Shakespeare Text Generator

A neural network-based text generator trained on Shakespeare's complete works. The model uses LSTM architecture to generate Shakespeare-style text.

Features

• Downloads Shakespeare's complete works automatically
• Uses PyTorch with CUDA/MPS/CPU support
• Optimized for different platforms (Windows, Mac, Linux)
• Supports Apple Silicon (M-series) processors
• Automatic worker optimization for data loading
• Learning rate scheduling for better training

Requirements

• Python 3.8-3.11 (tested with 3.11.9)
• PyTorch 2.0.1
• CUDA 11.8
• MPS 1.4
• Apple Silicon (M-series) processors
• TensorFlow 2.12.0 (for TF implementation)

pip install -r requirements.txt

Usage

Option 1: Use Pre-trained Model

A pre-trained model (best_model.pt) is included in the repository. To generate text immediately:

python generate.py

Option 2: Train Your Own Model

1. Train the model:

python train.py

This will download Shakespeare's text, preprocess it, and train the model. The model will be saved as best_model.pt. That means you will over write the pre-trained model. If you want to keep the pre-trained model, you can copy that file to a different name.

2. Generate text:

python generate.py

Model Architecture

• LSTM-based neural network
• Character-level text generation
• Embedding layer for character encoding
• Dropout for regularization
• AdamW optimizer with OneCycleLR scheduler

Files

• data_loader.py: Downloads and preprocesses Shakespeare's text
• model.py: Neural network model definition
• train.py: Training script with optimization
• generate.py: Text generation script
• requirements.txt: Required Python packages
• best_model.pt: Pre-trained model

Performance

The model automatically selects the best available hardware:

• CUDA for NVIDIA GPUs
• MPS for Apple Silicon
• CPU for other systems

Data loading is optimized based on the platform:

• Windows: Single worker for stability
• Other platforms: Automatically uses 80% of available CPU cores