Research project multi subject dreambooth (#1948)

* implemented multi subject dreambooth in research_projects

* minor edits to readme

* added style and quality fixes
Co-authored-by: Krista Opsahl-Ong <kristaopsahlong@gmail.com>

examples/research_projects/multi_subject_dreambooth/README.md

+# Multi Subject DreamBooth training
+
+[DreamBooth](https://arxiv.org/abs/2208.12242) is a method to personalize text2image models like stable diffusion given just a few(3~5) images of a subject.
+This `train_multi_subject_dreambooth.py` script shows how to implement the training procedure for one or more subjects and adapt it for stable diffusion. Note that this code is based off of the `examples/dreambooth/train_dreambooth.py` script as of 01/06/2022.
+
+This script was added by @kopsahlong, and is not actively maintained. However, if you come across anything that could use fixing, feel free to open an issue and tag @kopsahlong.
+
+## Running locally with PyTorch
+### Installing the dependencies
+
+Before running the script, make sure to install the library's training dependencies:
+
+To start, execute the following steps in a new virtual environment:
+```bash
+git clone https://github.com/huggingface/diffusers
+cd diffusers
+pip install -e .
+```
+
+Then cd into the folder `diffusers/examples/research_projects/multi_subject_dreambooth` and run the following:
+```bash
+pip install -r requirements.txt
+```
+
+And initialize an [🤗Accelerate](https://github.com/huggingface/accelerate/) environment with:
+
+```bash
+accelerate config
+```
+
+Or for a default accelerate configuration without answering questions about your environment
+
+```bash
+accelerate config default
+```
+
+Or if your environment doesn't support an interactive shell e.g. a notebook
+
+```python
+from accelerate.utils import write_basic_config
+write_basic_config()
+```
+
+### Multi Subject Training Example
+In order to have your model learn multiple concepts at once, we simply add in the additional data directories and prompts to our `instance_data_dir` and `instance_prompt` (as well as `class_data_dir` and `class_prompt` if `--with_prior_preservation` is specified) as one comma separated string.
+
+See an example with 2 subjects below, which learns a model for one dog subject and one human subject:
+
+```bash
+export MODEL_NAME="CompVis/stable-diffusion-v1-4"
+export OUTPUT_DIR="path-to-save-model"
+
+# Subject 1
+export INSTANCE_DIR_1="path-to-instance-images-concept-1"
+export INSTANCE_PROMPT_1="a photo of a sks dog"
+export CLASS_DIR_1="path-to-class-images-dog"
+export CLASS_PROMPT_1="a photo of a dog"
+
+# Subject 2
+export INSTANCE_DIR_2="path-to-instance-images-concept-2"
+export INSTANCE_PROMPT_2="a photo of a t@y person"
+export CLASS_DIR_2="path-to-class-images-person"
+export CLASS_PROMPT_2="a photo of a person"
+
+accelerate launch train_multi_subject_dreambooth.py \
+  --pretrained_model_name_or_path=$MODEL_NAME  \
+  --instance_data_dir="$INSTANCE_DIR_1,$INSTANCE_DIR_2" \
+  --output_dir=$OUTPUT_DIR \
+  --train_text_encoder \
+  --instance_prompt="$INSTANCE_PROMPT_1,$INSTANCE_PROMPT_2" \
+  --with_prior_preservation \
+  --prior_loss_weight=1.0 \
+  --class_data_dir="$CLASS_DIR_1,$CLASS_DIR_2" \
+  --class_prompt="$CLASS_PROMPT_1,$CLASS_PROMPT_2"\
+  --num_class_images=50 \
+  --resolution=512 \
+  --train_batch_size=1 \
+  --gradient_accumulation_steps=1 \
+  --learning_rate=1e-6 \
+  --lr_scheduler="constant" \
+  --lr_warmup_steps=0 \
+  --max_train_steps=1500
+```
+
+This example shows training for 2 subjects, but please note that the model can be trained on any number of new concepts. This can be done by continuing to add in the corresponding directories and prompts to the corresponding comma separated string.
+
+Note also that in this script, `sks` and `t@y` were used as tokens to learn the new subjects ([this thread](https://github.com/XavierXiao/Dreambooth-Stable-Diffusion/issues/71) inspired the use of `t@y` as our second identifier). However, there may be better rare tokens to experiment with, and results also seemed to be good when more intuitive words are used.
+
+### Inference
+
+Once you have trained a model using above command, the inference can be done simply using the `StableDiffusionPipeline`. Make sure to include the `identifier`(e.g. sks in above example) in your prompt.
+
+```python
+from diffusers import StableDiffusionPipeline
+import torch
+
+model_id = "path-to-your-trained-model"
+pipe = StableDiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.float16).to("cuda")
+
+prompt = "A photo of a t@y person petting an sks dog"
+image = pipe(prompt, num_inference_steps=200, guidance_scale=7.5).images[0]
+
+image.save("person-petting-dog.png")
+```
+
+### Inference from a training checkpoint
+
+You can also perform inference from one of the checkpoints saved during the training process, if you used the `--checkpointing_steps` argument. Please, refer to [the documentation](https://huggingface.co/docs/diffusers/main/en/training/dreambooth#performing-inference-using-a-saved-checkpoint) to see how to do it.
+
+## Additional Dreambooth documentation
+Because the `train_multi_subject_dreambooth.py` script here was forked from an original version of `train_dreambooth.py` in the `examples/dreambooth` folder, I've included the original applicable training documentation for single subject examples below.
+
+This should explain how to play with training variables such as prior preservation, fine tuning the text encoder, etc. which is still applicable to our multi subject training code. Note also that the examples below, which are single subject examples, also work with `train_multi_subject_dreambooth.py`, as this script supports 1 (or more) subjects.
+
+### Single subject dog toy example
+
+Let's get our dataset. Download images from [here](https://drive.google.com/drive/folders/1BO_dyz-p65qhBRRMRA4TbZ8qW4rB99JZ) and save them in a directory. This will be our training data.
+
+And launch the training using
+
+**___Note: Change the `resolution` to 768 if you are using the [stable-diffusion-2](https://huggingface.co/stabilityai/stable-diffusion-2) 768x768 model.___**
+
+```bash
+export MODEL_NAME="CompVis/stable-diffusion-v1-4"
+export INSTANCE_DIR="path-to-instance-images"
+export OUTPUT_DIR="path-to-save-model"
+
+accelerate launch train_dreambooth.py \
+  --pretrained_model_name_or_path=$MODEL_NAME  \
+  --instance_data_dir=$INSTANCE_DIR \
+  --output_dir=$OUTPUT_DIR \
+  --instance_prompt="a photo of sks dog" \
+  --resolution=512 \
+  --train_batch_size=1 \
+  --gradient_accumulation_steps=1 \
+  --learning_rate=5e-6 \
+  --lr_scheduler="constant" \
+  --lr_warmup_steps=0 \
+  --max_train_steps=400
+```
+
+### Training with prior-preservation loss
+
+Prior-preservation is used to avoid overfitting and language-drift. Refer to the paper to learn more about it. For prior-preservation we first generate images using the model with a class prompt and then use those during training along with our data.
+According to the paper, it's recommended to generate `num_epochs * num_samples` images for prior-preservation. 200-300 works well for most cases. The `num_class_images` flag sets the number of images to generate with the class prompt. You can place existing images in `class_data_dir`, and the training script will generate any additional images so that `num_class_images` are present in `class_data_dir` during training time.
+
+```bash
+export MODEL_NAME="CompVis/stable-diffusion-v1-4"
+export INSTANCE_DIR="path-to-instance-images"
+export CLASS_DIR="path-to-class-images"
+export OUTPUT_DIR="path-to-save-model"
+
+accelerate launch train_dreambooth.py \
+  --pretrained_model_name_or_path=$MODEL_NAME  \
+  --instance_data_dir=$INSTANCE_DIR \
+  --class_data_dir=$CLASS_DIR \
+  --output_dir=$OUTPUT_DIR \
+  --with_prior_preservation --prior_loss_weight=1.0 \
+  --instance_prompt="a photo of sks dog" \
+  --class_prompt="a photo of dog" \
+  --resolution=512 \
+  --train_batch_size=1 \
+  --gradient_accumulation_steps=1 \
+  --learning_rate=5e-6 \
+  --lr_scheduler="constant" \
+  --lr_warmup_steps=0 \
+  --num_class_images=200 \
+  --max_train_steps=800
+```
+
+
+### Training on a 16GB GPU:
+
+With the help of gradient checkpointing and the 8-bit optimizer from bitsandbytes it's possible to run train dreambooth on a 16GB GPU.
+
+To install `bitandbytes` please refer to this [readme](https://github.com/TimDettmers/bitsandbytes#requirements--installation).
+
+```bash
+export MODEL_NAME="CompVis/stable-diffusion-v1-4"
+export INSTANCE_DIR="path-to-instance-images"
+export CLASS_DIR="path-to-class-images"
+export OUTPUT_DIR="path-to-save-model"
+
+accelerate launch train_dreambooth.py \
+  --pretrained_model_name_or_path=$MODEL_NAME  \
+  --instance_data_dir=$INSTANCE_DIR \
+  --class_data_dir=$CLASS_DIR \
+  --output_dir=$OUTPUT_DIR \
+  --with_prior_preservation --prior_loss_weight=1.0 \
+  --instance_prompt="a photo of sks dog" \
+  --class_prompt="a photo of dog" \
+  --resolution=512 \
+  --train_batch_size=1 \
+  --gradient_accumulation_steps=2 --gradient_checkpointing \
+  --use_8bit_adam \
+  --learning_rate=5e-6 \
+  --lr_scheduler="constant" \
+  --lr_warmup_steps=0 \
+  --num_class_images=200 \
+  --max_train_steps=800
+```
+
+### Training on a 8 GB GPU:
+
+By using [DeepSpeed](https://www.deepspeed.ai/) it's possible to offload some
+tensors from VRAM to either CPU or NVME allowing to train with less VRAM.
+
+DeepSpeed needs to be enabled with `accelerate config`. During configuration
+answer yes to "Do you want to use DeepSpeed?". With DeepSpeed stage 2, fp16
+mixed precision and offloading both parameters and optimizer state to cpu it's
+possible to train on under 8 GB VRAM with a drawback of requiring significantly
+more RAM (about 25 GB). See [documentation](https://huggingface.co/docs/accelerate/usage_guides/deepspeed) for more DeepSpeed configuration options.
+
+Changing the default Adam optimizer to DeepSpeed's special version of Adam
+`deepspeed.ops.adam.DeepSpeedCPUAdam` gives a substantial speedup but enabling
+it requires CUDA toolchain with the same version as pytorch. 8-bit optimizer
+does not seem to be compatible with DeepSpeed at the moment.
+
+```bash
+export MODEL_NAME="CompVis/stable-diffusion-v1-4"
+export INSTANCE_DIR="path-to-instance-images"
+export CLASS_DIR="path-to-class-images"
+export OUTPUT_DIR="path-to-save-model"
+
+accelerate launch --mixed_precision="fp16" train_dreambooth.py \
+  --pretrained_model_name_or_path=$MODEL_NAME \
+  --instance_data_dir=$INSTANCE_DIR \
+  --class_data_dir=$CLASS_DIR \
+  --output_dir=$OUTPUT_DIR \
+  --with_prior_preservation --prior_loss_weight=1.0 \
+  --instance_prompt="a photo of sks dog" \
+  --class_prompt="a photo of dog" \
+  --resolution=512 \
+  --train_batch_size=1 \
+  --sample_batch_size=1 \
+  --gradient_accumulation_steps=1 --gradient_checkpointing \
+  --learning_rate=5e-6 \
+  --lr_scheduler="constant" \
+  --lr_warmup_steps=0 \
+  --num_class_images=200 \
+  --max_train_steps=800
+```
+
+### Fine-tune text encoder with the UNet.
+
+The script also allows to fine-tune the `text_encoder` along with the `unet`. It's been observed experimentally that fine-tuning `text_encoder` gives much better results especially on faces. 
+Pass the `--train_text_encoder` argument to the script to enable training `text_encoder`.
+
+___Note: Training text encoder requires more memory, with this option the training won't fit on 16GB GPU. It needs at least 24GB VRAM.___
+
+```bash
+export MODEL_NAME="CompVis/stable-diffusion-v1-4"
+export INSTANCE_DIR="path-to-instance-images"
+export CLASS_DIR="path-to-class-images"
+export OUTPUT_DIR="path-to-save-model"
+
+accelerate launch train_dreambooth.py \
+  --pretrained_model_name_or_path=$MODEL_NAME  \
+  --train_text_encoder \
+  --instance_data_dir=$INSTANCE_DIR \
+  --class_data_dir=$CLASS_DIR \
+  --output_dir=$OUTPUT_DIR \
+  --with_prior_preservation --prior_loss_weight=1.0 \
+  --instance_prompt="a photo of sks dog" \
+  --class_prompt="a photo of dog" \
+  --resolution=512 \
+  --train_batch_size=1 \
+  --use_8bit_adam \
+  --gradient_checkpointing \
+  --learning_rate=2e-6 \
+  --lr_scheduler="constant" \
+  --lr_warmup_steps=0 \
+  --num_class_images=200 \
+  --max_train_steps=800
+```
+
+### Using DreamBooth for other pipelines than Stable Diffusion
+
+Altdiffusion also support dreambooth now, the runing comman is basically the same as abouve, all you need to do is replace the `MODEL_NAME` like this:
+One can now simply change the `pretrained_model_name_or_path` to another architecture such as [`AltDiffusion`](https://huggingface.co/docs/diffusers/api/pipelines/alt_diffusion).
+
+```
+export MODEL_NAME="CompVis/stable-diffusion-v1-4" --> export MODEL_NAME="BAAI/AltDiffusion-m9"
+or
+export MODEL_NAME="CompVis/stable-diffusion-v1-4" --> export MODEL_NAME="BAAI/AltDiffusion"
+```
+
+### Training with xformers:
+You can enable memory efficient attention by [installing xFormers](https://github.com/facebookresearch/xformers#installing-xformers) and padding the `--enable_xformers_memory_efficient_attention` argument to the script. This is not available with the Flax/JAX implementation.
+
+You can also use Dreambooth to train the specialized in-painting model. See [the script in the research folder for details](https://github.com/huggingface/diffusers/tree/main/examples/research_projects/dreambooth_inpaint).
\ No newline at end of file

examples/research_projects/multi_subject_dreambooth/requirements.txt

+accelerate
+torchvision
+transformers>=4.25.1
+ftfy
+tensorboard
+modelcards
\ No newline at end of file

examples/research_projects/multi_subject_dreambooth/train_multi_subject_dreambooth.py

+import argparse
+import hashlib
+import itertools
+import logging
+import math
+import os
+import warnings
+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
+
+import datasets
+import diffusers
+import transformers
+from accelerate import Accelerator
+from accelerate.logging import get_logger
+from accelerate.utils import set_seed
+from diffusers import AutoencoderKL, DDPMScheduler, DiffusionPipeline, UNet2DConditionModel
+from diffusers.optimization import get_scheduler
+from diffusers.utils import check_min_version
+from diffusers.utils.import_utils import is_xformers_available
+from huggingface_hub import HfFolder, Repository, whoami
+from PIL import Image
+from torchvision import transforms
+from tqdm.auto import tqdm
+from transformers import AutoTokenizer, PretrainedConfig
+
+
+# Will error if the minimal version of diffusers is not installed. Remove at your own risks.
+check_min_version("0.10.0.dev0")
+
+logger = get_logger(__name__)
+
+
+def import_model_class_from_model_name_or_path(pretrained_model_name_or_path: str, revision: str):
+    text_encoder_config = PretrainedConfig.from_pretrained(
+        pretrained_model_name_or_path,
+        subfolder="text_encoder",
+        revision=revision,
+    )
+    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.")
+
+
+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(
+        "--revision",
+        type=str,
+        default=None,
+        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,
+        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,
+        required=True,
+        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(
+        "--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 there are not enough images already present in"
+            " class_data_dir, 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(
+        "--checkpointing_steps",
+        type=int,
+        default=500,
+        help=(
+            "Save a checkpoint of the training state every X updates. These checkpoints can be used both as final"
+            " checkpoints in case they are better than the last checkpoint, and are also suitable for resuming"
+            " training using `--resume_from_checkpoint`."
+        ),
+    )
+    parser.add_argument(
+        "--resume_from_checkpoint",
+        type=str,
+        default=None,
+        help=(
+            "Whether training should be resumed from a previous checkpoint. Use a path saved by"
+            ' `--checkpointing_steps`, or `"latest"` to automatically select the last available checkpoint.'
+        ),
+    )
+    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(
+        "--lr_num_cycles",
+        type=int,
+        default=1,
+        help="Number of hard resets of the lr in cosine_with_restarts scheduler.",
+    )
+    parser.add_argument("--lr_power", type=float, default=1.0, help="Power factor of the polynomial 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(
+        "--allow_tf32",
+        action="store_true",
+        help=(
+            "Whether or not to allow TF32 on Ampere GPUs. Can be used to speed up training. For more information, see"
+            " https://pytorch.org/docs/stable/notes/cuda.html#tensorfloat-32-tf32-on-ampere-devices"
+        ),
+    )
+    parser.add_argument(
+        "--report_to",
+        type=str,
+        default="tensorboard",
+        help=(
+            'The integration to report the results and logs to. Supported platforms are `"tensorboard"`'
+            ' (default), `"wandb"` and `"comet_ml"`. Use `"all"` to report to all integrations.'
+        ),
+    )
+    parser.add_argument(
+        "--mixed_precision",
+        type=str,
+        default=None,
+        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.  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."
+        ),
+    )
+    parser.add_argument(
+        "--prior_generation_precision",
+        type=str,
+        default=None,
+        choices=["no", "fp32", "fp16", "bf16"],
+        help=(
+            "Choose prior generation precision between fp32, fp16 and bf16 (bfloat16). Bf16 requires PyTorch >="
+            " 1.10.and an Nvidia Ampere GPU.  Default to  fp16 if a GPU is available else fp32."
+        ),
+    )
+    parser.add_argument("--local_rank", type=int, default=-1, help="For distributed training: local_rank")
+    parser.add_argument(
+        "--enable_xformers_memory_efficient_attention", action="store_true", help="Whether or not to use xformers."
+    )
+
+    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
+
+    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.")
+    else:
+        # logger is not available yet
+        if args.class_data_dir is not None:
+            warnings.warn("You need not use --class_data_dir without --with_prior_preservation.")
+        if args.class_prompt is not None:
+            warnings.warn("You need not use --class_prompt without --with_prior_preservation.")
+
+    return args
+
+
+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,
+        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 = []
+        self.instance_images_path = []
+        self.num_instance_images = []
+        self.instance_prompt = []
+        self.class_data_root = []
+        self.class_images_path = []
+        self.num_class_images = []
+        self.class_prompt = []
+        self._length = 0
+
+        for i in range(len(instance_data_root)):
+            self.instance_data_root.append(Path(instance_data_root[i]))
+            if not self.instance_data_root[i].exists():
+                raise ValueError("Instance images root doesn't exists.")
+
+            self.instance_images_path.append(list(Path(instance_data_root[i]).iterdir()))
+            self.num_instance_images.append(len(self.instance_images_path[i]))
+            self.instance_prompt.append(instance_prompt[i])
+            self._length += self.num_instance_images[i]
+
+            if class_data_root is not None:
+                self.class_data_root.append(Path(class_data_root[i]))
+                self.class_data_root[i].mkdir(parents=True, exist_ok=True)
+                self.class_images_path.append(list(self.class_data_root[i].iterdir()))
+                self.num_class_images.append(len(self.class_images_path))
+                if self.num_class_images[i] > self.num_instance_images[i]:
+                    self._length -= self.num_instance_images[i]
+                    self._length += self.num_class_images[i]
+                self.class_prompt.append(class_prompt[i])
+            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 = {}
+        for i in range(len(self.instance_images_path)):
+            instance_image = Image.open(self.instance_images_path[i][index % self.num_instance_images[i]])
+            if not instance_image.mode == "RGB":
+                instance_image = instance_image.convert("RGB")
+            example[f"instance_images_{i}"] = self.image_transforms(instance_image)
+            example[f"instance_prompt_ids_{i}"] = self.tokenizer(
+                self.instance_prompt[i],
+                truncation=True,
+                padding="max_length",
+                max_length=self.tokenizer.model_max_length,
+                return_tensors="pt",
+            ).input_ids
+
+        if self.class_data_root:
+            for i in range(len(self.class_data_root)):
+                class_image = Image.open(self.class_images_path[i][index % self.num_class_images[i]])
+                if not class_image.mode == "RGB":
+                    class_image = class_image.convert("RGB")
+                example[f"class_images_{i}"] = self.image_transforms(class_image)
+                example[f"class_prompt_ids_{i}"] = self.tokenizer(
+                    self.class_prompt[i],
+                    truncation=True,
+                    padding="max_length",
+                    max_length=self.tokenizer.model_max_length,
+                    return_tensors="pt",
+                ).input_ids
+
+        return example
+
+
+def collate_fn(num_instances, examples, with_prior_preservation=False):
+    input_ids = []
+    pixel_values = []
+
+    for i in range(num_instances):
+        input_ids += [example[f"instance_prompt_ids_{i}"] for example in examples]
+        pixel_values += [example[f"instance_images_{i}"] 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:
+        for i in range(num_instances):
+            input_ids += [example[f"class_prompt_ids_{i}"] for example in examples]
+            pixel_values += [example[f"class_images_{i}"] 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
+
+
+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}"
+
+
+def main(args):
+    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=args.report_to,
+        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."
+        )
+
+    # Parse instance and class inputs, and double check that lengths match
+    instance_data_dir = args.instance_data_dir.split(",")
+    instance_prompt = args.instance_prompt.split(",")
+    assert all(
+        x == len(instance_data_dir) for x in [len(instance_data_dir), len(instance_prompt)]
+    ), "Instance data dir and prompt inputs are not of the same length."
+
+    if args.with_prior_preservation:
+        class_data_dir = args.class_data_dir.split(",")
+        class_prompt = args.class_prompt.split(",")
+        assert all(
+            x == len(instance_data_dir)
+            for x in [len(instance_data_dir), len(instance_prompt), len(class_data_dir), len(class_prompt)]
+        ), "Instance & class data dir or prompt inputs are not of the same length."
+    else:
+        class_data_dir = args.class_data_dir
+        class_prompt = args.class_prompt
+
+    # Make one log on every process with the configuration for debugging.
+    logging.basicConfig(
+        format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
+        datefmt="%m/%d/%Y %H:%M:%S",
+        level=logging.INFO,
+    )
+    logger.info(accelerator.state, main_process_only=False)
+    if accelerator.is_local_main_process:
+        datasets.utils.logging.set_verbosity_warning()
+        transformers.utils.logging.set_verbosity_warning()
+        diffusers.utils.logging.set_verbosity_info()
+    else:
+        datasets.utils.logging.set_verbosity_error()
+        transformers.utils.logging.set_verbosity_error()
+        diffusers.utils.logging.set_verbosity_error()
+
+    # If passed along, set the training seed now.
+    if args.seed is not None:
+        set_seed(args.seed)
+
+    # Generate class images if prior preservation is enabled.
+    if args.with_prior_preservation:
+        for i in range(len(class_data_dir)):
+            class_images_dir = Path(class_data_dir[i])
+            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
+                if args.prior_generation_precision == "fp32":
+                    torch_dtype = torch.float32
+                elif args.prior_generation_precision == "fp16":
+                    torch_dtype = torch.float16
+                elif args.prior_generation_precision == "bf16":
+                    torch_dtype = torch.bfloat16
+                pipeline = DiffusionPipeline.from_pretrained(
+                    args.pretrained_model_name_or_path,
+                    torch_dtype=torch_dtype,
+                    safety_checker=None,
+                    revision=args.revision,
+                )
+                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(class_prompt[i], 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
+                ):
+                    images = pipeline(example["prompt"]).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()
+
+    # 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:
+        tokenizer = AutoTokenizer.from_pretrained(args.tokenizer_name, revision=args.revision, use_fast=False)
+    elif args.pretrained_model_name_or_path:
+        tokenizer = AutoTokenizer.from_pretrained(
+            args.pretrained_model_name_or_path,
+            subfolder="tokenizer",
+            revision=args.revision,
+            use_fast=False,
+        )
+
+    # import correct text encoder class
+    text_encoder_cls = import_model_class_from_model_name_or_path(args.pretrained_model_name_or_path, args.revision)
+
+    # Load scheduler and models
+    noise_scheduler = DDPMScheduler.from_pretrained(args.pretrained_model_name_or_path, subfolder="scheduler")
+    text_encoder = text_encoder_cls.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
+    )
+
+    vae.requires_grad_(False)
+    if not args.train_text_encoder:
+        text_encoder.requires_grad_(False)
+
+    if args.enable_xformers_memory_efficient_attention:
+        if is_xformers_available():
+            unet.enable_xformers_memory_efficient_attention()
+        else:
+            raise ValueError("xformers is not available. Make sure it is installed correctly")
+
+    if args.gradient_checkpointing:
+        unet.enable_gradient_checkpointing()
+        if args.train_text_encoder:
+            text_encoder.gradient_checkpointing_enable()
+
+    # Enable TF32 for faster training on Ampere GPUs,
+    # cf https://pytorch.org/docs/stable/notes/cuda.html#tensorfloat-32-tf32-on-ampere-devices
+    if args.allow_tf32:
+        torch.backends.cuda.matmul.allow_tf32 = True
+
+    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
+
+    # Optimizer creation
+    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,
+    )
+
+    # Dataset and DataLoaders creation:
+    train_dataset = DreamBoothDataset(
+        instance_data_root=instance_data_dir,
+        instance_prompt=instance_prompt,
+        class_data_root=class_data_dir if args.with_prior_preservation else None,
+        class_prompt=class_prompt,
+        tokenizer=tokenizer,
+        size=args.resolution,
+        center_crop=args.center_crop,
+    )
+
+    train_dataloader = torch.utils.data.DataLoader(
+        train_dataset,
+        batch_size=args.train_batch_size,
+        shuffle=True,
+        collate_fn=lambda examples: collate_fn(len(instance_data_dir), examples, args.with_prior_preservation),
+        num_workers=1,
+    )
+
+    # 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,
+        num_cycles=args.lr_num_cycles,
+        power=args.lr_power,
+    )
+
+    # Prepare everything with our `accelerator`.
+    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
+        )
+
+    # 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.
+    weight_dtype = torch.float32
+    if accelerator.mixed_precision == "fp16":
+        weight_dtype = torch.float16
+    elif accelerator.mixed_precision == "bf16":
+        weight_dtype = torch.bfloat16
+
+    # Move vae and text_encoder to device and cast to weight_dtype
+    vae.to(accelerator.device, dtype=weight_dtype)
+    if not args.train_text_encoder:
+        text_encoder.to(accelerator.device, dtype=weight_dtype)
+
+    # 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}")
+    global_step = 0
+    first_epoch = 0
+
+    # Potentially load in the weights and states from a previous save
+    if args.resume_from_checkpoint:
+        if args.resume_from_checkpoint != "latest":
+            path = os.path.basename(args.resume_from_checkpoint)
+        else:
+            # Get the mos recent checkpoint
+            dirs = os.listdir(args.output_dir)
+            dirs = [d for d in dirs if d.startswith("checkpoint")]
+            dirs = sorted(dirs, key=lambda x: int(x.split("-")[1]))
+            path = dirs[-1]
+        accelerator.print(f"Resuming from checkpoint {path}")
+        accelerator.load_state(os.path.join(args.output_dir, path))
+        global_step = int(path.split("-")[1])
+
+        resume_global_step = global_step * args.gradient_accumulation_steps
+        first_epoch = resume_global_step // num_update_steps_per_epoch
+        resume_step = resume_global_step % num_update_steps_per_epoch
+
+    # Only show the progress bar once on each machine.
+    progress_bar = tqdm(range(global_step, args.max_train_steps), disable=not accelerator.is_local_main_process)
+    progress_bar.set_description("Steps")
+
+    for epoch in range(first_epoch, args.num_train_epochs):
+        unet.train()
+        if args.train_text_encoder:
+            text_encoder.train()
+        for step, batch in enumerate(train_dataloader):
+            # Skip steps until we reach the resumed step
+            if args.resume_from_checkpoint and epoch == first_epoch and step < resume_step:
+                if step % args.gradient_accumulation_steps == 0:
+                    progress_bar.update(1)
+                continue
+
+            with accelerator.accumulate(unet):
+                # Convert images to latent space
+                latents = vae.encode(batch["pixel_values"].to(dtype=weight_dtype)).latent_dist.sample()
+                latents = latents * 0.18215
+
+                # 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
+                encoder_hidden_states = text_encoder(batch["input_ids"])[0]
+
+                # Predict the noise residual
+                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}")
+
+                if args.with_prior_preservation:
+                    # 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)
+
+                    # Compute instance loss
+                    loss = F.mse_loss(model_pred.float(), target.float(), reduction="mean")
+
+                    # Compute prior loss
+                    prior_loss = F.mse_loss(model_pred_prior.float(), target_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(model_pred.float(), target.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()
+
+            # 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.checkpointing_steps == 0:
+                    if accelerator.is_main_process:
+                        save_path = os.path.join(args.output_dir, f"checkpoint-{global_step}")
+                        accelerator.save_state(save_path)
+                        logger.info(f"Saved state to {save_path}")
+
+            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
+
+    # Create the pipeline using using the trained modules and save it.
+    accelerator.wait_for_everyone()
+    if accelerator.is_main_process:
+        pipeline = DiffusionPipeline.from_pretrained(
+            args.pretrained_model_name_or_path,
+            unet=accelerator.unwrap_model(unet),
+            text_encoder=accelerator.unwrap_model(text_encoder),
+            revision=args.revision,
+        )
+        pipeline.save_pretrained(args.output_dir)
+
+        if args.push_to_hub:
+            repo.push_to_hub(commit_message="End of training", blocking=False, auto_lfs_prune=True)
+
+    accelerator.end_training()
+
+
+if __name__ == "__main__":
+    args = parse_args()
+    main(args)