Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

feat: Treat HTTP request completion as final response #748

Merged
merged 18 commits into from
Jul 12, 2024

Conversation

nnshah1
Copy link
Contributor

@nnshah1 nnshah1 commented Jul 11, 2024

For HTTP streaming endpoints which do not explicitly send a [DONE] message, updated to treat the HTTP request completion as a final response.

This is backwards compatible and doesn't effect endpoints that do explicitly send [DONE].

This is one change needed for Triton Generate as well as some TGI interfaces / servers.

Checks are added to ensure only a single final response is sent.

Copy link
Contributor

@dyastremsky dyastremsky left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Do you have a quick repro/test we could do to the see the effect e.g. with the generate endpoint? We might want to add a CI test to verify this behavior.

@nnshah1
Copy link
Contributor Author

nnshah1 commented Jul 11, 2024

Do you have a quick repro/test we could do to the see the effect e.g. with the generate endpoint? We might want to add a CI test to verify this behavior.

will require further changes to support generate - was thinking to seperate the C++ changes from the python changes to make merging easier - but we ccould do an feature branch - with seperate MRs into it - if preferred -

I'm good either way

@nnshah1 nnshah1 merged commit b9bab20 into main Jul 12, 2024
5 checks passed
@nnshah1 nnshah1 deleted the nnshah1-openai-finish-on-close branch July 12, 2024 00:33
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Development

Successfully merging this pull request may close these issues.

5 participants