Add mistral as supported provider for AWS Bedrock

[jarib]

Mistral is now supported in AWS Bedrock.

$ aws bedrock list-foundation-models --region us-east-1 --query 'modelSummaries[*].modelId' | grep mistral
    "mistral.mistral-7b-instruct-v0:2",
    "mistral.mixtral-8x7b-instruct-v0:1"
$

Langchain has a duplicated list of valid models. I'd argue that they both can be removed. Langchain doesn't need to validate the models before passing them on to Bedrock.

[vercel]

The latest updates on your projects. Learn more about Vercel for Git ↗︎

Name	Status	Preview	Comments	Updated (UTC)
langchainjs-api-refs	✅ Ready (Inspect)	Visit Preview	💬 Add feedback	Mar 18, 2024 9:51pm
langchainjs-docs	✅ Ready (Inspect)	Visit Preview		Mar 18, 2024 9:51pm

[jacoblee93]

The issue is in the input/output parsing - I think there might be some other changes needed in the adapter?

[esoler-sage]

@jacoblee93 You are 💯 right! I created a PR against his fork (https://github.dev/jarib/langchainjs/pull/1) with the changes needed to implement it properly.
@jarib if you want to take a look it'll solve 4683 ;)

Thanks!

[jarib]

Ah, I missed that. I've merged @esoler-sage's fix, thanks!

[jacoblee93]

Thank you! FYI @efriis

[jarib]

Shouldn't this be added as separate tests? Claude support is pretty important too :D

[jacoblee93]

Well we've already got Claude-3 below, which most people should be using now.

[efriis]

Good PR!

A lot of these if statements / double validation will be removed in an upcoming refactor of bedrock on the python side, and @jacoblee93 will probably tap in to do something similar in js. I believe the reason for validation is that Bedrock models don't have a standard api for their inputs and arguments, so checking and throwing an error fails quickly instead of bedrock introducing a new model that doesn't have a supported input/args format, and producing an output that doesn't make sense.

[esoler-sage]

@efriis I was about to create 2 PR... token count on Bedrock Models and add different chat prompts to improve this line: https://github.com/langchain-ai/langchainjs/blob/main/libs/langchain-community/src/chat_models/bedrock/web.ts#L295.

For this second, probably it's better to wait until this refactoring is done?