[Bug]: The Anthropic provider does not support vision
Is there an existing issue for the same bug?
- [X] I have checked the troubleshooting document at https://docs.all-hands.dev/modules/usage/troubleshooting
- [X] I have checked the existing issues.
Describe the bug
The current model ID's used are incorrect and do not support vision.
Current OpenHands version
0.9.2
Installation and Configuration
N/A
Model and Agent
Anthropic models
Operating System
osx
Reproduction Steps
- Select Anthropic provider
- A model that is supposed to support vision
- Upload images
- Try to send
Logs, Errors, Screenshots, and Additional Context
No response
https://github.com/All-Hands-AI/OpenHands/pull/3773#issuecomment-2338920671
I'll test for openAI in a bit. FWIW I am not sure FE should change this, it feels like the right thing to try as best as we can to standardize with provider. More like, this pattern could be added to liteLLM's list, since it supports it, it just doesn't know it has vision if we write it this way, but it does know if we write it without "anthropic/"... hah. Maybe we can fix it somehow first, though, pending theirs.
Looking at the list of prices, this would be broken for OpenAI as well: https://github.com/BerriAI/litellm/blob/main/model_prices_and_context_window.json
All OAI models do not have a provider/ prefix.
I think it would make sense to request the additions within the litellm repo to standardize and prevent the need to maintaining a mapping within this repo.
Created an issue here: https://github.com/BerriAI/litellm/issues/5608
I've made a pull request in litellm that will resolve this: https://github.com/BerriAI/litellm/pull/5688
@ColeMurray That is awesome! Thanks a lot
That's really super, hopefully it'll get the ✅ soon. Thanks @ColeMurray !