xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-07-08 04:07:14 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Refactor] refactor freezing_value/cuda_event initialize outside try finally (#23758)

Browse Source 
Signed-off-by: Andy Xie <andy.xning@gmail.com>
This commit is contained in:
Ning Xie 2025-08-30 21:42:25 +08:00 committed by GitHub
parent fb4983e112
commit e80bca309e
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 3 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
vllm/v1/worker/cpu_model_runner.py
View File

@ -128,8 +128,8 @@ def _torch_cuda_wrapper():
self.record = lambda: None self.record = lambda: None
self.synchronize = lambda: None self.synchronize = lambda: None
cuda_event = torch.cuda.Event
try: try:
cuda_event = torch.cuda.Event
torch.cuda.Event = _EventPlaceholder torch.cuda.Event = _EventPlaceholder
yield yield
finally: finally:
@ -141,9 +141,9 @@ def _set_global_compilation_settings(config: VllmConfig):
import torch._inductor.config import torch._inductor.config
inductor_config = config.compilation_config.inductor_compile_config inductor_config = config.compilation_config.inductor_compile_config
# Note: The MKLDNN and CPPGEMM backend requires freezing parameters.
freezing_value = torch._inductor.config.freezing
try: try:
# Note: The MKLDNN and CPPGEMM backend requires freezing parameters.
freezing_value = torch._inductor.config.freezing
if inductor_config.get("max_autotune", False): if inductor_config.get("max_autotune", False):
torch._inductor.config.freezing = True torch._inductor.config.freezing = True
yield yield