app.py

import os
import base64
import gradio as gr
from huggingface_hub import upload_file, InferenceClient
import json

# --- Config ---
HF_DATASET_REPO = "OppaAI/Robot_MCP"
HF_VLM_MODEL = "Qwen/Qwen2.5-VL-7B-Instruct" 

# --- Helper Functions ---
def save_and_upload_image(image_b64, hf_token):
    """Save image to /tmp and upload to HF dataset."""
    image_bytes = base64.b64decode(image_b64)
    local_tmp_path = "/tmp/tmp.jpg"
    with open(local_tmp_path, "wb") as f:
        f.write(image_bytes)

    path_in_repo = "images/tmp.jpg"
    upload_file(
        path_or_fileobj=local_tmp_path,
        path_in_repo=path_in_repo,
        repo_id=HF_DATASET_REPO,
        token=hf_token,
        repo_type="dataset"
    )

    hf_image_url = f"https://huggingface.co/datasets/{HF_DATASET_REPO}/resolve/main/{path_in_repo}"
    return local_tmp_path, hf_image_url, path_in_repo, len(image_bytes)


# --- Main MCP function ---
def process_and_describe(payload: dict):
    try:
        hf_token = payload.get("hf_token")
        if not hf_token:
            return {"error": "HF token not provided in payload."}

        robot_id = payload.get("robot_id", "unknown")
        image_b64 = payload.get("image_b64")
        if not image_b64:
            return {"error": "No image provided."}

        # Save & upload
        local_tmp_path, hf_url, path_in_repo, size_bytes = save_and_upload_image(image_b64, hf_token)

        # Init HF client
        hf_client = InferenceClient(token=hf_token)

        # System prompt: describe + suggest action
        system_prompt = """
        You are a helpful robot assistant.
        1. Describe the image in detail.
        2. Suggest what the robot should do next based on what it sees:
           - Human figure → say 'Hi'.
           - Ball → move towards it.
           - Obstacles → stop or avoid.
           - Red button → press it.
        Always respond in JSON:
        {"description": "...", "action": {"move": "...", "interact": "..."}}
        """

        messages_payload = [
            {"role": "system", "content": system_prompt},
            {"role": "user", "content": [
                {"type": "text", "text": "Here is an image."},
                {"type": "image_url", "image_url": {"url": f"data:image/jpeg;base64,{image_b64}"}}
            ]}
        ]

        # Call VLM
        chat_completion = hf_client.chat.completions.create(
            model=HF_VLM_MODEL,
            messages=messages_payload,
            max_tokens=300
        )

        # Robustly extract text
        try:
            vlm_text = chat_completion.choices[0].message.content.strip()
        except Exception:
            # fallback if structure is different
            vlm_text = str(chat_completion)

        # Attempt to parse JSON from VLM
        action_data = {}
        try:
            action_data = json.loads(vlm_text)
        except Exception:
            # If VLM didn't return valid JSON, wrap text as description
            action_data = {"description": vlm_text, "action": {"move": "unknown", "interact": "unknown"}}

        return {
            "saved_to_hf_hub": True,
            "repo_id": HF_DATASET_REPO,
            "path_in_repo": path_in_repo,
            "image_url": hf_url,
            "file_size_bytes": size_bytes,
            "robot_id": robot_id,
            "vlm_response": vlm_text,
            "vlm_action": action_data.get("action", {}),
            "vlm_description": action_data.get("description", "")
        }

    except Exception as e:
        return {"error": f"An API error occurred: {str(e)}"}


# --- Gradio MCP Interface ---
demo = gr.Interface(
    fn=process_and_describe,
    inputs=gr.JSON(label="Input Payload (Dict format with 'image_b64')"),
    outputs=gr.JSON(label="Reply to Jetson"),
    api_name="predict"
)

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