app.py

import os
import re
import torch
from threading import Thread
from typing import Iterator
from mongoengine import connect, Document, StringField, SequenceField
import gradio as gr
import spaces
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig, TextIteratorStreamer
from peft import PeftModel
import openai
from openai import OpenAI



openai.api_key = os.environ.get("OPENAI_KEY")

def generate_image(text):
    try:
        response = openai.images.generate(
            model="dall-e-3",
            prompt="Create an illustration that accurately depicts the character and the setting of a story:"+text,
            n=1,
            size="1024x1024"
        )
    except Exception as error:
        print(str(error))
        raise gr.Error("An error occurred while generating the image. Please check your API key and try again.")

    
    return response.data[0].url
    

# Constants
MAX_MAX_NEW_TOKENS = 2048
DEFAULT_MAX_NEW_TOKENS = 1024
MAX_INPUT_TOKEN_LENGTH = int(os.getenv("MAX_INPUT_TOKEN_LENGTH", "4096"))

LICENSE = """
---
As a derivative work of [Llama-2-7b-hf](https://huggingface.co/meta-llama/Llama-2-7b-hf) by Meta,
this demo is governed by the original [license](https://huggingface.co/spaces/huggingface-projects/llama-2-7b-chat/blob/main/LICENSE.txt) and [acceptable use policy](https://huggingface.co/spaces/huggingface-projects/llama-2-7b-chat/blob/main/USE_POLICY.md).
"""

# GPU Check and add CPU warning
if not torch.cuda.is_available():
    DESCRIPTION += "\n<p>Running on CPU 🥶 This demo does not work on CPU.</p>"

if torch.cuda.is_available():

    # Model and Tokenizer Configuration
    model_id = "meta-llama/Llama-2-7b-chat-hf"
    bnb_config = BitsAndBytesConfig(
        load_in_4bit=True,
        bnb_4bit_use_double_quant=False,
        bnb_4bit_quant_type="nf4",
        bnb_4bit_compute_dtype=torch.bfloat16
    )
    base_model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto", quantization_config=bnb_config)
    model = PeftModel.from_pretrained(base_model, "ranamhamoud/storytell")
    tokenizer = AutoTokenizer.from_pretrained(model_id)
    tokenizer.pad_token = tokenizer.eos_token


def make_prompt(entry):
    return  f"### Human: When asked to explain use a story.Don't repeat the assesments, limit to 500 words.However keep context in mind if edits to the content is required. {entry} ### Assistant:"

def process_text(text):

    text = re.sub(r'\[answer:\]\s*', 'Answer: ', text)
    text = re.sub(r'\[.*?\](?<!Answer: )', '', text)
    
    return text
custom_css = """
body, input, button, textarea, label {
    font-family: Arial, sans-serif;
    font-size: 24px;
}
.gr-chat-interface .gr-chat-message-container {
    font-size: 14px;
}
.gr-button {
    font-size: 14px;
    padding: 12px 24px;
}
.gr-input {
    font-size: 14px;
}
"""

# Gradio Function
@spaces.GPU
def generate(
    message: str,
    chat_history: list[tuple[str, str]],
    max_new_tokens: int = DEFAULT_MAX_NEW_TOKENS,
    temperature: float = 0.6,
    top_p: float = 0.7,
    top_k: int = 20,
    repetition_penalty: float = 1.0,
) -> Iterator[str]:
    conversation = []
    for user, assistant in chat_history:
        conversation.extend([{"role": "user", "content": user}, {"role": "assistant", "content": assistant}])
    conversation.append({"role": "user", "content": make_prompt(message)})
    enc = tokenizer(make_prompt(message), return_tensors="pt", padding=True, truncation=True)
    input_ids = enc.input_ids.to(model.device)
    if input_ids.shape[1] > MAX_INPUT_TOKEN_LENGTH:
        input_ids = input_ids[:, -MAX_INPUT_TOKEN_LENGTH:]
        gr.Warning(f"Trimmed input from conversation as it was longer than {MAX_INPUT_TOKEN_LENGTH} tokens.")
    
    streamer = TextIteratorStreamer(tokenizer, timeout=10.0, skip_prompt=True, skip_special_tokens=False)
    generate_kwargs = dict(
        {"input_ids": input_ids},
        streamer=streamer,
        max_new_tokens=max_new_tokens,
        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:
        processed_text = process_text(text)
        outputs.append(processed_text)
        output = "".join(outputs)
        yield output

    final_story = "".join(outputs)
    image_url = generate_image(final_story)
    outputs.append(image_url)

chat_interface = gr.ChatInterface(
    fn=generate,
    fill_height=True,
    stop_btn=None,
    examples=[
        ["Can you explain briefly to me what is the Python programming language?"],
        ["Could you please provide an explanation about the concept of recursion?"],
        ["Could you explain what a URL is?"]
    ],
    theme='shivi/calm_seafoam',autofocus=True, 
)

# Gradio Web Interface
with gr.Blocks(css=custom_css,theme='shivi/calm_seafoam',fill_height=True) as demo:
        chat_interface.render()
    # gr.Markdown(LICENSE)

        
# Main Execution
if __name__ == "__main__":
    demo.queue(max_size=20)
    demo.launch(share=True)