UI-TARS-desktop icon indicating copy to clipboard operation
UI-TARS-desktop copied to clipboard

How to configure the UI-TARS-Model?

Open zengxi opened this issue 9 months ago • 5 comments

Image

I don't get it, how to set the UI-TARS-Model params?

zengxi avatar Mar 22 '25 06:03 zengxi

I’m encountering the same issue. So, which VLM does ui-tars-desktop actually use? Does it use a cloud-based model by default?

treblam avatar Mar 22 '25 09:03 treblam

@zengxi

I'm very sorry that the latest release seems to have caused some confusion for you. Agent TARS is our new Desktop App, an exploration for general agents. The introduction document can be found here: https://agent-tars.com/2025/03/18/announcing-agent-tars-app

If you are looking for UI-TARS Desktop, the latest version is 0.0.7, which you can find here: https://github.com/bytedance/UI-TARS-desktop/releases/tag/v0.0.7

Can you please tell me if you want the VLM+GUI Agent-driven UI-TARS Desktop or the Agent TARS Desktop?

We will give more detailed explanations in the official documentation later. Thank you for your feedback!

ulivz avatar Mar 22 '25 12:03 ulivz

I meet the same question.I am confused about how to set the local deployment model on this app.I try to set base_url="http://127.0.0.1:8000/v1" and api_key="empty" in the setting page(pic.1),but I get the 404 bad request (pic.2), and I think the reason is the argument "repetition_penalty" unsupported to vLLM OPENAI API server.

Image

Image

ginreedcho avatar Mar 24 '25 07:03 ginreedcho

@zengxi

I'm very sorry that the latest release seems to have caused some confusion for you. Agent TARS is our new Desktop App, an exploration for general agents. The introduction document can be found here: https://agent-tars.com/2025/03/18/announcing-agent-tars-app

If you are looking for UI-TARS Desktop, the latest version is 0.0.7, which you can find here: https://github.com/bytedance/UI-TARS-desktop/releases/tag/v0.0.7

Can you please tell me if you want the VLM+GUI Agent-driven UI-TARS Desktop or the Agent TARS Desktop?

We will give more detailed explanations in the official documentation later. Thank you for your feedback!

I’m really confused by these two concepts. I understand that the Agent TARS Desktop is a client based on the UI-TARS-Model. However, in the settings UI, the only available option is GPT-4o(and Azure), not the UI-TARS-Model, which makes me even more confused.

zengxi avatar Mar 28 '25 11:03 zengxi

@zengxi

I'm very sorry that the latest release seems to have caused some confusion for you. Agent TARS is our new Desktop App, an exploration for general agents. The introduction document can be found here: https://agent-tars.com/2025/03/18/announcing-agent-tars-app

If you are looking for UI-TARS Desktop, the latest version is 0.0.7, which you can find here: https://github.com/bytedance/UI-TARS-desktop/releases/tag/v0.0.7

Can you please tell me if you want the VLM+GUI Agent-driven UI-TARS Desktop or the Agent TARS Desktop?

We will give more detailed explanations in the official documentation later. Thank you for your feedback!

so, does Agent TARS Desktop use the UI-TARS Model under the hood, if it is, where is it deployed?

treblam avatar Mar 29 '25 09:03 treblam