TkTkT

A collection of Pythonic subword tokenisers and text preprocessing tools, with full backwards- and forwards-compatibility with HuggingFace tokenizers!

Quick navigation:

• Installation
• Features
• Examples
• Architecture

Features

Supported tokenisers

All subword tokenisers are defined under tktkt.models. Many of these can be instantiated without much background knowledge using the builders in tktkt.builders. Also, any HuggingFace tokeniser can be wrapped into a TkTkT tokeniser, and any TkTkT tokeniser can be wrapped into a HuggingFace tokeniser.

Currently, the package implements:

• Byte-pair encoding (BPE) tokenisers:
  • Classical BPE (Sennrich et al., 2016), with added support for n-ary merges (byte-tuple encoding, BTE) and any word boundary marker (start-of-words like GPT-2 and end-of-words like Sennrich).
  • BPE-dropout (Provilkov et al., 2020)
  • BPE-knockout (Bauwens & Delobelle, 2024)
  • TrimmedBPE (Cognetta et al., 2024)
  • Other variants:
    • EnsuredBPE: BPE where the last merges have been replaced by the merges necessary to ensure that a given list of strings is in the vocabulary.
    • ShuffledBPE: BPE but with merge priorities shuffled, although types are never shuffled to a priority before the ancestors in their merge tree.
• Unigram language model (ULM), dubbed KudoPiece in TkTkT (Kudo, 2018):
  • Wrapper around the SentencePiece package
  • Native implementation in TkTkT
• Greedy tokenisers (Bauwens, 2023 and later Uzan et al., 2024):
  • Left-to-right and right-to-left
  • Random-access, dubbed "FLOTA" by Hofmann et al., 2022.
• Derivative leverager (DeL) (Hofmann et al., 2021).
  • Segmentation
  • Trainer
• Character/byte N-grams.
• Randomised segmentation from a given subword vocabulary.

Currently work in progress:

• A family of Viterbi-driven tokenisers.
• Morfessor family

Evaluation metrics

TkTkT currently supports the following intrinsic tokeniser evaluation metrics:

• Fertility statistics: how many tokens the tokeniser produces per word, and how many segmentations its vocabulary could produce in theory.
• Morphological boundary recognition: using the tokeniser as a binary classifier for whether two morphemes meet at each position in a word.
• Richness of token contexts with accessor variety.
• Entropy-based measures, including Rényi efficiency.
• Comparison between two tokenisers: how much they tokenise words exactly the same, and how much their split points overlap.

Visualisers

The following tokenisation procedures can be visualised:

• BPE/BTE: the final merge tree (in regular LaTeX), as well as an animated progression of the merges (in LaTeX Beamer).

Architecture

The goal of TkTkT is to provide a straightforward Pythonic interface for everything-tokenisation, and to be as object-oriented as possible. The main interfaces are found under tktkt.interfaces.

Fundamentally, all tokenisers are a Tokeniser that have a Preprocessor.

• The Tokeniser class has two important methods:

  • .tokenise(pretoken: str) -> List[str]: segments a string as-is into parts.
  • .prepareAndTokenise(text: str) -> List[str]: applies the tokeniser's preprocessor and then tokenises each pre-token separately.

• The Preprocessor class is a pipeline of three components:

  1. a non-invertible text mapping
  2. an invertible text mapping
  3. a pretokeniser that splits strings into smaller strings.

To map tokens (string segments) to identifiers (integers) for indexing into an embedding matrix, three interfaces are supported:

• TokeniserWithFiniteIdRange: tokenisers with a finite range of IDs, but an infinite domain of tokens that map into that range. An example is a hashing vocabulary.
• TokeniserWithFiniteTypeDomain: same as above, but only supports mapping a predetermined set of unique tokens (types). No assumption is made about how this mapping happens. HuggingFace's PreTrainedTokenizer is an example of these.
• TokeniserWithVocabDict: same as above, but the mapping is made by an explicit dictionary.

TkTkT also has wrapper classes (in tktkt.wrappers) to extend existing tokenisers with an ID mapping.

Installation

Non-editable (recommended)

Simply run

pip install "tktkt[github] @ git+https://github.com/bauwenst/TkTkT"

where you should leave out the [github] suffix if you have an editable installation of bpe_knockout.

Editable

If you want to keep TkTkT out of your site-packages for easy editing, an editable install is always possible:

git clone https://github.com/bauwenst/TkTkT
cd TkTkT
pip install -e .[github]

where the same caveat applies about the [github] suffix.

Examples

Basic usage

Let's first instantiate a toy preprocessor in TkTkT:

from tktkt.preparation.instances import Preprocessor, KudoSpaceMarker, \
    Lowercaser, Replace, \
    PretokeniserSequence, WhitespacePretokeniser, PunctuationPretokeniser, AddWordBoundary

toy_preprocessor = Preprocessor(
    Lowercaser(),
    Replace("!", "."),
    PretokeniserSequence([
        WhitespacePretokeniser(),
        PunctuationPretokeniser(),
        AddWordBoundary(KudoSpaceMarker)
    ])
)

print(toy_preprocessor.do("This example will be preprocessed (even without a tokeniser)!"))

Now we instantiate a greedy TkTkT tokeniser with that preprocessor:

from tktkt.models.greedy.directional import L2R_Greedy

tokeniser = L2R_Greedy(
    preprocessor=toy_preprocessor,
    vocab={"a": 0, "b": 1, "c": 2, "d": 3, "ab": 4, "ba": 5, ".": 6, ",": 7, "▁": 8}
)

print(tokeniser.prepareAndTokenise("A bad cab, ABBA!"))
print(tokeniser.tokenise("abc."))

There are many more preprocessing classes available, some pre-made. Check out the ModernEnglishPreprocessor for typical modern use-cases.

HuggingFace compatibility

In the example below, a BPE tokeniser is loaded from the HuggingFace hub as a PreTrainedTokenizerFast and converted into a TkTkT Tokeniser object. Then, this object is itself converted into a HuggingFace PreTrainedTokenizer again.

# Backwards-compatibility:
from transformers import AutoTokenizer
from tktkt.models.huggingface.wrapper import HuggingFaceTokeniser

hf_roberta = AutoTokenizer.from_pretrained("roberta-base")
tktkt_roberta = HuggingFaceTokeniser(hf_roberta)

###
sentence = " That's so supercalifragilisticexpialidocious, Günther!"
print("Full tokenisation pipeline:")
print("\tHF Tk:", hf_roberta.tokenize(sentence))  # Note the lack of autocompletion on this.
print("\tTkTkT:", tktkt_roberta.prepareAndTokenise(sentence))
print("Only the preprocessing:")
print("\tTkTkT:", tktkt_roberta.preprocessor.do(sentence))
###

# Forwards-compatibility:
from tktkt.interfaces.huggingface import TktktToHuggingFace

hf_tktkt_roberta = TktktToHuggingFace(tktkt_roberta, specials_from=hf_roberta)
print(hf_tktkt_roberta.tokenize(sentence))

KudoPiece (ULM)

Let's say you want to train and load an English ULM tokeniser. You are, of course, scared of the sentencepiece library because its Python interface is a thin wrapper around a command-line call, not allowing autocompletion in your IDE. In TkTkT, you would proceed as follows (note that ULM is called "KudoPiece" in TkTkT because many tokenisers are based on a language model of unigrams).

First we call the trainer with relevant training arguments:

from tktkt.preparation.instances import IdentityMapper, AppendSpace, IdentityPretokeniser, Preprocessor
from tktkt.models.kudopiece.vocabularisation import *
from string import ascii_letters

### Your data iterator goes here.
sentence_corpus: Iterable[str] = ...
###

preprocessor = Preprocessor(
    IdentityMapper(),
    AppendSpace(front_not_back=True),
    IdentityPretokeniser()
)

args_alpha = KudoPieceArguments_Alphabet(
    required_chars=list(ascii_letters),
    byte_fallback=True,
    character_coverage=0.9995
)

args_algo = KudoPieceArguments_Algorithm()

trainer = KudoPieceTrainer(
    preprocessor=preprocessor,
    word_boundary_location=BoundaryMarkerLocation.START,
    final_vocab_size=40_000,
    alphabet_arguments=args_alpha,
    algorithm_arguments=args_algo,
    file_stem="tutorial"
)
model_path = trainer.vocabulariseFromStringIterable(sentence_corpus)

Once the final model is stored to disk, we can load it as an object (and give it a basic preprocessor). Note that all models are stored under tktkt.files.paths.DataPaths.pathToModels().

from tktkt.models.kudopiece.segmentation import KudoPieceTokeniser

# # If you need to recover the path:
# from tktkt.files.paths import TkTkTPaths
# model_path = TkTkTPaths.pathToModels() / "kudopiece" / "tutorial_xxxx-yy-zz_aa-bb-cc.model"

tokeniser = KudoPieceTokeniser(preprocessor=preprocessor, model_file=model_path)

print(tokeniser.prepareAndTokenise("Hello there, my good friend!"))

Why does this exist if we have HuggingFace tokenizers?

First of all, note again that TkTkT has backwards compatibility with HuggingFace tokenizers. There are wrapper classes for tokenisers under tktkt.models.huggingface and for normalisers/pretokenisers under tktkt.preparation.huggingface.

Here's a non-exhaustive list of reasons:

1. The HuggingFace tokenizers library has horrifically un(der)documented Python interfaces. Some classes even accept arguments that aren't in their signature.
2. The tokenizers library is implemented in Rust and hence there is no possibility of inspecting implementations in any Python IDE. Have fun using your black box.
3. The tokenizers interface does not allow separating preprocessing from the actual tokenisation algorithm.
   • The PreTrainedTokenizerBase class, from which the "slow" (Pythonic) PreTrainedTokenizer and "fast" (Rustic) PreTrainedTokenizerFast classes both inherit, only declares an end-to-end .tokenize() method (equivalent to TkTkT's .prepareAndTokenise()). The interface for these subclasses is different enough that both lack features of the other:
     • Whereas PreTrainedTokenizer does declare a ._tokenize() (equivalent to TkTkT's .tokenise()), I challenge you to find the equivalent for PreTrainedTokenizerFast. Best you'll find is .backend_tokenizer.model.tokenize(), which outputs unusable objects of class tokenizers.Token.
     • Whereas PreTrainedTokenizerFast has fields .backend_tokenizer.pre_tokenizer and .backend_tokenizer.normalizer (untyped of course, so you can't get autocompletion on their methods unless you manually assign them to a variable and annotate it yourself), PreTrainedTokenizer has no access to a pretokeniser. Preprocessing has to be defined inside ._tokenize(), which means you're doing two steps of preprocessing (one inside .tokenize() and one inside ._tokenize()) making this ._tokenize() no longer equivalent to TkTkT's .tokenise().
     • For PreTrainedTokenizerFast, the .backend_tokenizer.pre_tokenizer and .backend_tokenizer.normalizer fields can both be None, rather than an identity transform like in TkTkT, meaning you always have to check if they exist. Also funny: even when they are not None, you can't check if they exist with a simple if t.backend_tokenizer.normalizer: ... because somehow that's always False.
   • Also, the PreTrainedTokenizerBase interface is not defined with @abstractmethod but with an ever-increasing amount of raise NotImplementedError methods. In other words: it's hard to know which methods need to be implemented and there's no enforcement mechanism to ensure everything has been implemented.
4. The tokenizers.pre_tokenizers submodule has technical debt that can't be patched. Some examples:
   • The mapping from Unicode codepoints to UTF-8 bytes, as first used in GPT-2, is only implemented in the ByteLevel pretokeniser. Yet, it is concerned with more than this, since it splits on spaces and punctuation (optionally prefixed by a space) before applying the mapping. This is wrong for at least three reasons:
     • Users of the byte mapping don't necessary want the string to be split;
     • It synonymises prefixed spaces (converted to Ġ) with start-of-word boundaries whilst actually all words (even those directly preceded by punctuation) should be marked with such a boundary;
     • It assumes that such boundaries should always be at the start of a word.
   • The GPT-2 convention of having a word boundary at the start of (almost) all words is hardcoded throughout transformers and tokenizers (with options that commonly look like add_prefix_space) even though the original BPE paper used word boundaries at the end of words (</w>). Only supporting the start-of-word convention is bad because this deteriorates downstream performance for e.g. Germanic languages, where a compound has its head at the end and hence it should be allowed to tokenise the head with the exact same tokens as it would be if it was isolated.
   • There is literally a normaliser class called Precompiled which is just one big object stored in base64 in the tokeniser config JSON. No access to it in Python, no interface, no description of what it does. A black box. Probably a holdover from adapting the sentencepiece package to HuggingFace, yet TkTkT doesn't do it that way.
5. Did you know that their RoBERTa BPE implementation removes the highest-priority merge from the tokeniser unless the merge file is preceded by a #version tag? This doesn't conform to the BPE standard, and almost cost me a paper.
6. In the little documentation that does exist (e.g. for WordPiece and KudoPiece), there are so many theoretical inaccuracies that we shouldn't even have confidence in anything that isn't a BPE tokeniser implemented by them. Their explanation for KudoPiece, an algorithm which itself was already poorly explained originally, is mathematically absurd.
7. They offer very few core models (basically only BPE and KudoPiece, which sentencepiece already offers and keeps much more updated) whilst there exist many more in the literature, and the likelihood that someone who knows the literature comes along to implement all of them in C++ is rather low.

There is also the pyonmttok package which has better design, but also sticks to BPE and KudoPiece.

Pronunciation

The acronym stands for ToKeniser ToolKiT and is supposed to be pronounced fast, like a beatboxer mimicking hi-hats (kind of like "tuh-kuh-tuh-kuh-ts" but as fast as you can). It is mandatory that you do this, because I said so. If you are Brazilian, you may pronounce it "tuca tuca" while playing the official TkTkT theme song, which gives you a peek into my demented taste in music.

If you think all of this is cringe, too bad -- it is the only way not to fall into clinical depression after realising just how messy modern subword tokenisation is at the algorithmic and an implementational level.