app.py

import gradio as gr
import json
import base64
from io import BytesIO
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:
        image_b64 = payload["image_b64"]
        robot_id = payload.get("robot_id", "unknown")

        # Base64 → Bytes
        img_bytes = base64.b64decode(image_b64)

        # multipart file
        files = {
            "file": ("image.jpg", BytesIO(img_bytes), "image/jpeg")
        }

        # Router 要求的 payload 格式（不含 image_data）
        data = {
            "model": MODEL,
            "messages": [
                {
                    "role": "user",
                    "content": [
                        {"type": "text", "text": "Describe this image in detail."},
                        {"type": "file", "file": "image.jpg"}
                    ]
                }
            ]
        }

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

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

        out = resp.json()
        txt = out["choices"][0]["message"]["content"][0]["text"]

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

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

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

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