SmolBPE

Overview

SmolBPE is a lightweight and efficient Byte Pair Encoding (BPE) tokenizer designed for deep learning applications and large language models (LLMs) such as GPT-4. It provides a simple interface to tokenize textual data, facilitating better handling of out-of-vocabulary words and improving the performance of language models.

Features

• Efficient Tokenization: Implements the BPE algorithm for effective subword tokenization.
• Customizable Vocabulary Size: Allows you to specify the desired vocabulary size according to your needs.
• Unicode Support: Handles a wide range of characters, including Unicode characters, enabling multilingual tokenization.
• Easy Integration: Designed for seamless integration with existing Python projects and NLP pipelines.
• Command-Line Interface: Provides a CLI tool for training and using the tokenizer without writing additional code.
• Open Source: Licensed under the MIT License, promoting openness and collaboration.

Installation

You can install SmolBPE using pip:

pip install smolbpe

Alternatively, you can install it directly from the source code:

git clone https://github.com/T4ras123/SmolBPE.git
cd SmolBPE
pip install .

Quick Start Guide

Using the Tokenizer in Python

1.Importing the Tokenizer

from smolbpe.tokenizer import Tokenizer

2.Initializing the Tokenizer

tokenizer = Tokenizer()

You can specify a custom output file to save the vocab file to and regex pattern if needed:

tokenizer = Tokenizer(output='vocab.json', pattern=r"\p{L}+|\p{Z}+|\p{N}+|[\p{P}&&[^.]]")

3.Training the Tokenizer

Train the tokenizer on your dataset to build the vocabulary and merge rules:

with open("path_to_your_data", "r", encoding="utf-8") as f:
    text = f.read()

tokenizer.train(text, vocab_size=400)

4.Encoding Text

Convert text into a list of token IDs:

encoded_tokens = tokenizer.encode("Tokenizing isn't real")
print(encoded_tokens)

5.Decoding Tokens

Convert token IDs back into human-readable text:

decoded_text = tokenizer.decode(encoded_tokens)
print(decoded_text)

Command-Line Interface

SmolBPE provides a command-line interface for easy tokenization tasks.

Training the Tokenizer

tokenizer --text smth.txt --vocab_size 400 --output vocab.json 

Advanced Usage

Loading a Pre-trained Vocabulary

If you have a pre-trained vocabulary and merges file, you can load them directly:

tokenizer = Tokenizer()
tokenizer.load_vocab('vocab.json')

Custom Regex Pattern

Customize the tokenization by providing a different regex pattern:

custom_pattern = r"\w+|\s+|[^\s\w]+"
tokenizer = Tokenizer(pattern=custom_pattern)

Custom special tokens

Add custom special tokens that appear in your dataset

special_tokens = ['<|start_text|>', '<|good_luck|>']
tokenizer = Tokenizer(special_tokens=special_tokens)

Project Structure

SmolBPE/
├── smolbpe/
│   ├── __init__.py
│   └── tokenizer.py
├── LICENSE
├── MANIFEST.in
├── README.md
└── setup.py

Contributing

Contributions are welcome! To contribute:

1. Fork the repository on GitHub.
2. Create a new branch for your feature or bug fix.
3. Commit your changes with descriptive commit messages.
4. Push your branch to your forked repository.
5. Open a pull request on the main repository.

Please ensure your code adheres to the project's coding standards and includes appropriate tests.

License

This project is licensed under the MIT License. You are free to use, modify, and distribute this software in accordance with the license.

Contact

For any inquiries or feedback, please contact the author:

• Author: Vover
• Email: [email protected]
• GitHub: T4ras123

Acknowledgments

• Inspired by tokenization techniques used in GPT models.
• Special thanks to the open-source community for continuous support.

---

Happy tokenizing with SmolBPE!