OpenRouter/DeepSeek Model Support

[GratefulDave]

I'm trying to figure out how to add OpenRouter's deepseek/deepseek-chat model... Would rather pay tenths of a cent for my commit messages. :)

[Blarc]

Hey @GratefulDave,

there was a MR regarding OpenRouter (#267), but I noticed that it uses the same client as OpenAI. Could you maybe try to use OpenAI client and configure the url and token for OpenRouter?

[GratefulDave]

Yeah, I tried that:

[Blarc]

Hey @GratefulDave,

I created an accout for OpenRouter and tested it myself. I believe, you're using the wrong host url. Here's a a configuration that worked for me:

(Although the result is not good and contains prompt, as some models seems to do that.)

[GratefulDave]

You have a great product. Thanks for your prompt reply and sorry for my delated response.
I was using Phi4 and noticed the model responds with the prompt and sometimes too much data.

I keep getting Too many follow-up requests 21 even after I wait a few minutes.

On another note, I was trying to use a Ollama distilled R1 model, but when I refresh the models, it only shows Llama3 and Phi4.

Any suggestions?

[Blarc]

@GratefulDave Regarding the model responding with the prompt; I could maybe add an option to all? client configurations to remove the prompt. This is currently enabled only for HuggingFace provider, as I thought this was the only one the also return the prompt.

Regarding the too many requests, I can't really say much, since that depends entirely on the API you're calling.

Similarly, the models that are returned after the refresh depends on the API, however, the models combo box is editable and you can manually add the models.