app.py

import gradio as gr
import os
import time
from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer, BitsAndBytesConfig
import torch
from threading import Thread
import logging
import spaces
from functools import lru_cache

print(f"Is CUDA available: {torch.cuda.is_available()}")
# True
print(f"CUDA device: {torch.cuda.get_device_name(torch.cuda.current_device())}")

# Set up logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)

# Set an environment variable
HF_TOKEN = os.environ.get("HF_TOKEN", None)

DESCRIPTION = '''
<div>
<h1 style="text-align: center;">ContenteaseAI custom trained model</h1>
</div>
'''

LICENSE = """
<p/>
---
For more information, visit our [website](https://contentease.ai).
"""

PLACEHOLDER = """
<div style="padding: 30px; text-align: center; display: flex; flex-direction: column; align-items: center;">
   <h1 style="font-size: 28px; margin-bottom: 2px; opacity: 0.55;">ContenteaseAI Custom AI trained model</h1>
   <p style="font-size: 18px; margin-bottom: 2px; opacity: 0.65;">Enter the text extracted from the PDF:</p>
</div>
"""

css = """
h1 {
  text-align: center;
  display: block;
}
"""

# Load the tokenizer and model with quantization
model_id = "meta-llama/Meta-Llama-3.1-8B-Instruct"
bnb_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_use_double_quant=True,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_compute_dtype=torch.bfloat16
)

@lru_cache(maxsize=1)
def load_model_and_tokenizer():
    try:
        start_time = time.time()
        logger.info("Loading tokenizer...")
        tokenizer = AutoTokenizer.from_pretrained(model_id)
        logger.info("Loading model...")
        model = AutoModelForCausalLM.from_pretrained(
            model_id,
            device_map="auto",
            quantization_config=bnb_config,
            torch_dtype=torch.bfloat16
        )
        model.generation_config.pad_token_id = tokenizer.pad_token_id
        end_time = time.time()
        logger.info(f"Model and tokenizer loaded successfully in {end_time - start_time} seconds.")
        return model, tokenizer
    except Exception as e:
        logger.error(f"Error loading model or tokenizer: {e}")
        raise

try:
    model, tokenizer = load_model_and_tokenizer()
except Exception as e:
    logger.error(f"Failed to load model and tokenizer: {e}")
    raise

terminators = [
    tokenizer.eos_token_id,
    tokenizer.convert_tokens_to_ids("<|eot_id|>")
]

SYS_PROMPT = """
Given the text of a hotel property improvement plan, extract the items to be replaced for only the Guest Rooms/ Suites, Guest Bathrooms/Suite Bathrooms.
First, find the section of the pdf which describes improvements to be done on the Guest Rooms and Guest Bathrooms, then find the items to be replaced.
Ignore items from other sections of the hotel.
Items to be replaced are usually preceded by the words replace, install, or provide. 
Return the results as a JSON with "Guest Room" and "Guest Bathroom" as keys and each value the list of unique items to be replaced.
Return only the JSON with no extra text.

Example Text:
"
Site & Building Exterior
Replace all exterior decorative lighting
...
Guestrooms
Replace [ORG] C-Table.
Provide full length mirror.
Replace cabinets - Kitchen.
at doors where brass hardware finishes exist – replace with stainless
...
Guest Bathrooms - (FRCM) Replace mirrors. Install a vanity mirror that has integrated lighting
Guest Bathrooms - (FRCM) Replace artwork and decorative accessories.
...
Suites - Replace microwave, refrigerator, and associated casegood cabinet.
"
Example Response:
{
"Guest Room": [
  "C-Table",
  "full length mirror",
  "kitchen cabinets",
  "stainless steel door hardware",
  "microwave",
  "refrigerator",
  "casegood cabinet",],
"Guest Bathroom": [
  "vanity mirror with integrated lighting",
  "artwork",
  "decorative accessories",],
}
"""
def chunk_text(text, chunk_size=5000):
    """
    Splits the input text into chunks of specified size.
    
    Args:
        text (str): The input text to be chunked.
        chunk_size (int): The size of each chunk in tokens.
        
    Returns:
        list: A list of text chunks.
    """
    words = text.split()
    chunks = [' '.join(words[i:i + chunk_size]) for i in range(0, len(words), chunk_size)]
    logger.info(f"Total chunks created: {len(chunks)}")
    return chunks

def combine_responses(responses):
    """
    Combines the responses from all chunks into a final output string.
    
    Args:
        responses (list): A list of responses from each chunk.
        
    Returns:
        str: The combined output string.
    """
    combined_output = " ".join(responses)
    return combined_output

def generate_response_for_chunk(chunk, history, temperature, max_new_tokens):
    start_time = time.time()
    if len(history) == 0:
        pass
    else:
        history.pop()
    conversation = [{"role": "system", "content": SYS_PROMPT}]
    for user, assistant in history:
        conversation.extend([{"role": "user", "content": user}, {"role": "assistant", "content": assistant}])
    conversation.append({"role": "user", "content": chunk})

    input_ids = tokenizer.apply_chat_template(conversation, return_tensors="pt").to(model.device)
    
    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_new_tokens,
        do_sample=True,
        temperature=temperature,
        eos_token_id=terminators,
        pad_token_id=tokenizer.eos_token_id
    )
    if temperature == 0:
        generate_kwargs['do_sample'] = False
        
    t = Thread(target=model.generate, kwargs=generate_kwargs)
    t.start()

    outputs = []
    for text in streamer:
        outputs.append(text)
    
    end_time = time.time()
    logger.info(f"Time taken for generating response for a chunk: {end_time - start_time} seconds")
    
    return "".join(outputs)

@spaces.GPU(duration=110)
def chat_llama3_8b(message: str, history: list, temperature: float, max_new_tokens: int):
    """
    Generate a streaming response using the llama3-8b model with chunking.
    
    Args:
        message (str): The input message.
        history (list): The conversation history used by ChatInterface.
        temperature (float): The temperature for generating the response.
        max_new_tokens (int): The maximum number of new tokens to generate.
        
    Returns:
        str: The generated response.
    """
    try:
        start_time = time.time()
        
        chunks = chunk_text(message)
        responses = []
        count=0
        for chunk in chunks:
            logger.info(f"Processing chunk {count+1}/{len(chunks)}")
            response = generate_response_for_chunk(chunk, history, temperature, max_new_tokens)
            responses.append(response)
            count+=1
        final_output = combine_responses(responses)
        
        end_time = time.time()
        logger.info(f"Total time taken for generating response: {end_time - start_time} seconds")
        
        yield final_output
    except Exception as e:
        logger.error(f"Error generating response: {e}")
        yield "An error occurred while generating the response. Please try again."

# Gradio block
chatbot = gr.Chatbot(height=450, placeholder=PLACEHOLDER, label='Gradio ChatInterface')

with gr.Blocks(fill_height=True, css=css) as demo:
    gr.Markdown(DESCRIPTION)
    
    gr.ChatInterface(
        fn=chat_llama3_8b,
        chatbot=chatbot,
        fill_height=True,
        additional_inputs_accordion=gr.Accordion(label="⚙️ Parameters", open=False, render=False),
        additional_inputs=[
            gr.Slider(minimum=0, maximum=1, step=0.1, value=0.95, label="Temperature", render=False),
            gr.Slider(minimum=128, maximum=2000, step=1, value=700, label="Max new tokens", render=False),
        ]
    )
    
    gr.Markdown(LICENSE)
    
if __name__ == "__main__":
    try:
        demo.launch(show_error=True)
    except Exception as e:
        logger.error(f"Error launching Gradio demo: {e}")