app.py

import gradio as gr
import openai

# We assume the Hugging Face Inference API is OpenAI-compatible.
# For each LLM, set openai.api_base to the model's endpoint and then call openai.ChatCompletion.

# Your Hugging Face API key
HF_API_KEY = "hf_1234"

# Model endpoints on Hugging Face
MODEL_ENDPOINTS = {
    "Qwen2.5-72B-Instruct": "https://api-inference.huggingface.co/models/Qwen/Qwen2.5-72B-Instruct",
    "Llama3.3-70B-Instruct": "https://api-inference.huggingface.co/models/meta-llama/Llama-3.3-70B-Instruct",
    "Qwen2.5-Coder-32B-Instruct": "https://api-inference.huggingface.co/models/Qwen/Qwen2.5-Coder-32B-Instruct",
}

# Query a specific model using OpenAI-compatible ChatCompletion
def query_model(prompt, model_endpoint):
    openai.api_key = HF_API_KEY
    openai.api_base = model_endpoint
    response = openai.ChatCompletion.create(
        model="any-model-placeholder",  # placeholder name, not actually used by the HF endpoint
        messages=[{"role": "user", "content": prompt}],
        max_tokens=512,
        temperature=0.7
    )
    return response.choices[0].message["content"]

def chat_with_models(user_input, history):
    # Let each model provide its own contribution
    responses = []
    for model_name, endpoint in MODEL_ENDPOINTS.items():
        model_response = query_model(user_input, endpoint)
        responses.append(f"**{model_name}**: {model_response}")

    # Combine all responses in a single answer
    combined_answer = "\n\n".join(responses)
    history.append((user_input, combined_answer))
    return history, history

with gr.Blocks() as demo:
    gr.Markdown("# Multi-LLM Chatbot using Hugging Face Inference API")

    chatbot = gr.Chatbot()
    msg = gr.Textbox(label="Your Message")
    clear = gr.Button("Clear")

    def clear_chat():
        return [], []

    msg.submit(chat_with_models, [msg, chatbot], [chatbot, chatbot])
    clear.click(fn=clear_chat, outputs=[chatbot, chatbot])

demo.launch()