app.py

import gradio as gr
from openai import OpenAI
import os

ACCESS_TOKEN = os.getenv("HF_TOKEN")

def show_loading_status(msg):
    
    try:
        gr.toast(msg)
    except:
        pass
    print(msg)

show_loading_status("Access token loaded.")

# Initialize the Hugging Face Inference-based OpenAI client
client = OpenAI(
    base_url="https://api-inference.huggingface.co/v1/",
    api_key=ACCESS_TOKEN,
)
show_loading_status("OpenAI client initialized.")


def respond(
    message,
    history: list[tuple[str, str]],
    system_message,
    max_tokens,
    temperature,
    top_p,
    frequency_penalty,
    seed,
    custom_model
):
    show_loading_status(f"Received message: {message}")
    show_loading_status(f"History: {history}")
    show_loading_status(f"System message: {system_message}")
    show_loading_status(f"Max tokens: {max_tokens}, Temperature: {temperature}, Top-P: {top_p}")
    show_loading_status(f"Frequency Penalty: {frequency_penalty}, Seed: {seed}")
    show_loading_status(f"Selected model (custom_model): {custom_model}")

    # Convert seed to None if -1 (meaning random)
    seed = seed if seed != -1 else random.randint(1, 1000000000),

    messages = [{"role": "system", "content": system_message}]
    show_loading_status("Initial messages array constructed.")

    # Add conversation history to the context
    for val in history:
        user_part = val[0]
        assistant_part = val[1]
        if user_part:
            messages.append({"role": "user", "content": user_part})
            show_loading_status(f"Added user message to context: {user_part}")
        if assistant_part:
            messages.append({"role": "assistant", "content": assistant_part})
            show_loading_status(f"Added assistant message to context: {assistant_part}")

    # Append the latest user message
    messages.append({"role": "user", "content": message})
    show_loading_status("Latest user message appended.")

    # If user provided a model, use that; otherwise, fall back to a default
    model_to_use = custom_model.strip() if custom_model.strip() != "" else "meta-llama/Llama-3.3-70B-Instruct"
    show_loading_status(f"Model selected for inference: {model_to_use}")

    response_text = ""
    show_loading_status("Sending request to OpenAI API.")

    try:
        for message_chunk in client.chat.completions.create(
            model=model_to_use,
            max_tokens=max_tokens,
            stream=True,
            temperature=temperature,
            top_p=top_p,
            frequency_penalty=frequency_penalty,
            seed=seed,
            messages=messages,
        ):
            # Each chunk is a piece of the streaming text
            token_text = message_chunk.choices[0].delta.content
            show_loading_status(f"Received token: {token_text}")
            response_text += token_text
            yield response_text

        show_loading_status("Completed response generation.")

    except Exception as e:
        show_loading_status("Error encountered during completion streaming.")
        raise gr.Error(f"An unexpected error occurred: {str(e)}")


# GRADIO UI

chatbot = gr.Chatbot(
    height=600, 
    show_copy_button=True, 
    placeholder="Select a model and begin chatting", 
    likeable=True, 
    layout="panel"
)
show_loading_status("Chatbot interface created.")

system_message_box = gr.Textbox(
    value="", 
    placeholder="You are a helpful assistant.", 
    label="System Prompt"
)

max_tokens_slider = gr.Slider(
    minimum=1,
    maximum=4096,
    value=512,
    step=1,
    label="Max new tokens"
)
temperature_slider = gr.Slider(
    minimum=0.1,
    maximum=4.0,
    value=0.7,
    step=0.1,
    label="Temperature"
)
top_p_slider = gr.Slider(
    minimum=0.1,
    maximum=1.0,
    value=0.95,
    step=0.05,
    label="Top-P"
)
frequency_penalty_slider = gr.Slider(
    minimum=-2.0,
    maximum=2.0,
    value=0.0,
    step=0.1,
    label="Frequency Penalty"
)
seed_slider = gr.Slider(
    minimum=-1,
    maximum=1000000000,
    value=-1,
    step=1,
    label="Seed (-1 for random)"
)

custom_model_box = gr.Textbox(
    value="",
    label="Custom Model",
    info="(Optional) Provide a custom Hugging Face model path. Supports Warm and Cold models.",
    placeholder="meta-llama/Llama-3.3-70B-Instruct"
)

def set_custom_model_from_radio(selected):
    show_loading_status(f"Featured model selected: {selected}")
    return selected

demo = gr.ChatInterface(
    fn=respond,
    additional_inputs=[
        system_message_box,
        max_tokens_slider,
        temperature_slider,
        top_p_slider,
        frequency_penalty_slider,
        seed_slider,
        custom_model_box,
    ],
    fill_height=True,
    chatbot=chatbot,
    theme="Nymbo/Nymbo_Theme",
)
show_loading_status("ChatInterface object created.")

with demo:
    with gr.Accordion("Model Selection", open=False):
        model_search_box = gr.Textbox(
            label="Filter Models",
            placeholder="Search for a featured model...",
            lines=1
        )
        show_loading_status("Model search box created.")

        models_list = [
            "meta-llama/Llama-3.3-70B-Instruct",
            "meta-llama/Llama-3.2-3B-Instruct",
            "meta-llama/Llama-3.2-1B-Instruct",
            "meta-llama/Llama-3.1-8B-Instruct",
            "NousResearch/Hermes-3-Llama-3.1-8B",
            "mistralai/Mistral-Nemo-Instruct-2407",
            "mistralai/Mixtral-8x7B-Instruct-v0.1",
            "mistralai/Mistral-7B-Instruct-v0.3",
            "Qwen/Qwen2.5-72B-Instruct",
            "Qwen/QwQ-32B-Preview",
            "HuggingFaceTB/SmolLM2-1.7B-Instruct",
            "microsoft/Phi-3.5-mini-instruct",
        ]
        show_loading_status("Models list initialized.")

        featured_model_radio = gr.Radio(
            label="Select a model below",
            choices=models_list,
            value="meta-llama/Llama-3.3-70B-Instruct",
            interactive=True
        )
        show_loading_status("Featured models radio button created.")

        def filter_models(search_term):
            show_loading_status(f"Filtering models with search term: {search_term}")
            filtered = [m for m in models_list if search_term.lower() in m.lower()]
            show_loading_status(f"Filtered models: {filtered}")
            return gr.update(choices=filtered)

        model_search_box.change(
            fn=filter_models,
            inputs=model_search_box,
            outputs=featured_model_radio
        )
        show_loading_status("Model search box change event linked.")

        featured_model_radio.change(
            fn=set_custom_model_from_radio,
            inputs=featured_model_radio,
            outputs=custom_model_box
        )
        show_loading_status("Featured model radio button change event linked.")

show_loading_status("Gradio interface initialized.")

if __name__ == "__main__":
    show_loading_status("Launching the demo application.")
    demo.launch()