Python API with C extensions for faster training and encoding

[benarnav]

I've been working on a C version which is based on the GPT2 paper: bytephase

It provides a Python API with C extensions to significantly speed up training and encoding. The project has a thorough README and complete docstring for all class methods.
bytephase will be a good addition considering the 2nd point in todo section: "write an even more optimized C or Rust version (think through)"

I'd like to add bytephase to the community extension section of the README, encouraging more developers to review and contribute to this implementation and possibly build more features into it (ex. GPT4 support, loading other pretrained tokenizers).

Example usage:

from bytephase import Tokenizer

# Initialize and train
tokenizer = Tokenizer()
# OR select a custom regex pattern, defaults to the GPT2 pattern
custom_pattern = r'\w+|\s+|[^\w\s]+'
tokenizer = Tokenizer(pattern=custom_pattern)

tokenizer.train("path/to/your_data.txt", vocab_size=10000)

# Encode
encoded = tokenizer.encode("Hello, world!")
# [1869, 574, 111, 44, 1560, 33]

# Decode
decoded = tokenizer.decode(encoded)
# "Hello, world!"

# Save and load
tokenizer.save("saved_tokenizer")
tokenizer.load("saved_tokenizer.bpe")