fix: corrected bedrock model list

[zhaoninge]

litellm.model_list and litellm.model_cost are not the correct list of model ids. Such as, bedrock/us-west-2/mistral.mixtral-8x7b-instruct-v0:1 should be bedrock/mistral.mixtral-8x7b-instruct-v0:, without the rengion name. Now fix for AWS bedrock model, using the official sdk to get the model id in real time. Before using it, you need to configure AWS_REGION_NAME AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY.

[li-boxuan]

Question: any reason this fix isn't targeting litellm repo?

[zhaoninge]

Question: any reason this fix isn't targeting litellm repo?问题：此修复程序不针对 litellm 存储库的任何原因是什么？

litellm repo does not provide an API to query all provider models. litellm repo also does not have an interface that query the list of available models from each provider.

[rbren]

TBH I'm thinking we should redo this logic--there are way too many model names, and it's really easy to pick the wrong one (e.g. azure/gpt-4-turbo instead of gpt-4-turbo)

I generally agree we should depend on LiteLLM upstream and ask them to fix. I do like having the AWS settings though

[zhaoninge]

I agree with everyone. I have sent a request to litellm to provide an API that can query available models in real time. (https://github.com/BerriAI/litellm/issues/361, so many wishes) However, until litellm responds, I think this fix can be used as a temporary solution, otherwise we can't use the AWS bedrock model. Sorry, I can only partially solve the model id problem of AWS bedrock with my current energy. If we agree, we can work together to solve the problems of other providers.

[rbren]

I'm good with this one as-is. But I'm going to open a ticket to simplify the dropdown here, so we only have a handful of options to pick from, maybe separated by provider.

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 33.33333% with 6 lines in your changes are missing coverage. Please review.

:exclamation: No coverage uploaded for pull request base (main@356caf0). Click here to learn what that means.

Files	Patch %	Lines
opendevin/server/listen.py	0.00%	6 Missing :warning:

:exclamation: Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #1513   +/-   ##
=======================================
  Coverage        ?   62.56%           
=======================================
  Files           ?       92           
  Lines           ?     3745           
  Branches        ?        0           
=======================================
  Hits            ?     2343           
  Misses          ?     1402           
  Partials        ?        0           

:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.