Metadata-Version: 2.4
Name: tktkt
Version: 2024.6.1
Summary: Tokeniser toolkit: a collection of Pythonic subword tokenisers and supporting tools.
Author-email: Thomas Bauwens <thomas.bauwens@kuleuven.be>
Maintainer-email: Thomas Bauwens <thomas.bauwens@kuleuven.be>
License:                                  Apache License
                                   Version 2.0, January 2004
                                http://www.apache.org/licenses/
        
           TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
        
           1. Definitions.
        
              "License" shall mean the terms and conditions for use, reproduction,
              and distribution as defined by Sections 1 through 9 of this document.
        
              "Licensor" shall mean the copyright owner or entity authorized by
              the copyright owner that is granting the License.
        
              "Legal Entity" shall mean the union of the acting entity and all
              other entities that control, are controlled by, or are under common
              control with that entity. For the purposes of this definition,
              "control" means (i) the power, direct or indirect, to cause the
              direction or management of such entity, whether by contract or
              otherwise, or (ii) ownership of fifty percent (50%) or more of the
              outstanding shares, or (iii) beneficial ownership of such entity.
        
              "You" (or "Your") shall mean an individual or Legal Entity
              exercising permissions granted by this License.
        
              "Source" form shall mean the preferred form for making modifications,
              including but not limited to software source code, documentation
              source, and configuration files.
        
              "Object" form shall mean any form resulting from mechanical
              transformation or translation of a Source form, including but
              not limited to compiled object code, generated documentation,
              and conversions to other media types.
        
              "Work" shall mean the work of authorship, whether in Source or
              Object form, made available under the License, as indicated by a
              copyright notice that is included in or attached to the work
              (an example is provided in the Appendix below).
        
              "Derivative Works" shall mean any work, whether in Source or Object
              form, that is based on (or derived from) the Work and for which the
              editorial revisions, annotations, elaborations, or other modifications
              represent, as a whole, an original work of authorship. For the purposes
              of this License, Derivative Works shall not include works that remain
              separable from, or merely link (or bind by name) to the interfaces of,
              the Work and Derivative Works thereof.
        
              "Contribution" shall mean any work of authorship, including
              the original version of the Work and any modifications or additions
              to that Work or Derivative Works thereof, that is intentionally
              submitted to Licensor for inclusion in the Work by the copyright owner
              or by an individual or Legal Entity authorized to submit on behalf of
              the copyright owner. For the purposes of this definition, "submitted"
              means any form of electronic, verbal, or written communication sent
              to the Licensor or its representatives, including but not limited to
              communication on electronic mailing lists, source code control systems,
              and issue tracking systems that are managed by, or on behalf of, the
              Licensor for the purpose of discussing and improving the Work, but
              excluding communication that is conspicuously marked or otherwise
              designated in writing by the copyright owner as "Not a Contribution."
        
              "Contributor" shall mean Licensor and any individual or Legal Entity
              on behalf of whom a Contribution has been received by Licensor and
              subsequently incorporated within the Work.
        
           2. Grant of Copyright License. Subject to the terms and conditions of
              this License, each Contributor hereby grants to You a perpetual,
              worldwide, non-exclusive, no-charge, royalty-free, irrevocable
              copyright license to reproduce, prepare Derivative Works of,
              publicly display, publicly perform, sublicense, and distribute the
              Work and such Derivative Works in Source or Object form.
        
           3. Grant of Patent License. Subject to the terms and conditions of
              this License, each Contributor hereby grants to You a perpetual,
              worldwide, non-exclusive, no-charge, royalty-free, irrevocable
              (except as stated in this section) patent license to make, have made,
              use, offer to sell, sell, import, and otherwise transfer the Work,
              where such license applies only to those patent claims licensable
              by such Contributor that are necessarily infringed by their
              Contribution(s) alone or by combination of their Contribution(s)
              with the Work to which such Contribution(s) was submitted. If You
              institute patent litigation against any entity (including a
              cross-claim or counterclaim in a lawsuit) alleging that the Work
              or a Contribution incorporated within the Work constitutes direct
              or contributory patent infringement, then any patent licenses
              granted to You under this License for that Work shall terminate
              as of the date such litigation is filed.
        
           4. Redistribution. You may reproduce and distribute copies of the
              Work or Derivative Works thereof in any medium, with or without
              modifications, and in Source or Object form, provided that You
              meet the following conditions:
        
              (a) You must give any other recipients of the Work or
                  Derivative Works a copy of this License; and
        
              (b) You must cause any modified files to carry prominent notices
                  stating that You changed the files; and
        
              (c) You must retain, in the Source form of any Derivative Works
                  that You distribute, all copyright, patent, trademark, and
                  attribution notices from the Source form of the Work,
                  excluding those notices that do not pertain to any part of
                  the Derivative Works; and
        
              (d) If the Work includes a "NOTICE" text file as part of its
                  distribution, then any Derivative Works that You distribute must
                  include a readable copy of the attribution notices contained
                  within such NOTICE file, excluding those notices that do not
                  pertain to any part of the Derivative Works, in at least one
                  of the following places: within a NOTICE text file distributed
                  as part of the Derivative Works; within the Source form or
                  documentation, if provided along with the Derivative Works; or,
                  within a display generated by the Derivative Works, if and
                  wherever such third-party notices normally appear. The contents
                  of the NOTICE file are for informational purposes only and
                  do not modify the License. You may add Your own attribution
                  notices within Derivative Works that You distribute, alongside
                  or as an addendum to the NOTICE text from the Work, provided
                  that such additional attribution notices cannot be construed
                  as modifying the License.
        
              You may add Your own copyright statement to Your modifications and
              may provide additional or different license terms and conditions
              for use, reproduction, or distribution of Your modifications, or
              for any such Derivative Works as a whole, provided Your use,
              reproduction, and distribution of the Work otherwise complies with
              the conditions stated in this License.
        
           5. Submission of Contributions. Unless You explicitly state otherwise,
              any Contribution intentionally submitted for inclusion in the Work
              by You to the Licensor shall be under the terms and conditions of
              this License, without any additional terms or conditions.
              Notwithstanding the above, nothing herein shall supersede or modify
              the terms of any separate license agreement you may have executed
              with Licensor regarding such Contributions.
        
           6. Trademarks. This License does not grant permission to use the trade
              names, trademarks, service marks, or product names of the Licensor,
              except as required for reasonable and customary use in describing the
              origin of the Work and reproducing the content of the NOTICE file.
        
           7. Disclaimer of Warranty. Unless required by applicable law or
              agreed to in writing, Licensor provides the Work (and each
              Contributor provides its Contributions) on an "AS IS" BASIS,
              WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
              implied, including, without limitation, any warranties or conditions
              of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
              PARTICULAR PURPOSE. You are solely responsible for determining the
              appropriateness of using or redistributing the Work and assume any
              risks associated with Your exercise of permissions under this License.
        
           8. Limitation of Liability. In no event and under no legal theory,
              whether in tort (including negligence), contract, or otherwise,
              unless required by applicable law (such as deliberate and grossly
              negligent acts) or agreed to in writing, shall any Contributor be
              liable to You for damages, including any direct, indirect, special,
              incidental, or consequential damages of any character arising as a
              result of this License or out of the use or inability to use the
              Work (including but not limited to damages for loss of goodwill,
              work stoppage, computer failure or malfunction, or any and all
              other commercial damages or losses), even if such Contributor
              has been advised of the possibility of such damages.
        
           9. Accepting Warranty or Additional Liability. While redistributing
              the Work or Derivative Works thereof, You may choose to offer,
              and charge a fee for, acceptance of support, warranty, indemnity,
              or other liability obligations and/or rights consistent with this
              License. However, in accepting such obligations, You may act only
              on Your own behalf and on Your sole responsibility, not on behalf
              of any other Contributor, and only if You agree to indemnify,
              defend, and hold each Contributor harmless for any liability
              incurred by, or claims asserted against, such Contributor by reason
              of your accepting any such warranty or additional liability.
        
           END OF TERMS AND CONDITIONS
        
           APPENDIX: How to apply the Apache License to your work.
        
              To apply the Apache License to your work, attach the following
              boilerplate notice, with the fields enclosed by brackets "[]"
              replaced with your own identifying information. (Don't include
              the brackets!)  The text should be enclosed in the appropriate
              comment syntax for the file format. We also recommend that a
              file or class name and description of purpose be included on the
              same "printed page" as the copyright notice for easier
              identification within third-party archives.
        
           Copyright [yyyy] [name of copyright owner]
        
           Licensed under the Apache License, Version 2.0 (the "License");
           you may not use this file except in compliance with the License.
           You may obtain a copy of the License at
        
               http://www.apache.org/licenses/LICENSE-2.0
        
           Unless required by applicable law or agreed to in writing, software
           distributed under the License is distributed on an "AS IS" BASIS,
           WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
           See the License for the specific language governing permissions and
           limitations under the License.
        
Keywords: NLP,tokenizers,tokenization,subwords,segmentation,natural language
Classifier: Programming Language :: Python :: 3
Classifier: Operating System :: OS Independent
Requires-Python: >=3.9
Description-Content-Type: text/markdown
License-File: LICENSE
Requires-Dist: transformers
Requires-Dist: tokenizers
Requires-Dist: datasets
Requires-Dist: evaluate
Requires-Dist: sentencepiece
Requires-Dist: regex
Provides-Extra: github
Dynamic: license-file

# TkTkT
A collection of Pythonic subword tokenisers and text preprocessing tools, with full
backwards- *and* forwards-compatibility with HuggingFace `tokenizers`.

Quick navigation:
- <a href="#installation">Installation</a>
- <a href="#features">Features</a>
- <a href="#examples">Examples</a>
- <a href="#architecture">Architecture</a>

## 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](https://aclanthology.org/P16-1162/)), 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](https://aclanthology.org/2020.acl-main.170/))
  - BPE-knockout (Bauwens & Delobelle, 2024)
  - Trimmed BPE ([Cognetta et al., 2024](https://arxiv.org/abs/2404.00397))
- Unigram language model (ULM), dubbed *KudoPiece* in TkTkT ([Kudo, 2018](https://aclanthology.org/P18-1007/)):
  - Wrapper around the [SentencePiece](https://github.com/google/sentencepiece) package
  - Native implementation in TkTkT
- Greedy tokenisers ([Bauwens, 2023](https://bauwenst.github.io/cdn/doc/pdf/2023/masterthesis.pdf) and later [Uzan et al., 2024](https://arxiv.org/abs/2403.01289)):
  - Left-to-right and right-to-left
  - Random-access, dubbed "FLOTA" by [Hofmann et al., 2022](https://aclanthology.org/2022.acl-short.43/).
- Derivative leverager (DeL) ([Hofmann et al., 2021](https://aclanthology.org/2021.acl-long.279/)).
  - Segmentation
  - Trainer
- Character/byte N-grams.
- Randomised segmentation from a given subword vocabulary.
- A family of Viterbi-driven tokenisers (work in progress).
- Coming soon: 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.

### 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
```shell
pip install "tktkt[github] @ git+https://github.com/bauwenst/TkTkT.git"
```
where you should leave out the `[github]` suffix if you have an editable installation of [`bpe_knockout`](https://github.com/bauwenst/BPE-knockout).

### Editable
If you want to keep TkTkT out of your `site-packages` for easy editing, an editable install is always possible:
```shell
git clone https://github.com/bauwenst/TkTkT.git
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:
```python
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:
```python
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 `CommonsensePreprocessor` 
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.
```python
# 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 it is more technically accurate).

First we call the trainer with relevant training arguments:
```python
from tktkt.models.kudopiece.training import *
from string import ascii_letters

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

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

args_algo = KudoPieceArguments_Algorithm()

trainer = KudoPieceTrainer(
    word_boundary_location=BoundaryMarkerLocation.START,
    final_vocab_size=40_000,
    alphabet_arguments=args_alpha,
    algorithm_arguments=args_algo,
    file_stem="kudopiece_en"
)
model_path = trainer.train_from_iterator(sentence_corpus, strings_need_space_splitting=True)
```
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()`.
```python
from tktkt.models.kudopiece.segmentation import KudoPieceTokeniser
from tktkt.preparation.instances import IdentityMapper, AppendSpace, IdentityPretokeniser, Preprocessor

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

preprocessor = Preprocessor(
    IdentityMapper(),
    AppendSpace(front_not_back=True),
    IdentityPretokeniser()
)
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](https://github.com/huggingface/transformers/issues/26254).
        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](https://github.com/huggingface/transformers/blob/9b5a6450d481b0f02834684ffd8b3ba4cbbd6fe0/src/transformers/models/roberta/tokenization_roberta.py#L194)
    unless the merge file is preceded by a `#version` tag? This doesn't conform to [the BPE standard](https://github.com/rsennrich/subword-nmt/), 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](https://huggingface.co/learn/nlp-course/chapter6/7), 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`](github.com/google/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.

## 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](https://open.spotify.com/track/2aX7w5bdbES8A9H5FDydSA).

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.
