documentation-website icon indicating copy to clipboard operation
documentation-website copied to clipboard

[DOC] Update Cohere Connector documentation (v3 and PQ guidance)

Open dylan-tong-aws opened this issue 11 months ago • 6 comments

What do you want to do?

  • [ x] Request a change to existing documentation
  • [ ] Add new documentation
  • [ ] Report a technical problem with the documentation
  • [ ] Other

Tell us about your request. Provide a summary of the request and all versions that are affected.

Update the Cohere connector documentation: https://opensearch.org/docs/latest/ml-commons-plugin/remote-models/connectors/#cohere-connector

  1. Update the blueprint to use the latest version for cohere embed v3. Please work with @ylwu-amzn to obtain the latest blueprint. I can't find it in Github, and there are no examples in the documentation.

  2. Add a section for PQ support guidance:

Cohere embed v3 has been tuned for IVF PQ index support (https://opensearch.org/docs/latest/search-plugins/knn/knn-index/#supported-faiss-encoders). The following default parameters should work well:

code_size=8 Cohere recommend m=256 (for a 16x compression) or m=128 (for a 32x compression).

As similarity_metric, cohere has tested and tuned the model inner product

What other resources are available? Provide links to related issues, POCs, steps for testing, etc.

dylan-tong-aws avatar Mar 11 '24 23:03 dylan-tong-aws

Cohere embed blueprint (v2 and v3): https://github.com/opensearch-project/ml-commons/blob/2.x/docs/remote_inference_blueprints/cohere_connector_embedding_blueprint.md

ylwu-amzn avatar Mar 11 '24 23:03 ylwu-amzn

Thanks, @dylan-tong-aws and @ylwu-amzn ! Is this connected to a release?

hdhalter avatar Mar 12 '24 22:03 hdhalter

Cohere Connector documentation

This part is not for 2.13 release. It's just a doc for released feature.

ylwu-amzn avatar Mar 13 '24 21:03 ylwu-amzn

This PR addresses the first item above: https://github.com/opensearch-project/documentation-website/pull/6671

hdhalter avatar Mar 13 '24 22:03 hdhalter

The first item is merged. I'm working with jmazane on the second item.

hdhalter avatar Mar 14 '24 15:03 hdhalter

@jmazanec15 - Did we address #2 above?

hdhalter avatar Apr 26 '24 21:04 hdhalter