File size: 20,168 Bytes
455a40f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
# coding=utf-8
# Copyright 2018 HuggingFace Inc..
#
# 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 argparse
import json
import logging
import os
import sys
from unittest.mock import patch

import torch

from transformers import ViTMAEForPreTraining, Wav2Vec2ForPreTraining
from transformers.testing_utils import CaptureLogger, TestCasePlus, get_gpu_count, slow, torch_device
from transformers.utils import is_apex_available


SRC_DIRS = [
    os.path.join(os.path.dirname(__file__), dirname)
    for dirname in [
        "text-generation",
        "text-classification",
        "token-classification",
        "language-modeling",
        "multiple-choice",
        "question-answering",
        "summarization",
        "translation",
        "image-classification",
        "speech-recognition",
        "audio-classification",
        "speech-pretraining",
        "image-pretraining",
        "semantic-segmentation",
    ]
]
sys.path.extend(SRC_DIRS)


if SRC_DIRS is not None:
    import run_audio_classification
    import run_clm
    import run_generation
    import run_glue
    import run_image_classification
    import run_mae
    import run_mlm
    import run_ner
    import run_qa as run_squad
    import run_semantic_segmentation
    import run_seq2seq_qa as run_squad_seq2seq
    import run_speech_recognition_ctc
    import run_speech_recognition_seq2seq
    import run_summarization
    import run_swag
    import run_translation
    import run_wav2vec2_pretraining_no_trainer


logging.basicConfig(level=logging.DEBUG)

logger = logging.getLogger()


def get_setup_file():
    parser = argparse.ArgumentParser()
    parser.add_argument("-f")
    args = parser.parse_args()
    return args.f


def get_results(output_dir):
    results = {}
    path = os.path.join(output_dir, "all_results.json")
    if os.path.exists(path):
        with open(path, "r") as f:
            results = json.load(f)
    else:
        raise ValueError(f"can't find {path}")
    return results


def is_cuda_and_apex_available():
    is_using_cuda = torch.cuda.is_available() and torch_device == "cuda"
    return is_using_cuda and is_apex_available()


stream_handler = logging.StreamHandler(sys.stdout)
logger.addHandler(stream_handler)


class ExamplesTests(TestCasePlus):
    def test_run_glue(self):
        tmp_dir = self.get_auto_remove_tmp_dir()
        testargs = f"""
            run_glue.py
            --model_name_or_path distilbert-base-uncased
            --output_dir {tmp_dir}
            --overwrite_output_dir
            --train_file ./tests/fixtures/tests_samples/MRPC/train.csv
            --validation_file ./tests/fixtures/tests_samples/MRPC/dev.csv
            --do_train
            --do_eval
            --per_device_train_batch_size=2
            --per_device_eval_batch_size=1
            --learning_rate=1e-4
            --max_steps=10
            --warmup_steps=2
            --seed=42
            --max_seq_length=128
            """.split()

        if is_cuda_and_apex_available():
            testargs.append("--fp16")

        with patch.object(sys, "argv", testargs):
            run_glue.main()
            result = get_results(tmp_dir)
            self.assertGreaterEqual(result["eval_accuracy"], 0.75)

    def test_run_clm(self):
        tmp_dir = self.get_auto_remove_tmp_dir()
        testargs = f"""
            run_clm.py
            --model_name_or_path distilgpt2
            --train_file ./tests/fixtures/sample_text.txt
            --validation_file ./tests/fixtures/sample_text.txt
            --do_train
            --do_eval
            --block_size 128
            --per_device_train_batch_size 5
            --per_device_eval_batch_size 5
            --num_train_epochs 2
            --output_dir {tmp_dir}
            --overwrite_output_dir
            """.split()

        if torch.cuda.device_count() > 1:
            # Skipping because there are not enough batches to train the model + would need a drop_last to work.
            return

        if torch_device != "cuda":
            testargs.append("--no_cuda")

        with patch.object(sys, "argv", testargs):
            run_clm.main()
            result = get_results(tmp_dir)
            self.assertLess(result["perplexity"], 100)

    def test_run_clm_config_overrides(self):
        # test that config_overrides works, despite the misleading dumps of default un-updated
        # config via tokenizer

        tmp_dir = self.get_auto_remove_tmp_dir()
        testargs = f"""
            run_clm.py
            --model_type gpt2
            --tokenizer_name gpt2
            --train_file ./tests/fixtures/sample_text.txt
            --output_dir {tmp_dir}
            --config_overrides n_embd=10,n_head=2
            """.split()

        if torch_device != "cuda":
            testargs.append("--no_cuda")

        logger = run_clm.logger
        with patch.object(sys, "argv", testargs):
            with CaptureLogger(logger) as cl:
                run_clm.main()

        self.assertIn('"n_embd": 10', cl.out)
        self.assertIn('"n_head": 2', cl.out)

    def test_run_mlm(self):
        tmp_dir = self.get_auto_remove_tmp_dir()
        testargs = f"""
            run_mlm.py
            --model_name_or_path distilroberta-base
            --train_file ./tests/fixtures/sample_text.txt
            --validation_file ./tests/fixtures/sample_text.txt
            --output_dir {tmp_dir}
            --overwrite_output_dir
            --do_train
            --do_eval
            --prediction_loss_only
            --num_train_epochs=1
        """.split()

        if torch_device != "cuda":
            testargs.append("--no_cuda")

        with patch.object(sys, "argv", testargs):
            run_mlm.main()
            result = get_results(tmp_dir)
            self.assertLess(result["perplexity"], 42)

    def test_run_ner(self):
        # with so little data distributed training needs more epochs to get the score on par with 0/1 gpu
        epochs = 7 if get_gpu_count() > 1 else 2

        tmp_dir = self.get_auto_remove_tmp_dir()
        testargs = f"""
            run_ner.py
            --model_name_or_path bert-base-uncased
            --train_file tests/fixtures/tests_samples/conll/sample.json
            --validation_file tests/fixtures/tests_samples/conll/sample.json
            --output_dir {tmp_dir}
            --overwrite_output_dir
            --do_train
            --do_eval
            --warmup_steps=2
            --learning_rate=2e-4
            --per_device_train_batch_size=2
            --per_device_eval_batch_size=2
            --num_train_epochs={epochs}
            --seed 7
        """.split()

        if torch_device != "cuda":
            testargs.append("--no_cuda")

        with patch.object(sys, "argv", testargs):
            run_ner.main()
            result = get_results(tmp_dir)
            self.assertGreaterEqual(result["eval_accuracy"], 0.75)
            self.assertLess(result["eval_loss"], 0.5)

    def test_run_squad(self):
        tmp_dir = self.get_auto_remove_tmp_dir()
        testargs = f"""
            run_qa.py
            --model_name_or_path bert-base-uncased
            --version_2_with_negative
            --train_file tests/fixtures/tests_samples/SQUAD/sample.json
            --validation_file tests/fixtures/tests_samples/SQUAD/sample.json
            --output_dir {tmp_dir}
            --overwrite_output_dir
            --max_steps=10
            --warmup_steps=2
            --do_train
            --do_eval
            --learning_rate=2e-4
            --per_device_train_batch_size=2
            --per_device_eval_batch_size=1
        """.split()

        with patch.object(sys, "argv", testargs):
            run_squad.main()
            result = get_results(tmp_dir)
            self.assertGreaterEqual(result["eval_f1"], 30)
            self.assertGreaterEqual(result["eval_exact"], 30)

    def test_run_squad_seq2seq(self):
        tmp_dir = self.get_auto_remove_tmp_dir()
        testargs = f"""
            run_seq2seq_qa.py
            --model_name_or_path t5-small
            --context_column context
            --question_column question
            --answer_column answers
            --version_2_with_negative
            --train_file tests/fixtures/tests_samples/SQUAD/sample.json
            --validation_file tests/fixtures/tests_samples/SQUAD/sample.json
            --output_dir {tmp_dir}
            --overwrite_output_dir
            --max_steps=10
            --warmup_steps=2
            --do_train
            --do_eval
            --learning_rate=2e-4
            --per_device_train_batch_size=2
            --per_device_eval_batch_size=1
            --predict_with_generate
        """.split()

        with patch.object(sys, "argv", testargs):
            run_squad_seq2seq.main()
            result = get_results(tmp_dir)
            self.assertGreaterEqual(result["eval_f1"], 30)
            self.assertGreaterEqual(result["eval_exact"], 30)

    def test_run_swag(self):
        tmp_dir = self.get_auto_remove_tmp_dir()
        testargs = f"""
            run_swag.py
            --model_name_or_path bert-base-uncased
            --train_file tests/fixtures/tests_samples/swag/sample.json
            --validation_file tests/fixtures/tests_samples/swag/sample.json
            --output_dir {tmp_dir}
            --overwrite_output_dir
            --max_steps=20
            --warmup_steps=2
            --do_train
            --do_eval
            --learning_rate=2e-4
            --per_device_train_batch_size=2
            --per_device_eval_batch_size=1
        """.split()

        with patch.object(sys, "argv", testargs):
            run_swag.main()
            result = get_results(tmp_dir)
            self.assertGreaterEqual(result["eval_accuracy"], 0.8)

    def test_generation(self):
        testargs = ["run_generation.py", "--prompt=Hello", "--length=10", "--seed=42"]

        if is_cuda_and_apex_available():
            testargs.append("--fp16")

        model_type, model_name = (
            "--model_type=gpt2",
            "--model_name_or_path=sshleifer/tiny-gpt2",
        )
        with patch.object(sys, "argv", testargs + [model_type, model_name]):
            result = run_generation.main()
            self.assertGreaterEqual(len(result[0]), 10)

    @slow
    def test_run_summarization(self):
        tmp_dir = self.get_auto_remove_tmp_dir()
        testargs = f"""
            run_summarization.py
            --model_name_or_path t5-small
            --train_file tests/fixtures/tests_samples/xsum/sample.json
            --validation_file tests/fixtures/tests_samples/xsum/sample.json
            --output_dir {tmp_dir}
            --overwrite_output_dir
            --max_steps=50
            --warmup_steps=8
            --do_train
            --do_eval
            --learning_rate=2e-4
            --per_device_train_batch_size=2
            --per_device_eval_batch_size=1
            --predict_with_generate
        """.split()

        with patch.object(sys, "argv", testargs):
            run_summarization.main()
            result = get_results(tmp_dir)
            self.assertGreaterEqual(result["eval_rouge1"], 10)
            self.assertGreaterEqual(result["eval_rouge2"], 2)
            self.assertGreaterEqual(result["eval_rougeL"], 7)
            self.assertGreaterEqual(result["eval_rougeLsum"], 7)

    @slow
    def test_run_translation(self):
        tmp_dir = self.get_auto_remove_tmp_dir()
        testargs = f"""
            run_translation.py
            --model_name_or_path sshleifer/student_marian_en_ro_6_1
            --source_lang en
            --target_lang ro
            --train_file tests/fixtures/tests_samples/wmt16/sample.json
            --validation_file tests/fixtures/tests_samples/wmt16/sample.json
            --output_dir {tmp_dir}
            --overwrite_output_dir
            --max_steps=50
            --warmup_steps=8
            --do_train
            --do_eval
            --learning_rate=3e-3
            --per_device_train_batch_size=2
            --per_device_eval_batch_size=1
            --predict_with_generate
            --source_lang en_XX
            --target_lang ro_RO
        """.split()

        with patch.object(sys, "argv", testargs):
            run_translation.main()
            result = get_results(tmp_dir)
            self.assertGreaterEqual(result["eval_bleu"], 30)

    def test_run_image_classification(self):
        tmp_dir = self.get_auto_remove_tmp_dir()
        testargs = f"""
            run_image_classification.py
            --output_dir {tmp_dir}
            --model_name_or_path google/vit-base-patch16-224-in21k
            --dataset_name hf-internal-testing/cats_vs_dogs_sample
            --do_train
            --do_eval
            --learning_rate 1e-4
            --per_device_train_batch_size 2
            --per_device_eval_batch_size 1
            --remove_unused_columns False
            --overwrite_output_dir True
            --dataloader_num_workers 16
            --metric_for_best_model accuracy
            --max_steps 10
            --train_val_split 0.1
            --seed 42
        """.split()

        if is_cuda_and_apex_available():
            testargs.append("--fp16")

        with patch.object(sys, "argv", testargs):
            run_image_classification.main()
            result = get_results(tmp_dir)
            self.assertGreaterEqual(result["eval_accuracy"], 0.8)

    def test_run_speech_recognition_ctc(self):
        tmp_dir = self.get_auto_remove_tmp_dir()
        testargs = f"""
            run_speech_recognition_ctc.py
            --output_dir {tmp_dir}
            --model_name_or_path hf-internal-testing/tiny-random-wav2vec2
            --dataset_name hf-internal-testing/librispeech_asr_dummy
            --dataset_config_name clean
            --train_split_name validation
            --eval_split_name validation
            --do_train
            --do_eval
            --learning_rate 1e-4
            --per_device_train_batch_size 2
            --per_device_eval_batch_size 1
            --remove_unused_columns False
            --overwrite_output_dir True
            --preprocessing_num_workers 16
            --max_steps 10
            --seed 42
        """.split()

        if is_cuda_and_apex_available():
            testargs.append("--fp16")

        with patch.object(sys, "argv", testargs):
            run_speech_recognition_ctc.main()
            result = get_results(tmp_dir)
            self.assertLess(result["eval_loss"], result["train_loss"])

    def test_run_speech_recognition_seq2seq(self):
        tmp_dir = self.get_auto_remove_tmp_dir()
        testargs = f"""
            run_speech_recognition_seq2seq.py
            --output_dir {tmp_dir}
            --model_name_or_path hf-internal-testing/tiny-random-speech-encoder-decoder
            --dataset_name hf-internal-testing/librispeech_asr_dummy
            --dataset_config_name clean
            --train_split_name validation
            --eval_split_name validation
            --do_train
            --do_eval
            --learning_rate 1e-4
            --per_device_train_batch_size 2
            --per_device_eval_batch_size 4
            --remove_unused_columns False
            --overwrite_output_dir True
            --preprocessing_num_workers 16
            --max_steps 10
            --seed 42
        """.split()

        if is_cuda_and_apex_available():
            testargs.append("--fp16")

        with patch.object(sys, "argv", testargs):
            run_speech_recognition_seq2seq.main()
            result = get_results(tmp_dir)
            self.assertLess(result["eval_loss"], result["train_loss"])

    def test_run_audio_classification(self):
        tmp_dir = self.get_auto_remove_tmp_dir()
        testargs = f"""
            run_audio_classification.py
            --output_dir {tmp_dir}
            --model_name_or_path hf-internal-testing/tiny-random-wav2vec2
            --dataset_name anton-l/superb_demo
            --dataset_config_name ks
            --train_split_name test
            --eval_split_name test
            --audio_column_name audio
            --label_column_name label
            --do_train
            --do_eval
            --learning_rate 1e-4
            --per_device_train_batch_size 2
            --per_device_eval_batch_size 1
            --remove_unused_columns False
            --overwrite_output_dir True
            --num_train_epochs 10
            --max_steps 50
            --seed 42
        """.split()

        if is_cuda_and_apex_available():
            testargs.append("--fp16")

        with patch.object(sys, "argv", testargs):
            run_audio_classification.main()
            result = get_results(tmp_dir)
            self.assertLess(result["eval_loss"], result["train_loss"])

    def test_run_wav2vec2_pretraining(self):
        tmp_dir = self.get_auto_remove_tmp_dir()
        testargs = f"""
            run_wav2vec2_pretraining_no_trainer.py
            --output_dir {tmp_dir}
            --model_name_or_path hf-internal-testing/tiny-random-wav2vec2
            --dataset_name hf-internal-testing/librispeech_asr_dummy
            --dataset_config_names clean
            --dataset_split_names validation
            --learning_rate 1e-4
            --per_device_train_batch_size 4
            --per_device_eval_batch_size 4
            --preprocessing_num_workers 16
            --max_train_steps 2
            --validation_split_percentage 5
            --seed 42
        """.split()

        if is_cuda_and_apex_available():
            testargs.append("--fp16")

        with patch.object(sys, "argv", testargs):
            run_wav2vec2_pretraining_no_trainer.main()
            model = Wav2Vec2ForPreTraining.from_pretrained(tmp_dir)
            self.assertIsNotNone(model)

    def test_run_vit_mae_pretraining(self):
        tmp_dir = self.get_auto_remove_tmp_dir()
        testargs = f"""
            run_mae.py
            --output_dir {tmp_dir}
            --dataset_name hf-internal-testing/cats_vs_dogs_sample
            --do_train
            --do_eval
            --learning_rate 1e-4
            --per_device_train_batch_size 2
            --per_device_eval_batch_size 1
            --remove_unused_columns False
            --overwrite_output_dir True
            --dataloader_num_workers 16
            --metric_for_best_model accuracy
            --max_steps 10
            --train_val_split 0.1
            --seed 42
        """.split()

        if is_cuda_and_apex_available():
            testargs.append("--fp16")

        with patch.object(sys, "argv", testargs):
            run_mae.main()
            model = ViTMAEForPreTraining.from_pretrained(tmp_dir)
            self.assertIsNotNone(model)

    def test_run_semantic_segmentation(self):
        tmp_dir = self.get_auto_remove_tmp_dir()
        testargs = f"""
            run_semantic_segmentation.py
            --output_dir {tmp_dir}
            --dataset_name huggingface/semantic-segmentation-test-sample
            --do_train
            --do_eval
            --remove_unused_columns False
            --overwrite_output_dir True
            --max_steps 10
            --learning_rate=2e-4
            --per_device_train_batch_size=2
            --per_device_eval_batch_size=1
            --seed 32
        """.split()

        if is_cuda_and_apex_available():
            testargs.append("--fp16")

        with patch.object(sys, "argv", testargs):
            run_semantic_segmentation.main()
            result = get_results(tmp_dir)
            self.assertGreaterEqual(result["eval_overall_accuracy"], 0.1)