tokenizers 0.23.1rc0

No project description provided

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 -e .

Free-threaded Python (3.14t)

tokenizers ships dedicated wheels for the free-threaded build of CPython (python3.14t). These wheels declare Py_MOD_GIL_NOT_USED, so importing tokenizers does not force the GIL back on — multi-threaded code stays GIL-free.

The full mutable API works on 3.14t — the same as on regular CPython. Setters are thread-safe: the inner tokenizer state is wrapped in a std::sync::RwLock, so concurrent tokenizer.X = … from multiple threads serialize correctly and concurrent encode operations take a read guard that blocks writers only briefly.

from tokenizers import Tokenizer
from tokenizers.models import BPE
from tokenizers.pre_tokenizers import Whitespace
from tokenizers.processors import ByteLevel

tok = Tokenizer(BPE())
tok.pre_tokenizer = Whitespace()                 # ✅ thread-safe on 3.14t
tok.post_processor = ByteLevel(trim_offsets=True)

Caveat — compound mutations are not atomic. Statements like tokenizer.post_processor.special_tokens = X evaluate in two steps from Python's point of view (read attribute → set attribute on the result). If another thread swaps tokenizer.post_processor between those steps, the mutation lands on an orphaned component. This is the same class of race as dict[k] = v interleaved with dict.clear() — coordinate with a Python lock if you need the compound to be atomic.

For the full thread-safety analysis, see docs/free-threading-audit.md.

Load a pretrained tokenizer from the Hub

from tokenizers import Tokenizer

tokenizer = Tokenizer.from_pretrained("bert-base-cased")

Using the provided Tokenizers

We provide some pre-build tokenizers to cover the most common cases. You can easily load one of these using some vocab.json and merges.txt files:

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 them 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" ])

# Now, let's use it:
encoded = tokenizer.encode("I can feel the magic, can you?")

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

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

Whenever these provided tokenizers don't give you enough freedom, you can build your own tokenizer, by putting all the different parts you need together. You can check how we implemented the provided tokenizers and adapt them easily to your own needs.

Building a byte-level BPE

Here is an example showing how to build your own byte-level BPE by putting all the different pieces together, and then saving it to a single file:

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

# Initialize a tokenizer
tokenizer = Tokenizer(models.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 train
trainer = trainers.BpeTrainer(
    vocab_size=20000,
    min_frequency=2,
    initial_alphabet=pre_tokenizers.ByteLevel.alphabet()
)
tokenizer.train([
    "./path/to/dataset/1.txt",
    "./path/to/dataset/2.txt",
    "./path/to/dataset/3.txt"
], trainer=trainer)

# And Save it
tokenizer.save("byte-level-bpe.tokenizer.json", pretty=True)

Now, when you want to use this tokenizer, this is as simple as:

from tokenizers import Tokenizer

tokenizer = Tokenizer.from_file("byte-level-bpe.tokenizer.json")

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

Typing support and stub generation

The compiled PyO3 extension does not expose type annotations, so editors and type checkers would otherwise see most objects as Any. To provide full typing support, we use a two-step stub generation process:

1. Rust introspection (tools/stub-gen/): Uses pyo3-introspection to analyze the compiled extension and generate .pyi stub files
2. Python enrichment (stub.py): Adds docstrings from the runtime module and generates forwarding __init__.py shims

Running stub generation

The easiest way to regenerate stubs is via make style:

cd bindings/python
make style

This will:

1. Build the extension with maturin develop --release
2. Run introspection to generate .pyi files
3. Enrich stubs with docstrings via stub.py
4. Format with ruff

Running manually

To run the stub generator directly:

cd bindings/python
cargo run --manifest-path tools/stub-gen/Cargo.toml
python stub.py

The stub generator automatically:

• Builds the extension using maturin
• Copies the built .so to the project root for introspection
• Detects and sets PYTHONHOME for embedded Python (handles uv/venv environments)
• Generates stubs to py_src/tokenizers/

Troubleshooting

If you encounter Python initialization errors, you can manually set PYTHONHOME:

export PYTHONHOME=$(python3 -c 'import sys; print(sys.base_prefix)')
cargo run --manifest-path tools/stub-gen/Cargo.toml