Adding Llama FastTokenizer support. (#22264)

* Adding Llama FastTokenizer support.

- Requires https://github.com/huggingface/tokenizers/pull/1183 version
- Only support byte_fallback for llama, raise otherwise (safety net).
- Lots of questions are special tokens

How to test:

```python

from transformers.convert_slow_tokenizer import convert_slow_tokenizer
from transformers import AutoTokenizer
from tokenizers import Tokenizer

tokenizer = AutoTokenizer.from_pretrained("huggingface/llama-7b")

if False:
    new_tokenizer = Tokenizer.from_file("tok.json")
else:
    new_tokenizer = convert_slow_tokenizer(tokenizer)
    new_tokenizer.save("tok.json")

strings = [
    "This is a test",
    "生活的真谛是",
    "生活的真谛是[MASK]。",
    # XXX: This one is problematic because of special tokens
    # "<s> Something something",
]

for string in strings:
    encoded = tokenizer(string)["input_ids"]
    encoded2 = new_tokenizer.encode(string).ids

    assert encoded == encoded2, f"{encoded} != {encoded2}"

    decoded = tokenizer.decode(encoded)
    decoded2 = new_tokenizer.decode(encoded2)

    assert decoded.strip() == decoded2, f"{repr(decoded)} != {repr(decoded2)}"
```

The converter + some test script.

The test script.

Tmp save.

Adding Fast tokenizer + tests.

Adding the tokenization tests.

Correct combination.

Small fix.

Fixing tests.

Fixing with latest update.

Rebased.

fix copies + normalized added tokens  + copies.

Adding doc.

TMP.

Doc + split files.

Doc.

Versions + try import.

Fix Camembert + warnings -> Error.

Fix by ArthurZucker.

Not a decorator.

* Fixing comments.

* Adding more to docstring.

* Doc rewriting.

docs/source/en/index.mdx

@@ -336,7 +336,7 @@ Flax), PyTorch, and/or TensorFlow.
 |              LED              |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
 |             LeViT             |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
 |             LiLT              |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-|             LLaMA             |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
+|             LLaMA             |       ✅       |       ✅       |       ✅        |         ❌         |      ❌      |
 |          Longformer           |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
 |            LongT5             |       ❌       |       ❌       |       ✅        |         ❌         |      ✅      |
 |             LUKE              |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |

docs/source/en/model_doc/llama.mdx

@@ -59,6 +59,14 @@ This model was contributed by [zphang](https://huggingface.co/zphang) with contr
     - create_token_type_ids_from_sequences
     - save_vocabulary
 
+## LlamaTokenizerFast
+
+[[autodoc]] LlamaTokenizerFast
+    - build_inputs_with_special_tokens
+    - get_special_tokens_mask
+    - create_token_type_ids_from_sequences
+    - save_vocabulary
+
 ## LlamaModel
 
 [[autodoc]] LlamaModel

setup.py

@@ -78,7 +78,7 @@ import re
 import shutil
 from pathlib import Path
 
-from setuptools import setup, Command
+from setuptools import Command, setup
 
 
 # Remove stale transformers.egg-info directory to avoid https://github.com/pypa/pip/issues/5466
@@ -251,6 +251,7 @@ class DepsTableUpdateCommand(Command):
         with open(target, "w", encoding="utf-8", newline="\n") as f:
             f.write("\n".join(content))
 
+
 extras = {}
 
 extras["ja"] = deps_list("fugashi", "ipadic", "unidic_lite", "unidic", "sudachipy", "sudachidict_core", "rhoknp")

src/transformers/__init__.py

@@ -740,6 +740,7 @@ else:
     _import_structure["models.layoutlmv3"].append("LayoutLMv3TokenizerFast")
     _import_structure["models.layoutxlm"].append("LayoutXLMTokenizerFast")
     _import_structure["models.led"].append("LEDTokenizerFast")
+    _import_structure["models.llama"].append("LlamaTokenizerFast")
     _import_structure["models.longformer"].append("LongformerTokenizerFast")
     _import_structure["models.lxmert"].append("LxmertTokenizerFast")
     _import_structure["models.markuplm"].append("MarkupLMTokenizerFast")
@@ -4388,6 +4389,7 @@ if TYPE_CHECKING:
         from .models.layoutlmv3 import LayoutLMv3TokenizerFast
         from .models.layoutxlm import LayoutXLMTokenizerFast
         from .models.led import LEDTokenizerFast
+        from .models.llama import LlamaTokenizerFast
         from .models.longformer import LongformerTokenizerFast
         from .models.lxmert import LxmertTokenizerFast
         from .models.markuplm import MarkupLMTokenizerFast

src/transformers/convert_slow_tokenizer.py

@@ -19,10 +19,9 @@ All the conversions are grouped here to gather SentencePiece dependencies outsid
 allow to make our dependency on SentencePiece optional.
 """
 
-import warnings
 from typing import Dict, List, Tuple
 
-from tokenizers import Regex, Tokenizer, decoders, normalizers, pre_tokenizers, processors
+from tokenizers import AddedToken, Regex, Tokenizer, decoders, normalizers, pre_tokenizers, processors
 from tokenizers.models import BPE, Unigram, WordPiece
 
 from .utils import requires_backends
@@ -450,12 +449,13 @@ class SpmConverter(Converter):
         self.proto = m
 
         if self.proto.trainer_spec.byte_fallback:
-            warnings.warn(
-                "The sentencepiece tokenizer that you are converting to a fast tokenizer uses the byte fallback option"
-                " which is not implemented in the fast tokenizers. In practice this means that the fast version of the"
-                " tokenizer can produce unknown tokens whereas the sentencepiece version would have converted these "
-                "unknown tokens into a sequence of byte tokens matching the original piece of text."
-            )
+            if not getattr(self, "handle_byte_fallback", None):
+                raise RuntimeError(
+                    "The sentencepiece tokenizer that you are converting to a fast tokenizer uses the byte fallback option"
+                    " which is not implemented in the fast tokenizers. In practice this means that the fast version of the"
+                    " tokenizer can produce unknown tokens whereas the sentencepiece version would have converted these "
+                    "unknown tokens into a sequence of byte tokens matching the original piece of text."
+                )
 
     def vocab(self, proto):
         return [(piece.piece, piece.score) for piece in proto.pieces]
@@ -1094,6 +1094,78 @@ class XGLMConverter(SpmConverter):
         )
 
 
+class LlamaConverter(SpmConverter):
+    handle_byte_fallback = True
+
+    def vocab(self, proto):
+        vocab = [
+            ("<unk>", 0.0),
+            ("<s>", 0.0),
+            ("</s>", 0.0),
+        ]
+        vocab += [(piece.piece, piece.score) for piece in proto.pieces[3:]]
+        return vocab
+
+    def unk_id(self, proto):
+        unk_id = 0
+        return unk_id
+
+    def decoder(self, replacement, add_prefix_space):
+        return decoders.Sequence(
+            [
+                decoders.Replace("▁", " "),
+                decoders.ByteFallback(),
+                decoders.Fuse(),
+                decoders.Strip(content=" ", left=1),
+            ]
+        )
+
+    def tokenizer(self, proto):
+        model_type = proto.trainer_spec.model_type
+        vocab_scores = self.vocab(proto)
+        if model_type == 1:
+            raise RuntimeError("Llama is supposed to be a BPE model!")
+        elif model_type == 2:
+            _, merges = SentencePieceExtractor(self.original_tokenizer.vocab_file).extract(vocab_scores)
+            bpe_vocab = {word: i for i, (word, _score) in enumerate(vocab_scores)}
+            tokenizer = Tokenizer(
+                BPE(bpe_vocab, merges, unk_token=proto.trainer_spec.unk_piece, fuse_unk=True, byte_fallback=True)
+            )
+            tokenizer.add_special_tokens(
+                [
+                    AddedToken("<unk>", normalized=True),
+                    AddedToken("<s>", normalized=True),
+                    AddedToken("</s>", normalized=True),
+                ]
+            )
+        else:
+            raise Exception(
+                "You're trying to run a `Unigram` model but you're file was trained with a different algorithm"
+            )
+
+        return tokenizer
+
+    def normalizer(self, proto):
+        return normalizers.Sequence(
+            [
+                normalizers.Prepend(prepend="▁"),
+                normalizers.Replace(pattern=" ", content="▁"),
+            ]
+        )
+
+    def pre_tokenizer(self, replacement, add_prefix_space):
+        return None
+
+    def post_processor(self):
+        return processors.TemplateProcessing(
+            single="<s> $A",
+            pair="<s> $A $B",
+            special_tokens=[
+                ("<s>", self.original_tokenizer.convert_tokens_to_ids("<s>")),
+            ],
+        )
+
+
 class MarkupLMConverter(Converter):
     def converted(self) -> Tokenizer:
         ot = self.original_tokenizer
@@ -1183,6 +1255,7 @@ SLOW_TO_FAST_CONVERTERS = {
     "XLNetTokenizer": XLNetConverter,
     "SplinterTokenizer": SplinterConverter,
     "XGLMTokenizer": XGLMConverter,
+    "LlamaTokenizer": LlamaConverter,
 }
 
 

src/transformers/models/auto/tokenization_auto.py

@@ -172,7 +172,13 @@ else:
             ("layoutxlm", ("LayoutXLMTokenizer", "LayoutXLMTokenizerFast" if is_tokenizers_available() else None)),
             ("led", ("LEDTokenizer", "LEDTokenizerFast" if is_tokenizers_available() else None)),
             ("lilt", ("LayoutLMv3Tokenizer", "LayoutLMv3TokenizerFast" if is_tokenizers_available() else None)),
-            ("llama", ("LlamaTokenizer" if is_sentencepiece_available() else None, None)),
+            (
+                "llama",
+                (
+                    "LlamaTokenizer" if is_sentencepiece_available() else None,
+                    "LlamaTokenizerFast" if is_tokenizers_available() else None,
+                ),
+            ),
             ("longformer", ("LongformerTokenizer", "LongformerTokenizerFast" if is_tokenizers_available() else None)),
             (
                 "longt5",

src/transformers/models/llama/__init__.py

@@ -17,6 +17,7 @@ from ...utils import (
     OptionalDependencyNotAvailable,
     _LazyModule,
     is_sentencepiece_available,
+    is_tokenizers_available,
     is_torch_available,
 )
 
@@ -33,6 +34,14 @@ except OptionalDependencyNotAvailable:
 else:
     _import_structure["tokenization_llama"] = ["LlamaTokenizer"]
 
+try:
+    if not is_tokenizers_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
+    _import_structure["tokenization_llama_fast"] = ["LlamaTokenizerFast"]
+
 try:
     if not is_torch_available():
         raise OptionalDependencyNotAvailable()
@@ -58,6 +67,14 @@ if TYPE_CHECKING:
     else:
         from .tokenization_llama import LlamaTokenizer
 
+    try:
+        if not is_tokenizers_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
+        from .tokenization_llama_fast import LlamaTokenizerFast
+
     try:
         if not is_torch_available():
             raise OptionalDependencyNotAvailable()

src/transformers/models/llama/tokenization_llama_fast.py

@@ -0,0 +1,82 @@
+# coding=utf-8
+# Copyright 2020 The HuggingFace Inc. team.
+#
+# 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.
+from ...tokenization_utils_fast import PreTrainedTokenizerFast
+from ...utils.versions import require_version
+
+
+require_version("tokenizers>=0.13.3")
+
+
+class LlamaTokenizerFast(PreTrainedTokenizerFast):
+    """
+    Construct a Llama tokenizer. Based on byte-level Byte-Pair-Encoding.
+
+    This uses notably ByteFallback and no normalization.
+
+    ```
+    from transformers import LlamaTokenizerFast
+
+    tokenizer = LlaTokenizerFast.from_pretrained("hf-internal-testing/llama-tokenizer")
+    tokenizer.encode("Hello this is a test")
+    >>> [1, 15043, 445, 338, 263, 1243]
+    ```
+
+    This tokenizer inherits from [`PreTrainedTokenizerFast`] which contains most of the main methods. Users should
+    refer to this superclass for more information regarding those methods.
+
+    Args:
+        vocab_file (`str`):
+            [SentencePiece](https://github.com/google/sentencepiece) file (generally has a .model extension) that
+            contains the vocabulary necessary to instantiate a tokenizer.
+        tokenizer_file (`str`):
+            [tokenizers](https://github.com/huggingface/tokenizers) file (generally has a .json extension) that
+            contains everything needed to load the tokenizer.
+
+        clean_up_tokenization_spaces (`str`, *optional*, defaults to `False`):
+            Wether to cleanup spaces after decoding, cleanup consists in removing potential artifacts like extra
+            spaces.
+
+        bos_token (`str`, *optional*, defaults to `"<s>"`):
+            The beginning of sequence token that was used during pretraining. Can be used a sequence classifier token.
+
+        eos_token (`str`, *optional*, defaults to `"</s>"`):
+            The end of sequence token.
+
+        unk_token (`str`, *optional*, defaults to `"<unk>"`):
+            The unknown token. A token that is not in the vocabulary cannot be converted to an ID and is set to be this
+            token instead.
+    """
+
+    padding_side = "left"
+
+    def __init__(
+        self,
+        vocab_file=None,
+        tokenizer_file=None,
+        clean_up_tokenization_spaces=False,
+        unk_token="<unk>",
+        bos_token="<s>",
+        eos_token="</s>",
+        **kwargs,
+    ):
+        super().__init__(
+            vocab_file=vocab_file,
+            tokenizer_file=tokenizer_file,
+            clean_up_tokenization_spaces=clean_up_tokenization_spaces,
+            unk_token=unk_token,
+            bos_token=bos_token,
+            eos_token=eos_token,
+            **kwargs,
+        )

src/transformers/utils/dummy_tokenizers_objects.py

@@ -219,6 +219,13 @@ class LEDTokenizerFast(metaclass=DummyObject):
         requires_backends(self, ["tokenizers"])
 
 
+class LlamaTokenizerFast(metaclass=DummyObject):
+    _backends = ["tokenizers"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["tokenizers"])
+
+
 class LongformerTokenizerFast(metaclass=DummyObject):
     _backends = ["tokenizers"]
 

tests/models/llama/test_tokenization_llama.py

@@ -1,3 +1,4 @@
+# coding=utf-8
 # Copyright 2023 The HuggingFace Team. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -23,8 +24,10 @@ from transformers import (
     SPIECE_UNDERLINE,
     AddedToken,
     LlamaTokenizer,
+    LlamaTokenizerFast,
     is_torch_available,
 )
+from transformers.convert_slow_tokenizer import convert_slow_tokenizer
 from transformers.testing_utils import (
     get_tests_dir,
     nested_simplify,
@@ -287,13 +290,11 @@ class LlamaTokenizationTest(TokenizerTesterMixin, unittest.TestCase):
 @require_sentencepiece
 @require_tokenizers
 class LlamaIntegrationTest(unittest.TestCase):
-    checkpoint_name = "hf-internal-testing/llama-tokenizer"
-
     @classmethod
     def setUpClass(cls):
-        cls.tokenizer: LlamaTokenizer = LlamaTokenizer.from_pretrained(cls.checkpoint_name)
-        cls.rust_tokenizer = cls.tokenizer  # TODO @narsil replace with the rust one
-        cls.pad_token_id = 1
+        checkpoint_name = "hf-internal-testing/llama-tokenizer"
+        cls.tokenizer: LlamaTokenizer = LlamaTokenizer.from_pretrained(checkpoint_name)
+        cls.rust_tokenizer = LlamaTokenizerFast.from_pretrained(checkpoint_name)
         return cls
 
     @require_torch
@@ -314,6 +315,27 @@ class LlamaIntegrationTest(unittest.TestCase):
             },
         )
 
+    @slow
+    def test_conversion(self):
+        # This is excruciatingly slow since it has to recreate the entire merge
+        # list from the original vocabulary in spm
+        self.rust_tokenizer.save_pretrained("./out")
+        with tempfile.TemporaryDirectory() as dirname:
+            self.rust_tokenizer.save_pretrained(dirname)
+
+            with open(os.path.join(dirname, "tokenizer.json"), "r") as f:
+                old_serialized = f.read()
+
+        new_tokenizer = convert_slow_tokenizer(self.tokenizer)
+        with tempfile.NamedTemporaryFile() as f:
+            new_tokenizer.save(f.name)
+            # Re-opening since `f` is in bytes.
+            new_serialized = open(f.name, "r").read()
+            with open("out_tokenizer.json", "w") as g:
+                g.write(new_serialized)
+
+            self.assertEqual(old_serialized, new_serialized)
+
     def test_simple_encode_decode(self):
         pyth_tokenizer = self.tokenizer
         rust_tokenizer = self.rust_tokenizer
@@ -362,11 +384,27 @@ class LlamaIntegrationTest(unittest.TestCase):
         self.assertEqual(pyth_tokenizer.encode(" Hello"), [1, 29871, 15043])
         self.assertEqual(rust_tokenizer.encode(" Hello"), [1, 29871, 15043])
 
+    def test_no_differences_showcase(self):
+        pyth_tokenizer = self.tokenizer
+        rust_tokenizer = self.rust_tokenizer
+        self.assertEqual(pyth_tokenizer.encode(""), [1])
+        self.assertEqual(rust_tokenizer.encode(""), [1])
+
+        self.assertEqual(pyth_tokenizer.encode(" "), [1, 259])
+        self.assertEqual(rust_tokenizer.encode(" "), [1, 259])
+
+        self.assertEqual(pyth_tokenizer.encode("  "), [1, 1678])
+        self.assertEqual(rust_tokenizer.encode("  "), [1, 1678])
+
+        self.assertEqual(pyth_tokenizer.encode(" Hello"), [1, 29871, 15043])
+        self.assertEqual(rust_tokenizer.encode(" Hello"), [1, 29871, 15043])
+
         self.assertEqual(pyth_tokenizer.encode("<s>"), [1, 1])
         self.assertEqual(rust_tokenizer.encode("<s>"), [1, 1])
 
-        self.assertEqual(pyth_tokenizer.encode(""), [1])
-        self.assertEqual(rust_tokenizer.encode(""), [1])
+    def test_no_differences_decode(self):
+        pyth_tokenizer = self.tokenizer
+        rust_tokenizer = self.rust_tokenizer
 
         self.assertEqual(pyth_tokenizer.decode([869]), ".")
         self.assertEqual(rust_tokenizer.decode([869]), ".")
@@ -374,6 +412,15 @@ class LlamaIntegrationTest(unittest.TestCase):
         self.assertEqual(pyth_tokenizer.decode([30112, 869]), "ا .")
         self.assertEqual(rust_tokenizer.decode([30112, 869]), "ا .")
 
+    def test_no_differences_special_tokens(self):
+        pyth_tokenizer = self.tokenizer
+        rust_tokenizer = self.rust_tokenizer
+        self.assertEqual(pyth_tokenizer.encode(""), [1])
+        self.assertEqual(rust_tokenizer.encode(""), [1])
+
+        self.assertEqual(pyth_tokenizer.encode("<s>"), [1, 1])
+        self.assertEqual(rust_tokenizer.encode("<s>"), [1, 1])
+
     @unittest.skipIf(
         os.getenv("RUN_TOKENIZER_INTEGRATION", "0") == "0",
         "RUN_TOKENIZER_INTEGRATION=1 to run tokenizer integration tests",
@@ -392,8 +439,8 @@ class LlamaIntegrationTest(unittest.TestCase):
 
             self.assertEqual(encoded1, encoded2)
 
-            decoded1 = pyth_tokenizer.decode(encoded1)
-            decoded2 = rust_tokenizer.decode(encoded2)
+            decoded1 = pyth_tokenizer.decode(encoded1, skip_special_tokens=True)
+            decoded2 = rust_tokenizer.decode(encoded2, skip_special_tokens=True)
 
             self.assertEqual(decoded1, decoded2)
 
@@ -406,7 +453,7 @@ class LlamaIntegrationTest(unittest.TestCase):
 
                 self.assertEqual(encoded1, encoded2)
 
-                decoded1 = pyth_tokenizer.decode(encoded1)
-                decoded2 = rust_tokenizer.decode(encoded2)
+                decoded1 = pyth_tokenizer.decode(encoded1, skip_special_tokens=True)
+                decoded2 = rust_tokenizer.decode(encoded2, skip_special_tokens=True)
 
                 self.assertEqual(decoded1, decoded2)

tests/utils/test_convert_slow_tokenizer.py

@@ -24,11 +24,10 @@ class ConvertSlowTokenizerTest(unittest.TestCase):
 
         original_tokenizer_with_bytefallback = FakeOriginalTokenizer(vocab_file=spm_model_file_with_bytefallback)
 
-        with warnings.catch_warnings(record=True) as w:
+        with self.assertRaises(RuntimeError) as cm:
             _ = SpmConverter(original_tokenizer_with_bytefallback)
-        self.assertEqual(len(w), 1)
         self.assertIn(
             "The sentencepiece tokenizer that you are converting to a fast tokenizer uses the byte fallback option"
             " which is not implemented in the fast tokenizers.",
-            str(w[0].message),
+            str(cm.exception),
         )