app.py

import gradio as gr
import requests
import json
import os
from google.oauth2 import service_account
from google.auth.transport.requests import Request

# Hardcoded values
PROJECT_ID = "white-dispatch-460720-h7"  # Using project ID instead of project number
ENDPOINT_ID = "5591670836659486720"
LOCATION = "us-central1"

def query_vertex(user_input):
    # Try to get service account key from environment variable
    service_account_key = os.environ.get("GOOGLE_SERVICE_ACCOUNT_KEY")
    
    if not service_account_key:
        return "Error: GOOGLE_SERVICE_ACCOUNT_KEY environment variable not set. Please configure service account authentication in Hugging Face Spaces settings."
    
    try:
        # Parse the service account key and get credentials
        import json as json_lib
        service_account_info = json_lib.loads(service_account_key)
        credentials = service_account.Credentials.from_service_account_info(
            service_account_info,
            scopes=['https://www.googleapis.com/auth/cloud-platform']
        )
        credentials.refresh(Request())
        token = credentials.token
    except Exception as e:
        return f"Error authenticating with service account: {str(e)}"
    
    url = f"https://{ENDPOINT_ID}.{LOCATION}-104808504044.prediction.vertexai.goog/v1/projects/{PROJECT_ID}/locations/{LOCATION}/endpoints/{ENDPOINT_ID}:predict"

    payload = {
        "instances": [
            {
                "@requestFormat": "chatCompletions",
                "messages": [
                    {"role": "user", "content": user_input}
                ],
                "max_tokens": 200
            }
        ]
    }

    headers = {
        "Authorization": f"Bearer {token}",
        "Content-Type": "application/json"
    }

    response = requests.post(url, headers=headers, data=json.dumps(payload))
    try:
        result = response.json()
        # Extract the content from the response
        if 'predictions' in result:
            predictions = result['predictions']
            if 'choices' in predictions and len(predictions['choices']) > 0:
                choice = predictions['choices'][0]
                if 'message' in choice and 'content' in choice['message']:
                    return choice['message']['content']
        return f"Unexpected response format: {result}"
    except Exception as e:
        return f"Error: {response.text} - {str(e)}"

# Gradio interface
demo = gr.Interface(
    fn=query_vertex,
    inputs=gr.Textbox(lines=4, placeholder="Enter your product description..."),
    outputs=gr.Textbox(lines=15, placeholder="HTS classification will appear here..."),
    title="Atlas HTS Classification (via Vertex AI)",
    description="Enter a product description and Atlas (LLaMA-3.3-70B fine-tuned) will return the HTS classification."
)

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