Release v0.20.0: faster encode, better python support

Release v0.20.0

This release is focused on performances and user experience.

Performances:

First off, we did a bit of benchmarking, and found some place for improvement for us!
With a few minor changes (mostly #1587) here is what we get on Llama3 running on a g6 instances on AWS https://github.com/huggingface/tokenizers/blob/main/bindings/python/benches/test_tiktoken.py :

Python API

We shipped better deserialization errors in general, and support for __str__ and __repr__ for all the object. This allows for a lot easier debugging see this:

>>> from tokenizers import Tokenizer;
>>> tokenizer = Tokenizer.from_pretrained("bert-base-uncased");
>>> print(tokenizer)
Tokenizer(version="1.0", truncation=None, padding=None, added_tokens=[{"id":0, "content":"[PAD]", "single_word":False, "lstrip":False, "rstrip":False, ...}, {"id":100, "content":"[UNK]", "single_word":False, "lstrip":False, "rstrip":False, ...}, {"id":101, "content":"[CLS]", "single_word":False, "lstrip":False, "rstrip":False, ...}, {"id":102, "content":"[SEP]", "single_word":False, "lstrip":False, "rstrip":False, ...}, {"id":103, "content":"[MASK]", "single_word":False, "lstrip":False, "rstrip":False, ...}], normalizer=BertNormalizer(clean_text=True, handle_chinese_chars=True, strip_accents=None, lowercase=True), pre_tokenizer=BertPreTokenizer(), post_processor=TemplateProcessing(single=[SpecialToken(id="[CLS]", type_id=0), Sequence(id=A, type_id=0), SpecialToken(id="[SEP]", type_id=0)], pair=[SpecialToken(id="[CLS]", type_id=0), Sequence(id=A, type_id=0), SpecialToken(id="[SEP]", type_id=0), Sequence(id=B, type_id=1), SpecialToken(id="[SEP]", type_id=1)], special_tokens={"[CLS]":SpecialToken(id="[CLS]", ids=[101], tokens=["[CLS]"]), "[SEP]":SpecialToken(id="[SEP]", ids=[102], tokens=["[SEP]"])}), decoder=WordPiece(prefix="##", cleanup=True), model=WordPiece(unk_token="[UNK]", continuing_subword_prefix="##", max_input_chars_per_word=100, vocab={"[PAD]":0, "[unused0]":1, "[unused1]":2, "[unused2]":3, "[unused3]":4, ...}))

>>> tokenizer
Tokenizer(version="1.0", truncation=None, padding=None, added_tokens=[{"id":0, "content":"[PAD]", "single_word":False, "lstrip":False, "rstrip":False, "normalized":False, "special":True}, {"id":100, "content":"[UNK]", "single_word":False, "lstrip":False, "rstrip":False, "normalized":False, "special":True}, {"id":101, "content":"[CLS]", "single_word":False, "lstrip":False, "rstrip":False, "normalized":False, "special":True}, {"id":102, "content":"[SEP]", "single_word":False, "lstrip":False, "rstrip":False, "normalized":False, "special":True}, {"id":103, "content":"[MASK]", "single_word":False, "lstrip":False, "rstrip":False, "normalized":False, "special":True}], normalizer=BertNormalizer(clean_text=True, handle_chinese_chars=True, strip_accents=None, lowercase=True), pre_tokenizer=BertPreTokenizer(), post_processor=TemplateProcessing(single=[SpecialToken(id="[CLS]", type_id=0), Sequence(id=A, type_id=0), SpecialToken(id="[SEP]", type_id=0)], pair=[SpecialToken(id="[CLS]", type_id=0), Sequence(id=A, type_id=0), SpecialToken(id="[SEP]", type_id=0), Sequence(id=B, type_id=1), SpecialToken(id="[SEP]", type_id=1)], special_tokens={"[CLS]":SpecialToken(id="[CLS]", ids=[101], tokens=["[CLS]"]), "[SEP]":SpecialToken(id="[SEP]", ids=[102], tokens=["[SEP]"])}), decoder=WordPiece(prefix="##", cleanup=True), model=WordPiece(unk_token="[UNK]", continuing_subword_prefix="##", max_input_chars_per_word=100, vocab={"[PAD]":0, "[unused0]":1, "[unused1]":2, ...}))

The pre_tokenizer.Sequence and normalizer.Sequence are also more accessible now:

from tokenizers import normalizers
norm = normalizers.Sequence([normalizers.Strip(), normalizers.BertNormalizer()])
norm[0]
norm[1].lowercase=False

What's Changed

• remove enforcement of non special when adding tokens by @ArthurZucker in #1521
• [BREAKING CHANGE] Ignore added_tokens (both special and not) in the decoder by @Narsil in #1513
• Make USED_PARALLELISM atomic by @nathaniel-daniel in #1532
• Fixing for clippy 1.78 by @Narsil in #1548
• feat(ci): add trufflehog secrets detection by @McPatate in #1551
• Switch from cached_download to hf_hub_download in tests by @Wauplin in #1547
• Fix "dictionnary" typo by @nprisbrey in #1511
• make sure we don't warn on empty tokens by @ArthurZucker in #1554
• Enable dropout = 0.0 as an equivalent to none in BPE by @mcognetta in #1550
• Revert "[BREAKING CHANGE] Ignore added_tokens (both special and not) … by @ArthurZucker in #1569
• Add bytelevel normalizer to fix decode when adding tokens to BPE by @ArthurZucker in #1555
• Fix clippy + feature test management. by @Narsil in #1580
• Bump spm_precompiled to 0.1.3 by @MikeIvanichev in #1571
• Add benchmark vs tiktoken by @Narsil in #1582
• Fixing the benchmark. by @Narsil in #1583
• Tiny improvement by @Narsil in #1585
• Enable fancy regex by @Narsil in #1586
• Fixing release CI strict (taken from safetensors). by @Narsil in #1593
• Adding some serialization testing around the wrapper. by @Narsil in #1594
• Add-legacy-tests by @ArthurZucker in #1597
• Adding a few tests for decoder deserialization. by @Narsil in #1598
• Better serialization error by @Narsil in #1595
• Add test normalizers by @ArthurZucker in #1600
• Improve decoder deserialization by @Narsil in #1599
• Using serde (serde_pyo3) to get str and repr easily. by @Narsil in #1588
• Merges cannot handle tokens containing spaces. by @Narsil in #909
• Fix doc about split by @ArthurZucker in #1591
• Support None to reset pre_tokenizers and normalizers, and index sequences by @ArthurZucker in #1590
• Fix strip python type by @ArthurZucker in #1602
• Tests + Deserialization improvement for normalizers. by @Narsil in #1604
• add deserialize for pre tokenizers by @ArthurZucker in #1603
• Perf improvement 16% by removing offsets. by @Narsil in #1587

New Contributors

• @nathaniel-daniel made their first contribution in #1532
• @nprisbrey made their first contribution in #1511
• @mcognetta made their first contribution in #1550
• @MikeIvanichev made their first contribution in #1571

Full Changelog: v0.19.1...v0.20.0rc1

v0.19.1

What's Changed

• add serialization for ignore_merges by @ArthurZucker in #1504

Full Changelog: v0.19.0...v0.19.1

v0.19.0

What's Changed

• chore: Remove CLI - this was originally intended for local development by @bryantbiggs in #1442
• [remove black] And use ruff by @ArthurZucker in #1436
• Bump ip from 2.0.0 to 2.0.1 in /bindings/node by @dependabot in #1456
• Added ability to inspect a 'Sequence' decoder and the AddedVocabulary. by @eaplatanios in #1443
• 🚨🚨 BREAKING CHANGE 🚨🚨: (add_prefix_space dropped everything is using prepend_scheme enum instead) Refactor metaspace by @ArthurZucker in #1476
• Add more support for tiktoken based tokenizers by @ArthurZucker in #1493
• PyO3 0.21. by @Narsil in #1494
• Remove 3.13 (potential undefined behavior.) by @Narsil in #1497
• Bumping all versions 3 times (ty transformers :) ) by @Narsil in #1498
• Fixing doc. by @Narsil in #1499

Full Changelog: v0.15.2...v0.19.0

v0.19.0rc0

Bumping 3 versions because of this: https://github.com/huggingface/transformers/blob/60dea593edd0b94ee15dc3917900b26e3acfbbee/setup.py#L177

What's Changed

• chore: Remove CLI - this was originally intended for local development by @bryantbiggs in #1442
• [remove black] And use ruff by @ArthurZucker in #1436
• Bump ip from 2.0.0 to 2.0.1 in /bindings/node by @dependabot in #1456
• Added ability to inspect a 'Sequence' decoder and the AddedVocabulary. by @eaplatanios in #1443
• 🚨🚨 BREAKING CHANGE 🚨🚨: (add_prefix_space dropped everything is using prepend_scheme enum instead) Refactor metaspace by @ArthurZucker in #1476
• Add more support for tiktoken based tokenizers by @ArthurZucker in #1493
• PyO3 0.21. by @Narsil in #1494
• Remove 3.13 (potential undefined behavior.) by @Narsil in #1497
• Bumping all versions 3 times (ty transformers :) ) by @Narsil in #1498

Full Changelog: v0.15.2...v0.19.0rc0

v0.15.2

What's Changed

Big shoutout to @rlrs for the fast replace normalizers PR. This boosts the performances of the tokenizers:

• chore: Update dependencies to latest supported versions by @bryantbiggs in #1441
• Convert word counts to u64 by @stephenroller in #1433
• Efficient Replace normalizer by @rlrs in #1413

New Contributors

• @bryantbiggs made their first contribution in #1441
• @stephenroller made their first contribution in #1433
• @rlrs made their first contribution in #1413

Full Changelog: v0.15.1...v0.15.2rc1

v0.15.1

What's Changed

• udpate to version = "0.15.1-dev0" by @ArthurZucker in #1390
• Derive Clone on Tokenizer, add Encoding.into_tokens() method by @epwalsh in #1381
• Stale bot. by @Narsil in #1404
• Fix doc links in readme by @Pierrci in #1367
• Faster HF dataset iteration in docs by @mariosasko in #1414
• Add quick doc to byte_level.rs by @steventrouble in #1420
• Fix make bench. by @Narsil in #1428
• Bump follow-redirects from 1.15.1 to 1.15.4 in /tokenizers/examples/unstable_wasm/www by @dependabot in #1430
• pyo3: update to 0.20 by @mikelui in #1386
• Encode special tokens by @ArthurZucker in #1437
• Update release for python3.12 windows by @ArthurZucker in #1438

New Contributors

• @steventrouble made their first contribution in #1420

Full Changelog: v0.15.0...v0.15.1

v0.15.1.rc0

What's Changed

• pyo3: update to 0.19 by @mikelui in #1322
• Add expect() for disabling truncation by @boyleconnor in #1316
• Re-using scritpts from safetensors. by @Narsil in #1328
• Reduce number of different revisions by 1 by @Narsil in #1329
• Python 38 arm by @Narsil in #1330
• Move to maturing mimicking move for safetensors. + Rewritten node bindings. by @Narsil in #1331
• Updating the docs with the new command. by @Narsil in #1333
• Update added tokens by @ArthurZucker in #1335
• update package version for dev by @ArthurZucker in #1339
• Added ability to inspect a 'Sequence' pre-tokenizer. by @eaplatanios in #1341
• Let's allow hf_hub < 1.0 by @ArthurZucker in #1344
• Fixing the progressbar. by @Narsil in #1353
• Preparing release. by @Narsil in #1355
• fix a clerical error in the comment by @tiandiweizun in #1356
• fix: remove useless token by @rtrompier in #1371
• Bump @babel/traverse from 7.22.11 to 7.23.2 in /bindings/node by @dependabot in #1370
• Allow hf_hub 0.18 by @mariosasko in #1383
• Allow huggingface_hub<1.0 by @Wauplin in #1385
• [pre_tokenizers] Fix sentencepiece based Metaspace by @ArthurZucker in #1357
• udpate to version = "0.15.1-dev0" by @ArthurZucker in #1390
• Derive Clone on Tokenizer, add Encoding.into_tokens() method by @epwalsh in #1381
• Stale bot. by @Narsil in #1404
• Fix doc links in readme by @Pierrci in #1367
• Faster HF dataset iteration in docs by @mariosasko in #1414
• Add quick doc to byte_level.rs by @steventrouble in #1420
• Fix make bench. by @Narsil in #1428
• Bump follow-redirects from 1.15.1 to 1.15.4 in /tokenizers/examples/unstable_wasm/www by @dependabot in #1430
• pyo3: update to 0.20 by @mikelui in #1386

New Contributors

• @mikelui made their first contribution in #1322
• @eaplatanios made their first contribution in #1341
• @tiandiweizun made their first contribution in #1356
• @rtrompier made their first contribution in #1371
• @mariosasko made their first contribution in #1383
• @Wauplin made their first contribution in #1385
• @steventrouble made their first contribution in #1420

Full Changelog: v0.13.4.rc2...v0.15.1.rc0

v0.15.0

What's Changed

• fix a clerical error in the comment by @tiandiweizun in #1356
• fix: remove useless token by @rtrompier in #1371
• Bump @babel/traverse from 7.22.11 to 7.23.2 in /bindings/node by @dependabot in #1370
• Allow hf_hub 0.18 by @mariosasko in #1383
• Allow huggingface_hub<1.0 by @Wauplin in #1385
• [pre_tokenizers] Fix sentencepiece based Metaspace by @ArthurZucker in #1357

New Contributors

• @tiandiweizun made their first contribution in #1356
• @rtrompier made their first contribution in #1371
• @mariosasko made their first contribution in #1383
• @Wauplin made their first contribution in #1385

Full Changelog: v0.14.1...v0.15.0

v0.14.1

What's Changed

• Fix conda release by @ArthurZucker in #1211
• Fix node release by @ArthurZucker in #1212
• Printing warning to stderr. by @Narsil in #1222
• Fixing padding_left sequence_ids. by @Narsil in #1233
• Use LTO for release and benchmark builds by @csko in #1157
• fix unigram.rs test_sample() by @chris-ha458 in #1244
• implement a simple max_sentencepiece_length into BPE by @chris-ha458 in #1228
• Makes decode and decode_batch work on borrowed content. by @mfuntowicz in #1251
• Update all GH Actions with dependency on actions/checkout by @mfuntowicz in #1256
• Parallelize unigram trainer by @mishig25 in #976
• Update unigram/trainer.rs by @chris-ha458 in #1257
• Fixing broken link. by @Narsil in #1268
• fix documentation regarding regex by @chris-ha458 in #1264
• Update Cargo.toml by @chris-ha458 in #1266
• Update README.md - Broken link by @sbhavani in #1272
• [doc build] Use secrets by @mishig25 in #1273
• Improve error for truncation with too high stride by @boyleconnor in #1275
• Add unigram bytefallback by @ArthurZucker in #1217
• revise type specification by @hiroshi-matsuda-rit in #1289
• Bump tough-cookie from 4.0.0 to 4.1.3 in /bindings/node by @dependabot in #1291
• Update path name: master -> main by @bact in #1292
• import Tuple from typing by @kellymarchisio in #1295
• Fixing clippy warnings on 1.71. by @Narsil in #1296
• Bump word-wrap from 1.2.3 to 1.2.4 in /bindings/node by @dependabot in #1299
• feat: Added CITATION.cff. by @SamuelLarkin in #1302
• Single warning for holes. by @Narsil in #1303
• Give error when initializing tokenizer with too high stride by @boyleconnor in #1306
• Handle when precompiled charsmap is empty by @kellymarchisio in #1308
• Derive clone for TrainerWrapper by @jonatanklosko in #1317
• CD backports by @chris-ha458 in #1318
• 0.13.4.rc1 by @Narsil in #1319
• Release all at once for simplicity. by @Narsil in #1320
• Fix stride condition. by @Narsil in #1321
• pyo3: update to 0.19 by @mikelui in #1322
• Add expect() for disabling truncation by @boyleconnor in #1316
• Re-using scritpts from safetensors. by @Narsil in #1328
• Reduce number of different revisions by 1 by @Narsil in #1329
• Python 38 arm by @Narsil in #1330
• Move to maturing mimicking move for safetensors. + Rewritten node bindings. by @Narsil in #1331
• Updating the docs with the new command. by @Narsil in #1333
• Update added tokens by @ArthurZucker in #1335
• update package version for dev by @ArthurZucker in #1339
• Added ability to inspect a 'Sequence' pre-tokenizer. by @eaplatanios in #1341
• Let's allow hf_hub < 1.0 by @ArthurZucker in #1344
• Fixing the progressbar. by @Narsil in #1353
• Preparing release. by @Narsil in #1355

New Contributors

• @csko made their first contribution in #1157
• @chris-ha458 made their first contribution in #1244
• @sbhavani made their first contribution in #1272
• @boyleconnor made their first contribution in #1275
• @hiroshi-matsuda-rit made their first contribution in #1289
• @bact made their first contribution in #1292
• @kellymarchisio made their first contribution in #1295
• @SamuelLarkin made their first contribution in #1302
• @jonatanklosko made their first contribution in #1317
• @mikelui made their first contribution in #1322
• @eaplatanios made their first contribution in #1341

Full Changelog: v0.13.3...v0.14.1

v0.14.1rc1

What's Changed

• pyo3: update to 0.19 by @mikelui in #1322
• Add expect() for disabling truncation by @boyleconnor in #1316
• Re-using scritpts from safetensors. by @Narsil in #1328
• Reduce number of different revisions by 1 by @Narsil in #1329
• Python 38 arm by @Narsil in #1330
• Move to maturing mimicking move for safetensors. + Rewritten node bindings. by @Narsil in #1331
• Updating the docs with the new command. by @Narsil in #1333
• Update added tokens by @ArthurZucker in #1335
• update package version for dev by @ArthurZucker in #1339
• Added ability to inspect a 'Sequence' pre-tokenizer. by @eaplatanios in #1341
• Let's allow hf_hub < 1.0 by @ArthurZucker in #1344
• Fixing the progressbar. by @Narsil in #1353

New Contributors

• @mikelui made their first contribution in #1322
• @eaplatanios made their first contribution in #1341

Full Changelog: v0.13.4.rc2...v0.14.1rc1