app.py

import os
import math
import transformers
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch
import gradio as gr
import sentencepiece
from tokenization_xgen import XgenTokenizer

title = "Welcome to 🙋🏻‍♂️Tonic's🌷Tulu Chat!"
description = "[allenai/tulu-2-dpo-7b](https://huggingface.co/allenai/tulu-2-dpo-7b) and larger Tulu-2 models are Instruct Llama Finetunes using the [mistralai/Mistral-7B](https://huggingface.co/mistralai/Mistral-7B-v0.1) recipe. You can use [allenai/tulu-2-13b](https://huggingface.co/allenai/tulu-2-13b) here via API using Gradio by scrolling down and clicking Use 'Via API' or privately by [cloning this space on huggingface](https://huggingface.co/spaces/Tonic1/TuluDemo?duplicate=true) See also the large model here : [allenai/tulu-2-dpo-70b](https://huggingface.co/allenai/tulu-2-dpo-70b) . [Join my active builders' server on discord](https://discord.gg/VqTxc76K3u). Let's build together!."

os.environ['PYTORCH_CUDA_ALLOC_CONF'] = 'max_split_size_mb:50'
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
model_name = "allenai/tulu-2-dpo-13b"
tokenizer = AutoTokenizer.from_pretrained("allenai/tulu-2-dpo-13b")
model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.bfloat16, device_map="auto")

class TuluChatBot:
    def __init__(self, model, tokenizer, system_message="You are 🌷Tulu, an AI language model created by Tonic-AI. You are a cautious assistant. You carefully follow instructions. You are helpful and harmless and you follow ethical guidelines and promote positive behavior."):
        self.model = model
        self.tokenizer = tokenizer
        self.system_message = system_message

    def set_system_message(self, new_system_message):
        self.system_message = new_system_message

    def format_prompt(self, user_message):
        prompt = f"<|assistant|>\n {self.system_message}\n\n <|user|>{user_message}\n\n<|assistant|>\n"
        return prompt

    def predict(self, user_message, temperature=0.4, max_new_tokens=70, top_p=0.99, repetition_penalty=1.9):
        prompt = self.format_prompt(user_message)
        inputs = self.tokenizer(prompt, return_tensors='pt', add_special_tokens=False)
        input_ids = inputs["input_ids"].to(self.model.device)
        attention_mask = inputs["attention_mask"].to(self.model.device)


        output_ids = self.model.generate(
            input_ids,
            attention_mask=attention_mask,
            max_length=input_ids.shape[1] + max_new_tokens,
            temperature=temperature,
            top_p=top_p,
            repetition_penalty=repetition_penalty,
            do_sample=True
        )

        response = self.tokenizer.decode(output_ids[0], skip_special_tokens=True)
        return response

def gradio_predict(user_message, system_message, max_new_tokens, temperature, top_p, repetition_penalty):
    Tulu_bot.set_system_message(system_message)
    response = Tulu_bot.predict(user_message, temperature, max_new_tokens, top_p, repetition_penalty)
    return response
    
Tulu_bot = TuluChatBot(model, tokenizer)

iface = gr.Interface(
    fn=gradio_predict,
    title=title,
    description=description,
    inputs=[
        gr.Textbox(label="Your Message", type="text", lines=3),
        gr.Textbox(label="Introduce a Character Here or Set a Scene (system prompt)", type="text", lines=2),
        gr.Slider(label="Max new tokens", value=1269, minimum=550, maximum=3200, step=1),
        gr.Slider(label="Temperature", value=1.2, minimum=0.05, maximum=4.0, step=0.05),
        gr.Slider(label="Top-p (nucleus sampling)", value=0.90, minimum=0.01, maximum=0.99, step=0.05),
        gr.Slider(label="Repetition penalty", value=1.9, minimum=1.0, maximum=2.0, step=0.05)
    ],
    outputs="text",
    theme="ParityError/Anime"
)

iface.queue(max_size=5).launch()