GPT-2 Model Setup Guide

Installation Steps

1. Install Git LFS:
   • Mac: Use Homebrew - brew install git-lfs.
   • Linux/Windows: Download from Git LFS website.
2. Clone the GPT-2 repository: git clone https://huggingface.co/gpt2.
3. Go into the folder cd gpt2.
4. Create and activate a virtual environment:
   • Mac/Linux: python -m venv .env then source .env/bin/activate.
   • Windows: python -m venv .env then .env\Scripts\activate.
5. Install required libraries:
   • Install Transformers: pip install transformers.
   • Install PyTorch: Follow instructions at PyTorch Get Started.

Running main.py

Create a file named main.py inside your project folder. Copy and paste the following script into this file:

        
            import torch
            import logging
            from transformers import pipeline, set_seed

            # Set logging level to ERROR to suppress informational messages
            logging.getLogger('transformers').setLevel(logging.ERROR)

            # Function to generate text using the GPT-2 model
            def generate_text(seed_text, max_length=200, num_return_sequences=1):
                generated = generator(seed_text, max_length=max_length, num_return_sequences=num_return_sequences)[0]["generated_text"]

                # Remove the seed text from the beginning of the generated text
                if generated.startswith(seed_text):
                    return generated[len(seed_text):].strip()
                return generated

            # Set the device to MPS if available, else use CPU
            device = "mps" if torch.backends.mps.is_available() else "cpu"

            # Initialize the GPT-2 text-generation pipeline
            generator = pipeline('text-generation', model='gpt2', device=device)

            # Set a seed for reproducibility
            set_seed(420)

            # Get initial input from the user
            seed_text = input("What should the first sentence be?\n")
            max_length = int(input("Enter the maximum length for text generation (e.g., 200): "))
            num_sequences = int(input("Enter the number of sequences to generate (usually 1): "))

            # Generate initial text
            txt = generate_text(seed_text, max_length, num_sequences)
            print("Keep clicking enter to generate more...")
            print(txt)

            # Continue generating text on pressing Enter
            while input("") == "":
                # Dynamically determine the number of words for the new seed
                # based on a fraction of max_length, e.g., half of max_length
                num_words = max_length // 2
                seed_words = txt.split()

                # Ensure we don't exceed the actual number of words in txt
                num_words = min(num_words, len(seed_words))
                new_seed = " ".join(txt.split()[:num_words])

                # Generate more text
                txt = generate_text(new_seed, max_length, num_sequences)
                print(txt)
        
    

Run the script using python main.py in your terminal. If that does not work, try python3 main.py