import gradio as gr
from openai import OpenAI
import os

# Retrieve the access token from the environment variable
ACCESS_TOKEN = os.getenv("HF_TOKEN")
print("Access token loaded.")

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

def respond(
    message,
    history: list[tuple[str, str]],
    system_message,
    max_tokens,
    temperature,
    top_p,
    frequency_penalty,
    seed,
    custom_model,
    selected_model
):
    """
    Handles the chatbot response generation.
    """
    print(f"Received message: {message}")
    print(f"History: {history}")
    print(f"System message: {system_message}")
    print(f"Max tokens: {max_tokens}, Temperature: {temperature}, Top-P: {top_p}")
    print(f"Frequency Penalty: {frequency_penalty}, Seed: {seed}")
    print(f"Custom model: {custom_model}")
    print(f"Selected model: {selected_model}")

    # Convert seed to None if -1 (meaning random)
    if seed == -1:
        seed = None

    # Construct the messages array required by the API
    messages = [{"role": "system", "content": system_message}]

    # 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})
            print(f"Added user message to context: {user_part}")
        if assistant_part:
            messages.append({"role": "assistant", "content": assistant_part})
            print(f"Added assistant message to context: {assistant_part}")

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

    # Determine which model to use
    model_to_use = (
        custom_model.strip()
        if custom_model.strip() != ""
        else selected_model.strip()
    )
    print(f"Model selected for inference: {model_to_use}")

    # Start with an empty string to build the response as tokens stream in
    response = ""
    print("Sending request to OpenAI API.")

    # Make the streaming request to the HF Inference API via openai-like client
    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,
    ):
        # Extract the token text from the response chunk
        token_text = message_chunk.choices[0].delta.content
        print(f"Received token: {token_text}")
        response += token_text
        yield response

    print("Completed response generation.")

# Predefined list of placeholder models for the Featured Models accordion
models_list = [
    "meta-llama/Llama-3.3-70B-Instruct",
    "bigscience/bloom-7b1",
    "EleutherAI/gpt-neo-2.7B",
    "OpenAssistant/oasst-sft-4-pythia-12b-epoch-3.5",
    "HuggingFace/distilgpt2",
]

# Function to filter models based on search input
def filter_models(search_term):
    filtered_models = [m for m in models_list if search_term.lower() in m.lower()]
    return gr.update(choices=filtered_models)

# Create a Chatbot component with a specified height
chatbot = gr.Chatbot(height=600)
print("Chatbot interface created.")

# Create the Gradio ChatInterface
# Added "Featured Models" accordion and integrated filtering
demo = gr.Interface(
    fn=respond,
    inputs=[
        gr.Textbox(value="", label="System message"),
        gr.Slider(minimum=1, maximum=4096, 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"),
        gr.Slider(
            minimum=-2.0,
            maximum=2.0,
            value=0.0,
            step=0.1,
            label="Frequency Penalty"
        ),
        gr.Slider(
            minimum=-1,
            maximum=65535,  # Arbitrary upper limit for demonstration
            value=-1,
            step=1,
            label="Seed (-1 for random)"
        ),
        gr.Textbox(
            value="",
            label="Custom Model",
            info="(Optional) Provide a custom Hugging Face model path. This will override the default model if not empty.",
        ),
        # Add Featured Models accordion
        gr.Accordion("Featured Models", open=True, children=[
            gr.Textbox(label="Filter Models", placeholder="Search for a featured model...", lines=1).change(
                filter_models, inputs=["value"], outputs="choices"
            ),
            gr.Radio(
                label="Select a featured model",
                value="meta-llama/Llama-3.3-70B-Instruct",
                choices=models_list,
                elem_id="model-radio",
            )
        ]),
    ],
    outputs=gr.Chatbot(height=600),
    theme="Nymbo/Nymbo_Theme",
)

print("Gradio interface initialized.")

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