xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-14 00:05:36 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix incompatibililty with non cuda platform for nvfp4 (#23478)

Browse Source 
Signed-off-by: Lu Fang <fanglu@fb.com>
Co-authored-by: Lucia (Lu) Fang <fanglu@meta.com>
This commit is contained in:
Lucia Fang 2025-08-24 15:35:41 -07:00 committed by GitHub
parent ad78868450
commit c7fc6b1354
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 3 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
vllm/compilation/fusion.py
View File

@ -47,8 +47,10 @@ QUANT_OPS: dict[QuantKey, OpOverload] = {
torch.ops._C.dynamic_scaled_fp8_quant.default, # noqa: E501 torch.ops._C.dynamic_scaled_fp8_quant.default, # noqa: E501
kFp8DynamicTokenSym: kFp8DynamicTokenSym:
torch.ops._C.dynamic_per_token_scaled_fp8_quant.default, # noqa: E501 torch.ops._C.dynamic_per_token_scaled_fp8_quant.default, # noqa: E501
kNvfp4Quant: torch.ops._C.scaled_fp4_quant.default, # noqa: E501
} }
if current_platform.is_cuda() and hasattr(torch.ops._C, "scaled_fp4_quant"):
QUANT_OPS[
kNvfp4Quant] = torch.ops._C.scaled_fp4_quant.default # noqa: E501
class FusedRMSQuantKey(NamedTuple): class FusedRMSQuantKey(NamedTuple):