train_dreambooth.py 29.5 KB
Newer Older
1
import argparse
2
import hashlib
3
import inspect
4
import itertools
5
6
7
8
9
10
11
12
13
14
15
16
17
import math
import os
from pathlib import Path
from typing import Optional

import torch
import torch.nn.functional as F
import torch.utils.checkpoint
from torch.utils.data import Dataset

from accelerate import Accelerator
from accelerate.logging import get_logger
from accelerate.utils import set_seed
18
from diffusers import AutoencoderKL, DDPMScheduler, DiffusionPipeline, UNet2DConditionModel
19
from diffusers.optimization import get_scheduler
20
from diffusers.utils import check_min_version
21
from diffusers.utils.import_utils import is_xformers_available
22
23
24
25
from huggingface_hub import HfFolder, Repository, whoami
from PIL import Image
from torchvision import transforms
from tqdm.auto import tqdm
26
from transformers import AutoTokenizer, PretrainedConfig
27
28


29
30
31
# Will error if the minimal version of diffusers is not installed. Remove at your own risks.
check_min_version("0.10.0.dev0")

32
33
34
logger = get_logger(__name__)


35
def import_model_class_from_model_name_or_path(pretrained_model_name_or_path: str, revision: str):
36
37
38
    text_encoder_config = PretrainedConfig.from_pretrained(
        pretrained_model_name_or_path,
        subfolder="text_encoder",
39
        revision=revision,
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
    )
    model_class = text_encoder_config.architectures[0]

    if model_class == "CLIPTextModel":
        from transformers import CLIPTextModel

        return CLIPTextModel
    elif model_class == "RobertaSeriesModelWithTransformation":
        from diffusers.pipelines.alt_diffusion.modeling_roberta_series import RobertaSeriesModelWithTransformation

        return RobertaSeriesModelWithTransformation
    else:
        raise ValueError(f"{model_class} is not supported.")


Suraj Patil's avatar
Suraj Patil committed
55
def parse_args(input_args=None):
56
57
58
59
60
61
62
63
    parser = argparse.ArgumentParser(description="Simple example of a training script.")
    parser.add_argument(
        "--pretrained_model_name_or_path",
        type=str,
        default=None,
        required=True,
        help="Path to pretrained model or model identifier from huggingface.co/models.",
    )
64
65
66
67
68
69
70
    parser.add_argument(
        "--revision",
        type=str,
        default=None,
        required=False,
        help="Revision of pretrained model identifier from huggingface.co/models.",
    )
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
    parser.add_argument(
        "--tokenizer_name",
        type=str,
        default=None,
        help="Pretrained tokenizer name or path if not the same as model_name",
    )
    parser.add_argument(
        "--instance_data_dir",
        type=str,
        default=None,
        required=True,
        help="A folder containing the training data of instance images.",
    )
    parser.add_argument(
        "--class_data_dir",
        type=str,
        default=None,
        required=False,
        help="A folder containing the training data of class images.",
    )
    parser.add_argument(
        "--instance_prompt",
        type=str,
        default=None,
95
        required=True,
Yuta Hayashibe's avatar
Yuta Hayashibe committed
96
        help="The prompt with identifier specifying the instance",
97
98
99
100
101
    )
    parser.add_argument(
        "--class_prompt",
        type=str,
        default=None,
Yuta Hayashibe's avatar
Yuta Hayashibe committed
102
        help="The prompt to specify images in the same class as provided instance images.",
103
104
105
106
107
    )
    parser.add_argument(
        "--with_prior_preservation",
        default=False,
        action="store_true",
Yuta Hayashibe's avatar
Yuta Hayashibe committed
108
        help="Flag to add prior preservation loss.",
109
110
111
112
113
114
115
    )
    parser.add_argument("--prior_loss_weight", type=float, default=1.0, help="The weight of prior preservation loss.")
    parser.add_argument(
        "--num_class_images",
        type=int,
        default=100,
        help=(
116
117
            "Minimal class images for prior preservation loss. If there are not enough images already present in"
            " class_data_dir, additional images will be sampled with class_prompt."
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
        ),
    )
    parser.add_argument(
        "--output_dir",
        type=str,
        default="text-inversion-model",
        help="The output directory where the model predictions and checkpoints will be written.",
    )
    parser.add_argument("--seed", type=int, default=None, help="A seed for reproducible training.")
    parser.add_argument(
        "--resolution",
        type=int,
        default=512,
        help=(
            "The resolution for input images, all the images in the train/validation dataset will be resized to this"
            " resolution"
        ),
    )
    parser.add_argument(
        "--center_crop", action="store_true", help="Whether to center crop images before resizing to resolution"
    )
139
    parser.add_argument("--train_text_encoder", action="store_true", help="Whether to train the text encoder")
140
141
142
143
144
145
146
147
148
149
150
151
152
    parser.add_argument(
        "--train_batch_size", type=int, default=4, help="Batch size (per device) for the training dataloader."
    )
    parser.add_argument(
        "--sample_batch_size", type=int, default=4, help="Batch size (per device) for sampling images."
    )
    parser.add_argument("--num_train_epochs", type=int, default=1)
    parser.add_argument(
        "--max_train_steps",
        type=int,
        default=None,
        help="Total number of training steps to perform.  If provided, overrides num_train_epochs.",
    )
153
    parser.add_argument("--save_steps", type=int, default=500, help="Save checkpoint every X updates steps.")
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
    parser.add_argument(
        "--gradient_accumulation_steps",
        type=int,
        default=1,
        help="Number of updates steps to accumulate before performing a backward/update pass.",
    )
    parser.add_argument(
        "--gradient_checkpointing",
        action="store_true",
        help="Whether or not to use gradient checkpointing to save memory at the expense of slower backward pass.",
    )
    parser.add_argument(
        "--learning_rate",
        type=float,
        default=5e-6,
        help="Initial learning rate (after the potential warmup period) to use.",
    )
    parser.add_argument(
        "--scale_lr",
        action="store_true",
        default=False,
        help="Scale the learning rate by the number of GPUs, gradient accumulation steps, and batch size.",
    )
    parser.add_argument(
        "--lr_scheduler",
        type=str,
        default="constant",
        help=(
            'The scheduler type to use. Choose between ["linear", "cosine", "cosine_with_restarts", "polynomial",'
            ' "constant", "constant_with_warmup"]'
        ),
    )
    parser.add_argument(
        "--lr_warmup_steps", type=int, default=500, help="Number of steps for the warmup in the lr scheduler."
    )
    parser.add_argument(
        "--use_8bit_adam", action="store_true", help="Whether or not to use 8-bit Adam from bitsandbytes."
    )
    parser.add_argument("--adam_beta1", type=float, default=0.9, help="The beta1 parameter for the Adam optimizer.")
    parser.add_argument("--adam_beta2", type=float, default=0.999, help="The beta2 parameter for the Adam optimizer.")
    parser.add_argument("--adam_weight_decay", type=float, default=1e-2, help="Weight decay to use.")
    parser.add_argument("--adam_epsilon", type=float, default=1e-08, help="Epsilon value for the Adam optimizer")
    parser.add_argument("--max_grad_norm", default=1.0, type=float, help="Max gradient norm.")
    parser.add_argument("--push_to_hub", action="store_true", help="Whether or not to push the model to the Hub.")
    parser.add_argument("--hub_token", type=str, default=None, help="The token to use to push to the Model Hub.")
    parser.add_argument(
        "--hub_model_id",
        type=str,
        default=None,
        help="The name of the repository to keep in sync with the local `output_dir`.",
    )
    parser.add_argument(
        "--logging_dir",
        type=str,
        default="logs",
        help=(
            "[TensorBoard](https://www.tensorflow.org/tensorboard) log directory. Will default to"
            " *output_dir/runs/**CURRENT_DATETIME_HOSTNAME***."
        ),
    )
    parser.add_argument(
        "--mixed_precision",
        type=str,
217
        default=None,
218
219
        choices=["no", "fp16", "bf16"],
        help=(
220
221
222
            "Whether to use mixed precision. Choose between fp16 and bf16 (bfloat16). Bf16 requires PyTorch >="
            " 1.10.and an Nvidia Ampere GPU.  Default to the value of accelerate config of the current system or the"
            " flag passed with the `accelerate.launch` command. Use this argument to override the accelerate config."
223
224
225
226
        ),
    )
    parser.add_argument("--local_rank", type=int, default=-1, help="For distributed training: local_rank")

227
228
229
230
231
    if input_args is not None:
        args = parser.parse_args(input_args)
    else:
        args = parser.parse_args()

232
233
234
235
236
237
238
239
240
    env_local_rank = int(os.environ.get("LOCAL_RANK", -1))
    if env_local_rank != -1 and env_local_rank != args.local_rank:
        args.local_rank = env_local_rank

    if args.with_prior_preservation:
        if args.class_data_dir is None:
            raise ValueError("You must specify a data directory for class images.")
        if args.class_prompt is None:
            raise ValueError("You must specify prompt for class images.")
241
242
243
244
245
    else:
        if args.class_data_dir is not None:
            logger.warning("You need not use --class_data_dir without --with_prior_preservation.")
        if args.class_prompt is not None:
            logger.warning("You need not use --class_prompt without --with_prior_preservation.")
246
247
248
249
250
251

    return args


class DreamBoothDataset(Dataset):
    """
Yuta Hayashibe's avatar
Yuta Hayashibe committed
252
    A dataset to prepare the instance and class images with the prompts for fine-tuning the model.
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
    It pre-processes the images and the tokenizes prompts.
    """

    def __init__(
        self,
        instance_data_root,
        instance_prompt,
        tokenizer,
        class_data_root=None,
        class_prompt=None,
        size=512,
        center_crop=False,
    ):
        self.size = size
        self.center_crop = center_crop
        self.tokenizer = tokenizer

        self.instance_data_root = Path(instance_data_root)
        if not self.instance_data_root.exists():
            raise ValueError("Instance images root doesn't exists.")

        self.instance_images_path = list(Path(instance_data_root).iterdir())
        self.num_instance_images = len(self.instance_images_path)
        self.instance_prompt = instance_prompt
        self._length = self.num_instance_images

        if class_data_root is not None:
            self.class_data_root = Path(class_data_root)
            self.class_data_root.mkdir(parents=True, exist_ok=True)
282
            self.class_images_path = list(self.class_data_root.iterdir())
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
            self.num_class_images = len(self.class_images_path)
            self._length = max(self.num_class_images, self.num_instance_images)
            self.class_prompt = class_prompt
        else:
            self.class_data_root = None

        self.image_transforms = transforms.Compose(
            [
                transforms.Resize(size, interpolation=transforms.InterpolationMode.BILINEAR),
                transforms.CenterCrop(size) if center_crop else transforms.RandomCrop(size),
                transforms.ToTensor(),
                transforms.Normalize([0.5], [0.5]),
            ]
        )

    def __len__(self):
        return self._length

    def __getitem__(self, index):
        example = {}
        instance_image = Image.open(self.instance_images_path[index % self.num_instance_images])
        if not instance_image.mode == "RGB":
            instance_image = instance_image.convert("RGB")
        example["instance_images"] = self.image_transforms(instance_image)
        example["instance_prompt_ids"] = self.tokenizer(
            self.instance_prompt,
            truncation=True,
310
            padding="max_length",
311
            max_length=self.tokenizer.model_max_length,
312
            return_tensors="pt",
313
314
315
316
317
318
319
320
321
322
        ).input_ids

        if self.class_data_root:
            class_image = Image.open(self.class_images_path[index % self.num_class_images])
            if not class_image.mode == "RGB":
                class_image = class_image.convert("RGB")
            example["class_images"] = self.image_transforms(class_image)
            example["class_prompt_ids"] = self.tokenizer(
                self.class_prompt,
                truncation=True,
323
                padding="max_length",
324
                max_length=self.tokenizer.model_max_length,
325
                return_tensors="pt",
326
327
328
329
330
            ).input_ids

        return example


331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
def collate_fn(examples, with_prior_preservation=False):
    input_ids = [example["instance_prompt_ids"] for example in examples]
    pixel_values = [example["instance_images"] for example in examples]

    # Concat class and instance examples for prior preservation.
    # We do this to avoid doing two forward passes.
    if with_prior_preservation:
        input_ids += [example["class_prompt_ids"] for example in examples]
        pixel_values += [example["class_images"] for example in examples]

    pixel_values = torch.stack(pixel_values)
    pixel_values = pixel_values.to(memory_format=torch.contiguous_format).float()

    input_ids = torch.cat(input_ids, dim=0)

    batch = {
        "input_ids": input_ids,
        "pixel_values": pixel_values,
    }
    return batch


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
class PromptDataset(Dataset):
    "A simple dataset to prepare the prompts to generate class images on multiple GPUs."

    def __init__(self, prompt, num_samples):
        self.prompt = prompt
        self.num_samples = num_samples

    def __len__(self):
        return self.num_samples

    def __getitem__(self, index):
        example = {}
        example["prompt"] = self.prompt
        example["index"] = index
        return example


def get_full_repo_name(model_id: str, organization: Optional[str] = None, token: Optional[str] = None):
    if token is None:
        token = HfFolder.get_token()
    if organization is None:
        username = whoami(token)["name"]
        return f"{username}/{model_id}"
    else:
        return f"{organization}/{model_id}"


380
def main(args):
381
382
383
384
385
386
387
388
389
    logging_dir = Path(args.output_dir, args.logging_dir)

    accelerator = Accelerator(
        gradient_accumulation_steps=args.gradient_accumulation_steps,
        mixed_precision=args.mixed_precision,
        log_with="tensorboard",
        logging_dir=logging_dir,
    )

390
391
392
393
394
395
396
397
398
    # Currently, it's not possible to do gradient accumulation when training two models with accelerate.accumulate
    # This will be enabled soon in accelerate. For now, we don't allow gradient accumulation when training two models.
    # TODO (patil-suraj): Remove this check when gradient accumulation with two models is enabled in accelerate.
    if args.train_text_encoder and args.gradient_accumulation_steps > 1 and accelerator.num_processes > 1:
        raise ValueError(
            "Gradient accumulation is not supported when training the text encoder in distributed training. "
            "Please set gradient_accumulation_steps to 1. This feature will be supported in the future."
        )

399
400
401
402
403
404
405
406
407
408
409
    if args.seed is not None:
        set_seed(args.seed)

    if args.with_prior_preservation:
        class_images_dir = Path(args.class_data_dir)
        if not class_images_dir.exists():
            class_images_dir.mkdir(parents=True)
        cur_class_images = len(list(class_images_dir.iterdir()))

        if cur_class_images < args.num_class_images:
            torch_dtype = torch.float16 if accelerator.device.type == "cuda" else torch.float32
410
            pipeline = DiffusionPipeline.from_pretrained(
411
412
413
414
                args.pretrained_model_name_or_path,
                torch_dtype=torch_dtype,
                safety_checker=None,
                revision=args.revision,
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
            )
            pipeline.set_progress_bar_config(disable=True)

            num_new_images = args.num_class_images - cur_class_images
            logger.info(f"Number of class images to sample: {num_new_images}.")

            sample_dataset = PromptDataset(args.class_prompt, num_new_images)
            sample_dataloader = torch.utils.data.DataLoader(sample_dataset, batch_size=args.sample_batch_size)

            sample_dataloader = accelerator.prepare(sample_dataloader)
            pipeline.to(accelerator.device)

            for example in tqdm(
                sample_dataloader, desc="Generating class images", disable=not accelerator.is_local_main_process
            ):
430
                images = pipeline(example["prompt"]).images
431
432

                for i, image in enumerate(images):
433
434
435
                    hash_image = hashlib.sha1(image.tobytes()).hexdigest()
                    image_filename = class_images_dir / f"{example['index'][i] + cur_class_images}-{hash_image}.jpg"
                    image.save(image_filename)
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459

            del pipeline
            if torch.cuda.is_available():
                torch.cuda.empty_cache()

    # Handle the repository creation
    if accelerator.is_main_process:
        if args.push_to_hub:
            if args.hub_model_id is None:
                repo_name = get_full_repo_name(Path(args.output_dir).name, token=args.hub_token)
            else:
                repo_name = args.hub_model_id
            repo = Repository(args.output_dir, clone_from=repo_name)

            with open(os.path.join(args.output_dir, ".gitignore"), "w+") as gitignore:
                if "step_*" not in gitignore:
                    gitignore.write("step_*\n")
                if "epoch_*" not in gitignore:
                    gitignore.write("epoch_*\n")
        elif args.output_dir is not None:
            os.makedirs(args.output_dir, exist_ok=True)

    # Load the tokenizer
    if args.tokenizer_name:
460
        tokenizer = AutoTokenizer.from_pretrained(
461
462
            args.tokenizer_name,
            revision=args.revision,
463
            use_fast=False,
464
        )
465
    elif args.pretrained_model_name_or_path:
466
        tokenizer = AutoTokenizer.from_pretrained(
467
468
469
            args.pretrained_model_name_or_path,
            subfolder="tokenizer",
            revision=args.revision,
470
            use_fast=False,
471
        )
472

473
    # import correct text encoder class
474
    text_encoder_cls = import_model_class_from_model_name_or_path(args.pretrained_model_name_or_path, args.revision)
475

476
    # Load models and create wrapper for stable diffusion
477
    text_encoder = text_encoder_cls.from_pretrained(
478
479
480
481
482
483
484
485
486
487
488
489
490
491
        args.pretrained_model_name_or_path,
        subfolder="text_encoder",
        revision=args.revision,
    )
    vae = AutoencoderKL.from_pretrained(
        args.pretrained_model_name_or_path,
        subfolder="vae",
        revision=args.revision,
    )
    unet = UNet2DConditionModel.from_pretrained(
        args.pretrained_model_name_or_path,
        subfolder="unet",
        revision=args.revision,
    )
492

493
494
    if is_xformers_available():
        try:
495
            unet.enable_xformers_memory_efficient_attention()
496
497
498
499
500
501
        except Exception as e:
            logger.warning(
                "Could not enable memory efficient attention. Make sure xformers is installed"
                f" correctly and a GPU is available: {e}"
            )

502
503
504
505
    vae.requires_grad_(False)
    if not args.train_text_encoder:
        text_encoder.requires_grad_(False)

506
507
    if args.gradient_checkpointing:
        unet.enable_gradient_checkpointing()
508
509
        if args.train_text_encoder:
            text_encoder.gradient_checkpointing_enable()
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528

    if args.scale_lr:
        args.learning_rate = (
            args.learning_rate * args.gradient_accumulation_steps * args.train_batch_size * accelerator.num_processes
        )

    # Use 8-bit Adam for lower memory usage or to fine-tune the model in 16GB GPUs
    if args.use_8bit_adam:
        try:
            import bitsandbytes as bnb
        except ImportError:
            raise ImportError(
                "To use 8-bit Adam, please install the bitsandbytes library: `pip install bitsandbytes`."
            )

        optimizer_class = bnb.optim.AdamW8bit
    else:
        optimizer_class = torch.optim.AdamW

529
530
531
    params_to_optimize = (
        itertools.chain(unet.parameters(), text_encoder.parameters()) if args.train_text_encoder else unet.parameters()
    )
532
    optimizer = optimizer_class(
533
        params_to_optimize,
534
535
536
537
538
539
        lr=args.learning_rate,
        betas=(args.adam_beta1, args.adam_beta2),
        weight_decay=args.adam_weight_decay,
        eps=args.adam_epsilon,
    )

540
    noise_scheduler = DDPMScheduler.from_pretrained(args.pretrained_model_name_or_path, subfolder="scheduler")
541
542
543
544
545
546
547
548
549
550
551
552

    train_dataset = DreamBoothDataset(
        instance_data_root=args.instance_data_dir,
        instance_prompt=args.instance_prompt,
        class_data_root=args.class_data_dir if args.with_prior_preservation else None,
        class_prompt=args.class_prompt,
        tokenizer=tokenizer,
        size=args.resolution,
        center_crop=args.center_crop,
    )

    train_dataloader = torch.utils.data.DataLoader(
553
554
555
556
557
        train_dataset,
        batch_size=args.train_batch_size,
        shuffle=True,
        collate_fn=lambda examples: collate_fn(examples, args.with_prior_preservation),
        num_workers=1,
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
    )

    # Scheduler and math around the number of training steps.
    overrode_max_train_steps = False
    num_update_steps_per_epoch = math.ceil(len(train_dataloader) / args.gradient_accumulation_steps)
    if args.max_train_steps is None:
        args.max_train_steps = args.num_train_epochs * num_update_steps_per_epoch
        overrode_max_train_steps = True

    lr_scheduler = get_scheduler(
        args.lr_scheduler,
        optimizer=optimizer,
        num_warmup_steps=args.lr_warmup_steps * args.gradient_accumulation_steps,
        num_training_steps=args.max_train_steps * args.gradient_accumulation_steps,
    )

574
575
576
577
578
579
580
581
    if args.train_text_encoder:
        unet, text_encoder, optimizer, train_dataloader, lr_scheduler = accelerator.prepare(
            unet, text_encoder, optimizer, train_dataloader, lr_scheduler
        )
    else:
        unet, optimizer, train_dataloader, lr_scheduler = accelerator.prepare(
            unet, optimizer, train_dataloader, lr_scheduler
        )
582

583
    weight_dtype = torch.float32
584
    if accelerator.mixed_precision == "fp16":
585
        weight_dtype = torch.float16
586
    elif accelerator.mixed_precision == "bf16":
587
588
589
590
591
592
        weight_dtype = torch.bfloat16

    # Move text_encode and vae to gpu.
    # For mixed precision training we cast the text_encoder and vae weights to half-precision
    # as these models are only used for inference, keeping weights in full precision is not required.
    vae.to(accelerator.device, dtype=weight_dtype)
593
594
    if not args.train_text_encoder:
        text_encoder.to(accelerator.device, dtype=weight_dtype)
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625

    # We need to recalculate our total training steps as the size of the training dataloader may have changed.
    num_update_steps_per_epoch = math.ceil(len(train_dataloader) / args.gradient_accumulation_steps)
    if overrode_max_train_steps:
        args.max_train_steps = args.num_train_epochs * num_update_steps_per_epoch
    # Afterwards we recalculate our number of training epochs
    args.num_train_epochs = math.ceil(args.max_train_steps / num_update_steps_per_epoch)

    # We need to initialize the trackers we use, and also store our configuration.
    # The trackers initializes automatically on the main process.
    if accelerator.is_main_process:
        accelerator.init_trackers("dreambooth", config=vars(args))

    # Train!
    total_batch_size = args.train_batch_size * accelerator.num_processes * args.gradient_accumulation_steps

    logger.info("***** Running training *****")
    logger.info(f"  Num examples = {len(train_dataset)}")
    logger.info(f"  Num batches each epoch = {len(train_dataloader)}")
    logger.info(f"  Num Epochs = {args.num_train_epochs}")
    logger.info(f"  Instantaneous batch size per device = {args.train_batch_size}")
    logger.info(f"  Total train batch size (w. parallel, distributed & accumulation) = {total_batch_size}")
    logger.info(f"  Gradient Accumulation steps = {args.gradient_accumulation_steps}")
    logger.info(f"  Total optimization steps = {args.max_train_steps}")
    # Only show the progress bar once on each machine.
    progress_bar = tqdm(range(args.max_train_steps), disable=not accelerator.is_local_main_process)
    progress_bar.set_description("Steps")
    global_step = 0

    for epoch in range(args.num_train_epochs):
        unet.train()
626
627
        if args.train_text_encoder:
            text_encoder.train()
628
629
630
        for step, batch in enumerate(train_dataloader):
            with accelerator.accumulate(unet):
                # Convert images to latent space
631
632
                latents = vae.encode(batch["pixel_values"].to(dtype=weight_dtype)).latent_dist.sample()
                latents = latents * 0.18215
633
634

                # Sample noise that we'll add to the latents
635
                noise = torch.randn_like(latents)
636
637
638
639
640
641
642
643
644
645
                bsz = latents.shape[0]
                # Sample a random timestep for each image
                timesteps = torch.randint(0, noise_scheduler.config.num_train_timesteps, (bsz,), device=latents.device)
                timesteps = timesteps.long()

                # Add noise to the latents according to the noise magnitude at each timestep
                # (this is the forward diffusion process)
                noisy_latents = noise_scheduler.add_noise(latents, noise, timesteps)

                # Get the text embedding for conditioning
646
                encoder_hidden_states = text_encoder(batch["input_ids"])[0]
647
648

                # Predict the noise residual
649
650
651
652
653
654
655
656
657
                model_pred = unet(noisy_latents, timesteps, encoder_hidden_states).sample

                # Get the target for loss depending on the prediction type
                if noise_scheduler.config.prediction_type == "epsilon":
                    target = noise
                elif noise_scheduler.config.prediction_type == "v_prediction":
                    target = noise_scheduler.get_velocity(latents, noise, timesteps)
                else:
                    raise ValueError(f"Unknown prediction type {noise_scheduler.config.prediction_type}")
658
659

                if args.with_prior_preservation:
660
661
662
                    # Chunk the noise and model_pred into two parts and compute the loss on each part separately.
                    model_pred, model_pred_prior = torch.chunk(model_pred, 2, dim=0)
                    target, target_prior = torch.chunk(target, 2, dim=0)
663
664

                    # Compute instance loss
665
                    loss = F.mse_loss(model_pred.float(), target.float(), reduction="none").mean([1, 2, 3]).mean()
666
667

                    # Compute prior loss
668
                    prior_loss = F.mse_loss(model_pred_prior.float(), target_prior.float(), reduction="mean")
669
670
671
672

                    # Add the prior loss to the instance loss.
                    loss = loss + args.prior_loss_weight * prior_loss
                else:
673
                    loss = F.mse_loss(model_pred.float(), target.float(), reduction="mean")
674
675

                accelerator.backward(loss)
676
                if accelerator.sync_gradients:
677
678
679
680
681
682
                    params_to_clip = (
                        itertools.chain(unet.parameters(), text_encoder.parameters())
                        if args.train_text_encoder
                        else unet.parameters()
                    )
                    accelerator.clip_grad_norm_(params_to_clip, args.max_grad_norm)
683
684
685
686
687
688
689
690
691
                optimizer.step()
                lr_scheduler.step()
                optimizer.zero_grad()

            # Checks if the accelerator has performed an optimization step behind the scenes
            if accelerator.sync_gradients:
                progress_bar.update(1)
                global_step += 1

692
693
                if global_step % args.save_steps == 0:
                    if accelerator.is_main_process:
694
695
696
697
698
699
700
701
702
                        # When 'keep_fp32_wrapper' is `False` (the default), then the models are
                        # unwrapped and the mixed precision hooks are removed, so training crashes
                        # when the unwrapped models are used for further training.
                        # This is only supported in newer versions of `accelerate`.
                        # TODO(Pedro, Suraj): Remove `accepts_keep_fp32_wrapper` when forcing newer accelerate versions
                        accepts_keep_fp32_wrapper = "keep_fp32_wrapper" in set(
                            inspect.signature(accelerator.unwrap_model).parameters.keys()
                        )
                        extra_args = {"keep_fp32_wrapper": True} if accepts_keep_fp32_wrapper else {}
703
                        pipeline = DiffusionPipeline.from_pretrained(
704
                            args.pretrained_model_name_or_path,
705
706
                            unet=accelerator.unwrap_model(unet, **extra_args),
                            text_encoder=accelerator.unwrap_model(text_encoder, **extra_args),
707
708
709
710
711
                            revision=args.revision,
                        )
                        save_path = os.path.join(args.output_dir, f"checkpoint-{global_step}")
                        pipeline.save_pretrained(save_path)

712
713
714
715
716
717
718
719
720
721
722
            logs = {"loss": loss.detach().item(), "lr": lr_scheduler.get_last_lr()[0]}
            progress_bar.set_postfix(**logs)
            accelerator.log(logs, step=global_step)

            if global_step >= args.max_train_steps:
                break

        accelerator.wait_for_everyone()

    # Create the pipeline using using the trained modules and save it.
    if accelerator.is_main_process:
723
        pipeline = DiffusionPipeline.from_pretrained(
724
725
726
            args.pretrained_model_name_or_path,
            unet=accelerator.unwrap_model(unet),
            text_encoder=accelerator.unwrap_model(text_encoder),
727
            revision=args.revision,
728
729
730
731
        )
        pipeline.save_pretrained(args.output_dir)

        if args.push_to_hub:
732
            repo.push_to_hub(commit_message="End of training", blocking=False, auto_lfs_prune=True)
733
734
735
736
737

    accelerator.end_training()


if __name__ == "__main__":
738
739
    args = parse_args()
    main(args)