import gradio as gr
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

# Load the model and tokenizer locally in bfloat16 precision
model_name = "vietdata/llama32_1b_pub"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype=torch.bfloat16,  # Load model in bfloat16 precision
    device_map="auto" if torch.cuda.is_available() else None,  # Automatically map to available devices
)

# Define the respond function
def respond(
    message,
    history: list[tuple[str, str]],
    system_message,
    max_tokens,
    temperature,
    top_p,
):
    from transformers import TextGenerationPipeline

    # Build the conversation context
    prompt = system_message + "\n"
    for user_msg, bot_msg in history:
        if user_msg:
            prompt += f"User: {user_msg}\n"
        if bot_msg:
            prompt += f"Bot: {bot_msg}\n"
    prompt += f"User: {message}\nBot:"

    # Set up a text generation pipeline
    pipe = TextGenerationPipeline(
        model=model, 
        tokenizer=tokenizer, 
        device=torch.cuda.current_device() if torch.cuda.is_available() else -1
    )

    # Generate the response
    response = pipe(
        prompt,
        max_length=len(prompt) + max_tokens,
        temperature=temperature,
        top_p=top_p,
        pad_token_id=tokenizer.eos_token_id
    )[0]["generated_text"]

    # Extract the generated part only
    generated_response = response[len(prompt):]
    yield generated_response


# Gradio app definition
demo = gr.ChatInterface(
    respond,
    additional_inputs=[
        gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
        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)",
        ),
    ],
)

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