app.py

import gradio as gr
import json
import base64
import requests
import os

HF_TOKEN = os.environ.get("HF_CV_ROBOT_TOKEN")
MODEL = "Qwen/Qwen2.5-VL-7B-Instruct"

def process(payload: dict):
    try:
        if not HF_TOKEN:
            return {"error": "Missing token"}

        robot_id = payload.get("robot_id", "unknown")
        image_b64 = payload["image_b64"]

        data = {
            "model": MODEL,
            "messages": [
                {
                    "role": "user",
                    "content": [
                        {"type": "text", "text": "Describe this image in detail."},
                        {
                            "type": "image_data",
                            "image_data": {"b64": image_b64}
                        }
                    ]
                }
            ]
        }

        resp = requests.post(
            "https://router.huggingface.co/v1/chat/completions",
            headers={
                "Authorization": f"Bearer {HF_TOKEN}",
                "Content-Type": "application/json"
            },
            data=json.dumps(data),
            timeout=60
        )

        if resp.status_code != 200:
            return {"error": f"VLM API error: {resp.status_code}, {resp.text}"}

        try:
            content = resp.json()["choices"][0]["message"]["content"]
            first = content[0]["text"]
        except:
            return {"error": f"Bad response: {resp.text}"}

        return {
            "received": True,
            "robot_id": robot_id,
            "vllm_analysis": first
        }

    except Exception as e:
        return {"error": str(e)}

demo = gr.Interface(
    fn=process,
    inputs=gr.JSON(label="Input Payload"),
    outputs=gr.JSON(label="Reply"),
    api_name="predict"
)

if __name__ == "__main__":
    demo.launch(mcp_server=True)