A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 5, 09:45 UTC
[ROCm] Correctly guard fused_silu_mul_block_quant on ROCm (#39387)
Signed-off-by: Micah Williamson <micah.williamson@amd.com>
Micah Williamson committed
Apr 9, 2026 at 12:59 UTC
d5f75fdf508f034b4148b3c0ac9b0c691e9b746f
1 file changed
+1
-1
vllm/compilation/passes/fusion/act_quant_fusion.py
+1
-1
@@ -301,7 +301,7 @@ class ActivationQuantFusionPass(VllmPatternMatcherPass):
301
pattern_silu_mul_nvfp4 = SiluMulNvfp4QuantPattern()
302
pattern_silu_mul_nvfp4.register(self.patterns)
303
304
- if current_platform.is_cuda_alike():
304
+ if current_platform.is_cuda():
305
for quant_key in [kFp8Dynamic128Sym, kFp8Dynamic64Sym]:
306
for is_scale_transposed in [False, True]:
307
for is_e8m0 in [True, False]: