Skip to main content

Fast and Customizable Tokenizers

Project description



Build GitHub


Tokenizers

Provides an implementation of today's most used tokenizers, with a focus on performance and versatility.

Bindings over the Rust implementation. If you are interested in the High-level design, you can go check it there.

Otherwise, let's dive in!

Main features:

  • Train new vocabularies and tokenize using 4 pre-made tokenizers (Bert WordPiece and the 3 most common BPE versions).
  • Extremely fast (both training and tokenization), thanks to the Rust implementation. Takes less than 20 seconds to tokenize a GB of text on a server's CPU.
  • Easy to use, but also extremely versatile.
  • Designed for research and production.
  • Normalization comes with alignments tracking. It's always possible to get the part of the original sentence that corresponds to a given token.
  • Does all the pre-processing: Truncate, Pad, add the special tokens your model needs.

Installation

With pip:

pip install tokenizers

From sources:

To use this method, you need to have the Rust installed:

# Install with:
curl https://sh.rustup.rs -sSf | sh -s -- -y
export PATH="$HOME/.cargo/bin:$PATH"

Once Rust is installed, you can compile doing the following

git clone https://github.com/huggingface/tokenizers
cd tokenizers/bindings/python

# Create a virtual env (you can use yours as well)
python -m venv .env
source .env/bin/activate

# Install `tokenizers` in the current virtual env
pip install setuptools_rust
python setup.py install

Using the provided Tokenizers

Using a pre-trained tokenizer is really simple:

from tokenizers import CharBPETokenizer

# Initialize a tokenizer
vocab = "./path/to/vocab.json"
merges = "./path/to/merges.txt"
tokenizer = CharBPETokenizer(vocab, merges)

# And then encode:
encoded = tokenizer.encode("I can feel the magic, can you?")
print(encoded.ids)
print(encoded.tokens)

And you can train yours just as simply:

from tokenizers import CharBPETokenizer

# Initialize a tokenizer
tokenizer = CharBPETokenizer()

# Then train it!
tokenizer.train([ "./path/to/files/1.txt", "./path/to/files/2.txt" ])

# And you can use it
encoded = tokenizer.encode("I can feel the magic, can you?")

# And finally save it somewhere
tokenizer.save("./path/to/directory", "my-bpe")

Provided Tokenizers

  • CharBPETokenizer: The original BPE
  • ByteLevelBPETokenizer: The byte level version of the BPE
  • SentencePieceBPETokenizer: A BPE implementation compatible with the one used by SentencePiece
  • BertWordPieceTokenizer: The famous Bert tokenizer, using WordPiece

All of these can be used and trained as explained above!

Build your own

You can also easily build your own tokenizers, by putting all the different parts you need together:

Use a pre-trained tokenizer

from tokenizers import Tokenizer, models, pre_tokenizers, decoders, processors

# Load a BPE Model
vocab = "./path/to/vocab.json"
merges = "./path/to/merges.txt"
bpe = models.BPE.from_files(vocab, merges)

# Initialize a tokenizer
tokenizer = Tokenizer(bpe)

# Customize pre-tokenization and decoding
tokenizer.pre_tokenizer = pre_tokenizers.ByteLevel(add_prefix_space=True)
tokenizer.decoder = decoders.ByteLevel()
tokenizer.post_processor = processors.ByteLevel(trim_offsets=True)

# And then encode:
encoded = tokenizer.encode("I can feel the magic, can you?")
print(encoded.ids)
print(encoded.tokens)

# Or tokenize multiple sentences at once:
encoded = tokenizer.encode_batch([
	"I can feel the magic, can you?",
	"The quick brown fox jumps over the lazy dog"
])
print(encoded)

Train a new tokenizer

from tokenizers import Tokenizer, models, pre_tokenizers, decoders, trainers, processors

# Initialize a tokenizer
tokenizer = Tokenizer(models.BPE.empty())

# Customize pre-tokenization and decoding
tokenizer.pre_tokenizer = pre_tokenizers.ByteLevel(add_prefix_space=True)
tokenizer.decoder = decoders.ByteLevel()
tokenizer.post_processor = processors.ByteLevel(trim_offsets=True)

# And then train
trainer = trainers.BpeTrainer(vocab_size=20000, min_frequency=2)
tokenizer.train(trainer, [
	"./path/to/dataset/1.txt",
	"./path/to/dataset/2.txt",
	"./path/to/dataset/3.txt"
])

# Now we can encode
encoded = tokenizer.encode("I can feel the magic, can you?")
print(encoded)

Project details


Release history Release notifications | RSS feed

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

tokenizers-0.7.0rc2.tar.gz (78.8 kB view hashes)

Uploaded Source

Built Distributions

tokenizers-0.7.0rc2-cp38-cp38-win_amd64.whl (1.1 MB view hashes)

Uploaded CPython 3.8 Windows x86-64

tokenizers-0.7.0rc2-cp38-cp38-win32.whl (944.7 kB view hashes)

Uploaded CPython 3.8 Windows x86

tokenizers-0.7.0rc2-cp38-cp38-manylinux1_x86_64.whl (7.4 MB view hashes)

Uploaded CPython 3.8

tokenizers-0.7.0rc2-cp38-cp38-macosx_10_10_x86_64.whl (1.2 MB view hashes)

Uploaded CPython 3.8 macOS 10.10+ x86-64

tokenizers-0.7.0rc2-cp37-cp37m-win_amd64.whl (1.1 MB view hashes)

Uploaded CPython 3.7m Windows x86-64

tokenizers-0.7.0rc2-cp37-cp37m-win32.whl (944.7 kB view hashes)

Uploaded CPython 3.7m Windows x86

tokenizers-0.7.0rc2-cp37-cp37m-manylinux1_x86_64.whl (5.5 MB view hashes)

Uploaded CPython 3.7m

tokenizers-0.7.0rc2-cp37-cp37m-macosx_10_10_x86_64.whl (1.2 MB view hashes)

Uploaded CPython 3.7m macOS 10.10+ x86-64

tokenizers-0.7.0rc2-cp36-cp36m-win_amd64.whl (1.1 MB view hashes)

Uploaded CPython 3.6m Windows x86-64

tokenizers-0.7.0rc2-cp36-cp36m-win32.whl (945.0 kB view hashes)

Uploaded CPython 3.6m Windows x86

tokenizers-0.7.0rc2-cp36-cp36m-manylinux1_x86_64.whl (3.7 MB view hashes)

Uploaded CPython 3.6m

tokenizers-0.7.0rc2-cp36-cp36m-macosx_10_10_x86_64.whl (1.2 MB view hashes)

Uploaded CPython 3.6m macOS 10.10+ x86-64

tokenizers-0.7.0rc2-cp35-cp35m-win_amd64.whl (1.1 MB view hashes)

Uploaded CPython 3.5m Windows x86-64

tokenizers-0.7.0rc2-cp35-cp35m-win32.whl (945.0 kB view hashes)

Uploaded CPython 3.5m Windows x86

tokenizers-0.7.0rc2-cp35-cp35m-manylinux1_x86_64.whl (1.9 MB view hashes)

Uploaded CPython 3.5m

tokenizers-0.7.0rc2-cp35-cp35m-macosx_10_10_x86_64.whl (1.2 MB view hashes)

Uploaded CPython 3.5m macOS 10.10+ x86-64

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page