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

[torch.compile] Turn on silu+fp4 quant fusion by default for O1+ (#34718)

Browse Source 
Signed-off-by: Luka Govedič <lgovedic@redhat.com>
This commit is contained in:
Luka Govedič
2026-02-17 22:29:15 -05:00
committed by GitHub
parent 4a00a511bb
commit 02e8f26cea
1 changed files with 10 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

15
vllm/config/vllm.py
View File

@@ -95,11 +95,16 @@ def enable_norm_fusion(cfg: "VllmConfig") -> bool:
def enable_act_fusion(cfg: "VllmConfig") -> bool: def enable_act_fusion(cfg: "VllmConfig") -> bool:
"""Enable if either SiLU+Mul or quant FP8 custom op is active; """
otherwise Inductor handles fusion.""" Enable if either SiLU+Mul or quant FP8 custom op is active;
return cfg.compilation_config.is_custom_op_enabled( otherwise Inductor handles fusion.
"silu_and_mul" Also enable for FP4 models as FP4 quant is always custom so Inductor cannot fuse it.
) or cfg.compilation_config.is_custom_op_enabled("quant_fp8") """
return (
cfg.compilation_config.is_custom_op_enabled("silu_and_mul")
or cfg.compilation_config.is_custom_op_enabled("quant_fp8")
or (cfg.model_config is not None and cfg.model_config.is_nvfp4_quantized())
)
def enable_allreduce_rms_fusion(cfg: "VllmConfig") -> bool: def enable_allreduce_rms_fusion(cfg: "VllmConfig") -> bool: