Skip to content

Support partial rollout resume in Search-R1 example#2128

Open
OLIVER-XYP wants to merge 2 commits into
THUDM:mainfrom
OLIVER-XYP:search-r1-partial-rollout
Open

Support partial rollout resume in Search-R1 example#2128
OLIVER-XYP wants to merge 2 commits into
THUDM:mainfrom
OLIVER-XYP:search-r1-partial-rollout

Conversation

@OLIVER-XYP

Copy link
Copy Markdown

Summary

  • Enable partial rollout resume for the Search-R1 custom generate example.
  • Preserve partial response, tokens, loss masks, and rollout log probabilities across aborted rollouts.
  • Resume generation from existing Search-R1 context instead of restarting from the prompt.
  • Document partial rollout usage in the Search-R1 READMEs.

Tests

  • python -m py_compile examples\search-r1\generate_with_search.py tests\test_search_r1_partial_rollout.py
  • pytest tests/test_sample.py tests/test_search_r1_partial_rollout.py -q

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant