mystic-ai / pipeline

Pipeline is an open source python SDK for building AI/ML workflows

https://www.mystic.ai

Apache License 2.0

117 stars 21 forks source link

PC-1044 Streaming runs #422

Closed paulcjh closed 3 months ago

paulcjh commented 3 months ago

Pull request outline

Checklist:

[ ] Docs updated
[ ] Tests written
[ ] Check for breaking changes in Resource
[ ] Version bumped

Added:

A list of new things

Changed:

A list of things changed

Removed:

A list of things removed

Related issues:

none

You can add a reference to an issue using hash followed by the issue number
If a checklist item is to be ignored it must be struck through

linear[bot] commented 3 months ago

PC-1044 Streaming runs

rossgray commented 3 months ago

@paulcjh I wonder if we should make the streaming endpoint in the container a separate endpoint, as we're doing with the backend API, since previously we've tried to keep /v4/runs consistent between the container and pcore?

rossgray commented 3 months ago

the only other thought I had (and maybe we can always do this later) was whether it makes sense to return the full run result schema every time, or just the raw outputs? Since I have a feeling all of the serialising JSON in and out of Pydantic objects may add a bit of overhead. We could always add a separate endpoint for that later though. Maybe also an endpoint where the results are streamed directly from the resource too without going via catalyst and pcore