File size: 17,143 Bytes
2f5127c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
# Copyright 2020-2025 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 gc
import itertools
import tempfile
import unittest

import pytest
import torch
from accelerate.utils.memory import release_memory
from datasets import load_dataset
from parameterized import parameterized
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
from transformers.testing_utils import (
    backend_empty_cache,
    require_liger_kernel,
    require_peft,
    require_torch_accelerator,
    require_torch_multi_accelerator,
    torch_device,
)
from transformers.utils import is_peft_available

from trl import SFTConfig, SFTTrainer
from trl.models.utils import setup_chat_format

from ..testing_utils import require_bitsandbytes
from .testing_constants import DEVICE_MAP_OPTIONS, GRADIENT_CHECKPOINTING_KWARGS, MODELS_TO_TEST, PACKING_OPTIONS


if is_peft_available():
    from peft import LoraConfig, PeftModel


@pytest.mark.slow
@require_torch_accelerator
@require_peft
class SFTTrainerSlowTester(unittest.TestCase):
    def setUp(self):
        self.train_dataset = load_dataset("stanfordnlp/imdb", split="train[:10%]")
        self.eval_dataset = load_dataset("stanfordnlp/imdb", split="test[:10%]")
        self.max_length = 128
        self.peft_config = LoraConfig(
            lora_alpha=16,
            lora_dropout=0.1,
            r=8,
            bias="none",
            task_type="CAUSAL_LM",
        )

    def tearDown(self):
        gc.collect()
        backend_empty_cache(torch_device)
        gc.collect()

    @parameterized.expand(list(itertools.product(MODELS_TO_TEST, PACKING_OPTIONS)))
    def test_sft_trainer_str(self, model_name, packing):
        """
        Simply tests if passing a simple str to `SFTTrainer` loads and runs the trainer
        as expected.
        """
        with tempfile.TemporaryDirectory() as tmp_dir:
            training_args = SFTConfig(
                output_dir=tmp_dir,
                logging_strategy="no",
                report_to="none",
                per_device_train_batch_size=2,
                max_steps=10,
                packing=packing,
                max_length=self.max_length,
            )

            trainer = SFTTrainer(
                model_name,
                args=training_args,
                train_dataset=self.train_dataset,
                eval_dataset=self.eval_dataset,
            )

            trainer.train()

    @parameterized.expand(list(itertools.product(MODELS_TO_TEST, PACKING_OPTIONS)))
    def test_sft_trainer_transformers(self, model_name, packing):
        """
        Simply tests if passing a transformers model to `SFTTrainer` loads and runs the trainer
        as expected.
        """
        with tempfile.TemporaryDirectory() as tmp_dir:
            training_args = SFTConfig(
                output_dir=tmp_dir,
                logging_strategy="no",
                report_to="none",
                per_device_train_batch_size=2,
                max_steps=10,
                packing=packing,
                max_length=self.max_length,
            )

            model = AutoModelForCausalLM.from_pretrained(model_name)
            tokenizer = AutoTokenizer.from_pretrained(model_name)

            trainer = SFTTrainer(
                model,
                args=training_args,
                processing_class=tokenizer,
                train_dataset=self.train_dataset,
                eval_dataset=self.eval_dataset,
            )

            trainer.train()

        release_memory(model, trainer)

    @parameterized.expand(list(itertools.product(MODELS_TO_TEST, PACKING_OPTIONS)))
    @require_peft
    def test_sft_trainer_peft(self, model_name, packing):
        """
        Simply tests if passing a transformers model + peft config to `SFTTrainer` loads and runs the trainer
        as expected.
        """
        with tempfile.TemporaryDirectory() as tmp_dir:
            training_args = SFTConfig(
                output_dir=tmp_dir,
                logging_strategy="no",
                report_to="none",
                per_device_train_batch_size=2,
                max_steps=10,
                fp16=True,
                packing=packing,
                max_length=self.max_length,
            )

            model = AutoModelForCausalLM.from_pretrained(model_name)
            tokenizer = AutoTokenizer.from_pretrained(model_name)

            trainer = SFTTrainer(
                model,
                args=training_args,
                processing_class=tokenizer,
                train_dataset=self.train_dataset,
                eval_dataset=self.eval_dataset,
                peft_config=self.peft_config,
            )

            self.assertIsInstance(trainer.model, PeftModel)

            trainer.train()

        release_memory(model, trainer)

    @parameterized.expand(list(itertools.product(MODELS_TO_TEST, PACKING_OPTIONS)))
    def test_sft_trainer_transformers_mp(self, model_name, packing):
        """
        Simply tests if passing a transformers model to `SFTTrainer` loads and runs the trainer
        as expected in mixed precision.
        """
        with tempfile.TemporaryDirectory() as tmp_dir:
            training_args = SFTConfig(
                output_dir=tmp_dir,
                logging_strategy="no",
                report_to="none",
                per_device_train_batch_size=2,
                max_steps=10,
                fp16=True,  # this is sufficient to enable amp
                packing=packing,
                max_length=self.max_length,
            )

            model = AutoModelForCausalLM.from_pretrained(model_name)
            tokenizer = AutoTokenizer.from_pretrained(model_name)

            trainer = SFTTrainer(
                model,
                args=training_args,
                processing_class=tokenizer,
                train_dataset=self.train_dataset,
                eval_dataset=self.eval_dataset,
            )

            trainer.train()

        release_memory(model, trainer)

    @parameterized.expand(list(itertools.product(MODELS_TO_TEST, PACKING_OPTIONS, GRADIENT_CHECKPOINTING_KWARGS)))
    def test_sft_trainer_transformers_mp_gc(self, model_name, packing, gradient_checkpointing_kwargs):
        """
        Simply tests if passing a transformers model to `SFTTrainer` loads and runs the trainer
        as expected in mixed precision + different scenarios of gradient_checkpointing.
        """
        with tempfile.TemporaryDirectory() as tmp_dir:
            training_args = SFTConfig(
                output_dir=tmp_dir,
                logging_strategy="no",
                report_to="none",
                per_device_train_batch_size=2,
                max_steps=10,
                packing=packing,
                max_length=self.max_length,
                fp16=True,  # this is sufficient to enable amp
                gradient_checkpointing=True,
                gradient_checkpointing_kwargs=gradient_checkpointing_kwargs,
            )

            model = AutoModelForCausalLM.from_pretrained(model_name)
            tokenizer = AutoTokenizer.from_pretrained(model_name)

            trainer = SFTTrainer(
                model,
                args=training_args,
                processing_class=tokenizer,
                train_dataset=self.train_dataset,
                eval_dataset=self.eval_dataset,
            )

            trainer.train()

        release_memory(model, trainer)

    @parameterized.expand(list(itertools.product(MODELS_TO_TEST, PACKING_OPTIONS, GRADIENT_CHECKPOINTING_KWARGS)))
    @require_peft
    def test_sft_trainer_transformers_mp_gc_peft(self, model_name, packing, gradient_checkpointing_kwargs):
        """
        Simply tests if passing a transformers model + PEFT to `SFTTrainer` loads and runs the trainer
        as expected in mixed precision + different scenarios of gradient_checkpointing.
        """
        with tempfile.TemporaryDirectory() as tmp_dir:
            training_args = SFTConfig(
                output_dir=tmp_dir,
                logging_strategy="no",
                report_to="none",
                per_device_train_batch_size=2,
                max_steps=10,
                packing=packing,
                max_length=self.max_length,
                fp16=True,  # this is sufficient to enable amp
                gradient_checkpointing=True,
                gradient_checkpointing_kwargs=gradient_checkpointing_kwargs,
            )

            model = AutoModelForCausalLM.from_pretrained(model_name)
            tokenizer = AutoTokenizer.from_pretrained(model_name)

            trainer = SFTTrainer(
                model,
                args=training_args,
                processing_class=tokenizer,
                train_dataset=self.train_dataset,
                eval_dataset=self.eval_dataset,
                peft_config=self.peft_config,
            )

            self.assertIsInstance(trainer.model, PeftModel)

            trainer.train()

        release_memory(model, trainer)

    @parameterized.expand(
        list(itertools.product(MODELS_TO_TEST, PACKING_OPTIONS, GRADIENT_CHECKPOINTING_KWARGS, DEVICE_MAP_OPTIONS))
    )
    @require_torch_multi_accelerator
    def test_sft_trainer_transformers_mp_gc_device_map(
        self, model_name, packing, gradient_checkpointing_kwargs, device_map
    ):
        """
        Simply tests if passing a transformers model to `SFTTrainer` loads and runs the trainer
        as expected in mixed precision + different scenarios of gradient_checkpointing (single, multi-gpu, etc).
        """
        with tempfile.TemporaryDirectory() as tmp_dir:
            training_args = SFTConfig(
                output_dir=tmp_dir,
                logging_strategy="no",
                report_to="none",
                per_device_train_batch_size=2,
                max_steps=10,
                packing=packing,
                max_length=self.max_length,
                fp16=True,  # this is sufficient to enable amp
                gradient_checkpointing=True,
                gradient_checkpointing_kwargs=gradient_checkpointing_kwargs,
            )

            model = AutoModelForCausalLM.from_pretrained(model_name, device_map=device_map)
            tokenizer = AutoTokenizer.from_pretrained(model_name)

            trainer = SFTTrainer(
                model,
                args=training_args,
                processing_class=tokenizer,
                train_dataset=self.train_dataset,
                eval_dataset=self.eval_dataset,
            )

            trainer.train()

        release_memory(model, trainer)

    @parameterized.expand(list(itertools.product(MODELS_TO_TEST, PACKING_OPTIONS, GRADIENT_CHECKPOINTING_KWARGS)))
    @require_peft
    @require_bitsandbytes
    def test_sft_trainer_transformers_mp_gc_peft_qlora(self, model_name, packing, gradient_checkpointing_kwargs):
        """
        Simply tests if passing a transformers model + PEFT + bnb to `SFTTrainer` loads and runs the trainer
        as expected in mixed precision + different scenarios of gradient_checkpointing.
        """
        with tempfile.TemporaryDirectory() as tmp_dir:
            training_args = SFTConfig(
                output_dir=tmp_dir,
                logging_strategy="no",
                report_to="none",
                per_device_train_batch_size=2,
                max_steps=10,
                packing=packing,
                max_length=self.max_length,
                fp16=True,  # this is sufficient to enable amp
                gradient_checkpointing=True,
                gradient_checkpointing_kwargs=gradient_checkpointing_kwargs,
            )

            quantization_config = BitsAndBytesConfig(load_in_4bit=True, bnb_4bit_compute_dtype=torch.float16)

            model = AutoModelForCausalLM.from_pretrained(model_name, quantization_config=quantization_config)
            tokenizer = AutoTokenizer.from_pretrained(model_name)

            trainer = SFTTrainer(
                model,
                args=training_args,
                processing_class=tokenizer,
                train_dataset=self.train_dataset,
                eval_dataset=self.eval_dataset,
                peft_config=self.peft_config,
            )

            self.assertIsInstance(trainer.model, PeftModel)

            trainer.train()

        release_memory(model, trainer)

    @parameterized.expand(list(itertools.product(MODELS_TO_TEST, PACKING_OPTIONS)))
    @require_peft
    @require_bitsandbytes
    def test_sft_trainer_with_chat_format_qlora(self, model_name, packing):
        """
        Simply tests if using setup_chat_format with a transformers model + peft + bnb config to `SFTTrainer` loads and runs the trainer
        as expected.
        """
        with tempfile.TemporaryDirectory() as tmp_dir:
            train_dataset = load_dataset("trl-internal-testing/dolly-chatml-sft", split="train")

            training_args = SFTConfig(
                packing=packing,
                max_length=self.max_length,
                output_dir=tmp_dir,
                logging_strategy="no",
                report_to="none",
                per_device_train_batch_size=2,
                max_steps=10,
                fp16=True,
            )

            quantization_config = BitsAndBytesConfig(load_in_4bit=True, bnb_4bit_compute_dtype=torch.float16)

            model = AutoModelForCausalLM.from_pretrained(model_name, quantization_config=quantization_config)
            tokenizer = AutoTokenizer.from_pretrained(model_name)

            if tokenizer.chat_template is None:
                model, tokenizer = setup_chat_format(model, tokenizer)

            trainer = SFTTrainer(
                model,
                args=training_args,
                processing_class=tokenizer,
                train_dataset=train_dataset,
                peft_config=self.peft_config,
            )

            self.assertIsInstance(trainer.model, PeftModel)

            trainer.train()

        release_memory(model, trainer)

    @parameterized.expand(list(itertools.product(MODELS_TO_TEST, PACKING_OPTIONS)))
    @require_liger_kernel
    def test_sft_trainer_with_liger(self, model_name, packing):
        """
        Tests if passing use_liger=True to SFTConfig loads and runs the trainer
        with AutoLigerKernelForCausalLM as expected.
        """
        with tempfile.TemporaryDirectory() as tmp_dir:
            training_args = SFTConfig(
                output_dir=tmp_dir,
                logging_strategy="no",
                report_to="none",
                per_device_train_batch_size=2,
                max_steps=2,
                packing=packing,
                max_length=self.max_length,
                use_liger_kernel=True,
            )

            trainer = SFTTrainer(
                model_name,
                args=training_args,
                train_dataset=self.train_dataset,
                eval_dataset=self.eval_dataset,
            )

            trainer.train()

        release_memory(trainer.model, trainer)

    @parameterized.expand(list(itertools.product(MODELS_TO_TEST, PACKING_OPTIONS)))
    @require_torch_accelerator
    def test_train_offloading(self, model_name, packing):
        """Test that activation offloading works with SFTTrainer."""

        with tempfile.TemporaryDirectory() as tmp_dir:
            # Initialize the trainer
            training_args = SFTConfig(
                output_dir=tmp_dir,
                activation_offloading=True,
                report_to="none",
                per_device_train_batch_size=2,
                max_steps=2,
                packing=packing,
                max_length=self.max_length,
            )
            trainer = SFTTrainer(
                model=model_name, args=training_args, train_dataset=self.train_dataset, eval_dataset=self.eval_dataset
            )

            # Save the initial parameters to compare them later
            previous_trainable_params = {n: param.clone() for n, param in trainer.model.named_parameters()}

            # Train the model
            trainer.train()

            # Check that the training loss is not None
            self.assertIsNotNone(trainer.state.log_history[-1]["train_loss"])

            # Check the params have changed
            for n, param in previous_trainable_params.items():
                new_param = trainer.model.get_parameter(n)
                self.assertFalse(torch.allclose(param, new_param), f"Parameter {n} has not changed")

            release_memory(trainer.model, trainer)