app.py

import gradio as gr
import torch
from unsloth import FastLanguageModel
from transformers import TextStreamer
from unsloth.chat_templates import get_chat_template

# Initialize the model
max_seq_length = 2048
dtype = None
load_in_4bit = True

model, tokenizer = FastLanguageModel.from_pretrained(
    model_name="umair894/llama3",
    max_seq_length=max_seq_length,
    dtype=dtype,
    load_in_4bit=load_in_4bit,
)

tokenizer = get_chat_template(
    tokenizer,
    chat_template="llama-3",
    mapping={"role": "from", "content": "value", "user": "human", "assistant": "gpt"},
    map_eos_token=True,
)

FastLanguageModel.for_inference(model) # Enable native 2x faster inference

# VIKK introduction prompt
vikk_intro = """Consider you self a legal assistant in USA and your name is VIKK. You are very knowledgeable about all aspects of the law...
"""

# Function to get chat response
def get_response(message, history, system_message, max_tokens, temperature, top_p):
    messages = [{"role": "system", "content": system_message}] if system_message else []
    if not history:
        history = [{"role": "assistant", "content": vikk_intro}]
    
    for msg in history:
        if msg[0]:
            messages.append({"role": "user", "content": msg[0]})
        if msg[1]:
            messages.append({"role": "assistant", "content": msg[1]})
    
    messages.append({"role": "user", "content": message})

    formatted_messages = [{"from": "assistant", "value": vikk_intro}]
    for msg in messages[1:]:
        role = "human" if msg["role"] == "user" else "assistant"
        formatted_messages.append({"from": role, "value": msg["content"]})

    inputs = tokenizer.apply_chat_template(
        formatted_messages,
        tokenize=True,
        add_generation_prompt=True,
        return_tensors="pt",
    ).to("cuda")

    text_streamer = TextStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)
    
    output = ""
    for out in model.generate(input_ids=inputs["input_ids"], streamer=text_streamer, max_new_tokens=max_tokens, use_cache=True):
        output += out

    response = tokenizer.decode(output, skip_special_tokens=True).split(">>> Assistant: ")[-1].strip()
    
    return response

# Gradio interface
with gr.Blocks() as demo:
    gr.Markdown("# Chatbot Interface")

    with gr.Row():
        with gr.Column():
            system_message = gr.Textbox(value="You are a friendly Chatbot.", label="System message")
            max_tokens = gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens")
            temperature = gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature")
            top_p = gr.Slider(minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-p (nucleus sampling)")

        with gr.Column():
            chatbot = gr.Chatbot()

            user_input = gr.Textbox(label="You:")
            send_button = gr.Button("Send")

            def respond(message, history, system_message, max_tokens, temperature, top_p):
                response = get_response(message, history, system_message, max_tokens, temperature, top_p)
                history.append((message, response))
                return history

            send_button.click(respond, [user_input, chatbot, system_message, max_tokens, temperature, top_p], chatbot)

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