no action after thinking

[supperdsj]

i'v deplay hf.co/bytedance-research/UI-TARS-7B-gguf:latest with ollama in localhost,but get any action after thinking and screen shot

vlm provider: ollama
vlm base url: http://localhost:11434/api/generate
vlm model name: hf.co/bytedance-research/UI-TARS-7B-gguf:latest

[zakkor]

Hi, you need to use this for the base url: http://localhost:11434/v1

[zakkor]

that being said, it still doesn't work. seems totally braindead and just clicks a random spot over and over

[wrt-n]

Had the same issue and this worked for me:

Also getting the issue mentioned by zakkor with nulls in the click coordinates (tried official 16bit 2B version, as well as quants of 2B and 7B model)

[SebastianBoehler]

I am having the same issue with null coordinates...

[thepok]

i have the same proplems i tried the 7b and the 2b model. they seem to work fine in ollama, and i guess they produce tokens.

They take screenshots and print zeros to the chatlog

[naveenkasturi]

Seems to be same issue for me as well. Did any one find a way around it?

[DFin]

Can confirm the Null issue in coordinates. It works generating tokens with both GGUF models using ollama endpoint, but it constantly generates coordinates like this (Null, 0.xzy, Null, 0.xyz). Using the Mac desktop client.

[rolandgvc]

I'm having the same issues with the huggingface setup

[tomcharlesosman]

mac desktop client not working for me either.

[doubleLLL3]

I'm having the same issues with the huggingface setup我在使用 huggingface 设置时遇到了同样的问题

@rolandgvc Restart the App, and then fine.

And I'm also using the huggingface, is your url like this?
https://xxx.endpoints.huggingface.cloud/v1

need /v1 suffix

[DFin]

I am not having the problems with running via huggingface. Both UI-TARS-72B-SFT and UI-TARS-7B-SFT worked fine. The problem only happens using ollama.

[nivibilla]

I also get this when i use vllm 72b in bf16