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

[Log] Only Print Profiler Results on Rank 0 (#23370)

Browse Source 
Signed-off-by: yewentao256 <zhyanwentao@126.com>
This commit is contained in:
Wentao Ye
2025-09-02 14:53:34 -04:00
committed by GitHub
parent 598bd74cf8
commit 98aee612aa
2 changed files with 8 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
vllm/v1/worker/gpu_worker.py
View File

@@ -400,6 +400,8 @@ class Worker(WorkerBase):
self.profiler.start() self.profiler.start()
else: else:
self.profiler.stop() self.profiler.stop()
# only print profiler results on rank 0
if self.local_rank == 0:
print(self.profiler.key_averages().table( print(self.profiler.key_averages().table(
sort_by="self_cuda_time_total")) sort_by="self_cuda_time_total"))

6
vllm/worker/worker.py
View File

@@ -128,8 +128,10 @@ class Worker(LocalOrDistributedWorkerBase):
if self.profiler is None: if self.profiler is None:
raise RuntimeError("Profiler is not enabled.") raise RuntimeError("Profiler is not enabled.")
self.profiler.stop() self.profiler.stop()
print( # only print profiler results on rank 0
self.profiler.key_averages().table(sort_by="self_cuda_time_total")) if self.local_rank == 0:
print(self.profiler.key_averages().table(
sort_by="self_cuda_time_total"))
def sleep(self, level: int = 1) -> None: def sleep(self, level: int = 1) -> None:
free_bytes_before_sleep = torch.cuda.mem_get_info()[0] free_bytes_before_sleep = torch.cuda.mem_get_info()[0]