Support custom model names

pymike00 / tinychat

🔮 TinyChat is a lightweight Desktop client for modern Language Models designed for straightforward comprehension. Supports OpenAI, Anthropic, Meta, Mistral, Google and Cohere APIs.

MIT License

38 stars 7 forks source link

Support custom model names #7

Open johnd0e opened 4 months ago

johnd0e commented 4 months ago

Currently, there are two distinct OpenAI options in the providers list: "gpt-3.5 turbo" and "gpt-4 turbo". It would be more efficient to consolidate these into a single item, with the capability to switch between different models. Naturally, the ability to assign custom model names is necessary.

The same improvements should also be made for Claude, LLama, and Mistral.

pymike00 commented 4 months ago

Hi there,

Yes, duplicating the handler instantiation like that is probably not the cleanest of solutions, I will see what can be done without compromising simplicity too much while refactoring.

If you had any initial thoughts feel free to share them here!

johnd0e commented 4 months ago

If you had any initial thoughts feel free to share them here!

I see several alternatives

Common models list, with possibility to add there any model by any provider. Additionally there could be "Filter by provider" , drop-down control.
2 separate lists in dropdown controls: Provider / Model, with possibility to add any custom model. It would be good, if we could add custom providers as well. By "custom provider" I mean any OpenAI-, or Google Gemini-compatible service, which can be defined with its URL (#6)