Stop pipelined generation when consumer drops the stream#113
Merged
Conversation
When the consumer breaks out of the token stream (e.g. at EOS), the forwarding loop now detects the terminated yield result and finishes the inner token continuation, which signals the GPU producer to stop. Additionally, outputContinuation.onTermination finishes the inner stream so cancellation propagates even if the forwarding loop hasn't started iterating yet. Also cancel any prior active generation at the start of generate() on all engine types (sequential, static-shape, VLM). This ensures the cancel-and-replace contract is upheld regardless of which engine is in use, preventing state overlap between turns.
stikves
requested review from
alejandro-isaza,
carinapeng,
kevchengcodes and
tjia1818
July 17, 2026 16:56
stikves
marked this pull request as ready for review
July 17, 2026 16:56
tjia1818
approved these changes
Jul 17, 2026
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
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
When the consumer breaks out of the token stream (e.g. at EOS), the forwarding loop now detects the terminated yield result and finishes the inner token continuation, which signals the GPU producer to stop.
Additionally, outputContinuation.onTermination finishes the inner stream so cancellation propagates even if the forwarding loop hasn't started iterating yet.
Also cancel any prior active generation at the start of generate() on all engine types (sequential, static-shape, VLM). This ensures the cancel-and-replace contract is upheld regardless of which engine is in use, preventing state overlap between turns.