frozenleaves/vllm
1
0
Fork 0
mirror of https://github.com/vllm-project/vllm.git synced 2025-10-20 23:03:52 +08:00
Code Issues Packages Projects Releases Wiki Activity

[CI] Fix broken compile tests due to unsupported SiluMul+Nvfp4Quant fusion (#23973)

Browse Source 
Signed-off-by: Yong Hoon Shin <yhshin@meta.com>
Co-authored-by: Roger Wang <hey@rogerw.io>
This commit is contained in:
Yong Hoon Shin
2025-08-30 00:14:43 -07:00
committed by GitHub
parent ee52a32705
commit 9748c5198b
1 changed files with 6 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
vllm/compilation/activation_quant_fusion.py
View File

@ -29,8 +29,9 @@ SILU_MUL_OP = torch.ops._C.silu_and_mul.default
FUSED_OPS: dict[QuantKey, OpOverload] = { FUSED_OPS: dict[QuantKey, OpOverload] = {
kFp8StaticTensorSym: torch.ops._C.silu_and_mul_quant.default, # noqa: E501 kFp8StaticTensorSym: torch.ops._C.silu_and_mul_quant.default, # noqa: E501
} }
if current_platform.is_cuda() and hasattr(torch.ops._C, silu_and_mul_nvfp4_quant_supported = (current_platform.is_cuda() and hasattr(
"silu_and_mul_nvfp4_quant"): torch.ops._C, "silu_and_mul_nvfp4_quant"))
if silu_and_mul_nvfp4_quant_supported:
FUSED_OPS[ FUSED_OPS[
kNvfp4Quant] = torch.ops._C.silu_and_mul_nvfp4_quant.default # noqa: E501 kNvfp4Quant] = torch.ops._C.silu_and_mul_nvfp4_quant.default # noqa: E501
@ -171,8 +172,9 @@ class ActivationQuantFusionPass(VllmInductorPass):
pattern_silu_mul_fp8 = SiluMulFp8StaticQuantPattern() pattern_silu_mul_fp8 = SiluMulFp8StaticQuantPattern()
pattern_silu_mul_fp8.register(self.patterns) pattern_silu_mul_fp8.register(self.patterns)
pattern_silu_mul_nvfp4 = SiluMulNvfp4QuantPattern() if silu_and_mul_nvfp4_quant_supported:
pattern_silu_mul_nvfp4.register(self.patterns) pattern_silu_mul_nvfp4 = SiluMulNvfp4QuantPattern()
pattern_silu_mul_nvfp4.register(self.patterns)
def __call__(self, graph: torch.fx.Graph): def __call__(self, graph: torch.fx.Graph):
self.begin() self.begin()