app.py

import gradio as gr
import re
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch
import bitsandbytes
import accelerate
model_name_or_path = "teknium/OpenHermes-2.5-Mistral-7B"
dtype = torch.bfloat16
model = AutoModelForCausalLM.from_pretrained(model_name_or_path,
                                             device_map="auto",
                                             torch_dtype=dtype,
                                             trust_remote_code=False,
                                             load_in_4bit=True,
                                             revision="main")
tokenizer = AutoTokenizer.from_pretrained(model_name_or_path, use_fast=True)

BASE_SYSTEM_MESSAGE = "I carefully provide accurate, factual, thoughtful, nuanced answers and am brilliant at reasoning."

def clear_chat(chat_history_state, chat_message):
    chat_history_state = []
    chat_message = ''
    return chat_history_state, chat_message

def user(message, history):
    history = history or []
    history.append([message, ""])
    return "", history

def regenerate(chatbot, chat_history_state, system_msg, max_tokens, temperature, top_p, top_k, repetition_penalty):
    print("Regenerate function called")  # Debug print
    
    if not chat_history_state:
        print("Chat history is empty")  # Debug print
        return chatbot, chat_history_state, ""
    
    # Remove only the last assistant's message from the chat history
    if len(chat_history_state) > 0:
        print(f"Before: {chat_history_state[-1]}")  # Debug print
        chat_history_state[-1][1] = ""
        print(f"After: {chat_history_state[-1]}")  # Debug print
    
    # Re-run the chat function
    new_history, _, _ = chat(chat_history_state, system_msg, max_tokens, temperature, top_p, top_k, repetition_penalty)
    print(f"New history: {new_history}")  # Debug print
    
    return new_history, new_history, ""


def chat(history, system_message, max_tokens, temperature, top_p, top_k, repetition_penalty):
    print(f"Chat function called with history: {history}")
    history = history or []
    
    # Use BASE_SYSTEM_MESSAGE if system_message is empty
    system_message_to_use = system_message if system_message.strip() else BASE_SYSTEM_MESSAGE
    
    # A última mensagem do usuário
    user_prompt = history[-1][0] if history else ""
    print(f"User prompt used for generation: {user_prompt}")  # Debug print
    # Preparar a entrada para o modelo
    prompt_template = f'''system
{system_message_to_use.strip()}
user
{user_prompt}
assistant
'''
    input_ids = tokenizer(prompt_template, return_tensors='pt').input_ids.cuda()
    
    # Gerar a saída
    output = model.generate(
        input_ids=input_ids,
        max_length=max_tokens,
        temperature=temperature,
        top_p=top_p,
        top_k=top_k,
        repetition_penalty=repetition_penalty
    )
    
    # Decodificar a saída
    decoded_output = tokenizer.decode(output[0], skip_special_tokens=True)
    assistant_response = decoded_output.split('assistant')[-1].strip()  # Pegar apenas a última resposta do assistente
    print(f"Generated assistant response: {assistant_response}")  # Debug print
    # Atualizar o histórico
    if history:
        history[-1][1] += assistant_response
    else:
        history.append(["", assistant_response])
        
    print(f"Updated history: {history}")
    return history, history, ""


start_message = ""

with gr.Blocks() as demo:
    with gr.Row():
        with gr.Column():
            gr.Markdown("""
    ## OpenHermes-V2.5 Finetuned on Mistral 7B
    **Space created by [@artificialguybr](https://twitter.com/artificialguybr). Model by [@Teknium1](https://twitter.com/Teknium1). Thanks HF for GPU!**
    **OpenHermes-V2.5 is currently SOTA in some benchmarks for 7B models.**
    **Hermes 2 model was trained on 900,000 instructions, and surpasses all previous versions of Hermes 13B and below, and matches 70B on some benchmarks! Hermes 2 changes the game with strong multiturn chat skills, system prompt capabilities, and uses ChatML format. It's quality, diversity and scale is unmatched in the current OS LM landscape. Not only does it do well in benchmarks, but also in unmeasured capabilities, like Roleplaying, Tasks, and more.**
    """)
    with gr.Row():
        #chatbot = gr.Chatbot().style(height=500)
        chatbot = gr.Chatbot(elem_id="chatbot")
    with gr.Row():
        message = gr.Textbox(
            label="What do you want to chat about?",
            placeholder="Ask me anything.",
            lines=3,
        )
    with gr.Row():
        submit = gr.Button(value="Send message", variant="secondary", scale=1)
        clear = gr.Button(value="New topic", variant="secondary", scale=0)
        stop = gr.Button(value="Stop", variant="secondary", scale=0)
        regen_btn = gr.Button(value="Regenerate", variant="secondary", scale=0)
    with gr.Accordion("Show Model Parameters", open=False):
        with gr.Row():
            with gr.Column():
                max_tokens = gr.Slider(20, 512, label="Max Tokens", step=20, value=500)
                temperature = gr.Slider(0.0, 2.0, label="Temperature", step=0.1, value=0.7)
                top_p = gr.Slider(0.0, 1.0, label="Top P", step=0.05, value=0.95)
                top_k = gr.Slider(1, 100, label="Top K", step=1, value=40)
                repetition_penalty = gr.Slider(1.0, 2.0, label="Repetition Penalty", step=0.1, value=1.1)

        system_msg = gr.Textbox(
            start_message, label="System Message", interactive=True, visible=True, placeholder="System prompt. Provide instructions which you want the model to remember.", lines=5)

    chat_history_state = gr.State()
    clear.click(clear_chat, inputs=[chat_history_state, message], outputs=[chat_history_state, message], queue=False)
    clear.click(lambda: None, None, chatbot, queue=False)

    submit_click_event = submit.click(
    fn=user, inputs=[message, chat_history_state], outputs=[message, chat_history_state], queue=True
    ).then(
        fn=chat, inputs=[chat_history_state, system_msg, max_tokens, temperature, top_p, top_k, repetition_penalty], outputs=[chatbot, chat_history_state, message], queue=True
    )
    
    # Corrected the clear button click event
    clear.click(
        fn=clear_chat, inputs=[chat_history_state, message], outputs=[chat_history_state, message], queue=False
    )
    
    # Stop button remains the same
    stop.click(fn=None, inputs=None, outputs=None, cancels=[submit_click_event], queue=False)
    regen_click_event = regen_btn.click(
        fn=regenerate, 
        inputs=[chatbot, chat_history_state, system_msg, max_tokens, temperature, top_p, top_k, repetition_penalty], 
        outputs=[chatbot, chat_history_state, message], 
        queue=True
    )


demo.queue(max_size=128, concurrency_count=2)
demo.launch()