lmql icon indicating copy to clipboard operation
lmql copied to clipboard

Supporting Google and AWS LLMs available through APIs

Open RCegielaBCG opened this issue 1 year ago • 6 comments

Having an option to use e.g. GCP Bison model instead of GPT would be really helpful. Some organizations are very attached to their cloud vendors...

RCegielaBCG avatar Jul 14 '23 12:07 RCegielaBCG

Similar to Anthropic #118 models, GCP does not currently seem to offer support for model steering, i.e. masking of the model distribution during text generation: https://cloud.google.com/vertex-ai/docs/generative-ai/model-reference/text

This means, we can offer support, but advanced constraining/steering will not be available, due to API restrictions on their end.

lbeurerkellner avatar Jul 17 '23 08:07 lbeurerkellner

Does any of the pretrained models in AWS Bedrock support it (A21, Titan)? How about Luminous from Aleph Alpha?

RCegielaBCG avatar Jul 17 '23 09:07 RCegielaBCG

There are now models in Bedrock that return logits in different forms. Here are a couple of examples:

  1. Cohere's command model: https://docs.cohere.com/reference/generate
  2. AI21's jurassic model: https://docs.ai21.com/reference/j2-complete-ref

Sadly, Claude remains very restrictive.

I'd like to work on integrating it if its an option :)

shabie avatar Dec 04 '23 22:12 shabie

I would also vote for this; being able to run against a Bedrock back-end would be very helpful, even for a limited subset of models.

stevenbedrick avatar Feb 12 '24 18:02 stevenbedrick

Upvoting this! Very interested in integration for Cohere and generally Bedrock

vrtnis avatar Apr 02 '24 18:04 vrtnis

I'm also really hoping for Cohere Command-R integration.

Fuehnix avatar Apr 02 '24 18:04 Fuehnix