ceos/unsloth
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

[intel] skip xpu fbgemm fp8 (#3625)

Browse source 
* skip xpu fbgemm fp8

* Apply suggestion from @gemini-code-assist[bot]

Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* Apply suggestion from @danielhanchen

---------

Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: Daniel Han <danielhanchen@gmail.com>
This commit is contained in:
Lei Zhenyuan 2025-12-11 13:13:29 +08:00 • committed by GitHub
commit beb83d7f28
1 changed files with 14 additions and 11 deletions
Download patch file Download diff file Expand all files Collapse all files

25
unsloth/models/_utils.py
View file

@ -2311,17 +2311,20 @@ def verify_fp8_support_if_applicable(model_config):
raise ValueError(
f"Unsloth: FP8 quantization is only supported on CUDA GPUs. You are using {DEVICE_TYPE}."
)
major_version, minor_version = torch.cuda.get_device_capability()
if quant_method == "fbgemm_fp8" and major_version < 9:
# While L4 does support FP8 as data type, it doesn't have fbgemm (package) support yet. So we restrict it.
raise ValueError(
f"Unsloth: FBGEMM FP8 quantization is only supported on H100 and higher GPUs. L4 is not supported. You are using {torch.cuda.get_device_name()}. Refer to https://developer.nvidia.com/cuda-gpus for more details."
)
if quant_method == "fp8" and major_version * 10 + minor_version < 89:
# In case of block quantized, we allow L4 because we fall back to torchao kernels.
raise ValueError(
f"Unsloth: FP8 quantization is only supported on L4 and higher GPUs with compute capability 8.9 or higher. You are using {torch.cuda.get_device_name()}. Refer to https://developer.nvidia.com/cuda-gpus for more details."
)
# [TODO] Need to add FP8 support for Intel XPUs
if DEVICE_TYPE == "cuda":
major_version, minor_version = torch.cuda.get_device_capability()
if quant_method == "fbgemm_fp8" and major_version < 9:
# While L4 does support FP8 as data type, it doesn't have fbgemm (package) support yet. So we restrict it.
raise ValueError(
f"Unsloth: FBGEMM FP8 quantization is only supported on H100 and higher GPUs. L4 is not supported. You are using {torch.cuda.get_device_name()}. Refer to https://developer.nvidia.com/cuda-gpus for more details."
)
if quant_method == "fp8" and major_version * 10 + minor_version < 89:
# In case of block quantized, we allow L4 because we fall back to torchao kernels.
raise ValueError(
f"Unsloth: FP8 quantization is only supported on L4 and higher GPUs with compute capability 8.9 or higher. You are using {torch.cuda.get_device_name()}. Refer to https://developer.nvidia.com/cuda-gpus for more details."
)
def _get_inference_mode_context_manager(model: torch.nn.Module):