Add an OpenRouter provider #921
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
Oh I also need to add some tests. |
jhrozek
commented
Feb 4, 2025
|
This was my Cline config |
|
@danbarr when and if this is accepted, we'll probably want to document how to use OR with this provider. I noticed we suggest to use VLLM now in the docs for continue. |
|
Yeah vllm worked because both it and OpenRouter are an OpenAI-compatible API. Originally I thought we could use our existing /openai provider endpoint but that's where we ran into the LiteLLM automatic routing based on the model name. |
In order to properly support "muxing providers" like openrouter, we'll have to tell litellm (or in future a native implementation), what server do we want to proxy to. We were already doing that with Vllm, but since are about to do the same for OpenRouter, let's move the `_get_base_url` method to the base provider.
OpenRouter is a "muxing provider" which itself provides access to multiple models and providers. It speaks a dialect of the OpenAI protocol, but for our purposes, we can say it's OpenAI. There are some differences in handling the requests, though: 1) we need to know where to forward the request to, by default this is `https://openrouter.ai/api/v1`, this is done by setting the base_url parameter 2) we need to prefix the model with `openrouter/`. This is a lite-LLM-ism (see https://docs.litellm.ai/docs/providers/openrouter) which we'll be able to remove once we ditch litellm Initially I was considering just exposing the OpenAI provider on an additional route and handling the prefix based on the route, but I think having an explicit provider class is better as it allows us to handle any differences in OpenRouter dialect easily in the future. Related: #878
We can later alias it to openai if we decide to merge them.
jhrozek
force-pushed
the
cline_openrouter
branch
from
e0150f0 to
920d882
Compare
tests added |
aponcedeleonch
approved these changes
Feb 6, 2025
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
There are some differences in handling the requests, though:
https://openrouter.ai/api/v1, this is done by setting the base_url parameteropenrouter/. This is a lite-LLM-ism (see https://docs.litellm.ai/docs/providers/openrouter) which we'll be able to remove once we ditch litellmInitially I was considering just exposing the OpenAI provider on an
additional route and handling the prefix based on the route, but I think
having an explicit provider class is better as it allows us to handle
any differences in OpenRouter dialect easily in the future.
Initially I only tested this with Cline, using DeepSeek for Plan and Anthropic
for Act. I still need to test other assistants (Continue).
Related: #878