Repository navigation
fix(agent): propagate multi-turn rollout truncation - #1327
Merged
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
MultiTurnAgentExecutor drops vLLM's
finish_reason="length", so sample conversion marks those rollouts as untruncated. This underreports truncation metrics and bypasses the stop-properly reward penalty when enabled.Carry the final generation's truncation flag in the executor response, matching SingleTurnAgentExecutor and reusing the existing sample conversion and penalty logic.
Validation: 24 regression cases covering stop/length endings over one or two turns, with and without logprobs, and three penalty coefficients. The unpatched code fails all 12 length cases; the patched full test suite passes (46 tests). Compileall and pre-commit pass. A separate CPU probe using the shipped AgentInstance, 30 public texts and tiny-model generation also passes; no full GPU training run was performed.