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

Feature: Files and Batches as a unified route #862

Merged
merged 16 commits into from
Jan 17, 2025
Merged

Conversation

narengogi
Copy link
Collaborator

@narengogi narengogi commented Jan 13, 2025

Related Issues: (optional)

Feature to add batches and files as a unified route
This add the following routes:

POST /v1/files
GET /v1/files
GET /v1/files/:fileId
DELETE /v1/files/:fileId

POST /v1/batches
GET /v1/batches
GET /v1/batches/:batchId
POST /v1/batches/:batchId/cancel

✨ Gateway specific additional route for convenience (Not in OpenI spec) ✨
GET /v1/batches/:batchId/output

Description

  • For file uploads, the gateway needs to transform files to the provider format (like bedrock) on the fly, the files can be as large as 20GB, this implementation does not load the files into memory, rather it transforms the data streams on the fly with nodeJS streams and pushes to the other endpoint with a half duplex connection.

  • Similar streaming transformers for getting batch output and getting files is implemented, where the file stream is transformed on the fly

  • I've had to upgrade wrangler version for nodeJS compatibility to add Polyfill support for nodeJS crypto module which is used for signing the AWS requests, although the gateway has @smithy/signature-v4 package installed, I wasn't able to get it to work for the s3 file upload routes, I'll have to revisit this

Testing done

  • To make sure that we're not adding any additional memory overhead, and that there are no memory leaks, I've monitored the memory consumption with the top command on my machine
image during testing a 1gb file upload, memory went up by about 40MiB, but returned to normal shortly after

TO-DO for later

  • Revisit and simplify the AWS signer code, we should be able to use the @smithy/signature-v4 package without having to write all the signing code ourselves, or alternatively we can remove the smithy signing package and implement the functions ourselves

@narengogi narengogi requested a review from sk-portkey January 13, 2025 11:20
src/providers/ai21/api.ts Outdated Show resolved Hide resolved
@narengogi narengogi requested a review from VisargD January 14, 2025 13:35
VisargD
VisargD previously approved these changes Jan 16, 2025
@narengogi narengogi changed the title Feature Files and Batches Feature: Files and Batches as a unified route Jan 17, 2025
@VisargD VisargD merged commit c0a8fd4 into main Jan 17, 2025
2 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants