stable-diffusion-inpainting, 4/5 breast size, updated code (2000 steps)

args.json

@@ -7,12 +7,10 @@
   "class_data_dir": null,
   "instance_prompt": null,
   "class_prompt": null,
-  "save_sample_prompt": "close up of woman wearing bikini top, amfdk breast size|close up of woman wearing bikini top, boqnf breast size|close up of woman wearing bikini top, czufm breast size|close up of woman wearing bikini top, dpqjd breast size",
   "save_sample_negative_prompt": null,
   "n_save_sample": 4,
   "save_guidance_scale": 7.5,
   "save_infer_steps": 50,
-  "pad_tokens": true,
   "with_prior_preservation": true,
   "prior_loss_weight": 1.0,
   "num_class_images": 300,
@@ -46,6 +44,7 @@
   "save_min_steps": 0,
   "mixed_precision": "fp16",
   "not_cache_latents": true,
+  "hflip": false,
   "local_rank": -1,
   "concepts_list": [
     {

text_encoder/pytorch_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:18e4ec9289fee0ea712fa0be8a4cc42b7a0e697681033246b11d8c181a544146
+oid sha256:1791b109357aaec3009ab57c590775556825c134f27a143c235945c3fdc635d2
 size 492308087

train_inpainting_dreambooth.py

@@ -0,0 +1,866 @@
+import argparse
+import hashlib
+import itertools
+import json
+import math
+import os
+import random
+import shutil
+from contextlib import nullcontext
+from pathlib import Path
+from typing import Optional
+
+import torch
+import torch.nn.functional as F
+import torch.utils.checkpoint
+from accelerate import Accelerator
+from accelerate.logging import get_logger
+from accelerate.utils import set_seed
+from huggingface_hub import HfFolder, Repository, whoami
+from PIL import Image
+from torch.utils.data import Dataset
+from torchvision import transforms
+from tqdm.auto import tqdm
+from transformers import CLIPTextModel, CLIPTokenizer
+
+from diffusers import (AutoencoderKL, DDIMScheduler, DDPMScheduler,
+                       StableDiffusionInpaintPipeline, UNet2DConditionModel)
+from diffusers.optimization import get_scheduler
+
+torch.backends.cudnn.benchmark = True
+
+
+logger = get_logger(__name__)
+
+
+def parse_args(input_args=None):
+    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.",
+    )
+    parser.add_argument(
+        "--pretrained_vae_name_or_path",
+        type=str,
+        default=None,
+        help="Path to pretrained vae or vae identifier from huggingface.co/models.",
+    )
+    parser.add_argument(
+        "--revision",
+        type=str,
+        default="fp16",
+        required=False,
+        help="Revision of pretrained 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(
+        "--instance_data_dir",
+        type=str,
+        default=None,
+        help="A folder containing the training data of instance images.",
+    )
+    parser.add_argument(
+        "--class_data_dir",
+        type=str,
+        default=None,
+        help="A folder containing the training data of class images.",
+    )
+    parser.add_argument(
+        "--instance_prompt",
+        type=str,
+        default=None,
+        help="The prompt with identifier specifying the instance",
+    )
+    parser.add_argument(
+        "--class_prompt",
+        type=str,
+        default=None,
+        help="The prompt to specify images in the same class as provided instance images.",
+    )
+    # parser.add_argument(
+    #     "--save_sample_prompt",
+    #     type=str,
+    #     default=None,
+    #     help="The prompt used to generate sample outputs to save.",
+    # )
+    parser.add_argument(
+        "--save_sample_negative_prompt",
+        type=str,
+        default=None,
+        help="The negative prompt used to generate sample outputs to save.",
+    )
+    parser.add_argument(
+        "--n_save_sample",
+        type=int,
+        default=4,
+        help="The number of samples to save.",
+    )
+    parser.add_argument(
+        "--save_guidance_scale",
+        type=float,
+        default=7.5,
+        help="CFG for save sample.",
+    )
+    parser.add_argument(
+        "--save_infer_steps",
+        type=int,
+        default=50,
+        help="The number of inference steps for save sample.",
+    )
+    parser.add_argument(
+        "--with_prior_preservation",
+        default=False,
+        action="store_true",
+        help="Flag to add prior preservation loss.",
+    )
+    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=(
+            "Minimal class images for prior preservation loss. If not have enough images, additional images will be"
+            " sampled with class_prompt."
+        ),
+    )
+    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_text_encoder", action="store_true", help="Whether to train the text encoder")
+    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.",
+    )
+    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("--log_interval", type=int, default=10, help="Log every N steps.")
+    parser.add_argument("--save_interval", type=int, default=10_000, help="Save weights every N steps.")
+    parser.add_argument("--save_min_steps", type=int, default=0, help="Start saving weights after N steps.")
+    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("--not_cache_latents", action="store_true", help="Do not precompute and cache latents from VAE.")
+    parser.add_argument("--hflip", action="store_true", help="Apply horizontal flip data augmentation.")
+    parser.add_argument("--local_rank", type=int, default=-1, help="For distributed training: local_rank")
+    parser.add_argument(
+        "--concepts_list",
+        type=str,
+        default=None,
+        help="Path to json containing multiple concepts, will overwrite parameters like instance_prompt, class_prompt, etc.",
+    )
+
+    if input_args is not None:
+        args = parser.parse_args(input_args)
+    else:
+        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
+
+    return args
+
+
+def get_cutout_holes(height, width, min_holes=8, max_holes=32, min_height=32, max_height=128, min_width=32, max_width=128):
+    holes = []
+    for _n in range(random.randint(min_holes, max_holes)):
+        hole_height = random.randint(min_height, max_height)
+        hole_width = random.randint(min_width, max_width)
+        y1 = random.randint(0, height - hole_height)
+        x1 = random.randint(0, width - hole_width)
+        y2 = y1 + hole_height
+        x2 = x1 + hole_width
+        holes.append((x1, y1, x2, y2))
+    return holes
+
+
+def generate_random_mask(image):
+    mask = torch.zeros_like(image[:1])
+    holes = get_cutout_holes(mask.shape[1], mask.shape[2])
+    for (x1, y1, x2, y2) in holes:
+        mask[:, y1:y2, x1:x2] = 1.
+    if random.uniform(0, 1) < 0.25:
+        mask.fill_(1.)
+    masked_image = image * (mask < 0.5)
+    return mask, masked_image
+
+
+class DreamBoothDataset(Dataset):
+    """
+    A dataset to prepare the instance and class images with the prompts for fine-tuning the model.
+    It pre-processes the images and the tokenizes prompts.
+    """
+
+    def __init__(
+        self,
+        concepts_list,
+        tokenizer,
+        with_prior_preservation=True,
+        size=512,
+        center_crop=False,
+        num_class_images=None,
+        hflip=False
+    ):
+        self.size = size
+        self.center_crop = center_crop
+        self.tokenizer = tokenizer
+        self.with_prior_preservation = with_prior_preservation
+        self.instance_images_path = []
+        self.class_images_path = []
+
+        for concept in concepts_list:
+            inst_img_path = [(x, concept["instance_prompt"]) for x in Path(concept["instance_data_dir"]).iterdir() if x.is_file()]
+            self.instance_images_path.extend(inst_img_path)
+
+            if with_prior_preservation:
+                class_img_path = [(x, concept["class_prompt"]) for x in Path(concept["class_data_dir"]).iterdir() if x.is_file()]
+                self.class_images_path.extend(class_img_path[:num_class_images])
+
+        random.shuffle(self.instance_images_path)
+        self.num_instance_images = len(self.instance_images_path)
+        self.num_class_images = len(self.class_images_path)
+        self._length = max(self.num_class_images, self.num_instance_images)
+
+        self.image_transforms = transforms.Compose(
+            [
+                transforms.RandomHorizontalFlip(0.5 * hflip),
+                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_path, instance_prompt = self.instance_images_path[index % self.num_instance_images]
+        instance_image = Image.open(instance_path)
+        if not instance_image.mode == "RGB":
+            instance_image = instance_image.convert("RGB")
+        example["instance_images"] = self.image_transforms(instance_image)
+        example["instance_masks"], example["instance_masked_images"] = generate_random_mask(example["instance_images"])
+        example["instance_prompt_ids"] = self.tokenizer(
+            instance_prompt,
+            padding="max_length",
+            truncation=True,
+            max_length=self.tokenizer.model_max_length,
+        ).input_ids
+
+        if self.with_prior_preservation:
+            class_path, class_prompt = self.class_images_path[index % self.num_class_images]
+            class_image = Image.open(class_path)
+            if not class_image.mode == "RGB":
+                class_image = class_image.convert("RGB")
+            example["class_images"] = self.image_transforms(class_image)
+            example["class_masks"], example["class_masked_images"] = generate_random_mask(example["class_images"])
+            example["class_prompt_ids"] = self.tokenizer(
+                class_prompt,
+                padding="max_length",
+                truncation=True,
+                max_length=self.tokenizer.model_max_length,
+            ).input_ids
+
+        return example
+
+
+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
+
+
+class LatentsDataset(Dataset):
+    def __init__(self, latents_cache, text_encoder_cache):
+        self.latents_cache = latents_cache
+        self.text_encoder_cache = text_encoder_cache
+
+    def __len__(self):
+        return len(self.latents_cache)
+
+    def __getitem__(self, index):
+        return self.latents_cache[index], self.text_encoder_cache[index]
+
+
+class AverageMeter:
+    def __init__(self, name=None):
+        self.name = name
+        self.reset()
+
+    def reset(self):
+        self.sum = self.count = self.avg = 0
+
+    def update(self, val, n=1):
+        self.sum += val * n
+        self.count += n
+        self.avg = self.sum / self.count
+
+
+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 main(args):
+    logging_dir = Path(args.output_dir, "0", args.logging_dir)
+
+    accelerator = Accelerator(
+        gradient_accumulation_steps=args.gradient_accumulation_steps,
+        mixed_precision=args.mixed_precision,
+        log_with="tensorboard",
+        logging_dir=logging_dir,
+    )
+
+    # 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."
+        )
+
+    if args.seed is not None:
+        set_seed(args.seed)
+
+    if args.concepts_list is None:
+        args.concepts_list = [
+            {
+                "instance_prompt": args.instance_prompt,
+                "class_prompt": args.class_prompt,
+                "instance_data_dir": args.instance_data_dir,
+                "class_data_dir": args.class_data_dir
+            }
+        ]
+    else:
+        with open(args.concepts_list, "r") as f:
+            args.concepts_list = json.load(f)
+
+    if args.with_prior_preservation:
+        pipeline = None
+        for concept in args.concepts_list:
+            class_images_dir = Path(concept["class_data_dir"])
+            class_images_dir.mkdir(parents=True, exist_ok=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
+                if pipeline is None:
+                    pipeline = StableDiffusionInpaintPipeline.from_pretrained(
+                        args.pretrained_model_name_or_path,
+                        vae=AutoencoderKL.from_pretrained(
+                            args.pretrained_vae_name_or_path or args.pretrained_model_name_or_path,
+                            revision=None if args.pretrained_vae_name_or_path else args.revision
+                        ),
+                        torch_dtype=torch_dtype,
+                        safety_checker=None,
+                        revision=args.revision
+                    )
+                    pipeline.set_progress_bar_config(disable=True)
+                    pipeline.to(accelerator.device)
+
+                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(concept["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)
+
+                inp_img = Image.new("RGB", (512, 512), color=(0, 0, 0))
+                inp_mask = Image.new("L", (512, 512), color=255)
+
+                with torch.autocast("cuda"), torch.inference_mode():
+                    for example in tqdm(
+                        sample_dataloader, desc="Generating class images", disable=not accelerator.is_local_main_process
+                    ):
+                        images = pipeline(
+                            prompt=example["prompt"],
+                            image=inp_img,
+                            mask_image=inp_mask,
+                            num_inference_steps=args.save_infer_steps
+                        ).images
+
+                        for i, image in enumerate(images):
+                            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)
+
+        del pipeline
+        if torch.cuda.is_available():
+            torch.cuda.empty_cache()
+
+    # Load the tokenizer
+    if args.tokenizer_name:
+        tokenizer = CLIPTokenizer.from_pretrained(
+            args.tokenizer_name,
+            revision=args.revision,
+        )
+    elif args.pretrained_model_name_or_path:
+        tokenizer = CLIPTokenizer.from_pretrained(
+            args.pretrained_model_name_or_path,
+            subfolder="tokenizer",
+            revision=args.revision,
+        )
+
+    # Load models and create wrapper for stable diffusion
+    text_encoder = CLIPTextModel.from_pretrained(
+        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,
+        torch_dtype=torch.float32
+    )
+
+    vae.requires_grad_(False)
+    if not args.train_text_encoder:
+        text_encoder.requires_grad_(False)
+
+    if args.gradient_checkpointing:
+        unet.enable_gradient_checkpointing()
+        if args.train_text_encoder:
+            text_encoder.gradient_checkpointing_enable()
+
+    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
+
+    params_to_optimize = (
+        itertools.chain(unet.parameters(), text_encoder.parameters()) if args.train_text_encoder else unet.parameters()
+    )
+    optimizer = optimizer_class(
+        params_to_optimize,
+        lr=args.learning_rate,
+        betas=(args.adam_beta1, args.adam_beta2),
+        weight_decay=args.adam_weight_decay,
+        eps=args.adam_epsilon,
+    )
+
+    noise_scheduler = DDPMScheduler.from_config(args.pretrained_model_name_or_path, subfolder="scheduler")
+
+    train_dataset = DreamBoothDataset(
+        concepts_list=args.concepts_list,
+        tokenizer=tokenizer,
+        with_prior_preservation=args.with_prior_preservation,
+        size=args.resolution,
+        center_crop=args.center_crop,
+        num_class_images=args.num_class_images,
+        hflip=args.hflip
+    )
+
+    def collate_fn(examples):
+        input_ids = [example["instance_prompt_ids"] for example in examples]
+        pixel_values = [example["instance_images"] for example in examples]
+        mask_values = [example["instance_masks"] for example in examples]
+        masked_image_values = [example["instance_masked_images"] for example in examples]
+
+        # Concat class and instance examples for prior preservation.
+        # We do this to avoid doing two forward passes.
+        if args.with_prior_preservation:
+            input_ids += [example["class_prompt_ids"] for example in examples]
+            pixel_values += [example["class_images"] for example in examples]
+            mask_values += [example["class_masks"] for example in examples]
+            masked_image_values += [example["class_masked_images"] for example in examples]
+
+        pixel_values = torch.stack(pixel_values).to(memory_format=torch.contiguous_format).float()
+        mask_values = torch.stack(mask_values).to(memory_format=torch.contiguous_format).float()
+        masked_image_values = torch.stack(masked_image_values).to(memory_format=torch.contiguous_format).float()
+
+        input_ids = tokenizer.pad(
+            {"input_ids": input_ids},
+            padding="max_length",
+            max_length=tokenizer.model_max_length,
+            return_tensors="pt",
+        ).input_ids
+
+        batch = {
+            "input_ids": input_ids,
+            "pixel_values": pixel_values,
+            "mask_values": mask_values,
+            "masked_image_values": masked_image_values
+        }
+        return batch
+
+    train_dataloader = torch.utils.data.DataLoader(
+        train_dataset, batch_size=args.train_batch_size, shuffle=True, collate_fn=collate_fn, pin_memory=True, num_workers=8
+    )
+
+    weight_dtype = torch.float32
+    if args.mixed_precision == "fp16":
+        weight_dtype = torch.float16
+    elif args.mixed_precision == "bf16":
+        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)
+    if not args.train_text_encoder:
+        text_encoder.to(accelerator.device, dtype=weight_dtype)
+
+    if not args.not_cache_latents:
+        latents_cache = []
+        text_encoder_cache = []
+        for batch in tqdm(train_dataloader, desc="Caching latents"):
+            with torch.no_grad():
+                batch["pixel_values"] = batch["pixel_values"].to(accelerator.device, non_blocking=True, dtype=weight_dtype)
+                batch["input_ids"] = batch["input_ids"].to(accelerator.device, non_blocking=True)
+                latents_cache.append(vae.encode(batch["pixel_values"]).latent_dist)
+                if args.train_text_encoder:
+                    text_encoder_cache.append(batch["input_ids"])
+                else:
+                    text_encoder_cache.append(text_encoder(batch["input_ids"])[0])
+        train_dataset = LatentsDataset(latents_cache, text_encoder_cache)
+        train_dataloader = torch.utils.data.DataLoader(train_dataset, batch_size=1, collate_fn=lambda x: x, shuffle=True)
+
+        del vae
+        if not args.train_text_encoder:
+            del text_encoder
+        if torch.cuda.is_available():
+            torch.cuda.empty_cache()
+
+    # 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,
+    )
+
+    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
+        )
+
+    # 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")
+
+    # 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}")
+
+    def save_weights(step):
+        # Create the pipeline using using the trained modules and save it.
+        if accelerator.is_main_process:
+            if args.train_text_encoder:
+                text_enc_model = accelerator.unwrap_model(text_encoder)
+            else:
+                text_enc_model = CLIPTextModel.from_pretrained(args.pretrained_model_name_or_path, subfolder="text_encoder", revision=args.revision)
+            scheduler = DDIMScheduler(beta_start=0.00085, beta_end=0.012, beta_schedule="scaled_linear", clip_sample=False, set_alpha_to_one=False)
+            pipeline = StableDiffusionInpaintPipeline.from_pretrained(
+                args.pretrained_model_name_or_path,
+                unet=accelerator.unwrap_model(unet),
+                text_encoder=text_enc_model,
+                vae=AutoencoderKL.from_pretrained(
+                    args.pretrained_vae_name_or_path or args.pretrained_model_name_or_path,
+                    subfolder=None if args.pretrained_vae_name_or_path else "vae",
+                    revision=None if args.pretrained_vae_name_or_path else args.revision
+                ),
+                safety_checker=None,
+                scheduler=scheduler,
+                torch_dtype=torch.float16,
+                revision=args.revision,
+            )
+            save_dir = os.path.join(args.output_dir, f"{step}")
+            pipeline.save_pretrained(save_dir)
+            with open(os.path.join(save_dir, "args.json"), "w") as f:
+                json.dump(args.__dict__, f, indent=2)
+
+            shutil.copy("train_inpainting_dreambooth.py", save_dir)
+
+            pipeline = pipeline.to(accelerator.device)
+            pipeline.set_progress_bar_config(disable=True)
+            for idx, concept in enumerate(args.concepts_list):
+                g_cuda = torch.Generator(device=accelerator.device).manual_seed(args.seed)
+                sample_dir = os.path.join(save_dir, "samples", str(idx))
+                os.makedirs(sample_dir, exist_ok=True)
+                inp_img = Image.new("RGB", (512, 512), color=(0, 0, 0))
+                inp_mask = Image.new("L", (512, 512), color=255)
+                with torch.autocast("cuda"), torch.inference_mode():
+                    for i in tqdm(range(args.n_save_sample), desc="Generating samples"):
+                        images = pipeline(
+                            prompt=concept["instance_prompt"],
+                            image=inp_img,
+                            mask_image=inp_mask,
+                            negative_prompt=args.save_sample_negative_prompt,
+                            guidance_scale=args.save_guidance_scale,
+                            num_inference_steps=args.save_infer_steps,
+                            generator=g_cuda
+                        ).images
+                        images[0].save(os.path.join(sample_dir, f"{i}.png"))
+            del pipeline
+            if torch.cuda.is_available():
+                torch.cuda.empty_cache()
+            print(f"[*] Weights saved at {save_dir}")
+
+    # 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
+    loss_avg = AverageMeter()
+    text_enc_context = nullcontext() if args.train_text_encoder else torch.no_grad()
+    for epoch in range(args.num_train_epochs):
+        unet.train()
+        if args.train_text_encoder:
+            text_encoder.train()
+        random.shuffle(train_dataset.class_images_path)
+        for step, batch in enumerate(train_dataloader):
+            with accelerator.accumulate(unet):
+                # Convert images to latent space
+                with torch.no_grad():
+                    if not args.not_cache_latents:
+                        latent_dist = batch[0][0]
+                    else:
+                        latent_dist = vae.encode(batch["pixel_values"].to(dtype=weight_dtype)).latent_dist
+                        masked_latent_dist = vae.encode(batch["masked_image_values"].to(dtype=weight_dtype)).latent_dist
+                    latents = latent_dist.sample() * 0.18215
+                    masked_image_latents = masked_latent_dist.sample() * 0.18215
+                    mask = F.interpolate(batch["mask_values"], scale_factor=1 / 8)
+
+                # Sample noise that we'll add to the latents
+                noise = torch.randn_like(latents)
+                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
+                with text_enc_context:
+                    if not args.not_cache_latents:
+                        if args.train_text_encoder:
+                            encoder_hidden_states = text_encoder(batch[0][1])[0]
+                        else:
+                            encoder_hidden_states = batch[0][1]
+                    else:
+                        encoder_hidden_states = text_encoder(batch["input_ids"])[0]
+
+                encoder_hidden_states = F.dropout(encoder_hidden_states, p=0.1)
+
+                latent_model_input = torch.cat([noisy_latents, mask, masked_image_latents], dim=1)
+                # Predict the noise residual
+                noise_pred = unet(latent_model_input, timesteps, encoder_hidden_states).sample
+
+                if args.with_prior_preservation:
+                    # Chunk the noise and noise_pred into two parts and compute the loss on each part separately.
+                    noise_pred, noise_pred_prior = torch.chunk(noise_pred, 2, dim=0)
+                    noise, noise_prior = torch.chunk(noise, 2, dim=0)
+
+                    # Compute instance loss
+                    loss = F.mse_loss(noise_pred.float(), noise.float(), reduction="none").mean([1, 2, 3]).mean()
+
+                    # Compute prior loss
+                    prior_loss = F.mse_loss(noise_pred_prior.float(), noise_prior.float(), reduction="mean")
+
+                    # Add the prior loss to the instance loss.
+                    loss = loss + args.prior_loss_weight * prior_loss
+                else:
+                    loss = F.mse_loss(noise_pred.float(), noise.float(), reduction="mean")
+
+                accelerator.backward(loss)
+                # if accelerator.sync_gradients:
+                #     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)
+                optimizer.step()
+                lr_scheduler.step()
+                optimizer.zero_grad(set_to_none=True)
+                loss_avg.update(loss.detach_(), bsz)
+
+            if not global_step % args.log_interval:
+                logs = {"loss": loss_avg.avg.item(), "lr": lr_scheduler.get_last_lr()[0]}
+                progress_bar.set_postfix(**logs)
+                accelerator.log(logs, step=global_step)
+
+            if global_step > 0 and not global_step % args.save_interval and global_step >= args.save_min_steps:
+                save_weights(global_step)
+
+            progress_bar.update(1)
+            global_step += 1
+
+            if global_step >= args.max_train_steps:
+                break
+
+        accelerator.wait_for_everyone()
+
+    save_weights(global_step)
+
+    accelerator.end_training()
+
+
+if __name__ == "__main__":
+    args = parse_args()
+    main(args)

unet/diffusion_pytorch_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:42c5e779efd52a45b6e501b26277c17d8ea3b7948025a0202c350b690146e7a2
+oid sha256:e3619be9a04e9be8314bfa3ce5781a9a604ec19fe44ec8395863611a5885c62a
 size 3438421925