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

[CI][torch.compile] Reduce e2e fusion test time (#33293)

Browse Source 
Signed-off-by: Luka Govedič <lgovedic@redhat.com>
Signed-off-by: ProExpertProg <luka.govedic@gmail.com>
Signed-off-by: Luka Govedič <ProExpertProg@users.noreply.github.com>
This commit is contained in:
Luka Govedič
2026-02-04 19:09:03 -05:00
committed by GitHub
parent 439afa4eea
commit 4d9513537d
17 changed files with 1068 additions and 821 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
vllm/config/vllm.py
View File

@@ -766,7 +766,12 @@ class VllmConfig:
if self.compilation_config.pass_config.fuse_gemm_comms:
self.compilation_config.pass_config.enable_sp = True
if self.compilation_config.pass_config.enable_sp:
if "-rms_norm" in self.compilation_config.custom_ops:
if self.parallel_config.tensor_parallel_size == 1:
logger.warning("Sequence Parallelism requires TP>1, disabling")
self.compilation_config.pass_config.enable_sp = False
self.compilation_config.pass_config.fuse_gemm_comms = False
elif "-rms_norm" in self.compilation_config.custom_ops:
logger.warning(
"RMS norm force disabled, sequence parallelism might break"
)