app.py

import streamlit as st
import torch
import time
from threading import Thread
from transformers import (
    AutoTokenizer,
    AutoModelForCausalLM,
    TextIteratorStreamer
)


# App title
st.set_page_config(page_title="😶‍🌫️ FuseChat Model")

root_path = "FuseAI"

@st.cache_resource
def load_model(model_name):
    tokenizer = AutoTokenizer.from_pretrained(
        f"{root_path}/{model_name}",
        trust_remote_code=True,
    )

    if tokenizer.pad_token_id is None:
            if tokenizer.eos_token_id is not None:
                tokenizer.pad_token_id = tokenizer.eos_token_id
            else:
                tokenizer.pad_token_id = 0

    model = AutoModelForCausalLM.from_pretrained(
        f"{root_path}/{model_name}",
        device_map="auto",
        torch_dtype=torch.bfloat16,
        trust_remote_code=True,
    )

    model.eval()
    return model, tokenizer


with st.sidebar:
    st.title('😶‍🌫️ FuseChat')
    st.write('This chatbot is created using FuseChat, a model developed by FuseAI')
    st.subheader('Models and parameters')
    selected_model = st.sidebar.selectbox('Choose a FuseChat model', ['FuseChat-7B-VaRM', 'FuseChat-7B-Slerp', 'FuseChat-7B-TA'], key='selected_model')
    temperature = st.sidebar.slider('temperature', min_value=0.01, max_value=5.0, value=0.1, step=0.01)
    top_p = st.sidebar.slider('top_p', min_value=0.01, max_value=1.0, value=0.9, step=0.01)
    top_k = st.sidebar.slider('top_k', min_value=1, max_value=1000, value=50, step=1)
    repetition_penalty = st.sidebar.slider('repetition penalty', min_value=1., max_value=2., value=1.2, step=0.05)
    max_length = st.sidebar.slider('max new tokens', min_value=32, max_value=2000, value=240, step=8)

with st.spinner('loading model..'):
    model, tokenizer = load_model(selected_model)

# Store LLM generated responses
if "messages" not in st.session_state.keys():
    st.session_state.messages = [{"role": "assistant", "content": "How may I assist you today?"}]

# Display or clear chat messages
for message in st.session_state.messages:
    with st.chat_message(message["role"]):
        st.write(message["content"])

def clear_chat_history():
    st.session_state.messages = [{"role": "assistant", "content": "How may I assist you today?"}]
st.sidebar.button('Clear Chat History', on_click=clear_chat_history)


def generate_fusechat_response():
    string_dialogue = "You are a helpful and harmless assistant."
    for dict_message in st.session_state.messages:
        if dict_message["role"] == "user":
            string_dialogue += "GPT4 Correct User: " + dict_message["content"] + "<|end_of_turn|>"
        else:
            string_dialogue += "GPT4 Correct Assistant: " + dict_message["content"] + "<|end_of_turn|>"

    input_ids = tokenizer(f"{string_dialogue}GPT4 Correct Assistant: ", return_tensors="pt").input_ids
    streamer = TextIteratorStreamer(tokenizer, timeout=10.0, skip_prompt=True, skip_special_tokens=True)
    generate_kwargs = dict(
        {"input_ids": input_ids},
        streamer=streamer,
        max_new_tokens=max_length,
        do_sample=True,
        top_p=top_p,
        top_k=top_k,
        temperature=temperature,
        num_beams=1,
        repetition_penalty=repetition_penalty,
    )
    t = Thread(target=model.generate, kwargs=generate_kwargs)
    t.start()

    outputs = []
    for text in streamer:
        outputs.append(text)
    return "".join(outputs)

# User-provided prompt
if prompt := st.chat_input("Hello there! How are you doing?"):
    st.session_state.messages.append({"role": "user", "content": prompt})
    with st.chat_message("user"):
        st.write(prompt)

# Generate a new response if last message is not from assistant
if st.session_state.messages[-1]["role"] != "assistant":
    with st.chat_message("assistant"):
        with st.spinner("Thinking..."):
            response = generate_fusechat_response()
            placeholder = st.empty()
            full_response = ''
            for item in response:
                full_response += item
                time.sleep(0.05)
                placeholder.markdown(full_response + "▌")
            placeholder.markdown(full_response)
    message = {"role": "assistant", "content": full_response}
    st.session_state.messages.append(message)