No Ollama provider

[awaescher]

I just downloaded the installer for 0.0.2 on Windows.
In issue #11 I saw that there should be a provider for Ollama but it's missing for me.
Any idea how to overcome this?

[6xdd]

Version 0.0.1 supports Olllam:

Ollam was canceled in version 0.0.2: Why refer to https://github.com/bytedance/UI-TARS-desktop#%EF%B8%8F-important-announcement-gguf-model-performance

[awaescher]

This would be really sad! Thanks for the feedback.
But I doubt that this can be the reason as with 0.0.1 the Ollama provider is missing, too.

[RoversX]

At least let us give it a try 🤣

[DefamationStation]

Version 0.0.1 supports Olllam:

Ollam was canceled in version 0.0.2: Why refer to https://github.com/bytedance/UI-TARS-desktop#%EF%B8%8F-important-announcement-gguf-model-performance

This is not true. 0.0.1 doesn't have Ollama either.

[timothycarambat]

Just use the baseURL (http://localhost:11434/v1) for the ollama instance with the model name in ollama as the model name. For key just type random stuff. Its literally just an API endpoint - it uses the openai package in the backend so any OpenAI compatible service will work (Ollama, lmstudio, etc etc)

[JjjFangg]

Hello, we have observed that the performance of the Ollama + GGUF approach is currently inferior to cloud deployment. At this stage, we recommend using Hugging Face inference endpoints for optimal results. We will consider uploading to Ollama once local deployment achieves performance parity with online inference.

[awaescher]

I understand. But why actively disabling a feature because of performance concerns? People might have an killer AI machine with actually good performance.

Even more important: For prototyping, it would be amazingly helpful to use Ollama to run 100% locally. Think about building an early prototype in an enterprise scenario where going the full way to use external services and share data with them might need weeks of discussions before even being able to try it out.