app.py

from flask import Flask, render_template, request, redirect, url_for, send_file, jsonify
from transformers import GPT2LMHeadModel, GPT2Tokenizer
import os
import torch
import zipfile
import pandas as pd
from utils import preprocess_data, train_model

app = Flask(__name__)
app.config["UPLOAD_FOLDER"] = "uploads"
app.config["MODEL_FOLDER"] = "models"

# Initialize device
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

# Load tokenizer and set padding if needed
model_name = "gpt2"
tokenizer = GPT2Tokenizer.from_pretrained(model_name)
tokenizer.add_special_tokens({'pad_token': '[PAD]'})

# Cache for loaded models to avoid redundant loading
loaded_models = {}


@app.route("/")
def home():
    # List available models
    models = [model for model in os.listdir(app.config["MODEL_FOLDER"]) if
              os.path.isdir(os.path.join(app.config["MODEL_FOLDER"], model))]
    return render_template("home.html", models=models)


@app.route("/upload", methods=["POST"])
def upload_file():
    if "file" not in request.files or "model_name" not in request.form:
        return redirect(request.url)

    file = request.files["file"]
    model_name = request.form["model_name"]

    if not file.filename or not model_name:
        return redirect(request.url)

    # Prepare directories and paths
    model_path = os.path.join(app.config["MODEL_FOLDER"], model_name)
    os.makedirs(model_path, exist_ok=True)
    filepath = os.path.join(app.config["UPLOAD_FOLDER"], file.filename)
    file.save(filepath)

    # Load and preprocess data
    try:
        df = pd.read_csv(filepath)
        dataset = preprocess_data(df, tokenizer)
    except Exception as e:
        return f"Data processing error: {e}", 500

    # Train and save model
    try:
        # Clear any previous GPU memory allocation
        torch.cuda.empty_cache()

        model = GPT2LMHeadModel.from_pretrained("gpt2")
        model.resize_token_embeddings(len(tokenizer))
        model.to(device)

        # Train the model
        train_model(model, tokenizer, dataset, model_path)

        # Clear GPU memory right after training
        del model
        torch.cuda.empty_cache()
    except torch.cuda.OutOfMemoryError:
        # Clear memory in case of OOM error and return an appropriate message
        torch.cuda.empty_cache()
        return "CUDA out of memory error. Try a smaller model or reduce batch size.", 500
    except Exception as e:
        return f"Model training error: {e}", 500

    # Zip the model files for download
    model_zip_path = os.path.join(model_path, f"{model_name}.zip")
    with zipfile.ZipFile(model_zip_path, 'w') as model_zip:
        for folder, _, files in os.walk(model_path):
            for file_name in files:
                file_path = os.path.join(folder, file_name)
                model_zip.write(file_path, os.path.relpath(file_path, app.config["MODEL_FOLDER"]))

    return redirect(url_for("home"))


@app.route("/download/<model_name>")
def download_model(model_name):
    model_path = os.path.join(app.config["MODEL_FOLDER"], model_name, f"{model_name}.zip")
    if os.path.exists(model_path):
        return send_file(model_path, as_attachment=True)
    else:
        return "Model not found", 404


@app.route("/chat/<model_name>")
def chat(model_name):
    return render_template("chat.html", model_name=model_name)


@app.route("/generate/<model_name>", methods=["POST"])
def generate_response(model_name):
    prompt = request.json.get("prompt")
    if not prompt:
        return jsonify({"error": "No prompt provided"}), 400

    # Load the model if not already in cache
    if model_name not in loaded_models:
        model_path = os.path.join(app.config["MODEL_FOLDER"], model_name)
        if not os.path.exists(model_path):
            return jsonify({"error": f"Model '{model_name}' not found"}), 404
        try:
            # Clear GPU memory and load the model
            torch.cuda.empty_cache()
            model = GPT2LMHeadModel.from_pretrained(model_path)
            model.to(device)
            loaded_models[model_name] = model
        except Exception as e:
            return jsonify({"error": f"Failed to load model '{model_name}': {str(e)}"}), 500

    # Generate response
    model = loaded_models[model_name]
    try:
        inputs = tokenizer.encode(prompt, return_tensors="pt").to(device)
        outputs = model.generate(
            inputs,
            max_length=50,
            num_return_sequences=1,
            no_repeat_ngram_size=2,
            pad_token_id=tokenizer.eos_token_id
        )
        response = tokenizer.decode(outputs[0], skip_special_tokens=True)
    except torch.cuda.OutOfMemoryError:
        torch.cuda.empty_cache()
        return jsonify({"error": "Out of memory. Try a smaller model or shorter prompt."}), 500
    except Exception as e:
        return jsonify({"error": str(e)}), 500
    finally:
        # Clear GPU memory after generation to avoid leaks
        torch.cuda.empty_cache()

    return jsonify({"response": response})


if __name__ == "__main__":
    os.makedirs(app.config["UPLOAD_FOLDER"], exist_ok=True)
    os.makedirs(app.config["MODEL_FOLDER"], exist_ok=True)
    app.run(debug=True)