Hugging Face's logo

RedHatAI
/
quantization

kernel
Model card Files
xet
Community
2
danieldk HF Staff commited on
Commit
d5fd23d
·
1 Parent(s): 8aa00a3

Temporarily disable ROCm

Browse files
Files changed (1) hide show
  1. build.toml +48 -48
build.toml CHANGED
@@ -46,35 +46,35 @@ src = [
46
  "gptq_marlin/marlin_template.h",
47
  ]
48
 
49
- [kernel.fp8_common_rocm]
50
- backend = "rocm"
51
- depends = ["torch"]
52
- rocm-archs = [
53
- "gfx906",
54
- "gfx908",
55
- "gfx90a",
56
- "gfx940",
57
- "gfx941",
58
- "gfx942",
59
- "gfx1030",
60
- "gfx1100",
61
- "gfx1101",
62
- ]
63
- include = ["."]
64
- src = [
65
- "attention/attention_dtypes.h",
66
- "attention/attention_generic.cuh",
67
- "attention/dtype_bfloat16.cuh",
68
- "attention/dtype_float16.cuh",
69
- "attention/dtype_float32.cuh",
70
- "attention/dtype_fp8.cuh",
71
- "fp8/amd/quant_utils.cuh",
72
- "fp8/common.cu",
73
- "fp8/common.cuh",
74
- "dispatch_utils.h",
75
- "utils.cuh",
76
- "vectorization.cuh",
77
- ]
78
 
79
  [kernel.int8_common]
80
  backend = "cuda"
@@ -240,22 +240,22 @@ src = [
240
  "marlin/sparse/marlin_24_cuda_kernel.cu",
241
  ]
242
 
243
- [kernel.int8_common_rocm]
244
- backend = "rocm"
245
- depends = ["torch"]
246
- rocm-archs = [
247
- "gfx906",
248
- "gfx908",
249
- "gfx90a",
250
- "gfx940",
251
- "gfx941",
252
- "gfx942",
253
- "gfx1030",
254
- "gfx1100",
255
- "gfx1101",
256
- ]
257
- include = ["."]
258
- src = [
259
- "compressed_tensors/int8_quant_kernels.cu",
260
- "dispatch_utils.h",
261
- ]
 
46
  "gptq_marlin/marlin_template.h",
47
  ]
48
 
49
+ #[kernel.fp8_common_rocm]
50
+ #backend = "rocm"
51
+ #depends = ["torch"]
52
+ #rocm-archs = [
53
+ # "gfx906",
54
+ # "gfx908",
55
+ # "gfx90a",
56
+ # "gfx940",
57
+ # "gfx941",
58
+ # "gfx942",
59
+ # "gfx1030",
60
+ # "gfx1100",
61
+ # "gfx1101",
62
+ #]
63
+ #include = ["."]
64
+ #src = [
65
+ # "attention/attention_dtypes.h",
66
+ # "attention/attention_generic.cuh",
67
+ # "attention/dtype_bfloat16.cuh",
68
+ # "attention/dtype_float16.cuh",
69
+ # "attention/dtype_float32.cuh",
70
+ # "attention/dtype_fp8.cuh",
71
+ # "fp8/amd/quant_utils.cuh",
72
+ # "fp8/common.cu",
73
+ # "fp8/common.cuh",
74
+ # "dispatch_utils.h",
75
+ # "utils.cuh",
76
+ # "vectorization.cuh",
77
+ #]
78
 
79
  [kernel.int8_common]
80
  backend = "cuda"
 
240
  "marlin/sparse/marlin_24_cuda_kernel.cu",
241
  ]
242
 
243
+ #[kernel.int8_common_rocm]
244
+ #backend = "rocm"
245
+ #depends = ["torch"]
246
+ #rocm-archs = [
247
+ # "gfx906",
248
+ # "gfx908",
249
+ # "gfx90a",
250
+ # "gfx940",
251
+ # "gfx941",
252
+ # "gfx942",
253
+ # "gfx1030",
254
+ # "gfx1100",
255
+ # "gfx1101",
256
+ #]
257
+ #include = ["."]
258
+ #src = [
259
+ # "compressed_tensors/int8_quant_kernels.cu",
260
+ # "dispatch_utils.h",
261
+ #]