xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-03-27 05:59:06 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Debugging] Add annotation for easier trace analysis (#22496)

Browse Source
This commit is contained in:
Dayeol Lee 2025-11-05 16:52:52 -08:00 committed by GitHub
parent efe73e9b57
commit 1767658559
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 19 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

22
vllm/v1/worker/gpu_worker.py
View File

@ -509,6 +509,19 @@ class Worker(WorkerBase):
def get_supported_tasks(self) -> tuple[SupportedTask, ...]:
return self.model_runner.get_supported_tasks()
def annotate_profile(self, scheduler_output):
# add trace annotation so that we can easily distinguish
# new/cached request numbers in each iteration
if not self.profiler:
return nullcontext()
num_new = len(scheduler_output.scheduled_new_reqs)
num_cached = len(scheduler_output.scheduled_cached_reqs.req_ids)
return torch.profiler.record_function(
f"execute_new_{num_new}_cached_{num_cached}"
)
@torch.inference_mode()
def sample_tokens(
self, grammar_output: "GrammarOutput | None"
@ -536,9 +549,12 @@ class Worker(WorkerBase):
)
)
output = self.model_runner.execute_model(scheduler_output, intermediate_tensors)
if isinstance(output, (ModelRunnerOutput, NoneType)):
return output
with self.annotate_profile(scheduler_output):
output = self.model_runner.execute_model(
scheduler_output, intermediate_tensors
)
if isinstance(output, (ModelRunnerOutput, NoneType)):
return output
assert isinstance(output, IntermediateTensors)
parallel_config = self.vllm_config.parallel_config