Skip to content

[None][fix] Cherry-pick #10715: Disable short profile for tunable ops with MERGE strategy #10844

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
hyukn merged 1 commit into from
Jan 21, 2026
Merged

[None][fix] Cherry-pick #10715: Disable short profile for tunable ops with MERGE strategy #10844

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 4 additions & 1 deletion tensorrt_llm/_torch/autotuner.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1105,7 +1105,10 @@ def pure_profile(stream: torch.cuda.Stream, repeat: int):

disable_short_profile = os.environ.get(
"TLLM_AUTOTUNER_DISABLE_SHORT_PROFILE", "0") == "1"
if fewer_repeat_avg_time > short_profile_threshold_ms and not disable_short_profile:

# Disable this feature for merged tuning strategy to avoid potential hang due to asymmetric tuning.
if fewer_repeat_avg_time > short_profile_threshold_ms and not disable_short_profile \
and tuning_config.distributed_tuning_strategy != DistributedTuningStrategy.MERGE:
# directly use the few repeat estimated time to avoid redundant profiling
avg_time = fewer_repeat_avg_time
else:
Expand Down
Loading