Spaces:
Runtime error
Runtime error
File size: 18,119 Bytes
a1d409e |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 |
# Copyright 2020 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 unittest
from transformers import MODEL_FOR_MASKED_LM_MAPPING, TF_MODEL_FOR_MASKED_LM_MAPPING, FillMaskPipeline, pipeline
from transformers.pipelines import PipelineException
from transformers.testing_utils import (
is_pipeline_test,
nested_simplify,
require_tf,
require_torch,
require_torch_gpu,
slow,
)
from .test_pipelines_common import ANY
@is_pipeline_test
class FillMaskPipelineTests(unittest.TestCase):
model_mapping = MODEL_FOR_MASKED_LM_MAPPING
tf_model_mapping = TF_MODEL_FOR_MASKED_LM_MAPPING
@require_tf
def test_small_model_tf(self):
unmasker = pipeline(task="fill-mask", model="sshleifer/tiny-distilroberta-base", top_k=2, framework="tf")
outputs = unmasker("My name is <mask>")
self.assertEqual(
nested_simplify(outputs, decimals=6),
[
{"sequence": "My name is grouped", "score": 2.1e-05, "token": 38015, "token_str": " grouped"},
{"sequence": "My name is accuser", "score": 2.1e-05, "token": 25506, "token_str": " accuser"},
],
)
outputs = unmasker("The largest city in France is <mask>")
self.assertEqual(
nested_simplify(outputs, decimals=6),
[
{
"sequence": "The largest city in France is grouped",
"score": 2.1e-05,
"token": 38015,
"token_str": " grouped",
},
{
"sequence": "The largest city in France is accuser",
"score": 2.1e-05,
"token": 25506,
"token_str": " accuser",
},
],
)
outputs = unmasker("My name is <mask>", targets=[" Patrick", " Clara", " Teven"], top_k=3)
self.assertEqual(
nested_simplify(outputs, decimals=6),
[
{"sequence": "My name is Clara", "score": 2e-05, "token": 13606, "token_str": " Clara"},
{"sequence": "My name is Patrick", "score": 2e-05, "token": 3499, "token_str": " Patrick"},
{"sequence": "My name is Te", "score": 1.9e-05, "token": 2941, "token_str": " Te"},
],
)
@require_torch
def test_small_model_pt(self):
unmasker = pipeline(task="fill-mask", model="sshleifer/tiny-distilroberta-base", top_k=2, framework="pt")
outputs = unmasker("My name is <mask>")
self.assertEqual(
nested_simplify(outputs, decimals=6),
[
{"sequence": "My name is Maul", "score": 2.2e-05, "token": 35676, "token_str": " Maul"},
{"sequence": "My name isELS", "score": 2.2e-05, "token": 16416, "token_str": "ELS"},
],
)
outputs = unmasker("The largest city in France is <mask>")
self.assertEqual(
nested_simplify(outputs, decimals=6),
[
{
"sequence": "The largest city in France is Maul",
"score": 2.2e-05,
"token": 35676,
"token_str": " Maul",
},
{"sequence": "The largest city in France isELS", "score": 2.2e-05, "token": 16416, "token_str": "ELS"},
],
)
outputs = unmasker("My name is <mask>", targets=[" Patrick", " Clara", " Teven"], top_k=3)
self.assertEqual(
nested_simplify(outputs, decimals=6),
[
{"sequence": "My name is Patrick", "score": 2.1e-05, "token": 3499, "token_str": " Patrick"},
{"sequence": "My name is Te", "score": 2e-05, "token": 2941, "token_str": " Te"},
{"sequence": "My name is Clara", "score": 2e-05, "token": 13606, "token_str": " Clara"},
],
)
outputs = unmasker("My name is <mask> <mask>", top_k=2)
self.assertEqual(
nested_simplify(outputs, decimals=6),
[
[
{
"score": 2.2e-05,
"token": 35676,
"token_str": " Maul",
"sequence": "<s>My name is Maul<mask></s>",
},
{"score": 2.2e-05, "token": 16416, "token_str": "ELS", "sequence": "<s>My name isELS<mask></s>"},
],
[
{
"score": 2.2e-05,
"token": 35676,
"token_str": " Maul",
"sequence": "<s>My name is<mask> Maul</s>",
},
{"score": 2.2e-05, "token": 16416, "token_str": "ELS", "sequence": "<s>My name is<mask>ELS</s>"},
],
],
)
@require_torch_gpu
def test_fp16_casting(self):
pipe = pipeline("fill-mask", model="hf-internal-testing/tiny-random-distilbert", device=0, framework="pt")
# convert model to fp16
pipe.model.half()
response = pipe("Paris is the [MASK] of France.")
# We actually don't care about the result, we just want to make sure
# it works, meaning the float16 tensor got casted back to float32
# for postprocessing.
self.assertIsInstance(response, list)
@slow
@require_torch
def test_large_model_pt(self):
unmasker = pipeline(task="fill-mask", model="distilroberta-base", top_k=2, framework="pt")
self.run_large_test(unmasker)
@slow
@require_tf
def test_large_model_tf(self):
unmasker = pipeline(task="fill-mask", model="distilroberta-base", top_k=2, framework="tf")
self.run_large_test(unmasker)
def run_large_test(self, unmasker):
outputs = unmasker("My name is <mask>")
self.assertEqual(
nested_simplify(outputs),
[
{"sequence": "My name is John", "score": 0.008, "token": 610, "token_str": " John"},
{"sequence": "My name is Chris", "score": 0.007, "token": 1573, "token_str": " Chris"},
],
)
outputs = unmasker("The largest city in France is <mask>")
self.assertEqual(
nested_simplify(outputs),
[
{
"sequence": "The largest city in France is Paris",
"score": 0.251,
"token": 2201,
"token_str": " Paris",
},
{
"sequence": "The largest city in France is Lyon",
"score": 0.214,
"token": 12790,
"token_str": " Lyon",
},
],
)
outputs = unmasker("My name is <mask>", targets=[" Patrick", " Clara", " Teven"], top_k=3)
self.assertEqual(
nested_simplify(outputs),
[
{"sequence": "My name is Patrick", "score": 0.005, "token": 3499, "token_str": " Patrick"},
{"sequence": "My name is Clara", "score": 0.000, "token": 13606, "token_str": " Clara"},
{"sequence": "My name is Te", "score": 0.000, "token": 2941, "token_str": " Te"},
],
)
@require_torch
def test_model_no_pad_pt(self):
unmasker = pipeline(task="fill-mask", model="sshleifer/tiny-distilroberta-base", framework="pt")
unmasker.tokenizer.pad_token_id = None
unmasker.tokenizer.pad_token = None
self.run_pipeline_test(unmasker, [])
@require_tf
def test_model_no_pad_tf(self):
unmasker = pipeline(task="fill-mask", model="sshleifer/tiny-distilroberta-base", framework="tf")
unmasker.tokenizer.pad_token_id = None
unmasker.tokenizer.pad_token = None
self.run_pipeline_test(unmasker, [])
def get_test_pipeline(self, model, tokenizer, processor):
if tokenizer is None or tokenizer.mask_token_id is None:
self.skipTest("The provided tokenizer has no mask token, (probably reformer or wav2vec2)")
fill_masker = FillMaskPipeline(model=model, tokenizer=tokenizer)
examples = [
f"This is another {tokenizer.mask_token} test",
]
return fill_masker, examples
def run_pipeline_test(self, fill_masker, examples):
tokenizer = fill_masker.tokenizer
model = fill_masker.model
outputs = fill_masker(
f"This is a {tokenizer.mask_token}",
)
self.assertEqual(
outputs,
[
{"sequence": ANY(str), "score": ANY(float), "token": ANY(int), "token_str": ANY(str)},
{"sequence": ANY(str), "score": ANY(float), "token": ANY(int), "token_str": ANY(str)},
{"sequence": ANY(str), "score": ANY(float), "token": ANY(int), "token_str": ANY(str)},
{"sequence": ANY(str), "score": ANY(float), "token": ANY(int), "token_str": ANY(str)},
{"sequence": ANY(str), "score": ANY(float), "token": ANY(int), "token_str": ANY(str)},
],
)
outputs = fill_masker([f"This is a {tokenizer.mask_token}"])
self.assertEqual(
outputs,
[
{"sequence": ANY(str), "score": ANY(float), "token": ANY(int), "token_str": ANY(str)},
{"sequence": ANY(str), "score": ANY(float), "token": ANY(int), "token_str": ANY(str)},
{"sequence": ANY(str), "score": ANY(float), "token": ANY(int), "token_str": ANY(str)},
{"sequence": ANY(str), "score": ANY(float), "token": ANY(int), "token_str": ANY(str)},
{"sequence": ANY(str), "score": ANY(float), "token": ANY(int), "token_str": ANY(str)},
],
)
outputs = fill_masker([f"This is a {tokenizer.mask_token}", f"Another {tokenizer.mask_token} great test."])
self.assertEqual(
outputs,
[
[
{"sequence": ANY(str), "score": ANY(float), "token": ANY(int), "token_str": ANY(str)},
{"sequence": ANY(str), "score": ANY(float), "token": ANY(int), "token_str": ANY(str)},
{"sequence": ANY(str), "score": ANY(float), "token": ANY(int), "token_str": ANY(str)},
{"sequence": ANY(str), "score": ANY(float), "token": ANY(int), "token_str": ANY(str)},
{"sequence": ANY(str), "score": ANY(float), "token": ANY(int), "token_str": ANY(str)},
],
[
{"sequence": ANY(str), "score": ANY(float), "token": ANY(int), "token_str": ANY(str)},
{"sequence": ANY(str), "score": ANY(float), "token": ANY(int), "token_str": ANY(str)},
{"sequence": ANY(str), "score": ANY(float), "token": ANY(int), "token_str": ANY(str)},
{"sequence": ANY(str), "score": ANY(float), "token": ANY(int), "token_str": ANY(str)},
{"sequence": ANY(str), "score": ANY(float), "token": ANY(int), "token_str": ANY(str)},
],
],
)
with self.assertRaises(ValueError):
fill_masker([None])
# No mask_token is not supported
with self.assertRaises(PipelineException):
fill_masker("This is")
self.run_test_top_k(model, tokenizer)
self.run_test_targets(model, tokenizer)
self.run_test_top_k_targets(model, tokenizer)
self.fill_mask_with_duplicate_targets_and_top_k(model, tokenizer)
self.fill_mask_with_multiple_masks(model, tokenizer)
def run_test_targets(self, model, tokenizer):
vocab = tokenizer.get_vocab()
targets = sorted(vocab.keys())[:2]
# Pipeline argument
fill_masker = FillMaskPipeline(model=model, tokenizer=tokenizer, targets=targets)
outputs = fill_masker(f"This is a {tokenizer.mask_token}")
self.assertEqual(
outputs,
[
{"sequence": ANY(str), "score": ANY(float), "token": ANY(int), "token_str": ANY(str)},
{"sequence": ANY(str), "score": ANY(float), "token": ANY(int), "token_str": ANY(str)},
],
)
target_ids = {vocab[el] for el in targets}
self.assertEqual({el["token"] for el in outputs}, target_ids)
self.assertEqual({el["token_str"] for el in outputs}, set(targets))
# Call argument
fill_masker = FillMaskPipeline(model=model, tokenizer=tokenizer)
outputs = fill_masker(f"This is a {tokenizer.mask_token}", targets=targets)
self.assertEqual(
outputs,
[
{"sequence": ANY(str), "score": ANY(float), "token": ANY(int), "token_str": ANY(str)},
{"sequence": ANY(str), "score": ANY(float), "token": ANY(int), "token_str": ANY(str)},
],
)
target_ids = {vocab[el] for el in targets}
self.assertEqual({el["token"] for el in outputs}, target_ids)
self.assertEqual({el["token_str"] for el in outputs}, set(targets))
# Score equivalence
outputs = fill_masker(f"This is a {tokenizer.mask_token}", targets=targets)
tokens = [top_mask["token_str"] for top_mask in outputs]
scores = [top_mask["score"] for top_mask in outputs]
unmasked_targets = fill_masker(f"This is a {tokenizer.mask_token}", targets=tokens)
target_scores = [top_mask["score"] for top_mask in unmasked_targets]
self.assertEqual(nested_simplify(scores), nested_simplify(target_scores))
# Raises with invalid
with self.assertRaises(ValueError):
outputs = fill_masker(f"This is a {tokenizer.mask_token}", targets=[""])
with self.assertRaises(ValueError):
outputs = fill_masker(f"This is a {tokenizer.mask_token}", targets=[])
with self.assertRaises(ValueError):
outputs = fill_masker(f"This is a {tokenizer.mask_token}", targets="")
def run_test_top_k(self, model, tokenizer):
fill_masker = FillMaskPipeline(model=model, tokenizer=tokenizer, top_k=2)
outputs = fill_masker(f"This is a {tokenizer.mask_token}")
self.assertEqual(
outputs,
[
{"sequence": ANY(str), "score": ANY(float), "token": ANY(int), "token_str": ANY(str)},
{"sequence": ANY(str), "score": ANY(float), "token": ANY(int), "token_str": ANY(str)},
],
)
fill_masker = FillMaskPipeline(model=model, tokenizer=tokenizer)
outputs2 = fill_masker(f"This is a {tokenizer.mask_token}", top_k=2)
self.assertEqual(
outputs2,
[
{"sequence": ANY(str), "score": ANY(float), "token": ANY(int), "token_str": ANY(str)},
{"sequence": ANY(str), "score": ANY(float), "token": ANY(int), "token_str": ANY(str)},
],
)
self.assertEqual(nested_simplify(outputs), nested_simplify(outputs2))
def run_test_top_k_targets(self, model, tokenizer):
vocab = tokenizer.get_vocab()
fill_masker = FillMaskPipeline(model=model, tokenizer=tokenizer)
# top_k=2, ntargets=3
targets = sorted(vocab.keys())[:3]
outputs = fill_masker(f"This is a {tokenizer.mask_token}", top_k=2, targets=targets)
# If we use the most probably targets, and filter differently, we should still
# have the same results
targets2 = [el["token_str"] for el in sorted(outputs, key=lambda x: x["score"], reverse=True)]
outputs2 = fill_masker(f"This is a {tokenizer.mask_token}", top_k=3, targets=targets2)
# They should yield exactly the same result
self.assertEqual(nested_simplify(outputs), nested_simplify(outputs2))
def fill_mask_with_duplicate_targets_and_top_k(self, model, tokenizer):
fill_masker = FillMaskPipeline(model=model, tokenizer=tokenizer)
vocab = tokenizer.get_vocab()
# String duplicates + id duplicates
targets = sorted(vocab.keys())[:3]
targets = [targets[0], targets[1], targets[0], targets[2], targets[1]]
outputs = fill_masker(f"My name is {tokenizer.mask_token}", targets=targets, top_k=10)
# The target list contains duplicates, so we can't output more
# than them
self.assertEqual(len(outputs), 3)
def fill_mask_with_multiple_masks(self, model, tokenizer):
fill_masker = FillMaskPipeline(model=model, tokenizer=tokenizer)
outputs = fill_masker(
f"This is a {tokenizer.mask_token} {tokenizer.mask_token} {tokenizer.mask_token}", top_k=2
)
self.assertEqual(
outputs,
[
[
{"sequence": ANY(str), "score": ANY(float), "token": ANY(int), "token_str": ANY(str)},
{"sequence": ANY(str), "score": ANY(float), "token": ANY(int), "token_str": ANY(str)},
],
[
{"sequence": ANY(str), "score": ANY(float), "token": ANY(int), "token_str": ANY(str)},
{"sequence": ANY(str), "score": ANY(float), "token": ANY(int), "token_str": ANY(str)},
],
[
{"sequence": ANY(str), "score": ANY(float), "token": ANY(int), "token_str": ANY(str)},
{"sequence": ANY(str), "score": ANY(float), "token": ANY(int), "token_str": ANY(str)},
],
],
)
|