app.py

import gradio as gr
import json
import base64
from PIL import Image
import io
import requests
import os

HF_TOKEN = os.getenv("HF_CV_ROBOT_TOKEN")
MODEL = "Qwen/Qwen3-VL-32B-Instruct"

# -------------------------------
# 主處理函數
# -------------------------------
def process(payload: dict):
    try:
        robot_id = payload.get("robot_id", "unknown")
        image_b64 = payload["image_b64"]

        # Base64 解碼成圖片，用 PIL 開啟
        img_bytes = base64.b64decode(image_b64)
        img = Image.open(io.BytesIO(img_bytes)).convert("RGB")

        # Router API payload
        headers = {"Authorization": f"Bearer {HF_TOKEN}"}
        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=headers,
            json=data,
            timeout=60
        )

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

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

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

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

# -------------------------------
# Gradio MCP Server
# -------------------------------
demo = gr.Interface(
    fn=process,
    inputs=gr.JSON(label="Input Payload (Dict format)"),
    outputs=gr.JSON(label="Reply to Jetson"),
    api_name="predict"
)

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