biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[ROCm] Correctly guard fused_silu_mul_block_quant on ROCm (#39387)

Browse Source 
Signed-off-by: Micah Williamson <micah.williamson@amd.com>
This commit is contained in:
Micah Williamson
2026-04-09 12:59:03 -05:00
committed by GitHub
parent 827268e98d
commit d5f75fdf50
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
vllm/compilation/passes/fusion/act_quant_fusion.py
View File

@@ -301,7 +301,7 @@ class ActivationQuantFusionPass(VllmPatternMatcherPass):
pattern_silu_mul_nvfp4 = SiluMulNvfp4QuantPattern()
pattern_silu_mul_nvfp4.register(self.patterns)
if current_platform.is_cuda_alike():
if current_platform.is_cuda():
for quant_key in [kFp8Dynamic128Sym, kFp8Dynamic64Sym]:
for is_scale_transposed in [False, True]:
for is_e8m0 in [True, False]: