Skip to content

fix: clamp max_new_tokens on retry to prevent response_length overflow #2003

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

Open
YaoweiFan wants to merge 1 commit into
base: main
Choose a base branch
from
Open

fix: clamp max_new_tokens on retry to prevent response_length overflow #2003

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 slime/rollout/sglang_rollout.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -163,6 +163,15 @@ async def generate(args: Namespace, sample: Sample, sampling_params: dict[str, A

prompt_ids = _prepare_prompt_ids(sample, state.tokenizer, state.processor)

# Clamp on retry: ABORTED samples carry response_length from prior attempts.
sampling_params = dict(sampling_params)
if args.rollout_max_response_len is not None:
remaining = args.rollout_max_response_len - sample.response_length
if remaining <= 0:
sample.status = Sample.Status.TRUNCATED
return sample
sampling_params["max_new_tokens"] = min(sampling_params["max_new_tokens"], remaining)

assert (
sampling_params["max_new_tokens"] >= 0
), f"max_new_tokens: {sampling_params['max_new_tokens']} should not be less than 0"
Expand Down
Loading