xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-04-29 07:07:07 +08:00
Code Issues Packages Projects Releases Wiki Activity

[compile] Disable aot when eager backend is used. (#30810)

Browse Source 
Signed-off-by: zhxchen17 <zhxchen17@fb.com>
This commit is contained in:
Zhengxu Chen 2025-12-17 04:55:56 -05:00 committed by GitHub
parent 519ef9a911
commit 177c391db2
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 9 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
vllm/compilation/decorators.py
View File

@ -435,7 +435,10 @@ def _support_torch_compile(
return self.aot_compiled_fn(self, *args, **kwargs)
if self.compiled:
assert not envs.VLLM_USE_AOT_COMPILE
assert (
not envs.VLLM_USE_AOT_COMPILE
or self.vllm_config.compilation_config.backend == "eager"
)
return TorchCompileWithNoGuardsWrapper.__call__(self, *args, **kwargs)
# This is the path for the first compilation.
@ -508,7 +511,11 @@ def _support_torch_compile(
_torch27_patch_tensor_subclasses(),
torch._inductor.config.patch(**inductor_config_patches),
):
if envs.VLLM_USE_AOT_COMPILE:
use_aot_compile = envs.VLLM_USE_AOT_COMPILE
if self.vllm_config.compilation_config.backend == "eager":
logger.warning("Detected eager backend, disabling AOT compile.")
use_aot_compile = False
if use_aot_compile:
self.aot_compiled_fn = self.aot_compile(*args, **kwargs)
output = self.aot_compiled_fn(self, *args, **kwargs)
assert aot_compilation_path is not None