app.py

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

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

if not HF_TOKEN:
    print("ERROR: HF_CV_ROBOT_TOKEN environment variable not set.")

# -------------------------------
# Main Processing Function
# -------------------------------
def process(payload: dict):
    try:
        if not HF_TOKEN:
            return {"error": "Hugging Face token missing. Check environment variable."}

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

        # 正確格式：直接把 base64 放入 image_data
        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}"},
            json=data,
            timeout=60
        )

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

        try:
            vlm_text = resp.json()["choices"][0]["message"]["content"][0]["text"]
        except Exception as e:
            return {"error": f"Failed to parse VLM response: {e}, Response={resp.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)