app.py

import gradio as gr
from huggingface_hub import InferenceClient

# Function to create InferenceClient dynamically based on model selection
def get_client(model_name):
    return InferenceClient(model_name)

def respond(
    message,
    history: list[tuple[str, str]],
    max_tokens,
    temperature,
    top_p,
    model_name,  # Added model_name to the function arguments
):
    # Statically defined system message
    system_message = "You are a friendly Chatbot."
    
    # Create client for the selected model
    client = get_client(model_name)
    
    # Check if the model is one of the problematic models
    if model_name in ["indonlp/cendol-mt5-small-inst", "indonlp/cendol-mt5-small-chat"]:
        # For these models, we simply concatenate the conversation into a single string
        history_str = ""
        for user_msg, assistant_msg in history:
            if user_msg:
                history_str += f"{user_msg}\n"
            if assistant_msg:
                history_str += f"{assistant_msg}\n"
        
        # Add the latest user message
        history_str += f"{message}\n"
        
        # Pass the entire conversation history as a plain text prompt
        response = client.text_generation(
            history_str,  # Single string as input
            max_new_tokens=max_tokens,
            temperature=temperature,
            top_p=top_p
        )
        
        # Since response is a string, return it directly
        full_response = response
    else:
        # For other models, we use a structured format with roles
        messages = [{"role": "system", "content": system_message}]
        for val in history:
            if val[0]:
                messages.append({"role": "user", "content": val[0]})
            if val[1]:
                messages.append({"role": "assistant", "content": val[1]})
        
        # Add the latest user message
        messages.append({"role": "user", "content": message})
        
        # Make the request
        response = client.chat_completion(
            messages,
            max_tokens=max_tokens,
            temperature=temperature,
            top_p=top_p,
            stream=False
        )
        
        # Extract the full response for chat models
        full_response = response.choices[0].message["content"]
    
    return full_response


# Gradio ChatInterface setup with static system message and no Textbox for system message
demo = gr.ChatInterface(
    respond,
    additional_inputs=[
        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
        gr.Slider(
            minimum=0.1,
            maximum=1.0,
            value=0.95, step=0.05, label="Top-p (nucleus sampling)"
        ),
        # Dropdown to select model
        gr.Dropdown(
            choices=[
                "meta-llama/Meta-Llama-3-8B-Instruct",
                "mistralai/Mistral-7B-Instruct-v0.3",
                "HuggingFaceH4/zephyr-7b-beta"
            ],
            value="meta-llama/Meta-Llama-3-8B-Instruct",
            label="Choose Model"
        ),
    ],
)

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