Force EOS at max_tokens with importance-weight correction#145
Open
ClementeP wants to merge 1 commit into
Open
Conversation
SequenceModel previously truncated sequences without an EOS token when hitting max_tokens, producing samples that did not target the length-conditioned distribution. At the boundary step we now swap the proposal for a point mass on EOS and apply the corresponding IS correction via a new TokenSampler.logw_eos method, so particles are properly weighted with respect to the target conditioned on |y| <= max_tokens; every returned sequence ends with EOS. SetTokenSampler overrides logw_eos to use complete(ctx) - prefix(ctx), avoiding materializing the full logw_next vector that set samplers are designed to skip during regular sampling. Tests updated to reflect the new boundary semantics and a focused prefix-overlap test added to cover a case the existing property test was not exercising. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Codecov Report✅ All modified and coverable lines are covered by tests. 📢 Thoughts on this report? Let us know! |
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
SequenceModel previously truncated sequences without an EOS token when hitting max_tokens, producing samples that did not target the length-conditioned distribution. At the boundary step we now swap the proposal for a point mass on EOS and apply the corresponding IS correction via a new TokenSampler.logw_eos method, so particles are properly weighted with respect to the target conditioned on |y| <= max_tokens; every returned sequence ends with EOS.
SetTokenSampler overrides logw_eos to use complete(ctx) - prefix(ctx), avoiding materializing the full logw_next vector that set samplers are designed to skip during regular sampling.
Tests updated to reflect the new boundary semantics and a focused prefix-overlap test added to cover a case the existing property test was not exercising.