Fix OpenAI chat stream hanging #156
Merged
Conversation
This file contains 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
brainlid
approved these changes
Jul 18, 2024
There was a problem hiding this comment.
Awesome @stevehodgkiss! Thanks! I just noticed that hanging behavior the other day but hadn't diagnosed it yet.
❤️💛💙💜
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Outdated suggestions cannot be applied.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.
We recently noticed that chat streams would appear to "hang" in the middle of generating a response. After a bit of debugging I found that it was due to the message being split in the middle of the "data: " part (i.e. "dat" in the last message, then "a: {" in the next). This results in the stream failing to parse any future messages and appearing to "hang" or freeze.
Haven't seen this issue before so I'm guessing something has changed in the way OpenAI streams their responses.
I couldn't see any tests for this particular case in the official node & python clients, but I modified one of the node clients tests to check and it is supported (still passes). The node client appears to split by line endings first, then by
:to get the field name (i.e.data).The change in this PR pattern matches on their being something in the buffer (previous/partial message data), combines the buffered & new data then recursively calls the
decode_streamfunction, since the combined data may need splitting again on the "data: " key.