app.py

import gradio as gr
import requests
import os

api_url = "https://api-inference.huggingface.co/models/llava-hf/llava-v1.6-mistral-7b-hf"
api_key = os.getenv("HF_API_KEY")
headers = {"Authorization": f"Bearer {api_key}"}

def generate_response(prompt):
    response = requests.post(api_url, headers=headers, json={"inputs": prompt})
    if response.status_code == 200:
        return response.json()[0]['generated_text']
    else:
        return "Error: " + response.json().get('error', 'Unknown error')

iface = gr.Interface(fn=generate_response, inputs="text", outputs="text")
iface.launch()