app.py

import os
import sys
import uvicorn
from fastapi import FastAPI, Query, HTTPException, BackgroundTasks
from fastapi.responses import HTMLResponse
from starlette.middleware.cors import CORSMiddleware
from datasets import load_dataset, list_datasets
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
from loguru import logger
import concurrent.futures
import psutil
import asyncio
import torch
from tenacity import retry, stop_after_attempt, wait_fixed
from huggingface_hub import HfApi, RepositoryNotFoundError
from dotenv import load_dotenv

# Cargar variables de entorno
load_dotenv()

# Obtener el token de Hugging Face
HUGGINGFACE_TOKEN = os.getenv("HUGGINGFACE_TOKEN")
if not HUGGINGFACE_TOKEN:
    logger.error("Hugging Face token not found. Please set the HUGGINGFACE_TOKEN environment variable.")
    sys.exit(1)

# Inicializar diccionarios para datasets y ejemplos
datasets_dict = {}
example_usage_list = []

# Configuración de caché
CACHE_DIR = os.path.expanduser("~/.cache/huggingface")
os.makedirs(CACHE_DIR, exist_ok=True)
os.environ["HF_HOME"] = CACHE_DIR
os.environ["HF_TOKEN"] = HUGGINGFACE_TOKEN

pipeline_instance = None  # Solo un pipeline

# Flag para indicar si la inicialización está completa
initialization_complete = False

def initialize_model():
    global pipeline_instance
    try:
        logger.info("Initializing the base model and tokenizer.")
        base_model_repo = "meta-llama/Llama-3.2-1B"
        model = AutoModelForCausalLM.from_pretrained(
            base_model_repo,
            cache_dir=CACHE_DIR,
            ignore_mismatched_sizes=True  # Ignorar discrepancias de tamaño
        )
        tokenizer = AutoTokenizer.from_pretrained(base_model_repo, cache_dir=CACHE_DIR)
        if tokenizer.pad_token is None:
            tokenizer.pad_token = tokenizer.eos_token
        pipeline_instance = pipeline(
            "text-generation",
            model=model,
            tokenizer=tokenizer,
            device=0 if torch.cuda.is_available() else -1
        )
        logger.info("Model and tokenizer initialized successfully.")
    except Exception as e:
        logger.error(f"Error initializing model and tokenizer: {e}", exc_info=True)
        sys.exit(1)

@retry(stop=stop_after_attempt(3), wait=wait_fixed(5))
def download_dataset(dataset_name):
    try:
        logger.info(f"Starting download for dataset: {dataset_name}")
        # Eliminado 'trust_remote_code=True' para evitar el error con ParquetConfig
        datasets_dict[dataset_name] = load_dataset(dataset_name, cache_dir=CACHE_DIR)
        create_example_usage(dataset_name)
    except Exception as e:
        logger.error(f"Error loading dataset {dataset_name}: {e}", exc_info=True)
        raise

def upload_model_to_hub():
    try:
        api = HfApi()
        model_repo = "Yhhxhfh/test"
        try:
            api.repo_info(repo_id=model_repo)
            logger.info(f"Model repository {model_repo} already exists.")
        except RepositoryNotFoundError:
            api.create_repo(repo_id=model_repo, private=False, token=HUGGINGFACE_TOKEN)
            logger.info(f"Created model repository {model_repo}.")
        logger.info(f"Pushing the model and tokenizer to {model_repo}.")
        pipeline_instance.model.push_to_hub(model_repo, use_auth_token=HUGGINGFACE_TOKEN)
        pipeline_instance.tokenizer.push_to_hub(model_repo, use_auth_token=HUGGINGFACE_TOKEN)
        logger.info(f"Successfully pushed the model and tokenizer to {model_repo}.")
    except Exception as e:
        logger.error(f"Error uploading model to Hugging Face Hub: {e}", exc_info=True)

def create_example_usage(dataset_name):
    try:
        logger.info(f"Creating example usage for dataset {dataset_name}")
        example_prompts = [
            "Translate the following catering menu from English to French:",
            "Generate a catering menu for a wedding with vegetarian options:",
            "Convert the following catering menu to a gluten-free version:",
            "Provide a detailed catering menu for a corporate event including desserts:",
            "Generate a children's birthday party catering menu with allergen-free items:"
        ]
        examples = []
        for prompt in example_prompts:
            generated_text = pipeline_instance(prompt, max_length=50, num_return_sequences=1)[0]['generated_text']
            examples.append({"prompt": prompt, "response": generated_text})
        example_usage_list.append({"dataset_name": dataset_name, "examples": examples})
        logger.info(f"Example usage created for dataset {dataset_name}")
    except Exception as e:
        logger.error(f"Error creating example usage for dataset {dataset_name}: {e}", exc_info=True)

def unify_datasets():
    try:
        logger.info("Starting to unify datasets")
        unified_dataset = None
        for dataset in datasets_dict.values():
            if unified_dataset is None:
                unified_dataset = dataset
            else:
                unified_dataset = unified_dataset.concatenate(dataset)
        datasets_dict['unified'] = unified_dataset
        logger.info("Datasets successfully unified.")
    except Exception as e:
        logger.error(f"Error unifying datasets: {e}", exc_info=True)

# Configuración de concurrencia
cpu_count = psutil.cpu_count(logical=False) or 1
memory_available_mb = psutil.virtual_memory().available / (1024 * 1024)
memory_per_download_mb = 100
memory_available = int(memory_available_mb / memory_per_download_mb)
gpu_count = torch.cuda.device_count()
max_concurrent_downloads = min(cpu_count, memory_available, gpu_count * 2 if gpu_count else cpu_count)
max_concurrent_downloads = max(1, max_concurrent_downloads)
max_concurrent_downloads = min(10, max_concurrent_downloads)

logger.info(f"Using up to {max_concurrent_downloads} concurrent workers for downloading datasets.")

executor = concurrent.futures.ThreadPoolExecutor(max_workers=max_concurrent_downloads)

async def download_and_process_datasets():
    global initialization_complete
    try:
        dataset_names = list_datasets()
        logger.info(f"Found {len(dataset_names)} datasets to download.")
        loop = asyncio.get_event_loop()
        tasks = []
        for dataset_name in dataset_names:
            task = loop.run_in_executor(executor, download_dataset, dataset_name)
            tasks.append(task)
        await asyncio.gather(*tasks)
        unify_datasets()
        upload_model_to_hub()
        initialization_complete = True
        logger.info("All initialization tasks completed successfully.")
    except Exception as e:
        logger.error(f"Error during dataset processing: {e}", exc_info=True)

# Inicializar FastAPI
app = FastAPI()

# Configuración de CORS
app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],  # Cambia esto según tus necesidades
    allow_credentials=True,
    allow_methods=["*"],
    allow_headers=["*"]
)

message_history = []

@app.on_event("startup")
async def startup_event():
    logger.info("Application startup initiated.")
    loop = asyncio.get_event_loop()
    # Crear una tarea en segundo plano para inicializar el modelo y descargar datasets
    loop.create_task(run_initialization(loop))
    logger.info("Background initialization tasks started.")

async def run_initialization(loop):
    global initialization_complete
    try:
        # Inicializar el modelo en un hilo separado
        await loop.run_in_executor(None, initialize_model)
        # Descargar y procesar datasets
        await download_and_process_datasets()
    except Exception as e:
        logger.error(f"Error during startup tasks: {e}", exc_info=True)

@app.get('/')
async def index():
    html_code = """
    <!DOCTYPE html>
    <html lang="en">
    <head>
        <!-- Existing head content -->
        <meta charset="UTF-8">
        <meta name="viewport" content="width=device-width, initial-scale=1.0">
        <title>ChatGPT Chatbot</title>
        <style>
            /* Existing styles */
            /* Add styles for the model selector */
            .model-selector {
                margin-bottom: 10px;
            }
            body {
                font-family: Arial, sans-serif;
                margin: 0;
                padding: 0;
                background-color: #f4f4f4;
            }
            .container {
                max-width: 800px;
                margin: auto;
                padding: 20px;
            }
            .chat-container {
                background-color: #fff;
                border-radius: 8px;
                box-shadow: 0 0 10px rgba(0, 0, 0, 0.1);
                overflow: hidden;
                margin-bottom: 20px;
                animation: fadeInUp 0.5s ease forwards;
                display: flex;
                flex-direction: column;
            }
            .chat-box {
                flex: 1;
                overflow-y: auto;
                padding: 10px;
            }
            .chat-input {
                width: calc(100% - 20px);
                border: none;
                border-top: 1px solid #ddd;
                padding: 10px;
                font-size: 16px;
                outline: none;
            }
            .chat-input:focus {
                border-top: 1px solid #007bff;
            }
            .user-message {
                margin-bottom: 10px;
                padding: 8px 12px;
                border-radius: 8px;
                background-color: #007bff;
                color: #fff;
                max-width: 70%;
                word-wrap: break-word;
                align-self: flex-end;
            }
            .bot-message {
                margin-bottom: 10px;
                padding: 8px 12px;
                border-radius: 8px;
                background-color: #4CAF50;
                color: #fff;
                max-width: 70%;
                word-wrap: break-word;
            }
            .toggle-history {
                text-align: center;
                cursor: pointer;
                color: #007bff;
                margin-bottom: 10px;
            }
            .history-container {
                display: none;
            }
            .history-container.show {
                display: block;
            }
            .history-container .history-content {
                max-height: 200px;
                overflow-y: auto;
            }
            @keyframes fadeInUp {
                from {
                    opacity: 0;
                    transform: translateY(20px);
                }
                to {
                    opacity: 1;
                    transform: translateY(0);
                }
            }
        </style>
    </head>
    <body>
        <div class="container">
            <h1 style="text-align: center;">ChatGPT Chatbot</h1>
            <div class="chat-container" id="chat-container">
                <div class="chat-box" id="chat-box">
                </div>
                <input type="text" class="chat-input" id="user-input" placeholder="Type your message...">
                <button onclick="retryLastMessage()">Retry Last Message</button>
            </div>
            <div class="toggle-history" onclick="toggleHistory()">Toggle History</div>
            <div class="history-container" id="history-container">
                <h2>Chat History</h2>
                <div class="history-content" id="history-content"></div>
            </div>
        </div>
        <script>
            function toggleHistory() {
                const historyContainer = document.getElementById('history-container');
                historyContainer.classList.toggle('show');
            }

            function saveMessage(sender, message) {
                const historyContent = document.getElementById('history-content');
                const messageElement = document.createElement('div');
                messageElement.className = `${sender}-message`;
                messageElement.innerText = message;
                historyContent.appendChild(messageElement);
            }

            function appendMessage(sender, message) {
                const chatBox = document.getElementById('chat-box');
                const messageElement = document.createElement('div');
                messageElement.className = `${sender}-message`;
                messageElement.innerText = message;
                chatBox.appendChild(messageElement);
                chatBox.scrollTop = chatBox.scrollHeight;
            }

            const chatContainer = document.getElementById('chat-container');
            const chatBox = document.getElementById('chat-box');
            const userInput = document.getElementById('user-input');

            userInput.addEventListener('keyup', function(event) {
                if (event.keyCode === 13) {
                    event.preventDefault();
                    sendMessage();
                }
            });

            function sendMessage() {
                const userMessage = userInput.value.trim();
                if (userMessage === '') return;

                saveMessage('user', userMessage);
                appendMessage('user', userMessage);
                userInput.value = '';

                fetch(`/autocomplete?q=${encodeURIComponent(userMessage)}`)
                .then(response => {
                    if (response.status === 503) {
                        return response.json().then(data => { throw new Error(data.detail); });
                    }
                    return response.json();
                })
                .then(data => {
                    const botMessages = data.result;
                    botMessages.forEach(message => {
                        saveMessage('bot', message);
                        appendMessage('bot', message);
                    });
                })
                .catch(error => {
                    console.error('Error:', error);
                    appendMessage('bot', error.message || 'An error occurred. Please try again later.');
                });
            }

            function retryLastMessage() {
                const lastUserMessage = document.querySelector('.user-message:last-of-type');
                if (lastUserMessage) {
                    userInput.value = lastUserMessage.innerText;
                    sendMessage();
                }
            }
        </script>
    </body>
    </html>
    """
    return HTMLResponse(content=html_code, status_code=200)

@app.get('/autocomplete')
async def autocomplete(q: str = Query(..., title='query')):
    global message_history, pipeline_instance, initialization_complete
    message_history.append(('user', q))
    
    if not initialization_complete:
        logger.warning("Model is not initialized yet.")
        raise HTTPException(status_code=503, detail="Model is not initialized yet. Please try again later.")
    
    try:
        response = pipeline_instance(q, max_length=50, num_return_sequences=1)[0]['generated_text']
        logger.debug(f"Successfully autocomplete, q:{q}, res:{response}")
        return {"result": [response]}
    except Exception as e:
        logger.error(f"Ignored error in autocomplete: {e}", exc_info=True)
        raise HTTPException(status_code=500, detail="An error occurred while processing your request.")

if __name__ == '__main__':
    port = int(os.getenv("PORT", 7860))
    uvicorn.run(app=app, host='0.0.0.0', port=port)