import gradio as gr
from huggingface_hub import InferenceClient
import random

# Define the model to be used
model = "NousResearch/Nous-Hermes-2-Mixtral-8x7B-DPO"
client = InferenceClient(model)

def format_prompt_mixtral(message, history):
    prompt = "<s>"
    if history:
        for user_prompt, bot_response in history:
            prompt += f"[INST] {user_prompt} [/INST]"
            prompt += f" {bot_response}</s> "
    prompt += f"[INST] {message} [/INST]"
    return prompt

def chat_inf(system_prompt, prompt, history, seed, temp, tokens, top_p, rep_p):
    if system_prompt:
        system_prompt = f'{system_prompt}, '

    generate_kwargs = dict(
        temperature=temp,
        max_new_tokens=tokens,
        top_p=top_p,
        repetition_penalty=rep_p,
        do_sample=True,
        seed=seed,
    )

    formatted_prompt = format_prompt_mixtral(f"{system_prompt}{prompt}", history)
    stream = client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=False)
    output = ""
    for response in stream:
        output += response.token.text
        yield [(prompt, output)]
    history.append((prompt, output))
    yield history

def clear_fn():
    return None, None, None

rand_val = random.randint(1, 1111111111111111)

def check_rand(inp, val):
    if inp:
        return gr.Slider(label="Seed", minimum=1, maximum=1111111111111111, value=random.randint(1, 1111111111111111))
    else:
        return gr.Slider(label="Seed", minimum=1, maximum=1111111111111111, value=int(val))

with gr.Blocks() as app:
    gr.HTML("""<center><h1 style='font-size:xx-large;'>Chatbot</h1><br><h3>running on Huggingface Inference Client</h3><br><h7>EXPERIMENTAL""")
    with gr.Row():
        chat = gr.Chatbot(height=500)
    with gr.Group():
        with gr.Row():
            with gr.Column(scale=3):
                inp = gr.Textbox(label="Prompt")
                sys_inp = gr.Textbox(label="System Prompt (optional)")
                with gr.Row():
                    with gr.Column(scale=2):
                        btn = gr.Button("Chat")
                    with gr.Column(scale=1):
                        with gr.Group():
                            stop_btn = gr.Button("Stop")
                            clear_btn = gr.Button("Clear")
            with gr.Column(scale=1):
                with gr.Group():
                    rand = gr.Checkbox(label="Random Seed", value=True)
                    seed = gr.Slider(label="Seed", minimum=1, maximum=1111111111111111, step=1, value=rand_val)
                    tokens = gr.Slider(label="Max new tokens", value=3840, minimum=0, maximum=8000, step=64, interactive=True, visible=True, info="The maximum number of tokens")
                    temp = gr.Slider(label="Temperature", step=0.01, minimum=0.01, maximum=1.0, value=0.9)
                    top_p = gr.Slider(label="Top-P", step=0.01, minimum=0.01, maximum=1.0, value=0.9)
                    rep_p = gr.Slider(label="Repetition Penalty", step=0.1, minimum=0.1, maximum=2.0, value=1.0)

    hid1 = gr.Number(value=1, visible=False)

    go = btn.click(check_rand, [rand, seed], seed).then(chat_inf, [sys_inp, inp, chat, seed, temp, tokens, top_p, rep_p], chat)

    stop_btn.click(None, None, None, cancels=[go])
    clear_btn.click(clear_fn, None, [inp, sys_inp, chat])

app.queue(default_concurrency_limit=10).launch()