File size: 23,901 Bytes
510b9cd
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
import argparse
import itertools
import math
import os
import random
from pathlib import Path
from typing import Optional

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

import PIL
from accelerate import Accelerator
from accelerate.logging import get_logger
from accelerate.utils import set_seed
from diffusers import AutoencoderKL, DDPMScheduler, PNDMScheduler, StableDiffusionPipeline, UNet2DConditionModel
from diffusers.optimization import get_scheduler
from diffusers.pipelines.stable_diffusion import StableDiffusionSafetyChecker
from huggingface_hub import HfFolder, Repository, whoami
from PIL import Image
from torchvision import transforms
from tqdm.auto import tqdm
from transformers import CLIPFeatureExtractor, CLIPTextModel, CLIPTokenizer
import gc

logger = get_logger(__name__)


def save_progress(text_encoder, placeholder_token_id, accelerator, args):
    logger.info("Saving embeddings")
    learned_embeds = accelerator.unwrap_model(text_encoder).get_input_embeddings().weight[placeholder_token_id]
    learned_embeds_dict = {args.placeholder_token: learned_embeds.detach().cpu()}
    torch.save(learned_embeds_dict, os.path.join(args.output_dir, "learned_embeds.bin"))


def parse_args():
    parser = argparse.ArgumentParser(description="Simple example of a training script.")
    parser.add_argument(
        "--save_steps",
        type=int,
        default=500,
        help="Save learned_embeds.bin every X updates steps.",
    )
    parser.add_argument(
        "--pretrained_model_name_or_path",
        type=str,
        default=None,
        help="Path to pretrained model or model identifier from huggingface.co/models.",
    )
    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(
        "--train_data_dir", type=str, default=None, help="A folder containing the training data."
    )
    parser.add_argument(
        "--placeholder_token",
        type=str,
        default=None,
        help="A token to use as a placeholder for the concept.",
    )
    parser.add_argument(
        "--initializer_token", type=str, default=None, help="A token to use as initializer word."
    )
    parser.add_argument("--learnable_property", type=str, default="object", help="Choose between 'object' and 'style'")
    parser.add_argument("--repeats", type=int, default=100, help="How many times to repeat the training data.")
    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"
    )
    parser.add_argument(
        "--train_batch_size", type=int, default=16, help="Batch size (per device) for the training dataloader."
    )
    parser.add_argument("--num_train_epochs", type=int, default=100)
    parser.add_argument(
        "--max_train_steps",
        type=int,
        default=5000,
        help="Total number of training steps to perform.  If provided, overrides num_train_epochs.",
    )
    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(
        "--learning_rate",
        type=float,
        default=1e-4,
        help="Initial learning rate (after the potential warmup period) to use.",
    )
    parser.add_argument(
        "--scale_lr",
        action="store_true",
        default=True,
        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("--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("--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,
        default="no",
        choices=["no", "fp16", "bf16"],
        help=(
            "Whether to use mixed precision. Choose"
            "between fp16 and bf16 (bfloat16). Bf16 requires PyTorch >= 1.10."
            "and an Nvidia Ampere GPU."
        ),
    )
    parser.add_argument("--local_rank", type=int, default=-1, help="For distributed training: local_rank")

    args = parser.parse_args()
    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.train_data_dir is None:
        raise ValueError("You must specify a train data directory.")
    '''

    return args


imagenet_templates_small = [
    "a photo of a {}",
    "a rendering of a {}",
    "a cropped photo of the {}",
    "the photo of a {}",
    "a photo of a clean {}",
    "a photo of a dirty {}",
    "a dark photo of the {}",
    "a photo of my {}",
    "a photo of the cool {}",
    "a close-up photo of a {}",
    "a bright photo of the {}",
    "a cropped photo of a {}",
    "a photo of the {}",
    "a good photo of the {}",
    "a photo of one {}",
    "a close-up photo of the {}",
    "a rendition of the {}",
    "a photo of the clean {}",
    "a rendition of a {}",
    "a photo of a nice {}",
    "a good photo of a {}",
    "a photo of the nice {}",
    "a photo of the small {}",
    "a photo of the weird {}",
    "a photo of the large {}",
    "a photo of a cool {}",
    "a photo of a small {}",
]

imagenet_style_templates_small = [
    "a painting in the style of {}",
    "a rendering in the style of {}",
    "a cropped painting in the style of {}",
    "the painting in the style of {}",
    "a clean painting in the style of {}",
    "a dirty painting in the style of {}",
    "a dark painting in the style of {}",
    "a picture in the style of {}",
    "a cool painting in the style of {}",
    "a close-up painting in the style of {}",
    "a bright painting in the style of {}",
    "a cropped painting in the style of {}",
    "a good painting in the style of {}",
    "a close-up painting in the style of {}",
    "a rendition in the style of {}",
    "a nice painting in the style of {}",
    "a small painting in the style of {}",
    "a weird painting in the style of {}",
    "a large painting in the style of {}",
]


class TextualInversionDataset(Dataset):
    def __init__(
        self,
        data_root,
        tokenizer,
        learnable_property="object",  # [object, style]
        size=512,
        repeats=100,
        interpolation="bicubic",
        flip_p=0.5,
        set="train",
        placeholder_token="*",
        center_crop=False,
    ):
        self.data_root = data_root
        self.tokenizer = tokenizer
        self.learnable_property = learnable_property
        self.size = size
        self.placeholder_token = placeholder_token
        self.center_crop = center_crop
        self.flip_p = flip_p

        self.image_paths = [os.path.join(self.data_root, file_path) for file_path in os.listdir(self.data_root)]

        self.num_images = len(self.image_paths)
        self._length = self.num_images

        if set == "train":
            self._length = self.num_images * repeats

        self.interpolation = {
            "linear": PIL.Image.LINEAR,
            "bilinear": PIL.Image.BILINEAR,
            "bicubic": PIL.Image.BICUBIC,
            "lanczos": PIL.Image.LANCZOS,
        }[interpolation]

        self.templates = imagenet_style_templates_small if learnable_property == "style" else imagenet_templates_small
        self.flip_transform = transforms.RandomHorizontalFlip(p=self.flip_p)

    def __len__(self):
        return self._length

    def __getitem__(self, i):
        example = {}
        image = Image.open(self.image_paths[i % self.num_images])

        if not image.mode == "RGB":
            image = image.convert("RGB")

        placeholder_string = self.placeholder_token
        text = random.choice(self.templates).format(placeholder_string)

        example["input_ids"] = self.tokenizer(
            text,
            padding="max_length",
            truncation=True,
            max_length=self.tokenizer.model_max_length,
            return_tensors="pt",
        ).input_ids[0]

        # default to score-sde preprocessing
        img = np.array(image).astype(np.uint8)

        if self.center_crop:
            crop = min(img.shape[0], img.shape[1])
            h, w, = (
                img.shape[0],
                img.shape[1],
            )
            img = img[(h - crop) // 2 : (h + crop) // 2, (w - crop) // 2 : (w + crop) // 2]

        image = Image.fromarray(img)
        image = image.resize((self.size, self.size), resample=self.interpolation)

        image = self.flip_transform(image)
        image = np.array(image).astype(np.uint8)
        image = (image / 127.5 - 1.0).astype(np.float32)

        example["pixel_values"] = torch.from_numpy(image).permute(2, 0, 1)
        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}"


def freeze_params(params):
    for param in params:
        param.requires_grad = False


def merge_two_dicts(starting_dict: dict, updater_dict: dict) -> dict:
    """
    Starts from base starting dict and then adds the remaining key values from updater replacing the values from
    the first starting/base dict with the second updater dict.

    For later: how does d = {**d1, **d2} replace collision?

    :param starting_dict:
    :param updater_dict:
    :return:
    """
    new_dict: dict = starting_dict.copy()   # start with keys and values of starting_dict
    new_dict.update(updater_dict)    # modifies starting_dict with keys and values of updater_dict
    return new_dict

def merge_args(args1: argparse.Namespace, args2: argparse.Namespace) -> argparse.Namespace:
    """

    ref: https://stackoverflow.com/questions/56136549/how-can-i-merge-two-argparse-namespaces-in-python-2-x
    :param args1:
    :param args2:
    :return:
    """
    # - the merged args
    # The vars() function returns the __dict__ attribute to values of the given object e.g {field:value}.
    merged_key_values_for_namespace: dict = merge_two_dicts(vars(args1), vars(args2))
    args = argparse.Namespace(**merged_key_values_for_namespace)
    return args

def run_training(args_imported):
    args_default = parse_args()
    args = merge_args(args_default, args_imported)

    print(args)
    logging_dir = os.path.join(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,
    )

    # If passed along, set the training seed now.
    if args.seed is not None:
        set_seed(args.seed)

    # 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 and add the placeholder token as a additional special token
    if args.tokenizer_name:
        tokenizer = CLIPTokenizer.from_pretrained(args.tokenizer_name)
    elif args.pretrained_model_name_or_path:
        tokenizer = CLIPTokenizer.from_pretrained(args.pretrained_model_name_or_path, subfolder="tokenizer")

    # Add the placeholder token in tokenizer
    num_added_tokens = tokenizer.add_tokens(args.placeholder_token)
    if num_added_tokens == 0:
        raise ValueError(
            f"The tokenizer already contains the token {args.placeholder_token}. Please pass a different"
            " `placeholder_token` that is not already in the tokenizer."
        )

    # Convert the initializer_token, placeholder_token to ids
    token_ids = tokenizer.encode(args.initializer_token, add_special_tokens=False)
    # Check if initializer_token is a single token or a sequence of tokens
    if len(token_ids) > 1:
        raise ValueError("The initializer token must be a single token.")

    initializer_token_id = token_ids[0]
    placeholder_token_id = tokenizer.convert_tokens_to_ids(args.placeholder_token)

    # Load models and create wrapper for stable diffusion
    text_encoder = CLIPTextModel.from_pretrained(args.pretrained_model_name_or_path, subfolder="text_encoder")
    vae = AutoencoderKL.from_pretrained(args.pretrained_model_name_or_path, subfolder="vae")
    unet = UNet2DConditionModel.from_pretrained(args.pretrained_model_name_or_path, subfolder="unet")

    # Resize the token embeddings as we are adding new special tokens to the tokenizer
    text_encoder.resize_token_embeddings(len(tokenizer))

    # Initialise the newly added placeholder token with the embeddings of the initializer token
    token_embeds = text_encoder.get_input_embeddings().weight.data
    token_embeds[placeholder_token_id] = token_embeds[initializer_token_id]

    # Freeze vae and unet
    freeze_params(vae.parameters())
    freeze_params(unet.parameters())
    # Freeze all parameters except for the token embeddings in text encoder
    params_to_freeze = itertools.chain(
        text_encoder.text_model.encoder.parameters(),
        text_encoder.text_model.final_layer_norm.parameters(),
        text_encoder.text_model.embeddings.position_embedding.parameters(),
    )
    freeze_params(params_to_freeze)

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

    # Initialize the optimizer
    optimizer = torch.optim.AdamW(
        text_encoder.get_input_embeddings().parameters(),  # only optimize the embeddings
        lr=args.learning_rate,
        betas=(args.adam_beta1, args.adam_beta2),
        weight_decay=args.adam_weight_decay,
        eps=args.adam_epsilon,
    )

    # TODO (patil-suraj): load scheduler using args
    noise_scheduler = DDPMScheduler(
        beta_start=0.00085,
        beta_end=0.012,
        beta_schedule="scaled_linear",
        num_train_timesteps=1000,
    )

    train_dataset = TextualInversionDataset(
        data_root=args.train_data_dir,
        tokenizer=tokenizer,
        size=args.resolution,
        placeholder_token=args.placeholder_token,
        repeats=args.repeats,
        learnable_property=args.learnable_property,
        center_crop=args.center_crop,
        set="train",
    )
    train_dataloader = torch.utils.data.DataLoader(train_dataset, batch_size=args.train_batch_size, shuffle=True)

    # 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,
    )

    text_encoder, optimizer, train_dataloader, lr_scheduler = accelerator.prepare(
        text_encoder, optimizer, train_dataloader, lr_scheduler
    )

    # Move vae and unet to device
    vae.to(accelerator.device)
    unet.to(accelerator.device)

    # Keep vae and unet in eval model as we don't train these
    vae.eval()
    unet.eval()

    # 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("textual_inversion", 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 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):
        text_encoder.train()
        for step, batch in enumerate(train_dataloader):
            with accelerator.accumulate(text_encoder):
                # Convert images to latent space
                latents = vae.encode(batch["pixel_values"]).latent_dist.sample().detach()
                latents = latents * 0.18215

                # Sample noise that we'll add to the latents
                noise = torch.randn(latents.shape).to(latents.device)
                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
                ).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
                encoder_hidden_states = text_encoder(batch["input_ids"])[0]

                # Predict the noise residual
                noise_pred = unet(noisy_latents, timesteps, encoder_hidden_states).sample

                loss = F.mse_loss(noise_pred, noise, reduction="none").mean([1, 2, 3]).mean()
                accelerator.backward(loss)

                # Zero out the gradients for all token embeddings except the newly added
                # embeddings for the concept, as we only want to optimize the concept embeddings
                if accelerator.num_processes > 1:
                    grads = text_encoder.module.get_input_embeddings().weight.grad
                else:
                    grads = text_encoder.get_input_embeddings().weight.grad
                # Get the index for tokens that we want to zero the grads for
                index_grads_to_zero = torch.arange(len(tokenizer)) != placeholder_token_id
                grads.data[index_grads_to_zero, :] = grads.data[index_grads_to_zero, :].fill_(0)

                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
                if global_step % args.save_steps == 0:
                    save_progress(text_encoder, placeholder_token_id, accelerator, args)

            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:
        pipeline = StableDiffusionPipeline(
            text_encoder=accelerator.unwrap_model(text_encoder),
            vae=vae,
            unet=unet,
            tokenizer=tokenizer,
            scheduler=PNDMScheduler(
                beta_start=0.00085, beta_end=0.012, beta_schedule="scaled_linear", skip_prk_steps=True
            ),
            safety_checker=StableDiffusionSafetyChecker.from_pretrained("CompVis/stable-diffusion-safety-checker"),
            feature_extractor=CLIPFeatureExtractor.from_pretrained("openai/clip-vit-base-patch32"),
        )
        pipeline.save_pretrained(args.output_dir)
        # Also save the newly trained embeddings
        save_progress(text_encoder, placeholder_token_id, accelerator, args)

        if args.push_to_hub:
            repo.push_to_hub(commit_message="End of training", blocking=False, auto_lfs_prune=True)

    accelerator.end_training()
    torch.cuda.empty_cache()
    gc.collect()


if __name__ == "__main__":
    main()