app.py

import streamlit as st
from typing import Generator
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

st.set_page_config(
    page_icon="💬",
    page_title="Chat App",
    layout="wide",
)

model_name = "JuliaTsk/SuccinctLabs-chat-finetuned"

tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name, device_map="cpu", low_cpu_mem_usage=True)
st.title("ChatGPT-like clone 🎈")


def generate_chat_responses(chat_completion) -> Generator[str, None, None]:
    for chunk in chat_completion:
        if chunk.choices[0].delta.content:
            yield chunk.choices[0].delta.content


left, right = st.columns([2, 6], vertical_alignment="top")
max_tokens_range = 32768
max_tokens = left.slider(
    label="Max Tokens:",
    min_value=128,
    max_value=max_tokens_range,
    # Default value or max allowed if less
    value=min(1024, max_tokens_range),
    step=128,
    help=f"Adjust the maximum number of tokens (words) for the model's response."
)
temperature = left.slider(
    label="Temperature:",
    min_value=0.0,
    max_value=1.0,
    value=0.7,
    step=0.01,
    help=f"Controls randomness: a low value means less random responses."
)

if "messages" not in st.session_state:
    st.session_state.messages = []

for message in st.session_state.messages:
    avatar = '🤖' if message["role"] == "assistant" else '👨‍💻'
    with right.chat_message(message["role"], avatar=avatar):
        right.markdown(message["content"])


prompt = st.chat_input("Say something")
if prompt:
    with right.chat_message("user", avatar='👨‍💻'):
        right.markdown(prompt)
    st.session_state.messages.append({"role": "user", "content": prompt})
    with right.chat_message("assistant"):
        inputs = tokenizer(prompt, return_tensors="pt")
        outputs = model.generate(inputs["input_ids"], max_length=100, num_return_sequences=1)
        generated_text = tokenizer.decode(outputs[0], skip_special_tokens=True)

st.session_state.messages.append({"role": "assistant", "content": generated_text})