, pad = x_109_pad_0, pad_type = x_109_pad_type_0, strides = x_109_strides_0, weight = model_layers_4_conv_pointwise_conv2_weight_to_fp16_palettized, x = input_261_cast_fp16)[name = string("x_109_cast_fp16")];
+            tensor<int32, [3]> input_263_perm_0 = const()[name = string("input_263_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 126, 1024]> input_263_cast_fp16 = transpose(perm = input_263_perm_0, x = x_109_cast_fp16)[name = string("transpose_278")];
+            tensor<fp16, [1, 126, 1024]> input_265_cast_fp16 = add(x = input_247_cast_fp16, y = input_263_cast_fp16)[name = string("input_265_cast_fp16")];
+            tensor<int32, [1]> input_267_axes_0 = const()[name = string("input_267_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_4_norm_feed_forward2_weight_to_fp16 = const()[name = string("model_layers_4_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(61100288)))];
+            tensor<fp16, [1024]> model_layers_4_norm_feed_forward2_bias_to_fp16 = const()[name = string("model_layers_4_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(61102400)))];
+            tensor<fp16, [1, 126, 1024]> input_267_cast_fp16 = layer_norm(axes = input_267_axes_0, beta = model_layers_4_norm_feed_forward2_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_4_norm_feed_forward2_weight_to_fp16, x = input_265_cast_fp16)[name = string("input_267_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_4_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(61104512))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63201728))))[name = string("model_layers_4_feed_forward2_linear1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 4096]> linear_44_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_4_feed_forward2_linear1_weight_to_fp16_palettized, x = input_267_cast_fp16)[name = string("linear_44_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_271_cast_fp16 = silu(x = linear_44_cast_fp16)[name = string("input_271_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_4_feed_forward2_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63205888))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65303104))))[name = string("model_layers_4_feed_forward2_linear2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_45_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_4_feed_forward2_linear2_weight_to_fp16_palettized, x = input_271_cast_fp16)[name = string("linear_45_cast_fp16")];
+            fp16 var_1013_to_fp16 = const()[name = string("op_1013_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_1014_cast_fp16 = mul(x = linear_45_cast_fp16, y = var_1013_to_fp16)[name = string("op_1014_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_277_cast_fp16 = add(x = input_265_cast_fp16, y = var_1014_cast_fp16)[name = string("input_277_cast_fp16")];
+            tensor<int32, [1]> input_279_axes_0 = const()[name = string("input_279_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_4_norm_out_weight_to_fp16 = const()[name = string("model_layers_4_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65304192)))];
+            tensor<fp16, [1024]> model_layers_4_norm_out_bias_to_fp16 = const()[name = string("model_layers_4_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65306304)))];
+            tensor<fp16, [1, 126, 1024]> input_279_cast_fp16 = layer_norm(axes = input_279_axes_0, beta = model_layers_4_norm_out_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_4_norm_out_weight_to_fp16, x = input_277_cast_fp16)[name = string("input_279_cast_fp16")];
+            tensor<int32, [1]> input_281_axes_0 = const()[name = string("input_281_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_5_norm_feed_forward1_weight_to_fp16 = const()[name = string("model_layers_5_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65308416)))];
+            tensor<fp16, [1024]> model_layers_5_norm_feed_forward1_bias_to_fp16 = const()[name = string("model_layers_5_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65310528)))];
+            tensor<fp16, [1, 126, 1024]> input_281_cast_fp16 = layer_norm(axes = input_281_axes_0, beta = model_layers_5_norm_feed_forward1_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_5_norm_feed_forward1_weight_to_fp16, x = input_279_cast_fp16)[name = string("input_281_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_5_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65312640))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67409856))))[name = string("model_layers_5_feed_forward1_linear1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 4096]> linear_46_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_5_feed_forward1_linear1_weight_to_fp16_palettized, x = input_281_cast_fp16)[name = string("linear_46_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_285_cast_fp16 = silu(x = linear_46_cast_fp16)[name = string("input_285_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_5_feed_forward1_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67414016))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69511232))))[name = string("model_layers_5_feed_forward1_linear2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_47_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_5_feed_forward1_linear2_weight_to_fp16_palettized, x = input_285_cast_fp16)[name = string("linear_47_cast_fp16")];
+            fp16 var_1042_to_fp16 = const()[name = string("op_1042_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_1043_cast_fp16 = mul(x = linear_47_cast_fp16, y = var_1042_to_fp16)[name = string("op_1043_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_291_cast_fp16 = add(x = input_279_cast_fp16, y = var_1043_cast_fp16)[name = string("input_291_cast_fp16")];
+            tensor<int32, [1]> query_11_axes_0 = const()[name = string("query_11_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_5_norm_self_att_weight_to_fp16 = const()[name = string("model_layers_5_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69512320)))];
+            tensor<fp16, [1024]> model_layers_5_norm_self_att_bias_to_fp16 = const()[name = string("model_layers_5_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69514432)))];
+            tensor<fp16, [1, 126, 1024]> query_11_cast_fp16 = layer_norm(axes = query_11_axes_0, beta = model_layers_5_norm_self_att_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_5_norm_self_att_weight_to_fp16, x = input_291_cast_fp16)[name = string("query_11_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_5_self_attn_linear_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69516544))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70040896))))[name = string("model_layers_5_self_attn_linear_q_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_48_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_5_self_attn_linear_q_weight_to_fp16_palettized, x = query_11_cast_fp16)[name = string("linear_48_cast_fp16")];
+            tensor<int32, [4]> var_1059 = const()[name = string("op_1059"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> q_31_cast_fp16 = reshape(shape = var_1059, x = linear_48_cast_fp16)[name = string("q_31_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_5_self_attn_linear_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70041984))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70566336))))[name = string("model_layers_5_self_attn_linear_k_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_49_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_5_self_attn_linear_k_weight_to_fp16_palettized, x = query_11_cast_fp16)[name = string("linear_49_cast_fp16")];
+            tensor<int32, [4]> var_1063 = const()[name = string("op_1063"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> k_21_cast_fp16 = reshape(shape = var_1063, x = linear_49_cast_fp16)[name = string("k_21_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_5_self_attn_linear_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70567424))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71091776))))[name = string("model_layers_5_self_attn_linear_v_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_50_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_5_self_attn_linear_v_weight_to_fp16_palettized, x = query_11_cast_fp16)[name = string("linear_50_cast_fp16")];
+            tensor<int32, [4]> var_1067 = const()[name = string("op_1067"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> v_11_cast_fp16 = reshape(shape = var_1067, x = linear_50_cast_fp16)[name = string("v_11_cast_fp16")];
+            tensor<int32, [4]> value_11_perm_0 = const()[name = string("value_11_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> model_layers_5_self_attn_pos_bias_u_to_fp16 = const()[name = string("model_layers_5_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71092864)))];
+            tensor<fp16, [1, 126, 8, 128]> var_1079_cast_fp16 = add(x = q_31_cast_fp16, y = model_layers_5_self_attn_pos_bias_u_to_fp16)[name = string("op_1079_cast_fp16")];
+            tensor<fp16, [8, 128]> model_layers_5_self_attn_pos_bias_v_to_fp16 = const()[name = string("model_layers_5_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71094976)))];
+            tensor<fp16, [1, 126, 8, 128]> var_1081_cast_fp16 = add(x = q_31_cast_fp16, y = model_layers_5_self_attn_pos_bias_v_to_fp16)[name = string("op_1081_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_11_perm_0 = const()[name = string("q_with_bias_v_11_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_117_transpose_x_0 = const()[name = string("x_117_transpose_x_0"), val = bool(false)];
+            bool x_117_transpose_y_0 = const()[name = string("x_117_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 251]> var_1083_to_fp16 = const()[name = string("op_1083_to_fp16"), val = tensor<fp16, [1, 8, 128, 251]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71097088)))];
+            tensor<fp16, [1, 8, 126, 128]> q_with_bias_v_11_cast_fp16 = transpose(perm = q_with_bias_v_11_perm_0, x = var_1081_cast_fp16)[name = string("transpose_277")];
+            tensor<fp16, [1, 8, 126, 251]> x_117_cast_fp16 = matmul(transpose_x = x_117_transpose_x_0, transpose_y = x_117_transpose_y_0, x = q_with_bias_v_11_cast_fp16, y = var_1083_to_fp16)[name = string("x_117_cast_fp16")];
+            tensor<int32, [8]> x_119_pad_0 = const()[name = string("x_119_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_119_mode_0 = const()[name = string("x_119_mode_0"), val = string("constant")];
+            fp16 const_64_to_fp16 = const()[name = string("const_64_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 252]> x_119_cast_fp16 = pad(constant_val = const_64_to_fp16, mode = x_119_mode_0, pad = x_119_pad_0, x = x_117_cast_fp16)[name = string("x_119_cast_fp16")];
+            tensor<int32, [4]> var_1091 = const()[name = string("op_1091"), val = tensor<int32, [4]>([1, 8, -1, 126])];
+            tensor<fp16, [1, 8, 252, 126]> x_121_cast_fp16 = reshape(shape = var_1091, x = x_119_cast_fp16)[name = string("x_121_cast_fp16")];
+            tensor<int32, [4]> var_1095_begin_0 = const()[name = string("op_1095_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_1095_end_0 = const()[name = string("op_1095_end_0"), val = tensor<int32, [4]>([1, 8, 252, 126])];
+            tensor<bool, [4]> var_1095_end_mask_0 = const()[name = string("op_1095_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 251, 126]> var_1095_cast_fp16 = slice_by_index(begin = var_1095_begin_0, end = var_1095_end_0, end_mask = var_1095_end_mask_0, x = x_121_cast_fp16)[name = string("op_1095_cast_fp16")];
+            tensor<int32, [4]> var_1096 = const()[name = string("op_1096"), val = tensor<int32, [4]>([1, 8, 126, 251])];
+            tensor<fp16, [1, 8, 126, 251]> matrix_bd_21_cast_fp16 = reshape(shape = var_1096, x = var_1095_cast_fp16)[name = string("matrix_bd_21_cast_fp16")];
+            bool matrix_ac_11_transpose_x_0 = const()[name = string("matrix_ac_11_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_11_transpose_y_0 = const()[name = string("matrix_ac_11_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_106_perm_0 = const()[name = string("transpose_106_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_107_perm_0 = const()[name = string("transpose_107_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 126]> transpose_107 = transpose(perm = transpose_107_perm_0, x = k_21_cast_fp16)[name = string("transpose_275")];
+            tensor<fp16, [1, 8, 126, 128]> transpose_106 = transpose(perm = transpose_106_perm_0, x = var_1079_cast_fp16)[name = string("transpose_276")];
+            tensor<fp16, [1, 8, 126, 126]> matrix_ac_11_cast_fp16 = matmul(transpose_x = matrix_ac_11_transpose_x_0, transpose_y = matrix_ac_11_transpose_y_0, x = transpose_106, y = transpose_107)[name = string("matrix_ac_11_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_23_begin_0 = const()[name = string("matrix_bd_23_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_23_end_0 = const()[name = string("matrix_bd_23_end_0"), val = tensor<int32, [4]>([1, 8, 126, 126])];
+            tensor<bool, [4]> matrix_bd_23_end_mask_0 = const()[name = string("matrix_bd_23_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 126, 126]> matrix_bd_23_cast_fp16 = slice_by_index(begin = matrix_bd_23_begin_0, end = matrix_bd_23_end_0, end_mask = matrix_bd_23_end_mask_0, x = matrix_bd_21_cast_fp16)[name = string("matrix_bd_23_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_1105_cast_fp16 = add(x = matrix_ac_11_cast_fp16, y = matrix_bd_23_cast_fp16)[name = string("op_1105_cast_fp16")];
+            fp16 _inversed_scores_21_y_0_to_fp16 = const()[name = string("_inversed_scores_21_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 126, 126]> _inversed_scores_21_cast_fp16 = mul(x = var_1105_cast_fp16, y = _inversed_scores_21_y_0_to_fp16)[name = string("_inversed_scores_21_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> scores_23_cast_fp16 = select(a = var_7_to_fp16, b = _inversed_scores_21_cast_fp16, cond = mask_3)[name = string("scores_23_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_1111_cast_fp16 = softmax(axis = var_25, x = scores_23_cast_fp16)[name = string("op_1111_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> input_293_cast_fp16 = select(a = var_6_to_fp16, b = var_1111_cast_fp16, cond = mask_3)[name = string("input_293_cast_fp16")];
+            bool x_123_transpose_x_0 = const()[name = string("x_123_transpose_x_0"), val = bool(false)];
+            bool x_123_transpose_y_0 = const()[name = string("x_123_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 126, 128]> value_11_cast_fp16 = transpose(perm = value_11_perm_0, x = v_11_cast_fp16)[name = string("transpose_274")];
+            tensor<fp16, [1, 8, 126, 128]> x_123_cast_fp16 = matmul(transpose_x = x_123_transpose_x_0, transpose_y = x_123_transpose_y_0, x = input_293_cast_fp16, y = value_11_cast_fp16)[name = string("x_123_cast_fp16")];
+            tensor<int32, [4]> var_1115_perm_0 = const()[name = string("op_1115_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_1116 = const()[name = string("op_1116"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 126, 8, 128]> var_1115_cast_fp16 = transpose(perm = var_1115_perm_0, x = x_123_cast_fp16)[name = string("transpose_273")];
+            tensor<fp16, [1, 126, 1024]> input_295_cast_fp16 = reshape(shape = var_1116, x = var_1115_cast_fp16)[name = string("input_295_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_5_self_attn_linear_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71611200))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72135552))))[name = string("model_layers_5_self_attn_linear_out_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_52_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_5_self_attn_linear_out_weight_to_fp16_palettized, x = input_295_cast_fp16)[name = string("linear_52_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_299_cast_fp16 = add(x = input_291_cast_fp16, y = linear_52_cast_fp16)[name = string("input_299_cast_fp16")];
+            tensor<int32, [1]> x_127_axes_0 = const()[name = string("x_127_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_5_norm_conv_weight_to_fp16 = const()[name = string("model_layers_5_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72136640)))];
+            tensor<fp16, [1024]> model_layers_5_norm_conv_bias_to_fp16 = const()[name = string("model_layers_5_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72138752)))];
+            tensor<fp16, [1, 126, 1024]> x_127_cast_fp16 = layer_norm(axes = x_127_axes_0, beta = model_layers_5_norm_conv_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_5_norm_conv_weight_to_fp16, x = input_299_cast_fp16)[name = string("x_127_cast_fp16")];
+            tensor<int32, [3]> input_301_perm_0 = const()[name = string("input_301_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_303_pad_type_0 = const()[name = string("input_303_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_303_strides_0 = const()[name = string("input_303_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_303_pad_0 = const()[name = string("input_303_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_303_dilations_0 = const()[name = string("input_303_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_303_groups_0 = const()[name = string("input_303_groups_0"), val = int32(1)];
+            tensor<fp16, [2048, 1024, 1]> model_layers_5_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72140864))), lut = tensor<fp16, [64, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(73189504))))[name = string("model_layers_5_conv_pointwise_conv1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 1024, 126]> input_301_cast_fp16 = transpose(perm = input_301_perm_0, x = x_127_cast_fp16)[name = string("transpose_272")];
+            tensor<fp16, [1, 2048, 126]> input_303_cast_fp16 = conv(dilations = input_303_dilations_0, groups = input_303_groups_0, pad = input_303_pad_0, pad_type = input_303_pad_type_0, strides = input_303_strides_0, weight = model_layers_5_conv_pointwise_conv1_weight_to_fp16_palettized, x = input_301_cast_fp16)[name = string("input_303_cast_fp16")];
+            int32 x_129_split_num_splits_0 = const()[name = string("x_129_split_num_splits_0"), val = int32(2)];
+            int32 x_129_split_axis_0 = const()[name = string("x_129_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 126]> x_129_split_cast_fp16_0, tensor<fp16, [1, 1024, 126]> x_129_split_cast_fp16_1 = split(axis = x_129_split_axis_0, num_splits = x_129_split_num_splits_0, x = input_303_cast_fp16)[name = string("x_129_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_129_split_1_sigmoid_cast_fp16 = sigmoid(x = x_129_split_cast_fp16_1)[name = string("x_129_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_129_cast_fp16 = mul(x = x_129_split_cast_fp16_0, y = x_129_split_1_sigmoid_cast_fp16)[name = string("x_129_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_305_cast_fp16 = select(a = var_6_to_fp16, b = x_129_cast_fp16, cond = var_323)[name = string("input_305_cast_fp16")];
+            tensor<int32, [6]> input_307_pad_0 = const()[name = string("input_307_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_307_mode_0 = const()[name = string("input_307_mode_0"), val = string("constant")];
+            fp16 const_67_to_fp16 = const()[name = string("const_67_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 134]> input_307_cast_fp16 = pad(constant_val = const_67_to_fp16, mode = input_307_mode_0, pad = input_307_pad_0, x = input_305_cast_fp16)[name = string("input_307_cast_fp16")];
+            string input_309_pad_type_0 = const()[name = string("input_309_pad_type_0"), val = string("valid")];
+            int32 input_309_groups_0 = const()[name = string("input_309_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_309_strides_0 = const()[name = string("input_309_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_309_pad_0 = const()[name = string("input_309_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_309_dilations_0 = const()[name = string("input_309_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_258_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(73191616))), lut = tensor<fp16, [32, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(73196288))))[name = string("const_258_to_fp16_palettized")];
+            tensor<fp16, [1024]> const_259_to_fp16 = const()[name = string("const_259_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(73197376)))];
+            tensor<fp16, [1, 1024, 126]> input_311_cast_fp16 = conv(bias = const_259_to_fp16, dilations = input_309_dilations_0, groups = input_309_groups_0, pad = input_309_pad_0, pad_type = input_309_pad_type_0, strides = input_309_strides_0, weight = const_258_to_fp16_palettized, x = input_307_cast_fp16)[name = string("input_311_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_313_cast_fp16 = silu(x = input_311_cast_fp16)[name = string("input_313_cast_fp16")];
+            string x_131_pad_type_0 = const()[name = string("x_131_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_131_strides_0 = const()[name = string("x_131_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_131_pad_0 = const()[name = string("x_131_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_131_dilations_0 = const()[name = string("x_131_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_131_groups_0 = const()[name = string("x_131_groups_0"), val = int32(1)];
+            tensor<fp16, [1024, 1024, 1]> model_layers_5_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(73199488))), lut = tensor<fp16, [32, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(73723840))))[name = string("model_layers_5_conv_pointwise_conv2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 1024, 126]> x_131_cast_fp16 = conv(dilations = x_131_dilations_0, groups = x_131_groups_0, pad = x_131_pad_0, pad_type = x_131_pad_type_0, strides = x_131_strides_0, weight = model_layers_5_conv_pointwise_conv2_weight_to_fp16_palettized, x = input_313_cast_fp16)[name = string("x_131_cast_fp16")];
+            tensor<int32, [3]> input_315_perm_0 = const()[name = string("input_315_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 126, 1024]> input_315_cast_fp16 = transpose(perm = input_315_perm_0, x = x_131_cast_fp16)[name = string("transpose_271")];
+            tensor<fp16, [1, 126, 1024]> input_317_cast_fp16 = add(x = input_299_cast_fp16, y = input_315_cast_fp16)[name = string("input_317_cast_fp16")];
+            tensor<int32, [1]> input_319_axes_0 = const()[name = string("input_319_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_5_norm_feed_forward2_weight_to_fp16 = const()[name = string("model_layers_5_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(73724928)))];
+            tensor<fp16, [1024]> model_layers_5_norm_feed_forward2_bias_to_fp16 = const()[name = string("model_layers_5_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(73727040)))];
+            tensor<fp16, [1, 126, 1024]> input_319_cast_fp16 = layer_norm(axes = input_319_axes_0, beta = model_layers_5_norm_feed_forward2_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_5_norm_feed_forward2_weight_to_fp16, x = input_317_cast_fp16)[name = string("input_319_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_5_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(73729152))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(75826368))))[name = string("model_layers_5_feed_forward2_linear1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 4096]> linear_53_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_5_feed_forward2_linear1_weight_to_fp16_palettized, x = input_319_cast_fp16)[name = string("linear_53_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_323_cast_fp16 = silu(x = linear_53_cast_fp16)[name = string("input_323_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_5_feed_forward2_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(75830528))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77927744))))[name = string("model_layers_5_feed_forward2_linear2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_54_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_5_feed_forward2_linear2_weight_to_fp16_palettized, x = input_323_cast_fp16)[name = string("linear_54_cast_fp16")];
+            fp16 var_1176_to_fp16 = const()[name = string("op_1176_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_1177_cast_fp16 = mul(x = linear_54_cast_fp16, y = var_1176_to_fp16)[name = string("op_1177_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_329_cast_fp16 = add(x = input_317_cast_fp16, y = var_1177_cast_fp16)[name = string("input_329_cast_fp16")];
+            tensor<int32, [1]> input_331_axes_0 = const()[name = string("input_331_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_5_norm_out_weight_to_fp16 = const()[name = string("model_layers_5_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77928832)))];
+            tensor<fp16, [1024]> model_layers_5_norm_out_bias_to_fp16 = const()[name = string("model_layers_5_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77930944)))];
+            tensor<fp16, [1, 126, 1024]> input_331_cast_fp16 = layer_norm(axes = input_331_axes_0, beta = model_layers_5_norm_out_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_5_norm_out_weight_to_fp16, x = input_329_cast_fp16)[name = string("input_331_cast_fp16")];
+            tensor<int32, [1]> input_333_axes_0 = const()[name = string("input_333_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_6_norm_feed_forward1_weight_to_fp16 = const()[name = string("model_layers_6_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77933056)))];
+            tensor<fp16, [1024]> model_layers_6_norm_feed_forward1_bias_to_fp16 = const()[name = string("model_layers_6_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77935168)))];
+            tensor<fp16, [1, 126, 1024]> input_333_cast_fp16 = layer_norm(axes = input_333_axes_0, beta = model_layers_6_norm_feed_forward1_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_6_norm_feed_forward1_weight_to_fp16, x = input_331_cast_fp16)[name = string("input_333_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_6_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77937280))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80034496))))[name = string("model_layers_6_feed_forward1_linear1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 4096]> linear_55_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_6_feed_forward1_linear1_weight_to_fp16_palettized, x = input_333_cast_fp16)[name = string("linear_55_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_337_cast_fp16 = silu(x = linear_55_cast_fp16)[name = string("input_337_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_6_feed_forward1_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80038656))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82135872))))[name = string("model_layers_6_feed_forward1_linear2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_56_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_6_feed_forward1_linear2_weight_to_fp16_palettized, x = input_337_cast_fp16)[name = string("linear_56_cast_fp16")];
+            fp16 var_1205_to_fp16 = const()[name = string("op_1205_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_1206_cast_fp16 = mul(x = linear_56_cast_fp16, y = var_1205_to_fp16)[name = string("op_1206_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_343_cast_fp16 = add(x = input_331_cast_fp16, y = var_1206_cast_fp16)[name = string("input_343_cast_fp16")];
+            tensor<int32, [1]> query_13_axes_0 = const()[name = string("query_13_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_6_norm_self_att_weight_to_fp16 = const()[name = string("model_layers_6_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82136960)))];
+            tensor<fp16, [1024]> model_layers_6_norm_self_att_bias_to_fp16 = const()[name = string("model_layers_6_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82139072)))];
+            tensor<fp16, [1, 126, 1024]> query_13_cast_fp16 = layer_norm(axes = query_13_axes_0, beta = model_layers_6_norm_self_att_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_6_norm_self_att_weight_to_fp16, x = input_343_cast_fp16)[name = string("query_13_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_6_self_attn_linear_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82141184))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82665536))))[name = string("model_layers_6_self_attn_linear_q_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_57_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_6_self_attn_linear_q_weight_to_fp16_palettized, x = query_13_cast_fp16)[name = string("linear_57_cast_fp16")];
+            tensor<int32, [4]> var_1222 = const()[name = string("op_1222"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> q_37_cast_fp16 = reshape(shape = var_1222, x = linear_57_cast_fp16)[name = string("q_37_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_6_self_attn_linear_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82666624))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(83190976))))[name = string("model_layers_6_self_attn_linear_k_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_58_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_6_self_attn_linear_k_weight_to_fp16_palettized, x = query_13_cast_fp16)[name = string("linear_58_cast_fp16")];
+            tensor<int32, [4]> var_1226 = const()[name = string("op_1226"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> k_25_cast_fp16 = reshape(shape = var_1226, x = linear_58_cast_fp16)[name = string("k_25_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_6_self_attn_linear_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(83192064))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(83716416))))[name = string("model_layers_6_self_attn_linear_v_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_59_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_6_self_attn_linear_v_weight_to_fp16_palettized, x = query_13_cast_fp16)[name = string("linear_59_cast_fp16")];
+            tensor<int32, [4]> var_1230 = const()[name = string("op_1230"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> v_13_cast_fp16 = reshape(shape = var_1230, x = linear_59_cast_fp16)[name = string("v_13_cast_fp16")];
+            tensor<int32, [4]> value_13_perm_0 = const()[name = string("value_13_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> model_layers_6_self_attn_pos_bias_u_to_fp16 = const()[name = string("model_layers_6_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(83717504)))];
+            tensor<fp16, [1, 126, 8, 128]> var_1242_cast_fp16 = add(x = q_37_cast_fp16, y = model_layers_6_self_attn_pos_bias_u_to_fp16)[name = string("op_1242_cast_fp16")];
+            tensor<fp16, [8, 128]> model_layers_6_self_attn_pos_bias_v_to_fp16 = const()[name = string("model_layers_6_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(83719616)))];
+            tensor<fp16, [1, 126, 8, 128]> var_1244_cast_fp16 = add(x = q_37_cast_fp16, y = model_layers_6_self_attn_pos_bias_v_to_fp16)[name = string("op_1244_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_13_perm_0 = const()[name = string("q_with_bias_v_13_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_139_transpose_x_0 = const()[name = string("x_139_transpose_x_0"), val = bool(false)];
+            bool x_139_transpose_y_0 = const()[name = string("x_139_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 251]> var_1246_to_fp16 = const()[name = string("op_1246_to_fp16"), val = tensor<fp16, [1, 8, 128, 251]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(83721728)))];
+            tensor<fp16, [1, 8, 126, 128]> q_with_bias_v_13_cast_fp16 = transpose(perm = q_with_bias_v_13_perm_0, x = var_1244_cast_fp16)[name = string("transpose_270")];
+            tensor<fp16, [1, 8, 126, 251]> x_139_cast_fp16 = matmul(transpose_x = x_139_transpose_x_0, transpose_y = x_139_transpose_y_0, x = q_with_bias_v_13_cast_fp16, y = var_1246_to_fp16)[name = string("x_139_cast_fp16")];
+            tensor<int32, [8]> x_141_pad_0 = const()[name = string("x_141_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_141_mode_0 = const()[name = string("x_141_mode_0"), val = string("constant")];
+            fp16 const_74_to_fp16 = const()[name = string("const_74_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 252]> x_141_cast_fp16 = pad(constant_val = const_74_to_fp16, mode = x_141_mode_0, pad = x_141_pad_0, x = x_139_cast_fp16)[name = string("x_141_cast_fp16")];
+            tensor<int32, [4]> var_1254 = const()[name = string("op_1254"), val = tensor<int32, [4]>([1, 8, -1, 126])];
+            tensor<fp16, [1, 8, 252, 126]> x_143_cast_fp16 = reshape(shape = var_1254, x = x_141_cast_fp16)[name = string("x_143_cast_fp16")];
+            tensor<int32, [4]> var_1258_begin_0 = const()[name = string("op_1258_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_1258_end_0 = const()[name = string("op_1258_end_0"), val = tensor<int32, [4]>([1, 8, 252, 126])];
+            tensor<bool, [4]> var_1258_end_mask_0 = const()[name = string("op_1258_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 251, 126]> var_1258_cast_fp16 = slice_by_index(begin = var_1258_begin_0, end = var_1258_end_0, end_mask = var_1258_end_mask_0, x = x_143_cast_fp16)[name = string("op_1258_cast_fp16")];
+            tensor<int32, [4]> var_1259 = const()[name = string("op_1259"), val = tensor<int32, [4]>([1, 8, 126, 251])];
+            tensor<fp16, [1, 8, 126, 251]> matrix_bd_25_cast_fp16 = reshape(shape = var_1259, x = var_1258_cast_fp16)[name = string("matrix_bd_25_cast_fp16")];
+            bool matrix_ac_13_transpose_x_0 = const()[name = string("matrix_ac_13_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_13_transpose_y_0 = const()[name = string("matrix_ac_13_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_108_perm_0 = const()[name = string("transpose_108_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_109_perm_0 = const()[name = string("transpose_109_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 126]> transpose_109 = transpose(perm = transpose_109_perm_0, x = k_25_cast_fp16)[name = string("transpose_268")];
+            tensor<fp16, [1, 8, 126, 128]> transpose_108 = transpose(perm = transpose_108_perm_0, x = var_1242_cast_fp16)[name = string("transpose_269")];
+            tensor<fp16, [1, 8, 126, 126]> matrix_ac_13_cast_fp16 = matmul(transpose_x = matrix_ac_13_transpose_x_0, transpose_y = matrix_ac_13_transpose_y_0, x = transpose_108, y = transpose_109)[name = string("matrix_ac_13_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_27_begin_0 = const()[name = string("matrix_bd_27_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_27_end_0 = const()[name = string("matrix_bd_27_end_0"), val = tensor<int32, [4]>([1, 8, 126, 126])];
+            tensor<bool, [4]> matrix_bd_27_end_mask_0 = const()[name = string("matrix_bd_27_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 126, 126]> matrix_bd_27_cast_fp16 = slice_by_index(begin = matrix_bd_27_begin_0, end = matrix_bd_27_end_0, end_mask = matrix_bd_27_end_mask_0, x = matrix_bd_25_cast_fp16)[name = string("matrix_bd_27_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_1268_cast_fp16 = add(x = matrix_ac_13_cast_fp16, y = matrix_bd_27_cast_fp16)[name = string("op_1268_cast_fp16")];
+            fp16 _inversed_scores_25_y_0_to_fp16 = const()[name = string("_inversed_scores_25_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 126, 126]> _inversed_scores_25_cast_fp16 = mul(x = var_1268_cast_fp16, y = _inversed_scores_25_y_0_to_fp16)[name = string("_inversed_scores_25_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> scores_27_cast_fp16 = select(a = var_7_to_fp16, b = _inversed_scores_25_cast_fp16, cond = mask_3)[name = string("scores_27_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_1274_cast_fp16 = softmax(axis = var_25, x = scores_27_cast_fp16)[name = string("op_1274_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> input_345_cast_fp16 = select(a = var_6_to_fp16, b = var_1274_cast_fp16, cond = mask_3)[name = string("input_345_cast_fp16")];
+            bool x_145_transpose_x_0 = const()[name = string("x_145_transpose_x_0"), val = bool(false)];
+            bool x_145_transpose_y_0 = const()[name = string("x_145_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 126, 128]> value_13_cast_fp16 = transpose(perm = value_13_perm_0, x = v_13_cast_fp16)[name = string("transpose_267")];
+            tensor<fp16, [1, 8, 126, 128]> x_145_cast_fp16 = matmul(transpose_x = x_145_transpose_x_0, transpose_y = x_145_transpose_y_0, x = input_345_cast_fp16, y = value_13_cast_fp16)[name = string("x_145_cast_fp16")];
+            tensor<int32, [4]> var_1278_perm_0 = const()[name = string("op_1278_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_1279 = const()[name = string("op_1279"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 126, 8, 128]> var_1278_cast_fp16 = transpose(perm = var_1278_perm_0, x = x_145_cast_fp16)[name = string("transpose_266")];
+            tensor<fp16, [1, 126, 1024]> input_347_cast_fp16 = reshape(shape = var_1279, x = var_1278_cast_fp16)[name = string("input_347_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_6_self_attn_linear_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(84235840))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(84760192))))[name = string("model_layers_6_self_attn_linear_out_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_61_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_6_self_attn_linear_out_weight_to_fp16_palettized, x = input_347_cast_fp16)[name = string("linear_61_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_351_cast_fp16 = add(x = input_343_cast_fp16, y = linear_61_cast_fp16)[name = string("input_351_cast_fp16")];
+            tensor<int32, [1]> x_149_axes_0 = const()[name = string("x_149_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_6_norm_conv_weight_to_fp16 = const()[name = string("model_layers_6_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(84761280)))];
+            tensor<fp16, [1024]> model_layers_6_norm_conv_bias_to_fp16 = const()[name = string("model_layers_6_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(84763392)))];
+            tensor<fp16, [1, 126, 1024]> x_149_cast_fp16 = layer_norm(axes = x_149_axes_0, beta = model_layers_6_norm_conv_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_6_norm_conv_weight_to_fp16, x = input_351_cast_fp16)[name = string("x_149_cast_fp16")];
+            tensor<int32, [3]> input_353_perm_0 = const()[name = string("input_353_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_355_pad_type_0 = const()[name = string("input_355_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_355_strides_0 = const()[name = string("input_355_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_355_pad_0 = const()[name = string("input_355_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_355_dilations_0 = const()[name = string("input_355_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_355_groups_0 = const()[name = string("input_355_groups_0"), val = int32(1)];
+            tensor<fp16, [2048, 1024, 1]> model_layers_6_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(84765504))), lut = tensor<fp16, [64, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85814144))))[name = string("model_layers_6_conv_pointwise_conv1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 1024, 126]> input_353_cast_fp16 = transpose(perm = input_353_perm_0, x = x_149_cast_fp16)[name = string("transpose_265")];
+            tensor<fp16, [1, 2048, 126]> input_355_cast_fp16 = conv(dilations = input_355_dilations_0, groups = input_355_groups_0, pad = input_355_pad_0, pad_type = input_355_pad_type_0, strides = input_355_strides_0, weight = model_layers_6_conv_pointwise_conv1_weight_to_fp16_palettized, x = input_353_cast_fp16)[name = string("input_355_cast_fp16")];
+            int32 x_151_split_num_splits_0 = const()[name = string("x_151_split_num_splits_0"), val = int32(2)];
+            int32 x_151_split_axis_0 = const()[name = string("x_151_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 126]> x_151_split_cast_fp16_0, tensor<fp16, [1, 1024, 126]> x_151_split_cast_fp16_1 = split(axis = x_151_split_axis_0, num_splits = x_151_split_num_splits_0, x = input_355_cast_fp16)[name = string("x_151_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_151_split_1_sigmoid_cast_fp16 = sigmoid(x = x_151_split_cast_fp16_1)[name = string("x_151_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_151_cast_fp16 = mul(x = x_151_split_cast_fp16_0, y = x_151_split_1_sigmoid_cast_fp16)[name = string("x_151_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_357_cast_fp16 = select(a = var_6_to_fp16, b = x_151_cast_fp16, cond = var_323)[name = string("input_357_cast_fp16")];
+            tensor<int32, [6]> input_359_pad_0 = const()[name = string("input_359_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_359_mode_0 = const()[name = string("input_359_mode_0"), val = string("constant")];
+            fp16 const_77_to_fp16 = const()[name = string("const_77_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 134]> input_359_cast_fp16 = pad(constant_val = const_77_to_fp16, mode = input_359_mode_0, pad = input_359_pad_0, x = input_357_cast_fp16)[name = string("input_359_cast_fp16")];
+            string input_361_pad_type_0 = const()[name = string("input_361_pad_type_0"), val = string("valid")];
+            int32 input_361_groups_0 = const()[name = string("input_361_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_361_strides_0 = const()[name = string("input_361_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_361_pad_0 = const()[name = string("input_361_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_361_dilations_0 = const()[name = string("input_361_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_260_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85816256))), lut = tensor<fp16, [32, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85820928))))[name = string("const_260_to_fp16_palettized")];
+            tensor<fp16, [1024]> const_261_to_fp16 = const()[name = string("const_261_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85822016)))];
+            tensor<fp16, [1, 1024, 126]> input_363_cast_fp16 = conv(bias = const_261_to_fp16, dilations = input_361_dilations_0, groups = input_361_groups_0, pad = input_361_pad_0, pad_type = input_361_pad_type_0, strides = input_361_strides_0, weight = const_260_to_fp16_palettized, x = input_359_cast_fp16)[name = string("input_363_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_365_cast_fp16 = silu(x = input_363_cast_fp16)[name = string("input_365_cast_fp16")];
+            string x_153_pad_type_0 = const()[name = string("x_153_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_153_strides_0 = const()[name = string("x_153_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_153_pad_0 = const()[name = string("x_153_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_153_dilations_0 = const()[name = string("x_153_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_153_groups_0 = const()[name = string("x_153_groups_0"), val = int32(1)];
+            tensor<fp16, [1024, 1024, 1]> model_layers_6_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85824128))), lut = tensor<fp16, [32, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86348480))))[name = string("model_layers_6_conv_pointwise_conv2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 1024, 126]> x_153_cast_fp16 = conv(dilations = x_153_dilations_0, groups = x_153_groups_0, pad = x_153_pad_0, pad_type = x_153_pad_type_0, strides = x_153_strides_0, weight = model_layers_6_conv_pointwise_conv2_weight_to_fp16_palettized, x = input_365_cast_fp16)[name = string("x_153_cast_fp16")];
+            tensor<int32, [3]> input_367_perm_0 = const()[name = string("input_367_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 126, 1024]> input_367_cast_fp16 = transpose(perm = input_367_perm_0, x = x_153_cast_fp16)[name = string("transpose_264")];
+            tensor<fp16, [1, 126, 1024]> input_369_cast_fp16 = add(x = input_351_cast_fp16, y = input_367_cast_fp16)[name = string("input_369_cast_fp16")];
+            tensor<int32, [1]> input_371_axes_0 = const()[name = string("input_371_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_6_norm_feed_forward2_weight_to_fp16 = const()[name = string("model_layers_6_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86349568)))];
+            tensor<fp16, [1024]> model_layers_6_norm_feed_forward2_bias_to_fp16 = const()[name = string("model_layers_6_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86351680)))];
+            tensor<fp16, [1, 126, 1024]> input_371_cast_fp16 = layer_norm(axes = input_371_axes_0, beta = model_layers_6_norm_feed_forward2_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_6_norm_feed_forward2_weight_to_fp16, x = input_369_cast_fp16)[name = string("input_371_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_6_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86353792))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88451008))))[name = string("model_layers_6_feed_forward2_linear1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 4096]> linear_62_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_6_feed_forward2_linear1_weight_to_fp16_palettized, x = input_371_cast_fp16)[name = string("linear_62_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_375_cast_fp16 = silu(x = linear_62_cast_fp16)[name = string("input_375_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_6_feed_forward2_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88455168))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90552384))))[name = string("model_layers_6_feed_forward2_linear2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_63_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_6_feed_forward2_linear2_weight_to_fp16_palettized, x = input_375_cast_fp16)[name = string("linear_63_cast_fp16")];
+            fp16 var_1339_to_fp16 = const()[name = string("op_1339_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_1340_cast_fp16 = mul(x = linear_63_cast_fp16, y = var_1339_to_fp16)[name = string("op_1340_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_381_cast_fp16 = add(x = input_369_cast_fp16, y = var_1340_cast_fp16)[name = string("input_381_cast_fp16")];
+            tensor<int32, [1]> input_383_axes_0 = const()[name = string("input_383_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_6_norm_out_weight_to_fp16 = const()[name = string("model_layers_6_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90553472)))];
+            tensor<fp16, [1024]> model_layers_6_norm_out_bias_to_fp16 = const()[name = string("model_layers_6_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90555584)))];
+            tensor<fp16, [1, 126, 1024]> input_383_cast_fp16 = layer_norm(axes = input_383_axes_0, beta = model_layers_6_norm_out_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_6_norm_out_weight_to_fp16, x = input_381_cast_fp16)[name = string("input_383_cast_fp16")];
+            tensor<int32, [1]> input_385_axes_0 = const()[name = string("input_385_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_7_norm_feed_forward1_weight_to_fp16 = const()[name = string("model_layers_7_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90557696)))];
+            tensor<fp16, [1024]> model_layers_7_norm_feed_forward1_bias_to_fp16 = const()[name = string("model_layers_7_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90559808)))];
+            tensor<fp16, [1, 126, 1024]> input_385_cast_fp16 = layer_norm(axes = input_385_axes_0, beta = model_layers_7_norm_feed_forward1_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_7_norm_feed_forward1_weight_to_fp16, x = input_383_cast_fp16)[name = string("input_385_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_7_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90561920))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92659136))))[name = string("model_layers_7_feed_forward1_linear1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 4096]> linear_64_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_7_feed_forward1_linear1_weight_to_fp16_palettized, x = input_385_cast_fp16)[name = string("linear_64_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_389_cast_fp16 = silu(x = linear_64_cast_fp16)[name = string("input_389_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_7_feed_forward1_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92663296))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(94760512))))[name = string("model_layers_7_feed_forward1_linear2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_65_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_7_feed_forward1_linear2_weight_to_fp16_palettized, x = input_389_cast_fp16)[name = string("linear_65_cast_fp16")];
+            fp16 var_1368_to_fp16 = const()[name = string("op_1368_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_1369_cast_fp16 = mul(x = linear_65_cast_fp16, y = var_1368_to_fp16)[name = string("op_1369_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_395_cast_fp16 = add(x = input_383_cast_fp16, y = var_1369_cast_fp16)[name = string("input_395_cast_fp16")];
+            tensor<int32, [1]> query_15_axes_0 = const()[name = string("query_15_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_7_norm_self_att_weight_to_fp16 = const()[name = string("model_layers_7_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(94761600)))];
+            tensor<fp16, [1024]> model_layers_7_norm_self_att_bias_to_fp16 = const()[name = string("model_layers_7_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(94763712)))];
+            tensor<fp16, [1, 126, 1024]> query_15_cast_fp16 = layer_norm(axes = query_15_axes_0, beta = model_layers_7_norm_self_att_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_7_norm_self_att_weight_to_fp16, x = input_395_cast_fp16)[name = string("query_15_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_7_self_attn_linear_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(94765824))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95290176))))[name = string("model_layers_7_self_attn_linear_q_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_66_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_7_self_attn_linear_q_weight_to_fp16_palettized, x = query_15_cast_fp16)[name = string("linear_66_cast_fp16")];
+            tensor<int32, [4]> var_1385 = const()[name = string("op_1385"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> q_43_cast_fp16 = reshape(shape = var_1385, x = linear_66_cast_fp16)[name = string("q_43_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_7_self_attn_linear_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95291264))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95815616))))[name = string("model_layers_7_self_attn_linear_k_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_67_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_7_self_attn_linear_k_weight_to_fp16_palettized, x = query_15_cast_fp16)[name = string("linear_67_cast_fp16")];
+            tensor<int32, [4]> var_1389 = const()[name = string("op_1389"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> k_29_cast_fp16 = reshape(shape = var_1389, x = linear_67_cast_fp16)[name = string("k_29_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_7_self_attn_linear_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95816704))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96341056))))[name = string("model_layers_7_self_attn_linear_v_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_68_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_7_self_attn_linear_v_weight_to_fp16_palettized, x = query_15_cast_fp16)[name = string("linear_68_cast_fp16")];
+            tensor<int32, [4]> var_1393 = const()[name = string("op_1393"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> v_15_cast_fp16 = reshape(shape = var_1393, x = linear_68_cast_fp16)[name = string("v_15_cast_fp16")];
+            tensor<int32, [4]> value_15_perm_0 = const()[name = string("value_15_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> model_layers_7_self_attn_pos_bias_u_to_fp16 = const()[name = string("model_layers_7_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96342144)))];
+            tensor<fp16, [1, 126, 8, 128]> var_1405_cast_fp16 = add(x = q_43_cast_fp16, y = model_layers_7_self_attn_pos_bias_u_to_fp16)[name = string("op_1405_cast_fp16")];
+            tensor<fp16, [8, 128]> model_layers_7_self_attn_pos_bias_v_to_fp16 = const()[name = string("model_layers_7_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96344256)))];
+            tensor<fp16, [1, 126, 8, 128]> var_1407_cast_fp16 = add(x = q_43_cast_fp16, y = model_layers_7_self_attn_pos_bias_v_to_fp16)[name = string("op_1407_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_15_perm_0 = const()[name = string("q_with_bias_v_15_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_161_transpose_x_0 = const()[name = string("x_161_transpose_x_0"), val = bool(false)];
+            bool x_161_transpose_y_0 = const()[name = string("x_161_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 251]> var_1409_to_fp16 = const()[name = string("op_1409_to_fp16"), val = tensor<fp16, [1, 8, 128, 251]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96346368)))];
+            tensor<fp16, [1, 8, 126, 128]> q_with_bias_v_15_cast_fp16 = transpose(perm = q_with_bias_v_15_perm_0, x = var_1407_cast_fp16)[name = string("transpose_263")];
+            tensor<fp16, [1, 8, 126, 251]> x_161_cast_fp16 = matmul(transpose_x = x_161_transpose_x_0, transpose_y = x_161_transpose_y_0, x = q_with_bias_v_15_cast_fp16, y = var_1409_to_fp16)[name = string("x_161_cast_fp16")];
+            tensor<int32, [8]> x_163_pad_0 = const()[name = string("x_163_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_163_mode_0 = const()[name = string("x_163_mode_0"), val = string("constant")];
+            fp16 const_84_to_fp16 = const()[name = string("const_84_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 252]> x_163_cast_fp16 = pad(constant_val = const_84_to_fp16, mode = x_163_mode_0, pad = x_163_pad_0, x = x_161_cast_fp16)[name = string("x_163_cast_fp16")];
+            tensor<int32, [4]> var_1417 = const()[name = string("op_1417"), val = tensor<int32, [4]>([1, 8, -1, 126])];
+            tensor<fp16, [1, 8, 252, 126]> x_165_cast_fp16 = reshape(shape = var_1417, x = x_163_cast_fp16)[name = string("x_165_cast_fp16")];
+            tensor<int32, [4]> var_1421_begin_0 = const()[name = string("op_1421_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_1421_end_0 = const()[name = string("op_1421_end_0"), val = tensor<int32, [4]>([1, 8, 252, 126])];
+            tensor<bool, [4]> var_1421_end_mask_0 = const()[name = string("op_1421_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 251, 126]> var_1421_cast_fp16 = slice_by_index(begin = var_1421_begin_0, end = var_1421_end_0, end_mask = var_1421_end_mask_0, x = x_165_cast_fp16)[name = string("op_1421_cast_fp16")];
+            tensor<int32, [4]> var_1422 = const()[name = string("op_1422"), val = tensor<int32, [4]>([1, 8, 126, 251])];
+            tensor<fp16, [1, 8, 126, 251]> matrix_bd_29_cast_fp16 = reshape(shape = var_1422, x = var_1421_cast_fp16)[name = string("matrix_bd_29_cast_fp16")];
+            bool matrix_ac_15_transpose_x_0 = const()[name = string("matrix_ac_15_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_15_transpose_y_0 = const()[name = string("matrix_ac_15_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_110_perm_0 = const()[name = string("transpose_110_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_111_perm_0 = const()[name = string("transpose_111_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 126]> transpose_111 = transpose(perm = transpose_111_perm_0, x = k_29_cast_fp16)[name = string("transpose_261")];
+            tensor<fp16, [1, 8, 126, 128]> transpose_110 = transpose(perm = transpose_110_perm_0, x = var_1405_cast_fp16)[name = string("transpose_262")];
+            tensor<fp16, [1, 8, 126, 126]> matrix_ac_15_cast_fp16 = matmul(transpose_x = matrix_ac_15_transpose_x_0, transpose_y = matrix_ac_15_transpose_y_0, x = transpose_110, y = transpose_111)[name = string("matrix_ac_15_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_31_begin_0 = const()[name = string("matrix_bd_31_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_31_end_0 = const()[name = string("matrix_bd_31_end_0"), val = tensor<int32, [4]>([1, 8, 126, 126])];
+            tensor<bool, [4]> matrix_bd_31_end_mask_0 = const()[name = string("matrix_bd_31_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 126, 126]> matrix_bd_31_cast_fp16 = slice_by_index(begin = matrix_bd_31_begin_0, end = matrix_bd_31_end_0, end_mask = matrix_bd_31_end_mask_0, x = matrix_bd_29_cast_fp16)[name = string("matrix_bd_31_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_1431_cast_fp16 = add(x = matrix_ac_15_cast_fp16, y = matrix_bd_31_cast_fp16)[name = string("op_1431_cast_fp16")];
+            fp16 _inversed_scores_29_y_0_to_fp16 = const()[name = string("_inversed_scores_29_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 126, 126]> _inversed_scores_29_cast_fp16 = mul(x = var_1431_cast_fp16, y = _inversed_scores_29_y_0_to_fp16)[name = string("_inversed_scores_29_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> scores_31_cast_fp16 = select(a = var_7_to_fp16, b = _inversed_scores_29_cast_fp16, cond = mask_3)[name = string("scores_31_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_1437_cast_fp16 = softmax(axis = var_25, x = scores_31_cast_fp16)[name = string("op_1437_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> input_397_cast_fp16 = select(a = var_6_to_fp16, b = var_1437_cast_fp16, cond = mask_3)[name = string("input_397_cast_fp16")];
+            bool x_167_transpose_x_0 = const()[name = string("x_167_transpose_x_0"), val = bool(false)];
+            bool x_167_transpose_y_0 = const()[name = string("x_167_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 126, 128]> value_15_cast_fp16 = transpose(perm = value_15_perm_0, x = v_15_cast_fp16)[name = string("transpose_260")];
+            tensor<fp16, [1, 8, 126, 128]> x_167_cast_fp16 = matmul(transpose_x = x_167_transpose_x_0, transpose_y = x_167_transpose_y_0, x = input_397_cast_fp16, y = value_15_cast_fp16)[name = string("x_167_cast_fp16")];
+            tensor<int32, [4]> var_1441_perm_0 = const()[name = string("op_1441_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_1442 = const()[name = string("op_1442"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 126, 8, 128]> var_1441_cast_fp16 = transpose(perm = var_1441_perm_0, x = x_167_cast_fp16)[name = string("transpose_259")];
+            tensor<fp16, [1, 126, 1024]> input_399_cast_fp16 = reshape(shape = var_1442, x = var_1441_cast_fp16)[name = string("input_399_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_7_self_attn_linear_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96860480))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(97384832))))[name = string("model_layers_7_self_attn_linear_out_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_70_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_7_self_attn_linear_out_weight_to_fp16_palettized, x = input_399_cast_fp16)[name = string("linear_70_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_403_cast_fp16 = add(x = input_395_cast_fp16, y = linear_70_cast_fp16)[name = string("input_403_cast_fp16")];
+            tensor<int32, [1]> x_171_axes_0 = const()[name = string("x_171_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_7_norm_conv_weight_to_fp16 = const()[name = string("model_layers_7_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(97385920)))];
+            tensor<fp16, [1024]> model_layers_7_norm_conv_bias_to_fp16 = const()[name = string("model_layers_7_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(97388032)))];
+            tensor<fp16, [1, 126, 1024]> x_171_cast_fp16 = layer_norm(axes = x_171_axes_0, beta = model_layers_7_norm_conv_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_7_norm_conv_weight_to_fp16, x = input_403_cast_fp16)[name = string("x_171_cast_fp16")];
+            tensor<int32, [3]> input_405_perm_0 = const()[name = string("input_405_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_407_pad_type_0 = const()[name = string("input_407_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_407_strides_0 = const()[name = string("input_407_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_407_pad_0 = const()[name = string("input_407_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_407_dilations_0 = const()[name = string("input_407_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_407_groups_0 = const()[name = string("input_407_groups_0"), val = int32(1)];
+            tensor<fp16, [2048, 1024, 1]> model_layers_7_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(97390144))), lut = tensor<fp16, [64, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(98438784))))[name = string("model_layers_7_conv_pointwise_conv1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 1024, 126]> input_405_cast_fp16 = transpose(perm = input_405_perm_0, x = x_171_cast_fp16)[name = string("transpose_258")];
+            tensor<fp16, [1, 2048, 126]> input_407_cast_fp16 = conv(dilations = input_407_dilations_0, groups = input_407_groups_0, pad = input_407_pad_0, pad_type = input_407_pad_type_0, strides = input_407_strides_0, weight = model_layers_7_conv_pointwise_conv1_weight_to_fp16_palettized, x = input_405_cast_fp16)[name = string("input_407_cast_fp16")];
+            int32 x_173_split_num_splits_0 = const()[name = string("x_173_split_num_splits_0"), val = int32(2)];
+            int32 x_173_split_axis_0 = const()[name = string("x_173_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 126]> x_173_split_cast_fp16_0, tensor<fp16, [1, 1024, 126]> x_173_split_cast_fp16_1 = split(axis = x_173_split_axis_0, num_splits = x_173_split_num_splits_0, x = input_407_cast_fp16)[name = string("x_173_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_173_split_1_sigmoid_cast_fp16 = sigmoid(x = x_173_split_cast_fp16_1)[name = string("x_173_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_173_cast_fp16 = mul(x = x_173_split_cast_fp16_0, y = x_173_split_1_sigmoid_cast_fp16)[name = string("x_173_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_409_cast_fp16 = select(a = var_6_to_fp16, b = x_173_cast_fp16, cond = var_323)[name = string("input_409_cast_fp16")];
+            tensor<int32, [6]> input_411_pad_0 = const()[name = string("input_411_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_411_mode_0 = const()[name = string("input_411_mode_0"), val = string("constant")];
+            fp16 const_87_to_fp16 = const()[name = string("const_87_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 134]> input_411_cast_fp16 = pad(constant_val = const_87_to_fp16, mode = input_411_mode_0, pad = input_411_pad_0, x = input_409_cast_fp16)[name = string("input_411_cast_fp16")];
+            string input_413_pad_type_0 = const()[name = string("input_413_pad_type_0"), val = string("valid")];
+            int32 input_413_groups_0 = const()[name = string("input_413_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_413_strides_0 = const()[name = string("input_413_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_413_pad_0 = const()[name = string("input_413_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_413_dilations_0 = const()[name = string("input_413_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_262_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(98440896))), lut = tensor<fp16, [32, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(98445568))))[name = string("const_262_to_fp16_palettized")];
+            tensor<fp16, [1024]> const_263_to_fp16 = const()[name = string("const_263_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(98446656)))];
+            tensor<fp16, [1, 1024, 126]> input_415_cast_fp16 = conv(bias = const_263_to_fp16, dilations = input_413_dilations_0, groups = input_413_groups_0, pad = input_413_pad_0, pad_type = input_413_pad_type_0, strides = input_413_strides_0, weight = const_262_to_fp16_palettized, x = input_411_cast_fp16)[name = string("input_415_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_417_cast_fp16 = silu(x = input_415_cast_fp16)[name = string("input_417_cast_fp16")];
+            string x_175_pad_type_0 = const()[name = string("x_175_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_175_strides_0 = const()[name = string("x_175_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_175_pad_0 = const()[name = string("x_175_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_175_dilations_0 = const()[name = string("x_175_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_175_groups_0 = const()[name = string("x_175_groups_0"), val = int32(1)];
+            tensor<fp16, [1024, 1024, 1]> model_layers_7_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(98448768))), lut = tensor<fp16, [32, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(98973120))))[name = string("model_layers_7_conv_pointwise_conv2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 1024, 126]> x_175_cast_fp16 = conv(dilations = x_175_dilations_0, groups = x_175_groups_0, pad = x_175_pad_0, pad_type = x_175_pad_type_0, strides = x_175_strides_0, weight = model_layers_7_conv_pointwise_conv2_weight_to_fp16_palettized, x = input_417_cast_fp16)[name = string("x_175_cast_fp16")];
+            tensor<int32, [3]> input_419_perm_0 = const()[name = string("input_419_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 126, 1024]> input_419_cast_fp16 = transpose(perm = input_419_perm_0, x = x_175_cast_fp16)[name = string("transpose_257")];
+            tensor<fp16, [1, 126, 1024]> input_421_cast_fp16 = add(x = input_403_cast_fp16, y = input_419_cast_fp16)[name = string("input_421_cast_fp16")];
+            tensor<int32, [1]> input_423_axes_0 = const()[name = string("input_423_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_7_norm_feed_forward2_weight_to_fp16 = const()[name = string("model_layers_7_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(98974208)))];
+            tensor<fp16, [1024]> model_layers_7_norm_feed_forward2_bias_to_fp16 = const()[name = string("model_layers_7_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(98976320)))];
+            tensor<fp16, [1, 126, 1024]> input_423_cast_fp16 = layer_norm(axes = input_423_axes_0, beta = model_layers_7_norm_feed_forward2_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_7_norm_feed_forward2_weight_to_fp16, x = input_421_cast_fp16)[name = string("input_423_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_7_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(98978432))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101075648))))[name = string("model_layers_7_feed_forward2_linear1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 4096]> linear_71_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_7_feed_forward2_linear1_weight_to_fp16_palettized, x = input_423_cast_fp16)[name = string("linear_71_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_427_cast_fp16 = silu(x = linear_71_cast_fp16)[name = string("input_427_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_7_feed_forward2_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101079808))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103177024))))[name = string("model_layers_7_feed_forward2_linear2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_72_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_7_feed_forward2_linear2_weight_to_fp16_palettized, x = input_427_cast_fp16)[name = string("linear_72_cast_fp16")];
+            fp16 var_1502_to_fp16 = const()[name = string("op_1502_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_1503_cast_fp16 = mul(x = linear_72_cast_fp16, y = var_1502_to_fp16)[name = string("op_1503_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_433_cast_fp16 = add(x = input_421_cast_fp16, y = var_1503_cast_fp16)[name = string("input_433_cast_fp16")];
+            tensor<int32, [1]> input_435_axes_0 = const()[name = string("input_435_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_7_norm_out_weight_to_fp16 = const()[name = string("model_layers_7_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103178112)))];
+            tensor<fp16, [1024]> model_layers_7_norm_out_bias_to_fp16 = const()[name = string("model_layers_7_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103180224)))];
+            tensor<fp16, [1, 126, 1024]> input_435_cast_fp16 = layer_norm(axes = input_435_axes_0, beta = model_layers_7_norm_out_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_7_norm_out_weight_to_fp16, x = input_433_cast_fp16)[name = string("input_435_cast_fp16")];
+            tensor<int32, [1]> input_437_axes_0 = const()[name = string("input_437_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_8_norm_feed_forward1_weight_to_fp16 = const()[name = string("model_layers_8_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103182336)))];
+            tensor<fp16, [1024]> model_layers_8_norm_feed_forward1_bias_to_fp16 = const()[name = string("model_layers_8_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103184448)))];
+            tensor<fp16, [1, 126, 1024]> input_437_cast_fp16 = layer_norm(axes = input_437_axes_0, beta = model_layers_8_norm_feed_forward1_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_8_norm_feed_forward1_weight_to_fp16, x = input_435_cast_fp16)[name = string("input_437_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_8_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103186560))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105283776))))[name = string("model_layers_8_feed_forward1_linear1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 4096]> linear_73_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_8_feed_forward1_linear1_weight_to_fp16_palettized, x = input_437_cast_fp16)[name = string("linear_73_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_441_cast_fp16 = silu(x = linear_73_cast_fp16)[name = string("input_441_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_8_feed_forward1_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105287936))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107385152))))[name = string("model_layers_8_feed_forward1_linear2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_74_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_8_feed_forward1_linear2_weight_to_fp16_palettized, x = input_441_cast_fp16)[name = string("linear_74_cast_fp16")];
+            fp16 var_1531_to_fp16 = const()[name = string("op_1531_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_1532_cast_fp16 = mul(x = linear_74_cast_fp16, y = var_1531_to_fp16)[name = string("op_1532_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_447_cast_fp16 = add(x = input_435_cast_fp16, y = var_1532_cast_fp16)[name = string("input_447_cast_fp16")];
+            tensor<int32, [1]> query_17_axes_0 = const()[name = string("query_17_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_8_norm_self_att_weight_to_fp16 = const()[name = string("model_layers_8_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107386240)))];
+            tensor<fp16, [1024]> model_layers_8_norm_self_att_bias_to_fp16 = const()[name = string("model_layers_8_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107388352)))];
+            tensor<fp16, [1, 126, 1024]> query_17_cast_fp16 = layer_norm(axes = query_17_axes_0, beta = model_layers_8_norm_self_att_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_8_norm_self_att_weight_to_fp16, x = input_447_cast_fp16)[name = string("query_17_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_8_self_attn_linear_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107390464))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107914816))))[name = string("model_layers_8_self_attn_linear_q_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_75_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_8_self_attn_linear_q_weight_to_fp16_palettized, x = query_17_cast_fp16)[name = string("linear_75_cast_fp16")];
+            tensor<int32, [4]> var_1548 = const()[name = string("op_1548"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> q_49_cast_fp16 = reshape(shape = var_1548, x = linear_75_cast_fp16)[name = string("q_49_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_8_self_attn_linear_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107915904))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108440256))))[name = string("model_layers_8_self_attn_linear_k_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_76_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_8_self_attn_linear_k_weight_to_fp16_palettized, x = query_17_cast_fp16)[name = string("linear_76_cast_fp16")];
+            tensor<int32, [4]> var_1552 = const()[name = string("op_1552"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> k_33_cast_fp16 = reshape(shape = var_1552, x = linear_76_cast_fp16)[name = string("k_33_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_8_self_attn_linear_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108441344))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108965696))))[name = string("model_layers_8_self_attn_linear_v_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_77_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_8_self_attn_linear_v_weight_to_fp16_palettized, x = query_17_cast_fp16)[name = string("linear_77_cast_fp16")];
+            tensor<int32, [4]> var_1556 = const()[name = string("op_1556"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> v_17_cast_fp16 = reshape(shape = var_1556, x = linear_77_cast_fp16)[name = string("v_17_cast_fp16")];
+            tensor<int32, [4]> value_17_perm_0 = const()[name = string("value_17_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> model_layers_8_self_attn_pos_bias_u_to_fp16 = const()[name = string("model_layers_8_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108966784)))];
+            tensor<fp16, [1, 126, 8, 128]> var_1568_cast_fp16 = add(x = q_49_cast_fp16, y = model_layers_8_self_attn_pos_bias_u_to_fp16)[name = string("op_1568_cast_fp16")];
+            tensor<fp16, [8, 128]> model_layers_8_self_attn_pos_bias_v_to_fp16 = const()[name = string("model_layers_8_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108968896)))];
+            tensor<fp16, [1, 126, 8, 128]> var_1570_cast_fp16 = add(x = q_49_cast_fp16, y = model_layers_8_self_attn_pos_bias_v_to_fp16)[name = string("op_1570_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_17_perm_0 = const()[name = string("q_with_bias_v_17_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_183_transpose_x_0 = const()[name = string("x_183_transpose_x_0"), val = bool(false)];
+            bool x_183_transpose_y_0 = const()[name = string("x_183_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 251]> var_1572_to_fp16 = const()[name = string("op_1572_to_fp16"), val = tensor<fp16, [1, 8, 128, 251]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108971008)))];
+            tensor<fp16, [1, 8, 126, 128]> q_with_bias_v_17_cast_fp16 = transpose(perm = q_with_bias_v_17_perm_0, x = var_1570_cast_fp16)[name = string("transpose_256")];
+            tensor<fp16, [1, 8, 126, 251]> x_183_cast_fp16 = matmul(transpose_x = x_183_transpose_x_0, transpose_y = x_183_transpose_y_0, x = q_with_bias_v_17_cast_fp16, y = var_1572_to_fp16)[name = string("x_183_cast_fp16")];
+            tensor<int32, [8]> x_185_pad_0 = const()[name = string("x_185_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_185_mode_0 = const()[name = string("x_185_mode_0"), val = string("constant")];
+            fp16 const_94_to_fp16 = const()[name = string("const_94_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 252]> x_185_cast_fp16 = pad(constant_val = const_94_to_fp16, mode = x_185_mode_0, pad = x_185_pad_0, x = x_183_cast_fp16)[name = string("x_185_cast_fp16")];
+            tensor<int32, [4]> var_1580 = const()[name = string("op_1580"), val = tensor<int32, [4]>([1, 8, -1, 126])];
+            tensor<fp16, [1, 8, 252, 126]> x_187_cast_fp16 = reshape(shape = var_1580, x = x_185_cast_fp16)[name = string("x_187_cast_fp16")];
+            tensor<int32, [4]> var_1584_begin_0 = const()[name = string("op_1584_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_1584_end_0 = const()[name = string("op_1584_end_0"), val = tensor<int32, [4]>([1, 8, 252, 126])];
+            tensor<bool, [4]> var_1584_end_mask_0 = const()[name = string("op_1584_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 251, 126]> var_1584_cast_fp16 = slice_by_index(begin = var_1584_begin_0, end = var_1584_end_0, end_mask = var_1584_end_mask_0, x = x_187_cast_fp16)[name = string("op_1584_cast_fp16")];
+            tensor<int32, [4]> var_1585 = const()[name = string("op_1585"), val = tensor<int32, [4]>([1, 8, 126, 251])];
+            tensor<fp16, [1, 8, 126, 251]> matrix_bd_33_cast_fp16 = reshape(shape = var_1585, x = var_1584_cast_fp16)[name = string("matrix_bd_33_cast_fp16")];
+            bool matrix_ac_17_transpose_x_0 = const()[name = string("matrix_ac_17_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_17_transpose_y_0 = const()[name = string("matrix_ac_17_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_112_perm_0 = const()[name = string("transpose_112_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_113_perm_0 = const()[name = string("transpose_113_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 126]> transpose_113 = transpose(perm = transpose_113_perm_0, x = k_33_cast_fp16)[name = string("transpose_254")];
+            tensor<fp16, [1, 8, 126, 128]> transpose_112 = transpose(perm = transpose_112_perm_0, x = var_1568_cast_fp16)[name = string("transpose_255")];
+            tensor<fp16, [1, 8, 126, 126]> matrix_ac_17_cast_fp16 = matmul(transpose_x = matrix_ac_17_transpose_x_0, transpose_y = matrix_ac_17_transpose_y_0, x = transpose_112, y = transpose_113)[name = string("matrix_ac_17_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_35_begin_0 = const()[name = string("matrix_bd_35_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_35_end_0 = const()[name = string("matrix_bd_35_end_0"), val = tensor<int32, [4]>([1, 8, 126, 126])];
+            tensor<bool, [4]> matrix_bd_35_end_mask_0 = const()[name = string("matrix_bd_35_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 126, 126]> matrix_bd_35_cast_fp16 = slice_by_index(begin = matrix_bd_35_begin_0, end = matrix_bd_35_end_0, end_mask = matrix_bd_35_end_mask_0, x = matrix_bd_33_cast_fp16)[name = string("matrix_bd_35_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_1594_cast_fp16 = add(x = matrix_ac_17_cast_fp16, y = matrix_bd_35_cast_fp16)[name = string("op_1594_cast_fp16")];
+            fp16 _inversed_scores_33_y_0_to_fp16 = const()[name = string("_inversed_scores_33_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 126, 126]> _inversed_scores_33_cast_fp16 = mul(x = var_1594_cast_fp16, y = _inversed_scores_33_y_0_to_fp16)[name = string("_inversed_scores_33_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> scores_35_cast_fp16 = select(a = var_7_to_fp16, b = _inversed_scores_33_cast_fp16, cond = mask_3)[name = string("scores_35_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_1600_cast_fp16 = softmax(axis = var_25, x = scores_35_cast_fp16)[name = string("op_1600_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> input_449_cast_fp16 = select(a = var_6_to_fp16, b = var_1600_cast_fp16, cond = mask_3)[name = string("input_449_cast_fp16")];
+            bool x_189_transpose_x_0 = const()[name = string("x_189_transpose_x_0"), val = bool(false)];
+            bool x_189_transpose_y_0 = const()[name = string("x_189_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 126, 128]> value_17_cast_fp16 = transpose(perm = value_17_perm_0, x = v_17_cast_fp16)[name = string("transpose_253")];
+            tensor<fp16, [1, 8, 126, 128]> x_189_cast_fp16 = matmul(transpose_x = x_189_transpose_x_0, transpose_y = x_189_transpose_y_0, x = input_449_cast_fp16, y = value_17_cast_fp16)[name = string("x_189_cast_fp16")];
+            tensor<int32, [4]> var_1604_perm_0 = const()[name = string("op_1604_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_1605 = const()[name = string("op_1605"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 126, 8, 128]> var_1604_cast_fp16 = transpose(perm = var_1604_perm_0, x = x_189_cast_fp16)[name = string("transpose_252")];
+            tensor<fp16, [1, 126, 1024]> input_451_cast_fp16 = reshape(shape = var_1605, x = var_1604_cast_fp16)[name = string("input_451_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_8_self_attn_linear_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(109485120))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110009472))))[name = string("model_layers_8_self_attn_linear_out_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_79_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_8_self_attn_linear_out_weight_to_fp16_palettized, x = input_451_cast_fp16)[name = string("linear_79_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_455_cast_fp16 = add(x = input_447_cast_fp16, y = linear_79_cast_fp16)[name = string("input_455_cast_fp16")];
+            tensor<int32, [1]> x_193_axes_0 = const()[name = string("x_193_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_8_norm_conv_weight_to_fp16 = const()[name = string("model_layers_8_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110010560)))];
+            tensor<fp16, [1024]> model_layers_8_norm_conv_bias_to_fp16 = const()[name = string("model_layers_8_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110012672)))];
+            tensor<fp16, [1, 126, 1024]> x_193_cast_fp16 = layer_norm(axes = x_193_axes_0, beta = model_layers_8_norm_conv_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_8_norm_conv_weight_to_fp16, x = input_455_cast_fp16)[name = string("x_193_cast_fp16")];
+            tensor<int32, [3]> input_457_perm_0 = const()[name = string("input_457_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_459_pad_type_0 = const()[name = string("input_459_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_459_strides_0 = const()[name = string("input_459_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_459_pad_0 = const()[name = string("input_459_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_459_dilations_0 = const()[name = string("input_459_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_459_groups_0 = const()[name = string("input_459_groups_0"), val = int32(1)];
+            tensor<fp16, [2048, 1024, 1]> model_layers_8_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110014784))), lut = tensor<fp16, [64, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111063424))))[name = string("model_layers_8_conv_pointwise_conv1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 1024, 126]> input_457_cast_fp16 = transpose(perm = input_457_perm_0, x = x_193_cast_fp16)[name = string("transpose_251")];
+            tensor<fp16, [1, 2048, 126]> input_459_cast_fp16 = conv(dilations = input_459_dilations_0, groups = input_459_groups_0, pad = input_459_pad_0, pad_type = input_459_pad_type_0, strides = input_459_strides_0, weight = model_layers_8_conv_pointwise_conv1_weight_to_fp16_palettized, x = input_457_cast_fp16)[name = string("input_459_cast_fp16")];
+            int32 x_195_split_num_splits_0 = const()[name = string("x_195_split_num_splits_0"), val = int32(2)];
+            int32 x_195_split_axis_0 = const()[name = string("x_195_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 126]> x_195_split_cast_fp16_0, tensor<fp16, [1, 1024, 126]> x_195_split_cast_fp16_1 = split(axis = x_195_split_axis_0, num_splits = x_195_split_num_splits_0, x = input_459_cast_fp16)[name = string("x_195_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_195_split_1_sigmoid_cast_fp16 = sigmoid(x = x_195_split_cast_fp16_1)[name = string("x_195_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_195_cast_fp16 = mul(x = x_195_split_cast_fp16_0, y = x_195_split_1_sigmoid_cast_fp16)[name = string("x_195_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_461_cast_fp16 = select(a = var_6_to_fp16, b = x_195_cast_fp16, cond = var_323)[name = string("input_461_cast_fp16")];
+            tensor<int32, [6]> input_463_pad_0 = const()[name = string("input_463_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_463_mode_0 = const()[name = string("input_463_mode_0"), val = string("constant")];
+            fp16 const_97_to_fp16 = const()[name = string("const_97_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 134]> input_463_cast_fp16 = pad(constant_val = const_97_to_fp16, mode = input_463_mode_0, pad = input_463_pad_0, x = input_461_cast_fp16)[name = string("input_463_cast_fp16")];
+            string input_465_pad_type_0 = const()[name = string("input_465_pad_type_0"), val = string("valid")];
+            int32 input_465_groups_0 = const()[name = string("input_465_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_465_strides_0 = const()[name = string("input_465_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_465_pad_0 = const()[name = string("input_465_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_465_dilations_0 = const()[name = string("input_465_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_264_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111065536))), lut = tensor<fp16, [32, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111070208))))[name = string("const_264_to_fp16_palettized")];
+            tensor<fp16, [1024]> const_265_to_fp16 = const()[name = string("const_265_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111071296)))];
+            tensor<fp16, [1, 1024, 126]> input_467_cast_fp16 = conv(bias = const_265_to_fp16, dilations = input_465_dilations_0, groups = input_465_groups_0, pad = input_465_pad_0, pad_type = input_465_pad_type_0, strides = input_465_strides_0, weight = const_264_to_fp16_palettized, x = input_463_cast_fp16)[name = string("input_467_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_469_cast_fp16 = silu(x = input_467_cast_fp16)[name = string("input_469_cast_fp16")];
+            string x_197_pad_type_0 = const()[name = string("x_197_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_197_strides_0 = const()[name = string("x_197_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_197_pad_0 = const()[name = string("x_197_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_197_dilations_0 = const()[name = string("x_197_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_197_groups_0 = const()[name = string("x_197_groups_0"), val = int32(1)];
+            tensor<fp16, [1024, 1024, 1]> model_layers_8_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111073408))), lut = tensor<fp16, [32, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111597760))))[name = string("model_layers_8_conv_pointwise_conv2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 1024, 126]> x_197_cast_fp16 = conv(dilations = x_197_dilations_0, groups = x_197_groups_0, pad = x_197_pad_0, pad_type = x_197_pad_type_0, strides = x_197_strides_0, weight = model_layers_8_conv_pointwise_conv2_weight_to_fp16_palettized, x = input_469_cast_fp16)[name = string("x_197_cast_fp16")];
+            tensor<int32, [3]> input_471_perm_0 = const()[name = string("input_471_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 126, 1024]> input_471_cast_fp16 = transpose(perm = input_471_perm_0, x = x_197_cast_fp16)[name = string("transpose_250")];
+            tensor<fp16, [1, 126, 1024]> input_473_cast_fp16 = add(x = input_455_cast_fp16, y = input_471_cast_fp16)[name = string("input_473_cast_fp16")];
+            tensor<int32, [1]> input_475_axes_0 = const()[name = string("input_475_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_8_norm_feed_forward2_weight_to_fp16 = const()[name = string("model_layers_8_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111598848)))];
+            tensor<fp16, [1024]> model_layers_8_norm_feed_forward2_bias_to_fp16 = const()[name = string("model_layers_8_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111600960)))];
+            tensor<fp16, [1, 126, 1024]> input_475_cast_fp16 = layer_norm(axes = input_475_axes_0, beta = model_layers_8_norm_feed_forward2_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_8_norm_feed_forward2_weight_to_fp16, x = input_473_cast_fp16)[name = string("input_475_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_8_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111603072))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113700288))))[name = string("model_layers_8_feed_forward2_linear1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 4096]> linear_80_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_8_feed_forward2_linear1_weight_to_fp16_palettized, x = input_475_cast_fp16)[name = string("linear_80_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_479_cast_fp16 = silu(x = linear_80_cast_fp16)[name = string("input_479_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_8_feed_forward2_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113704448))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115801664))))[name = string("model_layers_8_feed_forward2_linear2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_81_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_8_feed_forward2_linear2_weight_to_fp16_palettized, x = input_479_cast_fp16)[name = string("linear_81_cast_fp16")];
+            fp16 var_1665_to_fp16 = const()[name = string("op_1665_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_1666_cast_fp16 = mul(x = linear_81_cast_fp16, y = var_1665_to_fp16)[name = string("op_1666_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_485_cast_fp16 = add(x = input_473_cast_fp16, y = var_1666_cast_fp16)[name = string("input_485_cast_fp16")];
+            tensor<int32, [1]> input_487_axes_0 = const()[name = string("input_487_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_8_norm_out_weight_to_fp16 = const()[name = string("model_layers_8_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115802752)))];
+            tensor<fp16, [1024]> model_layers_8_norm_out_bias_to_fp16 = const()[name = string("model_layers_8_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115804864)))];
+            tensor<fp16, [1, 126, 1024]> input_487_cast_fp16 = layer_norm(axes = input_487_axes_0, beta = model_layers_8_norm_out_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_8_norm_out_weight_to_fp16, x = input_485_cast_fp16)[name = string("input_487_cast_fp16")];
+            tensor<int32, [1]> input_489_axes_0 = const()[name = string("input_489_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_9_norm_feed_forward1_weight_to_fp16 = const()[name = string("model_layers_9_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115806976)))];
+            tensor<fp16, [1024]> model_layers_9_norm_feed_forward1_bias_to_fp16 = const()[name = string("model_layers_9_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115809088)))];
+            tensor<fp16, [1, 126, 1024]> input_489_cast_fp16 = layer_norm(axes = input_489_axes_0, beta = model_layers_9_norm_feed_forward1_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_9_norm_feed_forward1_weight_to_fp16, x = input_487_cast_fp16)[name = string("input_489_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_9_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115811200))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117908416))))[name = string("model_layers_9_feed_forward1_linear1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 4096]> linear_82_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_9_feed_forward1_linear1_weight_to_fp16_palettized, x = input_489_cast_fp16)[name = string("linear_82_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_493_cast_fp16 = silu(x = linear_82_cast_fp16)[name = string("input_493_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_9_feed_forward1_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117912576))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120009792))))[name = string("model_layers_9_feed_forward1_linear2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_83_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_9_feed_forward1_linear2_weight_to_fp16_palettized, x = input_493_cast_fp16)[name = string("linear_83_cast_fp16")];
+            fp16 var_1694_to_fp16 = const()[name = string("op_1694_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_1695_cast_fp16 = mul(x = linear_83_cast_fp16, y = var_1694_to_fp16)[name = string("op_1695_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_499_cast_fp16 = add(x = input_487_cast_fp16, y = var_1695_cast_fp16)[name = string("input_499_cast_fp16")];
+            tensor<int32, [1]> query_19_axes_0 = const()[name = string("query_19_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_9_norm_self_att_weight_to_fp16 = const()[name = string("model_layers_9_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120010880)))];
+            tensor<fp16, [1024]> model_layers_9_norm_self_att_bias_to_fp16 = const()[name = string("model_layers_9_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120012992)))];
+            tensor<fp16, [1, 126, 1024]> query_19_cast_fp16 = layer_norm(axes = query_19_axes_0, beta = model_layers_9_norm_self_att_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_9_norm_self_att_weight_to_fp16, x = input_499_cast_fp16)[name = string("query_19_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_9_self_attn_linear_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120015104))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120539456))))[name = string("model_layers_9_self_attn_linear_q_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_84_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_9_self_attn_linear_q_weight_to_fp16_palettized, x = query_19_cast_fp16)[name = string("linear_84_cast_fp16")];
+            tensor<int32, [4]> var_1711 = const()[name = string("op_1711"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> q_55_cast_fp16 = reshape(shape = var_1711, x = linear_84_cast_fp16)[name = string("q_55_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_9_self_attn_linear_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120540544))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(121064896))))[name = string("model_layers_9_self_attn_linear_k_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_85_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_9_self_attn_linear_k_weight_to_fp16_palettized, x = query_19_cast_fp16)[name = string("linear_85_cast_fp16")];
+            tensor<int32, [4]> var_1715 = const()[name = string("op_1715"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> k_37_cast_fp16 = reshape(shape = var_1715, x = linear_85_cast_fp16)[name = string("k_37_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_9_self_attn_linear_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(121065984))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(121590336))))[name = string("model_layers_9_self_attn_linear_v_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_86_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_9_self_attn_linear_v_weight_to_fp16_palettized, x = query_19_cast_fp16)[name = string("linear_86_cast_fp16")];
+            tensor<int32, [4]> var_1719 = const()[name = string("op_1719"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> v_19_cast_fp16 = reshape(shape = var_1719, x = linear_86_cast_fp16)[name = string("v_19_cast_fp16")];
+            tensor<int32, [4]> value_19_perm_0 = const()[name = string("value_19_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> model_layers_9_self_attn_pos_bias_u_to_fp16 = const()[name = string("model_layers_9_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(121591424)))];
+            tensor<fp16, [1, 126, 8, 128]> var_1731_cast_fp16 = add(x = q_55_cast_fp16, y = model_layers_9_self_attn_pos_bias_u_to_fp16)[name = string("op_1731_cast_fp16")];
+            tensor<fp16, [8, 128]> model_layers_9_self_attn_pos_bias_v_to_fp16 = const()[name = string("model_layers_9_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(121593536)))];
+            tensor<fp16, [1, 126, 8, 128]> var_1733_cast_fp16 = add(x = q_55_cast_fp16, y = model_layers_9_self_attn_pos_bias_v_to_fp16)[name = string("op_1733_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_19_perm_0 = const()[name = string("q_with_bias_v_19_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_205_transpose_x_0 = const()[name = string("x_205_transpose_x_0"), val = bool(false)];
+            bool x_205_transpose_y_0 = const()[name = string("x_205_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 251]> var_1735_to_fp16 = const()[name = string("op_1735_to_fp16"), val = tensor<fp16, [1, 8, 128, 251]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(121595648)))];
+            tensor<fp16, [1, 8, 126, 128]> q_with_bias_v_19_cast_fp16 = transpose(perm = q_with_bias_v_19_perm_0, x = var_1733_cast_fp16)[name = string("transpose_249")];
+            tensor<fp16, [1, 8, 126, 251]> x_205_cast_fp16 = matmul(transpose_x = x_205_transpose_x_0, transpose_y = x_205_transpose_y_0, x = q_with_bias_v_19_cast_fp16, y = var_1735_to_fp16)[name = string("x_205_cast_fp16")];
+            tensor<int32, [8]> x_207_pad_0 = const()[name = string("x_207_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_207_mode_0 = const()[name = string("x_207_mode_0"), val = string("constant")];
+            fp16 const_104_to_fp16 = const()[name = string("const_104_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 252]> x_207_cast_fp16 = pad(constant_val = const_104_to_fp16, mode = x_207_mode_0, pad = x_207_pad_0, x = x_205_cast_fp16)[name = string("x_207_cast_fp16")];
+            tensor<int32, [4]> var_1743 = const()[name = string("op_1743"), val = tensor<int32, [4]>([1, 8, -1, 126])];
+            tensor<fp16, [1, 8, 252, 126]> x_209_cast_fp16 = reshape(shape = var_1743, x = x_207_cast_fp16)[name = string("x_209_cast_fp16")];
+            tensor<int32, [4]> var_1747_begin_0 = const()[name = string("op_1747_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_1747_end_0 = const()[name = string("op_1747_end_0"), val = tensor<int32, [4]>([1, 8, 252, 126])];
+            tensor<bool, [4]> var_1747_end_mask_0 = const()[name = string("op_1747_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 251, 126]> var_1747_cast_fp16 = slice_by_index(begin = var_1747_begin_0, end = var_1747_end_0, end_mask = var_1747_end_mask_0, x = x_209_cast_fp16)[name = string("op_1747_cast_fp16")];
+            tensor<int32, [4]> var_1748 = const()[name = string("op_1748"), val = tensor<int32, [4]>([1, 8, 126, 251])];
+            tensor<fp16, [1, 8, 126, 251]> matrix_bd_37_cast_fp16 = reshape(shape = var_1748, x = var_1747_cast_fp16)[name = string("matrix_bd_37_cast_fp16")];
+            bool matrix_ac_19_transpose_x_0 = const()[name = string("matrix_ac_19_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_19_transpose_y_0 = const()[name = string("matrix_ac_19_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_114_perm_0 = const()[name = string("transpose_114_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_115_perm_0 = const()[name = string("transpose_115_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 126]> transpose_115 = transpose(perm = transpose_115_perm_0, x = k_37_cast_fp16)[name = string("transpose_247")];
+            tensor<fp16, [1, 8, 126, 128]> transpose_114 = transpose(perm = transpose_114_perm_0, x = var_1731_cast_fp16)[name = string("transpose_248")];
+            tensor<fp16, [1, 8, 126, 126]> matrix_ac_19_cast_fp16 = matmul(transpose_x = matrix_ac_19_transpose_x_0, transpose_y = matrix_ac_19_transpose_y_0, x = transpose_114, y = transpose_115)[name = string("matrix_ac_19_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_39_begin_0 = const()[name = string("matrix_bd_39_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_39_end_0 = const()[name = string("matrix_bd_39_end_0"), val = tensor<int32, [4]>([1, 8, 126, 126])];
+            tensor<bool, [4]> matrix_bd_39_end_mask_0 = const()[name = string("matrix_bd_39_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 126, 126]> matrix_bd_39_cast_fp16 = slice_by_index(begin = matrix_bd_39_begin_0, end = matrix_bd_39_end_0, end_mask = matrix_bd_39_end_mask_0, x = matrix_bd_37_cast_fp16)[name = string("matrix_bd_39_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_1757_cast_fp16 = add(x = matrix_ac_19_cast_fp16, y = matrix_bd_39_cast_fp16)[name = string("op_1757_cast_fp16")];
+            fp16 _inversed_scores_37_y_0_to_fp16 = const()[name = string("_inversed_scores_37_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 126, 126]> _inversed_scores_37_cast_fp16 = mul(x = var_1757_cast_fp16, y = _inversed_scores_37_y_0_to_fp16)[name = string("_inversed_scores_37_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> scores_39_cast_fp16 = select(a = var_7_to_fp16, b = _inversed_scores_37_cast_fp16, cond = mask_3)[name = string("scores_39_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_1763_cast_fp16 = softmax(axis = var_25, x = scores_39_cast_fp16)[name = string("op_1763_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> input_501_cast_fp16 = select(a = var_6_to_fp16, b = var_1763_cast_fp16, cond = mask_3)[name = string("input_501_cast_fp16")];
+            bool x_211_transpose_x_0 = const()[name = string("x_211_transpose_x_0"), val = bool(false)];
+            bool x_211_transpose_y_0 = const()[name = string("x_211_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 126, 128]> value_19_cast_fp16 = transpose(perm = value_19_perm_0, x = v_19_cast_fp16)[name = string("transpose_246")];
+            tensor<fp16, [1, 8, 126, 128]> x_211_cast_fp16 = matmul(transpose_x = x_211_transpose_x_0, transpose_y = x_211_transpose_y_0, x = input_501_cast_fp16, y = value_19_cast_fp16)[name = string("x_211_cast_fp16")];
+            tensor<int32, [4]> var_1767_perm_0 = const()[name = string("op_1767_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_1768 = const()[name = string("op_1768"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 126, 8, 128]> var_1767_cast_fp16 = transpose(perm = var_1767_perm_0, x = x_211_cast_fp16)[name = string("transpose_245")];
+            tensor<fp16, [1, 126, 1024]> input_503_cast_fp16 = reshape(shape = var_1768, x = var_1767_cast_fp16)[name = string("input_503_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_9_self_attn_linear_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122109760))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122634112))))[name = string("model_layers_9_self_attn_linear_out_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_88_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_9_self_attn_linear_out_weight_to_fp16_palettized, x = input_503_cast_fp16)[name = string("linear_88_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_507_cast_fp16 = add(x = input_499_cast_fp16, y = linear_88_cast_fp16)[name = string("input_507_cast_fp16")];
+            tensor<int32, [1]> x_215_axes_0 = const()[name = string("x_215_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_9_norm_conv_weight_to_fp16 = const()[name = string("model_layers_9_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122635200)))];
+            tensor<fp16, [1024]> model_layers_9_norm_conv_bias_to_fp16 = const()[name = string("model_layers_9_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122637312)))];
+            tensor<fp16, [1, 126, 1024]> x_215_cast_fp16 = layer_norm(axes = x_215_axes_0, beta = model_layers_9_norm_conv_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_9_norm_conv_weight_to_fp16, x = input_507_cast_fp16)[name = string("x_215_cast_fp16")];
+            tensor<int32, [3]> input_509_perm_0 = const()[name = string("input_509_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_511_pad_type_0 = const()[name = string("input_511_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_511_strides_0 = const()[name = string("input_511_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_511_pad_0 = const()[name = string("input_511_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_511_dilations_0 = const()[name = string("input_511_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_511_groups_0 = const()[name = string("input_511_groups_0"), val = int32(1)];
+            tensor<fp16, [2048, 1024, 1]> model_layers_9_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122639424))), lut = tensor<fp16, [64, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123688064))))[name = string("model_layers_9_conv_pointwise_conv1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 1024, 126]> input_509_cast_fp16 = transpose(perm = input_509_perm_0, x = x_215_cast_fp16)[name = string("transpose_244")];
+            tensor<fp16, [1, 2048, 126]> input_511_cast_fp16 = conv(dilations = input_511_dilations_0, groups = input_511_groups_0, pad = input_511_pad_0, pad_type = input_511_pad_type_0, strides = input_511_strides_0, weight = model_layers_9_conv_pointwise_conv1_weight_to_fp16_palettized, x = input_509_cast_fp16)[name = string("input_511_cast_fp16")];
+            int32 x_217_split_num_splits_0 = const()[name = string("x_217_split_num_splits_0"), val = int32(2)];
+            int32 x_217_split_axis_0 = const()[name = string("x_217_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 126]> x_217_split_cast_fp16_0, tensor<fp16, [1, 1024, 126]> x_217_split_cast_fp16_1 = split(axis = x_217_split_axis_0, num_splits = x_217_split_num_splits_0, x = input_511_cast_fp16)[name = string("x_217_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_217_split_1_sigmoid_cast_fp16 = sigmoid(x = x_217_split_cast_fp16_1)[name = string("x_217_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_217_cast_fp16 = mul(x = x_217_split_cast_fp16_0, y = x_217_split_1_sigmoid_cast_fp16)[name = string("x_217_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_513_cast_fp16 = select(a = var_6_to_fp16, b = x_217_cast_fp16, cond = var_323)[name = string("input_513_cast_fp16")];
+            tensor<int32, [6]> input_515_pad_0 = const()[name = string("input_515_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_515_mode_0 = const()[name = string("input_515_mode_0"), val = string("constant")];
+            fp16 const_107_to_fp16 = const()[name = string("const_107_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 134]> input_515_cast_fp16 = pad(constant_val = const_107_to_fp16, mode = input_515_mode_0, pad = input_515_pad_0, x = input_513_cast_fp16)[name = string("input_515_cast_fp16")];
+            string input_517_pad_type_0 = const()[name = string("input_517_pad_type_0"), val = string("valid")];
+            int32 input_517_groups_0 = const()[name = string("input_517_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_517_strides_0 = const()[name = string("input_517_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_517_pad_0 = const()[name = string("input_517_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_517_dilations_0 = const()[name = string("input_517_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_266_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123690176))), lut = tensor<fp16, [32, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123694848))))[name = string("const_266_to_fp16_palettized")];
+            tensor<fp16, [1024]> const_267_to_fp16 = const()[name = string("const_267_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123695936)))];
+            tensor<fp16, [1, 1024, 126]> input_519_cast_fp16 = conv(bias = const_267_to_fp16, dilations = input_517_dilations_0, groups = input_517_groups_0, pad = input_517_pad_0, pad_type = input_517_pad_type_0, strides = input_517_strides_0, weight = const_266_to_fp16_palettized, x = input_515_cast_fp16)[name = string("input_519_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_521_cast_fp16 = silu(x = input_519_cast_fp16)[name = string("input_521_cast_fp16")];
+            string x_219_pad_type_0 = const()[name = string("x_219_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_219_strides_0 = const()[name = string("x_219_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_219_pad_0 = const()[name = string("x_219_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_219_dilations_0 = const()[name = string("x_219_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_219_groups_0 = const()[name = string("x_219_groups_0"), val = int32(1)];
+            tensor<fp16, [1024, 1024, 1]> model_layers_9_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123698048))), lut = tensor<fp16, [32, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124222400))))[name = string("model_layers_9_conv_pointwise_conv2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 1024, 126]> x_219_cast_fp16 = conv(dilations = x_219_dilations_0, groups = x_219_groups_0, pad = x_219_pad_0, pad_type = x_219_pad_type_0, strides = x_219_strides_0, weight = model_layers_9_conv_pointwise_conv2_weight_to_fp16_palettized, x = input_521_cast_fp16)[name = string("x_219_cast_fp16")];
+            tensor<int32, [3]> input_523_perm_0 = const()[name = string("input_523_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 126, 1024]> input_523_cast_fp16 = transpose(perm = input_523_perm_0, x = x_219_cast_fp16)[name = string("transpose_243")];
+            tensor<fp16, [1, 126, 1024]> input_525_cast_fp16 = add(x = input_507_cast_fp16, y = input_523_cast_fp16)[name = string("input_525_cast_fp16")];
+            tensor<int32, [1]> input_527_axes_0 = const()[name = string("input_527_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_9_norm_feed_forward2_weight_to_fp16 = const()[name = string("model_layers_9_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124223488)))];
+            tensor<fp16, [1024]> model_layers_9_norm_feed_forward2_bias_to_fp16 = const()[name = string("model_layers_9_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124225600)))];
+            tensor<fp16, [1, 126, 1024]> input_527_cast_fp16 = layer_norm(axes = input_527_axes_0, beta = model_layers_9_norm_feed_forward2_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_9_norm_feed_forward2_weight_to_fp16, x = input_525_cast_fp16)[name = string("input_527_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_9_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124227712))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126324928))))[name = string("model_layers_9_feed_forward2_linear1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 4096]> linear_89_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_9_feed_forward2_linear1_weight_to_fp16_palettized, x = input_527_cast_fp16)[name = string("linear_89_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_531_cast_fp16 = silu(x = linear_89_cast_fp16)[name = string("input_531_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_9_feed_forward2_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126329088))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128426304))))[name = string("model_layers_9_feed_forward2_linear2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_90_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_9_feed_forward2_linear2_weight_to_fp16_palettized, x = input_531_cast_fp16)[name = string("linear_90_cast_fp16")];
+            fp16 var_1828_to_fp16 = const()[name = string("op_1828_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_1829_cast_fp16 = mul(x = linear_90_cast_fp16, y = var_1828_to_fp16)[name = string("op_1829_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_537_cast_fp16 = add(x = input_525_cast_fp16, y = var_1829_cast_fp16)[name = string("input_537_cast_fp16")];
+            tensor<int32, [1]> input_539_axes_0 = const()[name = string("input_539_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_9_norm_out_weight_to_fp16 = const()[name = string("model_layers_9_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128427392)))];
+            tensor<fp16, [1024]> model_layers_9_norm_out_bias_to_fp16 = const()[name = string("model_layers_9_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128429504)))];
+            tensor<fp16, [1, 126, 1024]> input_539_cast_fp16 = layer_norm(axes = input_539_axes_0, beta = model_layers_9_norm_out_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_9_norm_out_weight_to_fp16, x = input_537_cast_fp16)[name = string("input_539_cast_fp16")];
+            tensor<int32, [1]> input_541_axes_0 = const()[name = string("input_541_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_10_norm_feed_forward1_weight_to_fp16 = const()[name = string("model_layers_10_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128431616)))];
+            tensor<fp16, [1024]> model_layers_10_norm_feed_forward1_bias_to_fp16 = const()[name = string("model_layers_10_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128433728)))];
+            tensor<fp16, [1, 126, 1024]> input_541_cast_fp16 = layer_norm(axes = input_541_axes_0, beta = model_layers_10_norm_feed_forward1_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_10_norm_feed_forward1_weight_to_fp16, x = input_539_cast_fp16)[name = string("input_541_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_10_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128435840))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(130533056))))[name = string("model_layers_10_feed_forward1_linear1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 4096]> linear_91_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_10_feed_forward1_linear1_weight_to_fp16_palettized, x = input_541_cast_fp16)[name = string("linear_91_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_545_cast_fp16 = silu(x = linear_91_cast_fp16)[name = string("input_545_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_10_feed_forward1_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(130537216))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132634432))))[name = string("model_layers_10_feed_forward1_linear2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_92_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_10_feed_forward1_linear2_weight_to_fp16_palettized, x = input_545_cast_fp16)[name = string("linear_92_cast_fp16")];
+            fp16 var_1857_to_fp16 = const()[name = string("op_1857_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_1858_cast_fp16 = mul(x = linear_92_cast_fp16, y = var_1857_to_fp16)[name = string("op_1858_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_551_cast_fp16 = add(x = input_539_cast_fp16, y = var_1858_cast_fp16)[name = string("input_551_cast_fp16")];
+            tensor<int32, [1]> query_21_axes_0 = const()[name = string("query_21_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_10_norm_self_att_weight_to_fp16 = const()[name = string("model_layers_10_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132635520)))];
+            tensor<fp16, [1024]> model_layers_10_norm_self_att_bias_to_fp16 = const()[name = string("model_layers_10_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132637632)))];
+            tensor<fp16, [1, 126, 1024]> query_21_cast_fp16 = layer_norm(axes = query_21_axes_0, beta = model_layers_10_norm_self_att_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_10_norm_self_att_weight_to_fp16, x = input_551_cast_fp16)[name = string("query_21_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_10_self_attn_linear_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132639744))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133164096))))[name = string("model_layers_10_self_attn_linear_q_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_93_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_10_self_attn_linear_q_weight_to_fp16_palettized, x = query_21_cast_fp16)[name = string("linear_93_cast_fp16")];
+            tensor<int32, [4]> var_1874 = const()[name = string("op_1874"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> q_61_cast_fp16 = reshape(shape = var_1874, x = linear_93_cast_fp16)[name = string("q_61_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_10_self_attn_linear_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133165184))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133689536))))[name = string("model_layers_10_self_attn_linear_k_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_94_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_10_self_attn_linear_k_weight_to_fp16_palettized, x = query_21_cast_fp16)[name = string("linear_94_cast_fp16")];
+            tensor<int32, [4]> var_1878 = const()[name = string("op_1878"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> k_41_cast_fp16 = reshape(shape = var_1878, x = linear_94_cast_fp16)[name = string("k_41_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_10_self_attn_linear_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133690624))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134214976))))[name = string("model_layers_10_self_attn_linear_v_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_95_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_10_self_attn_linear_v_weight_to_fp16_palettized, x = query_21_cast_fp16)[name = string("linear_95_cast_fp16")];
+            tensor<int32, [4]> var_1882 = const()[name = string("op_1882"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> v_21_cast_fp16 = reshape(shape = var_1882, x = linear_95_cast_fp16)[name = string("v_21_cast_fp16")];
+            tensor<int32, [4]> value_21_perm_0 = const()[name = string("value_21_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> model_layers_10_self_attn_pos_bias_u_to_fp16 = const()[name = string("model_layers_10_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134216064)))];
+            tensor<fp16, [1, 126, 8, 128]> var_1894_cast_fp16 = add(x = q_61_cast_fp16, y = model_layers_10_self_attn_pos_bias_u_to_fp16)[name = string("op_1894_cast_fp16")];
+            tensor<fp16, [8, 128]> model_layers_10_self_attn_pos_bias_v_to_fp16 = const()[name = string("model_layers_10_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134218176)))];
+            tensor<fp16, [1, 126, 8, 128]> var_1896_cast_fp16 = add(x = q_61_cast_fp16, y = model_layers_10_self_attn_pos_bias_v_to_fp16)[name = string("op_1896_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_21_perm_0 = const()[name = string("q_with_bias_v_21_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_227_transpose_x_0 = const()[name = string("x_227_transpose_x_0"), val = bool(false)];
+            bool x_227_transpose_y_0 = const()[name = string("x_227_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 251]> var_1898_to_fp16 = const()[name = string("op_1898_to_fp16"), val = tensor<fp16, [1, 8, 128, 251]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134220288)))];
+            tensor<fp16, [1, 8, 126, 128]> q_with_bias_v_21_cast_fp16 = transpose(perm = q_with_bias_v_21_perm_0, x = var_1896_cast_fp16)[name = string("transpose_242")];
+            tensor<fp16, [1, 8, 126, 251]> x_227_cast_fp16 = matmul(transpose_x = x_227_transpose_x_0, transpose_y = x_227_transpose_y_0, x = q_with_bias_v_21_cast_fp16, y = var_1898_to_fp16)[name = string("x_227_cast_fp16")];
+            tensor<int32, [8]> x_229_pad_0 = const()[name = string("x_229_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_229_mode_0 = const()[name = string("x_229_mode_0"), val = string("constant")];
+            fp16 const_114_to_fp16 = const()[name = string("const_114_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 252]> x_229_cast_fp16 = pad(constant_val = const_114_to_fp16, mode = x_229_mode_0, pad = x_229_pad_0, x = x_227_cast_fp16)[name = string("x_229_cast_fp16")];
+            tensor<int32, [4]> var_1906 = const()[name = string("op_1906"), val = tensor<int32, [4]>([1, 8, -1, 126])];
+            tensor<fp16, [1, 8, 252, 126]> x_231_cast_fp16 = reshape(shape = var_1906, x = x_229_cast_fp16)[name = string("x_231_cast_fp16")];
+            tensor<int32, [4]> var_1910_begin_0 = const()[name = string("op_1910_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_1910_end_0 = const()[name = string("op_1910_end_0"), val = tensor<int32, [4]>([1, 8, 252, 126])];
+            tensor<bool, [4]> var_1910_end_mask_0 = const()[name = string("op_1910_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 251, 126]> var_1910_cast_fp16 = slice_by_index(begin = var_1910_begin_0, end = var_1910_end_0, end_mask = var_1910_end_mask_0, x = x_231_cast_fp16)[name = string("op_1910_cast_fp16")];
+            tensor<int32, [4]> var_1911 = const()[name = string("op_1911"), val = tensor<int32, [4]>([1, 8, 126, 251])];
+            tensor<fp16, [1, 8, 126, 251]> matrix_bd_41_cast_fp16 = reshape(shape = var_1911, x = var_1910_cast_fp16)[name = string("matrix_bd_41_cast_fp16")];
+            bool matrix_ac_21_transpose_x_0 = const()[name = string("matrix_ac_21_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_21_transpose_y_0 = const()[name = string("matrix_ac_21_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_116_perm_0 = const()[name = string("transpose_116_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_117_perm_0 = const()[name = string("transpose_117_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 126]> transpose_117 = transpose(perm = transpose_117_perm_0, x = k_41_cast_fp16)[name = string("transpose_240")];
+            tensor<fp16, [1, 8, 126, 128]> transpose_116 = transpose(perm = transpose_116_perm_0, x = var_1894_cast_fp16)[name = string("transpose_241")];
+            tensor<fp16, [1, 8, 126, 126]> matrix_ac_21_cast_fp16 = matmul(transpose_x = matrix_ac_21_transpose_x_0, transpose_y = matrix_ac_21_transpose_y_0, x = transpose_116, y = transpose_117)[name = string("matrix_ac_21_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_43_begin_0 = const()[name = string("matrix_bd_43_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_43_end_0 = const()[name = string("matrix_bd_43_end_0"), val = tensor<int32, [4]>([1, 8, 126, 126])];
+            tensor<bool, [4]> matrix_bd_43_end_mask_0 = const()[name = string("matrix_bd_43_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 126, 126]> matrix_bd_43_cast_fp16 = slice_by_index(begin = matrix_bd_43_begin_0, end = matrix_bd_43_end_0, end_mask = matrix_bd_43_end_mask_0, x = matrix_bd_41_cast_fp16)[name = string("matrix_bd_43_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_1920_cast_fp16 = add(x = matrix_ac_21_cast_fp16, y = matrix_bd_43_cast_fp16)[name = string("op_1920_cast_fp16")];
+            fp16 _inversed_scores_41_y_0_to_fp16 = const()[name = string("_inversed_scores_41_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 126, 126]> _inversed_scores_41_cast_fp16 = mul(x = var_1920_cast_fp16, y = _inversed_scores_41_y_0_to_fp16)[name = string("_inversed_scores_41_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> scores_43_cast_fp16 = select(a = var_7_to_fp16, b = _inversed_scores_41_cast_fp16, cond = mask_3)[name = string("scores_43_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_1926_cast_fp16 = softmax(axis = var_25, x = scores_43_cast_fp16)[name = string("op_1926_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> input_553_cast_fp16 = select(a = var_6_to_fp16, b = var_1926_cast_fp16, cond = mask_3)[name = string("input_553_cast_fp16")];
+            bool x_233_transpose_x_0 = const()[name = string("x_233_transpose_x_0"), val = bool(false)];
+            bool x_233_transpose_y_0 = const()[name = string("x_233_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 126, 128]> value_21_cast_fp16 = transpose(perm = value_21_perm_0, x = v_21_cast_fp16)[name = string("transpose_239")];
+            tensor<fp16, [1, 8, 126, 128]> x_233_cast_fp16 = matmul(transpose_x = x_233_transpose_x_0, transpose_y = x_233_transpose_y_0, x = input_553_cast_fp16, y = value_21_cast_fp16)[name = string("x_233_cast_fp16")];
+            tensor<int32, [4]> var_1930_perm_0 = const()[name = string("op_1930_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_1931 = const()[name = string("op_1931"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 126, 8, 128]> var_1930_cast_fp16 = transpose(perm = var_1930_perm_0, x = x_233_cast_fp16)[name = string("transpose_238")];
+            tensor<fp16, [1, 126, 1024]> input_555_cast_fp16 = reshape(shape = var_1931, x = var_1930_cast_fp16)[name = string("input_555_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_10_self_attn_linear_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134734400))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(135258752))))[name = string("model_layers_10_self_attn_linear_out_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_97_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_10_self_attn_linear_out_weight_to_fp16_palettized, x = input_555_cast_fp16)[name = string("linear_97_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_559_cast_fp16 = add(x = input_551_cast_fp16, y = linear_97_cast_fp16)[name = string("input_559_cast_fp16")];
+            tensor<int32, [1]> x_237_axes_0 = const()[name = string("x_237_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_10_norm_conv_weight_to_fp16 = const()[name = string("model_layers_10_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(135259840)))];
+            tensor<fp16, [1024]> model_layers_10_norm_conv_bias_to_fp16 = const()[name = string("model_layers_10_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(135261952)))];
+            tensor<fp16, [1, 126, 1024]> x_237_cast_fp16 = layer_norm(axes = x_237_axes_0, beta = model_layers_10_norm_conv_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_10_norm_conv_weight_to_fp16, x = input_559_cast_fp16)[name = string("x_237_cast_fp16")];
+            tensor<int32, [3]> input_561_perm_0 = const()[name = string("input_561_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_563_pad_type_0 = const()[name = string("input_563_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_563_strides_0 = const()[name = string("input_563_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_563_pad_0 = const()[name = string("input_563_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_563_dilations_0 = const()[name = string("input_563_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_563_groups_0 = const()[name = string("input_563_groups_0"), val = int32(1)];
+            tensor<fp16, [2048, 1024, 1]> model_layers_10_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(135264064))), lut = tensor<fp16, [64, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136312704))))[name = string("model_layers_10_conv_pointwise_conv1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 1024, 126]> input_561_cast_fp16 = transpose(perm = input_561_perm_0, x = x_237_cast_fp16)[name = string("transpose_237")];
+            tensor<fp16, [1, 2048, 126]> input_563_cast_fp16 = conv(dilations = input_563_dilations_0, groups = input_563_groups_0, pad = input_563_pad_0, pad_type = input_563_pad_type_0, strides = input_563_strides_0, weight = model_layers_10_conv_pointwise_conv1_weight_to_fp16_palettized, x = input_561_cast_fp16)[name = string("input_563_cast_fp16")];
+            int32 x_239_split_num_splits_0 = const()[name = string("x_239_split_num_splits_0"), val = int32(2)];
+            int32 x_239_split_axis_0 = const()[name = string("x_239_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 126]> x_239_split_cast_fp16_0, tensor<fp16, [1, 1024, 126]> x_239_split_cast_fp16_1 = split(axis = x_239_split_axis_0, num_splits = x_239_split_num_splits_0, x = input_563_cast_fp16)[name = string("x_239_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_239_split_1_sigmoid_cast_fp16 = sigmoid(x = x_239_split_cast_fp16_1)[name = string("x_239_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_239_cast_fp16 = mul(x = x_239_split_cast_fp16_0, y = x_239_split_1_sigmoid_cast_fp16)[name = string("x_239_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_565_cast_fp16 = select(a = var_6_to_fp16, b = x_239_cast_fp16, cond = var_323)[name = string("input_565_cast_fp16")];
+            tensor<int32, [6]> input_567_pad_0 = const()[name = string("input_567_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_567_mode_0 = const()[name = string("input_567_mode_0"), val = string("constant")];
+            fp16 const_117_to_fp16 = const()[name = string("const_117_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 134]> input_567_cast_fp16 = pad(constant_val = const_117_to_fp16, mode = input_567_mode_0, pad = input_567_pad_0, x = input_565_cast_fp16)[name = string("input_567_cast_fp16")];
+            string input_569_pad_type_0 = const()[name = string("input_569_pad_type_0"), val = string("valid")];
+            int32 input_569_groups_0 = const()[name = string("input_569_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_569_strides_0 = const()[name = string("input_569_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_569_pad_0 = const()[name = string("input_569_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_569_dilations_0 = const()[name = string("input_569_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_268_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136314816))), lut = tensor<fp16, [32, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136319488))))[name = string("const_268_to_fp16_palettized")];
+            tensor<fp16, [1024]> const_269_to_fp16 = const()[name = string("const_269_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136320576)))];
+            tensor<fp16, [1, 1024, 126]> input_571_cast_fp16 = conv(bias = const_269_to_fp16, dilations = input_569_dilations_0, groups = input_569_groups_0, pad = input_569_pad_0, pad_type = input_569_pad_type_0, strides = input_569_strides_0, weight = const_268_to_fp16_palettized, x = input_567_cast_fp16)[name = string("input_571_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_573_cast_fp16 = silu(x = input_571_cast_fp16)[name = string("input_573_cast_fp16")];
+            string x_241_pad_type_0 = const()[name = string("x_241_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_241_strides_0 = const()[name = string("x_241_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_241_pad_0 = const()[name = string("x_241_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_241_dilations_0 = const()[name = string("x_241_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_241_groups_0 = const()[name = string("x_241_groups_0"), val = int32(1)];
+            tensor<fp16, [1024, 1024, 1]> model_layers_10_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136322688))), lut = tensor<fp16, [32, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136847040))))[name = string("model_layers_10_conv_pointwise_conv2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 1024, 126]> x_241_cast_fp16 = conv(dilations = x_241_dilations_0, groups = x_241_groups_0, pad = x_241_pad_0, pad_type = x_241_pad_type_0, strides = x_241_strides_0, weight = model_layers_10_conv_pointwise_conv2_weight_to_fp16_palettized, x = input_573_cast_fp16)[name = string("x_241_cast_fp16")];
+            tensor<int32, [3]> input_575_perm_0 = const()[name = string("input_575_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 126, 1024]> input_575_cast_fp16 = transpose(perm = input_575_perm_0, x = x_241_cast_fp16)[name = string("transpose_236")];
+            tensor<fp16, [1, 126, 1024]> input_577_cast_fp16 = add(x = input_559_cast_fp16, y = input_575_cast_fp16)[name = string("input_577_cast_fp16")];
+            tensor<int32, [1]> input_579_axes_0 = const()[name = string("input_579_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_10_norm_feed_forward2_weight_to_fp16 = const()[name = string("model_layers_10_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136848128)))];
+            tensor<fp16, [1024]> model_layers_10_norm_feed_forward2_bias_to_fp16 = const()[name = string("model_layers_10_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136850240)))];
+            tensor<fp16, [1, 126, 1024]> input_579_cast_fp16 = layer_norm(axes = input_579_axes_0, beta = model_layers_10_norm_feed_forward2_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_10_norm_feed_forward2_weight_to_fp16, x = input_577_cast_fp16)[name = string("input_579_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_10_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136852352))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138949568))))[name = string("model_layers_10_feed_forward2_linear1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 4096]> linear_98_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_10_feed_forward2_linear1_weight_to_fp16_palettized, x = input_579_cast_fp16)[name = string("linear_98_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_583_cast_fp16 = silu(x = linear_98_cast_fp16)[name = string("input_583_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_10_feed_forward2_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138953728))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141050944))))[name = string("model_layers_10_feed_forward2_linear2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_99_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_10_feed_forward2_linear2_weight_to_fp16_palettized, x = input_583_cast_fp16)[name = string("linear_99_cast_fp16")];
+            fp16 var_1991_to_fp16 = const()[name = string("op_1991_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_1992_cast_fp16 = mul(x = linear_99_cast_fp16, y = var_1991_to_fp16)[name = string("op_1992_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_589_cast_fp16 = add(x = input_577_cast_fp16, y = var_1992_cast_fp16)[name = string("input_589_cast_fp16")];
+            tensor<int32, [1]> input_591_axes_0 = const()[name = string("input_591_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_10_norm_out_weight_to_fp16 = const()[name = string("model_layers_10_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141052032)))];
+            tensor<fp16, [1024]> model_layers_10_norm_out_bias_to_fp16 = const()[name = string("model_layers_10_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141054144)))];
+            tensor<fp16, [1, 126, 1024]> input_591_cast_fp16 = layer_norm(axes = input_591_axes_0, beta = model_layers_10_norm_out_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_10_norm_out_weight_to_fp16, x = input_589_cast_fp16)[name = string("input_591_cast_fp16")];
+            tensor<int32, [1]> input_593_axes_0 = const()[name = string("input_593_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_11_norm_feed_forward1_weight_to_fp16 = const()[name = string("model_layers_11_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141056256)))];
+            tensor<fp16, [1024]> model_layers_11_norm_feed_forward1_bias_to_fp16 = const()[name = string("model_layers_11_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141058368)))];
+            tensor<fp16, [1, 126, 1024]> input_593_cast_fp16 = layer_norm(axes = input_593_axes_0, beta = model_layers_11_norm_feed_forward1_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_11_norm_feed_forward1_weight_to_fp16, x = input_591_cast_fp16)[name = string("input_593_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_11_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141060480))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143157696))))[name = string("model_layers_11_feed_forward1_linear1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 4096]> linear_100_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_11_feed_forward1_linear1_weight_to_fp16_palettized, x = input_593_cast_fp16)[name = string("linear_100_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_597_cast_fp16 = silu(x = linear_100_cast_fp16)[name = string("input_597_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_11_feed_forward1_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143161856))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145259072))))[name = string("model_layers_11_feed_forward1_linear2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_101_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_11_feed_forward1_linear2_weight_to_fp16_palettized, x = input_597_cast_fp16)[name = string("linear_101_cast_fp16")];
+            fp16 var_2020_to_fp16 = const()[name = string("op_2020_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_2021_cast_fp16 = mul(x = linear_101_cast_fp16, y = var_2020_to_fp16)[name = string("op_2021_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_603_cast_fp16 = add(x = input_591_cast_fp16, y = var_2021_cast_fp16)[name = string("input_603_cast_fp16")];
+            tensor<int32, [1]> query_23_axes_0 = const()[name = string("query_23_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_11_norm_self_att_weight_to_fp16 = const()[name = string("model_layers_11_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145260160)))];
+            tensor<fp16, [1024]> model_layers_11_norm_self_att_bias_to_fp16 = const()[name = string("model_layers_11_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145262272)))];
+            tensor<fp16, [1, 126, 1024]> query_23_cast_fp16 = layer_norm(axes = query_23_axes_0, beta = model_layers_11_norm_self_att_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_11_norm_self_att_weight_to_fp16, x = input_603_cast_fp16)[name = string("query_23_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_11_self_attn_linear_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145264384))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145788736))))[name = string("model_layers_11_self_attn_linear_q_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_102_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_11_self_attn_linear_q_weight_to_fp16_palettized, x = query_23_cast_fp16)[name = string("linear_102_cast_fp16")];
+            tensor<int32, [4]> var_2037 = const()[name = string("op_2037"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> q_67_cast_fp16 = reshape(shape = var_2037, x = linear_102_cast_fp16)[name = string("q_67_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_11_self_attn_linear_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145789824))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146314176))))[name = string("model_layers_11_self_attn_linear_k_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_103_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_11_self_attn_linear_k_weight_to_fp16_palettized, x = query_23_cast_fp16)[name = string("linear_103_cast_fp16")];
+            tensor<int32, [4]> var_2041 = const()[name = string("op_2041"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> k_45_cast_fp16 = reshape(shape = var_2041, x = linear_103_cast_fp16)[name = string("k_45_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_11_self_attn_linear_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146315264))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146839616))))[name = string("model_layers_11_self_attn_linear_v_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_104_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_11_self_attn_linear_v_weight_to_fp16_palettized, x = query_23_cast_fp16)[name = string("linear_104_cast_fp16")];
+            tensor<int32, [4]> var_2045 = const()[name = string("op_2045"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> v_23_cast_fp16 = reshape(shape = var_2045, x = linear_104_cast_fp16)[name = string("v_23_cast_fp16")];
+            tensor<int32, [4]> value_23_perm_0 = const()[name = string("value_23_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> model_layers_11_self_attn_pos_bias_u_to_fp16 = const()[name = string("model_layers_11_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146840704)))];
+            tensor<fp16, [1, 126, 8, 128]> var_2057_cast_fp16 = add(x = q_67_cast_fp16, y = model_layers_11_self_attn_pos_bias_u_to_fp16)[name = string("op_2057_cast_fp16")];
+            tensor<fp16, [8, 128]> model_layers_11_self_attn_pos_bias_v_to_fp16 = const()[name = string("model_layers_11_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146842816)))];
+            tensor<fp16, [1, 126, 8, 128]> var_2059_cast_fp16 = add(x = q_67_cast_fp16, y = model_layers_11_self_attn_pos_bias_v_to_fp16)[name = string("op_2059_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_23_perm_0 = const()[name = string("q_with_bias_v_23_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_249_transpose_x_0 = const()[name = string("x_249_transpose_x_0"), val = bool(false)];
+            bool x_249_transpose_y_0 = const()[name = string("x_249_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 251]> var_2061_to_fp16 = const()[name = string("op_2061_to_fp16"), val = tensor<fp16, [1, 8, 128, 251]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146844928)))];
+            tensor<fp16, [1, 8, 126, 128]> q_with_bias_v_23_cast_fp16 = transpose(perm = q_with_bias_v_23_perm_0, x = var_2059_cast_fp16)[name = string("transpose_235")];
+            tensor<fp16, [1, 8, 126, 251]> x_249_cast_fp16 = matmul(transpose_x = x_249_transpose_x_0, transpose_y = x_249_transpose_y_0, x = q_with_bias_v_23_cast_fp16, y = var_2061_to_fp16)[name = string("x_249_cast_fp16")];
+            tensor<int32, [8]> x_251_pad_0 = const()[name = string("x_251_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_251_mode_0 = const()[name = string("x_251_mode_0"), val = string("constant")];
+            fp16 const_124_to_fp16 = const()[name = string("const_124_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 252]> x_251_cast_fp16 = pad(constant_val = const_124_to_fp16, mode = x_251_mode_0, pad = x_251_pad_0, x = x_249_cast_fp16)[name = string("x_251_cast_fp16")];
+            tensor<int32, [4]> var_2069 = const()[name = string("op_2069"), val = tensor<int32, [4]>([1, 8, -1, 126])];
+            tensor<fp16, [1, 8, 252, 126]> x_253_cast_fp16 = reshape(shape = var_2069, x = x_251_cast_fp16)[name = string("x_253_cast_fp16")];
+            tensor<int32, [4]> var_2073_begin_0 = const()[name = string("op_2073_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_2073_end_0 = const()[name = string("op_2073_end_0"), val = tensor<int32, [4]>([1, 8, 252, 126])];
+            tensor<bool, [4]> var_2073_end_mask_0 = const()[name = string("op_2073_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 251, 126]> var_2073_cast_fp16 = slice_by_index(begin = var_2073_begin_0, end = var_2073_end_0, end_mask = var_2073_end_mask_0, x = x_253_cast_fp16)[name = string("op_2073_cast_fp16")];
+            tensor<int32, [4]> var_2074 = const()[name = string("op_2074"), val = tensor<int32, [4]>([1, 8, 126, 251])];
+            tensor<fp16, [1, 8, 126, 251]> matrix_bd_45_cast_fp16 = reshape(shape = var_2074, x = var_2073_cast_fp16)[name = string("matrix_bd_45_cast_fp16")];
+            bool matrix_ac_23_transpose_x_0 = const()[name = string("matrix_ac_23_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_23_transpose_y_0 = const()[name = string("matrix_ac_23_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_118_perm_0 = const()[name = string("transpose_118_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_119_perm_0 = const()[name = string("transpose_119_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 126]> transpose_119 = transpose(perm = transpose_119_perm_0, x = k_45_cast_fp16)[name = string("transpose_233")];
+            tensor<fp16, [1, 8, 126, 128]> transpose_118 = transpose(perm = transpose_118_perm_0, x = var_2057_cast_fp16)[name = string("transpose_234")];
+            tensor<fp16, [1, 8, 126, 126]> matrix_ac_23_cast_fp16 = matmul(transpose_x = matrix_ac_23_transpose_x_0, transpose_y = matrix_ac_23_transpose_y_0, x = transpose_118, y = transpose_119)[name = string("matrix_ac_23_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_47_begin_0 = const()[name = string("matrix_bd_47_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_47_end_0 = const()[name = string("matrix_bd_47_end_0"), val = tensor<int32, [4]>([1, 8, 126, 126])];
+            tensor<bool, [4]> matrix_bd_47_end_mask_0 = const()[name = string("matrix_bd_47_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 126, 126]> matrix_bd_47_cast_fp16 = slice_by_index(begin = matrix_bd_47_begin_0, end = matrix_bd_47_end_0, end_mask = matrix_bd_47_end_mask_0, x = matrix_bd_45_cast_fp16)[name = string("matrix_bd_47_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_2083_cast_fp16 = add(x = matrix_ac_23_cast_fp16, y = matrix_bd_47_cast_fp16)[name = string("op_2083_cast_fp16")];
+            fp16 _inversed_scores_45_y_0_to_fp16 = const()[name = string("_inversed_scores_45_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 126, 126]> _inversed_scores_45_cast_fp16 = mul(x = var_2083_cast_fp16, y = _inversed_scores_45_y_0_to_fp16)[name = string("_inversed_scores_45_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> scores_47_cast_fp16 = select(a = var_7_to_fp16, b = _inversed_scores_45_cast_fp16, cond = mask_3)[name = string("scores_47_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_2089_cast_fp16 = softmax(axis = var_25, x = scores_47_cast_fp16)[name = string("op_2089_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> input_605_cast_fp16 = select(a = var_6_to_fp16, b = var_2089_cast_fp16, cond = mask_3)[name = string("input_605_cast_fp16")];
+            bool x_255_transpose_x_0 = const()[name = string("x_255_transpose_x_0"), val = bool(false)];
+            bool x_255_transpose_y_0 = const()[name = string("x_255_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 126, 128]> value_23_cast_fp16 = transpose(perm = value_23_perm_0, x = v_23_cast_fp16)[name = string("transpose_232")];
+            tensor<fp16, [1, 8, 126, 128]> x_255_cast_fp16 = matmul(transpose_x = x_255_transpose_x_0, transpose_y = x_255_transpose_y_0, x = input_605_cast_fp16, y = value_23_cast_fp16)[name = string("x_255_cast_fp16")];
+            tensor<int32, [4]> var_2093_perm_0 = const()[name = string("op_2093_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_2094 = const()[name = string("op_2094"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 126, 8, 128]> var_2093_cast_fp16 = transpose(perm = var_2093_perm_0, x = x_255_cast_fp16)[name = string("transpose_231")];
+            tensor<fp16, [1, 126, 1024]> input_607_cast_fp16 = reshape(shape = var_2094, x = var_2093_cast_fp16)[name = string("input_607_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_11_self_attn_linear_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147359040))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147883392))))[name = string("model_layers_11_self_attn_linear_out_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_106_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_11_self_attn_linear_out_weight_to_fp16_palettized, x = input_607_cast_fp16)[name = string("linear_106_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_611_cast_fp16 = add(x = input_603_cast_fp16, y = linear_106_cast_fp16)[name = string("input_611_cast_fp16")];
+            tensor<int32, [1]> x_259_axes_0 = const()[name = string("x_259_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_11_norm_conv_weight_to_fp16 = const()[name = string("model_layers_11_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147884480)))];
+            tensor<fp16, [1024]> model_layers_11_norm_conv_bias_to_fp16 = const()[name = string("model_layers_11_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147886592)))];
+            tensor<fp16, [1, 126, 1024]> x_259_cast_fp16 = layer_norm(axes = x_259_axes_0, beta = model_layers_11_norm_conv_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_11_norm_conv_weight_to_fp16, x = input_611_cast_fp16)[name = string("x_259_cast_fp16")];
+            tensor<int32, [3]> input_613_perm_0 = const()[name = string("input_613_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_615_pad_type_0 = const()[name = string("input_615_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_615_strides_0 = const()[name = string("input_615_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_615_pad_0 = const()[name = string("input_615_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_615_dilations_0 = const()[name = string("input_615_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_615_groups_0 = const()[name = string("input_615_groups_0"), val = int32(1)];
+            tensor<fp16, [2048, 1024, 1]> model_layers_11_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147888704))), lut = tensor<fp16, [64, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(148937344))))[name = string("model_layers_11_conv_pointwise_conv1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 1024, 126]> input_613_cast_fp16 = transpose(perm = input_613_perm_0, x = x_259_cast_fp16)[name = string("transpose_230")];
+            tensor<fp16, [1, 2048, 126]> input_615_cast_fp16 = conv(dilations = input_615_dilations_0, groups = input_615_groups_0, pad = input_615_pad_0, pad_type = input_615_pad_type_0, strides = input_615_strides_0, weight = model_layers_11_conv_pointwise_conv1_weight_to_fp16_palettized, x = input_613_cast_fp16)[name = string("input_615_cast_fp16")];
+            int32 x_261_split_num_splits_0 = const()[name = string("x_261_split_num_splits_0"), val = int32(2)];
+            int32 x_261_split_axis_0 = const()[name = string("x_261_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 126]> x_261_split_cast_fp16_0, tensor<fp16, [1, 1024, 126]> x_261_split_cast_fp16_1 = split(axis = x_261_split_axis_0, num_splits = x_261_split_num_splits_0, x = input_615_cast_fp16)[name = string("x_261_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_261_split_1_sigmoid_cast_fp16 = sigmoid(x = x_261_split_cast_fp16_1)[name = string("x_261_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_261_cast_fp16 = mul(x = x_261_split_cast_fp16_0, y = x_261_split_1_sigmoid_cast_fp16)[name = string("x_261_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_617_cast_fp16 = select(a = var_6_to_fp16, b = x_261_cast_fp16, cond = var_323)[name = string("input_617_cast_fp16")];
+            tensor<int32, [6]> input_619_pad_0 = const()[name = string("input_619_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_619_mode_0 = const()[name = string("input_619_mode_0"), val = string("constant")];
+            fp16 const_127_to_fp16 = const()[name = string("const_127_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 134]> input_619_cast_fp16 = pad(constant_val = const_127_to_fp16, mode = input_619_mode_0, pad = input_619_pad_0, x = input_617_cast_fp16)[name = string("input_619_cast_fp16")];
+            string input_621_pad_type_0 = const()[name = string("input_621_pad_type_0"), val = string("valid")];
+            int32 input_621_groups_0 = const()[name = string("input_621_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_621_strides_0 = const()[name = string("input_621_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_621_pad_0 = const()[name = string("input_621_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_621_dilations_0 = const()[name = string("input_621_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_270_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(148939456))), lut = tensor<fp16, [32, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(148944128))))[name = string("const_270_to_fp16_palettized")];
+            tensor<fp16, [1024]> const_271_to_fp16 = const()[name = string("const_271_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(148945216)))];
+            tensor<fp16, [1, 1024, 126]> input_623_cast_fp16 = conv(bias = const_271_to_fp16, dilations = input_621_dilations_0, groups = input_621_groups_0, pad = input_621_pad_0, pad_type = input_621_pad_type_0, strides = input_621_strides_0, weight = const_270_to_fp16_palettized, x = input_619_cast_fp16)[name = string("input_623_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_625_cast_fp16 = silu(x = input_623_cast_fp16)[name = string("input_625_cast_fp16")];
+            string x_263_pad_type_0 = const()[name = string("x_263_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_263_strides_0 = const()[name = string("x_263_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_263_pad_0 = const()[name = string("x_263_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_263_dilations_0 = const()[name = string("x_263_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_263_groups_0 = const()[name = string("x_263_groups_0"), val = int32(1)];
+            tensor<fp16, [1024, 1024, 1]> model_layers_11_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(148947328))), lut = tensor<fp16, [32, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149471680))))[name = string("model_layers_11_conv_pointwise_conv2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 1024, 126]> x_263_cast_fp16 = conv(dilations = x_263_dilations_0, groups = x_263_groups_0, pad = x_263_pad_0, pad_type = x_263_pad_type_0, strides = x_263_strides_0, weight = model_layers_11_conv_pointwise_conv2_weight_to_fp16_palettized, x = input_625_cast_fp16)[name = string("x_263_cast_fp16")];
+            tensor<int32, [3]> input_627_perm_0 = const()[name = string("input_627_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 126, 1024]> input_627_cast_fp16 = transpose(perm = input_627_perm_0, x = x_263_cast_fp16)[name = string("transpose_229")];
+            tensor<fp16, [1, 126, 1024]> input_629_cast_fp16 = add(x = input_611_cast_fp16, y = input_627_cast_fp16)[name = string("input_629_cast_fp16")];
+            tensor<int32, [1]> input_631_axes_0 = const()[name = string("input_631_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_11_norm_feed_forward2_weight_to_fp16 = const()[name = string("model_layers_11_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149472768)))];
+            tensor<fp16, [1024]> model_layers_11_norm_feed_forward2_bias_to_fp16 = const()[name = string("model_layers_11_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149474880)))];
+            tensor<fp16, [1, 126, 1024]> input_631_cast_fp16 = layer_norm(axes = input_631_axes_0, beta = model_layers_11_norm_feed_forward2_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_11_norm_feed_forward2_weight_to_fp16, x = input_629_cast_fp16)[name = string("input_631_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_11_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149476992))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151574208))))[name = string("model_layers_11_feed_forward2_linear1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 4096]> linear_107_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_11_feed_forward2_linear1_weight_to_fp16_palettized, x = input_631_cast_fp16)[name = string("linear_107_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_635_cast_fp16 = silu(x = linear_107_cast_fp16)[name = string("input_635_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_11_feed_forward2_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151578368))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153675584))))[name = string("model_layers_11_feed_forward2_linear2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_108_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_11_feed_forward2_linear2_weight_to_fp16_palettized, x = input_635_cast_fp16)[name = string("linear_108_cast_fp16")];
+            fp16 var_2154_to_fp16 = const()[name = string("op_2154_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_2155_cast_fp16 = mul(x = linear_108_cast_fp16, y = var_2154_to_fp16)[name = string("op_2155_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_641_cast_fp16 = add(x = input_629_cast_fp16, y = var_2155_cast_fp16)[name = string("input_641_cast_fp16")];
+            tensor<int32, [1]> input_643_axes_0 = const()[name = string("input_643_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_11_norm_out_weight_to_fp16 = const()[name = string("model_layers_11_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153676672)))];
+            tensor<fp16, [1024]> model_layers_11_norm_out_bias_to_fp16 = const()[name = string("model_layers_11_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153678784)))];
+            tensor<fp16, [1, 126, 1024]> input_643_cast_fp16 = layer_norm(axes = input_643_axes_0, beta = model_layers_11_norm_out_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_11_norm_out_weight_to_fp16, x = input_641_cast_fp16)[name = string("input_643_cast_fp16")];
+            tensor<int32, [1]> input_645_axes_0 = const()[name = string("input_645_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_12_norm_feed_forward1_weight_to_fp16 = const()[name = string("model_layers_12_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153680896)))];
+            tensor<fp16, [1024]> model_layers_12_norm_feed_forward1_bias_to_fp16 = const()[name = string("model_layers_12_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153683008)))];
+            tensor<fp16, [1, 126, 1024]> input_645_cast_fp16 = layer_norm(axes = input_645_axes_0, beta = model_layers_12_norm_feed_forward1_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_12_norm_feed_forward1_weight_to_fp16, x = input_643_cast_fp16)[name = string("input_645_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_12_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153685120))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(155782336))))[name = string("model_layers_12_feed_forward1_linear1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 4096]> linear_109_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_12_feed_forward1_linear1_weight_to_fp16_palettized, x = input_645_cast_fp16)[name = string("linear_109_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_649_cast_fp16 = silu(x = linear_109_cast_fp16)[name = string("input_649_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_12_feed_forward1_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(155786496))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157883712))))[name = string("model_layers_12_feed_forward1_linear2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_110_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_12_feed_forward1_linear2_weight_to_fp16_palettized, x = input_649_cast_fp16)[name = string("linear_110_cast_fp16")];
+            fp16 var_2183_to_fp16 = const()[name = string("op_2183_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_2184_cast_fp16 = mul(x = linear_110_cast_fp16, y = var_2183_to_fp16)[name = string("op_2184_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_655_cast_fp16 = add(x = input_643_cast_fp16, y = var_2184_cast_fp16)[name = string("input_655_cast_fp16")];
+            tensor<int32, [1]> query_25_axes_0 = const()[name = string("query_25_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_12_norm_self_att_weight_to_fp16 = const()[name = string("model_layers_12_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157884800)))];
+            tensor<fp16, [1024]> model_layers_12_norm_self_att_bias_to_fp16 = const()[name = string("model_layers_12_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157886912)))];
+            tensor<fp16, [1, 126, 1024]> query_25_cast_fp16 = layer_norm(axes = query_25_axes_0, beta = model_layers_12_norm_self_att_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_12_norm_self_att_weight_to_fp16, x = input_655_cast_fp16)[name = string("query_25_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_12_self_attn_linear_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157889024))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(158413376))))[name = string("model_layers_12_self_attn_linear_q_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_111_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_12_self_attn_linear_q_weight_to_fp16_palettized, x = query_25_cast_fp16)[name = string("linear_111_cast_fp16")];
+            tensor<int32, [4]> var_2200 = const()[name = string("op_2200"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> q_73_cast_fp16 = reshape(shape = var_2200, x = linear_111_cast_fp16)[name = string("q_73_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_12_self_attn_linear_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(158414464))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(158938816))))[name = string("model_layers_12_self_attn_linear_k_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_112_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_12_self_attn_linear_k_weight_to_fp16_palettized, x = query_25_cast_fp16)[name = string("linear_112_cast_fp16")];
+            tensor<int32, [4]> var_2204 = const()[name = string("op_2204"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> k_49_cast_fp16 = reshape(shape = var_2204, x = linear_112_cast_fp16)[name = string("k_49_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_12_self_attn_linear_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(158939904))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(159464256))))[name = string("model_layers_12_self_attn_linear_v_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_113_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_12_self_attn_linear_v_weight_to_fp16_palettized, x = query_25_cast_fp16)[name = string("linear_113_cast_fp16")];
+            tensor<int32, [4]> var_2208 = const()[name = string("op_2208"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> v_25_cast_fp16 = reshape(shape = var_2208, x = linear_113_cast_fp16)[name = string("v_25_cast_fp16")];
+            tensor<int32, [4]> value_25_perm_0 = const()[name = string("value_25_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> model_layers_12_self_attn_pos_bias_u_to_fp16 = const()[name = string("model_layers_12_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(159465344)))];
+            tensor<fp16, [1, 126, 8, 128]> var_2220_cast_fp16 = add(x = q_73_cast_fp16, y = model_layers_12_self_attn_pos_bias_u_to_fp16)[name = string("op_2220_cast_fp16")];
+            tensor<fp16, [8, 128]> model_layers_12_self_attn_pos_bias_v_to_fp16 = const()[name = string("model_layers_12_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(159467456)))];
+            tensor<fp16, [1, 126, 8, 128]> var_2222_cast_fp16 = add(x = q_73_cast_fp16, y = model_layers_12_self_attn_pos_bias_v_to_fp16)[name = string("op_2222_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_25_perm_0 = const()[name = string("q_with_bias_v_25_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_271_transpose_x_0 = const()[name = string("x_271_transpose_x_0"), val = bool(false)];
+            bool x_271_transpose_y_0 = const()[name = string("x_271_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 251]> var_2224_to_fp16 = const()[name = string("op_2224_to_fp16"), val = tensor<fp16, [1, 8, 128, 251]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(159469568)))];
+            tensor<fp16, [1, 8, 126, 128]> q_with_bias_v_25_cast_fp16 = transpose(perm = q_with_bias_v_25_perm_0, x = var_2222_cast_fp16)[name = string("transpose_228")];
+            tensor<fp16, [1, 8, 126, 251]> x_271_cast_fp16 = matmul(transpose_x = x_271_transpose_x_0, transpose_y = x_271_transpose_y_0, x = q_with_bias_v_25_cast_fp16, y = var_2224_to_fp16)[name = string("x_271_cast_fp16")];
+            tensor<int32, [8]> x_273_pad_0 = const()[name = string("x_273_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_273_mode_0 = const()[name = string("x_273_mode_0"), val = string("constant")];
+            fp16 const_134_to_fp16 = const()[name = string("const_134_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 252]> x_273_cast_fp16 = pad(constant_val = const_134_to_fp16, mode = x_273_mode_0, pad = x_273_pad_0, x = x_271_cast_fp16)[name = string("x_273_cast_fp16")];
+            tensor<int32, [4]> var_2232 = const()[name = string("op_2232"), val = tensor<int32, [4]>([1, 8, -1, 126])];
+            tensor<fp16, [1, 8, 252, 126]> x_275_cast_fp16 = reshape(shape = var_2232, x = x_273_cast_fp16)[name = string("x_275_cast_fp16")];
+            tensor<int32, [4]> var_2236_begin_0 = const()[name = string("op_2236_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_2236_end_0 = const()[name = string("op_2236_end_0"), val = tensor<int32, [4]>([1, 8, 252, 126])];
+            tensor<bool, [4]> var_2236_end_mask_0 = const()[name = string("op_2236_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 251, 126]> var_2236_cast_fp16 = slice_by_index(begin = var_2236_begin_0, end = var_2236_end_0, end_mask = var_2236_end_mask_0, x = x_275_cast_fp16)[name = string("op_2236_cast_fp16")];
+            tensor<int32, [4]> var_2237 = const()[name = string("op_2237"), val = tensor<int32, [4]>([1, 8, 126, 251])];
+            tensor<fp16, [1, 8, 126, 251]> matrix_bd_49_cast_fp16 = reshape(shape = var_2237, x = var_2236_cast_fp16)[name = string("matrix_bd_49_cast_fp16")];
+            bool matrix_ac_25_transpose_x_0 = const()[name = string("matrix_ac_25_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_25_transpose_y_0 = const()[name = string("matrix_ac_25_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_120_perm_0 = const()[name = string("transpose_120_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_121_perm_0 = const()[name = string("transpose_121_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 126]> transpose_121 = transpose(perm = transpose_121_perm_0, x = k_49_cast_fp16)[name = string("transpose_226")];
+            tensor<fp16, [1, 8, 126, 128]> transpose_120 = transpose(perm = transpose_120_perm_0, x = var_2220_cast_fp16)[name = string("transpose_227")];
+            tensor<fp16, [1, 8, 126, 126]> matrix_ac_25_cast_fp16 = matmul(transpose_x = matrix_ac_25_transpose_x_0, transpose_y = matrix_ac_25_transpose_y_0, x = transpose_120, y = transpose_121)[name = string("matrix_ac_25_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_51_begin_0 = const()[name = string("matrix_bd_51_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_51_end_0 = const()[name = string("matrix_bd_51_end_0"), val = tensor<int32, [4]>([1, 8, 126, 126])];
+            tensor<bool, [4]> matrix_bd_51_end_mask_0 = const()[name = string("matrix_bd_51_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 126, 126]> matrix_bd_51_cast_fp16 = slice_by_index(begin = matrix_bd_51_begin_0, end = matrix_bd_51_end_0, end_mask = matrix_bd_51_end_mask_0, x = matrix_bd_49_cast_fp16)[name = string("matrix_bd_51_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_2246_cast_fp16 = add(x = matrix_ac_25_cast_fp16, y = matrix_bd_51_cast_fp16)[name = string("op_2246_cast_fp16")];
+            fp16 _inversed_scores_49_y_0_to_fp16 = const()[name = string("_inversed_scores_49_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 126, 126]> _inversed_scores_49_cast_fp16 = mul(x = var_2246_cast_fp16, y = _inversed_scores_49_y_0_to_fp16)[name = string("_inversed_scores_49_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> scores_51_cast_fp16 = select(a = var_7_to_fp16, b = _inversed_scores_49_cast_fp16, cond = mask_3)[name = string("scores_51_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_2252_cast_fp16 = softmax(axis = var_25, x = scores_51_cast_fp16)[name = string("op_2252_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> input_657_cast_fp16 = select(a = var_6_to_fp16, b = var_2252_cast_fp16, cond = mask_3)[name = string("input_657_cast_fp16")];
+            bool x_277_transpose_x_0 = const()[name = string("x_277_transpose_x_0"), val = bool(false)];
+            bool x_277_transpose_y_0 = const()[name = string("x_277_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 126, 128]> value_25_cast_fp16 = transpose(perm = value_25_perm_0, x = v_25_cast_fp16)[name = string("transpose_225")];
+            tensor<fp16, [1, 8, 126, 128]> x_277_cast_fp16 = matmul(transpose_x = x_277_transpose_x_0, transpose_y = x_277_transpose_y_0, x = input_657_cast_fp16, y = value_25_cast_fp16)[name = string("x_277_cast_fp16")];
+            tensor<int32, [4]> var_2256_perm_0 = const()[name = string("op_2256_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_2257 = const()[name = string("op_2257"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 126, 8, 128]> var_2256_cast_fp16 = transpose(perm = var_2256_perm_0, x = x_277_cast_fp16)[name = string("transpose_224")];
+            tensor<fp16, [1, 126, 1024]> input_659_cast_fp16 = reshape(shape = var_2257, x = var_2256_cast_fp16)[name = string("input_659_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_12_self_attn_linear_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(159983680))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(160508032))))[name = string("model_layers_12_self_attn_linear_out_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_115_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_12_self_attn_linear_out_weight_to_fp16_palettized, x = input_659_cast_fp16)[name = string("linear_115_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_663_cast_fp16 = add(x = input_655_cast_fp16, y = linear_115_cast_fp16)[name = string("input_663_cast_fp16")];
+            tensor<int32, [1]> x_281_axes_0 = const()[name = string("x_281_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_12_norm_conv_weight_to_fp16 = const()[name = string("model_layers_12_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(160509120)))];
+            tensor<fp16, [1024]> model_layers_12_norm_conv_bias_to_fp16 = const()[name = string("model_layers_12_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(160511232)))];
+            tensor<fp16, [1, 126, 1024]> x_281_cast_fp16 = layer_norm(axes = x_281_axes_0, beta = model_layers_12_norm_conv_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_12_norm_conv_weight_to_fp16, x = input_663_cast_fp16)[name = string("x_281_cast_fp16")];
+            tensor<int32, [3]> input_665_perm_0 = const()[name = string("input_665_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_667_pad_type_0 = const()[name = string("input_667_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_667_strides_0 = const()[name = string("input_667_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_667_pad_0 = const()[name = string("input_667_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_667_dilations_0 = const()[name = string("input_667_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_667_groups_0 = const()[name = string("input_667_groups_0"), val = int32(1)];
+            tensor<fp16, [2048, 1024, 1]> model_layers_12_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(160513344))), lut = tensor<fp16, [64, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161561984))))[name = string("model_layers_12_conv_pointwise_conv1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 1024, 126]> input_665_cast_fp16 = transpose(perm = input_665_perm_0, x = x_281_cast_fp16)[name = string("transpose_223")];
+            tensor<fp16, [1, 2048, 126]> input_667_cast_fp16 = conv(dilations = input_667_dilations_0, groups = input_667_groups_0, pad = input_667_pad_0, pad_type = input_667_pad_type_0, strides = input_667_strides_0, weight = model_layers_12_conv_pointwise_conv1_weight_to_fp16_palettized, x = input_665_cast_fp16)[name = string("input_667_cast_fp16")];
+            int32 x_283_split_num_splits_0 = const()[name = string("x_283_split_num_splits_0"), val = int32(2)];
+            int32 x_283_split_axis_0 = const()[name = string("x_283_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 126]> x_283_split_cast_fp16_0, tensor<fp16, [1, 1024, 126]> x_283_split_cast_fp16_1 = split(axis = x_283_split_axis_0, num_splits = x_283_split_num_splits_0, x = input_667_cast_fp16)[name = string("x_283_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_283_split_1_sigmoid_cast_fp16 = sigmoid(x = x_283_split_cast_fp16_1)[name = string("x_283_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_283_cast_fp16 = mul(x = x_283_split_cast_fp16_0, y = x_283_split_1_sigmoid_cast_fp16)[name = string("x_283_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_669_cast_fp16 = select(a = var_6_to_fp16, b = x_283_cast_fp16, cond = var_323)[name = string("input_669_cast_fp16")];
+            tensor<int32, [6]> input_671_pad_0 = const()[name = string("input_671_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_671_mode_0 = const()[name = string("input_671_mode_0"), val = string("constant")];
+            fp16 const_137_to_fp16 = const()[name = string("const_137_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 134]> input_671_cast_fp16 = pad(constant_val = const_137_to_fp16, mode = input_671_mode_0, pad = input_671_pad_0, x = input_669_cast_fp16)[name = string("input_671_cast_fp16")];
+            string input_673_pad_type_0 = const()[name = string("input_673_pad_type_0"), val = string("valid")];
+            int32 input_673_groups_0 = const()[name = string("input_673_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_673_strides_0 = const()[name = string("input_673_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_673_pad_0 = const()[name = string("input_673_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_673_dilations_0 = const()[name = string("input_673_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_272_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161564096))), lut = tensor<fp16, [32, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161568768))))[name = string("const_272_to_fp16_palettized")];
+            tensor<fp16, [1024]> const_273_to_fp16 = const()[name = string("const_273_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161569856)))];
+            tensor<fp16, [1, 1024, 126]> input_675_cast_fp16 = conv(bias = const_273_to_fp16, dilations = input_673_dilations_0, groups = input_673_groups_0, pad = input_673_pad_0, pad_type = input_673_pad_type_0, strides = input_673_strides_0, weight = const_272_to_fp16_palettized, x = input_671_cast_fp16)[name = string("input_675_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_677_cast_fp16 = silu(x = input_675_cast_fp16)[name = string("input_677_cast_fp16")];
+            string x_285_pad_type_0 = const()[name = string("x_285_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_285_strides_0 = const()[name = string("x_285_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_285_pad_0 = const()[name = string("x_285_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_285_dilations_0 = const()[name = string("x_285_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_285_groups_0 = const()[name = string("x_285_groups_0"), val = int32(1)];
+            tensor<fp16, [1024, 1024, 1]> model_layers_12_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161571968))), lut = tensor<fp16, [32, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162096320))))[name = string("model_layers_12_conv_pointwise_conv2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 1024, 126]> x_285_cast_fp16 = conv(dilations = x_285_dilations_0, groups = x_285_groups_0, pad = x_285_pad_0, pad_type = x_285_pad_type_0, strides = x_285_strides_0, weight = model_layers_12_conv_pointwise_conv2_weight_to_fp16_palettized, x = input_677_cast_fp16)[name = string("x_285_cast_fp16")];
+            tensor<int32, [3]> input_679_perm_0 = const()[name = string("input_679_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 126, 1024]> input_679_cast_fp16 = transpose(perm = input_679_perm_0, x = x_285_cast_fp16)[name = string("transpose_222")];
+            tensor<fp16, [1, 126, 1024]> input_681_cast_fp16 = add(x = input_663_cast_fp16, y = input_679_cast_fp16)[name = string("input_681_cast_fp16")];
+            tensor<int32, [1]> input_683_axes_0 = const()[name = string("input_683_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_12_norm_feed_forward2_weight_to_fp16 = const()[name = string("model_layers_12_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162097408)))];
+            tensor<fp16, [1024]> model_layers_12_norm_feed_forward2_bias_to_fp16 = const()[name = string("model_layers_12_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162099520)))];
+            tensor<fp16, [1, 126, 1024]> input_683_cast_fp16 = layer_norm(axes = input_683_axes_0, beta = model_layers_12_norm_feed_forward2_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_12_norm_feed_forward2_weight_to_fp16, x = input_681_cast_fp16)[name = string("input_683_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_12_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162101632))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164198848))))[name = string("model_layers_12_feed_forward2_linear1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 4096]> linear_116_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_12_feed_forward2_linear1_weight_to_fp16_palettized, x = input_683_cast_fp16)[name = string("linear_116_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_687_cast_fp16 = silu(x = linear_116_cast_fp16)[name = string("input_687_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_12_feed_forward2_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164203008))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166300224))))[name = string("model_layers_12_feed_forward2_linear2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_117_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_12_feed_forward2_linear2_weight_to_fp16_palettized, x = input_687_cast_fp16)[name = string("linear_117_cast_fp16")];
+            fp16 var_2317_to_fp16 = const()[name = string("op_2317_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_2318_cast_fp16 = mul(x = linear_117_cast_fp16, y = var_2317_to_fp16)[name = string("op_2318_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_693_cast_fp16 = add(x = input_681_cast_fp16, y = var_2318_cast_fp16)[name = string("input_693_cast_fp16")];
+            tensor<int32, [1]> input_695_axes_0 = const()[name = string("input_695_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_12_norm_out_weight_to_fp16 = const()[name = string("model_layers_12_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166301312)))];
+            tensor<fp16, [1024]> model_layers_12_norm_out_bias_to_fp16 = const()[name = string("model_layers_12_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166303424)))];
+            tensor<fp16, [1, 126, 1024]> input_695_cast_fp16 = layer_norm(axes = input_695_axes_0, beta = model_layers_12_norm_out_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_12_norm_out_weight_to_fp16, x = input_693_cast_fp16)[name = string("input_695_cast_fp16")];
+            tensor<int32, [1]> input_697_axes_0 = const()[name = string("input_697_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_13_norm_feed_forward1_weight_to_fp16 = const()[name = string("model_layers_13_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166305536)))];
+            tensor<fp16, [1024]> model_layers_13_norm_feed_forward1_bias_to_fp16 = const()[name = string("model_layers_13_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166307648)))];
+            tensor<fp16, [1, 126, 1024]> input_697_cast_fp16 = layer_norm(axes = input_697_axes_0, beta = model_layers_13_norm_feed_forward1_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_13_norm_feed_forward1_weight_to_fp16, x = input_695_cast_fp16)[name = string("input_697_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_13_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166309760))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(168406976))))[name = string("model_layers_13_feed_forward1_linear1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 4096]> linear_118_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_13_feed_forward1_linear1_weight_to_fp16_palettized, x = input_697_cast_fp16)[name = string("linear_118_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_701_cast_fp16 = silu(x = linear_118_cast_fp16)[name = string("input_701_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_13_feed_forward1_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(168411136))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170508352))))[name = string("model_layers_13_feed_forward1_linear2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_119_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_13_feed_forward1_linear2_weight_to_fp16_palettized, x = input_701_cast_fp16)[name = string("linear_119_cast_fp16")];
+            fp16 var_2346_to_fp16 = const()[name = string("op_2346_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_2347_cast_fp16 = mul(x = linear_119_cast_fp16, y = var_2346_to_fp16)[name = string("op_2347_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_707_cast_fp16 = add(x = input_695_cast_fp16, y = var_2347_cast_fp16)[name = string("input_707_cast_fp16")];
+            tensor<int32, [1]> query_27_axes_0 = const()[name = string("query_27_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_13_norm_self_att_weight_to_fp16 = const()[name = string("model_layers_13_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170509440)))];
+            tensor<fp16, [1024]> model_layers_13_norm_self_att_bias_to_fp16 = const()[name = string("model_layers_13_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170511552)))];
+            tensor<fp16, [1, 126, 1024]> query_27_cast_fp16 = layer_norm(axes = query_27_axes_0, beta = model_layers_13_norm_self_att_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_13_norm_self_att_weight_to_fp16, x = input_707_cast_fp16)[name = string("query_27_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_13_self_attn_linear_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170513664))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(171038016))))[name = string("model_layers_13_self_attn_linear_q_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_120_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_13_self_attn_linear_q_weight_to_fp16_palettized, x = query_27_cast_fp16)[name = string("linear_120_cast_fp16")];
+            tensor<int32, [4]> var_2363 = const()[name = string("op_2363"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> q_79_cast_fp16 = reshape(shape = var_2363, x = linear_120_cast_fp16)[name = string("q_79_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_13_self_attn_linear_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(171039104))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(171563456))))[name = string("model_layers_13_self_attn_linear_k_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_121_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_13_self_attn_linear_k_weight_to_fp16_palettized, x = query_27_cast_fp16)[name = string("linear_121_cast_fp16")];
+            tensor<int32, [4]> var_2367 = const()[name = string("op_2367"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> k_53_cast_fp16 = reshape(shape = var_2367, x = linear_121_cast_fp16)[name = string("k_53_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_13_self_attn_linear_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(171564544))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172088896))))[name = string("model_layers_13_self_attn_linear_v_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_122_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_13_self_attn_linear_v_weight_to_fp16_palettized, x = query_27_cast_fp16)[name = string("linear_122_cast_fp16")];
+            tensor<int32, [4]> var_2371 = const()[name = string("op_2371"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> v_27_cast_fp16 = reshape(shape = var_2371, x = linear_122_cast_fp16)[name = string("v_27_cast_fp16")];
+            tensor<int32, [4]> value_27_perm_0 = const()[name = string("value_27_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> model_layers_13_self_attn_pos_bias_u_to_fp16 = const()[name = string("model_layers_13_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172089984)))];
+            tensor<fp16, [1, 126, 8, 128]> var_2383_cast_fp16 = add(x = q_79_cast_fp16, y = model_layers_13_self_attn_pos_bias_u_to_fp16)[name = string("op_2383_cast_fp16")];
+            tensor<fp16, [8, 128]> model_layers_13_self_attn_pos_bias_v_to_fp16 = const()[name = string("model_layers_13_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172092096)))];
+            tensor<fp16, [1, 126, 8, 128]> var_2385_cast_fp16 = add(x = q_79_cast_fp16, y = model_layers_13_self_attn_pos_bias_v_to_fp16)[name = string("op_2385_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_27_perm_0 = const()[name = string("q_with_bias_v_27_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_293_transpose_x_0 = const()[name = string("x_293_transpose_x_0"), val = bool(false)];
+            bool x_293_transpose_y_0 = const()[name = string("x_293_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 251]> var_2387_to_fp16 = const()[name = string("op_2387_to_fp16"), val = tensor<fp16, [1, 8, 128, 251]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172094208)))];
+            tensor<fp16, [1, 8, 126, 128]> q_with_bias_v_27_cast_fp16 = transpose(perm = q_with_bias_v_27_perm_0, x = var_2385_cast_fp16)[name = string("transpose_221")];
+            tensor<fp16, [1, 8, 126, 251]> x_293_cast_fp16 = matmul(transpose_x = x_293_transpose_x_0, transpose_y = x_293_transpose_y_0, x = q_with_bias_v_27_cast_fp16, y = var_2387_to_fp16)[name = string("x_293_cast_fp16")];
+            tensor<int32, [8]> x_295_pad_0 = const()[name = string("x_295_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_295_mode_0 = const()[name = string("x_295_mode_0"), val = string("constant")];
+            fp16 const_144_to_fp16 = const()[name = string("const_144_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 252]> x_295_cast_fp16 = pad(constant_val = const_144_to_fp16, mode = x_295_mode_0, pad = x_295_pad_0, x = x_293_cast_fp16)[name = string("x_295_cast_fp16")];
+            tensor<int32, [4]> var_2395 = const()[name = string("op_2395"), val = tensor<int32, [4]>([1, 8, -1, 126])];
+            tensor<fp16, [1, 8, 252, 126]> x_297_cast_fp16 = reshape(shape = var_2395, x = x_295_cast_fp16)[name = string("x_297_cast_fp16")];
+            tensor<int32, [4]> var_2399_begin_0 = const()[name = string("op_2399_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_2399_end_0 = const()[name = string("op_2399_end_0"), val = tensor<int32, [4]>([1, 8, 252, 126])];
+            tensor<bool, [4]> var_2399_end_mask_0 = const()[name = string("op_2399_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 251, 126]> var_2399_cast_fp16 = slice_by_index(begin = var_2399_begin_0, end = var_2399_end_0, end_mask = var_2399_end_mask_0, x = x_297_cast_fp16)[name = string("op_2399_cast_fp16")];
+            tensor<int32, [4]> var_2400 = const()[name = string("op_2400"), val = tensor<int32, [4]>([1, 8, 126, 251])];
+            tensor<fp16, [1, 8, 126, 251]> matrix_bd_53_cast_fp16 = reshape(shape = var_2400, x = var_2399_cast_fp16)[name = string("matrix_bd_53_cast_fp16")];
+            bool matrix_ac_27_transpose_x_0 = const()[name = string("matrix_ac_27_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_27_transpose_y_0 = const()[name = string("matrix_ac_27_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_122_perm_0 = const()[name = string("transpose_122_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_123_perm_0 = const()[name = string("transpose_123_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 126]> transpose_123 = transpose(perm = transpose_123_perm_0, x = k_53_cast_fp16)[name = string("transpose_219")];
+            tensor<fp16, [1, 8, 126, 128]> transpose_122 = transpose(perm = transpose_122_perm_0, x = var_2383_cast_fp16)[name = string("transpose_220")];
+            tensor<fp16, [1, 8, 126, 126]> matrix_ac_27_cast_fp16 = matmul(transpose_x = matrix_ac_27_transpose_x_0, transpose_y = matrix_ac_27_transpose_y_0, x = transpose_122, y = transpose_123)[name = string("matrix_ac_27_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_55_begin_0 = const()[name = string("matrix_bd_55_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_55_end_0 = const()[name = string("matrix_bd_55_end_0"), val = tensor<int32, [4]>([1, 8, 126, 126])];
+            tensor<bool, [4]> matrix_bd_55_end_mask_0 = const()[name = string("matrix_bd_55_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 126, 126]> matrix_bd_55_cast_fp16 = slice_by_index(begin = matrix_bd_55_begin_0, end = matrix_bd_55_end_0, end_mask = matrix_bd_55_end_mask_0, x = matrix_bd_53_cast_fp16)[name = string("matrix_bd_55_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_2409_cast_fp16 = add(x = matrix_ac_27_cast_fp16, y = matrix_bd_55_cast_fp16)[name = string("op_2409_cast_fp16")];
+            fp16 _inversed_scores_53_y_0_to_fp16 = const()[name = string("_inversed_scores_53_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 126, 126]> _inversed_scores_53_cast_fp16 = mul(x = var_2409_cast_fp16, y = _inversed_scores_53_y_0_to_fp16)[name = string("_inversed_scores_53_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> scores_55_cast_fp16 = select(a = var_7_to_fp16, b = _inversed_scores_53_cast_fp16, cond = mask_3)[name = string("scores_55_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_2415_cast_fp16 = softmax(axis = var_25, x = scores_55_cast_fp16)[name = string("op_2415_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> input_709_cast_fp16 = select(a = var_6_to_fp16, b = var_2415_cast_fp16, cond = mask_3)[name = string("input_709_cast_fp16")];
+            bool x_299_transpose_x_0 = const()[name = string("x_299_transpose_x_0"), val = bool(false)];
+            bool x_299_transpose_y_0 = const()[name = string("x_299_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 126, 128]> value_27_cast_fp16 = transpose(perm = value_27_perm_0, x = v_27_cast_fp16)[name = string("transpose_218")];
+            tensor<fp16, [1, 8, 126, 128]> x_299_cast_fp16 = matmul(transpose_x = x_299_transpose_x_0, transpose_y = x_299_transpose_y_0, x = input_709_cast_fp16, y = value_27_cast_fp16)[name = string("x_299_cast_fp16")];
+            tensor<int32, [4]> var_2419_perm_0 = const()[name = string("op_2419_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_2420 = const()[name = string("op_2420"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 126, 8, 128]> var_2419_cast_fp16 = transpose(perm = var_2419_perm_0, x = x_299_cast_fp16)[name = string("transpose_217")];
+            tensor<fp16, [1, 126, 1024]> input_711_cast_fp16 = reshape(shape = var_2420, x = var_2419_cast_fp16)[name = string("input_711_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_13_self_attn_linear_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172608320))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(173132672))))[name = string("model_layers_13_self_attn_linear_out_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_124_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_13_self_attn_linear_out_weight_to_fp16_palettized, x = input_711_cast_fp16)[name = string("linear_124_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_715_cast_fp16 = add(x = input_707_cast_fp16, y = linear_124_cast_fp16)[name = string("input_715_cast_fp16")];
+            tensor<int32, [1]> x_303_axes_0 = const()[name = string("x_303_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_13_norm_conv_weight_to_fp16 = const()[name = string("model_layers_13_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(173133760)))];
+            tensor<fp16, [1024]> model_layers_13_norm_conv_bias_to_fp16 = const()[name = string("model_layers_13_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(173135872)))];
+            tensor<fp16, [1, 126, 1024]> x_303_cast_fp16 = layer_norm(axes = x_303_axes_0, beta = model_layers_13_norm_conv_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_13_norm_conv_weight_to_fp16, x = input_715_cast_fp16)[name = string("x_303_cast_fp16")];
+            tensor<int32, [3]> input_717_perm_0 = const()[name = string("input_717_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_719_pad_type_0 = const()[name = string("input_719_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_719_strides_0 = const()[name = string("input_719_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_719_pad_0 = const()[name = string("input_719_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_719_dilations_0 = const()[name = string("input_719_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_719_groups_0 = const()[name = string("input_719_groups_0"), val = int32(1)];
+            tensor<fp16, [2048, 1024, 1]> model_layers_13_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(173137984))), lut = tensor<fp16, [64, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174186624))))[name = string("model_layers_13_conv_pointwise_conv1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 1024, 126]> input_717_cast_fp16 = transpose(perm = input_717_perm_0, x = x_303_cast_fp16)[name = string("transpose_216")];
+            tensor<fp16, [1, 2048, 126]> input_719_cast_fp16 = conv(dilations = input_719_dilations_0, groups = input_719_groups_0, pad = input_719_pad_0, pad_type = input_719_pad_type_0, strides = input_719_strides_0, weight = model_layers_13_conv_pointwise_conv1_weight_to_fp16_palettized, x = input_717_cast_fp16)[name = string("input_719_cast_fp16")];
+            int32 x_305_split_num_splits_0 = const()[name = string("x_305_split_num_splits_0"), val = int32(2)];
+            int32 x_305_split_axis_0 = const()[name = string("x_305_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 126]> x_305_split_cast_fp16_0, tensor<fp16, [1, 1024, 126]> x_305_split_cast_fp16_1 = split(axis = x_305_split_axis_0, num_splits = x_305_split_num_splits_0, x = input_719_cast_fp16)[name = string("x_305_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_305_split_1_sigmoid_cast_fp16 = sigmoid(x = x_305_split_cast_fp16_1)[name = string("x_305_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_305_cast_fp16 = mul(x = x_305_split_cast_fp16_0, y = x_305_split_1_sigmoid_cast_fp16)[name = string("x_305_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_721_cast_fp16 = select(a = var_6_to_fp16, b = x_305_cast_fp16, cond = var_323)[name = string("input_721_cast_fp16")];
+            tensor<int32, [6]> input_723_pad_0 = const()[name = string("input_723_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_723_mode_0 = const()[name = string("input_723_mode_0"), val = string("constant")];
+            fp16 const_147_to_fp16 = const()[name = string("const_147_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 134]> input_723_cast_fp16 = pad(constant_val = const_147_to_fp16, mode = input_723_mode_0, pad = input_723_pad_0, x = input_721_cast_fp16)[name = string("input_723_cast_fp16")];
+            string input_725_pad_type_0 = const()[name = string("input_725_pad_type_0"), val = string("valid")];
+            int32 input_725_groups_0 = const()[name = string("input_725_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_725_strides_0 = const()[name = string("input_725_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_725_pad_0 = const()[name = string("input_725_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_725_dilations_0 = const()[name = string("input_725_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_274_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174188736))), lut = tensor<fp16, [32, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174193408))))[name = string("const_274_to_fp16_palettized")];
+            tensor<fp16, [1024]> const_275_to_fp16 = const()[name = string("const_275_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174194496)))];
+            tensor<fp16, [1, 1024, 126]> input_727_cast_fp16 = conv(bias = const_275_to_fp16, dilations = input_725_dilations_0, groups = input_725_groups_0, pad = input_725_pad_0, pad_type = input_725_pad_type_0, strides = input_725_strides_0, weight = const_274_to_fp16_palettized, x = input_723_cast_fp16)[name = string("input_727_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_729_cast_fp16 = silu(x = input_727_cast_fp16)[name = string("input_729_cast_fp16")];
+            string x_307_pad_type_0 = const()[name = string("x_307_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_307_strides_0 = const()[name = string("x_307_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_307_pad_0 = const()[name = string("x_307_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_307_dilations_0 = const()[name = string("x_307_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_307_groups_0 = const()[name = string("x_307_groups_0"), val = int32(1)];
+            tensor<fp16, [1024, 1024, 1]> model_layers_13_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174196608))), lut = tensor<fp16, [32, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174720960))))[name = string("model_layers_13_conv_pointwise_conv2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 1024, 126]> x_307_cast_fp16 = conv(dilations = x_307_dilations_0, groups = x_307_groups_0, pad = x_307_pad_0, pad_type = x_307_pad_type_0, strides = x_307_strides_0, weight = model_layers_13_conv_pointwise_conv2_weight_to_fp16_palettized, x = input_729_cast_fp16)[name = string("x_307_cast_fp16")];
+            tensor<int32, [3]> input_731_perm_0 = const()[name = string("input_731_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 126, 1024]> input_731_cast_fp16 = transpose(perm = input_731_perm_0, x = x_307_cast_fp16)[name = string("transpose_215")];
+            tensor<fp16, [1, 126, 1024]> input_733_cast_fp16 = add(x = input_715_cast_fp16, y = input_731_cast_fp16)[name = string("input_733_cast_fp16")];
+            tensor<int32, [1]> input_735_axes_0 = const()[name = string("input_735_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_13_norm_feed_forward2_weight_to_fp16 = const()[name = string("model_layers_13_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174722048)))];
+            tensor<fp16, [1024]> model_layers_13_norm_feed_forward2_bias_to_fp16 = const()[name = string("model_layers_13_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174724160)))];
+            tensor<fp16, [1, 126, 1024]> input_735_cast_fp16 = layer_norm(axes = input_735_axes_0, beta = model_layers_13_norm_feed_forward2_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_13_norm_feed_forward2_weight_to_fp16, x = input_733_cast_fp16)[name = string("input_735_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_13_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174726272))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176823488))))[name = string("model_layers_13_feed_forward2_linear1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 4096]> linear_125_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_13_feed_forward2_linear1_weight_to_fp16_palettized, x = input_735_cast_fp16)[name = string("linear_125_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_739_cast_fp16 = silu(x = linear_125_cast_fp16)[name = string("input_739_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_13_feed_forward2_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176827648))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178924864))))[name = string("model_layers_13_feed_forward2_linear2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_126_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_13_feed_forward2_linear2_weight_to_fp16_palettized, x = input_739_cast_fp16)[name = string("linear_126_cast_fp16")];
+            fp16 var_2480_to_fp16 = const()[name = string("op_2480_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_2481_cast_fp16 = mul(x = linear_126_cast_fp16, y = var_2480_to_fp16)[name = string("op_2481_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_745_cast_fp16 = add(x = input_733_cast_fp16, y = var_2481_cast_fp16)[name = string("input_745_cast_fp16")];
+            tensor<int32, [1]> input_747_axes_0 = const()[name = string("input_747_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_13_norm_out_weight_to_fp16 = const()[name = string("model_layers_13_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178925952)))];
+            tensor<fp16, [1024]> model_layers_13_norm_out_bias_to_fp16 = const()[name = string("model_layers_13_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178928064)))];
+            tensor<fp16, [1, 126, 1024]> input_747_cast_fp16 = layer_norm(axes = input_747_axes_0, beta = model_layers_13_norm_out_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_13_norm_out_weight_to_fp16, x = input_745_cast_fp16)[name = string("input_747_cast_fp16")];
+            tensor<int32, [1]> input_749_axes_0 = const()[name = string("input_749_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_14_norm_feed_forward1_weight_to_fp16 = const()[name = string("model_layers_14_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178930176)))];
+            tensor<fp16, [1024]> model_layers_14_norm_feed_forward1_bias_to_fp16 = const()[name = string("model_layers_14_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178932288)))];
+            tensor<fp16, [1, 126, 1024]> input_749_cast_fp16 = layer_norm(axes = input_749_axes_0, beta = model_layers_14_norm_feed_forward1_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_14_norm_feed_forward1_weight_to_fp16, x = input_747_cast_fp16)[name = string("input_749_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_14_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178934400))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181031616))))[name = string("model_layers_14_feed_forward1_linear1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 4096]> linear_127_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_14_feed_forward1_linear1_weight_to_fp16_palettized, x = input_749_cast_fp16)[name = string("linear_127_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_753_cast_fp16 = silu(x = linear_127_cast_fp16)[name = string("input_753_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_14_feed_forward1_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181035776))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183132992))))[name = string("model_layers_14_feed_forward1_linear2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_128_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_14_feed_forward1_linear2_weight_to_fp16_palettized, x = input_753_cast_fp16)[name = string("linear_128_cast_fp16")];
+            fp16 var_2509_to_fp16 = const()[name = string("op_2509_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_2510_cast_fp16 = mul(x = linear_128_cast_fp16, y = var_2509_to_fp16)[name = string("op_2510_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_759_cast_fp16 = add(x = input_747_cast_fp16, y = var_2510_cast_fp16)[name = string("input_759_cast_fp16")];
+            tensor<int32, [1]> query_29_axes_0 = const()[name = string("query_29_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_14_norm_self_att_weight_to_fp16 = const()[name = string("model_layers_14_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183134080)))];
+            tensor<fp16, [1024]> model_layers_14_norm_self_att_bias_to_fp16 = const()[name = string("model_layers_14_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183136192)))];
+            tensor<fp16, [1, 126, 1024]> query_29_cast_fp16 = layer_norm(axes = query_29_axes_0, beta = model_layers_14_norm_self_att_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_14_norm_self_att_weight_to_fp16, x = input_759_cast_fp16)[name = string("query_29_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_14_self_attn_linear_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183138304))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183662656))))[name = string("model_layers_14_self_attn_linear_q_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_129_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_14_self_attn_linear_q_weight_to_fp16_palettized, x = query_29_cast_fp16)[name = string("linear_129_cast_fp16")];
+            tensor<int32, [4]> var_2526 = const()[name = string("op_2526"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> q_85_cast_fp16 = reshape(shape = var_2526, x = linear_129_cast_fp16)[name = string("q_85_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_14_self_attn_linear_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183663744))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184188096))))[name = string("model_layers_14_self_attn_linear_k_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_130_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_14_self_attn_linear_k_weight_to_fp16_palettized, x = query_29_cast_fp16)[name = string("linear_130_cast_fp16")];
+            tensor<int32, [4]> var_2530 = const()[name = string("op_2530"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> k_57_cast_fp16 = reshape(shape = var_2530, x = linear_130_cast_fp16)[name = string("k_57_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_14_self_attn_linear_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184189184))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184713536))))[name = string("model_layers_14_self_attn_linear_v_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_131_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_14_self_attn_linear_v_weight_to_fp16_palettized, x = query_29_cast_fp16)[name = string("linear_131_cast_fp16")];
+            tensor<int32, [4]> var_2534 = const()[name = string("op_2534"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> v_29_cast_fp16 = reshape(shape = var_2534, x = linear_131_cast_fp16)[name = string("v_29_cast_fp16")];
+            tensor<int32, [4]> value_29_perm_0 = const()[name = string("value_29_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> model_layers_14_self_attn_pos_bias_u_to_fp16 = const()[name = string("model_layers_14_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184714624)))];
+            tensor<fp16, [1, 126, 8, 128]> var_2546_cast_fp16 = add(x = q_85_cast_fp16, y = model_layers_14_self_attn_pos_bias_u_to_fp16)[name = string("op_2546_cast_fp16")];
+            tensor<fp16, [8, 128]> model_layers_14_self_attn_pos_bias_v_to_fp16 = const()[name = string("model_layers_14_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184716736)))];
+            tensor<fp16, [1, 126, 8, 128]> var_2548_cast_fp16 = add(x = q_85_cast_fp16, y = model_layers_14_self_attn_pos_bias_v_to_fp16)[name = string("op_2548_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_29_perm_0 = const()[name = string("q_with_bias_v_29_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_315_transpose_x_0 = const()[name = string("x_315_transpose_x_0"), val = bool(false)];
+            bool x_315_transpose_y_0 = const()[name = string("x_315_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 251]> var_2550_to_fp16 = const()[name = string("op_2550_to_fp16"), val = tensor<fp16, [1, 8, 128, 251]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184718848)))];
+            tensor<fp16, [1, 8, 126, 128]> q_with_bias_v_29_cast_fp16 = transpose(perm = q_with_bias_v_29_perm_0, x = var_2548_cast_fp16)[name = string("transpose_214")];
+            tensor<fp16, [1, 8, 126, 251]> x_315_cast_fp16 = matmul(transpose_x = x_315_transpose_x_0, transpose_y = x_315_transpose_y_0, x = q_with_bias_v_29_cast_fp16, y = var_2550_to_fp16)[name = string("x_315_cast_fp16")];
+            tensor<int32, [8]> x_317_pad_0 = const()[name = string("x_317_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_317_mode_0 = const()[name = string("x_317_mode_0"), val = string("constant")];
+            fp16 const_154_to_fp16 = const()[name = string("const_154_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 252]> x_317_cast_fp16 = pad(constant_val = const_154_to_fp16, mode = x_317_mode_0, pad = x_317_pad_0, x = x_315_cast_fp16)[name = string("x_317_cast_fp16")];
+            tensor<int32, [4]> var_2558 = const()[name = string("op_2558"), val = tensor<int32, [4]>([1, 8, -1, 126])];
+            tensor<fp16, [1, 8, 252, 126]> x_319_cast_fp16 = reshape(shape = var_2558, x = x_317_cast_fp16)[name = string("x_319_cast_fp16")];
+            tensor<int32, [4]> var_2562_begin_0 = const()[name = string("op_2562_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_2562_end_0 = const()[name = string("op_2562_end_0"), val = tensor<int32, [4]>([1, 8, 252, 126])];
+            tensor<bool, [4]> var_2562_end_mask_0 = const()[name = string("op_2562_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 251, 126]> var_2562_cast_fp16 = slice_by_index(begin = var_2562_begin_0, end = var_2562_end_0, end_mask = var_2562_end_mask_0, x = x_319_cast_fp16)[name = string("op_2562_cast_fp16")];
+            tensor<int32, [4]> var_2563 = const()[name = string("op_2563"), val = tensor<int32, [4]>([1, 8, 126, 251])];
+            tensor<fp16, [1, 8, 126, 251]> matrix_bd_57_cast_fp16 = reshape(shape = var_2563, x = var_2562_cast_fp16)[name = string("matrix_bd_57_cast_fp16")];
+            bool matrix_ac_29_transpose_x_0 = const()[name = string("matrix_ac_29_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_29_transpose_y_0 = const()[name = string("matrix_ac_29_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_124_perm_0 = const()[name = string("transpose_124_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_125_perm_0 = const()[name = string("transpose_125_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 126]> transpose_125 = transpose(perm = transpose_125_perm_0, x = k_57_cast_fp16)[name = string("transpose_212")];
+            tensor<fp16, [1, 8, 126, 128]> transpose_124 = transpose(perm = transpose_124_perm_0, x = var_2546_cast_fp16)[name = string("transpose_213")];
+            tensor<fp16, [1, 8, 126, 126]> matrix_ac_29_cast_fp16 = matmul(transpose_x = matrix_ac_29_transpose_x_0, transpose_y = matrix_ac_29_transpose_y_0, x = transpose_124, y = transpose_125)[name = string("matrix_ac_29_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_59_begin_0 = const()[name = string("matrix_bd_59_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_59_end_0 = const()[name = string("matrix_bd_59_end_0"), val = tensor<int32, [4]>([1, 8, 126, 126])];
+            tensor<bool, [4]> matrix_bd_59_end_mask_0 = const()[name = string("matrix_bd_59_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 126, 126]> matrix_bd_59_cast_fp16 = slice_by_index(begin = matrix_bd_59_begin_0, end = matrix_bd_59_end_0, end_mask = matrix_bd_59_end_mask_0, x = matrix_bd_57_cast_fp16)[name = string("matrix_bd_59_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_2572_cast_fp16 = add(x = matrix_ac_29_cast_fp16, y = matrix_bd_59_cast_fp16)[name = string("op_2572_cast_fp16")];
+            fp16 _inversed_scores_57_y_0_to_fp16 = const()[name = string("_inversed_scores_57_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 126, 126]> _inversed_scores_57_cast_fp16 = mul(x = var_2572_cast_fp16, y = _inversed_scores_57_y_0_to_fp16)[name = string("_inversed_scores_57_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> scores_59_cast_fp16 = select(a = var_7_to_fp16, b = _inversed_scores_57_cast_fp16, cond = mask_3)[name = string("scores_59_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_2578_cast_fp16 = softmax(axis = var_25, x = scores_59_cast_fp16)[name = string("op_2578_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> input_761_cast_fp16 = select(a = var_6_to_fp16, b = var_2578_cast_fp16, cond = mask_3)[name = string("input_761_cast_fp16")];
+            bool x_321_transpose_x_0 = const()[name = string("x_321_transpose_x_0"), val = bool(false)];
+            bool x_321_transpose_y_0 = const()[name = string("x_321_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 126, 128]> value_29_cast_fp16 = transpose(perm = value_29_perm_0, x = v_29_cast_fp16)[name = string("transpose_211")];
+            tensor<fp16, [1, 8, 126, 128]> x_321_cast_fp16 = matmul(transpose_x = x_321_transpose_x_0, transpose_y = x_321_transpose_y_0, x = input_761_cast_fp16, y = value_29_cast_fp16)[name = string("x_321_cast_fp16")];
+            tensor<int32, [4]> var_2582_perm_0 = const()[name = string("op_2582_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_2583 = const()[name = string("op_2583"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 126, 8, 128]> var_2582_cast_fp16 = transpose(perm = var_2582_perm_0, x = x_321_cast_fp16)[name = string("transpose_210")];
+            tensor<fp16, [1, 126, 1024]> input_763_cast_fp16 = reshape(shape = var_2583, x = var_2582_cast_fp16)[name = string("input_763_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_14_self_attn_linear_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185232960))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185757312))))[name = string("model_layers_14_self_attn_linear_out_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_133_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_14_self_attn_linear_out_weight_to_fp16_palettized, x = input_763_cast_fp16)[name = string("linear_133_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_767_cast_fp16 = add(x = input_759_cast_fp16, y = linear_133_cast_fp16)[name = string("input_767_cast_fp16")];
+            tensor<int32, [1]> x_325_axes_0 = const()[name = string("x_325_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_14_norm_conv_weight_to_fp16 = const()[name = string("model_layers_14_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185758400)))];
+            tensor<fp16, [1024]> model_layers_14_norm_conv_bias_to_fp16 = const()[name = string("model_layers_14_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185760512)))];
+            tensor<fp16, [1, 126, 1024]> x_325_cast_fp16 = layer_norm(axes = x_325_axes_0, beta = model_layers_14_norm_conv_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_14_norm_conv_weight_to_fp16, x = input_767_cast_fp16)[name = string("x_325_cast_fp16")];
+            tensor<int32, [3]> input_769_perm_0 = const()[name = string("input_769_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_771_pad_type_0 = const()[name = string("input_771_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_771_strides_0 = const()[name = string("input_771_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_771_pad_0 = const()[name = string("input_771_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_771_dilations_0 = const()[name = string("input_771_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_771_groups_0 = const()[name = string("input_771_groups_0"), val = int32(1)];
+            tensor<fp16, [2048, 1024, 1]> model_layers_14_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185762624))), lut = tensor<fp16, [64, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186811264))))[name = string("model_layers_14_conv_pointwise_conv1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 1024, 126]> input_769_cast_fp16 = transpose(perm = input_769_perm_0, x = x_325_cast_fp16)[name = string("transpose_209")];
+            tensor<fp16, [1, 2048, 126]> input_771_cast_fp16 = conv(dilations = input_771_dilations_0, groups = input_771_groups_0, pad = input_771_pad_0, pad_type = input_771_pad_type_0, strides = input_771_strides_0, weight = model_layers_14_conv_pointwise_conv1_weight_to_fp16_palettized, x = input_769_cast_fp16)[name = string("input_771_cast_fp16")];
+            int32 x_327_split_num_splits_0 = const()[name = string("x_327_split_num_splits_0"), val = int32(2)];
+            int32 x_327_split_axis_0 = const()[name = string("x_327_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 126]> x_327_split_cast_fp16_0, tensor<fp16, [1, 1024, 126]> x_327_split_cast_fp16_1 = split(axis = x_327_split_axis_0, num_splits = x_327_split_num_splits_0, x = input_771_cast_fp16)[name = string("x_327_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_327_split_1_sigmoid_cast_fp16 = sigmoid(x = x_327_split_cast_fp16_1)[name = string("x_327_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_327_cast_fp16 = mul(x = x_327_split_cast_fp16_0, y = x_327_split_1_sigmoid_cast_fp16)[name = string("x_327_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_773_cast_fp16 = select(a = var_6_to_fp16, b = x_327_cast_fp16, cond = var_323)[name = string("input_773_cast_fp16")];
+            tensor<int32, [6]> input_775_pad_0 = const()[name = string("input_775_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_775_mode_0 = const()[name = string("input_775_mode_0"), val = string("constant")];
+            fp16 const_157_to_fp16 = const()[name = string("const_157_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 134]> input_775_cast_fp16 = pad(constant_val = const_157_to_fp16, mode = input_775_mode_0, pad = input_775_pad_0, x = input_773_cast_fp16)[name = string("input_775_cast_fp16")];
+            string input_777_pad_type_0 = const()[name = string("input_777_pad_type_0"), val = string("valid")];
+            int32 input_777_groups_0 = const()[name = string("input_777_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_777_strides_0 = const()[name = string("input_777_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_777_pad_0 = const()[name = string("input_777_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_777_dilations_0 = const()[name = string("input_777_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_276_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186813376))), lut = tensor<fp16, [32, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186818048))))[name = string("const_276_to_fp16_palettized")];
+            tensor<fp16, [1024]> const_277_to_fp16 = const()[name = string("const_277_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186819136)))];
+            tensor<fp16, [1, 1024, 126]> input_779_cast_fp16 = conv(bias = const_277_to_fp16, dilations = input_777_dilations_0, groups = input_777_groups_0, pad = input_777_pad_0, pad_type = input_777_pad_type_0, strides = input_777_strides_0, weight = const_276_to_fp16_palettized, x = input_775_cast_fp16)[name = string("input_779_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_781_cast_fp16 = silu(x = input_779_cast_fp16)[name = string("input_781_cast_fp16")];
+            string x_329_pad_type_0 = const()[name = string("x_329_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_329_strides_0 = const()[name = string("x_329_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_329_pad_0 = const()[name = string("x_329_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_329_dilations_0 = const()[name = string("x_329_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_329_groups_0 = const()[name = string("x_329_groups_0"), val = int32(1)];
+            tensor<fp16, [1024, 1024, 1]> model_layers_14_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186821248))), lut = tensor<fp16, [32, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187345600))))[name = string("model_layers_14_conv_pointwise_conv2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 1024, 126]> x_329_cast_fp16 = conv(dilations = x_329_dilations_0, groups = x_329_groups_0, pad = x_329_pad_0, pad_type = x_329_pad_type_0, strides = x_329_strides_0, weight = model_layers_14_conv_pointwise_conv2_weight_to_fp16_palettized, x = input_781_cast_fp16)[name = string("x_329_cast_fp16")];
+            tensor<int32, [3]> input_783_perm_0 = const()[name = string("input_783_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 126, 1024]> input_783_cast_fp16 = transpose(perm = input_783_perm_0, x = x_329_cast_fp16)[name = string("transpose_208")];
+            tensor<fp16, [1, 126, 1024]> input_785_cast_fp16 = add(x = input_767_cast_fp16, y = input_783_cast_fp16)[name = string("input_785_cast_fp16")];
+            tensor<int32, [1]> input_787_axes_0 = const()[name = string("input_787_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_14_norm_feed_forward2_weight_to_fp16 = const()[name = string("model_layers_14_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187346688)))];
+            tensor<fp16, [1024]> model_layers_14_norm_feed_forward2_bias_to_fp16 = const()[name = string("model_layers_14_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187348800)))];
+            tensor<fp16, [1, 126, 1024]> input_787_cast_fp16 = layer_norm(axes = input_787_axes_0, beta = model_layers_14_norm_feed_forward2_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_14_norm_feed_forward2_weight_to_fp16, x = input_785_cast_fp16)[name = string("input_787_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_14_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187350912))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189448128))))[name = string("model_layers_14_feed_forward2_linear1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 4096]> linear_134_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_14_feed_forward2_linear1_weight_to_fp16_palettized, x = input_787_cast_fp16)[name = string("linear_134_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_791_cast_fp16 = silu(x = linear_134_cast_fp16)[name = string("input_791_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_14_feed_forward2_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189452288))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(191549504))))[name = string("model_layers_14_feed_forward2_linear2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_135_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_14_feed_forward2_linear2_weight_to_fp16_palettized, x = input_791_cast_fp16)[name = string("linear_135_cast_fp16")];
+            fp16 var_2643_to_fp16 = const()[name = string("op_2643_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_2644_cast_fp16 = mul(x = linear_135_cast_fp16, y = var_2643_to_fp16)[name = string("op_2644_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_797_cast_fp16 = add(x = input_785_cast_fp16, y = var_2644_cast_fp16)[name = string("input_797_cast_fp16")];
+            tensor<int32, [1]> input_799_axes_0 = const()[name = string("input_799_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_14_norm_out_weight_to_fp16 = const()[name = string("model_layers_14_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(191550592)))];
+            tensor<fp16, [1024]> model_layers_14_norm_out_bias_to_fp16 = const()[name = string("model_layers_14_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(191552704)))];
+            tensor<fp16, [1, 126, 1024]> input_799_cast_fp16 = layer_norm(axes = input_799_axes_0, beta = model_layers_14_norm_out_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_14_norm_out_weight_to_fp16, x = input_797_cast_fp16)[name = string("input_799_cast_fp16")];
+            tensor<int32, [1]> input_801_axes_0 = const()[name = string("input_801_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_15_norm_feed_forward1_weight_to_fp16 = const()[name = string("model_layers_15_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(191554816)))];
+            tensor<fp16, [1024]> model_layers_15_norm_feed_forward1_bias_to_fp16 = const()[name = string("model_layers_15_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(191556928)))];
+            tensor<fp16, [1, 126, 1024]> input_801_cast_fp16 = layer_norm(axes = input_801_axes_0, beta = model_layers_15_norm_feed_forward1_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_15_norm_feed_forward1_weight_to_fp16, x = input_799_cast_fp16)[name = string("input_801_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_15_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(191559040))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(193656256))))[name = string("model_layers_15_feed_forward1_linear1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 4096]> linear_136_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_15_feed_forward1_linear1_weight_to_fp16_palettized, x = input_801_cast_fp16)[name = string("linear_136_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_805_cast_fp16 = silu(x = linear_136_cast_fp16)[name = string("input_805_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_15_feed_forward1_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(193660416))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195757632))))[name = string("model_layers_15_feed_forward1_linear2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_137_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_15_feed_forward1_linear2_weight_to_fp16_palettized, x = input_805_cast_fp16)[name = string("linear_137_cast_fp16")];
+            fp16 var_2672_to_fp16 = const()[name = string("op_2672_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_2673_cast_fp16 = mul(x = linear_137_cast_fp16, y = var_2672_to_fp16)[name = string("op_2673_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_811_cast_fp16 = add(x = input_799_cast_fp16, y = var_2673_cast_fp16)[name = string("input_811_cast_fp16")];
+            tensor<int32, [1]> query_31_axes_0 = const()[name = string("query_31_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_15_norm_self_att_weight_to_fp16 = const()[name = string("model_layers_15_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195758720)))];
+            tensor<fp16, [1024]> model_layers_15_norm_self_att_bias_to_fp16 = const()[name = string("model_layers_15_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195760832)))];
+            tensor<fp16, [1, 126, 1024]> query_31_cast_fp16 = layer_norm(axes = query_31_axes_0, beta = model_layers_15_norm_self_att_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_15_norm_self_att_weight_to_fp16, x = input_811_cast_fp16)[name = string("query_31_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_15_self_attn_linear_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195762944))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196287296))))[name = string("model_layers_15_self_attn_linear_q_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_138_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_15_self_attn_linear_q_weight_to_fp16_palettized, x = query_31_cast_fp16)[name = string("linear_138_cast_fp16")];
+            tensor<int32, [4]> var_2689 = const()[name = string("op_2689"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> q_91_cast_fp16 = reshape(shape = var_2689, x = linear_138_cast_fp16)[name = string("q_91_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_15_self_attn_linear_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196288384))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196812736))))[name = string("model_layers_15_self_attn_linear_k_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_139_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_15_self_attn_linear_k_weight_to_fp16_palettized, x = query_31_cast_fp16)[name = string("linear_139_cast_fp16")];
+            tensor<int32, [4]> var_2693 = const()[name = string("op_2693"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> k_61_cast_fp16 = reshape(shape = var_2693, x = linear_139_cast_fp16)[name = string("k_61_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_15_self_attn_linear_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196813824))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197338176))))[name = string("model_layers_15_self_attn_linear_v_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_140_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_15_self_attn_linear_v_weight_to_fp16_palettized, x = query_31_cast_fp16)[name = string("linear_140_cast_fp16")];
+            tensor<int32, [4]> var_2697 = const()[name = string("op_2697"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> v_31_cast_fp16 = reshape(shape = var_2697, x = linear_140_cast_fp16)[name = string("v_31_cast_fp16")];
+            tensor<int32, [4]> value_31_perm_0 = const()[name = string("value_31_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> model_layers_15_self_attn_pos_bias_u_to_fp16 = const()[name = string("model_layers_15_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197339264)))];
+            tensor<fp16, [1, 126, 8, 128]> var_2709_cast_fp16 = add(x = q_91_cast_fp16, y = model_layers_15_self_attn_pos_bias_u_to_fp16)[name = string("op_2709_cast_fp16")];
+            tensor<fp16, [8, 128]> model_layers_15_self_attn_pos_bias_v_to_fp16 = const()[name = string("model_layers_15_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197341376)))];
+            tensor<fp16, [1, 126, 8, 128]> var_2711_cast_fp16 = add(x = q_91_cast_fp16, y = model_layers_15_self_attn_pos_bias_v_to_fp16)[name = string("op_2711_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_31_perm_0 = const()[name = string("q_with_bias_v_31_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_337_transpose_x_0 = const()[name = string("x_337_transpose_x_0"), val = bool(false)];
+            bool x_337_transpose_y_0 = const()[name = string("x_337_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 251]> var_2713_to_fp16 = const()[name = string("op_2713_to_fp16"), val = tensor<fp16, [1, 8, 128, 251]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197343488)))];
+            tensor<fp16, [1, 8, 126, 128]> q_with_bias_v_31_cast_fp16 = transpose(perm = q_with_bias_v_31_perm_0, x = var_2711_cast_fp16)[name = string("transpose_207")];
+            tensor<fp16, [1, 8, 126, 251]> x_337_cast_fp16 = matmul(transpose_x = x_337_transpose_x_0, transpose_y = x_337_transpose_y_0, x = q_with_bias_v_31_cast_fp16, y = var_2713_to_fp16)[name = string("x_337_cast_fp16")];
+            tensor<int32, [8]> x_339_pad_0 = const()[name = string("x_339_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_339_mode_0 = const()[name = string("x_339_mode_0"), val = string("constant")];
+            fp16 const_164_to_fp16 = const()[name = string("const_164_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 252]> x_339_cast_fp16 = pad(constant_val = const_164_to_fp16, mode = x_339_mode_0, pad = x_339_pad_0, x = x_337_cast_fp16)[name = string("x_339_cast_fp16")];
+            tensor<int32, [4]> var_2721 = const()[name = string("op_2721"), val = tensor<int32, [4]>([1, 8, -1, 126])];
+            tensor<fp16, [1, 8, 252, 126]> x_341_cast_fp16 = reshape(shape = var_2721, x = x_339_cast_fp16)[name = string("x_341_cast_fp16")];
+            tensor<int32, [4]> var_2725_begin_0 = const()[name = string("op_2725_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_2725_end_0 = const()[name = string("op_2725_end_0"), val = tensor<int32, [4]>([1, 8, 252, 126])];
+            tensor<bool, [4]> var_2725_end_mask_0 = const()[name = string("op_2725_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 251, 126]> var_2725_cast_fp16 = slice_by_index(begin = var_2725_begin_0, end = var_2725_end_0, end_mask = var_2725_end_mask_0, x = x_341_cast_fp16)[name = string("op_2725_cast_fp16")];
+            tensor<int32, [4]> var_2726 = const()[name = string("op_2726"), val = tensor<int32, [4]>([1, 8, 126, 251])];
+            tensor<fp16, [1, 8, 126, 251]> matrix_bd_61_cast_fp16 = reshape(shape = var_2726, x = var_2725_cast_fp16)[name = string("matrix_bd_61_cast_fp16")];
+            bool matrix_ac_31_transpose_x_0 = const()[name = string("matrix_ac_31_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_31_transpose_y_0 = const()[name = string("matrix_ac_31_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_126_perm_0 = const()[name = string("transpose_126_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_127_perm_0 = const()[name = string("transpose_127_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 126]> transpose_127 = transpose(perm = transpose_127_perm_0, x = k_61_cast_fp16)[name = string("transpose_205")];
+            tensor<fp16, [1, 8, 126, 128]> transpose_126 = transpose(perm = transpose_126_perm_0, x = var_2709_cast_fp16)[name = string("transpose_206")];
+            tensor<fp16, [1, 8, 126, 126]> matrix_ac_31_cast_fp16 = matmul(transpose_x = matrix_ac_31_transpose_x_0, transpose_y = matrix_ac_31_transpose_y_0, x = transpose_126, y = transpose_127)[name = string("matrix_ac_31_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_63_begin_0 = const()[name = string("matrix_bd_63_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_63_end_0 = const()[name = string("matrix_bd_63_end_0"), val = tensor<int32, [4]>([1, 8, 126, 126])];
+            tensor<bool, [4]> matrix_bd_63_end_mask_0 = const()[name = string("matrix_bd_63_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 126, 126]> matrix_bd_63_cast_fp16 = slice_by_index(begin = matrix_bd_63_begin_0, end = matrix_bd_63_end_0, end_mask = matrix_bd_63_end_mask_0, x = matrix_bd_61_cast_fp16)[name = string("matrix_bd_63_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_2735_cast_fp16 = add(x = matrix_ac_31_cast_fp16, y = matrix_bd_63_cast_fp16)[name = string("op_2735_cast_fp16")];
+            fp16 _inversed_scores_61_y_0_to_fp16 = const()[name = string("_inversed_scores_61_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 126, 126]> _inversed_scores_61_cast_fp16 = mul(x = var_2735_cast_fp16, y = _inversed_scores_61_y_0_to_fp16)[name = string("_inversed_scores_61_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> scores_63_cast_fp16 = select(a = var_7_to_fp16, b = _inversed_scores_61_cast_fp16, cond = mask_3)[name = string("scores_63_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_2741_cast_fp16 = softmax(axis = var_25, x = scores_63_cast_fp16)[name = string("op_2741_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> input_813_cast_fp16 = select(a = var_6_to_fp16, b = var_2741_cast_fp16, cond = mask_3)[name = string("input_813_cast_fp16")];
+            bool x_343_transpose_x_0 = const()[name = string("x_343_transpose_x_0"), val = bool(false)];
+            bool x_343_transpose_y_0 = const()[name = string("x_343_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 126, 128]> value_31_cast_fp16 = transpose(perm = value_31_perm_0, x = v_31_cast_fp16)[name = string("transpose_204")];
+            tensor<fp16, [1, 8, 126, 128]> x_343_cast_fp16 = matmul(transpose_x = x_343_transpose_x_0, transpose_y = x_343_transpose_y_0, x = input_813_cast_fp16, y = value_31_cast_fp16)[name = string("x_343_cast_fp16")];
+            tensor<int32, [4]> var_2745_perm_0 = const()[name = string("op_2745_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_2746 = const()[name = string("op_2746"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 126, 8, 128]> var_2745_cast_fp16 = transpose(perm = var_2745_perm_0, x = x_343_cast_fp16)[name = string("transpose_203")];
+            tensor<fp16, [1, 126, 1024]> input_815_cast_fp16 = reshape(shape = var_2746, x = var_2745_cast_fp16)[name = string("input_815_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_15_self_attn_linear_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197857600))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198381952))))[name = string("model_layers_15_self_attn_linear_out_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_142_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_15_self_attn_linear_out_weight_to_fp16_palettized, x = input_815_cast_fp16)[name = string("linear_142_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_819_cast_fp16 = add(x = input_811_cast_fp16, y = linear_142_cast_fp16)[name = string("input_819_cast_fp16")];
+            tensor<int32, [1]> x_347_axes_0 = const()[name = string("x_347_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_15_norm_conv_weight_to_fp16 = const()[name = string("model_layers_15_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198383040)))];
+            tensor<fp16, [1024]> model_layers_15_norm_conv_bias_to_fp16 = const()[name = string("model_layers_15_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198385152)))];
+            tensor<fp16, [1, 126, 1024]> x_347_cast_fp16 = layer_norm(axes = x_347_axes_0, beta = model_layers_15_norm_conv_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_15_norm_conv_weight_to_fp16, x = input_819_cast_fp16)[name = string("x_347_cast_fp16")];
+            tensor<int32, [3]> input_821_perm_0 = const()[name = string("input_821_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_823_pad_type_0 = const()[name = string("input_823_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_823_strides_0 = const()[name = string("input_823_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_823_pad_0 = const()[name = string("input_823_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_823_dilations_0 = const()[name = string("input_823_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_823_groups_0 = const()[name = string("input_823_groups_0"), val = int32(1)];
+            tensor<fp16, [2048, 1024, 1]> model_layers_15_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198387264))), lut = tensor<fp16, [64, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199435904))))[name = string("model_layers_15_conv_pointwise_conv1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 1024, 126]> input_821_cast_fp16 = transpose(perm = input_821_perm_0, x = x_347_cast_fp16)[name = string("transpose_202")];
+            tensor<fp16, [1, 2048, 126]> input_823_cast_fp16 = conv(dilations = input_823_dilations_0, groups = input_823_groups_0, pad = input_823_pad_0, pad_type = input_823_pad_type_0, strides = input_823_strides_0, weight = model_layers_15_conv_pointwise_conv1_weight_to_fp16_palettized, x = input_821_cast_fp16)[name = string("input_823_cast_fp16")];
+            int32 x_349_split_num_splits_0 = const()[name = string("x_349_split_num_splits_0"), val = int32(2)];
+            int32 x_349_split_axis_0 = const()[name = string("x_349_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 126]> x_349_split_cast_fp16_0, tensor<fp16, [1, 1024, 126]> x_349_split_cast_fp16_1 = split(axis = x_349_split_axis_0, num_splits = x_349_split_num_splits_0, x = input_823_cast_fp16)[name = string("x_349_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_349_split_1_sigmoid_cast_fp16 = sigmoid(x = x_349_split_cast_fp16_1)[name = string("x_349_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_349_cast_fp16 = mul(x = x_349_split_cast_fp16_0, y = x_349_split_1_sigmoid_cast_fp16)[name = string("x_349_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_825_cast_fp16 = select(a = var_6_to_fp16, b = x_349_cast_fp16, cond = var_323)[name = string("input_825_cast_fp16")];
+            tensor<int32, [6]> input_827_pad_0 = const()[name = string("input_827_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_827_mode_0 = const()[name = string("input_827_mode_0"), val = string("constant")];
+            fp16 const_167_to_fp16 = const()[name = string("const_167_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 134]> input_827_cast_fp16 = pad(constant_val = const_167_to_fp16, mode = input_827_mode_0, pad = input_827_pad_0, x = input_825_cast_fp16)[name = string("input_827_cast_fp16")];
+            string input_829_pad_type_0 = const()[name = string("input_829_pad_type_0"), val = string("valid")];
+            int32 input_829_groups_0 = const()[name = string("input_829_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_829_strides_0 = const()[name = string("input_829_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_829_pad_0 = const()[name = string("input_829_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_829_dilations_0 = const()[name = string("input_829_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_278_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199438016))), lut = tensor<fp16, [32, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199442688))))[name = string("const_278_to_fp16_palettized")];
+            tensor<fp16, [1024]> const_279_to_fp16 = const()[name = string("const_279_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199443776)))];
+            tensor<fp16, [1, 1024, 126]> input_831_cast_fp16 = conv(bias = const_279_to_fp16, dilations = input_829_dilations_0, groups = input_829_groups_0, pad = input_829_pad_0, pad_type = input_829_pad_type_0, strides = input_829_strides_0, weight = const_278_to_fp16_palettized, x = input_827_cast_fp16)[name = string("input_831_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_833_cast_fp16 = silu(x = input_831_cast_fp16)[name = string("input_833_cast_fp16")];
+            string x_351_pad_type_0 = const()[name = string("x_351_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_351_strides_0 = const()[name = string("x_351_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_351_pad_0 = const()[name = string("x_351_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_351_dilations_0 = const()[name = string("x_351_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_351_groups_0 = const()[name = string("x_351_groups_0"), val = int32(1)];
+            tensor<fp16, [1024, 1024, 1]> model_layers_15_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199445888))), lut = tensor<fp16, [32, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199970240))))[name = string("model_layers_15_conv_pointwise_conv2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 1024, 126]> x_351_cast_fp16 = conv(dilations = x_351_dilations_0, groups = x_351_groups_0, pad = x_351_pad_0, pad_type = x_351_pad_type_0, strides = x_351_strides_0, weight = model_layers_15_conv_pointwise_conv2_weight_to_fp16_palettized, x = input_833_cast_fp16)[name = string("x_351_cast_fp16")];
+            tensor<int32, [3]> input_835_perm_0 = const()[name = string("input_835_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 126, 1024]> input_835_cast_fp16 = transpose(perm = input_835_perm_0, x = x_351_cast_fp16)[name = string("transpose_201")];
+            tensor<fp16, [1, 126, 1024]> input_837_cast_fp16 = add(x = input_819_cast_fp16, y = input_835_cast_fp16)[name = string("input_837_cast_fp16")];
+            tensor<int32, [1]> input_839_axes_0 = const()[name = string("input_839_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_15_norm_feed_forward2_weight_to_fp16 = const()[name = string("model_layers_15_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199971328)))];
+            tensor<fp16, [1024]> model_layers_15_norm_feed_forward2_bias_to_fp16 = const()[name = string("model_layers_15_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199973440)))];
+            tensor<fp16, [1, 126, 1024]> input_839_cast_fp16 = layer_norm(axes = input_839_axes_0, beta = model_layers_15_norm_feed_forward2_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_15_norm_feed_forward2_weight_to_fp16, x = input_837_cast_fp16)[name = string("input_839_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_15_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199975552))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202072768))))[name = string("model_layers_15_feed_forward2_linear1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 4096]> linear_143_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_15_feed_forward2_linear1_weight_to_fp16_palettized, x = input_839_cast_fp16)[name = string("linear_143_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_843_cast_fp16 = silu(x = linear_143_cast_fp16)[name = string("input_843_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_15_feed_forward2_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202076928))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(204174144))))[name = string("model_layers_15_feed_forward2_linear2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_144_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_15_feed_forward2_linear2_weight_to_fp16_palettized, x = input_843_cast_fp16)[name = string("linear_144_cast_fp16")];
+            fp16 var_2806_to_fp16 = const()[name = string("op_2806_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_2807_cast_fp16 = mul(x = linear_144_cast_fp16, y = var_2806_to_fp16)[name = string("op_2807_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_849_cast_fp16 = add(x = input_837_cast_fp16, y = var_2807_cast_fp16)[name = string("input_849_cast_fp16")];
+            tensor<int32, [1]> input_851_axes_0 = const()[name = string("input_851_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_15_norm_out_weight_to_fp16 = const()[name = string("model_layers_15_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(204175232)))];
+            tensor<fp16, [1024]> model_layers_15_norm_out_bias_to_fp16 = const()[name = string("model_layers_15_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(204177344)))];
+            tensor<fp16, [1, 126, 1024]> input_851_cast_fp16 = layer_norm(axes = input_851_axes_0, beta = model_layers_15_norm_out_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_15_norm_out_weight_to_fp16, x = input_849_cast_fp16)[name = string("input_851_cast_fp16")];
+            tensor<int32, [1]> input_853_axes_0 = const()[name = string("input_853_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_16_norm_feed_forward1_weight_to_fp16 = const()[name = string("model_layers_16_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(204179456)))];
+            tensor<fp16, [1024]> model_layers_16_norm_feed_forward1_bias_to_fp16 = const()[name = string("model_layers_16_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(204181568)))];
+            tensor<fp16, [1, 126, 1024]> input_853_cast_fp16 = layer_norm(axes = input_853_axes_0, beta = model_layers_16_norm_feed_forward1_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_16_norm_feed_forward1_weight_to_fp16, x = input_851_cast_fp16)[name = string("input_853_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_16_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(204183680))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(206280896))))[name = string("model_layers_16_feed_forward1_linear1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 4096]> linear_145_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_16_feed_forward1_linear1_weight_to_fp16_palettized, x = input_853_cast_fp16)[name = string("linear_145_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_857_cast_fp16 = silu(x = linear_145_cast_fp16)[name = string("input_857_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_16_feed_forward1_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(206285056))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(208382272))))[name = string("model_layers_16_feed_forward1_linear2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_146_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_16_feed_forward1_linear2_weight_to_fp16_palettized, x = input_857_cast_fp16)[name = string("linear_146_cast_fp16")];
+            fp16 var_2835_to_fp16 = const()[name = string("op_2835_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_2836_cast_fp16 = mul(x = linear_146_cast_fp16, y = var_2835_to_fp16)[name = string("op_2836_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_863_cast_fp16 = add(x = input_851_cast_fp16, y = var_2836_cast_fp16)[name = string("input_863_cast_fp16")];
+            tensor<int32, [1]> query_33_axes_0 = const()[name = string("query_33_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_16_norm_self_att_weight_to_fp16 = const()[name = string("model_layers_16_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(208383360)))];
+            tensor<fp16, [1024]> model_layers_16_norm_self_att_bias_to_fp16 = const()[name = string("model_layers_16_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(208385472)))];
+            tensor<fp16, [1, 126, 1024]> query_33_cast_fp16 = layer_norm(axes = query_33_axes_0, beta = model_layers_16_norm_self_att_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_16_norm_self_att_weight_to_fp16, x = input_863_cast_fp16)[name = string("query_33_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_16_self_attn_linear_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(208387584))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(208911936))))[name = string("model_layers_16_self_attn_linear_q_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_147_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_16_self_attn_linear_q_weight_to_fp16_palettized, x = query_33_cast_fp16)[name = string("linear_147_cast_fp16")];
+            tensor<int32, [4]> var_2852 = const()[name = string("op_2852"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> q_97_cast_fp16 = reshape(shape = var_2852, x = linear_147_cast_fp16)[name = string("q_97_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_16_self_attn_linear_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(208913024))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(209437376))))[name = string("model_layers_16_self_attn_linear_k_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_148_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_16_self_attn_linear_k_weight_to_fp16_palettized, x = query_33_cast_fp16)[name = string("linear_148_cast_fp16")];
+            tensor<int32, [4]> var_2856 = const()[name = string("op_2856"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> k_65_cast_fp16 = reshape(shape = var_2856, x = linear_148_cast_fp16)[name = string("k_65_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_16_self_attn_linear_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(209438464))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(209962816))))[name = string("model_layers_16_self_attn_linear_v_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_149_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_16_self_attn_linear_v_weight_to_fp16_palettized, x = query_33_cast_fp16)[name = string("linear_149_cast_fp16")];
+            tensor<int32, [4]> var_2860 = const()[name = string("op_2860"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> v_33_cast_fp16 = reshape(shape = var_2860, x = linear_149_cast_fp16)[name = string("v_33_cast_fp16")];
+            tensor<int32, [4]> value_33_perm_0 = const()[name = string("value_33_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> model_layers_16_self_attn_pos_bias_u_to_fp16 = const()[name = string("model_layers_16_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(209963904)))];
+            tensor<fp16, [1, 126, 8, 128]> var_2872_cast_fp16 = add(x = q_97_cast_fp16, y = model_layers_16_self_attn_pos_bias_u_to_fp16)[name = string("op_2872_cast_fp16")];
+            tensor<fp16, [8, 128]> model_layers_16_self_attn_pos_bias_v_to_fp16 = const()[name = string("model_layers_16_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(209966016)))];
+            tensor<fp16, [1, 126, 8, 128]> var_2874_cast_fp16 = add(x = q_97_cast_fp16, y = model_layers_16_self_attn_pos_bias_v_to_fp16)[name = string("op_2874_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_33_perm_0 = const()[name = string("q_with_bias_v_33_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_359_transpose_x_0 = const()[name = string("x_359_transpose_x_0"), val = bool(false)];
+            bool x_359_transpose_y_0 = const()[name = string("x_359_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 251]> var_2876_to_fp16 = const()[name = string("op_2876_to_fp16"), val = tensor<fp16, [1, 8, 128, 251]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(209968128)))];
+            tensor<fp16, [1, 8, 126, 128]> q_with_bias_v_33_cast_fp16 = transpose(perm = q_with_bias_v_33_perm_0, x = var_2874_cast_fp16)[name = string("transpose_200")];
+            tensor<fp16, [1, 8, 126, 251]> x_359_cast_fp16 = matmul(transpose_x = x_359_transpose_x_0, transpose_y = x_359_transpose_y_0, x = q_with_bias_v_33_cast_fp16, y = var_2876_to_fp16)[name = string("x_359_cast_fp16")];
+            tensor<int32, [8]> x_361_pad_0 = const()[name = string("x_361_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_361_mode_0 = const()[name = string("x_361_mode_0"), val = string("constant")];
+            fp16 const_174_to_fp16 = const()[name = string("const_174_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 252]> x_361_cast_fp16 = pad(constant_val = const_174_to_fp16, mode = x_361_mode_0, pad = x_361_pad_0, x = x_359_cast_fp16)[name = string("x_361_cast_fp16")];
+            tensor<int32, [4]> var_2884 = const()[name = string("op_2884"), val = tensor<int32, [4]>([1, 8, -1, 126])];
+            tensor<fp16, [1, 8, 252, 126]> x_363_cast_fp16 = reshape(shape = var_2884, x = x_361_cast_fp16)[name = string("x_363_cast_fp16")];
+            tensor<int32, [4]> var_2888_begin_0 = const()[name = string("op_2888_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_2888_end_0 = const()[name = string("op_2888_end_0"), val = tensor<int32, [4]>([1, 8, 252, 126])];
+            tensor<bool, [4]> var_2888_end_mask_0 = const()[name = string("op_2888_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 251, 126]> var_2888_cast_fp16 = slice_by_index(begin = var_2888_begin_0, end = var_2888_end_0, end_mask = var_2888_end_mask_0, x = x_363_cast_fp16)[name = string("op_2888_cast_fp16")];
+            tensor<int32, [4]> var_2889 = const()[name = string("op_2889"), val = tensor<int32, [4]>([1, 8, 126, 251])];
+            tensor<fp16, [1, 8, 126, 251]> matrix_bd_65_cast_fp16 = reshape(shape = var_2889, x = var_2888_cast_fp16)[name = string("matrix_bd_65_cast_fp16")];
+            bool matrix_ac_33_transpose_x_0 = const()[name = string("matrix_ac_33_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_33_transpose_y_0 = const()[name = string("matrix_ac_33_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_128_perm_0 = const()[name = string("transpose_128_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_129_perm_0 = const()[name = string("transpose_129_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 126]> transpose_129 = transpose(perm = transpose_129_perm_0, x = k_65_cast_fp16)[name = string("transpose_198")];
+            tensor<fp16, [1, 8, 126, 128]> transpose_128 = transpose(perm = transpose_128_perm_0, x = var_2872_cast_fp16)[name = string("transpose_199")];
+            tensor<fp16, [1, 8, 126, 126]> matrix_ac_33_cast_fp16 = matmul(transpose_x = matrix_ac_33_transpose_x_0, transpose_y = matrix_ac_33_transpose_y_0, x = transpose_128, y = transpose_129)[name = string("matrix_ac_33_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_67_begin_0 = const()[name = string("matrix_bd_67_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_67_end_0 = const()[name = string("matrix_bd_67_end_0"), val = tensor<int32, [4]>([1, 8, 126, 126])];
+            tensor<bool, [4]> matrix_bd_67_end_mask_0 = const()[name = string("matrix_bd_67_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 126, 126]> matrix_bd_67_cast_fp16 = slice_by_index(begin = matrix_bd_67_begin_0, end = matrix_bd_67_end_0, end_mask = matrix_bd_67_end_mask_0, x = matrix_bd_65_cast_fp16)[name = string("matrix_bd_67_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_2898_cast_fp16 = add(x = matrix_ac_33_cast_fp16, y = matrix_bd_67_cast_fp16)[name = string("op_2898_cast_fp16")];
+            fp16 _inversed_scores_65_y_0_to_fp16 = const()[name = string("_inversed_scores_65_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 126, 126]> _inversed_scores_65_cast_fp16 = mul(x = var_2898_cast_fp16, y = _inversed_scores_65_y_0_to_fp16)[name = string("_inversed_scores_65_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> scores_67_cast_fp16 = select(a = var_7_to_fp16, b = _inversed_scores_65_cast_fp16, cond = mask_3)[name = string("scores_67_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_2904_cast_fp16 = softmax(axis = var_25, x = scores_67_cast_fp16)[name = string("op_2904_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> input_865_cast_fp16 = select(a = var_6_to_fp16, b = var_2904_cast_fp16, cond = mask_3)[name = string("input_865_cast_fp16")];
+            bool x_365_transpose_x_0 = const()[name = string("x_365_transpose_x_0"), val = bool(false)];
+            bool x_365_transpose_y_0 = const()[name = string("x_365_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 126, 128]> value_33_cast_fp16 = transpose(perm = value_33_perm_0, x = v_33_cast_fp16)[name = string("transpose_197")];
+            tensor<fp16, [1, 8, 126, 128]> x_365_cast_fp16 = matmul(transpose_x = x_365_transpose_x_0, transpose_y = x_365_transpose_y_0, x = input_865_cast_fp16, y = value_33_cast_fp16)[name = string("x_365_cast_fp16")];
+            tensor<int32, [4]> var_2908_perm_0 = const()[name = string("op_2908_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_2909 = const()[name = string("op_2909"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 126, 8, 128]> var_2908_cast_fp16 = transpose(perm = var_2908_perm_0, x = x_365_cast_fp16)[name = string("transpose_196")];
+            tensor<fp16, [1, 126, 1024]> input_867_cast_fp16 = reshape(shape = var_2909, x = var_2908_cast_fp16)[name = string("input_867_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_16_self_attn_linear_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(210482240))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211006592))))[name = string("model_layers_16_self_attn_linear_out_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_151_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_16_self_attn_linear_out_weight_to_fp16_palettized, x = input_867_cast_fp16)[name = string("linear_151_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_871_cast_fp16 = add(x = input_863_cast_fp16, y = linear_151_cast_fp16)[name = string("input_871_cast_fp16")];
+            tensor<int32, [1]> x_369_axes_0 = const()[name = string("x_369_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_16_norm_conv_weight_to_fp16 = const()[name = string("model_layers_16_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211007680)))];
+            tensor<fp16, [1024]> model_layers_16_norm_conv_bias_to_fp16 = const()[name = string("model_layers_16_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211009792)))];
+            tensor<fp16, [1, 126, 1024]> x_369_cast_fp16 = layer_norm(axes = x_369_axes_0, beta = model_layers_16_norm_conv_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_16_norm_conv_weight_to_fp16, x = input_871_cast_fp16)[name = string("x_369_cast_fp16")];
+            tensor<int32, [3]> input_873_perm_0 = const()[name = string("input_873_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_875_pad_type_0 = const()[name = string("input_875_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_875_strides_0 = const()[name = string("input_875_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_875_pad_0 = const()[name = string("input_875_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_875_dilations_0 = const()[name = string("input_875_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_875_groups_0 = const()[name = string("input_875_groups_0"), val = int32(1)];
+            tensor<fp16, [2048, 1024, 1]> model_layers_16_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211011904))), lut = tensor<fp16, [64, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212060544))))[name = string("model_layers_16_conv_pointwise_conv1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 1024, 126]> input_873_cast_fp16 = transpose(perm = input_873_perm_0, x = x_369_cast_fp16)[name = string("transpose_195")];
+            tensor<fp16, [1, 2048, 126]> input_875_cast_fp16 = conv(dilations = input_875_dilations_0, groups = input_875_groups_0, pad = input_875_pad_0, pad_type = input_875_pad_type_0, strides = input_875_strides_0, weight = model_layers_16_conv_pointwise_conv1_weight_to_fp16_palettized, x = input_873_cast_fp16)[name = string("input_875_cast_fp16")];
+            int32 x_371_split_num_splits_0 = const()[name = string("x_371_split_num_splits_0"), val = int32(2)];
+            int32 x_371_split_axis_0 = const()[name = string("x_371_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 126]> x_371_split_cast_fp16_0, tensor<fp16, [1, 1024, 126]> x_371_split_cast_fp16_1 = split(axis = x_371_split_axis_0, num_splits = x_371_split_num_splits_0, x = input_875_cast_fp16)[name = string("x_371_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_371_split_1_sigmoid_cast_fp16 = sigmoid(x = x_371_split_cast_fp16_1)[name = string("x_371_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_371_cast_fp16 = mul(x = x_371_split_cast_fp16_0, y = x_371_split_1_sigmoid_cast_fp16)[name = string("x_371_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_877_cast_fp16 = select(a = var_6_to_fp16, b = x_371_cast_fp16, cond = var_323)[name = string("input_877_cast_fp16")];
+            tensor<int32, [6]> input_879_pad_0 = const()[name = string("input_879_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_879_mode_0 = const()[name = string("input_879_mode_0"), val = string("constant")];
+            fp16 const_177_to_fp16 = const()[name = string("const_177_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 134]> input_879_cast_fp16 = pad(constant_val = const_177_to_fp16, mode = input_879_mode_0, pad = input_879_pad_0, x = input_877_cast_fp16)[name = string("input_879_cast_fp16")];
+            string input_881_pad_type_0 = const()[name = string("input_881_pad_type_0"), val = string("valid")];
+            int32 input_881_groups_0 = const()[name = string("input_881_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_881_strides_0 = const()[name = string("input_881_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_881_pad_0 = const()[name = string("input_881_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_881_dilations_0 = const()[name = string("input_881_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_280_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212062656))), lut = tensor<fp16, [32, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212067328))))[name = string("const_280_to_fp16_palettized")];
+            tensor<fp16, [1024]> const_281_to_fp16 = const()[name = string("const_281_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212068416)))];
+            tensor<fp16, [1, 1024, 126]> input_883_cast_fp16 = conv(bias = const_281_to_fp16, dilations = input_881_dilations_0, groups = input_881_groups_0, pad = input_881_pad_0, pad_type = input_881_pad_type_0, strides = input_881_strides_0, weight = const_280_to_fp16_palettized, x = input_879_cast_fp16)[name = string("input_883_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_885_cast_fp16 = silu(x = input_883_cast_fp16)[name = string("input_885_cast_fp16")];
+            string x_373_pad_type_0 = const()[name = string("x_373_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_373_strides_0 = const()[name = string("x_373_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_373_pad_0 = const()[name = string("x_373_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_373_dilations_0 = const()[name = string("x_373_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_373_groups_0 = const()[name = string("x_373_groups_0"), val = int32(1)];
+            tensor<fp16, [1024, 1024, 1]> model_layers_16_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212070528))), lut = tensor<fp16, [32, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212594880))))[name = string("model_layers_16_conv_pointwise_conv2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 1024, 126]> x_373_cast_fp16 = conv(dilations = x_373_dilations_0, groups = x_373_groups_0, pad = x_373_pad_0, pad_type = x_373_pad_type_0, strides = x_373_strides_0, weight = model_layers_16_conv_pointwise_conv2_weight_to_fp16_palettized, x = input_885_cast_fp16)[name = string("x_373_cast_fp16")];
+            tensor<int32, [3]> input_887_perm_0 = const()[name = string("input_887_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 126, 1024]> input_887_cast_fp16 = transpose(perm = input_887_perm_0, x = x_373_cast_fp16)[name = string("transpose_194")];
+            tensor<fp16, [1, 126, 1024]> input_889_cast_fp16 = add(x = input_871_cast_fp16, y = input_887_cast_fp16)[name = string("input_889_cast_fp16")];
+            tensor<int32, [1]> input_891_axes_0 = const()[name = string("input_891_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_16_norm_feed_forward2_weight_to_fp16 = const()[name = string("model_layers_16_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212595968)))];
+            tensor<fp16, [1024]> model_layers_16_norm_feed_forward2_bias_to_fp16 = const()[name = string("model_layers_16_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212598080)))];
+            tensor<fp16, [1, 126, 1024]> input_891_cast_fp16 = layer_norm(axes = input_891_axes_0, beta = model_layers_16_norm_feed_forward2_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_16_norm_feed_forward2_weight_to_fp16, x = input_889_cast_fp16)[name = string("input_891_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_16_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212600192))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214697408))))[name = string("model_layers_16_feed_forward2_linear1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 4096]> linear_152_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_16_feed_forward2_linear1_weight_to_fp16_palettized, x = input_891_cast_fp16)[name = string("linear_152_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_895_cast_fp16 = silu(x = linear_152_cast_fp16)[name = string("input_895_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_16_feed_forward2_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214701568))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(216798784))))[name = string("model_layers_16_feed_forward2_linear2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_153_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_16_feed_forward2_linear2_weight_to_fp16_palettized, x = input_895_cast_fp16)[name = string("linear_153_cast_fp16")];
+            fp16 var_2969_to_fp16 = const()[name = string("op_2969_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_2970_cast_fp16 = mul(x = linear_153_cast_fp16, y = var_2969_to_fp16)[name = string("op_2970_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_901_cast_fp16 = add(x = input_889_cast_fp16, y = var_2970_cast_fp16)[name = string("input_901_cast_fp16")];
+            tensor<int32, [1]> input_903_axes_0 = const()[name = string("input_903_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_16_norm_out_weight_to_fp16 = const()[name = string("model_layers_16_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(216799872)))];
+            tensor<fp16, [1024]> model_layers_16_norm_out_bias_to_fp16 = const()[name = string("model_layers_16_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(216801984)))];
+            tensor<fp16, [1, 126, 1024]> input_903_cast_fp16 = layer_norm(axes = input_903_axes_0, beta = model_layers_16_norm_out_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_16_norm_out_weight_to_fp16, x = input_901_cast_fp16)[name = string("input_903_cast_fp16")];
+            tensor<int32, [1]> input_905_axes_0 = const()[name = string("input_905_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_17_norm_feed_forward1_weight_to_fp16 = const()[name = string("model_layers_17_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(216804096)))];
+            tensor<fp16, [1024]> model_layers_17_norm_feed_forward1_bias_to_fp16 = const()[name = string("model_layers_17_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(216806208)))];
+            tensor<fp16, [1, 126, 1024]> input_905_cast_fp16 = layer_norm(axes = input_905_axes_0, beta = model_layers_17_norm_feed_forward1_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_17_norm_feed_forward1_weight_to_fp16, x = input_903_cast_fp16)[name = string("input_905_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_17_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(216808320))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(218905536))))[name = string("model_layers_17_feed_forward1_linear1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 4096]> linear_154_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_17_feed_forward1_linear1_weight_to_fp16_palettized, x = input_905_cast_fp16)[name = string("linear_154_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_909_cast_fp16 = silu(x = linear_154_cast_fp16)[name = string("input_909_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_17_feed_forward1_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(218909696))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221006912))))[name = string("model_layers_17_feed_forward1_linear2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_155_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_17_feed_forward1_linear2_weight_to_fp16_palettized, x = input_909_cast_fp16)[name = string("linear_155_cast_fp16")];
+            fp16 var_2998_to_fp16 = const()[name = string("op_2998_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_2999_cast_fp16 = mul(x = linear_155_cast_fp16, y = var_2998_to_fp16)[name = string("op_2999_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_915_cast_fp16 = add(x = input_903_cast_fp16, y = var_2999_cast_fp16)[name = string("input_915_cast_fp16")];
+            tensor<int32, [1]> query_35_axes_0 = const()[name = string("query_35_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_17_norm_self_att_weight_to_fp16 = const()[name = string("model_layers_17_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221008000)))];
+            tensor<fp16, [1024]> model_layers_17_norm_self_att_bias_to_fp16 = const()[name = string("model_layers_17_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221010112)))];
+            tensor<fp16, [1, 126, 1024]> query_35_cast_fp16 = layer_norm(axes = query_35_axes_0, beta = model_layers_17_norm_self_att_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_17_norm_self_att_weight_to_fp16, x = input_915_cast_fp16)[name = string("query_35_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_17_self_attn_linear_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221012224))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221536576))))[name = string("model_layers_17_self_attn_linear_q_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_156_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_17_self_attn_linear_q_weight_to_fp16_palettized, x = query_35_cast_fp16)[name = string("linear_156_cast_fp16")];
+            tensor<int32, [4]> var_3015 = const()[name = string("op_3015"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> q_103_cast_fp16 = reshape(shape = var_3015, x = linear_156_cast_fp16)[name = string("q_103_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_17_self_attn_linear_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221537664))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(222062016))))[name = string("model_layers_17_self_attn_linear_k_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_157_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_17_self_attn_linear_k_weight_to_fp16_palettized, x = query_35_cast_fp16)[name = string("linear_157_cast_fp16")];
+            tensor<int32, [4]> var_3019 = const()[name = string("op_3019"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> k_69_cast_fp16 = reshape(shape = var_3019, x = linear_157_cast_fp16)[name = string("k_69_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_17_self_attn_linear_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(222063104))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(222587456))))[name = string("model_layers_17_self_attn_linear_v_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_158_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_17_self_attn_linear_v_weight_to_fp16_palettized, x = query_35_cast_fp16)[name = string("linear_158_cast_fp16")];
+            tensor<int32, [4]> var_3023 = const()[name = string("op_3023"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> v_35_cast_fp16 = reshape(shape = var_3023, x = linear_158_cast_fp16)[name = string("v_35_cast_fp16")];
+            tensor<int32, [4]> value_35_perm_0 = const()[name = string("value_35_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> model_layers_17_self_attn_pos_bias_u_to_fp16 = const()[name = string("model_layers_17_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(222588544)))];
+            tensor<fp16, [1, 126, 8, 128]> var_3035_cast_fp16 = add(x = q_103_cast_fp16, y = model_layers_17_self_attn_pos_bias_u_to_fp16)[name = string("op_3035_cast_fp16")];
+            tensor<fp16, [8, 128]> model_layers_17_self_attn_pos_bias_v_to_fp16 = const()[name = string("model_layers_17_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(222590656)))];
+            tensor<fp16, [1, 126, 8, 128]> var_3037_cast_fp16 = add(x = q_103_cast_fp16, y = model_layers_17_self_attn_pos_bias_v_to_fp16)[name = string("op_3037_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_35_perm_0 = const()[name = string("q_with_bias_v_35_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_381_transpose_x_0 = const()[name = string("x_381_transpose_x_0"), val = bool(false)];
+            bool x_381_transpose_y_0 = const()[name = string("x_381_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 251]> var_3039_to_fp16 = const()[name = string("op_3039_to_fp16"), val = tensor<fp16, [1, 8, 128, 251]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(222592768)))];
+            tensor<fp16, [1, 8, 126, 128]> q_with_bias_v_35_cast_fp16 = transpose(perm = q_with_bias_v_35_perm_0, x = var_3037_cast_fp16)[name = string("transpose_193")];
+            tensor<fp16, [1, 8, 126, 251]> x_381_cast_fp16 = matmul(transpose_x = x_381_transpose_x_0, transpose_y = x_381_transpose_y_0, x = q_with_bias_v_35_cast_fp16, y = var_3039_to_fp16)[name = string("x_381_cast_fp16")];
+            tensor<int32, [8]> x_383_pad_0 = const()[name = string("x_383_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_383_mode_0 = const()[name = string("x_383_mode_0"), val = string("constant")];
+            fp16 const_184_to_fp16 = const()[name = string("const_184_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 252]> x_383_cast_fp16 = pad(constant_val = const_184_to_fp16, mode = x_383_mode_0, pad = x_383_pad_0, x = x_381_cast_fp16)[name = string("x_383_cast_fp16")];
+            tensor<int32, [4]> var_3047 = const()[name = string("op_3047"), val = tensor<int32, [4]>([1, 8, -1, 126])];
+            tensor<fp16, [1, 8, 252, 126]> x_385_cast_fp16 = reshape(shape = var_3047, x = x_383_cast_fp16)[name = string("x_385_cast_fp16")];
+            tensor<int32, [4]> var_3051_begin_0 = const()[name = string("op_3051_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_3051_end_0 = const()[name = string("op_3051_end_0"), val = tensor<int32, [4]>([1, 8, 252, 126])];
+            tensor<bool, [4]> var_3051_end_mask_0 = const()[name = string("op_3051_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 251, 126]> var_3051_cast_fp16 = slice_by_index(begin = var_3051_begin_0, end = var_3051_end_0, end_mask = var_3051_end_mask_0, x = x_385_cast_fp16)[name = string("op_3051_cast_fp16")];
+            tensor<int32, [4]> var_3052 = const()[name = string("op_3052"), val = tensor<int32, [4]>([1, 8, 126, 251])];
+            tensor<fp16, [1, 8, 126, 251]> matrix_bd_69_cast_fp16 = reshape(shape = var_3052, x = var_3051_cast_fp16)[name = string("matrix_bd_69_cast_fp16")];
+            bool matrix_ac_35_transpose_x_0 = const()[name = string("matrix_ac_35_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_35_transpose_y_0 = const()[name = string("matrix_ac_35_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_130_perm_0 = const()[name = string("transpose_130_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_131_perm_0 = const()[name = string("transpose_131_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 126]> transpose_131 = transpose(perm = transpose_131_perm_0, x = k_69_cast_fp16)[name = string("transpose_191")];
+            tensor<fp16, [1, 8, 126, 128]> transpose_130 = transpose(perm = transpose_130_perm_0, x = var_3035_cast_fp16)[name = string("transpose_192")];
+            tensor<fp16, [1, 8, 126, 126]> matrix_ac_35_cast_fp16 = matmul(transpose_x = matrix_ac_35_transpose_x_0, transpose_y = matrix_ac_35_transpose_y_0, x = transpose_130, y = transpose_131)[name = string("matrix_ac_35_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_71_begin_0 = const()[name = string("matrix_bd_71_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_71_end_0 = const()[name = string("matrix_bd_71_end_0"), val = tensor<int32, [4]>([1, 8, 126, 126])];
+            tensor<bool, [4]> matrix_bd_71_end_mask_0 = const()[name = string("matrix_bd_71_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 126, 126]> matrix_bd_71_cast_fp16 = slice_by_index(begin = matrix_bd_71_begin_0, end = matrix_bd_71_end_0, end_mask = matrix_bd_71_end_mask_0, x = matrix_bd_69_cast_fp16)[name = string("matrix_bd_71_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_3061_cast_fp16 = add(x = matrix_ac_35_cast_fp16, y = matrix_bd_71_cast_fp16)[name = string("op_3061_cast_fp16")];
+            fp16 _inversed_scores_69_y_0_to_fp16 = const()[name = string("_inversed_scores_69_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 126, 126]> _inversed_scores_69_cast_fp16 = mul(x = var_3061_cast_fp16, y = _inversed_scores_69_y_0_to_fp16)[name = string("_inversed_scores_69_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> scores_71_cast_fp16 = select(a = var_7_to_fp16, b = _inversed_scores_69_cast_fp16, cond = mask_3)[name = string("scores_71_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_3067_cast_fp16 = softmax(axis = var_25, x = scores_71_cast_fp16)[name = string("op_3067_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> input_917_cast_fp16 = select(a = var_6_to_fp16, b = var_3067_cast_fp16, cond = mask_3)[name = string("input_917_cast_fp16")];
+            bool x_387_transpose_x_0 = const()[name = string("x_387_transpose_x_0"), val = bool(false)];
+            bool x_387_transpose_y_0 = const()[name = string("x_387_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 126, 128]> value_35_cast_fp16 = transpose(perm = value_35_perm_0, x = v_35_cast_fp16)[name = string("transpose_190")];
+            tensor<fp16, [1, 8, 126, 128]> x_387_cast_fp16 = matmul(transpose_x = x_387_transpose_x_0, transpose_y = x_387_transpose_y_0, x = input_917_cast_fp16, y = value_35_cast_fp16)[name = string("x_387_cast_fp16")];
+            tensor<int32, [4]> var_3071_perm_0 = const()[name = string("op_3071_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_3072 = const()[name = string("op_3072"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 126, 8, 128]> var_3071_cast_fp16 = transpose(perm = var_3071_perm_0, x = x_387_cast_fp16)[name = string("transpose_189")];
+            tensor<fp16, [1, 126, 1024]> input_919_cast_fp16 = reshape(shape = var_3072, x = var_3071_cast_fp16)[name = string("input_919_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_17_self_attn_linear_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(223106880))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(223631232))))[name = string("model_layers_17_self_attn_linear_out_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_160_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_17_self_attn_linear_out_weight_to_fp16_palettized, x = input_919_cast_fp16)[name = string("linear_160_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_923_cast_fp16 = add(x = input_915_cast_fp16, y = linear_160_cast_fp16)[name = string("input_923_cast_fp16")];
+            tensor<int32, [1]> x_391_axes_0 = const()[name = string("x_391_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_17_norm_conv_weight_to_fp16 = const()[name = string("model_layers_17_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(223632320)))];
+            tensor<fp16, [1024]> model_layers_17_norm_conv_bias_to_fp16 = const()[name = string("model_layers_17_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(223634432)))];
+            tensor<fp16, [1, 126, 1024]> x_391_cast_fp16 = layer_norm(axes = x_391_axes_0, beta = model_layers_17_norm_conv_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_17_norm_conv_weight_to_fp16, x = input_923_cast_fp16)[name = string("x_391_cast_fp16")];
+            tensor<int32, [3]> input_925_perm_0 = const()[name = string("input_925_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_927_pad_type_0 = const()[name = string("input_927_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_927_strides_0 = const()[name = string("input_927_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_927_pad_0 = const()[name = string("input_927_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_927_dilations_0 = const()[name = string("input_927_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_927_groups_0 = const()[name = string("input_927_groups_0"), val = int32(1)];
+            tensor<fp16, [2048, 1024, 1]> model_layers_17_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(223636544))), lut = tensor<fp16, [64, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224685184))))[name = string("model_layers_17_conv_pointwise_conv1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 1024, 126]> input_925_cast_fp16 = transpose(perm = input_925_perm_0, x = x_391_cast_fp16)[name = string("transpose_188")];
+            tensor<fp16, [1, 2048, 126]> input_927_cast_fp16 = conv(dilations = input_927_dilations_0, groups = input_927_groups_0, pad = input_927_pad_0, pad_type = input_927_pad_type_0, strides = input_927_strides_0, weight = model_layers_17_conv_pointwise_conv1_weight_to_fp16_palettized, x = input_925_cast_fp16)[name = string("input_927_cast_fp16")];
+            int32 x_393_split_num_splits_0 = const()[name = string("x_393_split_num_splits_0"), val = int32(2)];
+            int32 x_393_split_axis_0 = const()[name = string("x_393_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 126]> x_393_split_cast_fp16_0, tensor<fp16, [1, 1024, 126]> x_393_split_cast_fp16_1 = split(axis = x_393_split_axis_0, num_splits = x_393_split_num_splits_0, x = input_927_cast_fp16)[name = string("x_393_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_393_split_1_sigmoid_cast_fp16 = sigmoid(x = x_393_split_cast_fp16_1)[name = string("x_393_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_393_cast_fp16 = mul(x = x_393_split_cast_fp16_0, y = x_393_split_1_sigmoid_cast_fp16)[name = string("x_393_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_929_cast_fp16 = select(a = var_6_to_fp16, b = x_393_cast_fp16, cond = var_323)[name = string("input_929_cast_fp16")];
+            tensor<int32, [6]> input_931_pad_0 = const()[name = string("input_931_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_931_mode_0 = const()[name = string("input_931_mode_0"), val = string("constant")];
+            fp16 const_187_to_fp16 = const()[name = string("const_187_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 134]> input_931_cast_fp16 = pad(constant_val = const_187_to_fp16, mode = input_931_mode_0, pad = input_931_pad_0, x = input_929_cast_fp16)[name = string("input_931_cast_fp16")];
+            string input_933_pad_type_0 = const()[name = string("input_933_pad_type_0"), val = string("valid")];
+            int32 input_933_groups_0 = const()[name = string("input_933_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_933_strides_0 = const()[name = string("input_933_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_933_pad_0 = const()[name = string("input_933_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_933_dilations_0 = const()[name = string("input_933_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_282_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224687296))), lut = tensor<fp16, [32, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224691968))))[name = string("const_282_to_fp16_palettized")];
+            tensor<fp16, [1024]> const_283_to_fp16 = const()[name = string("const_283_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224693056)))];
+            tensor<fp16, [1, 1024, 126]> input_935_cast_fp16 = conv(bias = const_283_to_fp16, dilations = input_933_dilations_0, groups = input_933_groups_0, pad = input_933_pad_0, pad_type = input_933_pad_type_0, strides = input_933_strides_0, weight = const_282_to_fp16_palettized, x = input_931_cast_fp16)[name = string("input_935_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_937_cast_fp16 = silu(x = input_935_cast_fp16)[name = string("input_937_cast_fp16")];
+            string x_395_pad_type_0 = const()[name = string("x_395_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_395_strides_0 = const()[name = string("x_395_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_395_pad_0 = const()[name = string("x_395_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_395_dilations_0 = const()[name = string("x_395_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_395_groups_0 = const()[name = string("x_395_groups_0"), val = int32(1)];
+            tensor<fp16, [1024, 1024, 1]> model_layers_17_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224695168))), lut = tensor<fp16, [32, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225219520))))[name = string("model_layers_17_conv_pointwise_conv2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 1024, 126]> x_395_cast_fp16 = conv(dilations = x_395_dilations_0, groups = x_395_groups_0, pad = x_395_pad_0, pad_type = x_395_pad_type_0, strides = x_395_strides_0, weight = model_layers_17_conv_pointwise_conv2_weight_to_fp16_palettized, x = input_937_cast_fp16)[name = string("x_395_cast_fp16")];
+            tensor<int32, [3]> input_939_perm_0 = const()[name = string("input_939_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 126, 1024]> input_939_cast_fp16 = transpose(perm = input_939_perm_0, x = x_395_cast_fp16)[name = string("transpose_187")];
+            tensor<fp16, [1, 126, 1024]> input_941_cast_fp16 = add(x = input_923_cast_fp16, y = input_939_cast_fp16)[name = string("input_941_cast_fp16")];
+            tensor<int32, [1]> input_943_axes_0 = const()[name = string("input_943_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_17_norm_feed_forward2_weight_to_fp16 = const()[name = string("model_layers_17_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225220608)))];
+            tensor<fp16, [1024]> model_layers_17_norm_feed_forward2_bias_to_fp16 = const()[name = string("model_layers_17_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225222720)))];
+            tensor<fp16, [1, 126, 1024]> input_943_cast_fp16 = layer_norm(axes = input_943_axes_0, beta = model_layers_17_norm_feed_forward2_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_17_norm_feed_forward2_weight_to_fp16, x = input_941_cast_fp16)[name = string("input_943_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_17_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225224832))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227322048))))[name = string("model_layers_17_feed_forward2_linear1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 4096]> linear_161_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_17_feed_forward2_linear1_weight_to_fp16_palettized, x = input_943_cast_fp16)[name = string("linear_161_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_947_cast_fp16 = silu(x = linear_161_cast_fp16)[name = string("input_947_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_17_feed_forward2_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227326208))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(229423424))))[name = string("model_layers_17_feed_forward2_linear2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_162_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_17_feed_forward2_linear2_weight_to_fp16_palettized, x = input_947_cast_fp16)[name = string("linear_162_cast_fp16")];
+            fp16 var_3132_to_fp16 = const()[name = string("op_3132_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_3133_cast_fp16 = mul(x = linear_162_cast_fp16, y = var_3132_to_fp16)[name = string("op_3133_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_953_cast_fp16 = add(x = input_941_cast_fp16, y = var_3133_cast_fp16)[name = string("input_953_cast_fp16")];
+            tensor<int32, [1]> input_955_axes_0 = const()[name = string("input_955_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_17_norm_out_weight_to_fp16 = const()[name = string("model_layers_17_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(229424512)))];
+            tensor<fp16, [1024]> model_layers_17_norm_out_bias_to_fp16 = const()[name = string("model_layers_17_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(229426624)))];
+            tensor<fp16, [1, 126, 1024]> input_955_cast_fp16 = layer_norm(axes = input_955_axes_0, beta = model_layers_17_norm_out_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_17_norm_out_weight_to_fp16, x = input_953_cast_fp16)[name = string("input_955_cast_fp16")];
+            tensor<int32, [1]> input_957_axes_0 = const()[name = string("input_957_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_18_norm_feed_forward1_weight_to_fp16 = const()[name = string("model_layers_18_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(229428736)))];
+            tensor<fp16, [1024]> model_layers_18_norm_feed_forward1_bias_to_fp16 = const()[name = string("model_layers_18_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(229430848)))];
+            tensor<fp16, [1, 126, 1024]> input_957_cast_fp16 = layer_norm(axes = input_957_axes_0, beta = model_layers_18_norm_feed_forward1_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_18_norm_feed_forward1_weight_to_fp16, x = input_955_cast_fp16)[name = string("input_957_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_18_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(229432960))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(231530176))))[name = string("model_layers_18_feed_forward1_linear1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 4096]> linear_163_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_18_feed_forward1_linear1_weight_to_fp16_palettized, x = input_957_cast_fp16)[name = string("linear_163_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_961_cast_fp16 = silu(x = linear_163_cast_fp16)[name = string("input_961_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_18_feed_forward1_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(231534336))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(233631552))))[name = string("model_layers_18_feed_forward1_linear2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_164_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_18_feed_forward1_linear2_weight_to_fp16_palettized, x = input_961_cast_fp16)[name = string("linear_164_cast_fp16")];
+            fp16 var_3161_to_fp16 = const()[name = string("op_3161_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_3162_cast_fp16 = mul(x = linear_164_cast_fp16, y = var_3161_to_fp16)[name = string("op_3162_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_967_cast_fp16 = add(x = input_955_cast_fp16, y = var_3162_cast_fp16)[name = string("input_967_cast_fp16")];
+            tensor<int32, [1]> query_37_axes_0 = const()[name = string("query_37_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_18_norm_self_att_weight_to_fp16 = const()[name = string("model_layers_18_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(233632640)))];
+            tensor<fp16, [1024]> model_layers_18_norm_self_att_bias_to_fp16 = const()[name = string("model_layers_18_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(233634752)))];
+            tensor<fp16, [1, 126, 1024]> query_37_cast_fp16 = layer_norm(axes = query_37_axes_0, beta = model_layers_18_norm_self_att_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_18_norm_self_att_weight_to_fp16, x = input_967_cast_fp16)[name = string("query_37_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_18_self_attn_linear_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(233636864))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(234161216))))[name = string("model_layers_18_self_attn_linear_q_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_165_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_18_self_attn_linear_q_weight_to_fp16_palettized, x = query_37_cast_fp16)[name = string("linear_165_cast_fp16")];
+            tensor<int32, [4]> var_3178 = const()[name = string("op_3178"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> q_109_cast_fp16 = reshape(shape = var_3178, x = linear_165_cast_fp16)[name = string("q_109_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_18_self_attn_linear_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(234162304))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(234686656))))[name = string("model_layers_18_self_attn_linear_k_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_166_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_18_self_attn_linear_k_weight_to_fp16_palettized, x = query_37_cast_fp16)[name = string("linear_166_cast_fp16")];
+            tensor<int32, [4]> var_3182 = const()[name = string("op_3182"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> k_73_cast_fp16 = reshape(shape = var_3182, x = linear_166_cast_fp16)[name = string("k_73_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_18_self_attn_linear_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(234687744))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(235212096))))[name = string("model_layers_18_self_attn_linear_v_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_167_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_18_self_attn_linear_v_weight_to_fp16_palettized, x = query_37_cast_fp16)[name = string("linear_167_cast_fp16")];
+            tensor<int32, [4]> var_3186 = const()[name = string("op_3186"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> v_37_cast_fp16 = reshape(shape = var_3186, x = linear_167_cast_fp16)[name = string("v_37_cast_fp16")];
+            tensor<int32, [4]> value_37_perm_0 = const()[name = string("value_37_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> model_layers_18_self_attn_pos_bias_u_to_fp16 = const()[name = string("model_layers_18_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(235213184)))];
+            tensor<fp16, [1, 126, 8, 128]> var_3198_cast_fp16 = add(x = q_109_cast_fp16, y = model_layers_18_self_attn_pos_bias_u_to_fp16)[name = string("op_3198_cast_fp16")];
+            tensor<fp16, [8, 128]> model_layers_18_self_attn_pos_bias_v_to_fp16 = const()[name = string("model_layers_18_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(235215296)))];
+            tensor<fp16, [1, 126, 8, 128]> var_3200_cast_fp16 = add(x = q_109_cast_fp16, y = model_layers_18_self_attn_pos_bias_v_to_fp16)[name = string("op_3200_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_37_perm_0 = const()[name = string("q_with_bias_v_37_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_403_transpose_x_0 = const()[name = string("x_403_transpose_x_0"), val = bool(false)];
+            bool x_403_transpose_y_0 = const()[name = string("x_403_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 251]> var_3202_to_fp16 = const()[name = string("op_3202_to_fp16"), val = tensor<fp16, [1, 8, 128, 251]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(235217408)))];
+            tensor<fp16, [1, 8, 126, 128]> q_with_bias_v_37_cast_fp16 = transpose(perm = q_with_bias_v_37_perm_0, x = var_3200_cast_fp16)[name = string("transpose_186")];
+            tensor<fp16, [1, 8, 126, 251]> x_403_cast_fp16 = matmul(transpose_x = x_403_transpose_x_0, transpose_y = x_403_transpose_y_0, x = q_with_bias_v_37_cast_fp16, y = var_3202_to_fp16)[name = string("x_403_cast_fp16")];
+            tensor<int32, [8]> x_405_pad_0 = const()[name = string("x_405_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_405_mode_0 = const()[name = string("x_405_mode_0"), val = string("constant")];
+            fp16 const_194_to_fp16 = const()[name = string("const_194_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 252]> x_405_cast_fp16 = pad(constant_val = const_194_to_fp16, mode = x_405_mode_0, pad = x_405_pad_0, x = x_403_cast_fp16)[name = string("x_405_cast_fp16")];
+            tensor<int32, [4]> var_3210 = const()[name = string("op_3210"), val = tensor<int32, [4]>([1, 8, -1, 126])];
+            tensor<fp16, [1, 8, 252, 126]> x_407_cast_fp16 = reshape(shape = var_3210, x = x_405_cast_fp16)[name = string("x_407_cast_fp16")];
+            tensor<int32, [4]> var_3214_begin_0 = const()[name = string("op_3214_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_3214_end_0 = const()[name = string("op_3214_end_0"), val = tensor<int32, [4]>([1, 8, 252, 126])];
+            tensor<bool, [4]> var_3214_end_mask_0 = const()[name = string("op_3214_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 251, 126]> var_3214_cast_fp16 = slice_by_index(begin = var_3214_begin_0, end = var_3214_end_0, end_mask = var_3214_end_mask_0, x = x_407_cast_fp16)[name = string("op_3214_cast_fp16")];
+            tensor<int32, [4]> var_3215 = const()[name = string("op_3215"), val = tensor<int32, [4]>([1, 8, 126, 251])];
+            tensor<fp16, [1, 8, 126, 251]> matrix_bd_73_cast_fp16 = reshape(shape = var_3215, x = var_3214_cast_fp16)[name = string("matrix_bd_73_cast_fp16")];
+            bool matrix_ac_37_transpose_x_0 = const()[name = string("matrix_ac_37_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_37_transpose_y_0 = const()[name = string("matrix_ac_37_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_132_perm_0 = const()[name = string("transpose_132_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_133_perm_0 = const()[name = string("transpose_133_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 126]> transpose_133 = transpose(perm = transpose_133_perm_0, x = k_73_cast_fp16)[name = string("transpose_184")];
+            tensor<fp16, [1, 8, 126, 128]> transpose_132 = transpose(perm = transpose_132_perm_0, x = var_3198_cast_fp16)[name = string("transpose_185")];
+            tensor<fp16, [1, 8, 126, 126]> matrix_ac_37_cast_fp16 = matmul(transpose_x = matrix_ac_37_transpose_x_0, transpose_y = matrix_ac_37_transpose_y_0, x = transpose_132, y = transpose_133)[name = string("matrix_ac_37_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_75_begin_0 = const()[name = string("matrix_bd_75_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_75_end_0 = const()[name = string("matrix_bd_75_end_0"), val = tensor<int32, [4]>([1, 8, 126, 126])];
+            tensor<bool, [4]> matrix_bd_75_end_mask_0 = const()[name = string("matrix_bd_75_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 126, 126]> matrix_bd_75_cast_fp16 = slice_by_index(begin = matrix_bd_75_begin_0, end = matrix_bd_75_end_0, end_mask = matrix_bd_75_end_mask_0, x = matrix_bd_73_cast_fp16)[name = string("matrix_bd_75_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_3224_cast_fp16 = add(x = matrix_ac_37_cast_fp16, y = matrix_bd_75_cast_fp16)[name = string("op_3224_cast_fp16")];
+            fp16 _inversed_scores_73_y_0_to_fp16 = const()[name = string("_inversed_scores_73_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 126, 126]> _inversed_scores_73_cast_fp16 = mul(x = var_3224_cast_fp16, y = _inversed_scores_73_y_0_to_fp16)[name = string("_inversed_scores_73_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> scores_75_cast_fp16 = select(a = var_7_to_fp16, b = _inversed_scores_73_cast_fp16, cond = mask_3)[name = string("scores_75_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_3230_cast_fp16 = softmax(axis = var_25, x = scores_75_cast_fp16)[name = string("op_3230_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> input_969_cast_fp16 = select(a = var_6_to_fp16, b = var_3230_cast_fp16, cond = mask_3)[name = string("input_969_cast_fp16")];
+            bool x_409_transpose_x_0 = const()[name = string("x_409_transpose_x_0"), val = bool(false)];
+            bool x_409_transpose_y_0 = const()[name = string("x_409_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 126, 128]> value_37_cast_fp16 = transpose(perm = value_37_perm_0, x = v_37_cast_fp16)[name = string("transpose_183")];
+            tensor<fp16, [1, 8, 126, 128]> x_409_cast_fp16 = matmul(transpose_x = x_409_transpose_x_0, transpose_y = x_409_transpose_y_0, x = input_969_cast_fp16, y = value_37_cast_fp16)[name = string("x_409_cast_fp16")];
+            tensor<int32, [4]> var_3234_perm_0 = const()[name = string("op_3234_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_3235 = const()[name = string("op_3235"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 126, 8, 128]> var_3234_cast_fp16 = transpose(perm = var_3234_perm_0, x = x_409_cast_fp16)[name = string("transpose_182")];
+            tensor<fp16, [1, 126, 1024]> input_971_cast_fp16 = reshape(shape = var_3235, x = var_3234_cast_fp16)[name = string("input_971_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_18_self_attn_linear_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(235731520))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236255872))))[name = string("model_layers_18_self_attn_linear_out_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_169_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_18_self_attn_linear_out_weight_to_fp16_palettized, x = input_971_cast_fp16)[name = string("linear_169_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_975_cast_fp16 = add(x = input_967_cast_fp16, y = linear_169_cast_fp16)[name = string("input_975_cast_fp16")];
+            tensor<int32, [1]> x_413_axes_0 = const()[name = string("x_413_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_18_norm_conv_weight_to_fp16 = const()[name = string("model_layers_18_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236256960)))];
+            tensor<fp16, [1024]> model_layers_18_norm_conv_bias_to_fp16 = const()[name = string("model_layers_18_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236259072)))];
+            tensor<fp16, [1, 126, 1024]> x_413_cast_fp16 = layer_norm(axes = x_413_axes_0, beta = model_layers_18_norm_conv_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_18_norm_conv_weight_to_fp16, x = input_975_cast_fp16)[name = string("x_413_cast_fp16")];
+            tensor<int32, [3]> input_977_perm_0 = const()[name = string("input_977_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_979_pad_type_0 = const()[name = string("input_979_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_979_strides_0 = const()[name = string("input_979_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_979_pad_0 = const()[name = string("input_979_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_979_dilations_0 = const()[name = string("input_979_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_979_groups_0 = const()[name = string("input_979_groups_0"), val = int32(1)];
+            tensor<fp16, [2048, 1024, 1]> model_layers_18_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236261184))), lut = tensor<fp16, [64, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237309824))))[name = string("model_layers_18_conv_pointwise_conv1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 1024, 126]> input_977_cast_fp16 = transpose(perm = input_977_perm_0, x = x_413_cast_fp16)[name = string("transpose_181")];
+            tensor<fp16, [1, 2048, 126]> input_979_cast_fp16 = conv(dilations = input_979_dilations_0, groups = input_979_groups_0, pad = input_979_pad_0, pad_type = input_979_pad_type_0, strides = input_979_strides_0, weight = model_layers_18_conv_pointwise_conv1_weight_to_fp16_palettized, x = input_977_cast_fp16)[name = string("input_979_cast_fp16")];
+            int32 x_415_split_num_splits_0 = const()[name = string("x_415_split_num_splits_0"), val = int32(2)];
+            int32 x_415_split_axis_0 = const()[name = string("x_415_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 126]> x_415_split_cast_fp16_0, tensor<fp16, [1, 1024, 126]> x_415_split_cast_fp16_1 = split(axis = x_415_split_axis_0, num_splits = x_415_split_num_splits_0, x = input_979_cast_fp16)[name = string("x_415_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_415_split_1_sigmoid_cast_fp16 = sigmoid(x = x_415_split_cast_fp16_1)[name = string("x_415_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_415_cast_fp16 = mul(x = x_415_split_cast_fp16_0, y = x_415_split_1_sigmoid_cast_fp16)[name = string("x_415_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_981_cast_fp16 = select(a = var_6_to_fp16, b = x_415_cast_fp16, cond = var_323)[name = string("input_981_cast_fp16")];
+            tensor<int32, [6]> input_983_pad_0 = const()[name = string("input_983_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_983_mode_0 = const()[name = string("input_983_mode_0"), val = string("constant")];
+            fp16 const_197_to_fp16 = const()[name = string("const_197_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 134]> input_983_cast_fp16 = pad(constant_val = const_197_to_fp16, mode = input_983_mode_0, pad = input_983_pad_0, x = input_981_cast_fp16)[name = string("input_983_cast_fp16")];
+            string input_985_pad_type_0 = const()[name = string("input_985_pad_type_0"), val = string("valid")];
+            int32 input_985_groups_0 = const()[name = string("input_985_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_985_strides_0 = const()[name = string("input_985_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_985_pad_0 = const()[name = string("input_985_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_985_dilations_0 = const()[name = string("input_985_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_284_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237311936))), lut = tensor<fp16, [32, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237316608))))[name = string("const_284_to_fp16_palettized")];
+            tensor<fp16, [1024]> const_285_to_fp16 = const()[name = string("const_285_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237317696)))];
+            tensor<fp16, [1, 1024, 126]> input_987_cast_fp16 = conv(bias = const_285_to_fp16, dilations = input_985_dilations_0, groups = input_985_groups_0, pad = input_985_pad_0, pad_type = input_985_pad_type_0, strides = input_985_strides_0, weight = const_284_to_fp16_palettized, x = input_983_cast_fp16)[name = string("input_987_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_989_cast_fp16 = silu(x = input_987_cast_fp16)[name = string("input_989_cast_fp16")];
+            string x_417_pad_type_0 = const()[name = string("x_417_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_417_strides_0 = const()[name = string("x_417_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_417_pad_0 = const()[name = string("x_417_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_417_dilations_0 = const()[name = string("x_417_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_417_groups_0 = const()[name = string("x_417_groups_0"), val = int32(1)];
+            tensor<fp16, [1024, 1024, 1]> model_layers_18_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237319808))), lut = tensor<fp16, [32, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237844160))))[name = string("model_layers_18_conv_pointwise_conv2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 1024, 126]> x_417_cast_fp16 = conv(dilations = x_417_dilations_0, groups = x_417_groups_0, pad = x_417_pad_0, pad_type = x_417_pad_type_0, strides = x_417_strides_0, weight = model_layers_18_conv_pointwise_conv2_weight_to_fp16_palettized, x = input_989_cast_fp16)[name = string("x_417_cast_fp16")];
+            tensor<int32, [3]> input_991_perm_0 = const()[name = string("input_991_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 126, 1024]> input_991_cast_fp16 = transpose(perm = input_991_perm_0, x = x_417_cast_fp16)[name = string("transpose_180")];
+            tensor<fp16, [1, 126, 1024]> input_993_cast_fp16 = add(x = input_975_cast_fp16, y = input_991_cast_fp16)[name = string("input_993_cast_fp16")];
+            tensor<int32, [1]> input_995_axes_0 = const()[name = string("input_995_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_18_norm_feed_forward2_weight_to_fp16 = const()[name = string("model_layers_18_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237845248)))];
+            tensor<fp16, [1024]> model_layers_18_norm_feed_forward2_bias_to_fp16 = const()[name = string("model_layers_18_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237847360)))];
+            tensor<fp16, [1, 126, 1024]> input_995_cast_fp16 = layer_norm(axes = input_995_axes_0, beta = model_layers_18_norm_feed_forward2_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_18_norm_feed_forward2_weight_to_fp16, x = input_993_cast_fp16)[name = string("input_995_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_18_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237849472))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(239946688))))[name = string("model_layers_18_feed_forward2_linear1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 4096]> linear_170_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_18_feed_forward2_linear1_weight_to_fp16_palettized, x = input_995_cast_fp16)[name = string("linear_170_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_999_cast_fp16 = silu(x = linear_170_cast_fp16)[name = string("input_999_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_18_feed_forward2_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(239950848))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(242048064))))[name = string("model_layers_18_feed_forward2_linear2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_171_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_18_feed_forward2_linear2_weight_to_fp16_palettized, x = input_999_cast_fp16)[name = string("linear_171_cast_fp16")];
+            fp16 var_3295_to_fp16 = const()[name = string("op_3295_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_3296_cast_fp16 = mul(x = linear_171_cast_fp16, y = var_3295_to_fp16)[name = string("op_3296_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_1005_cast_fp16 = add(x = input_993_cast_fp16, y = var_3296_cast_fp16)[name = string("input_1005_cast_fp16")];
+            tensor<int32, [1]> input_1007_axes_0 = const()[name = string("input_1007_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_18_norm_out_weight_to_fp16 = const()[name = string("model_layers_18_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(242049152)))];
+            tensor<fp16, [1024]> model_layers_18_norm_out_bias_to_fp16 = const()[name = string("model_layers_18_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(242051264)))];
+            tensor<fp16, [1, 126, 1024]> input_1007_cast_fp16 = layer_norm(axes = input_1007_axes_0, beta = model_layers_18_norm_out_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_18_norm_out_weight_to_fp16, x = input_1005_cast_fp16)[name = string("input_1007_cast_fp16")];
+            tensor<int32, [1]> input_1009_axes_0 = const()[name = string("input_1009_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_19_norm_feed_forward1_weight_to_fp16 = const()[name = string("model_layers_19_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(242053376)))];
+            tensor<fp16, [1024]> model_layers_19_norm_feed_forward1_bias_to_fp16 = const()[name = string("model_layers_19_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(242055488)))];
+            tensor<fp16, [1, 126, 1024]> input_1009_cast_fp16 = layer_norm(axes = input_1009_axes_0, beta = model_layers_19_norm_feed_forward1_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_19_norm_feed_forward1_weight_to_fp16, x = input_1007_cast_fp16)[name = string("input_1009_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_19_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(242057600))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244154816))))[name = string("model_layers_19_feed_forward1_linear1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 4096]> linear_172_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_19_feed_forward1_linear1_weight_to_fp16_palettized, x = input_1009_cast_fp16)[name = string("linear_172_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_1013_cast_fp16 = silu(x = linear_172_cast_fp16)[name = string("input_1013_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_19_feed_forward1_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244158976))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246256192))))[name = string("model_layers_19_feed_forward1_linear2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_173_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_19_feed_forward1_linear2_weight_to_fp16_palettized, x = input_1013_cast_fp16)[name = string("linear_173_cast_fp16")];
+            fp16 var_3324_to_fp16 = const()[name = string("op_3324_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_3325_cast_fp16 = mul(x = linear_173_cast_fp16, y = var_3324_to_fp16)[name = string("op_3325_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_1019_cast_fp16 = add(x = input_1007_cast_fp16, y = var_3325_cast_fp16)[name = string("input_1019_cast_fp16")];
+            tensor<int32, [1]> query_39_axes_0 = const()[name = string("query_39_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_19_norm_self_att_weight_to_fp16 = const()[name = string("model_layers_19_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246257280)))];
+            tensor<fp16, [1024]> model_layers_19_norm_self_att_bias_to_fp16 = const()[name = string("model_layers_19_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246259392)))];
+            tensor<fp16, [1, 126, 1024]> query_39_cast_fp16 = layer_norm(axes = query_39_axes_0, beta = model_layers_19_norm_self_att_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_19_norm_self_att_weight_to_fp16, x = input_1019_cast_fp16)[name = string("query_39_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_19_self_attn_linear_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246261504))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246785856))))[name = string("model_layers_19_self_attn_linear_q_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_174_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_19_self_attn_linear_q_weight_to_fp16_palettized, x = query_39_cast_fp16)[name = string("linear_174_cast_fp16")];
+            tensor<int32, [4]> var_3341 = const()[name = string("op_3341"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> q_115_cast_fp16 = reshape(shape = var_3341, x = linear_174_cast_fp16)[name = string("q_115_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_19_self_attn_linear_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246786944))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(247311296))))[name = string("model_layers_19_self_attn_linear_k_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_175_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_19_self_attn_linear_k_weight_to_fp16_palettized, x = query_39_cast_fp16)[name = string("linear_175_cast_fp16")];
+            tensor<int32, [4]> var_3345 = const()[name = string("op_3345"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> k_77_cast_fp16 = reshape(shape = var_3345, x = linear_175_cast_fp16)[name = string("k_77_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_19_self_attn_linear_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(247312384))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(247836736))))[name = string("model_layers_19_self_attn_linear_v_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_176_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_19_self_attn_linear_v_weight_to_fp16_palettized, x = query_39_cast_fp16)[name = string("linear_176_cast_fp16")];
+            tensor<int32, [4]> var_3349 = const()[name = string("op_3349"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> v_39_cast_fp16 = reshape(shape = var_3349, x = linear_176_cast_fp16)[name = string("v_39_cast_fp16")];
+            tensor<int32, [4]> value_39_perm_0 = const()[name = string("value_39_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> model_layers_19_self_attn_pos_bias_u_to_fp16 = const()[name = string("model_layers_19_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(247837824)))];
+            tensor<fp16, [1, 126, 8, 128]> var_3361_cast_fp16 = add(x = q_115_cast_fp16, y = model_layers_19_self_attn_pos_bias_u_to_fp16)[name = string("op_3361_cast_fp16")];
+            tensor<fp16, [8, 128]> model_layers_19_self_attn_pos_bias_v_to_fp16 = const()[name = string("model_layers_19_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(247839936)))];
+            tensor<fp16, [1, 126, 8, 128]> var_3363_cast_fp16 = add(x = q_115_cast_fp16, y = model_layers_19_self_attn_pos_bias_v_to_fp16)[name = string("op_3363_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_39_perm_0 = const()[name = string("q_with_bias_v_39_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_425_transpose_x_0 = const()[name = string("x_425_transpose_x_0"), val = bool(false)];
+            bool x_425_transpose_y_0 = const()[name = string("x_425_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 251]> var_3365_to_fp16 = const()[name = string("op_3365_to_fp16"), val = tensor<fp16, [1, 8, 128, 251]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(247842048)))];
+            tensor<fp16, [1, 8, 126, 128]> q_with_bias_v_39_cast_fp16 = transpose(perm = q_with_bias_v_39_perm_0, x = var_3363_cast_fp16)[name = string("transpose_179")];
+            tensor<fp16, [1, 8, 126, 251]> x_425_cast_fp16 = matmul(transpose_x = x_425_transpose_x_0, transpose_y = x_425_transpose_y_0, x = q_with_bias_v_39_cast_fp16, y = var_3365_to_fp16)[name = string("x_425_cast_fp16")];
+            tensor<int32, [8]> x_427_pad_0 = const()[name = string("x_427_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_427_mode_0 = const()[name = string("x_427_mode_0"), val = string("constant")];
+            fp16 const_204_to_fp16 = const()[name = string("const_204_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 252]> x_427_cast_fp16 = pad(constant_val = const_204_to_fp16, mode = x_427_mode_0, pad = x_427_pad_0, x = x_425_cast_fp16)[name = string("x_427_cast_fp16")];
+            tensor<int32, [4]> var_3373 = const()[name = string("op_3373"), val = tensor<int32, [4]>([1, 8, -1, 126])];
+            tensor<fp16, [1, 8, 252, 126]> x_429_cast_fp16 = reshape(shape = var_3373, x = x_427_cast_fp16)[name = string("x_429_cast_fp16")];
+            tensor<int32, [4]> var_3377_begin_0 = const()[name = string("op_3377_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_3377_end_0 = const()[name = string("op_3377_end_0"), val = tensor<int32, [4]>([1, 8, 252, 126])];
+            tensor<bool, [4]> var_3377_end_mask_0 = const()[name = string("op_3377_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 251, 126]> var_3377_cast_fp16 = slice_by_index(begin = var_3377_begin_0, end = var_3377_end_0, end_mask = var_3377_end_mask_0, x = x_429_cast_fp16)[name = string("op_3377_cast_fp16")];
+            tensor<int32, [4]> var_3378 = const()[name = string("op_3378"), val = tensor<int32, [4]>([1, 8, 126, 251])];
+            tensor<fp16, [1, 8, 126, 251]> matrix_bd_77_cast_fp16 = reshape(shape = var_3378, x = var_3377_cast_fp16)[name = string("matrix_bd_77_cast_fp16")];
+            bool matrix_ac_39_transpose_x_0 = const()[name = string("matrix_ac_39_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_39_transpose_y_0 = const()[name = string("matrix_ac_39_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_134_perm_0 = const()[name = string("transpose_134_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_135_perm_0 = const()[name = string("transpose_135_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 126]> transpose_135 = transpose(perm = transpose_135_perm_0, x = k_77_cast_fp16)[name = string("transpose_177")];
+            tensor<fp16, [1, 8, 126, 128]> transpose_134 = transpose(perm = transpose_134_perm_0, x = var_3361_cast_fp16)[name = string("transpose_178")];
+            tensor<fp16, [1, 8, 126, 126]> matrix_ac_39_cast_fp16 = matmul(transpose_x = matrix_ac_39_transpose_x_0, transpose_y = matrix_ac_39_transpose_y_0, x = transpose_134, y = transpose_135)[name = string("matrix_ac_39_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_79_begin_0 = const()[name = string("matrix_bd_79_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_79_end_0 = const()[name = string("matrix_bd_79_end_0"), val = tensor<int32, [4]>([1, 8, 126, 126])];
+            tensor<bool, [4]> matrix_bd_79_end_mask_0 = const()[name = string("matrix_bd_79_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 126, 126]> matrix_bd_79_cast_fp16 = slice_by_index(begin = matrix_bd_79_begin_0, end = matrix_bd_79_end_0, end_mask = matrix_bd_79_end_mask_0, x = matrix_bd_77_cast_fp16)[name = string("matrix_bd_79_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_3387_cast_fp16 = add(x = matrix_ac_39_cast_fp16, y = matrix_bd_79_cast_fp16)[name = string("op_3387_cast_fp16")];
+            fp16 _inversed_scores_77_y_0_to_fp16 = const()[name = string("_inversed_scores_77_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 126, 126]> _inversed_scores_77_cast_fp16 = mul(x = var_3387_cast_fp16, y = _inversed_scores_77_y_0_to_fp16)[name = string("_inversed_scores_77_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> scores_79_cast_fp16 = select(a = var_7_to_fp16, b = _inversed_scores_77_cast_fp16, cond = mask_3)[name = string("scores_79_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_3393_cast_fp16 = softmax(axis = var_25, x = scores_79_cast_fp16)[name = string("op_3393_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> input_1021_cast_fp16 = select(a = var_6_to_fp16, b = var_3393_cast_fp16, cond = mask_3)[name = string("input_1021_cast_fp16")];
+            bool x_431_transpose_x_0 = const()[name = string("x_431_transpose_x_0"), val = bool(false)];
+            bool x_431_transpose_y_0 = const()[name = string("x_431_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 126, 128]> value_39_cast_fp16 = transpose(perm = value_39_perm_0, x = v_39_cast_fp16)[name = string("transpose_176")];
+            tensor<fp16, [1, 8, 126, 128]> x_431_cast_fp16 = matmul(transpose_x = x_431_transpose_x_0, transpose_y = x_431_transpose_y_0, x = input_1021_cast_fp16, y = value_39_cast_fp16)[name = string("x_431_cast_fp16")];
+            tensor<int32, [4]> var_3397_perm_0 = const()[name = string("op_3397_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_3398 = const()[name = string("op_3398"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 126, 8, 128]> var_3397_cast_fp16 = transpose(perm = var_3397_perm_0, x = x_431_cast_fp16)[name = string("transpose_175")];
+            tensor<fp16, [1, 126, 1024]> input_1023_cast_fp16 = reshape(shape = var_3398, x = var_3397_cast_fp16)[name = string("input_1023_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_19_self_attn_linear_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(248356160))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(248880512))))[name = string("model_layers_19_self_attn_linear_out_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_178_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_19_self_attn_linear_out_weight_to_fp16_palettized, x = input_1023_cast_fp16)[name = string("linear_178_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_1027_cast_fp16 = add(x = input_1019_cast_fp16, y = linear_178_cast_fp16)[name = string("input_1027_cast_fp16")];
+            tensor<int32, [1]> x_435_axes_0 = const()[name = string("x_435_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_19_norm_conv_weight_to_fp16 = const()[name = string("model_layers_19_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(248881600)))];
+            tensor<fp16, [1024]> model_layers_19_norm_conv_bias_to_fp16 = const()[name = string("model_layers_19_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(248883712)))];
+            tensor<fp16, [1, 126, 1024]> x_435_cast_fp16 = layer_norm(axes = x_435_axes_0, beta = model_layers_19_norm_conv_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_19_norm_conv_weight_to_fp16, x = input_1027_cast_fp16)[name = string("x_435_cast_fp16")];
+            tensor<int32, [3]> input_1029_perm_0 = const()[name = string("input_1029_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_1031_pad_type_0 = const()[name = string("input_1031_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_1031_strides_0 = const()[name = string("input_1031_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_1031_pad_0 = const()[name = string("input_1031_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_1031_dilations_0 = const()[name = string("input_1031_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_1031_groups_0 = const()[name = string("input_1031_groups_0"), val = int32(1)];
+            tensor<fp16, [2048, 1024, 1]> model_layers_19_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(248885824))), lut = tensor<fp16, [64, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(249934464))))[name = string("model_layers_19_conv_pointwise_conv1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 1024, 126]> input_1029_cast_fp16 = transpose(perm = input_1029_perm_0, x = x_435_cast_fp16)[name = string("transpose_174")];
+            tensor<fp16, [1, 2048, 126]> input_1031_cast_fp16 = conv(dilations = input_1031_dilations_0, groups = input_1031_groups_0, pad = input_1031_pad_0, pad_type = input_1031_pad_type_0, strides = input_1031_strides_0, weight = model_layers_19_conv_pointwise_conv1_weight_to_fp16_palettized, x = input_1029_cast_fp16)[name = string("input_1031_cast_fp16")];
+            int32 x_437_split_num_splits_0 = const()[name = string("x_437_split_num_splits_0"), val = int32(2)];
+            int32 x_437_split_axis_0 = const()[name = string("x_437_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 126]> x_437_split_cast_fp16_0, tensor<fp16, [1, 1024, 126]> x_437_split_cast_fp16_1 = split(axis = x_437_split_axis_0, num_splits = x_437_split_num_splits_0, x = input_1031_cast_fp16)[name = string("x_437_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_437_split_1_sigmoid_cast_fp16 = sigmoid(x = x_437_split_cast_fp16_1)[name = string("x_437_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_437_cast_fp16 = mul(x = x_437_split_cast_fp16_0, y = x_437_split_1_sigmoid_cast_fp16)[name = string("x_437_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_1033_cast_fp16 = select(a = var_6_to_fp16, b = x_437_cast_fp16, cond = var_323)[name = string("input_1033_cast_fp16")];
+            tensor<int32, [6]> input_1035_pad_0 = const()[name = string("input_1035_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_1035_mode_0 = const()[name = string("input_1035_mode_0"), val = string("constant")];
+            fp16 const_207_to_fp16 = const()[name = string("const_207_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 134]> input_1035_cast_fp16 = pad(constant_val = const_207_to_fp16, mode = input_1035_mode_0, pad = input_1035_pad_0, x = input_1033_cast_fp16)[name = string("input_1035_cast_fp16")];
+            string input_1037_pad_type_0 = const()[name = string("input_1037_pad_type_0"), val = string("valid")];
+            int32 input_1037_groups_0 = const()[name = string("input_1037_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_1037_strides_0 = const()[name = string("input_1037_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_1037_pad_0 = const()[name = string("input_1037_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_1037_dilations_0 = const()[name = string("input_1037_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_286_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(249936576))), lut = tensor<fp16, [32, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(249941248))))[name = string("const_286_to_fp16_palettized")];
+            tensor<fp16, [1024]> const_287_to_fp16 = const()[name = string("const_287_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(249942336)))];
+            tensor<fp16, [1, 1024, 126]> input_1039_cast_fp16 = conv(bias = const_287_to_fp16, dilations = input_1037_dilations_0, groups = input_1037_groups_0, pad = input_1037_pad_0, pad_type = input_1037_pad_type_0, strides = input_1037_strides_0, weight = const_286_to_fp16_palettized, x = input_1035_cast_fp16)[name = string("input_1039_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_1041_cast_fp16 = silu(x = input_1039_cast_fp16)[name = string("input_1041_cast_fp16")];
+            string x_439_pad_type_0 = const()[name = string("x_439_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_439_strides_0 = const()[name = string("x_439_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_439_pad_0 = const()[name = string("x_439_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_439_dilations_0 = const()[name = string("x_439_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_439_groups_0 = const()[name = string("x_439_groups_0"), val = int32(1)];
+            tensor<fp16, [1024, 1024, 1]> model_layers_19_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(249944448))), lut = tensor<fp16, [32, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250468800))))[name = string("model_layers_19_conv_pointwise_conv2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 1024, 126]> x_439_cast_fp16 = conv(dilations = x_439_dilations_0, groups = x_439_groups_0, pad = x_439_pad_0, pad_type = x_439_pad_type_0, strides = x_439_strides_0, weight = model_layers_19_conv_pointwise_conv2_weight_to_fp16_palettized, x = input_1041_cast_fp16)[name = string("x_439_cast_fp16")];
+            tensor<int32, [3]> input_1043_perm_0 = const()[name = string("input_1043_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 126, 1024]> input_1043_cast_fp16 = transpose(perm = input_1043_perm_0, x = x_439_cast_fp16)[name = string("transpose_173")];
+            tensor<fp16, [1, 126, 1024]> input_1045_cast_fp16 = add(x = input_1027_cast_fp16, y = input_1043_cast_fp16)[name = string("input_1045_cast_fp16")];
+            tensor<int32, [1]> input_1047_axes_0 = const()[name = string("input_1047_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_19_norm_feed_forward2_weight_to_fp16 = const()[name = string("model_layers_19_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250469888)))];
+            tensor<fp16, [1024]> model_layers_19_norm_feed_forward2_bias_to_fp16 = const()[name = string("model_layers_19_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250472000)))];
+            tensor<fp16, [1, 126, 1024]> input_1047_cast_fp16 = layer_norm(axes = input_1047_axes_0, beta = model_layers_19_norm_feed_forward2_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_19_norm_feed_forward2_weight_to_fp16, x = input_1045_cast_fp16)[name = string("input_1047_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_19_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250474112))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(252571328))))[name = string("model_layers_19_feed_forward2_linear1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 4096]> linear_179_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_19_feed_forward2_linear1_weight_to_fp16_palettized, x = input_1047_cast_fp16)[name = string("linear_179_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_1051_cast_fp16 = silu(x = linear_179_cast_fp16)[name = string("input_1051_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_19_feed_forward2_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(252575488))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(254672704))))[name = string("model_layers_19_feed_forward2_linear2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_180_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_19_feed_forward2_linear2_weight_to_fp16_palettized, x = input_1051_cast_fp16)[name = string("linear_180_cast_fp16")];
+            fp16 var_3458_to_fp16 = const()[name = string("op_3458_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_3459_cast_fp16 = mul(x = linear_180_cast_fp16, y = var_3458_to_fp16)[name = string("op_3459_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_1057_cast_fp16 = add(x = input_1045_cast_fp16, y = var_3459_cast_fp16)[name = string("input_1057_cast_fp16")];
+            tensor<int32, [1]> input_1059_axes_0 = const()[name = string("input_1059_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_19_norm_out_weight_to_fp16 = const()[name = string("model_layers_19_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(254673792)))];
+            tensor<fp16, [1024]> model_layers_19_norm_out_bias_to_fp16 = const()[name = string("model_layers_19_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(254675904)))];
+            tensor<fp16, [1, 126, 1024]> input_1059_cast_fp16 = layer_norm(axes = input_1059_axes_0, beta = model_layers_19_norm_out_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_19_norm_out_weight_to_fp16, x = input_1057_cast_fp16)[name = string("input_1059_cast_fp16")];
+            tensor<int32, [1]> input_1061_axes_0 = const()[name = string("input_1061_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_20_norm_feed_forward1_weight_to_fp16 = const()[name = string("model_layers_20_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(254678016)))];
+            tensor<fp16, [1024]> model_layers_20_norm_feed_forward1_bias_to_fp16 = const()[name = string("model_layers_20_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(254680128)))];
+            tensor<fp16, [1, 126, 1024]> input_1061_cast_fp16 = layer_norm(axes = input_1061_axes_0, beta = model_layers_20_norm_feed_forward1_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_20_norm_feed_forward1_weight_to_fp16, x = input_1059_cast_fp16)[name = string("input_1061_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_20_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(254682240))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(256779456))))[name = string("model_layers_20_feed_forward1_linear1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 4096]> linear_181_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_20_feed_forward1_linear1_weight_to_fp16_palettized, x = input_1061_cast_fp16)[name = string("linear_181_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_1065_cast_fp16 = silu(x = linear_181_cast_fp16)[name = string("input_1065_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_20_feed_forward1_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(256783616))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(258880832))))[name = string("model_layers_20_feed_forward1_linear2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_182_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_20_feed_forward1_linear2_weight_to_fp16_palettized, x = input_1065_cast_fp16)[name = string("linear_182_cast_fp16")];
+            fp16 var_3487_to_fp16 = const()[name = string("op_3487_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_3488_cast_fp16 = mul(x = linear_182_cast_fp16, y = var_3487_to_fp16)[name = string("op_3488_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_1071_cast_fp16 = add(x = input_1059_cast_fp16, y = var_3488_cast_fp16)[name = string("input_1071_cast_fp16")];
+            tensor<int32, [1]> query_41_axes_0 = const()[name = string("query_41_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_20_norm_self_att_weight_to_fp16 = const()[name = string("model_layers_20_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(258881920)))];
+            tensor<fp16, [1024]> model_layers_20_norm_self_att_bias_to_fp16 = const()[name = string("model_layers_20_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(258884032)))];
+            tensor<fp16, [1, 126, 1024]> query_41_cast_fp16 = layer_norm(axes = query_41_axes_0, beta = model_layers_20_norm_self_att_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_20_norm_self_att_weight_to_fp16, x = input_1071_cast_fp16)[name = string("query_41_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_20_self_attn_linear_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(258886144))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(259410496))))[name = string("model_layers_20_self_attn_linear_q_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_183_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_20_self_attn_linear_q_weight_to_fp16_palettized, x = query_41_cast_fp16)[name = string("linear_183_cast_fp16")];
+            tensor<int32, [4]> var_3504 = const()[name = string("op_3504"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> q_121_cast_fp16 = reshape(shape = var_3504, x = linear_183_cast_fp16)[name = string("q_121_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_20_self_attn_linear_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(259411584))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(259935936))))[name = string("model_layers_20_self_attn_linear_k_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_184_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_20_self_attn_linear_k_weight_to_fp16_palettized, x = query_41_cast_fp16)[name = string("linear_184_cast_fp16")];
+            tensor<int32, [4]> var_3508 = const()[name = string("op_3508"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> k_81_cast_fp16 = reshape(shape = var_3508, x = linear_184_cast_fp16)[name = string("k_81_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_20_self_attn_linear_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(259937024))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(260461376))))[name = string("model_layers_20_self_attn_linear_v_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_185_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_20_self_attn_linear_v_weight_to_fp16_palettized, x = query_41_cast_fp16)[name = string("linear_185_cast_fp16")];
+            tensor<int32, [4]> var_3512 = const()[name = string("op_3512"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> v_41_cast_fp16 = reshape(shape = var_3512, x = linear_185_cast_fp16)[name = string("v_41_cast_fp16")];
+            tensor<int32, [4]> value_41_perm_0 = const()[name = string("value_41_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> model_layers_20_self_attn_pos_bias_u_to_fp16 = const()[name = string("model_layers_20_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(260462464)))];
+            tensor<fp16, [1, 126, 8, 128]> var_3524_cast_fp16 = add(x = q_121_cast_fp16, y = model_layers_20_self_attn_pos_bias_u_to_fp16)[name = string("op_3524_cast_fp16")];
+            tensor<fp16, [8, 128]> model_layers_20_self_attn_pos_bias_v_to_fp16 = const()[name = string("model_layers_20_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(260464576)))];
+            tensor<fp16, [1, 126, 8, 128]> var_3526_cast_fp16 = add(x = q_121_cast_fp16, y = model_layers_20_self_attn_pos_bias_v_to_fp16)[name = string("op_3526_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_41_perm_0 = const()[name = string("q_with_bias_v_41_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_447_transpose_x_0 = const()[name = string("x_447_transpose_x_0"), val = bool(false)];
+            bool x_447_transpose_y_0 = const()[name = string("x_447_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 251]> var_3528_to_fp16 = const()[name = string("op_3528_to_fp16"), val = tensor<fp16, [1, 8, 128, 251]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(260466688)))];
+            tensor<fp16, [1, 8, 126, 128]> q_with_bias_v_41_cast_fp16 = transpose(perm = q_with_bias_v_41_perm_0, x = var_3526_cast_fp16)[name = string("transpose_172")];
+            tensor<fp16, [1, 8, 126, 251]> x_447_cast_fp16 = matmul(transpose_x = x_447_transpose_x_0, transpose_y = x_447_transpose_y_0, x = q_with_bias_v_41_cast_fp16, y = var_3528_to_fp16)[name = string("x_447_cast_fp16")];
+            tensor<int32, [8]> x_449_pad_0 = const()[name = string("x_449_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_449_mode_0 = const()[name = string("x_449_mode_0"), val = string("constant")];
+            fp16 const_214_to_fp16 = const()[name = string("const_214_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 252]> x_449_cast_fp16 = pad(constant_val = const_214_to_fp16, mode = x_449_mode_0, pad = x_449_pad_0, x = x_447_cast_fp16)[name = string("x_449_cast_fp16")];
+            tensor<int32, [4]> var_3536 = const()[name = string("op_3536"), val = tensor<int32, [4]>([1, 8, -1, 126])];
+            tensor<fp16, [1, 8, 252, 126]> x_451_cast_fp16 = reshape(shape = var_3536, x = x_449_cast_fp16)[name = string("x_451_cast_fp16")];
+            tensor<int32, [4]> var_3540_begin_0 = const()[name = string("op_3540_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_3540_end_0 = const()[name = string("op_3540_end_0"), val = tensor<int32, [4]>([1, 8, 252, 126])];
+            tensor<bool, [4]> var_3540_end_mask_0 = const()[name = string("op_3540_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 251, 126]> var_3540_cast_fp16 = slice_by_index(begin = var_3540_begin_0, end = var_3540_end_0, end_mask = var_3540_end_mask_0, x = x_451_cast_fp16)[name = string("op_3540_cast_fp16")];
+            tensor<int32, [4]> var_3541 = const()[name = string("op_3541"), val = tensor<int32, [4]>([1, 8, 126, 251])];
+            tensor<fp16, [1, 8, 126, 251]> matrix_bd_81_cast_fp16 = reshape(shape = var_3541, x = var_3540_cast_fp16)[name = string("matrix_bd_81_cast_fp16")];
+            bool matrix_ac_41_transpose_x_0 = const()[name = string("matrix_ac_41_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_41_transpose_y_0 = const()[name = string("matrix_ac_41_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_136_perm_0 = const()[name = string("transpose_136_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_137_perm_0 = const()[name = string("transpose_137_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 126]> transpose_137 = transpose(perm = transpose_137_perm_0, x = k_81_cast_fp16)[name = string("transpose_170")];
+            tensor<fp16, [1, 8, 126, 128]> transpose_136 = transpose(perm = transpose_136_perm_0, x = var_3524_cast_fp16)[name = string("transpose_171")];
+            tensor<fp16, [1, 8, 126, 126]> matrix_ac_41_cast_fp16 = matmul(transpose_x = matrix_ac_41_transpose_x_0, transpose_y = matrix_ac_41_transpose_y_0, x = transpose_136, y = transpose_137)[name = string("matrix_ac_41_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_83_begin_0 = const()[name = string("matrix_bd_83_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_83_end_0 = const()[name = string("matrix_bd_83_end_0"), val = tensor<int32, [4]>([1, 8, 126, 126])];
+            tensor<bool, [4]> matrix_bd_83_end_mask_0 = const()[name = string("matrix_bd_83_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 126, 126]> matrix_bd_83_cast_fp16 = slice_by_index(begin = matrix_bd_83_begin_0, end = matrix_bd_83_end_0, end_mask = matrix_bd_83_end_mask_0, x = matrix_bd_81_cast_fp16)[name = string("matrix_bd_83_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_3550_cast_fp16 = add(x = matrix_ac_41_cast_fp16, y = matrix_bd_83_cast_fp16)[name = string("op_3550_cast_fp16")];
+            fp16 _inversed_scores_81_y_0_to_fp16 = const()[name = string("_inversed_scores_81_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 126, 126]> _inversed_scores_81_cast_fp16 = mul(x = var_3550_cast_fp16, y = _inversed_scores_81_y_0_to_fp16)[name = string("_inversed_scores_81_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> scores_83_cast_fp16 = select(a = var_7_to_fp16, b = _inversed_scores_81_cast_fp16, cond = mask_3)[name = string("scores_83_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_3556_cast_fp16 = softmax(axis = var_25, x = scores_83_cast_fp16)[name = string("op_3556_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> input_1073_cast_fp16 = select(a = var_6_to_fp16, b = var_3556_cast_fp16, cond = mask_3)[name = string("input_1073_cast_fp16")];
+            bool x_453_transpose_x_0 = const()[name = string("x_453_transpose_x_0"), val = bool(false)];
+            bool x_453_transpose_y_0 = const()[name = string("x_453_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 126, 128]> value_41_cast_fp16 = transpose(perm = value_41_perm_0, x = v_41_cast_fp16)[name = string("transpose_169")];
+            tensor<fp16, [1, 8, 126, 128]> x_453_cast_fp16 = matmul(transpose_x = x_453_transpose_x_0, transpose_y = x_453_transpose_y_0, x = input_1073_cast_fp16, y = value_41_cast_fp16)[name = string("x_453_cast_fp16")];
+            tensor<int32, [4]> var_3560_perm_0 = const()[name = string("op_3560_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_3561 = const()[name = string("op_3561"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 126, 8, 128]> var_3560_cast_fp16 = transpose(perm = var_3560_perm_0, x = x_453_cast_fp16)[name = string("transpose_168")];
+            tensor<fp16, [1, 126, 1024]> input_1075_cast_fp16 = reshape(shape = var_3561, x = var_3560_cast_fp16)[name = string("input_1075_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_20_self_attn_linear_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(260980800))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261505152))))[name = string("model_layers_20_self_attn_linear_out_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_187_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_20_self_attn_linear_out_weight_to_fp16_palettized, x = input_1075_cast_fp16)[name = string("linear_187_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_1079_cast_fp16 = add(x = input_1071_cast_fp16, y = linear_187_cast_fp16)[name = string("input_1079_cast_fp16")];
+            tensor<int32, [1]> x_457_axes_0 = const()[name = string("x_457_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_20_norm_conv_weight_to_fp16 = const()[name = string("model_layers_20_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261506240)))];
+            tensor<fp16, [1024]> model_layers_20_norm_conv_bias_to_fp16 = const()[name = string("model_layers_20_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261508352)))];
+            tensor<fp16, [1, 126, 1024]> x_457_cast_fp16 = layer_norm(axes = x_457_axes_0, beta = model_layers_20_norm_conv_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_20_norm_conv_weight_to_fp16, x = input_1079_cast_fp16)[name = string("x_457_cast_fp16")];
+            tensor<int32, [3]> input_1081_perm_0 = const()[name = string("input_1081_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_1083_pad_type_0 = const()[name = string("input_1083_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_1083_strides_0 = const()[name = string("input_1083_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_1083_pad_0 = const()[name = string("input_1083_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_1083_dilations_0 = const()[name = string("input_1083_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_1083_groups_0 = const()[name = string("input_1083_groups_0"), val = int32(1)];
+            tensor<fp16, [2048, 1024, 1]> model_layers_20_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261510464))), lut = tensor<fp16, [64, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262559104))))[name = string("model_layers_20_conv_pointwise_conv1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 1024, 126]> input_1081_cast_fp16 = transpose(perm = input_1081_perm_0, x = x_457_cast_fp16)[name = string("transpose_167")];
+            tensor<fp16, [1, 2048, 126]> input_1083_cast_fp16 = conv(dilations = input_1083_dilations_0, groups = input_1083_groups_0, pad = input_1083_pad_0, pad_type = input_1083_pad_type_0, strides = input_1083_strides_0, weight = model_layers_20_conv_pointwise_conv1_weight_to_fp16_palettized, x = input_1081_cast_fp16)[name = string("input_1083_cast_fp16")];
+            int32 x_459_split_num_splits_0 = const()[name = string("x_459_split_num_splits_0"), val = int32(2)];
+            int32 x_459_split_axis_0 = const()[name = string("x_459_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 126]> x_459_split_cast_fp16_0, tensor<fp16, [1, 1024, 126]> x_459_split_cast_fp16_1 = split(axis = x_459_split_axis_0, num_splits = x_459_split_num_splits_0, x = input_1083_cast_fp16)[name = string("x_459_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_459_split_1_sigmoid_cast_fp16 = sigmoid(x = x_459_split_cast_fp16_1)[name = string("x_459_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_459_cast_fp16 = mul(x = x_459_split_cast_fp16_0, y = x_459_split_1_sigmoid_cast_fp16)[name = string("x_459_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_1085_cast_fp16 = select(a = var_6_to_fp16, b = x_459_cast_fp16, cond = var_323)[name = string("input_1085_cast_fp16")];
+            tensor<int32, [6]> input_1087_pad_0 = const()[name = string("input_1087_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_1087_mode_0 = const()[name = string("input_1087_mode_0"), val = string("constant")];
+            fp16 const_217_to_fp16 = const()[name = string("const_217_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 134]> input_1087_cast_fp16 = pad(constant_val = const_217_to_fp16, mode = input_1087_mode_0, pad = input_1087_pad_0, x = input_1085_cast_fp16)[name = string("input_1087_cast_fp16")];
+            string input_1089_pad_type_0 = const()[name = string("input_1089_pad_type_0"), val = string("valid")];
+            int32 input_1089_groups_0 = const()[name = string("input_1089_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_1089_strides_0 = const()[name = string("input_1089_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_1089_pad_0 = const()[name = string("input_1089_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_1089_dilations_0 = const()[name = string("input_1089_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_288_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262561216))), lut = tensor<fp16, [32, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262565888))))[name = string("const_288_to_fp16_palettized")];
+            tensor<fp16, [1024]> const_289_to_fp16 = const()[name = string("const_289_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262566976)))];
+            tensor<fp16, [1, 1024, 126]> input_1091_cast_fp16 = conv(bias = const_289_to_fp16, dilations = input_1089_dilations_0, groups = input_1089_groups_0, pad = input_1089_pad_0, pad_type = input_1089_pad_type_0, strides = input_1089_strides_0, weight = const_288_to_fp16_palettized, x = input_1087_cast_fp16)[name = string("input_1091_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_1093_cast_fp16 = silu(x = input_1091_cast_fp16)[name = string("input_1093_cast_fp16")];
+            string x_461_pad_type_0 = const()[name = string("x_461_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_461_strides_0 = const()[name = string("x_461_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_461_pad_0 = const()[name = string("x_461_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_461_dilations_0 = const()[name = string("x_461_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_461_groups_0 = const()[name = string("x_461_groups_0"), val = int32(1)];
+            tensor<fp16, [1024, 1024, 1]> model_layers_20_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262569088))), lut = tensor<fp16, [32, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263093440))))[name = string("model_layers_20_conv_pointwise_conv2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 1024, 126]> x_461_cast_fp16 = conv(dilations = x_461_dilations_0, groups = x_461_groups_0, pad = x_461_pad_0, pad_type = x_461_pad_type_0, strides = x_461_strides_0, weight = model_layers_20_conv_pointwise_conv2_weight_to_fp16_palettized, x = input_1093_cast_fp16)[name = string("x_461_cast_fp16")];
+            tensor<int32, [3]> input_1095_perm_0 = const()[name = string("input_1095_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 126, 1024]> input_1095_cast_fp16 = transpose(perm = input_1095_perm_0, x = x_461_cast_fp16)[name = string("transpose_166")];
+            tensor<fp16, [1, 126, 1024]> input_1097_cast_fp16 = add(x = input_1079_cast_fp16, y = input_1095_cast_fp16)[name = string("input_1097_cast_fp16")];
+            tensor<int32, [1]> input_1099_axes_0 = const()[name = string("input_1099_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_20_norm_feed_forward2_weight_to_fp16 = const()[name = string("model_layers_20_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263094528)))];
+            tensor<fp16, [1024]> model_layers_20_norm_feed_forward2_bias_to_fp16 = const()[name = string("model_layers_20_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263096640)))];
+            tensor<fp16, [1, 126, 1024]> input_1099_cast_fp16 = layer_norm(axes = input_1099_axes_0, beta = model_layers_20_norm_feed_forward2_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_20_norm_feed_forward2_weight_to_fp16, x = input_1097_cast_fp16)[name = string("input_1099_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_20_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263098752))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265195968))))[name = string("model_layers_20_feed_forward2_linear1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 4096]> linear_188_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_20_feed_forward2_linear1_weight_to_fp16_palettized, x = input_1099_cast_fp16)[name = string("linear_188_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_1103_cast_fp16 = silu(x = linear_188_cast_fp16)[name = string("input_1103_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_20_feed_forward2_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265200128))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(267297344))))[name = string("model_layers_20_feed_forward2_linear2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_189_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_20_feed_forward2_linear2_weight_to_fp16_palettized, x = input_1103_cast_fp16)[name = string("linear_189_cast_fp16")];
+            fp16 var_3621_to_fp16 = const()[name = string("op_3621_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_3622_cast_fp16 = mul(x = linear_189_cast_fp16, y = var_3621_to_fp16)[name = string("op_3622_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_1109_cast_fp16 = add(x = input_1097_cast_fp16, y = var_3622_cast_fp16)[name = string("input_1109_cast_fp16")];
+            tensor<int32, [1]> input_1111_axes_0 = const()[name = string("input_1111_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_20_norm_out_weight_to_fp16 = const()[name = string("model_layers_20_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(267298432)))];
+            tensor<fp16, [1024]> model_layers_20_norm_out_bias_to_fp16 = const()[name = string("model_layers_20_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(267300544)))];
+            tensor<fp16, [1, 126, 1024]> input_1111_cast_fp16 = layer_norm(axes = input_1111_axes_0, beta = model_layers_20_norm_out_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_20_norm_out_weight_to_fp16, x = input_1109_cast_fp16)[name = string("input_1111_cast_fp16")];
+            tensor<int32, [1]> input_1113_axes_0 = const()[name = string("input_1113_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_21_norm_feed_forward1_weight_to_fp16 = const()[name = string("model_layers_21_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(267302656)))];
+            tensor<fp16, [1024]> model_layers_21_norm_feed_forward1_bias_to_fp16 = const()[name = string("model_layers_21_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(267304768)))];
+            tensor<fp16, [1, 126, 1024]> input_1113_cast_fp16 = layer_norm(axes = input_1113_axes_0, beta = model_layers_21_norm_feed_forward1_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_21_norm_feed_forward1_weight_to_fp16, x = input_1111_cast_fp16)[name = string("input_1113_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_21_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(267306880))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269404096))))[name = string("model_layers_21_feed_forward1_linear1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 4096]> linear_190_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_21_feed_forward1_linear1_weight_to_fp16_palettized, x = input_1113_cast_fp16)[name = string("linear_190_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_1117_cast_fp16 = silu(x = linear_190_cast_fp16)[name = string("input_1117_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_21_feed_forward1_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269408256))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271505472))))[name = string("model_layers_21_feed_forward1_linear2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_191_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_21_feed_forward1_linear2_weight_to_fp16_palettized, x = input_1117_cast_fp16)[name = string("linear_191_cast_fp16")];
+            fp16 var_3650_to_fp16 = const()[name = string("op_3650_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_3651_cast_fp16 = mul(x = linear_191_cast_fp16, y = var_3650_to_fp16)[name = string("op_3651_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_1123_cast_fp16 = add(x = input_1111_cast_fp16, y = var_3651_cast_fp16)[name = string("input_1123_cast_fp16")];
+            tensor<int32, [1]> query_43_axes_0 = const()[name = string("query_43_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_21_norm_self_att_weight_to_fp16 = const()[name = string("model_layers_21_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271506560)))];
+            tensor<fp16, [1024]> model_layers_21_norm_self_att_bias_to_fp16 = const()[name = string("model_layers_21_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271508672)))];
+            tensor<fp16, [1, 126, 1024]> query_43_cast_fp16 = layer_norm(axes = query_43_axes_0, beta = model_layers_21_norm_self_att_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_21_norm_self_att_weight_to_fp16, x = input_1123_cast_fp16)[name = string("query_43_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_21_self_attn_linear_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271510784))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(272035136))))[name = string("model_layers_21_self_attn_linear_q_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_192_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_21_self_attn_linear_q_weight_to_fp16_palettized, x = query_43_cast_fp16)[name = string("linear_192_cast_fp16")];
+            tensor<int32, [4]> var_3667 = const()[name = string("op_3667"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> q_127_cast_fp16 = reshape(shape = var_3667, x = linear_192_cast_fp16)[name = string("q_127_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_21_self_attn_linear_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(272036224))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(272560576))))[name = string("model_layers_21_self_attn_linear_k_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_193_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_21_self_attn_linear_k_weight_to_fp16_palettized, x = query_43_cast_fp16)[name = string("linear_193_cast_fp16")];
+            tensor<int32, [4]> var_3671 = const()[name = string("op_3671"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> k_85_cast_fp16 = reshape(shape = var_3671, x = linear_193_cast_fp16)[name = string("k_85_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_21_self_attn_linear_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(272561664))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(273086016))))[name = string("model_layers_21_self_attn_linear_v_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_194_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_21_self_attn_linear_v_weight_to_fp16_palettized, x = query_43_cast_fp16)[name = string("linear_194_cast_fp16")];
+            tensor<int32, [4]> var_3675 = const()[name = string("op_3675"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> v_43_cast_fp16 = reshape(shape = var_3675, x = linear_194_cast_fp16)[name = string("v_43_cast_fp16")];
+            tensor<int32, [4]> value_43_perm_0 = const()[name = string("value_43_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> model_layers_21_self_attn_pos_bias_u_to_fp16 = const()[name = string("model_layers_21_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(273087104)))];
+            tensor<fp16, [1, 126, 8, 128]> var_3687_cast_fp16 = add(x = q_127_cast_fp16, y = model_layers_21_self_attn_pos_bias_u_to_fp16)[name = string("op_3687_cast_fp16")];
+            tensor<fp16, [8, 128]> model_layers_21_self_attn_pos_bias_v_to_fp16 = const()[name = string("model_layers_21_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(273089216)))];
+            tensor<fp16, [1, 126, 8, 128]> var_3689_cast_fp16 = add(x = q_127_cast_fp16, y = model_layers_21_self_attn_pos_bias_v_to_fp16)[name = string("op_3689_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_43_perm_0 = const()[name = string("q_with_bias_v_43_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_469_transpose_x_0 = const()[name = string("x_469_transpose_x_0"), val = bool(false)];
+            bool x_469_transpose_y_0 = const()[name = string("x_469_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 251]> var_3691_to_fp16 = const()[name = string("op_3691_to_fp16"), val = tensor<fp16, [1, 8, 128, 251]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(273091328)))];
+            tensor<fp16, [1, 8, 126, 128]> q_with_bias_v_43_cast_fp16 = transpose(perm = q_with_bias_v_43_perm_0, x = var_3689_cast_fp16)[name = string("transpose_165")];
+            tensor<fp16, [1, 8, 126, 251]> x_469_cast_fp16 = matmul(transpose_x = x_469_transpose_x_0, transpose_y = x_469_transpose_y_0, x = q_with_bias_v_43_cast_fp16, y = var_3691_to_fp16)[name = string("x_469_cast_fp16")];
+            tensor<int32, [8]> x_471_pad_0 = const()[name = string("x_471_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_471_mode_0 = const()[name = string("x_471_mode_0"), val = string("constant")];
+            fp16 const_224_to_fp16 = const()[name = string("const_224_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 252]> x_471_cast_fp16 = pad(constant_val = const_224_to_fp16, mode = x_471_mode_0, pad = x_471_pad_0, x = x_469_cast_fp16)[name = string("x_471_cast_fp16")];
+            tensor<int32, [4]> var_3699 = const()[name = string("op_3699"), val = tensor<int32, [4]>([1, 8, -1, 126])];
+            tensor<fp16, [1, 8, 252, 126]> x_473_cast_fp16 = reshape(shape = var_3699, x = x_471_cast_fp16)[name = string("x_473_cast_fp16")];
+            tensor<int32, [4]> var_3703_begin_0 = const()[name = string("op_3703_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_3703_end_0 = const()[name = string("op_3703_end_0"), val = tensor<int32, [4]>([1, 8, 252, 126])];
+            tensor<bool, [4]> var_3703_end_mask_0 = const()[name = string("op_3703_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 251, 126]> var_3703_cast_fp16 = slice_by_index(begin = var_3703_begin_0, end = var_3703_end_0, end_mask = var_3703_end_mask_0, x = x_473_cast_fp16)[name = string("op_3703_cast_fp16")];
+            tensor<int32, [4]> var_3704 = const()[name = string("op_3704"), val = tensor<int32, [4]>([1, 8, 126, 251])];
+            tensor<fp16, [1, 8, 126, 251]> matrix_bd_85_cast_fp16 = reshape(shape = var_3704, x = var_3703_cast_fp16)[name = string("matrix_bd_85_cast_fp16")];
+            bool matrix_ac_43_transpose_x_0 = const()[name = string("matrix_ac_43_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_43_transpose_y_0 = const()[name = string("matrix_ac_43_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_138_perm_0 = const()[name = string("transpose_138_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_139_perm_0 = const()[name = string("transpose_139_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 126]> transpose_139 = transpose(perm = transpose_139_perm_0, x = k_85_cast_fp16)[name = string("transpose_163")];
+            tensor<fp16, [1, 8, 126, 128]> transpose_138 = transpose(perm = transpose_138_perm_0, x = var_3687_cast_fp16)[name = string("transpose_164")];
+            tensor<fp16, [1, 8, 126, 126]> matrix_ac_43_cast_fp16 = matmul(transpose_x = matrix_ac_43_transpose_x_0, transpose_y = matrix_ac_43_transpose_y_0, x = transpose_138, y = transpose_139)[name = string("matrix_ac_43_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_87_begin_0 = const()[name = string("matrix_bd_87_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_87_end_0 = const()[name = string("matrix_bd_87_end_0"), val = tensor<int32, [4]>([1, 8, 126, 126])];
+            tensor<bool, [4]> matrix_bd_87_end_mask_0 = const()[name = string("matrix_bd_87_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 126, 126]> matrix_bd_87_cast_fp16 = slice_by_index(begin = matrix_bd_87_begin_0, end = matrix_bd_87_end_0, end_mask = matrix_bd_87_end_mask_0, x = matrix_bd_85_cast_fp16)[name = string("matrix_bd_87_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_3713_cast_fp16 = add(x = matrix_ac_43_cast_fp16, y = matrix_bd_87_cast_fp16)[name = string("op_3713_cast_fp16")];
+            fp16 _inversed_scores_85_y_0_to_fp16 = const()[name = string("_inversed_scores_85_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 126, 126]> _inversed_scores_85_cast_fp16 = mul(x = var_3713_cast_fp16, y = _inversed_scores_85_y_0_to_fp16)[name = string("_inversed_scores_85_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> scores_87_cast_fp16 = select(a = var_7_to_fp16, b = _inversed_scores_85_cast_fp16, cond = mask_3)[name = string("scores_87_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_3719_cast_fp16 = softmax(axis = var_25, x = scores_87_cast_fp16)[name = string("op_3719_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> input_1125_cast_fp16 = select(a = var_6_to_fp16, b = var_3719_cast_fp16, cond = mask_3)[name = string("input_1125_cast_fp16")];
+            bool x_475_transpose_x_0 = const()[name = string("x_475_transpose_x_0"), val = bool(false)];
+            bool x_475_transpose_y_0 = const()[name = string("x_475_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 126, 128]> value_43_cast_fp16 = transpose(perm = value_43_perm_0, x = v_43_cast_fp16)[name = string("transpose_162")];
+            tensor<fp16, [1, 8, 126, 128]> x_475_cast_fp16 = matmul(transpose_x = x_475_transpose_x_0, transpose_y = x_475_transpose_y_0, x = input_1125_cast_fp16, y = value_43_cast_fp16)[name = string("x_475_cast_fp16")];
+            tensor<int32, [4]> var_3723_perm_0 = const()[name = string("op_3723_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_3724 = const()[name = string("op_3724"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 126, 8, 128]> var_3723_cast_fp16 = transpose(perm = var_3723_perm_0, x = x_475_cast_fp16)[name = string("transpose_161")];
+            tensor<fp16, [1, 126, 1024]> input_1127_cast_fp16 = reshape(shape = var_3724, x = var_3723_cast_fp16)[name = string("input_1127_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_21_self_attn_linear_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(273605440))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274129792))))[name = string("model_layers_21_self_attn_linear_out_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_196_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_21_self_attn_linear_out_weight_to_fp16_palettized, x = input_1127_cast_fp16)[name = string("linear_196_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_1131_cast_fp16 = add(x = input_1123_cast_fp16, y = linear_196_cast_fp16)[name = string("input_1131_cast_fp16")];
+            tensor<int32, [1]> x_479_axes_0 = const()[name = string("x_479_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_21_norm_conv_weight_to_fp16 = const()[name = string("model_layers_21_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274130880)))];
+            tensor<fp16, [1024]> model_layers_21_norm_conv_bias_to_fp16 = const()[name = string("model_layers_21_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274132992)))];
+            tensor<fp16, [1, 126, 1024]> x_479_cast_fp16 = layer_norm(axes = x_479_axes_0, beta = model_layers_21_norm_conv_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_21_norm_conv_weight_to_fp16, x = input_1131_cast_fp16)[name = string("x_479_cast_fp16")];
+            tensor<int32, [3]> input_1133_perm_0 = const()[name = string("input_1133_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_1135_pad_type_0 = const()[name = string("input_1135_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_1135_strides_0 = const()[name = string("input_1135_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_1135_pad_0 = const()[name = string("input_1135_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_1135_dilations_0 = const()[name = string("input_1135_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_1135_groups_0 = const()[name = string("input_1135_groups_0"), val = int32(1)];
+            tensor<fp16, [2048, 1024, 1]> model_layers_21_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274135104))), lut = tensor<fp16, [64, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275183744))))[name = string("model_layers_21_conv_pointwise_conv1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 1024, 126]> input_1133_cast_fp16 = transpose(perm = input_1133_perm_0, x = x_479_cast_fp16)[name = string("transpose_160")];
+            tensor<fp16, [1, 2048, 126]> input_1135_cast_fp16 = conv(dilations = input_1135_dilations_0, groups = input_1135_groups_0, pad = input_1135_pad_0, pad_type = input_1135_pad_type_0, strides = input_1135_strides_0, weight = model_layers_21_conv_pointwise_conv1_weight_to_fp16_palettized, x = input_1133_cast_fp16)[name = string("input_1135_cast_fp16")];
+            int32 x_481_split_num_splits_0 = const()[name = string("x_481_split_num_splits_0"), val = int32(2)];
+            int32 x_481_split_axis_0 = const()[name = string("x_481_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 126]> x_481_split_cast_fp16_0, tensor<fp16, [1, 1024, 126]> x_481_split_cast_fp16_1 = split(axis = x_481_split_axis_0, num_splits = x_481_split_num_splits_0, x = input_1135_cast_fp16)[name = string("x_481_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_481_split_1_sigmoid_cast_fp16 = sigmoid(x = x_481_split_cast_fp16_1)[name = string("x_481_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_481_cast_fp16 = mul(x = x_481_split_cast_fp16_0, y = x_481_split_1_sigmoid_cast_fp16)[name = string("x_481_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_1137_cast_fp16 = select(a = var_6_to_fp16, b = x_481_cast_fp16, cond = var_323)[name = string("input_1137_cast_fp16")];
+            tensor<int32, [6]> input_1139_pad_0 = const()[name = string("input_1139_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_1139_mode_0 = const()[name = string("input_1139_mode_0"), val = string("constant")];
+            fp16 const_227_to_fp16 = const()[name = string("const_227_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 134]> input_1139_cast_fp16 = pad(constant_val = const_227_to_fp16, mode = input_1139_mode_0, pad = input_1139_pad_0, x = input_1137_cast_fp16)[name = string("input_1139_cast_fp16")];
+            string input_1141_pad_type_0 = const()[name = string("input_1141_pad_type_0"), val = string("valid")];
+            int32 input_1141_groups_0 = const()[name = string("input_1141_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_1141_strides_0 = const()[name = string("input_1141_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_1141_pad_0 = const()[name = string("input_1141_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_1141_dilations_0 = const()[name = string("input_1141_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_290_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275185856))), lut = tensor<fp16, [32, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275190528))))[name = string("const_290_to_fp16_palettized")];
+            tensor<fp16, [1024]> const_291_to_fp16 = const()[name = string("const_291_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275191616)))];
+            tensor<fp16, [1, 1024, 126]> input_1143_cast_fp16 = conv(bias = const_291_to_fp16, dilations = input_1141_dilations_0, groups = input_1141_groups_0, pad = input_1141_pad_0, pad_type = input_1141_pad_type_0, strides = input_1141_strides_0, weight = const_290_to_fp16_palettized, x = input_1139_cast_fp16)[name = string("input_1143_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_1145_cast_fp16 = silu(x = input_1143_cast_fp16)[name = string("input_1145_cast_fp16")];
+            string x_483_pad_type_0 = const()[name = string("x_483_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_483_strides_0 = const()[name = string("x_483_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_483_pad_0 = const()[name = string("x_483_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_483_dilations_0 = const()[name = string("x_483_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_483_groups_0 = const()[name = string("x_483_groups_0"), val = int32(1)];
+            tensor<fp16, [1024, 1024, 1]> model_layers_21_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275193728))), lut = tensor<fp16, [32, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275718080))))[name = string("model_layers_21_conv_pointwise_conv2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 1024, 126]> x_483_cast_fp16 = conv(dilations = x_483_dilations_0, groups = x_483_groups_0, pad = x_483_pad_0, pad_type = x_483_pad_type_0, strides = x_483_strides_0, weight = model_layers_21_conv_pointwise_conv2_weight_to_fp16_palettized, x = input_1145_cast_fp16)[name = string("x_483_cast_fp16")];
+            tensor<int32, [3]> input_1147_perm_0 = const()[name = string("input_1147_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 126, 1024]> input_1147_cast_fp16 = transpose(perm = input_1147_perm_0, x = x_483_cast_fp16)[name = string("transpose_159")];
+            tensor<fp16, [1, 126, 1024]> input_1149_cast_fp16 = add(x = input_1131_cast_fp16, y = input_1147_cast_fp16)[name = string("input_1149_cast_fp16")];
+            tensor<int32, [1]> input_1151_axes_0 = const()[name = string("input_1151_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_21_norm_feed_forward2_weight_to_fp16 = const()[name = string("model_layers_21_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275719168)))];
+            tensor<fp16, [1024]> model_layers_21_norm_feed_forward2_bias_to_fp16 = const()[name = string("model_layers_21_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275721280)))];
+            tensor<fp16, [1, 126, 1024]> input_1151_cast_fp16 = layer_norm(axes = input_1151_axes_0, beta = model_layers_21_norm_feed_forward2_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_21_norm_feed_forward2_weight_to_fp16, x = input_1149_cast_fp16)[name = string("input_1151_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_21_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275723392))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(277820608))))[name = string("model_layers_21_feed_forward2_linear1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 4096]> linear_197_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_21_feed_forward2_linear1_weight_to_fp16_palettized, x = input_1151_cast_fp16)[name = string("linear_197_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_1155_cast_fp16 = silu(x = linear_197_cast_fp16)[name = string("input_1155_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_21_feed_forward2_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(277824768))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(279921984))))[name = string("model_layers_21_feed_forward2_linear2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_198_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_21_feed_forward2_linear2_weight_to_fp16_palettized, x = input_1155_cast_fp16)[name = string("linear_198_cast_fp16")];
+            fp16 var_3784_to_fp16 = const()[name = string("op_3784_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_3785_cast_fp16 = mul(x = linear_198_cast_fp16, y = var_3784_to_fp16)[name = string("op_3785_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_1161_cast_fp16 = add(x = input_1149_cast_fp16, y = var_3785_cast_fp16)[name = string("input_1161_cast_fp16")];
+            tensor<int32, [1]> input_1163_axes_0 = const()[name = string("input_1163_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_21_norm_out_weight_to_fp16 = const()[name = string("model_layers_21_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(279923072)))];
+            tensor<fp16, [1024]> model_layers_21_norm_out_bias_to_fp16 = const()[name = string("model_layers_21_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(279925184)))];
+            tensor<fp16, [1, 126, 1024]> input_1163_cast_fp16 = layer_norm(axes = input_1163_axes_0, beta = model_layers_21_norm_out_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_21_norm_out_weight_to_fp16, x = input_1161_cast_fp16)[name = string("input_1163_cast_fp16")];
+            tensor<int32, [1]> input_1165_axes_0 = const()[name = string("input_1165_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_22_norm_feed_forward1_weight_to_fp16 = const()[name = string("model_layers_22_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(279927296)))];
+            tensor<fp16, [1024]> model_layers_22_norm_feed_forward1_bias_to_fp16 = const()[name = string("model_layers_22_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(279929408)))];
+            tensor<fp16, [1, 126, 1024]> input_1165_cast_fp16 = layer_norm(axes = input_1165_axes_0, beta = model_layers_22_norm_feed_forward1_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_22_norm_feed_forward1_weight_to_fp16, x = input_1163_cast_fp16)[name = string("input_1165_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_22_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(279931520))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(282028736))))[name = string("model_layers_22_feed_forward1_linear1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 4096]> linear_199_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_22_feed_forward1_linear1_weight_to_fp16_palettized, x = input_1165_cast_fp16)[name = string("linear_199_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_1169_cast_fp16 = silu(x = linear_199_cast_fp16)[name = string("input_1169_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_22_feed_forward1_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(282032896))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(284130112))))[name = string("model_layers_22_feed_forward1_linear2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_200_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_22_feed_forward1_linear2_weight_to_fp16_palettized, x = input_1169_cast_fp16)[name = string("linear_200_cast_fp16")];
+            fp16 var_3813_to_fp16 = const()[name = string("op_3813_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_3814_cast_fp16 = mul(x = linear_200_cast_fp16, y = var_3813_to_fp16)[name = string("op_3814_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_1175_cast_fp16 = add(x = input_1163_cast_fp16, y = var_3814_cast_fp16)[name = string("input_1175_cast_fp16")];
+            tensor<int32, [1]> query_45_axes_0 = const()[name = string("query_45_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_22_norm_self_att_weight_to_fp16 = const()[name = string("model_layers_22_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(284131200)))];
+            tensor<fp16, [1024]> model_layers_22_norm_self_att_bias_to_fp16 = const()[name = string("model_layers_22_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(284133312)))];
+            tensor<fp16, [1, 126, 1024]> query_45_cast_fp16 = layer_norm(axes = query_45_axes_0, beta = model_layers_22_norm_self_att_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_22_norm_self_att_weight_to_fp16, x = input_1175_cast_fp16)[name = string("query_45_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_22_self_attn_linear_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(284135424))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(284659776))))[name = string("model_layers_22_self_attn_linear_q_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_201_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_22_self_attn_linear_q_weight_to_fp16_palettized, x = query_45_cast_fp16)[name = string("linear_201_cast_fp16")];
+            tensor<int32, [4]> var_3830 = const()[name = string("op_3830"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> q_133_cast_fp16 = reshape(shape = var_3830, x = linear_201_cast_fp16)[name = string("q_133_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_22_self_attn_linear_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(284660864))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(285185216))))[name = string("model_layers_22_self_attn_linear_k_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_202_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_22_self_attn_linear_k_weight_to_fp16_palettized, x = query_45_cast_fp16)[name = string("linear_202_cast_fp16")];
+            tensor<int32, [4]> var_3834 = const()[name = string("op_3834"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> k_89_cast_fp16 = reshape(shape = var_3834, x = linear_202_cast_fp16)[name = string("k_89_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_22_self_attn_linear_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(285186304))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(285710656))))[name = string("model_layers_22_self_attn_linear_v_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_203_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_22_self_attn_linear_v_weight_to_fp16_palettized, x = query_45_cast_fp16)[name = string("linear_203_cast_fp16")];
+            tensor<int32, [4]> var_3838 = const()[name = string("op_3838"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> v_45_cast_fp16 = reshape(shape = var_3838, x = linear_203_cast_fp16)[name = string("v_45_cast_fp16")];
+            tensor<int32, [4]> value_45_perm_0 = const()[name = string("value_45_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> model_layers_22_self_attn_pos_bias_u_to_fp16 = const()[name = string("model_layers_22_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(285711744)))];
+            tensor<fp16, [1, 126, 8, 128]> var_3850_cast_fp16 = add(x = q_133_cast_fp16, y = model_layers_22_self_attn_pos_bias_u_to_fp16)[name = string("op_3850_cast_fp16")];
+            tensor<fp16, [8, 128]> model_layers_22_self_attn_pos_bias_v_to_fp16 = const()[name = string("model_layers_22_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(285713856)))];
+            tensor<fp16, [1, 126, 8, 128]> var_3852_cast_fp16 = add(x = q_133_cast_fp16, y = model_layers_22_self_attn_pos_bias_v_to_fp16)[name = string("op_3852_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_45_perm_0 = const()[name = string("q_with_bias_v_45_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_491_transpose_x_0 = const()[name = string("x_491_transpose_x_0"), val = bool(false)];
+            bool x_491_transpose_y_0 = const()[name = string("x_491_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 251]> var_3854_to_fp16 = const()[name = string("op_3854_to_fp16"), val = tensor<fp16, [1, 8, 128, 251]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(285715968)))];
+            tensor<fp16, [1, 8, 126, 128]> q_with_bias_v_45_cast_fp16 = transpose(perm = q_with_bias_v_45_perm_0, x = var_3852_cast_fp16)[name = string("transpose_158")];
+            tensor<fp16, [1, 8, 126, 251]> x_491_cast_fp16 = matmul(transpose_x = x_491_transpose_x_0, transpose_y = x_491_transpose_y_0, x = q_with_bias_v_45_cast_fp16, y = var_3854_to_fp16)[name = string("x_491_cast_fp16")];
+            tensor<int32, [8]> x_493_pad_0 = const()[name = string("x_493_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_493_mode_0 = const()[name = string("x_493_mode_0"), val = string("constant")];
+            fp16 const_234_to_fp16 = const()[name = string("const_234_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 252]> x_493_cast_fp16 = pad(constant_val = const_234_to_fp16, mode = x_493_mode_0, pad = x_493_pad_0, x = x_491_cast_fp16)[name = string("x_493_cast_fp16")];
+            tensor<int32, [4]> var_3862 = const()[name = string("op_3862"), val = tensor<int32, [4]>([1, 8, -1, 126])];
+            tensor<fp16, [1, 8, 252, 126]> x_495_cast_fp16 = reshape(shape = var_3862, x = x_493_cast_fp16)[name = string("x_495_cast_fp16")];
+            tensor<int32, [4]> var_3866_begin_0 = const()[name = string("op_3866_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_3866_end_0 = const()[name = string("op_3866_end_0"), val = tensor<int32, [4]>([1, 8, 252, 126])];
+            tensor<bool, [4]> var_3866_end_mask_0 = const()[name = string("op_3866_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 251, 126]> var_3866_cast_fp16 = slice_by_index(begin = var_3866_begin_0, end = var_3866_end_0, end_mask = var_3866_end_mask_0, x = x_495_cast_fp16)[name = string("op_3866_cast_fp16")];
+            tensor<int32, [4]> var_3867 = const()[name = string("op_3867"), val = tensor<int32, [4]>([1, 8, 126, 251])];
+            tensor<fp16, [1, 8, 126, 251]> matrix_bd_89_cast_fp16 = reshape(shape = var_3867, x = var_3866_cast_fp16)[name = string("matrix_bd_89_cast_fp16")];
+            bool matrix_ac_45_transpose_x_0 = const()[name = string("matrix_ac_45_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_45_transpose_y_0 = const()[name = string("matrix_ac_45_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_140_perm_0 = const()[name = string("transpose_140_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_141_perm_0 = const()[name = string("transpose_141_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 126]> transpose_141 = transpose(perm = transpose_141_perm_0, x = k_89_cast_fp16)[name = string("transpose_156")];
+            tensor<fp16, [1, 8, 126, 128]> transpose_140 = transpose(perm = transpose_140_perm_0, x = var_3850_cast_fp16)[name = string("transpose_157")];
+            tensor<fp16, [1, 8, 126, 126]> matrix_ac_45_cast_fp16 = matmul(transpose_x = matrix_ac_45_transpose_x_0, transpose_y = matrix_ac_45_transpose_y_0, x = transpose_140, y = transpose_141)[name = string("matrix_ac_45_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_91_begin_0 = const()[name = string("matrix_bd_91_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_91_end_0 = const()[name = string("matrix_bd_91_end_0"), val = tensor<int32, [4]>([1, 8, 126, 126])];
+            tensor<bool, [4]> matrix_bd_91_end_mask_0 = const()[name = string("matrix_bd_91_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 126, 126]> matrix_bd_91_cast_fp16 = slice_by_index(begin = matrix_bd_91_begin_0, end = matrix_bd_91_end_0, end_mask = matrix_bd_91_end_mask_0, x = matrix_bd_89_cast_fp16)[name = string("matrix_bd_91_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_3876_cast_fp16 = add(x = matrix_ac_45_cast_fp16, y = matrix_bd_91_cast_fp16)[name = string("op_3876_cast_fp16")];
+            fp16 _inversed_scores_89_y_0_to_fp16 = const()[name = string("_inversed_scores_89_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 126, 126]> _inversed_scores_89_cast_fp16 = mul(x = var_3876_cast_fp16, y = _inversed_scores_89_y_0_to_fp16)[name = string("_inversed_scores_89_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> scores_91_cast_fp16 = select(a = var_7_to_fp16, b = _inversed_scores_89_cast_fp16, cond = mask_3)[name = string("scores_91_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_3882_cast_fp16 = softmax(axis = var_25, x = scores_91_cast_fp16)[name = string("op_3882_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> input_1177_cast_fp16 = select(a = var_6_to_fp16, b = var_3882_cast_fp16, cond = mask_3)[name = string("input_1177_cast_fp16")];
+            bool x_497_transpose_x_0 = const()[name = string("x_497_transpose_x_0"), val = bool(false)];
+            bool x_497_transpose_y_0 = const()[name = string("x_497_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 126, 128]> value_45_cast_fp16 = transpose(perm = value_45_perm_0, x = v_45_cast_fp16)[name = string("transpose_155")];
+            tensor<fp16, [1, 8, 126, 128]> x_497_cast_fp16 = matmul(transpose_x = x_497_transpose_x_0, transpose_y = x_497_transpose_y_0, x = input_1177_cast_fp16, y = value_45_cast_fp16)[name = string("x_497_cast_fp16")];
+            tensor<int32, [4]> var_3886_perm_0 = const()[name = string("op_3886_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_3887 = const()[name = string("op_3887"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 126, 8, 128]> var_3886_cast_fp16 = transpose(perm = var_3886_perm_0, x = x_497_cast_fp16)[name = string("transpose_154")];
+            tensor<fp16, [1, 126, 1024]> input_1179_cast_fp16 = reshape(shape = var_3887, x = var_3886_cast_fp16)[name = string("input_1179_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_22_self_attn_linear_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(286230080))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(286754432))))[name = string("model_layers_22_self_attn_linear_out_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_205_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_22_self_attn_linear_out_weight_to_fp16_palettized, x = input_1179_cast_fp16)[name = string("linear_205_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_1183_cast_fp16 = add(x = input_1175_cast_fp16, y = linear_205_cast_fp16)[name = string("input_1183_cast_fp16")];
+            tensor<int32, [1]> x_501_axes_0 = const()[name = string("x_501_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_22_norm_conv_weight_to_fp16 = const()[name = string("model_layers_22_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(286755520)))];
+            tensor<fp16, [1024]> model_layers_22_norm_conv_bias_to_fp16 = const()[name = string("model_layers_22_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(286757632)))];
+            tensor<fp16, [1, 126, 1024]> x_501_cast_fp16 = layer_norm(axes = x_501_axes_0, beta = model_layers_22_norm_conv_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_22_norm_conv_weight_to_fp16, x = input_1183_cast_fp16)[name = string("x_501_cast_fp16")];
+            tensor<int32, [3]> input_1185_perm_0 = const()[name = string("input_1185_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_1187_pad_type_0 = const()[name = string("input_1187_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_1187_strides_0 = const()[name = string("input_1187_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_1187_pad_0 = const()[name = string("input_1187_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_1187_dilations_0 = const()[name = string("input_1187_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_1187_groups_0 = const()[name = string("input_1187_groups_0"), val = int32(1)];
+            tensor<fp16, [2048, 1024, 1]> model_layers_22_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(286759744))), lut = tensor<fp16, [64, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287808384))))[name = string("model_layers_22_conv_pointwise_conv1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 1024, 126]> input_1185_cast_fp16 = transpose(perm = input_1185_perm_0, x = x_501_cast_fp16)[name = string("transpose_153")];
+            tensor<fp16, [1, 2048, 126]> input_1187_cast_fp16 = conv(dilations = input_1187_dilations_0, groups = input_1187_groups_0, pad = input_1187_pad_0, pad_type = input_1187_pad_type_0, strides = input_1187_strides_0, weight = model_layers_22_conv_pointwise_conv1_weight_to_fp16_palettized, x = input_1185_cast_fp16)[name = string("input_1187_cast_fp16")];
+            int32 x_503_split_num_splits_0 = const()[name = string("x_503_split_num_splits_0"), val = int32(2)];
+            int32 x_503_split_axis_0 = const()[name = string("x_503_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 126]> x_503_split_cast_fp16_0, tensor<fp16, [1, 1024, 126]> x_503_split_cast_fp16_1 = split(axis = x_503_split_axis_0, num_splits = x_503_split_num_splits_0, x = input_1187_cast_fp16)[name = string("x_503_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_503_split_1_sigmoid_cast_fp16 = sigmoid(x = x_503_split_cast_fp16_1)[name = string("x_503_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_503_cast_fp16 = mul(x = x_503_split_cast_fp16_0, y = x_503_split_1_sigmoid_cast_fp16)[name = string("x_503_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_1189_cast_fp16 = select(a = var_6_to_fp16, b = x_503_cast_fp16, cond = var_323)[name = string("input_1189_cast_fp16")];
+            tensor<int32, [6]> input_1191_pad_0 = const()[name = string("input_1191_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_1191_mode_0 = const()[name = string("input_1191_mode_0"), val = string("constant")];
+            fp16 const_237_to_fp16 = const()[name = string("const_237_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 134]> input_1191_cast_fp16 = pad(constant_val = const_237_to_fp16, mode = input_1191_mode_0, pad = input_1191_pad_0, x = input_1189_cast_fp16)[name = string("input_1191_cast_fp16")];
+            string input_1193_pad_type_0 = const()[name = string("input_1193_pad_type_0"), val = string("valid")];
+            int32 input_1193_groups_0 = const()[name = string("input_1193_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_1193_strides_0 = const()[name = string("input_1193_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_1193_pad_0 = const()[name = string("input_1193_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_1193_dilations_0 = const()[name = string("input_1193_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_292_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287810496))), lut = tensor<fp16, [32, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287815168))))[name = string("const_292_to_fp16_palettized")];
+            tensor<fp16, [1024]> const_293_to_fp16 = const()[name = string("const_293_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287816256)))];
+            tensor<fp16, [1, 1024, 126]> input_1195_cast_fp16 = conv(bias = const_293_to_fp16, dilations = input_1193_dilations_0, groups = input_1193_groups_0, pad = input_1193_pad_0, pad_type = input_1193_pad_type_0, strides = input_1193_strides_0, weight = const_292_to_fp16_palettized, x = input_1191_cast_fp16)[name = string("input_1195_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_1197_cast_fp16 = silu(x = input_1195_cast_fp16)[name = string("input_1197_cast_fp16")];
+            string x_505_pad_type_0 = const()[name = string("x_505_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_505_strides_0 = const()[name = string("x_505_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_505_pad_0 = const()[name = string("x_505_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_505_dilations_0 = const()[name = string("x_505_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_505_groups_0 = const()[name = string("x_505_groups_0"), val = int32(1)];
+            tensor<fp16, [1024, 1024, 1]> model_layers_22_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287818368))), lut = tensor<fp16, [32, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288342720))))[name = string("model_layers_22_conv_pointwise_conv2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 1024, 126]> x_505_cast_fp16 = conv(dilations = x_505_dilations_0, groups = x_505_groups_0, pad = x_505_pad_0, pad_type = x_505_pad_type_0, strides = x_505_strides_0, weight = model_layers_22_conv_pointwise_conv2_weight_to_fp16_palettized, x = input_1197_cast_fp16)[name = string("x_505_cast_fp16")];
+            tensor<int32, [3]> input_1199_perm_0 = const()[name = string("input_1199_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 126, 1024]> input_1199_cast_fp16 = transpose(perm = input_1199_perm_0, x = x_505_cast_fp16)[name = string("transpose_152")];
+            tensor<fp16, [1, 126, 1024]> input_1201_cast_fp16 = add(x = input_1183_cast_fp16, y = input_1199_cast_fp16)[name = string("input_1201_cast_fp16")];
+            tensor<int32, [1]> input_1203_axes_0 = const()[name = string("input_1203_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_22_norm_feed_forward2_weight_to_fp16 = const()[name = string("model_layers_22_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288343808)))];
+            tensor<fp16, [1024]> model_layers_22_norm_feed_forward2_bias_to_fp16 = const()[name = string("model_layers_22_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288345920)))];
+            tensor<fp16, [1, 126, 1024]> input_1203_cast_fp16 = layer_norm(axes = input_1203_axes_0, beta = model_layers_22_norm_feed_forward2_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_22_norm_feed_forward2_weight_to_fp16, x = input_1201_cast_fp16)[name = string("input_1203_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_22_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288348032))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290445248))))[name = string("model_layers_22_feed_forward2_linear1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 4096]> linear_206_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_22_feed_forward2_linear1_weight_to_fp16_palettized, x = input_1203_cast_fp16)[name = string("linear_206_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_1207_cast_fp16 = silu(x = linear_206_cast_fp16)[name = string("input_1207_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_22_feed_forward2_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290449408))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(292546624))))[name = string("model_layers_22_feed_forward2_linear2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_207_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_22_feed_forward2_linear2_weight_to_fp16_palettized, x = input_1207_cast_fp16)[name = string("linear_207_cast_fp16")];
+            fp16 var_3947_to_fp16 = const()[name = string("op_3947_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_3948_cast_fp16 = mul(x = linear_207_cast_fp16, y = var_3947_to_fp16)[name = string("op_3948_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_1213_cast_fp16 = add(x = input_1201_cast_fp16, y = var_3948_cast_fp16)[name = string("input_1213_cast_fp16")];
+            tensor<int32, [1]> input_1215_axes_0 = const()[name = string("input_1215_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_22_norm_out_weight_to_fp16 = const()[name = string("model_layers_22_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(292547712)))];
+            tensor<fp16, [1024]> model_layers_22_norm_out_bias_to_fp16 = const()[name = string("model_layers_22_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(292549824)))];
+            tensor<fp16, [1, 126, 1024]> input_1215_cast_fp16 = layer_norm(axes = input_1215_axes_0, beta = model_layers_22_norm_out_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_22_norm_out_weight_to_fp16, x = input_1213_cast_fp16)[name = string("input_1215_cast_fp16")];
+            tensor<int32, [1]> input_1217_axes_0 = const()[name = string("input_1217_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_23_norm_feed_forward1_weight_to_fp16 = const()[name = string("model_layers_23_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(292551936)))];
+            tensor<fp16, [1024]> model_layers_23_norm_feed_forward1_bias_to_fp16 = const()[name = string("model_layers_23_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(292554048)))];
+            tensor<fp16, [1, 126, 1024]> input_1217_cast_fp16 = layer_norm(axes = input_1217_axes_0, beta = model_layers_23_norm_feed_forward1_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_23_norm_feed_forward1_weight_to_fp16, x = input_1215_cast_fp16)[name = string("input_1217_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_23_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(292556160))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(294653376))))[name = string("model_layers_23_feed_forward1_linear1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 4096]> linear_208_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_23_feed_forward1_linear1_weight_to_fp16_palettized, x = input_1217_cast_fp16)[name = string("linear_208_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_1221_cast_fp16 = silu(x = linear_208_cast_fp16)[name = string("input_1221_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_23_feed_forward1_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(294657536))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(296754752))))[name = string("model_layers_23_feed_forward1_linear2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_209_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_23_feed_forward1_linear2_weight_to_fp16_palettized, x = input_1221_cast_fp16)[name = string("linear_209_cast_fp16")];
+            fp16 var_3976_to_fp16 = const()[name = string("op_3976_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_3977_cast_fp16 = mul(x = linear_209_cast_fp16, y = var_3976_to_fp16)[name = string("op_3977_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_1227_cast_fp16 = add(x = input_1215_cast_fp16, y = var_3977_cast_fp16)[name = string("input_1227_cast_fp16")];
+            tensor<int32, [1]> query_axes_0 = const()[name = string("query_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_23_norm_self_att_weight_to_fp16 = const()[name = string("model_layers_23_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(296755840)))];
+            tensor<fp16, [1024]> model_layers_23_norm_self_att_bias_to_fp16 = const()[name = string("model_layers_23_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(296757952)))];
+            tensor<fp16, [1, 126, 1024]> query_cast_fp16 = layer_norm(axes = query_axes_0, beta = model_layers_23_norm_self_att_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_23_norm_self_att_weight_to_fp16, x = input_1227_cast_fp16)[name = string("query_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_23_self_attn_linear_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(296760064))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(297284416))))[name = string("model_layers_23_self_attn_linear_q_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_210_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_23_self_attn_linear_q_weight_to_fp16_palettized, x = query_cast_fp16)[name = string("linear_210_cast_fp16")];
+            tensor<int32, [4]> var_3993 = const()[name = string("op_3993"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> q_139_cast_fp16 = reshape(shape = var_3993, x = linear_210_cast_fp16)[name = string("q_139_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_23_self_attn_linear_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(297285504))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(297809856))))[name = string("model_layers_23_self_attn_linear_k_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_211_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_23_self_attn_linear_k_weight_to_fp16_palettized, x = query_cast_fp16)[name = string("linear_211_cast_fp16")];
+            tensor<int32, [4]> var_3997 = const()[name = string("op_3997"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> k_93_cast_fp16 = reshape(shape = var_3997, x = linear_211_cast_fp16)[name = string("k_93_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_23_self_attn_linear_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(297810944))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(298335296))))[name = string("model_layers_23_self_attn_linear_v_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_212_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_23_self_attn_linear_v_weight_to_fp16_palettized, x = query_cast_fp16)[name = string("linear_212_cast_fp16")];
+            tensor<int32, [4]> var_4001 = const()[name = string("op_4001"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> v_cast_fp16 = reshape(shape = var_4001, x = linear_212_cast_fp16)[name = string("v_cast_fp16")];
+            tensor<int32, [4]> value_perm_0 = const()[name = string("value_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<fp16, [8, 128]> model_layers_23_self_attn_pos_bias_u_to_fp16 = const()[name = string("model_layers_23_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(298336384)))];
+            tensor<fp16, [1, 126, 8, 128]> var_4013_cast_fp16 = add(x = q_139_cast_fp16, y = model_layers_23_self_attn_pos_bias_u_to_fp16)[name = string("op_4013_cast_fp16")];
+            tensor<fp16, [8, 128]> model_layers_23_self_attn_pos_bias_v_to_fp16 = const()[name = string("model_layers_23_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(298338496)))];
+            tensor<fp16, [1, 126, 8, 128]> var_4015_cast_fp16 = add(x = q_139_cast_fp16, y = model_layers_23_self_attn_pos_bias_v_to_fp16)[name = string("op_4015_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_perm_0 = const()[name = string("q_with_bias_v_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            bool x_513_transpose_x_0 = const()[name = string("x_513_transpose_x_0"), val = bool(false)];
+            bool x_513_transpose_y_0 = const()[name = string("x_513_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 128, 251]> var_4017_to_fp16 = const()[name = string("op_4017_to_fp16"), val = tensor<fp16, [1, 8, 128, 251]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(298340608)))];
+            tensor<fp16, [1, 8, 126, 128]> q_with_bias_v_cast_fp16 = transpose(perm = q_with_bias_v_perm_0, x = var_4015_cast_fp16)[name = string("transpose_151")];
+            tensor<fp16, [1, 8, 126, 251]> x_513_cast_fp16 = matmul(transpose_x = x_513_transpose_x_0, transpose_y = x_513_transpose_y_0, x = q_with_bias_v_cast_fp16, y = var_4017_to_fp16)[name = string("x_513_cast_fp16")];
+            tensor<int32, [8]> x_515_pad_0 = const()[name = string("x_515_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            string x_515_mode_0 = const()[name = string("x_515_mode_0"), val = string("constant")];
+            fp16 const_244_to_fp16 = const()[name = string("const_244_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 252]> x_515_cast_fp16 = pad(constant_val = const_244_to_fp16, mode = x_515_mode_0, pad = x_515_pad_0, x = x_513_cast_fp16)[name = string("x_515_cast_fp16")];
+            tensor<int32, [4]> var_4025 = const()[name = string("op_4025"), val = tensor<int32, [4]>([1, 8, -1, 126])];
+            tensor<fp16, [1, 8, 252, 126]> x_517_cast_fp16 = reshape(shape = var_4025, x = x_515_cast_fp16)[name = string("x_517_cast_fp16")];
+            tensor<int32, [4]> var_4029_begin_0 = const()[name = string("op_4029_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_4029_end_0 = const()[name = string("op_4029_end_0"), val = tensor<int32, [4]>([1, 8, 252, 126])];
+            tensor<bool, [4]> var_4029_end_mask_0 = const()[name = string("op_4029_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 251, 126]> var_4029_cast_fp16 = slice_by_index(begin = var_4029_begin_0, end = var_4029_end_0, end_mask = var_4029_end_mask_0, x = x_517_cast_fp16)[name = string("op_4029_cast_fp16")];
+            tensor<int32, [4]> var_4030 = const()[name = string("op_4030"), val = tensor<int32, [4]>([1, 8, 126, 251])];
+            tensor<fp16, [1, 8, 126, 251]> matrix_bd_93_cast_fp16 = reshape(shape = var_4030, x = var_4029_cast_fp16)[name = string("matrix_bd_93_cast_fp16")];
+            bool matrix_ac_transpose_x_0 = const()[name = string("matrix_ac_transpose_x_0"), val = bool(false)];
+            bool matrix_ac_transpose_y_0 = const()[name = string("matrix_ac_transpose_y_0"), val = bool(false)];
+            tensor<int32, [4]> transpose_142_perm_0 = const()[name = string("transpose_142_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_143_perm_0 = const()[name = string("transpose_143_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 126]> transpose_143 = transpose(perm = transpose_143_perm_0, x = k_93_cast_fp16)[name = string("transpose_149")];
+            tensor<fp16, [1, 8, 126, 128]> transpose_142 = transpose(perm = transpose_142_perm_0, x = var_4013_cast_fp16)[name = string("transpose_150")];
+            tensor<fp16, [1, 8, 126, 126]> matrix_ac_cast_fp16 = matmul(transpose_x = matrix_ac_transpose_x_0, transpose_y = matrix_ac_transpose_y_0, x = transpose_142, y = transpose_143)[name = string("matrix_ac_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_begin_0 = const()[name = string("matrix_bd_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_end_0 = const()[name = string("matrix_bd_end_0"), val = tensor<int32, [4]>([1, 8, 126, 126])];
+            tensor<bool, [4]> matrix_bd_end_mask_0 = const()[name = string("matrix_bd_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 126, 126]> matrix_bd_cast_fp16 = slice_by_index(begin = matrix_bd_begin_0, end = matrix_bd_end_0, end_mask = matrix_bd_end_mask_0, x = matrix_bd_93_cast_fp16)[name = string("matrix_bd_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_4039_cast_fp16 = add(x = matrix_ac_cast_fp16, y = matrix_bd_cast_fp16)[name = string("op_4039_cast_fp16")];
+            fp16 _inversed_scores_93_y_0_to_fp16 = const()[name = string("_inversed_scores_93_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 126, 126]> _inversed_scores_93_cast_fp16 = mul(x = var_4039_cast_fp16, y = _inversed_scores_93_y_0_to_fp16)[name = string("_inversed_scores_93_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> scores_cast_fp16 = select(a = var_7_to_fp16, b = _inversed_scores_93_cast_fp16, cond = mask_3)[name = string("scores_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_4045_cast_fp16 = softmax(axis = var_25, x = scores_cast_fp16)[name = string("op_4045_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> input_1229_cast_fp16 = select(a = var_6_to_fp16, b = var_4045_cast_fp16, cond = mask_3)[name = string("input_1229_cast_fp16")];
+            bool x_519_transpose_x_0 = const()[name = string("x_519_transpose_x_0"), val = bool(false)];
+            bool x_519_transpose_y_0 = const()[name = string("x_519_transpose_y_0"), val = bool(false)];
+            tensor<fp16, [1, 8, 126, 128]> value_cast_fp16 = transpose(perm = value_perm_0, x = v_cast_fp16)[name = string("transpose_148")];
+            tensor<fp16, [1, 8, 126, 128]> x_519_cast_fp16 = matmul(transpose_x = x_519_transpose_x_0, transpose_y = x_519_transpose_y_0, x = input_1229_cast_fp16, y = value_cast_fp16)[name = string("x_519_cast_fp16")];
+            tensor<int32, [4]> var_4049_perm_0 = const()[name = string("op_4049_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_4050 = const()[name = string("op_4050"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 126, 8, 128]> var_4049_cast_fp16 = transpose(perm = var_4049_perm_0, x = x_519_cast_fp16)[name = string("transpose_147")];
+            tensor<fp16, [1, 126, 1024]> input_1231_cast_fp16 = reshape(shape = var_4050, x = var_4049_cast_fp16)[name = string("input_1231_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_23_self_attn_linear_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(298854720))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299379072))))[name = string("model_layers_23_self_attn_linear_out_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_214_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_23_self_attn_linear_out_weight_to_fp16_palettized, x = input_1231_cast_fp16)[name = string("linear_214_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_1235_cast_fp16 = add(x = input_1227_cast_fp16, y = linear_214_cast_fp16)[name = string("input_1235_cast_fp16")];
+            tensor<int32, [1]> x_523_axes_0 = const()[name = string("x_523_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_23_norm_conv_weight_to_fp16 = const()[name = string("model_layers_23_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299380160)))];
+            tensor<fp16, [1024]> model_layers_23_norm_conv_bias_to_fp16 = const()[name = string("model_layers_23_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299382272)))];
+            tensor<fp16, [1, 126, 1024]> x_523_cast_fp16 = layer_norm(axes = x_523_axes_0, beta = model_layers_23_norm_conv_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_23_norm_conv_weight_to_fp16, x = input_1235_cast_fp16)[name = string("x_523_cast_fp16")];
+            tensor<int32, [3]> input_1237_perm_0 = const()[name = string("input_1237_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            string input_1239_pad_type_0 = const()[name = string("input_1239_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> input_1239_strides_0 = const()[name = string("input_1239_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_1239_pad_0 = const()[name = string("input_1239_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_1239_dilations_0 = const()[name = string("input_1239_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 input_1239_groups_0 = const()[name = string("input_1239_groups_0"), val = int32(1)];
+            tensor<fp16, [2048, 1024, 1]> model_layers_23_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299384384))), lut = tensor<fp16, [64, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300433024))))[name = string("model_layers_23_conv_pointwise_conv1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 1024, 126]> input_1237_cast_fp16 = transpose(perm = input_1237_perm_0, x = x_523_cast_fp16)[name = string("transpose_146")];
+            tensor<fp16, [1, 2048, 126]> input_1239_cast_fp16 = conv(dilations = input_1239_dilations_0, groups = input_1239_groups_0, pad = input_1239_pad_0, pad_type = input_1239_pad_type_0, strides = input_1239_strides_0, weight = model_layers_23_conv_pointwise_conv1_weight_to_fp16_palettized, x = input_1237_cast_fp16)[name = string("input_1239_cast_fp16")];
+            int32 x_525_split_num_splits_0 = const()[name = string("x_525_split_num_splits_0"), val = int32(2)];
+            int32 x_525_split_axis_0 = const()[name = string("x_525_split_axis_0"), val = int32(1)];
+            tensor<fp16, [1, 1024, 126]> x_525_split_cast_fp16_0, tensor<fp16, [1, 1024, 126]> x_525_split_cast_fp16_1 = split(axis = x_525_split_axis_0, num_splits = x_525_split_num_splits_0, x = input_1239_cast_fp16)[name = string("x_525_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_525_split_1_sigmoid_cast_fp16 = sigmoid(x = x_525_split_cast_fp16_1)[name = string("x_525_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_525_cast_fp16 = mul(x = x_525_split_cast_fp16_0, y = x_525_split_1_sigmoid_cast_fp16)[name = string("x_525_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_1241_cast_fp16 = select(a = var_6_to_fp16, b = x_525_cast_fp16, cond = var_323)[name = string("input_1241_cast_fp16")];
+            tensor<int32, [6]> input_1243_pad_0 = const()[name = string("input_1243_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            string input_1243_mode_0 = const()[name = string("input_1243_mode_0"), val = string("constant")];
+            fp16 const_247_to_fp16 = const()[name = string("const_247_to_fp16"), val = fp16(0x0p+0)];
+            tensor<fp16, [1, 1024, 134]> input_1243_cast_fp16 = pad(constant_val = const_247_to_fp16, mode = input_1243_mode_0, pad = input_1243_pad_0, x = input_1241_cast_fp16)[name = string("input_1243_cast_fp16")];
+            string input_1245_pad_type_0 = const()[name = string("input_1245_pad_type_0"), val = string("valid")];
+            int32 input_1245_groups_0 = const()[name = string("input_1245_groups_0"), val = int32(1024)];
+            tensor<int32, [1]> input_1245_strides_0 = const()[name = string("input_1245_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> input_1245_pad_0 = const()[name = string("input_1245_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> input_1245_dilations_0 = const()[name = string("input_1245_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1024, 1, 9]> const_294_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300435136))), lut = tensor<fp16, [32, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300439808))))[name = string("const_294_to_fp16_palettized")];
+            tensor<fp16, [1024]> const_295_to_fp16 = const()[name = string("const_295_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300440896)))];
+            tensor<fp16, [1, 1024, 126]> input_1247_cast_fp16 = conv(bias = const_295_to_fp16, dilations = input_1245_dilations_0, groups = input_1245_groups_0, pad = input_1245_pad_0, pad_type = input_1245_pad_type_0, strides = input_1245_strides_0, weight = const_294_to_fp16_palettized, x = input_1243_cast_fp16)[name = string("input_1247_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_1249_cast_fp16 = silu(x = input_1247_cast_fp16)[name = string("input_1249_cast_fp16")];
+            string x_527_pad_type_0 = const()[name = string("x_527_pad_type_0"), val = string("valid")];
+            tensor<int32, [1]> x_527_strides_0 = const()[name = string("x_527_strides_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [2]> x_527_pad_0 = const()[name = string("x_527_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> x_527_dilations_0 = const()[name = string("x_527_dilations_0"), val = tensor<int32, [1]>([1])];
+            int32 x_527_groups_0 = const()[name = string("x_527_groups_0"), val = int32(1)];
+            tensor<fp16, [1024, 1024, 1]> model_layers_23_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300443008))), lut = tensor<fp16, [32, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300967360))))[name = string("model_layers_23_conv_pointwise_conv2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 1024, 126]> x_527_cast_fp16 = conv(dilations = x_527_dilations_0, groups = x_527_groups_0, pad = x_527_pad_0, pad_type = x_527_pad_type_0, strides = x_527_strides_0, weight = model_layers_23_conv_pointwise_conv2_weight_to_fp16_palettized, x = input_1249_cast_fp16)[name = string("x_527_cast_fp16")];
+            tensor<int32, [3]> input_1251_perm_0 = const()[name = string("input_1251_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 126, 1024]> input_1251_cast_fp16 = transpose(perm = input_1251_perm_0, x = x_527_cast_fp16)[name = string("transpose_145")];
+            tensor<fp16, [1, 126, 1024]> input_1253_cast_fp16 = add(x = input_1235_cast_fp16, y = input_1251_cast_fp16)[name = string("input_1253_cast_fp16")];
+            tensor<int32, [1]> input_1255_axes_0 = const()[name = string("input_1255_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_23_norm_feed_forward2_weight_to_fp16 = const()[name = string("model_layers_23_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300968448)))];
+            tensor<fp16, [1024]> model_layers_23_norm_feed_forward2_bias_to_fp16 = const()[name = string("model_layers_23_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300970560)))];
+            tensor<fp16, [1, 126, 1024]> input_1255_cast_fp16 = layer_norm(axes = input_1255_axes_0, beta = model_layers_23_norm_feed_forward2_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_23_norm_feed_forward2_weight_to_fp16, x = input_1253_cast_fp16)[name = string("input_1255_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_23_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300972672))), lut = tensor<fp16, [128, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303069888))))[name = string("model_layers_23_feed_forward2_linear1_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 4096]> linear_215_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_23_feed_forward2_linear1_weight_to_fp16_palettized, x = input_1255_cast_fp16)[name = string("linear_215_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_1259_cast_fp16 = silu(x = linear_215_cast_fp16)[name = string("input_1259_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_23_feed_forward2_linear2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303074048))), lut = tensor<fp16, [32, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(305171264))))[name = string("model_layers_23_feed_forward2_linear2_weight_to_fp16_palettized")];
+            tensor<fp16, [1, 126, 1024]> linear_216_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_23_feed_forward2_linear2_weight_to_fp16_palettized, x = input_1259_cast_fp16)[name = string("linear_216_cast_fp16")];
+            fp16 var_4110_to_fp16 = const()[name = string("op_4110_to_fp16"), val = fp16(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_4111_cast_fp16 = mul(x = linear_216_cast_fp16, y = var_4110_to_fp16)[name = string("op_4111_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_cast_fp16 = add(x = input_1253_cast_fp16, y = var_4111_cast_fp16)[name = string("input_cast_fp16")];
+            tensor<int32, [1]> audio_signal_axes_0 = const()[name = string("audio_signal_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_23_norm_out_weight_to_fp16 = const()[name = string("model_layers_23_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(305172352)))];
+            tensor<fp16, [1024]> model_layers_23_norm_out_bias_to_fp16 = const()[name = string("model_layers_23_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(305174464)))];
+            tensor<fp16, [1, 126, 1024]> audio_signal_cast_fp16 = layer_norm(axes = audio_signal_axes_0, beta = model_layers_23_norm_out_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_23_norm_out_weight_to_fp16, x = input_cast_fp16)[name = string("audio_signal_cast_fp16")];
+            tensor<int32, [3]> obj_1_perm_0 = const()[name = string("obj_1_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 1024, 126]> encoder_output = transpose(perm = obj_1_perm_0, x = audio_signal_cast_fp16)[name = string("transpose_144")];
+        } -> (encoder_output, encoder_output_length);
+}
\ No newline at end of file