app.py

import gradio as gr
import spaces
import torch


@spaces.GPU
def check_gpu():
    """Runs on ZeroGPU — proves the GPU is actually allocated to this call."""
    available = torch.cuda.is_available()
    name = torch.cuda.get_device_name(0) if available else "none"
    x = torch.tensor([1.0]).cuda()          # forces a real GPU op
    return {
        "cuda_available": available,
        "device": str(x.device),            # should be 'cuda:0'
        "gpu_name": name,
    }


with gr.Blocks() as demo:
    gr.Markdown("# knowledge-pipeline — GPU + API smoke test")
    out = gr.JSON(label="result")
    gr.Button("Check GPU").click(fn=check_gpu, inputs=None, outputs=out,
                                 api_name="check_gpu")

demo.launch()