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

[torch.compile] Enable AR+rms fusion by default available for -O2 (#34299)

Browse Source 
Signed-off-by: Luka Govedič <lgovedic@redhat.com>
This commit is contained in:
Luka Govedič
2026-02-11 03:30:00 -05:00
committed by GitHub
parent 675a22ed66
commit addac0e653
2 changed files with 16 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
vllm/config/compilation.py
View File

@@ -115,7 +115,7 @@ class PassConfig:
"""Fuse the custom SiluMul + quant ops."""
fuse_attn_quant: bool = Field(default=None)
"""Fuse the custom attention + quant ops."""
eliminate_noops: bool = Field(default=None)
eliminate_noops: bool = Field(default=True)
"""Eliminate no-op ops."""
enable_sp: bool = Field(default=None)
"""Enable sequence parallelism."""
@@ -194,7 +194,6 @@ class PassConfig:
"fuse_norm_quant",
"fuse_act_quant",
"fuse_attn_quant",
"eliminate_noops",
"enable_sp",
"fuse_gemm_comms",
"fuse_allreduce_rms",