tinker seam: unpack get_rollout_data tuple in fb-only/forward paths - #4
Merged
Merged
Conversation
Upstream changed get_rollout_data to return (RolloutBatch, ObjectStoreGetResult); the seam's forward_backward_only and forward_logprobs still bound the tuple, so every call failed in get_data_iterator. Unpack and hold the release context for the duration of data use, mirroring train().
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Upstream changed
get_rollout_datato return(RolloutBatch, ObjectStoreGetResult); the seam'sforward_backward_onlyandforward_logprobs(rebased in #3) still bound the tuple, so every fb call failed inget_data_iterator(AttributeError: 'tuple' object has no attribute 'get').Fix: unpack, and hold the release context for the duration of data use, mirroring
train().Validated on 4xH200 (DP=4, Qwen2.5-0.5B LoRA r32): grad-accum split-invariance ratio exactly 1.0 — grad_norm bit-identical (178.9398358212894) for 1xfb(8) vs 2xfb(4) + step, and across reruns.
Merge via rebase-merge to keep main linear.