Fix for Exception raised while parsing Chat Completions streaming response, in some rare cases #39741
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
github-actions
bot
added
the
AI Model Inference
nick863
approved these changes
Feb 14, 2025
howieleung
reviewed
Feb 14, 2025
…' into dargilco/azure-ai-inference-pr-utf8-streaming-fix-part2
|
API change check APIView has identified API level changes in this PR and created following API reviews. |
dargilco
merged commit 3dc6388
into
feature/azure-ai-inference-beta9
21 of 24 checks passed
dargilco
deleted the
dargilco/azure-ai-inference-pr-utf8-streaming-fix-part2
branch
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Outdated suggestions cannot be applied.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.
Description
I finally got one repro of the GitHub issue, while taking SDK logs and using DeepSeek model. That confirmed what I suspected. But I could not understand why the unit-test I added a couple of days ago, streaming Chinese characters broken across lines, did not exhibit the same issue. That unit-test passed.
After investigation turns out I had a bug in the unit-test! Missing comma after one of the SSE lines in the input array for the test. Which miraculously made the unit-test pass, when they should have failed exactly as reported in the GitHub issue.
After I discovered that, it was easy to implement a fix. I updated the logic in the SSE parsing to make sure UTF-8 decoding was moved further down, at the point where we are guaranteed we have a complete line of JSON string, just before deserializing it into the output chunk object. Until that point (including caching the previous incomplete line) the input is still handled as a "bytes" object.
I also updated all streaming samples with extra checks during printing of the streaming response and also printed the token usage. We already did that in the GitHub samples.