File size: 16,533 Bytes
b381930 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 |
# Copyright 2022 The HuggingFace Team. All rights reserved.
#
# 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.
import json
import os
import re
import unittest
from functools import lru_cache
from transformers import CodeGenTokenizer, CodeGenTokenizerFast
from transformers.models.codegen.tokenization_codegen import VOCAB_FILES_NAMES
from transformers.testing_utils import require_tokenizers, slow
from ...test_tokenization_common import TokenizerTesterMixin, use_cache_if_possible
@require_tokenizers
class CodeGenTokenizationTest(TokenizerTesterMixin, unittest.TestCase):
from_pretrained_id = "Salesforce/codegen-350M-mono"
tokenizer_class = CodeGenTokenizer
rust_tokenizer_class = CodeGenTokenizerFast
test_rust_tokenizer = True
from_pretrained_kwargs = {"add_prefix_space": True}
test_seq2seq = False
@classmethod
def setUpClass(cls):
super().setUpClass()
# Adapted from Sennrich et al. 2015 and https://github.com/rsennrich/subword-nmt
vocab = [
"l",
"o",
"w",
"e",
"r",
"s",
"t",
"i",
"d",
"n",
"\u0120",
"\u0120l",
"\u0120n",
"\u0120lo",
"\u0120low",
"er",
"\u0120lowest",
"\u0120newer",
"\u0120wider",
"<unk>",
"<|endoftext|>",
]
vocab_tokens = dict(zip(vocab, range(len(vocab))))
merges = ["#version: 0.2", "\u0120 l", "\u0120l o", "\u0120lo w", "e r", ""]
cls.special_tokens_map = {"unk_token": "<unk>"}
cls.vocab_file = os.path.join(cls.tmpdirname, VOCAB_FILES_NAMES["vocab_file"])
cls.merges_file = os.path.join(cls.tmpdirname, VOCAB_FILES_NAMES["merges_file"])
with open(cls.vocab_file, "w", encoding="utf-8") as fp:
fp.write(json.dumps(vocab_tokens) + "\n")
with open(cls.merges_file, "w", encoding="utf-8") as fp:
fp.write("\n".join(merges))
@classmethod
@use_cache_if_possible
@lru_cache(maxsize=64)
def get_tokenizer(cls, pretrained_name=None, **kwargs):
kwargs.update(cls.special_tokens_map)
pretrained_name = pretrained_name or cls.tmpdirname
return CodeGenTokenizer.from_pretrained(pretrained_name, **kwargs)
@classmethod
@use_cache_if_possible
@lru_cache(maxsize=64)
def get_rust_tokenizer(cls, pretrained_name=None, **kwargs):
kwargs.update(cls.special_tokens_map)
pretrained_name = pretrained_name or cls.tmpdirname
return CodeGenTokenizerFast.from_pretrained(pretrained_name, **kwargs)
def get_input_output_texts(self, tokenizer):
input_text = "lower newer"
output_text = "lower newer"
return input_text, output_text
def test_full_tokenizer(self):
tokenizer = CodeGenTokenizer(self.vocab_file, self.merges_file, **self.special_tokens_map)
text = "lower newer"
bpe_tokens = ["\u0120low", "er", "\u0120", "n", "e", "w", "er"]
tokens = tokenizer.tokenize(text, add_prefix_space=True)
self.assertListEqual(tokens, bpe_tokens)
input_tokens = tokens + [tokenizer.unk_token]
input_bpe_tokens = [14, 15, 10, 9, 3, 2, 15, 19]
self.assertListEqual(tokenizer.convert_tokens_to_ids(input_tokens), input_bpe_tokens)
def test_rust_and_python_full_tokenizers(self):
if not self.test_rust_tokenizer:
self.skipTest(reason="test_rust_tokenizer is set to False")
tokenizer = self.get_tokenizer()
rust_tokenizer = self.get_rust_tokenizer(add_prefix_space=True)
sequence = "lower newer"
# Testing tokenization
tokens = tokenizer.tokenize(sequence, add_prefix_space=True)
rust_tokens = rust_tokenizer.tokenize(sequence)
self.assertListEqual(tokens, rust_tokens)
# Testing conversion to ids without special tokens
ids = tokenizer.encode(sequence, add_special_tokens=False, add_prefix_space=True)
rust_ids = rust_tokenizer.encode(sequence, add_special_tokens=False)
self.assertListEqual(ids, rust_ids)
# Testing conversion to ids with special tokens
rust_tokenizer = self.get_rust_tokenizer(add_prefix_space=True)
ids = tokenizer.encode(sequence, add_prefix_space=True)
rust_ids = rust_tokenizer.encode(sequence)
self.assertListEqual(ids, rust_ids)
# Testing the unknown token
input_tokens = tokens + [rust_tokenizer.unk_token]
input_bpe_tokens = [14, 15, 10, 9, 3, 2, 15, 19]
self.assertListEqual(rust_tokenizer.convert_tokens_to_ids(input_tokens), input_bpe_tokens)
@unittest.skip
def test_pretokenized_inputs(self, *args, **kwargs):
# It's very difficult to mix/test pretokenization with byte-level
# And get both CodeGen and Roberta to work at the same time (mostly an issue of adding a space before the string)
pass
def test_padding(self, max_length=15):
for tokenizer, pretrained_name, kwargs in self.tokenizers_list:
with self.subTest(f"{tokenizer.__class__.__name__} ({pretrained_name})"):
tokenizer_r = self.get_rust_tokenizer(pretrained_name, **kwargs)
# Simple input
s = "This is a simple input"
s2 = ["This is a simple input 1", "This is a simple input 2"]
p = ("This is a simple input", "This is a pair")
p2 = [
("This is a simple input 1", "This is a simple input 2"),
("This is a simple pair 1", "This is a simple pair 2"),
]
# Simple input tests
self.assertRaises(ValueError, tokenizer_r.encode, s, max_length=max_length, padding="max_length")
# Simple input
self.assertRaises(ValueError, tokenizer_r.encode_plus, s, max_length=max_length, padding="max_length")
# Simple input
self.assertRaises(
ValueError,
tokenizer_r.batch_encode_plus,
s2,
max_length=max_length,
padding="max_length",
)
# Pair input
self.assertRaises(ValueError, tokenizer_r.encode, p, max_length=max_length, padding="max_length")
# Pair input
self.assertRaises(ValueError, tokenizer_r.encode_plus, p, max_length=max_length, padding="max_length")
# Pair input
self.assertRaises(
ValueError,
tokenizer_r.batch_encode_plus,
p2,
max_length=max_length,
padding="max_length",
)
def test_padding_if_pad_token_set_slow(self):
tokenizer = CodeGenTokenizer.from_pretrained(self.tmpdirname, pad_token="<pad>")
# Simple input
s = "This is a simple input"
s2 = ["This is a simple input looooooooong", "This is a simple input"]
p = ("This is a simple input", "This is a pair")
p2 = [
("This is a simple input loooooong", "This is a simple input"),
("This is a simple pair loooooong", "This is a simple pair"),
]
pad_token_id = tokenizer.pad_token_id
out_s = tokenizer(s, padding="max_length", max_length=30, return_tensors="np")
out_s2 = tokenizer(s2, padding=True, truncate=True, return_tensors="np")
out_p = tokenizer(*p, padding="max_length", max_length=60, return_tensors="np")
out_p2 = tokenizer(p2, padding=True, truncate=True, return_tensors="np")
# s
# test single string max_length padding
self.assertEqual(out_s["input_ids"].shape[-1], 30)
self.assertTrue(pad_token_id in out_s["input_ids"])
self.assertTrue(0 in out_s["attention_mask"])
# s2
# test automatic padding
self.assertEqual(out_s2["input_ids"].shape[-1], 33)
# long slice doesn't have padding
self.assertFalse(pad_token_id in out_s2["input_ids"][0])
self.assertFalse(0 in out_s2["attention_mask"][0])
# short slice does have padding
self.assertTrue(pad_token_id in out_s2["input_ids"][1])
self.assertTrue(0 in out_s2["attention_mask"][1])
# p
# test single pair max_length padding
self.assertEqual(out_p["input_ids"].shape[-1], 60)
self.assertTrue(pad_token_id in out_p["input_ids"])
self.assertTrue(0 in out_p["attention_mask"])
# p2
# test automatic padding pair
self.assertEqual(out_p2["input_ids"].shape[-1], 52)
# long slice pair doesn't have padding
self.assertFalse(pad_token_id in out_p2["input_ids"][0])
self.assertFalse(0 in out_p2["attention_mask"][0])
# short slice pair does have padding
self.assertTrue(pad_token_id in out_p2["input_ids"][1])
self.assertTrue(0 in out_p2["attention_mask"][1])
def test_add_bos_token_slow(self):
bos_token = "$$$"
tokenizer = CodeGenTokenizer.from_pretrained(self.tmpdirname, bos_token=bos_token, add_bos_token=True)
s = "This is a simple input"
s2 = ["This is a simple input 1", "This is a simple input 2"]
bos_token_id = tokenizer.bos_token_id
out_s = tokenizer(s)
out_s2 = tokenizer(s2)
self.assertEqual(out_s.input_ids[0], bos_token_id)
self.assertTrue(all(o[0] == bos_token_id for o in out_s2.input_ids))
decode_s = tokenizer.decode(out_s.input_ids)
decode_s2 = tokenizer.batch_decode(out_s2.input_ids)
self.assertTrue(decode_s.startswith(bos_token))
self.assertTrue(all(d.startswith(bos_token) for d in decode_s2))
@slow
def test_truncation(self):
tokenizer = CodeGenTokenizer.from_pretrained("Salesforce/codegen-350M-mono")
text = "\nif len_a > len_b:\n result = a\nelse:\n result = b\n\n\n\n#"
expected_truncated_text = "\nif len_a > len_b:\n result = a\nelse:\n result = b"
input_ids = tokenizer.encode(text)
truncation_pattern = ["^#", re.escape("<|endoftext|>"), "^'''", '^"""', "\n\n\n"]
decoded_text = tokenizer.decode(input_ids, truncate_before_pattern=truncation_pattern)
self.assertEqual(decoded_text, expected_truncated_text)
# TODO @ArthurZ outputs of the fast tokenizer are different in this case, un-related to the PR
# tokenizer has no padding token
@unittest.skip(reason="tokenizer has no padding token")
def test_padding_different_model_input_name(self):
pass
@slow
def test_tokenizer_integration(self):
# Custom test since this tokenizer takes return_token_type_ids as an init argument for backward compatibility.
sequences = [
"Transformers (formerly known as pytorch-transformers and pytorch-pretrained-bert) provides "
"general-purpose architectures (BERT, GPT-2, RoBERTa, XLM, DistilBert, XLNet...) for Natural "
"Language Understanding (NLU) and Natural Language Generation (NLG) with over 32+ pretrained "
"models in 100+ languages and deep interoperability between Jax, PyTorch and TensorFlow.",
"BERT is designed to pre-train deep bidirectional representations from unlabeled text by jointly "
"conditioning on both left and right context in all layers.",
"The quick brown fox jumps over the lazy dog.",
]
tokenizer_classes = [self.tokenizer_class]
if self.test_rust_tokenizer:
tokenizer_classes.append(self.rust_tokenizer_class)
# Test default case. i.e. return_token_type_ids is False.
for tokenizer_class in tokenizer_classes:
tokenizer = tokenizer_class.from_pretrained("Salesforce/codegen-350M-mono")
encoding = tokenizer(sequences)
decoded_sequences = [tokenizer.decode(seq, skip_special_tokens=True) for seq in encoding["input_ids"]]
# fmt: off
expected_encoding = {'input_ids': [[41762, 364, 357, 36234, 1900, 355, 12972, 13165, 354, 12, 35636, 364, 290, 12972, 13165, 354, 12, 5310, 13363, 12, 4835, 8, 3769, 2276, 12, 29983, 45619, 357, 13246, 51, 11, 402, 11571, 12, 17, 11, 5564, 13246, 38586, 11, 16276, 44, 11, 4307, 346, 33, 861, 11, 16276, 7934, 23029, 329, 12068, 15417, 28491, 357, 32572, 52, 8, 290, 12068, 15417, 16588, 357, 32572, 38, 8, 351, 625, 3933, 10, 2181, 13363, 4981, 287, 1802, 10, 8950, 290, 2769, 48817, 1799, 1022, 449, 897, 11, 9485, 15884, 354, 290, 309, 22854, 37535, 13], [13246, 51, 318, 3562, 284, 662, 12, 27432, 2769, 8406, 4154, 282, 24612, 422, 9642, 9608, 276, 2420, 416, 26913, 21143, 319, 1111, 1364, 290, 826, 4732, 287, 477, 11685, 13], [464, 2068, 7586, 21831, 18045, 625, 262, 16931, 3290, 13]], 'attention_mask': [[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], [1, 1, 1, 1, 1, 1, 1, 1, 1, 1]]} # noqa: E501
# fmt: on
encoding_data = encoding.data
self.assertDictEqual(encoding_data, expected_encoding)
for expected, decoded in zip(sequences, decoded_sequences):
self.assertEqual(expected, decoded)
# Test return_token_type_ids is True case.
for tokenizer_class in tokenizer_classes:
tokenizer = tokenizer_class.from_pretrained("Salesforce/codegen-350M-mono", return_token_type_ids=True)
encoding = tokenizer(sequences)
decoded_sequences = [tokenizer.decode(seq, skip_special_tokens=True) for seq in encoding["input_ids"]]
# fmt: off
expected_encoding = {'input_ids': [[41762, 364, 357, 36234, 1900, 355, 12972, 13165, 354, 12, 35636, 364, 290, 12972, 13165, 354, 12, 5310, 13363, 12, 4835, 8, 3769, 2276, 12, 29983, 45619, 357, 13246, 51, 11, 402, 11571, 12, 17, 11, 5564, 13246, 38586, 11, 16276, 44, 11, 4307, 346, 33, 861, 11, 16276, 7934, 23029, 329, 12068, 15417, 28491, 357, 32572, 52, 8, 290, 12068, 15417, 16588, 357, 32572, 38, 8, 351, 625, 3933, 10, 2181, 13363, 4981, 287, 1802, 10, 8950, 290, 2769, 48817, 1799, 1022, 449, 897, 11, 9485, 15884, 354, 290, 309, 22854, 37535, 13], [13246, 51, 318, 3562, 284, 662, 12, 27432, 2769, 8406, 4154, 282, 24612, 422, 9642, 9608, 276, 2420, 416, 26913, 21143, 319, 1111, 1364, 290, 826, 4732, 287, 477, 11685, 13], [464, 2068, 7586, 21831, 18045, 625, 262, 16931, 3290, 13]], 'token_type_ids': [[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], [0, 0, 0, 0, 0, 0, 0, 0, 0, 0]], 'attention_mask': [[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], [1, 1, 1, 1, 1, 1, 1, 1, 1, 1]]} # noqa: E501
# fmt: on
encoding_data = encoding.data
self.assertDictEqual(encoding_data, expected_encoding)
for expected, decoded in zip(sequences, decoded_sequences):
self.assertEqual(expected, decoded)
|