Spaces:

---

moonshotai

/

Kimi-VL-A3B-Thinking

like 88

Running on Zero

Hi @teowu Looks like the Space is using ZeroGPU. Didn't it work?

We are facing an error

    yield from gen_fn(prompt, *args, **kwargs)
  File "/usr/local/lib/python3.10/site-packages/spaces/zero/wrappers.py", line 353, in gradio_handler
    raise error("ZeroGPU worker error", "GPU task aborted")
gradio.exceptions.Error: 'GPU task aborted' 

this repeatedly
(while the model is generating text, and then fails)

@teowu The error is raised when the function decorated with @spaces.GPU takes longer than the specified duration parameter. The default value of it is 60 seconds, so you might want to adjust it. https://huggingface.co/spaces/moonshotai/Kimi-VL-A3B-Thinking/blob/main/app.py#L93

You can find more info about ZeroGPU in this page: https://huggingface.co/zero-gpu-explorers

Dear hysts,

Thank you so much.
As Kimi-VL-Thinking is a long-CoT thinking model (generates long long long), 60s might be not enough for many inference scenarios.
I have changed it to 180s.

Best
Haoning