Skip to content

scheduler: skip speculative decoding when all scheduled requests need <=1 output token #437

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

Closed
malaiwah wants to merge 1 commit into from
Closed

scheduler: skip speculative decoding when all scheduled requests need <=1 output token #437

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
9 changes: 9 additions & 0 deletions vllm/v1/core/sched/scheduler.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1160,6 +1160,15 @@ def schedule(self, throttle_prefills: bool = False) -> SchedulerOutput:
self.dynamic_sd_lookup[len(num_scheduled_tokens)],
)

# Skip spec decode when all scheduled requests need ≤1 output token.
# Speculative decoding is useless for 1-token outputs and adds ~40ms
# of draft generation + verification overhead.
if num_spec_tokens_to_schedule > 0:
if (scheduled_new_reqs
and not scheduled_running_reqs
and all(req.max_tokens <= 1 for req in scheduled_new_reqs)):
num_spec_tokens_to_schedule = 0

scheduled_encoder_input_stats = None
if (
self.log_stats
Expand Down
Loading