uday610 commited on
Commit
91f9c0e
·
verified ·
1 Parent(s): 5693eb5

Update genai_config.json

Browse files
Files changed (1) hide show
  1. genai_config.json +59 -59
genai_config.json CHANGED
@@ -1,60 +1,60 @@
1
- {
2
- "model": {
3
- "bos_token_id": 151643,
4
- "context_length": 32768,
5
- "decoder": {
6
- "session_options": {
7
- "log_id": "onnxruntime-genai",
8
- "custom_ops_library": "PATH\\onnx_custom_ops.dll",
9
- "external_data_file": "Qwen2.5-7B-Instruct_jit.pb.bin",
10
- "custom_allocator": "shared_d3d_xrt",
11
- "hybrid_opt_gpu_jit": "1",
12
- "hybrid_opt_free_after_prefill": "1",
13
- "hybrid_opt_enable_dynamic_dpm": "1",
14
- "rmm_memory_limit_mb": 5427,
15
- "provider_options": []
16
- },
17
- "filename": "Qwen2.5-7B-Instruct_jit.onnx",
18
- "head_size": 128,
19
- "hidden_size": 3584,
20
- "inputs": {
21
- "input_ids": "input_ids",
22
- "attention_mask": "attention_mask",
23
- "position_ids": "position_ids",
24
- "past_key_names": "past_key_values.%d.key",
25
- "past_value_names": "past_key_values.%d.value"
26
- },
27
- "outputs": {
28
- "logits": "logits",
29
- "present_key_names": "present.%d.key",
30
- "present_value_names": "present.%d.value"
31
- },
32
- "num_attention_heads": 28,
33
- "num_hidden_layers": 28,
34
- "num_key_value_heads": 4
35
- },
36
- "eos_token_id": [
37
- 151645,
38
- 151643
39
- ],
40
- "pad_token_id": 151643,
41
- "type": "qwen2",
42
- "vocab_size": 152064
43
- },
44
- "search": {
45
- "diversity_penalty": 0.0,
46
- "do_sample": true,
47
- "early_stopping": true,
48
- "length_penalty": 1.0,
49
- "max_length": 32768,
50
- "min_length": 0,
51
- "no_repeat_ngram_size": 0,
52
- "num_beams": 1,
53
- "num_return_sequences": 1,
54
- "past_present_share_buffer": true,
55
- "repetition_penalty": 1.05,
56
- "temperature": 0.7,
57
- "top_k": 1,
58
- "top_p": 0.8
59
- }
60
  }
 
1
+ {
2
+ "model": {
3
+ "bos_token_id": 151643,
4
+ "context_length": 32768,
5
+ "decoder": {
6
+ "session_options": {
7
+ "log_id": "onnxruntime-genai",
8
+ "custom_ops_library": "onnx_custom_ops.dll",
9
+ "external_data_file": "Qwen2.5-7B-Instruct_jit.pb.bin",
10
+ "custom_allocator": "shared_d3d_xrt",
11
+ "hybrid_opt_gpu_jit": "1",
12
+ "hybrid_opt_free_after_prefill": "1",
13
+ "hybrid_opt_enable_dynamic_dpm": "1",
14
+ "rmm_memory_limit_mb": 5427,
15
+ "provider_options": []
16
+ },
17
+ "filename": "Qwen2.5-7B-Instruct_jit.onnx",
18
+ "head_size": 128,
19
+ "hidden_size": 3584,
20
+ "inputs": {
21
+ "input_ids": "input_ids",
22
+ "attention_mask": "attention_mask",
23
+ "position_ids": "position_ids",
24
+ "past_key_names": "past_key_values.%d.key",
25
+ "past_value_names": "past_key_values.%d.value"
26
+ },
27
+ "outputs": {
28
+ "logits": "logits",
29
+ "present_key_names": "present.%d.key",
30
+ "present_value_names": "present.%d.value"
31
+ },
32
+ "num_attention_heads": 28,
33
+ "num_hidden_layers": 28,
34
+ "num_key_value_heads": 4
35
+ },
36
+ "eos_token_id": [
37
+ 151645,
38
+ 151643
39
+ ],
40
+ "pad_token_id": 151643,
41
+ "type": "qwen2",
42
+ "vocab_size": 152064
43
+ },
44
+ "search": {
45
+ "diversity_penalty": 0.0,
46
+ "do_sample": true,
47
+ "early_stopping": true,
48
+ "length_penalty": 1.0,
49
+ "max_length": 32768,
50
+ "min_length": 0,
51
+ "no_repeat_ngram_size": 0,
52
+ "num_beams": 1,
53
+ "num_return_sequences": 1,
54
+ "past_present_share_buffer": true,
55
+ "repetition_penalty": 1.05,
56
+ "temperature": 0.7,
57
+ "top_k": 1,
58
+ "top_p": 0.8
59
+ }
60
  }