app.py

import os
os.system("pip install torch transformers accelerate")
from transformers import AutoModelForCausalLM, AutoTokenizer
import gradio as gr
import requests
import json

from accelerate import Accelerator

SYSTEM_PROMPT = "As a generative chatbot (you are not a GPT but your structure is 50% the same), your primary function is to provide helpful and friendly responses to user queries. Feel free to add some personality, but make sure your responses are accurate and helpful. Your owner and developer is: @Costikoooo (Discord user) other developers are unknown. Your name is Chattybot."
TITLE = "Chattybot"
EXAMPLE_INPUT = "hello"

# Use your provided tokenizer and model
tokenizer = AutoTokenizer.from_pretrained('stabilityai/stablelm-zephyr-3b')
model = AutoModelForCausalLM.from_pretrained(
    'stabilityai/stablelm-zephyr-3b',
    trust_remote_code=True,
    device_map="auto"
)

HF_TOKEN = os.getenv("HF_TOKEN")
HEADERS = {"Authorization": f"Bearer {HF_TOKEN}"}

# Initialize Accelerator
accelerator = Accelerator()

# Wrap model and tokenizer with accelerator
model, tokenizer = accelerator.prepare(model, tokenizer)

def build_input_prompt(message, chatbot, system_prompt):
    input_prompt = "\n" + system_prompt + "</s>\n\n"
    for interaction in chatbot:
        input_prompt = input_prompt + str(interaction[0]) + "</s>\n\n" + str(interaction[1]) + "\n</s>\n\n"

    input_prompt = input_prompt + str(message) + "</s>\n"
    return input_prompt

def predict_beta(message, chatbot=[], system_prompt=""):
    input_prompt = build_input_prompt(message, chatbot, system_prompt)
    inputs = tokenizer(input_prompt, return_tensors="pt")

    with accelerator.device():
        tokens = model.generate(
            inputs["input_ids"],
            max_length=1024,
            temperature=0.8,
            do_sample=True
        )
    bot_message = tokenizer.decode(tokens[0], skip_special_tokens=True)
    return bot_message

def test_preview_chatbot(message, history):
    response = predict_beta(message, history, SYSTEM_PROMPT)
    text_start = response.rfind("") + len("")
    response = response[text_start:]
    return response

welcome_preview_message = f"""
Welcome to **{TITLE}**! Say something like: 
"{EXAMPLE_INPUT}"
"""

chatbot_preview = gr.Chatbot(layout="panel", value=[(None, welcome_preview_message)])
textbox_preview = gr.Textbox(scale=7, container=False, value=EXAMPLE_INPUT)

demo = gr.ChatInterface(test_preview_chatbot, chatbot=chatbot_preview, textbox=textbox_preview)

demo.launch()