xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-19 22:15:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Log] Save profiler results to file instead of stdout (#28144)

Browse Source 
Signed-off-by: Randall Smith <ransmith@amd.com>
Co-authored-by: Randall Smith <ransmith@amd.com>
This commit is contained in:
rasmith 2025-11-14 17:26:39 -06:00 committed by GitHub
parent e0c910bb89
commit ba041d980b
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 13 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

21
vllm/v1/worker/gpu_worker.py
View File

@ -596,14 +596,19 @@ 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 isinstance(self.profiler, torch.profiler.profile):
if ( rank = self.local_rank
isinstance(self.profiler, torch.profiler.profile) profiler_dir = envs.VLLM_TORCH_PROFILER_DIR
and self.local_rank == 0 profiler_out_file = f"{profiler_dir}/profiler_out_{rank}.txt"
): sort_key = "self_cuda_time_total"
print( table = self.profiler.key_averages().table(sort_by=sort_key)
self.profiler.key_averages().table(sort_by="self_cuda_time_total")
) with open(profiler_out_file, "w") as f:
print(table, file=f)
# only print profiler results on rank 0
if rank == 0:
print(table)
def execute_dummy_batch(self) -> None: def execute_dummy_batch(self) -> None:
self.model_runner._dummy_run(1, uniform_decode=True) self.model_runner._dummy_run(1, uniform_decode=True)