Enable efficient KV-cache based generation for Coconut latent reasoning#10
Open
somepatt wants to merge 1 commit intoInternLM:mainfrom
Open
Enable efficient KV-cache based generation for Coconut latent reasoning#10somepatt wants to merge 1 commit intoInternLM:mainfrom
somepatt wants to merge 1 commit intoInternLM:mainfrom
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
This PR refactors the generate() logic in Coconut to correctly and efficiently support KV-cache–based autoregressive decoding after latent token replacement.
Previously, generation recomputed the entire prefix on every decoding step by passing the full inputs_embeds sequence, which:
disabled effective KV-cache usage,
led to unnecessary quadratic computation,
and did not reflect how causal LMs are intended to be decoded.
This change aligns Coconut generation with standard causal LM decoding while preserving Coconut’s latent-replacement semantics.