更新diffusers文件夹

diffusers-0.27.0/examples/text_to_image/run-4.sh

-export MODEL_NAME="/path/to/stable-diffusion-2-1-base/"
+# source /public/home/chenxi/hopt/dtk24042-miopen.sh 
+export MODEL_NAME="/path/to/stable-diffusion-2-1-base"
 export OUTPUT_DIR="./output/sd2.1"
-export DATASET_NAME="/path/to/data/train-00000-of-00001-566cc9b19d7203f8.parquet"
+export DATASET_NAME="/path/to/train-00000-of-00001-566cc9b19d7203f8.parquet"
 #export DATASET_NAME="./pokemon-blip-captions"
-#export HIP_VISIBLE_DEVICES=7
+export HIP_VISIBLE_DEVICES=4,5,6,7
 #export LD_LIBRARY_PATH=/opt/rocblas-install/lib/:$LD_LIBRARY_PATH
 #export PYTORCH_HIP_ALLOC_CONF=max_split_size_mb:25314
-export HIP_VISIBLE_DEVICES=4,5,6,7
 
 torchrun --nproc_per_node=4 train_text_to_image_lora.py \
   --pretrained_model_name_or_path=$MODEL_NAME \
@@ -14,7 +14,7 @@ torchrun --nproc_per_node=4 train_text_to_image_lora.py \
   --resolution=960 \
   --center_crop \
   --random_flip \
-  --train_batch_size=8 \
+  --train_batch_size=2 \
   --gradient_accumulation_steps=4 \
   --max_train_steps=101 \
   --learning_rate=1e-04 \

diffusers-0.27.0/examples/text_to_image/run.sh

 # source /public/home/chenxi/hopt/dtk24042-miopen.sh 
-export MODEL_NAME="/path/to/stable-diffusion-2-1-base/"
+export MODEL_NAME="/path/to/stable-diffusion-2-1-base"
 export OUTPUT_DIR="./output/sd2.1"
-export DATASET_NAME="/path/to/data/train-00000-of-00001-566cc9b19d7203f8.parquet"
+export DATASET_NAME="path/to/train-00000-of-00001-566cc9b19d7203f8.parquet"
 #export DATASET_NAME="./pokemon-blip-captions"
-export HIP_VISIBLE_DEVICES=4,5,6,7
+export HIP_VISIBLE_DEVICES=7
 #export LD_LIBRARY_PATH=/opt/rocblas-install/lib/:$LD_LIBRARY_PATH
 #export PYTORCH_HIP_ALLOC_CONF=max_split_size_mb:25314
 
-python train_text_to_image_lora.py \
+python  train_text_to_image_lora.py \
   --pretrained_model_name_or_path=$MODEL_NAME \
   --dataset_name=$DATASET_NAME \
   --dataloader_num_workers=8 \
   --resolution=960 \
   --center_crop \
   --random_flip \
-  --train_batch_size=2 \
+  --train_batch_size=8 \
   --gradient_accumulation_steps=4 \
   --max_train_steps=101 \
   --learning_rate=1e-04 \

diffusers-0.27.0/examples/text_to_image/train_text_to_image_lora.py

@@ -766,7 +766,7 @@ def main():
     for epoch in range(first_epoch, args.num_train_epochs):
         unet.train()
         train_loss = 0.0
-        # from layer_check_pt import acc_check_hook, register_hook 
+        from layer_check_pt import acc_check_hook, register_hook 
         for step, batch in enumerate(train_dataloader):
             with accelerator.accumulate(unet):
                 # Convert images to latent space

diffusers-0.27.0/examples/textual_inversion/README.md

+## Textual Inversion fine-tuning example
+
+[Textual inversion](https://arxiv.org/abs/2208.01618) is a method to personalize text2image models like stable diffusion on your own images using just 3-5 examples.
+The `textual_inversion.py` script shows how to implement the training procedure and adapt it for stable diffusion.
+
+## Running on Colab
+
+Colab for training
+[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/sd_textual_inversion_training.ipynb)
+
+Colab for inference
+[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/stable_conceptualizer_inference.ipynb)
+
+## Running locally with PyTorch
+### Installing the dependencies
+
+Before running the scripts, make sure to install the library's training dependencies:
+
+**Important**
+
+To make sure you can successfully run the latest versions of the example scripts, we highly recommend **installing from source** and keeping the install up to date as we update the example scripts frequently and install some example-specific requirements. To do this, execute the following steps in a new virtual environment:
+```bash
+git clone https://github.com/huggingface/diffusers
+cd diffusers
+pip install .
+```
+
+Then cd in the example folder and run:
+```bash
+pip install -r requirements.txt
+```
+
+And initialize an [🤗 Accelerate](https://github.com/huggingface/accelerate/) environment with:
+
+```bash
+accelerate config
+```
+
+### Cat toy example
+
+First, let's login so that we can upload the checkpoint to the Hub during training:
+
+```bash
+huggingface-cli login
+```
+
+Now let's get our dataset. For this example we will use some cat images: https://huggingface.co/datasets/diffusers/cat_toy_example .
+
+Let's first download it locally:
+
+```py
+from huggingface_hub import snapshot_download
+
+local_dir = "./cat"
+snapshot_download("diffusers/cat_toy_example", local_dir=local_dir, repo_type="dataset", ignore_patterns=".gitattributes")
+```
+
+This will be our training data.
+Now we can 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.___**
+
+**___Note: Please follow the [README_sdxl.md](./README_sdxl.md) if you are using the [stable-diffusion-xl](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0).___**
+
+```bash
+export MODEL_NAME="runwayml/stable-diffusion-v1-5"
+export DATA_DIR="./cat"
+
+accelerate launch textual_inversion.py \
+  --pretrained_model_name_or_path=$MODEL_NAME \
+  --train_data_dir=$DATA_DIR \
+  --learnable_property="object" \
+  --placeholder_token="<cat-toy>" \
+  --initializer_token="toy" \
+  --resolution=512 \
+  --train_batch_size=1 \
+  --gradient_accumulation_steps=4 \
+  --max_train_steps=3000 \
+  --learning_rate=5.0e-04 \
+  --scale_lr \
+  --lr_scheduler="constant" \
+  --lr_warmup_steps=0 \
+  --push_to_hub \
+  --output_dir="textual_inversion_cat"
+```
+
+A full training run takes ~1 hour on one V100 GPU.
+
+**Note**: As described in [the official paper](https://arxiv.org/abs/2208.01618)
+only one embedding vector is used for the placeholder token, *e.g.* `"<cat-toy>"`.
+However, one can also add multiple embedding vectors for the placeholder token
+to increase the number of fine-tuneable parameters. This can help the model to learn
+more complex details. To use multiple embedding vectors, you should define `--num_vectors`
+to a number larger than one, *e.g.*:
+```bash
+--num_vectors 5
+```
+
+The saved textual inversion vectors will then be larger in size compared to the default case.
+
+### Inference
+
+Once you have trained a model using above command, the inference can be done simply using the `StableDiffusionPipeline`. Make sure to include the `placeholder_token` 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 <cat-toy> backpack"
+
+image = pipe(prompt, num_inference_steps=50, guidance_scale=7.5).images[0]
+
+image.save("cat-backpack.png")
+```
+
+
+## Training with Flax/JAX
+
+For faster training on TPUs and GPUs you can leverage the flax training example. Follow the instructions above to get the model and dataset before running the script.
+
+Before running the scripts, make sure to install the library's training dependencies:
+
+```bash
+pip install -U -r requirements_flax.txt
+```
+
+```bash
+export MODEL_NAME="duongna/stable-diffusion-v1-4-flax"
+export DATA_DIR="path-to-dir-containing-images"
+
+python textual_inversion_flax.py \
+  --pretrained_model_name_or_path=$MODEL_NAME \
+  --train_data_dir=$DATA_DIR \
+  --learnable_property="object" \
+  --placeholder_token="<cat-toy>" \
+  --initializer_token="toy" \
+  --resolution=512 \
+  --train_batch_size=1 \
+  --max_train_steps=3000 \
+  --learning_rate=5.0e-04 \
+  --scale_lr \
+  --output_dir="textual_inversion_cat"
+```
+It should be at least 70% faster than the PyTorch script with the same configuration.
+
+### 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.

diffusers-0.27.0/examples/textual_inversion/README_sdxl.md

+## Textual Inversion fine-tuning example for SDXL
+
+```
+export MODEL_NAME="stabilityai/stable-diffusion-xl-base-1.0"
+export DATA_DIR="./cat"
+
+accelerate launch textual_inversion_sdxl.py \
+  --pretrained_model_name_or_path=$MODEL_NAME \
+  --train_data_dir=$DATA_DIR \
+  --learnable_property="object" \
+  --placeholder_token="<cat-toy>" \
+  --initializer_token="toy" \
+  --mixed_precision="bf16" \
+  --resolution=768 \
+  --train_batch_size=1 \
+  --gradient_accumulation_steps=4 \
+  --max_train_steps=500 \
+  --learning_rate=5.0e-04 \
+  --scale_lr \
+  --lr_scheduler="constant" \
+  --lr_warmup_steps=0 \
+  --save_as_full_pipeline \
+  --output_dir="./textual_inversion_cat_sdxl"
+```
+
+For now, only training of the first text encoder is supported. 
\ No newline at end of file

diffusers-0.27.0/examples/textual_inversion/requirements.txt

+accelerate>=0.16.0
+torchvision
+transformers>=4.25.1
+ftfy
+tensorboard
+Jinja2

diffusers-0.27.0/examples/textual_inversion/requirements_flax.txt

+transformers>=4.25.1
+flax
+optax
+torch
+torchvision
+ftfy
+tensorboard
+Jinja2

diffusers-0.27.0/examples/textual_inversion/test_textual_inversion.py

+# coding=utf-8
+# Copyright 2024 HuggingFace Inc.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import logging
+import os
+import sys
+import tempfile
+
+
+sys.path.append("..")
+from test_examples_utils import ExamplesTestsAccelerate, run_command  # noqa: E402
+
+
+logging.basicConfig(level=logging.DEBUG)
+
+logger = logging.getLogger()
+stream_handler = logging.StreamHandler(sys.stdout)
+logger.addHandler(stream_handler)
+
+
+class TextualInversion(ExamplesTestsAccelerate):
+    def test_textual_inversion(self):
+        with tempfile.TemporaryDirectory() as tmpdir:
+            test_args = f"""
+                examples/textual_inversion/textual_inversion.py
+                --pretrained_model_name_or_path hf-internal-testing/tiny-stable-diffusion-pipe
+                --train_data_dir docs/source/en/imgs
+                --learnable_property object
+                --placeholder_token <cat-toy>
+                --initializer_token a
+                --save_steps 1
+                --num_vectors 2
+                --resolution 64
+                --train_batch_size 1
+                --gradient_accumulation_steps 1
+                --max_train_steps 2
+                --learning_rate 5.0e-04
+                --scale_lr
+                --lr_scheduler constant
+                --lr_warmup_steps 0
+                --output_dir {tmpdir}
+                """.split()
+
+            run_command(self._launch_args + test_args)
+            # save_pretrained smoke test
+            self.assertTrue(os.path.isfile(os.path.join(tmpdir, "learned_embeds.safetensors")))
+
+    def test_textual_inversion_checkpointing(self):
+        with tempfile.TemporaryDirectory() as tmpdir:
+            test_args = f"""
+                examples/textual_inversion/textual_inversion.py
+                --pretrained_model_name_or_path hf-internal-testing/tiny-stable-diffusion-pipe
+                --train_data_dir docs/source/en/imgs
+                --learnable_property object
+                --placeholder_token <cat-toy>
+                --initializer_token a
+                --save_steps 1
+                --num_vectors 2
+                --resolution 64
+                --train_batch_size 1
+                --gradient_accumulation_steps 1
+                --max_train_steps 3
+                --learning_rate 5.0e-04
+                --scale_lr
+                --lr_scheduler constant
+                --lr_warmup_steps 0
+                --output_dir {tmpdir}
+                --checkpointing_steps=1
+                --checkpoints_total_limit=2
+                """.split()
+
+            run_command(self._launch_args + test_args)
+
+            # check checkpoint directories exist
+            self.assertEqual(
+                {x for x in os.listdir(tmpdir) if "checkpoint" in x},
+                {"checkpoint-2", "checkpoint-3"},
+            )
+
+    def test_textual_inversion_checkpointing_checkpoints_total_limit_removes_multiple_checkpoints(self):
+        with tempfile.TemporaryDirectory() as tmpdir:
+            test_args = f"""
+                examples/textual_inversion/textual_inversion.py
+                --pretrained_model_name_or_path hf-internal-testing/tiny-stable-diffusion-pipe
+                --train_data_dir docs/source/en/imgs
+                --learnable_property object
+                --placeholder_token <cat-toy>
+                --initializer_token a
+                --save_steps 1
+                --num_vectors 2
+                --resolution 64
+                --train_batch_size 1
+                --gradient_accumulation_steps 1
+                --max_train_steps 2
+                --learning_rate 5.0e-04
+                --scale_lr
+                --lr_scheduler constant
+                --lr_warmup_steps 0
+                --output_dir {tmpdir}
+                --checkpointing_steps=1
+                """.split()
+
+            run_command(self._launch_args + test_args)
+
+            # check checkpoint directories exist
+            self.assertEqual(
+                {x for x in os.listdir(tmpdir) if "checkpoint" in x},
+                {"checkpoint-1", "checkpoint-2"},
+            )
+
+            resume_run_args = f"""
+                examples/textual_inversion/textual_inversion.py
+                --pretrained_model_name_or_path hf-internal-testing/tiny-stable-diffusion-pipe
+                --train_data_dir docs/source/en/imgs
+                --learnable_property object
+                --placeholder_token <cat-toy>
+                --initializer_token a
+                --save_steps 1
+                --num_vectors 2
+                --resolution 64
+                --train_batch_size 1
+                --gradient_accumulation_steps 1
+                --max_train_steps 2
+                --learning_rate 5.0e-04
+                --scale_lr
+                --lr_scheduler constant
+                --lr_warmup_steps 0
+                --output_dir {tmpdir}
+                --checkpointing_steps=1
+                --resume_from_checkpoint=checkpoint-2
+                --checkpoints_total_limit=2
+                """.split()
+
+            run_command(self._launch_args + resume_run_args)
+
+            # check checkpoint directories exist
+            self.assertEqual(
+                {x for x in os.listdir(tmpdir) if "checkpoint" in x},
+                {"checkpoint-2", "checkpoint-3"},
+            )

diffusers-0.27.0/examples/textual_inversion/test_textual_inversion_sdxl.py

+# coding=utf-8
+# Copyright 2024 HuggingFace Inc.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import logging
+import os
+import sys
+import tempfile
+
+
+sys.path.append("..")
+from test_examples_utils import ExamplesTestsAccelerate, run_command  # noqa: E402
+
+
+logging.basicConfig(level=logging.DEBUG)
+
+logger = logging.getLogger()
+stream_handler = logging.StreamHandler(sys.stdout)
+logger.addHandler(stream_handler)
+
+
+class TextualInversionSdxl(ExamplesTestsAccelerate):
+    def test_textual_inversion_sdxl(self):
+        with tempfile.TemporaryDirectory() as tmpdir:
+            test_args = f"""
+                examples/textual_inversion/textual_inversion_sdxl.py
+                --pretrained_model_name_or_path hf-internal-testing/tiny-sdxl-pipe
+                --train_data_dir docs/source/en/imgs
+                --learnable_property object
+                --placeholder_token <cat-toy>
+                --initializer_token a
+                --save_steps 1
+                --num_vectors 2
+                --resolution 64
+                --train_batch_size 1
+                --gradient_accumulation_steps 1
+                --max_train_steps 2
+                --learning_rate 5.0e-04
+                --scale_lr
+                --lr_scheduler constant
+                --lr_warmup_steps 0
+                --output_dir {tmpdir}
+                """.split()
+
+            run_command(self._launch_args + test_args)
+            # save_pretrained smoke test
+            self.assertTrue(os.path.isfile(os.path.join(tmpdir, "learned_embeds.safetensors")))
+
+    def test_textual_inversion_sdxl_checkpointing(self):
+        with tempfile.TemporaryDirectory() as tmpdir:
+            test_args = f"""
+                examples/textual_inversion/textual_inversion_sdxl.py
+                --pretrained_model_name_or_path hf-internal-testing/tiny-sdxl-pipe
+                --train_data_dir docs/source/en/imgs
+                --learnable_property object
+                --placeholder_token <cat-toy>
+                --initializer_token a
+                --save_steps 1
+                --num_vectors 2
+                --resolution 64
+                --train_batch_size 1
+                --gradient_accumulation_steps 1
+                --max_train_steps 3
+                --learning_rate 5.0e-04
+                --scale_lr
+                --lr_scheduler constant
+                --lr_warmup_steps 0
+                --output_dir {tmpdir}
+                --checkpointing_steps=1
+                --checkpoints_total_limit=2
+                """.split()
+
+            run_command(self._launch_args + test_args)
+
+            # check checkpoint directories exist
+            self.assertEqual(
+                {x for x in os.listdir(tmpdir) if "checkpoint" in x},
+                {"checkpoint-2", "checkpoint-3"},
+            )
+
+    def test_textual_inversion_sdxl_checkpointing_checkpoints_total_limit_removes_multiple_checkpoints(self):
+        with tempfile.TemporaryDirectory() as tmpdir:
+            test_args = f"""
+                examples/textual_inversion/textual_inversion_sdxl.py
+                --pretrained_model_name_or_path hf-internal-testing/tiny-sdxl-pipe
+                --train_data_dir docs/source/en/imgs
+                --learnable_property object
+                --placeholder_token <cat-toy>
+                --initializer_token a
+                --save_steps 1
+                --num_vectors 2
+                --resolution 64
+                --train_batch_size 1
+                --gradient_accumulation_steps 1
+                --max_train_steps 2
+                --learning_rate 5.0e-04
+                --scale_lr
+                --lr_scheduler constant
+                --lr_warmup_steps 0
+                --output_dir {tmpdir}
+                --checkpointing_steps=1
+                """.split()
+
+            run_command(self._launch_args + test_args)
+
+            # check checkpoint directories exist
+            self.assertEqual(
+                {x for x in os.listdir(tmpdir) if "checkpoint" in x},
+                {"checkpoint-1", "checkpoint-2"},
+            )
+
+            resume_run_args = f"""
+                examples/textual_inversion/textual_inversion_sdxl.py
+                --pretrained_model_name_or_path hf-internal-testing/tiny-sdxl-pipe
+                --train_data_dir docs/source/en/imgs
+                --learnable_property object
+                --placeholder_token <cat-toy>
+                --initializer_token a
+                --save_steps 1
+                --num_vectors 2
+                --resolution 64
+                --train_batch_size 1
+                --gradient_accumulation_steps 1
+                --max_train_steps 2
+                --learning_rate 5.0e-04
+                --scale_lr
+                --lr_scheduler constant
+                --lr_warmup_steps 0
+                --output_dir {tmpdir}
+                --checkpointing_steps=1
+                --resume_from_checkpoint=checkpoint-2
+                --checkpoints_total_limit=2
+                """.split()
+
+            run_command(self._launch_args + resume_run_args)
+
+            # check checkpoint directories exist
+            self.assertEqual(
+                {x for x in os.listdir(tmpdir) if "checkpoint" in x},
+                {"checkpoint-2", "checkpoint-3"},
+            )

diffusers-0.27.0/examples/textual_inversion/textual_inversion.py

+#!/usr/bin/env python
+# coding=utf-8
+# Copyright 2024 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+
+import argparse
+import logging
+import math
+import os
+import random
+import shutil
+import warnings
+from pathlib import Path
+
+import numpy as np
+import PIL
+import safetensors
+import torch
+import torch.nn.functional as F
+import torch.utils.checkpoint
+import transformers
+from accelerate import Accelerator
+from accelerate.logging import get_logger
+from accelerate.utils import ProjectConfiguration, set_seed
+from huggingface_hub import create_repo, upload_folder
+
+# TODO: remove and import from diffusers.utils when the new version of diffusers is released
+from packaging import version
+from PIL import Image
+from torch.utils.data import Dataset
+from torchvision import transforms
+from tqdm.auto import tqdm
+from transformers import CLIPTextModel, CLIPTokenizer
+
+import diffusers
+from diffusers import (
+    AutoencoderKL,
+    DDPMScheduler,
+    DiffusionPipeline,
+    DPMSolverMultistepScheduler,
+    StableDiffusionPipeline,
+    UNet2DConditionModel,
+)
+from diffusers.optimization import get_scheduler
+from diffusers.utils import check_min_version, is_wandb_available
+from diffusers.utils.hub_utils import load_or_create_model_card, populate_model_card
+from diffusers.utils.import_utils import is_xformers_available
+
+
+if is_wandb_available():
+    import wandb
+
+if version.parse(version.parse(PIL.__version__).base_version) >= version.parse("9.1.0"):
+    PIL_INTERPOLATION = {
+        "linear": PIL.Image.Resampling.BILINEAR,
+        "bilinear": PIL.Image.Resampling.BILINEAR,
+        "bicubic": PIL.Image.Resampling.BICUBIC,
+        "lanczos": PIL.Image.Resampling.LANCZOS,
+        "nearest": PIL.Image.Resampling.NEAREST,
+    }
+else:
+    PIL_INTERPOLATION = {
+        "linear": PIL.Image.LINEAR,
+        "bilinear": PIL.Image.BILINEAR,
+        "bicubic": PIL.Image.BICUBIC,
+        "lanczos": PIL.Image.LANCZOS,
+        "nearest": PIL.Image.NEAREST,
+    }
+# ------------------------------------------------------------------------------
+
+
+# Will error if the minimal version of diffusers is not installed. Remove at your own risks.
+check_min_version("0.27.0")
+
+logger = get_logger(__name__)
+
+
+def save_model_card(repo_id: str, images: list = None, base_model: str = None, repo_folder: str = None):
+    img_str = ""
+    if images is not None:
+        for i, image in enumerate(images):
+            image.save(os.path.join(repo_folder, f"image_{i}.png"))
+            img_str += f"![img_{i}](./image_{i}.png)\n"
+    model_description = f"""
+# Textual inversion text2image fine-tuning - {repo_id}
+These are textual inversion adaption weights for {base_model}. You can find some example images in the following. \n
+{img_str}
+"""
+    model_card = load_or_create_model_card(
+        repo_id_or_path=repo_id,
+        from_training=True,
+        license="creativeml-openrail-m",
+        base_model=base_model,
+        model_description=model_description,
+        inference=True,
+    )
+
+    tags = [
+        "stable-diffusion",
+        "stable-diffusion-diffusers",
+        "text-to-image",
+        "diffusers",
+        "textual_inversion",
+        "diffusers-training",
+    ]
+    model_card = populate_model_card(model_card, tags=tags)
+
+    model_card.save(os.path.join(repo_folder, "README.md"))
+
+
+def log_validation(text_encoder, tokenizer, unet, vae, args, accelerator, weight_dtype, epoch):
+    logger.info(
+        f"Running validation... \n Generating {args.num_validation_images} images with prompt:"
+        f" {args.validation_prompt}."
+    )
+    # create pipeline (note: unet and vae are loaded again in float32)
+    pipeline = DiffusionPipeline.from_pretrained(
+        args.pretrained_model_name_or_path,
+        text_encoder=accelerator.unwrap_model(text_encoder),
+        tokenizer=tokenizer,
+        unet=unet,
+        vae=vae,
+        safety_checker=None,
+        revision=args.revision,
+        variant=args.variant,
+        torch_dtype=weight_dtype,
+    )
+    pipeline.scheduler = DPMSolverMultistepScheduler.from_config(pipeline.scheduler.config)
+    pipeline = pipeline.to(accelerator.device)
+    pipeline.set_progress_bar_config(disable=True)
+
+    # run inference
+    generator = None if args.seed is None else torch.Generator(device=accelerator.device).manual_seed(args.seed)
+    images = []
+    for _ in range(args.num_validation_images):
+        with torch.autocast("cuda"):
+            image = pipeline(args.validation_prompt, num_inference_steps=25, generator=generator).images[0]
+        images.append(image)
+
+    for tracker in accelerator.trackers:
+        if tracker.name == "tensorboard":
+            np_images = np.stack([np.asarray(img) for img in images])
+            tracker.writer.add_images("validation", np_images, epoch, dataformats="NHWC")
+        if tracker.name == "wandb":
+            tracker.log(
+                {
+                    "validation": [
+                        wandb.Image(image, caption=f"{i}: {args.validation_prompt}") for i, image in enumerate(images)
+                    ]
+                }
+            )
+
+    del pipeline
+    torch.cuda.empty_cache()
+    return images
+
+
+def save_progress(text_encoder, placeholder_token_ids, accelerator, args, save_path, safe_serialization=True):
+    logger.info("Saving embeddings")
+    learned_embeds = (
+        accelerator.unwrap_model(text_encoder)
+        .get_input_embeddings()
+        .weight[min(placeholder_token_ids) : max(placeholder_token_ids) + 1]
+    )
+    learned_embeds_dict = {args.placeholder_token: learned_embeds.detach().cpu()}
+
+    if safe_serialization:
+        safetensors.torch.save_file(learned_embeds_dict, save_path, metadata={"format": "pt"})
+    else:
+        torch.save(learned_embeds_dict, save_path)
+
+
+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(
+        "--save_as_full_pipeline",
+        action="store_true",
+        help="Save the complete stable diffusion pipeline.",
+    )
+    parser.add_argument(
+        "--num_vectors",
+        type=int,
+        default=1,
+        help="How many textual inversion vectors shall be used to learn the concept.",
+    )
+    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(
+        "--variant",
+        type=str,
+        default=None,
+        help="Variant of the model files of the pretrained model identifier from huggingface.co/models, 'e.g.' fp16",
+    )
+    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, required=True, help="A folder containing the training data."
+    )
+    parser.add_argument(
+        "--placeholder_token",
+        type=str,
+        default=None,
+        required=True,
+        help="A token to use as a placeholder for the concept.",
+    )
+    parser.add_argument(
+        "--initializer_token", type=str, default=None, required=True, 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(
+        "--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=1e-4,
+        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(
+        "--dataloader_num_workers",
+        type=int,
+        default=0,
+        help=(
+            "Number of subprocesses to use for data loading. 0 means that the data will be loaded in the main process."
+        ),
+    )
+    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 Nvidia Ampere GPU or Intel Gen 4 Xeon (and later) ."
+        ),
+    )
+    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(
+        "--validation_prompt",
+        type=str,
+        default=None,
+        help="A prompt that is used during validation to verify that the model is learning.",
+    )
+    parser.add_argument(
+        "--num_validation_images",
+        type=int,
+        default=4,
+        help="Number of images that should be generated during validation with `validation_prompt`.",
+    )
+    parser.add_argument(
+        "--validation_steps",
+        type=int,
+        default=100,
+        help=(
+            "Run validation every X steps. Validation consists of running the prompt"
+            " `args.validation_prompt` multiple times: `args.num_validation_images`"
+            " and logging the images."
+        ),
+    )
+    parser.add_argument(
+        "--validation_epochs",
+        type=int,
+        default=None,
+        help=(
+            "Deprecated in favor of validation_steps. Run validation every X epochs. Validation consists of running the prompt"
+            " `args.validation_prompt` multiple times: `args.num_validation_images`"
+            " and logging the images."
+        ),
+    )
+    parser.add_argument("--local_rank", type=int, default=-1, help="For distributed training: local_rank")
+    parser.add_argument(
+        "--checkpointing_steps",
+        type=int,
+        default=500,
+        help=(
+            "Save a checkpoint of the training state every X updates. These checkpoints are only suitable for resuming"
+            " training using `--resume_from_checkpoint`."
+        ),
+    )
+    parser.add_argument(
+        "--checkpoints_total_limit",
+        type=int,
+        default=None,
+        help=("Max number of checkpoints to store."),
+    )
+    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(
+        "--enable_xformers_memory_efficient_attention", action="store_true", help="Whether or not to use xformers."
+    )
+    parser.add_argument(
+        "--no_safe_serialization",
+        action="store_true",
+        help="If specified save the checkpoint not in `safetensors` format, but in original PyTorch format instead.",
+    )
+
+    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_INTERPOLATION["linear"],
+            "bilinear": PIL_INTERPOLATION["bilinear"],
+            "bicubic": PIL_INTERPOLATION["bicubic"],
+            "lanczos": PIL_INTERPOLATION["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 main():
+    args = parse_args()
+    if args.report_to == "wandb" and args.hub_token is not None:
+        raise ValueError(
+            "You cannot use both --report_to=wandb and --hub_token due to a security risk of exposing your token."
+            " Please use `huggingface-cli login` to authenticate with the Hub."
+        )
+
+    logging_dir = os.path.join(args.output_dir, args.logging_dir)
+    accelerator_project_config = ProjectConfiguration(project_dir=args.output_dir, logging_dir=logging_dir)
+    accelerator = Accelerator(
+        gradient_accumulation_steps=args.gradient_accumulation_steps,
+        mixed_precision=args.mixed_precision,
+        log_with=args.report_to,
+        project_config=accelerator_project_config,
+    )
+
+    if args.report_to == "wandb":
+        if not is_wandb_available():
+            raise ImportError("Make sure to install wandb if you want to use it for logging during training.")
+
+    # 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:
+        transformers.utils.logging.set_verbosity_warning()
+        diffusers.utils.logging.set_verbosity_info()
+    else:
+        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)
+
+    # Handle the repository creation
+    if accelerator.is_main_process:
+        if args.output_dir is not None:
+            os.makedirs(args.output_dir, exist_ok=True)
+
+        if args.push_to_hub:
+            repo_id = create_repo(
+                repo_id=args.hub_model_id or Path(args.output_dir).name, exist_ok=True, token=args.hub_token
+            ).repo_id
+
+    # Load tokenizer
+    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")
+
+    # Load scheduler and models
+    noise_scheduler = DDPMScheduler.from_pretrained(args.pretrained_model_name_or_path, subfolder="scheduler")
+    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, variant=args.variant
+    )
+    unet = UNet2DConditionModel.from_pretrained(
+        args.pretrained_model_name_or_path, subfolder="unet", revision=args.revision, variant=args.variant
+    )
+
+    # Add the placeholder token in tokenizer
+    placeholder_tokens = [args.placeholder_token]
+
+    if args.num_vectors < 1:
+        raise ValueError(f"--num_vectors has to be larger or equal to 1, but is {args.num_vectors}")
+
+    # add dummy tokens for multi-vector
+    additional_tokens = []
+    for i in range(1, args.num_vectors):
+        additional_tokens.append(f"{args.placeholder_token}_{i}")
+    placeholder_tokens += additional_tokens
+
+    num_added_tokens = tokenizer.add_tokens(placeholder_tokens)
+    if num_added_tokens != args.num_vectors:
+        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_ids = tokenizer.convert_tokens_to_ids(placeholder_tokens)
+
+    # 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
+    with torch.no_grad():
+        for token_id in placeholder_token_ids:
+            token_embeds[token_id] = token_embeds[initializer_token_id].clone()
+
+    # Freeze vae and unet
+    vae.requires_grad_(False)
+    unet.requires_grad_(False)
+    # Freeze all parameters except for the token embeddings in text encoder
+    text_encoder.text_model.encoder.requires_grad_(False)
+    text_encoder.text_model.final_layer_norm.requires_grad_(False)
+    text_encoder.text_model.embeddings.position_embedding.requires_grad_(False)
+
+    if args.gradient_checkpointing:
+        # Keep unet in train mode if we are using gradient checkpointing to save memory.
+        # The dropout cannot be != 0 so it doesn't matter if we are in eval or train mode.
+        unet.train()
+        text_encoder.gradient_checkpointing_enable()
+        unet.enable_gradient_checkpointing()
+
+    if args.enable_xformers_memory_efficient_attention:
+        if is_xformers_available():
+            import xformers
+
+            xformers_version = version.parse(xformers.__version__)
+            if xformers_version == version.parse("0.0.16"):
+                logger.warning(
+                    "xFormers 0.0.16 cannot be used for training in some GPUs. If you observe problems during training, please update xFormers to at least 0.0.17. See https://huggingface.co/docs/diffusers/main/en/optimization/xformers for more details."
+                )
+            unet.enable_xformers_memory_efficient_attention()
+        else:
+            raise ValueError("xformers is not available. Make sure it is installed correctly")
+
+    # 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
+        )
+
+    # 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,
+    )
+
+    # Dataset and DataLoaders creation:
+    train_dataset = TextualInversionDataset(
+        data_root=args.train_data_dir,
+        tokenizer=tokenizer,
+        size=args.resolution,
+        placeholder_token=(" ".join(tokenizer.convert_ids_to_tokens(placeholder_token_ids))),
+        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, num_workers=args.dataloader_num_workers
+    )
+    if args.validation_epochs is not None:
+        warnings.warn(
+            f"FutureWarning: You are doing logging with validation_epochs={args.validation_epochs}."
+            " Deprecated validation_epochs in favor of `validation_steps`"
+            f"Setting `args.validation_steps` to {args.validation_epochs * len(train_dataset)}",
+            FutureWarning,
+            stacklevel=2,
+        )
+        args.validation_steps = args.validation_epochs * len(train_dataset)
+
+    # 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 * accelerator.num_processes,
+        num_training_steps=args.max_train_steps * accelerator.num_processes,
+        num_cycles=args.lr_num_cycles,
+    )
+
+    text_encoder.train()
+    # Prepare everything with our `accelerator`.
+    text_encoder, optimizer, train_dataloader, lr_scheduler = accelerator.prepare(
+        text_encoder, optimizer, train_dataloader, lr_scheduler
+    )
+
+    # For mixed precision training we cast all non-trainable weigths (vae, non-lora text_encoder and non-lora unet) to half-precision
+    # as these weights 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 unet to device and cast to weight_dtype
+    unet.to(accelerator.device, dtype=weight_dtype)
+    vae.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("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}")
+    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 most 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] if len(dirs) > 0 else None
+
+        if path is None:
+            accelerator.print(
+                f"Checkpoint '{args.resume_from_checkpoint}' does not exist. Starting a new training run."
+            )
+            args.resume_from_checkpoint = None
+            initial_global_step = 0
+        else:
+            accelerator.print(f"Resuming from checkpoint {path}")
+            accelerator.load_state(os.path.join(args.output_dir, path))
+            global_step = int(path.split("-")[1])
+
+            initial_global_step = global_step
+            first_epoch = global_step // num_update_steps_per_epoch
+
+    else:
+        initial_global_step = 0
+
+    progress_bar = tqdm(
+        range(0, args.max_train_steps),
+        initial=initial_global_step,
+        desc="Steps",
+        # Only show the progress bar once on each machine.
+        disable=not accelerator.is_local_main_process,
+    )
+
+    # keep original embeddings as reference
+    orig_embeds_params = accelerator.unwrap_model(text_encoder).get_input_embeddings().weight.data.clone()
+
+    for epoch in range(first_epoch, 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"].to(dtype=weight_dtype)).latent_dist.sample().detach()
+                latents = latents * vae.config.scaling_factor
+
+                # 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].to(dtype=weight_dtype)
+
+                # 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}")
+
+                loss = F.mse_loss(model_pred.float(), target.float(), reduction="mean")
+
+                accelerator.backward(loss)
+
+                optimizer.step()
+                lr_scheduler.step()
+                optimizer.zero_grad()
+
+                # Let's make sure we don't update any embedding weights besides the newly added token
+                index_no_updates = torch.ones((len(tokenizer),), dtype=torch.bool)
+                index_no_updates[min(placeholder_token_ids) : max(placeholder_token_ids) + 1] = False
+
+                with torch.no_grad():
+                    accelerator.unwrap_model(text_encoder).get_input_embeddings().weight[
+                        index_no_updates
+                    ] = orig_embeds_params[index_no_updates]
+
+            # Checks if the accelerator has performed an optimization step behind the scenes
+            if accelerator.sync_gradients:
+                images = []
+                progress_bar.update(1)
+                global_step += 1
+                if global_step % args.save_steps == 0:
+                    weight_name = (
+                        f"learned_embeds-steps-{global_step}.bin"
+                        if args.no_safe_serialization
+                        else f"learned_embeds-steps-{global_step}.safetensors"
+                    )
+                    save_path = os.path.join(args.output_dir, weight_name)
+                    save_progress(
+                        text_encoder,
+                        placeholder_token_ids,
+                        accelerator,
+                        args,
+                        save_path,
+                        safe_serialization=not args.no_safe_serialization,
+                    )
+
+                if accelerator.is_main_process:
+                    if global_step % args.checkpointing_steps == 0:
+                        # _before_ saving state, check if this save would set us over the `checkpoints_total_limit`
+                        if args.checkpoints_total_limit is not None:
+                            checkpoints = os.listdir(args.output_dir)
+                            checkpoints = [d for d in checkpoints if d.startswith("checkpoint")]
+                            checkpoints = sorted(checkpoints, key=lambda x: int(x.split("-")[1]))
+
+                            # before we save the new checkpoint, we need to have at _most_ `checkpoints_total_limit - 1` checkpoints
+                            if len(checkpoints) >= args.checkpoints_total_limit:
+                                num_to_remove = len(checkpoints) - args.checkpoints_total_limit + 1
+                                removing_checkpoints = checkpoints[0:num_to_remove]
+
+                                logger.info(
+                                    f"{len(checkpoints)} checkpoints already exist, removing {len(removing_checkpoints)} checkpoints"
+                                )
+                                logger.info(f"removing checkpoints: {', '.join(removing_checkpoints)}")
+
+                                for removing_checkpoint in removing_checkpoints:
+                                    removing_checkpoint = os.path.join(args.output_dir, removing_checkpoint)
+                                    shutil.rmtree(removing_checkpoint)
+
+                        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}")
+
+                    if args.validation_prompt is not None and global_step % args.validation_steps == 0:
+                        images = log_validation(
+                            text_encoder, tokenizer, unet, vae, args, accelerator, weight_dtype, epoch
+                        )
+
+            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 the trained modules and save it.
+    accelerator.wait_for_everyone()
+    if accelerator.is_main_process:
+        if args.push_to_hub and not args.save_as_full_pipeline:
+            logger.warning("Enabling full model saving because --push_to_hub=True was specified.")
+            save_full_model = True
+        else:
+            save_full_model = args.save_as_full_pipeline
+        if save_full_model:
+            pipeline = StableDiffusionPipeline.from_pretrained(
+                args.pretrained_model_name_or_path,
+                text_encoder=accelerator.unwrap_model(text_encoder),
+                vae=vae,
+                unet=unet,
+                tokenizer=tokenizer,
+            )
+            pipeline.save_pretrained(args.output_dir)
+        # Save the newly trained embeddings
+        weight_name = "learned_embeds.bin" if args.no_safe_serialization else "learned_embeds.safetensors"
+        save_path = os.path.join(args.output_dir, weight_name)
+        save_progress(
+            text_encoder,
+            placeholder_token_ids,
+            accelerator,
+            args,
+            save_path,
+            safe_serialization=not args.no_safe_serialization,
+        )
+
+        if args.push_to_hub:
+            save_model_card(
+                repo_id,
+                images=images,
+                base_model=args.pretrained_model_name_or_path,
+                repo_folder=args.output_dir,
+            )
+            upload_folder(
+                repo_id=repo_id,
+                folder_path=args.output_dir,
+                commit_message="End of training",
+                ignore_patterns=["step_*", "epoch_*"],
+            )
+
+    accelerator.end_training()
+
+
+if __name__ == "__main__":
+    main()

diffusers-0.27.0/examples/textual_inversion/textual_inversion_flax.py

+import argparse
+import logging
+import math
+import os
+import random
+from pathlib import Path
+
+import jax
+import jax.numpy as jnp
+import numpy as np
+import optax
+import PIL
+import torch
+import torch.utils.checkpoint
+import transformers
+from flax import jax_utils
+from flax.training import train_state
+from flax.training.common_utils import shard
+from huggingface_hub import create_repo, upload_folder
+
+# TODO: remove and import from diffusers.utils when the new version of diffusers is released
+from packaging import version
+from PIL import Image
+from torch.utils.data import Dataset
+from torchvision import transforms
+from tqdm.auto import tqdm
+from transformers import CLIPImageProcessor, CLIPTokenizer, FlaxCLIPTextModel, set_seed
+
+from diffusers import (
+    FlaxAutoencoderKL,
+    FlaxDDPMScheduler,
+    FlaxPNDMScheduler,
+    FlaxStableDiffusionPipeline,
+    FlaxUNet2DConditionModel,
+)
+from diffusers.pipelines.stable_diffusion import FlaxStableDiffusionSafetyChecker
+from diffusers.utils import check_min_version
+
+
+if version.parse(version.parse(PIL.__version__).base_version) >= version.parse("9.1.0"):
+    PIL_INTERPOLATION = {
+        "linear": PIL.Image.Resampling.BILINEAR,
+        "bilinear": PIL.Image.Resampling.BILINEAR,
+        "bicubic": PIL.Image.Resampling.BICUBIC,
+        "lanczos": PIL.Image.Resampling.LANCZOS,
+        "nearest": PIL.Image.Resampling.NEAREST,
+    }
+else:
+    PIL_INTERPOLATION = {
+        "linear": PIL.Image.LINEAR,
+        "bilinear": PIL.Image.BILINEAR,
+        "bicubic": PIL.Image.BICUBIC,
+        "lanczos": PIL.Image.LANCZOS,
+        "nearest": PIL.Image.NEAREST,
+    }
+# ------------------------------------------------------------------------------
+
+# Will error if the minimal version of diffusers is not installed. Remove at your own risks.
+check_min_version("0.27.0")
+
+logger = logging.getLogger(__name__)
+
+
+def parse_args():
+    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(
+        "--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, required=True, help="A folder containing the training data."
+    )
+    parser.add_argument(
+        "--placeholder_token",
+        type=str,
+        default=None,
+        required=True,
+        help="A token to use as a placeholder for the concept.",
+    )
+    parser.add_argument(
+        "--initializer_token", type=str, default=None, required=True, 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=42, 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(
+        "--save_steps",
+        type=int,
+        default=500,
+        help="Save learned_embeds.bin every X updates steps.",
+    )
+    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_warmup_steps", type=int, default=500, help="Number of steps for the warmup in the lr scheduler."
+    )
+    parser.add_argument(
+        "--revision",
+        type=str,
+        default=None,
+        required=False,
+        help="Revision of pretrained model identifier from huggingface.co/models.",
+    )
+    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("--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(
+        "--use_auth_token",
+        action="store_true",
+        help=(
+            "Will use the token generated when running `huggingface-cli login` (necessary to use this script with"
+            " private models)."
+        ),
+    )
+    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("--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_INTERPOLATION["linear"],
+            "bilinear": PIL_INTERPOLATION["bilinear"],
+            "bicubic": PIL_INTERPOLATION["bicubic"],
+            "lanczos": PIL_INTERPOLATION["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 resize_token_embeddings(model, new_num_tokens, initializer_token_id, placeholder_token_id, rng):
+    if model.config.vocab_size == new_num_tokens or new_num_tokens is None:
+        return
+    model.config.vocab_size = new_num_tokens
+
+    params = model.params
+    old_embeddings = params["text_model"]["embeddings"]["token_embedding"]["embedding"]
+    old_num_tokens, emb_dim = old_embeddings.shape
+
+    initializer = jax.nn.initializers.normal()
+
+    new_embeddings = initializer(rng, (new_num_tokens, emb_dim))
+    new_embeddings = new_embeddings.at[:old_num_tokens].set(old_embeddings)
+    new_embeddings = new_embeddings.at[placeholder_token_id].set(new_embeddings[initializer_token_id])
+    params["text_model"]["embeddings"]["token_embedding"]["embedding"] = new_embeddings
+
+    model.params = params
+    return model
+
+
+def get_params_to_save(params):
+    return jax.device_get(jax.tree_util.tree_map(lambda x: x[0], params))
+
+
+def main():
+    args = parse_args()
+
+    if args.seed is not None:
+        set_seed(args.seed)
+
+    if jax.process_index() == 0:
+        if args.output_dir is not None:
+            os.makedirs(args.output_dir, exist_ok=True)
+
+        if args.push_to_hub:
+            repo_id = create_repo(
+                repo_id=args.hub_model_id or Path(args.output_dir).name, exist_ok=True, token=args.hub_token
+            ).repo_id
+
+    # 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,
+    )
+    # Setup logging, we only want one process per machine to log things on the screen.
+    logger.setLevel(logging.INFO if jax.process_index() == 0 else logging.ERROR)
+    if jax.process_index() == 0:
+        transformers.utils.logging.set_verbosity_info()
+    else:
+        transformers.utils.logging.set_verbosity_error()
+
+    # 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 = FlaxCLIPTextModel.from_pretrained(
+        args.pretrained_model_name_or_path, subfolder="text_encoder", revision=args.revision
+    )
+    vae, vae_params = FlaxAutoencoderKL.from_pretrained(
+        args.pretrained_model_name_or_path, subfolder="vae", revision=args.revision
+    )
+    unet, unet_params = FlaxUNet2DConditionModel.from_pretrained(
+        args.pretrained_model_name_or_path, subfolder="unet", revision=args.revision
+    )
+
+    # Create sampling rng
+    rng = jax.random.PRNGKey(args.seed)
+    rng, _ = jax.random.split(rng)
+    # Resize the token embeddings as we are adding new special tokens to the tokenizer
+    text_encoder = resize_token_embeddings(
+        text_encoder, len(tokenizer), initializer_token_id, placeholder_token_id, rng
+    )
+    original_token_embeds = text_encoder.params["text_model"]["embeddings"]["token_embedding"]["embedding"]
+
+    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",
+    )
+
+    def collate_fn(examples):
+        pixel_values = torch.stack([example["pixel_values"] for example in examples])
+        input_ids = torch.stack([example["input_ids"] for example in examples])
+
+        batch = {"pixel_values": pixel_values, "input_ids": input_ids}
+        batch = {k: v.numpy() for k, v in batch.items()}
+
+        return batch
+
+    total_train_batch_size = args.train_batch_size * jax.local_device_count()
+    train_dataloader = torch.utils.data.DataLoader(
+        train_dataset, batch_size=total_train_batch_size, shuffle=True, drop_last=True, collate_fn=collate_fn
+    )
+
+    # Optimization
+    if args.scale_lr:
+        args.learning_rate = args.learning_rate * total_train_batch_size
+
+    constant_scheduler = optax.constant_schedule(args.learning_rate)
+
+    optimizer = optax.adamw(
+        learning_rate=constant_scheduler,
+        b1=args.adam_beta1,
+        b2=args.adam_beta2,
+        eps=args.adam_epsilon,
+        weight_decay=args.adam_weight_decay,
+    )
+
+    def create_mask(params, label_fn):
+        def _map(params, mask, label_fn):
+            for k in params:
+                if label_fn(k):
+                    mask[k] = "token_embedding"
+                else:
+                    if isinstance(params[k], dict):
+                        mask[k] = {}
+                        _map(params[k], mask[k], label_fn)
+                    else:
+                        mask[k] = "zero"
+
+        mask = {}
+        _map(params, mask, label_fn)
+        return mask
+
+    def zero_grads():
+        # from https://github.com/deepmind/optax/issues/159#issuecomment-896459491
+        def init_fn(_):
+            return ()
+
+        def update_fn(updates, state, params=None):
+            return jax.tree_util.tree_map(jnp.zeros_like, updates), ()
+
+        return optax.GradientTransformation(init_fn, update_fn)
+
+    # Zero out gradients of layers other than the token embedding layer
+    tx = optax.multi_transform(
+        {"token_embedding": optimizer, "zero": zero_grads()},
+        create_mask(text_encoder.params, lambda s: s == "token_embedding"),
+    )
+
+    state = train_state.TrainState.create(apply_fn=text_encoder.__call__, params=text_encoder.params, tx=tx)
+
+    noise_scheduler = FlaxDDPMScheduler(
+        beta_start=0.00085, beta_end=0.012, beta_schedule="scaled_linear", num_train_timesteps=1000
+    )
+    noise_scheduler_state = noise_scheduler.create_state()
+
+    # Initialize our training
+    train_rngs = jax.random.split(rng, jax.local_device_count())
+
+    # Define gradient train step fn
+    def train_step(state, vae_params, unet_params, batch, train_rng):
+        dropout_rng, sample_rng, new_train_rng = jax.random.split(train_rng, 3)
+
+        def compute_loss(params):
+            vae_outputs = vae.apply(
+                {"params": vae_params}, batch["pixel_values"], deterministic=True, method=vae.encode
+            )
+            latents = vae_outputs.latent_dist.sample(sample_rng)
+            # (NHWC) -> (NCHW)
+            latents = jnp.transpose(latents, (0, 3, 1, 2))
+            latents = latents * vae.config.scaling_factor
+
+            noise_rng, timestep_rng = jax.random.split(sample_rng)
+            noise = jax.random.normal(noise_rng, latents.shape)
+            bsz = latents.shape[0]
+            timesteps = jax.random.randint(
+                timestep_rng,
+                (bsz,),
+                0,
+                noise_scheduler.config.num_train_timesteps,
+            )
+            noisy_latents = noise_scheduler.add_noise(noise_scheduler_state, latents, noise, timesteps)
+            encoder_hidden_states = state.apply_fn(
+                batch["input_ids"], params=params, dropout_rng=dropout_rng, train=True
+            )[0]
+            # Predict the noise residual and compute loss
+            model_pred = unet.apply(
+                {"params": unet_params}, noisy_latents, timesteps, encoder_hidden_states, train=False
+            ).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(noise_scheduler_state, latents, noise, timesteps)
+            else:
+                raise ValueError(f"Unknown prediction type {noise_scheduler.config.prediction_type}")
+
+            loss = (target - model_pred) ** 2
+            loss = loss.mean()
+
+            return loss
+
+        grad_fn = jax.value_and_grad(compute_loss)
+        loss, grad = grad_fn(state.params)
+        grad = jax.lax.pmean(grad, "batch")
+        new_state = state.apply_gradients(grads=grad)
+
+        # Keep the token embeddings fixed except the newly added embeddings for the concept,
+        # as we only want to optimize the concept embeddings
+        token_embeds = original_token_embeds.at[placeholder_token_id].set(
+            new_state.params["text_model"]["embeddings"]["token_embedding"]["embedding"][placeholder_token_id]
+        )
+        new_state.params["text_model"]["embeddings"]["token_embedding"]["embedding"] = token_embeds
+
+        metrics = {"loss": loss}
+        metrics = jax.lax.pmean(metrics, axis_name="batch")
+        return new_state, metrics, new_train_rng
+
+    # Create parallel version of the train and eval step
+    p_train_step = jax.pmap(train_step, "batch", donate_argnums=(0,))
+
+    # Replicate the train state on each device
+    state = jax_utils.replicate(state)
+    vae_params = jax_utils.replicate(vae_params)
+    unet_params = jax_utils.replicate(unet_params)
+
+    # Train!
+    num_update_steps_per_epoch = math.ceil(len(train_dataloader))
+
+    # Scheduler and math around the number of training steps.
+    if args.max_train_steps is None:
+        args.max_train_steps = args.num_train_epochs * num_update_steps_per_epoch
+
+    args.num_train_epochs = math.ceil(args.max_train_steps / num_update_steps_per_epoch)
+
+    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) = {total_train_batch_size}")
+    logger.info(f"  Total optimization steps = {args.max_train_steps}")
+
+    global_step = 0
+
+    epochs = tqdm(range(args.num_train_epochs), desc=f"Epoch ... (1/{args.num_train_epochs})", position=0)
+    for epoch in epochs:
+        # ======================== Training ================================
+
+        train_metrics = []
+
+        steps_per_epoch = len(train_dataset) // total_train_batch_size
+        train_step_progress_bar = tqdm(total=steps_per_epoch, desc="Training...", position=1, leave=False)
+        # train
+        for batch in train_dataloader:
+            batch = shard(batch)
+            state, train_metric, train_rngs = p_train_step(state, vae_params, unet_params, batch, train_rngs)
+            train_metrics.append(train_metric)
+
+            train_step_progress_bar.update(1)
+            global_step += 1
+
+            if global_step >= args.max_train_steps:
+                break
+            if global_step % args.save_steps == 0:
+                learned_embeds = get_params_to_save(state.params)["text_model"]["embeddings"]["token_embedding"][
+                    "embedding"
+                ][placeholder_token_id]
+                learned_embeds_dict = {args.placeholder_token: learned_embeds}
+                jnp.save(
+                    os.path.join(args.output_dir, "learned_embeds-" + str(global_step) + ".npy"), learned_embeds_dict
+                )
+
+        train_metric = jax_utils.unreplicate(train_metric)
+
+        train_step_progress_bar.close()
+        epochs.write(f"Epoch... ({epoch + 1}/{args.num_train_epochs} | Loss: {train_metric['loss']})")
+
+    # Create the pipeline using using the trained modules and save it.
+    if jax.process_index() == 0:
+        scheduler = FlaxPNDMScheduler(
+            beta_start=0.00085, beta_end=0.012, beta_schedule="scaled_linear", skip_prk_steps=True
+        )
+        safety_checker = FlaxStableDiffusionSafetyChecker.from_pretrained(
+            "CompVis/stable-diffusion-safety-checker", from_pt=True
+        )
+        pipeline = FlaxStableDiffusionPipeline(
+            text_encoder=text_encoder,
+            vae=vae,
+            unet=unet,
+            tokenizer=tokenizer,
+            scheduler=scheduler,
+            safety_checker=safety_checker,
+            feature_extractor=CLIPImageProcessor.from_pretrained("openai/clip-vit-base-patch32"),
+        )
+
+        pipeline.save_pretrained(
+            args.output_dir,
+            params={
+                "text_encoder": get_params_to_save(state.params),
+                "vae": get_params_to_save(vae_params),
+                "unet": get_params_to_save(unet_params),
+                "safety_checker": safety_checker.params,
+            },
+        )
+
+        # Also save the newly trained embeddings
+        learned_embeds = get_params_to_save(state.params)["text_model"]["embeddings"]["token_embedding"]["embedding"][
+            placeholder_token_id
+        ]
+        learned_embeds_dict = {args.placeholder_token: learned_embeds}
+        jnp.save(os.path.join(args.output_dir, "learned_embeds.npy"), learned_embeds_dict)
+
+        if args.push_to_hub:
+            upload_folder(
+                repo_id=repo_id,
+                folder_path=args.output_dir,
+                commit_message="End of training",
+                ignore_patterns=["step_*", "epoch_*"],
+            )
+
+
+if __name__ == "__main__":
+    main()

diffusers-0.27.0/examples/textual_inversion/textual_inversion_sdxl.py

+#!/usr/bin/env python
+# coding=utf-8
+# Copyright 2024 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+
+import argparse
+import logging
+import math
+import os
+import random
+import shutil
+from pathlib import Path
+
+import numpy as np
+import PIL
+import safetensors
+import torch
+import torch.nn.functional as F
+import torch.utils.checkpoint
+import transformers
+from accelerate import Accelerator
+from accelerate.logging import get_logger
+from accelerate.utils import ProjectConfiguration, set_seed
+from huggingface_hub import create_repo, upload_folder
+from packaging import version
+from PIL import Image
+from torch.utils.data import Dataset
+from torchvision import transforms
+from tqdm.auto import tqdm
+from transformers import CLIPTextModel, CLIPTextModelWithProjection, CLIPTokenizer
+
+import diffusers
+from diffusers import (
+    AutoencoderKL,
+    DDPMScheduler,
+    DiffusionPipeline,
+    DPMSolverMultistepScheduler,
+    UNet2DConditionModel,
+)
+from diffusers.optimization import get_scheduler
+from diffusers.utils import check_min_version, is_wandb_available
+from diffusers.utils.hub_utils import load_or_create_model_card, populate_model_card
+from diffusers.utils.import_utils import is_xformers_available
+
+
+if is_wandb_available():
+    import wandb
+
+if version.parse(version.parse(PIL.__version__).base_version) >= version.parse("9.1.0"):
+    PIL_INTERPOLATION = {
+        "linear": PIL.Image.Resampling.BILINEAR,
+        "bilinear": PIL.Image.Resampling.BILINEAR,
+        "bicubic": PIL.Image.Resampling.BICUBIC,
+        "lanczos": PIL.Image.Resampling.LANCZOS,
+        "nearest": PIL.Image.Resampling.NEAREST,
+    }
+else:
+    PIL_INTERPOLATION = {
+        "linear": PIL.Image.LINEAR,
+        "bilinear": PIL.Image.BILINEAR,
+        "bicubic": PIL.Image.BICUBIC,
+        "lanczos": PIL.Image.LANCZOS,
+        "nearest": PIL.Image.NEAREST,
+    }
+# ------------------------------------------------------------------------------
+
+
+# Will error if the minimal version of diffusers is not installed. Remove at your own risks.
+check_min_version("0.27.0")
+
+logger = get_logger(__name__)
+
+
+def save_model_card(repo_id: str, images=None, base_model=str, repo_folder=None):
+    img_str = ""
+    for i, image in enumerate(images):
+        image.save(os.path.join(repo_folder, f"image_{i}.png"))
+        img_str += f"![img_{i}](./image_{i}.png)\n"
+
+    model_description = f"""
+# Textual inversion text2image fine-tuning - {repo_id}
+These are textual inversion adaption weights for {base_model}. You can find some example images in the following. \n
+{img_str}
+"""
+    model_card = load_or_create_model_card(
+        repo_id_or_path=repo_id,
+        from_training=True,
+        license="creativeml-openrail-m",
+        base_model=base_model,
+        model_description=model_description,
+        inference=True,
+    )
+
+    tags = [
+        "stable-diffusion-xl",
+        "stable-diffusion-xl-diffusers",
+        "text-to-image",
+        "diffusers",
+        "diffusers-training",
+        "textual_inversion",
+    ]
+
+    model_card = populate_model_card(model_card, tags=tags)
+
+    model_card.save(os.path.join(repo_folder, "README.md"))
+
+
+def log_validation(
+    text_encoder_1,
+    text_encoder_2,
+    tokenizer_1,
+    tokenizer_2,
+    unet,
+    vae,
+    args,
+    accelerator,
+    weight_dtype,
+    epoch,
+    is_final_validation=False,
+):
+    logger.info(
+        f"Running validation... \n Generating {args.num_validation_images} images with prompt:"
+        f" {args.validation_prompt}."
+    )
+    pipeline = DiffusionPipeline.from_pretrained(
+        args.pretrained_model_name_or_path,
+        text_encoder=accelerator.unwrap_model(text_encoder_1),
+        text_encoder_2=text_encoder_2,
+        tokenizer=tokenizer_1,
+        tokenizer_2=tokenizer_2,
+        unet=unet,
+        vae=vae,
+        safety_checker=None,
+        revision=args.revision,
+        variant=args.variant,
+        torch_dtype=weight_dtype,
+    )
+    pipeline.scheduler = DPMSolverMultistepScheduler.from_config(pipeline.scheduler.config)
+    pipeline = pipeline.to(accelerator.device)
+    pipeline.set_progress_bar_config(disable=True)
+
+    # run inference
+    generator = None if args.seed is None else torch.Generator(device=accelerator.device).manual_seed(args.seed)
+    images = []
+    for _ in range(args.num_validation_images):
+        image = pipeline(args.validation_prompt, num_inference_steps=25, generator=generator).images[0]
+        images.append(image)
+
+    tracker_key = "test" if is_final_validation else "validation"
+    for tracker in accelerator.trackers:
+        if tracker.name == "tensorboard":
+            np_images = np.stack([np.asarray(img) for img in images])
+            tracker.writer.add_images(tracker_key, np_images, epoch, dataformats="NHWC")
+        if tracker.name == "wandb":
+            tracker.log(
+                {
+                    tracker_key: [
+                        wandb.Image(image, caption=f"{i}: {args.validation_prompt}") for i, image in enumerate(images)
+                    ]
+                }
+            )
+
+    del pipeline
+    torch.cuda.empty_cache()
+    return images
+
+
+def save_progress(text_encoder, placeholder_token_ids, accelerator, args, save_path, safe_serialization=True):
+    logger.info("Saving embeddings")
+    learned_embeds = (
+        accelerator.unwrap_model(text_encoder)
+        .get_input_embeddings()
+        .weight[min(placeholder_token_ids) : max(placeholder_token_ids) + 1]
+    )
+    learned_embeds_dict = {args.placeholder_token: learned_embeds.detach().cpu()}
+
+    if safe_serialization:
+        safetensors.torch.save_file(learned_embeds_dict, save_path, metadata={"format": "pt"})
+    else:
+        torch.save(learned_embeds_dict, save_path)
+
+
+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(
+        "--save_as_full_pipeline",
+        action="store_true",
+        help="Save the complete stable diffusion pipeline.",
+    )
+    parser.add_argument(
+        "--num_vectors",
+        type=int,
+        default=1,
+        help="How many textual inversion vectors shall be used to learn the concept.",
+    )
+    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(
+        "--variant",
+        type=str,
+        default=None,
+        help="Variant of the model files of the pretrained model identifier from huggingface.co/models, 'e.g.' fp16",
+    )
+    parser.add_argument(
+        "--train_data_dir", type=str, default=None, required=True, help="A folder containing the training data."
+    )
+    parser.add_argument(
+        "--placeholder_token",
+        type=str,
+        default=None,
+        required=True,
+        help="A token to use as a placeholder for the concept.",
+    )
+    parser.add_argument(
+        "--initializer_token", type=str, default=None, required=True, 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(
+        "--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=1e-4,
+        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(
+        "--dataloader_num_workers",
+        type=int,
+        default=0,
+        help=(
+            "Number of subprocesses to use for data loading. 0 means that the data will be loaded in the main process."
+        ),
+    )
+    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("--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(
+        "--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(
+        "--validation_prompt",
+        type=str,
+        default=None,
+        help="A prompt that is used during validation to verify that the model is learning.",
+    )
+    parser.add_argument(
+        "--num_validation_images",
+        type=int,
+        default=4,
+        help="Number of images that should be generated during validation with `validation_prompt`.",
+    )
+    parser.add_argument(
+        "--validation_steps",
+        type=int,
+        default=100,
+        help=(
+            "Run validation every X steps. Validation consists of running the prompt"
+            " `args.validation_prompt` multiple times: `args.num_validation_images`"
+            " and logging the images."
+        ),
+    )
+    parser.add_argument("--local_rank", type=int, default=-1, help="For distributed training: local_rank")
+    parser.add_argument(
+        "--checkpointing_steps",
+        type=int,
+        default=500,
+        help=(
+            "Save a checkpoint of the training state every X updates. These checkpoints are only suitable for resuming"
+            " training using `--resume_from_checkpoint`."
+        ),
+    )
+    parser.add_argument(
+        "--checkpoints_total_limit",
+        type=int,
+        default=None,
+        help=("Max number of checkpoints to store."),
+    )
+    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(
+        "--enable_xformers_memory_efficient_attention", action="store_true", help="Whether or not to use xformers."
+    )
+
+    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_1,
+        tokenizer_2,
+        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_1 = tokenizer_1
+        self.tokenizer_2 = tokenizer_2
+        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_INTERPOLATION["linear"],
+            "bilinear": PIL_INTERPOLATION["bilinear"],
+            "bicubic": PIL_INTERPOLATION["bicubic"],
+            "lanczos": PIL_INTERPOLATION["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)
+        self.crop = transforms.CenterCrop(size) if center_crop else transforms.RandomCrop(size)
+
+    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["original_size"] = (image.height, image.width)
+
+        image = image.resize((self.size, self.size), resample=self.interpolation)
+
+        if self.center_crop:
+            y1 = max(0, int(round((image.height - self.size) / 2.0)))
+            x1 = max(0, int(round((image.width - self.size) / 2.0)))
+            image = self.crop(image)
+        else:
+            y1, x1, h, w = self.crop.get_params(image, (self.size, self.size))
+            image = transforms.functional.crop(image, y1, x1, h, w)
+
+        example["crop_top_left"] = (y1, x1)
+
+        example["input_ids_1"] = self.tokenizer_1(
+            text,
+            padding="max_length",
+            truncation=True,
+            max_length=self.tokenizer_1.model_max_length,
+            return_tensors="pt",
+        ).input_ids[0]
+
+        example["input_ids_2"] = self.tokenizer_2(
+            text,
+            padding="max_length",
+            truncation=True,
+            max_length=self.tokenizer_2.model_max_length,
+            return_tensors="pt",
+        ).input_ids[0]
+
+        # default to score-sde preprocessing
+        img = np.array(image).astype(np.uint8)
+
+        image = Image.fromarray(img)
+
+        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 main():
+    args = parse_args()
+    if args.report_to == "wandb" and args.hub_token is not None:
+        raise ValueError(
+            "You cannot use both --report_to=wandb and --hub_token due to a security risk of exposing your token."
+            " Please use `huggingface-cli login` to authenticate with the Hub."
+        )
+
+    logging_dir = os.path.join(args.output_dir, args.logging_dir)
+    accelerator_project_config = ProjectConfiguration(project_dir=args.output_dir, logging_dir=logging_dir)
+    accelerator = Accelerator(
+        gradient_accumulation_steps=args.gradient_accumulation_steps,
+        mixed_precision=args.mixed_precision,
+        log_with=args.report_to,
+        project_config=accelerator_project_config,
+    )
+
+    if args.report_to == "wandb":
+        if not is_wandb_available():
+            raise ImportError("Make sure to install wandb if you want to use it for logging during training.")
+
+    # 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:
+        transformers.utils.logging.set_verbosity_warning()
+        diffusers.utils.logging.set_verbosity_info()
+    else:
+        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)
+
+    # Handle the repository creation
+    if accelerator.is_main_process:
+        if args.output_dir is not None:
+            os.makedirs(args.output_dir, exist_ok=True)
+
+        if args.push_to_hub:
+            repo_id = create_repo(
+                repo_id=args.hub_model_id or Path(args.output_dir).name, exist_ok=True, token=args.hub_token
+            ).repo_id
+
+    # Load tokenizer
+    tokenizer_1 = CLIPTokenizer.from_pretrained(args.pretrained_model_name_or_path, subfolder="tokenizer")
+    tokenizer_2 = CLIPTokenizer.from_pretrained(args.pretrained_model_name_or_path, subfolder="tokenizer_2")
+
+    # Load scheduler and models
+    noise_scheduler = DDPMScheduler.from_pretrained(args.pretrained_model_name_or_path, subfolder="scheduler")
+    text_encoder_1 = CLIPTextModel.from_pretrained(
+        args.pretrained_model_name_or_path, subfolder="text_encoder", revision=args.revision
+    )
+    text_encoder_2 = CLIPTextModelWithProjection.from_pretrained(
+        args.pretrained_model_name_or_path, subfolder="text_encoder_2", revision=args.revision
+    )
+    vae = AutoencoderKL.from_pretrained(
+        args.pretrained_model_name_or_path, subfolder="vae", revision=args.revision, variant=args.variant
+    )
+    unet = UNet2DConditionModel.from_pretrained(
+        args.pretrained_model_name_or_path, subfolder="unet", revision=args.revision, variant=args.variant
+    )
+
+    # Add the placeholder token in tokenizer_1
+    placeholder_tokens = [args.placeholder_token]
+
+    if args.num_vectors < 1:
+        raise ValueError(f"--num_vectors has to be larger or equal to 1, but is {args.num_vectors}")
+
+    # add dummy tokens for multi-vector
+    additional_tokens = []
+    for i in range(1, args.num_vectors):
+        additional_tokens.append(f"{args.placeholder_token}_{i}")
+    placeholder_tokens += additional_tokens
+
+    num_added_tokens = tokenizer_1.add_tokens(placeholder_tokens)
+    if num_added_tokens != args.num_vectors:
+        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_1.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_ids = tokenizer_1.convert_tokens_to_ids(placeholder_tokens)
+
+    # Resize the token embeddings as we are adding new special tokens to the tokenizer
+    text_encoder_1.resize_token_embeddings(len(tokenizer_1))
+
+    # Initialise the newly added placeholder token with the embeddings of the initializer token
+    token_embeds = text_encoder_1.get_input_embeddings().weight.data
+    with torch.no_grad():
+        for token_id in placeholder_token_ids:
+            token_embeds[token_id] = token_embeds[initializer_token_id].clone()
+
+    # Freeze vae and unet
+    vae.requires_grad_(False)
+    unet.requires_grad_(False)
+    text_encoder_2.requires_grad_(False)
+    # Freeze all parameters except for the token embeddings in text encoder
+    text_encoder_1.text_model.encoder.requires_grad_(False)
+    text_encoder_1.text_model.final_layer_norm.requires_grad_(False)
+    text_encoder_1.text_model.embeddings.position_embedding.requires_grad_(False)
+
+    if args.gradient_checkpointing:
+        text_encoder_1.gradient_checkpointing_enable()
+
+    if args.enable_xformers_memory_efficient_attention:
+        if is_xformers_available():
+            import xformers
+
+            xformers_version = version.parse(xformers.__version__)
+            if xformers_version == version.parse("0.0.16"):
+                logger.warning(
+                    "xFormers 0.0.16 cannot be used for training in some GPUs. If you observe problems during training, please update xFormers to at least 0.0.17. See https://huggingface.co/docs/diffusers/main/en/optimization/xformers for more details."
+                )
+            unet.enable_xformers_memory_efficient_attention()
+        else:
+            raise ValueError("xformers is not available. Make sure it is installed correctly")
+
+    # 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
+        )
+
+    # Initialize the optimizer
+    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 = optimizer_class(
+        text_encoder_1.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,
+    )
+
+    placeholder_token = " ".join(tokenizer_1.convert_ids_to_tokens(placeholder_token_ids))
+    # Dataset and DataLoaders creation:
+    train_dataset = TextualInversionDataset(
+        data_root=args.train_data_dir,
+        tokenizer_1=tokenizer_1,
+        tokenizer_2=tokenizer_2,
+        size=args.resolution,
+        placeholder_token=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, num_workers=args.dataloader_num_workers
+    )
+
+    # 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 * accelerator.num_processes,
+        num_training_steps=args.max_train_steps * accelerator.num_processes,
+        num_cycles=args.lr_num_cycles,
+    )
+
+    text_encoder_1.train()
+    # Prepare everything with our `accelerator`.
+    text_encoder_1, optimizer, train_dataloader, lr_scheduler = accelerator.prepare(
+        text_encoder_1, optimizer, train_dataloader, lr_scheduler
+    )
+
+    # For mixed precision training we cast all non-trainable weigths (vae, non-lora text_encoder and non-lora unet) to half-precision
+    # as these weights 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 unet and text_encoder_2 to device and cast to weight_dtype
+    unet.to(accelerator.device, dtype=weight_dtype)
+    vae.to(accelerator.device, dtype=weight_dtype)
+    text_encoder_2.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("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}")
+    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 most 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] if len(dirs) > 0 else None
+
+        if path is None:
+            accelerator.print(
+                f"Checkpoint '{args.resume_from_checkpoint}' does not exist. Starting a new training run."
+            )
+            args.resume_from_checkpoint = None
+            initial_global_step = 0
+        else:
+            accelerator.print(f"Resuming from checkpoint {path}")
+            accelerator.load_state(os.path.join(args.output_dir, path))
+            global_step = int(path.split("-")[1])
+
+            initial_global_step = global_step
+            first_epoch = global_step // num_update_steps_per_epoch
+
+    else:
+        initial_global_step = 0
+
+    progress_bar = tqdm(
+        range(0, args.max_train_steps),
+        initial=initial_global_step,
+        desc="Steps",
+        # Only show the progress bar once on each machine.
+        disable=not accelerator.is_local_main_process,
+    )
+
+    # keep original embeddings as reference
+    orig_embeds_params = accelerator.unwrap_model(text_encoder_1).get_input_embeddings().weight.data.clone()
+
+    for epoch in range(first_epoch, args.num_train_epochs):
+        text_encoder_1.train()
+        for step, batch in enumerate(train_dataloader):
+            with accelerator.accumulate(text_encoder_1):
+                # Convert images to latent space
+                latents = vae.encode(batch["pixel_values"].to(dtype=weight_dtype)).latent_dist.sample().detach()
+                latents = latents * vae.config.scaling_factor
+
+                # 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_1 = (
+                    text_encoder_1(batch["input_ids_1"], output_hidden_states=True)
+                    .hidden_states[-2]
+                    .to(dtype=weight_dtype)
+                )
+                encoder_output_2 = text_encoder_2(
+                    batch["input_ids_2"].reshape(batch["input_ids_1"].shape[0], -1), output_hidden_states=True
+                )
+                encoder_hidden_states_2 = encoder_output_2.hidden_states[-2].to(dtype=weight_dtype)
+                original_size = [
+                    (batch["original_size"][0][i].item(), batch["original_size"][1][i].item())
+                    for i in range(args.train_batch_size)
+                ]
+                crop_top_left = [
+                    (batch["crop_top_left"][0][i].item(), batch["crop_top_left"][1][i].item())
+                    for i in range(args.train_batch_size)
+                ]
+                target_size = (args.resolution, args.resolution)
+                add_time_ids = torch.cat(
+                    [
+                        torch.tensor(original_size[i] + crop_top_left[i] + target_size)
+                        for i in range(args.train_batch_size)
+                    ]
+                ).to(accelerator.device, dtype=weight_dtype)
+                added_cond_kwargs = {"text_embeds": encoder_output_2[0], "time_ids": add_time_ids}
+                encoder_hidden_states = torch.cat([encoder_hidden_states_1, encoder_hidden_states_2], dim=-1)
+
+                # Predict the noise residual
+                model_pred = unet(
+                    noisy_latents, timesteps, encoder_hidden_states, added_cond_kwargs=added_cond_kwargs
+                ).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}")
+
+                loss = F.mse_loss(model_pred.float(), target.float(), reduction="mean")
+
+                accelerator.backward(loss)
+
+                optimizer.step()
+                lr_scheduler.step()
+                optimizer.zero_grad()
+
+                # Let's make sure we don't update any embedding weights besides the newly added token
+                index_no_updates = torch.ones((len(tokenizer_1),), dtype=torch.bool)
+                index_no_updates[min(placeholder_token_ids) : max(placeholder_token_ids) + 1] = False
+
+                with torch.no_grad():
+                    accelerator.unwrap_model(text_encoder_1).get_input_embeddings().weight[
+                        index_no_updates
+                    ] = orig_embeds_params[index_no_updates]
+
+            # Checks if the accelerator has performed an optimization step behind the scenes
+            if accelerator.sync_gradients:
+                images = []
+                progress_bar.update(1)
+                global_step += 1
+                if global_step % args.save_steps == 0:
+                    weight_name = f"learned_embeds-steps-{global_step}.safetensors"
+                    save_path = os.path.join(args.output_dir, weight_name)
+                    save_progress(
+                        text_encoder_1,
+                        placeholder_token_ids,
+                        accelerator,
+                        args,
+                        save_path,
+                        safe_serialization=True,
+                    )
+
+                if accelerator.is_main_process:
+                    if global_step % args.checkpointing_steps == 0:
+                        # _before_ saving state, check if this save would set us over the `checkpoints_total_limit`
+                        if args.checkpoints_total_limit is not None:
+                            checkpoints = os.listdir(args.output_dir)
+                            checkpoints = [d for d in checkpoints if d.startswith("checkpoint")]
+                            checkpoints = sorted(checkpoints, key=lambda x: int(x.split("-")[1]))
+
+                            # before we save the new checkpoint, we need to have at _most_ `checkpoints_total_limit - 1` checkpoints
+                            if len(checkpoints) >= args.checkpoints_total_limit:
+                                num_to_remove = len(checkpoints) - args.checkpoints_total_limit + 1
+                                removing_checkpoints = checkpoints[0:num_to_remove]
+
+                                logger.info(
+                                    f"{len(checkpoints)} checkpoints already exist, removing {len(removing_checkpoints)} checkpoints"
+                                )
+                                logger.info(f"removing checkpoints: {', '.join(removing_checkpoints)}")
+
+                                for removing_checkpoint in removing_checkpoints:
+                                    removing_checkpoint = os.path.join(args.output_dir, removing_checkpoint)
+                                    shutil.rmtree(removing_checkpoint)
+
+                        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}")
+
+                    if args.validation_prompt is not None and global_step % args.validation_steps == 0:
+                        images = log_validation(
+                            text_encoder_1,
+                            text_encoder_2,
+                            tokenizer_1,
+                            tokenizer_2,
+                            unet,
+                            vae,
+                            args,
+                            accelerator,
+                            weight_dtype,
+                            epoch,
+                        )
+
+            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 the trained modules and save it.
+    accelerator.wait_for_everyone()
+    if accelerator.is_main_process:
+        if args.validation_prompt:
+            images = log_validation(
+                text_encoder_1,
+                text_encoder_2,
+                tokenizer_1,
+                tokenizer_2,
+                unet,
+                vae,
+                args,
+                accelerator,
+                weight_dtype,
+                epoch,
+                is_final_validation=True,
+            )
+
+        if args.push_to_hub and not args.save_as_full_pipeline:
+            logger.warning("Enabling full model saving because --push_to_hub=True was specified.")
+            save_full_model = True
+        else:
+            save_full_model = args.save_as_full_pipeline
+        if save_full_model:
+            pipeline = DiffusionPipeline.from_pretrained(
+                args.pretrained_model_name_or_path,
+                text_encoder=accelerator.unwrap_model(text_encoder_1),
+                text_encoder_2=text_encoder_2,
+                vae=vae,
+                unet=unet,
+                tokenizer=tokenizer_1,
+                tokenizer_2=tokenizer_2,
+            )
+            pipeline.save_pretrained(args.output_dir)
+        # Save the newly trained embeddings
+        weight_name = "learned_embeds.safetensors"
+        save_path = os.path.join(args.output_dir, weight_name)
+        save_progress(
+            text_encoder_1,
+            placeholder_token_ids,
+            accelerator,
+            args,
+            save_path,
+            safe_serialization=True,
+        )
+
+        if args.push_to_hub:
+            save_model_card(
+                repo_id,
+                images=images,
+                base_model=args.pretrained_model_name_or_path,
+                repo_folder=args.output_dir,
+            )
+            upload_folder(
+                repo_id=repo_id,
+                folder_path=args.output_dir,
+                commit_message="End of training",
+                ignore_patterns=["step_*", "epoch_*"],
+            )
+
+    accelerator.end_training()
+
+
+if __name__ == "__main__":
+    main()

diffusers-0.27.0/examples/unconditional_image_generation/README.md

+## Training an unconditional diffusion model
+
+Creating a training image set is [described in a different document](https://huggingface.co/docs/datasets/image_process#image-datasets).
+
+### Installing the dependencies
+
+Before running the scripts, make sure to install the library's training dependencies:
+
+**Important**
+
+To make sure you can successfully run the latest versions of the example scripts, we highly recommend **installing from source** and keeping the install up to date as we update the example scripts frequently and install some example-specific requirements. To do this, execute the following steps in a new virtual environment:
+```bash
+git clone https://github.com/huggingface/diffusers
+cd diffusers
+pip install .
+```
+
+Then cd in the example folder  and run
+```bash
+pip install -r requirements.txt
+```
+
+
+And initialize an [🤗Accelerate](https://github.com/huggingface/accelerate/) environment with:
+
+```bash
+accelerate config
+```
+
+### Unconditional Flowers
+
+The command to train a DDPM UNet model on the Oxford Flowers dataset:
+
+```bash
+accelerate launch train_unconditional.py \
+  --dataset_name="huggan/flowers-102-categories" \
+  --resolution=64 --center_crop --random_flip \
+  --output_dir="ddpm-ema-flowers-64" \
+  --train_batch_size=16 \
+  --num_epochs=100 \
+  --gradient_accumulation_steps=1 \
+  --use_ema \
+  --learning_rate=1e-4 \
+  --lr_warmup_steps=500 \
+  --mixed_precision=no \
+  --push_to_hub
+```
+An example trained model: https://huggingface.co/anton-l/ddpm-ema-flowers-64
+
+A full training run takes 2 hours on 4xV100 GPUs.
+
+<img src="https://user-images.githubusercontent.com/26864830/180248660-a0b143d0-b89a-42c5-8656-2ebf6ece7e52.png" width="700" />
+
+
+### Unconditional Pokemon
+
+The command to train a DDPM UNet model on the Pokemon dataset:
+
+```bash
+accelerate launch train_unconditional.py \
+  --dataset_name="huggan/pokemon" \
+  --resolution=64 --center_crop --random_flip \
+  --output_dir="ddpm-ema-pokemon-64" \
+  --train_batch_size=16 \
+  --num_epochs=100 \
+  --gradient_accumulation_steps=1 \
+  --use_ema \
+  --learning_rate=1e-4 \
+  --lr_warmup_steps=500 \
+  --mixed_precision=no \
+  --push_to_hub
+```
+An example trained model: https://huggingface.co/anton-l/ddpm-ema-pokemon-64
+
+A full training run takes 2 hours on 4xV100 GPUs.
+
+<img src="https://user-images.githubusercontent.com/26864830/180248200-928953b4-db38-48db-b0c6-8b740fe6786f.png" width="700" />
+
+### Training with multiple GPUs
+
+`accelerate` allows for seamless multi-GPU training. Follow the instructions [here](https://huggingface.co/docs/accelerate/basic_tutorials/launch)
+for running distributed training with `accelerate`. Here is an example command:
+
+```bash
+accelerate launch --mixed_precision="fp16" --multi_gpu train_unconditional.py \
+  --dataset_name="huggan/pokemon" \
+  --resolution=64 --center_crop --random_flip \
+  --output_dir="ddpm-ema-pokemon-64" \
+  --train_batch_size=16 \
+  --num_epochs=100 \
+  --gradient_accumulation_steps=1 \
+  --use_ema \
+  --learning_rate=1e-4 \
+  --lr_warmup_steps=500 \
+  --mixed_precision="fp16" \
+  --logger="wandb"
+```
+
+To be able to use Weights and Biases (`wandb`) as a logger you need to install the library: `pip install wandb`.
+
+### Using your own data
+
+To use your own dataset, there are 2 ways:
+- you can either provide your own folder as `--train_data_dir`
+- or you can upload your dataset to the hub (possibly as a private repo, if you prefer so), and simply pass the `--dataset_name` argument.
+
+Below, we explain both in more detail.
+
+#### Provide the dataset as a folder
+
+If you provide your own folders with images, the script expects the following directory structure:
+
+```bash
+data_dir/xxx.png
+data_dir/xxy.png
+data_dir/[...]/xxz.png
+```
+
+In other words, the script will take care of gathering all images inside the folder. You can then run the script like this:
+
+```bash
+accelerate launch train_unconditional.py \
+    --train_data_dir <path-to-train-directory> \
+    <other-arguments>
+```
+
+Internally, the script will use the [`ImageFolder`](https://huggingface.co/docs/datasets/v2.0.0/en/image_process#imagefolder) feature which will automatically turn the folders into 🤗 Dataset objects.
+
+#### Upload your data to the hub, as a (possibly private) repo
+
+It's very easy (and convenient) to upload your image dataset to the hub using the [`ImageFolder`](https://huggingface.co/docs/datasets/v2.0.0/en/image_process#imagefolder) feature available in 🤗 Datasets. Simply do the following:
+
+```python
+from datasets import load_dataset
+
+# example 1: local folder
+dataset = load_dataset("imagefolder", data_dir="path_to_your_folder")
+
+# example 2: local files (supported formats are tar, gzip, zip, xz, rar, zstd)
+dataset = load_dataset("imagefolder", data_files="path_to_zip_file")
+
+# example 3: remote files (supported formats are tar, gzip, zip, xz, rar, zstd)
+dataset = load_dataset("imagefolder", data_files="https://download.microsoft.com/download/3/E/1/3E1C3F21-ECDB-4869-8368-6DEBA77B919F/kagglecatsanddogs_3367a.zip")
+
+# example 4: providing several splits
+dataset = load_dataset("imagefolder", data_files={"train": ["path/to/file1", "path/to/file2"], "test": ["path/to/file3", "path/to/file4"]})
+```
+
+`ImageFolder` will create an `image` column containing the PIL-encoded images.
+
+Next, push it to the hub!
+
+```python
+# assuming you have ran the huggingface-cli login command in a terminal
+dataset.push_to_hub("name_of_your_dataset")
+
+# if you want to push to a private repo, simply pass private=True:
+dataset.push_to_hub("name_of_your_dataset", private=True)
+```
+
+and that's it! You can now train your model by simply setting the `--dataset_name` argument to the name of your dataset on the hub.
+
+More on this can also be found in [this blog post](https://huggingface.co/blog/image-search-datasets).

diffusers-0.27.0/examples/unconditional_image_generation/requirements.txt

+accelerate>=0.16.0
+torchvision
+datasets

diffusers-0.27.0/examples/unconditional_image_generation/test_unconditional.py

+# coding=utf-8
+# Copyright 2024 HuggingFace Inc.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import logging
+import os
+import sys
+import tempfile
+
+
+sys.path.append("..")
+from test_examples_utils import ExamplesTestsAccelerate, run_command  # noqa: E402
+
+
+logging.basicConfig(level=logging.DEBUG)
+
+logger = logging.getLogger()
+stream_handler = logging.StreamHandler(sys.stdout)
+logger.addHandler(stream_handler)
+
+
+class Unconditional(ExamplesTestsAccelerate):
+    def test_train_unconditional(self):
+        with tempfile.TemporaryDirectory() as tmpdir:
+            test_args = f"""
+                examples/unconditional_image_generation/train_unconditional.py
+                --dataset_name hf-internal-testing/dummy_image_class_data
+                --model_config_name_or_path diffusers/ddpm_dummy
+                --resolution 64
+                --output_dir {tmpdir}
+                --train_batch_size 2
+                --num_epochs 1
+                --gradient_accumulation_steps 1
+                --ddpm_num_inference_steps 2
+                --learning_rate 1e-3
+                --lr_warmup_steps 5
+                """.split()
+
+            run_command(self._launch_args + test_args, return_stdout=True)
+            # save_pretrained smoke test
+            self.assertTrue(os.path.isfile(os.path.join(tmpdir, "unet", "diffusion_pytorch_model.safetensors")))
+            self.assertTrue(os.path.isfile(os.path.join(tmpdir, "scheduler", "scheduler_config.json")))
+
+    def test_unconditional_checkpointing_checkpoints_total_limit(self):
+        with tempfile.TemporaryDirectory() as tmpdir:
+            initial_run_args = f"""
+                examples/unconditional_image_generation/train_unconditional.py
+                --dataset_name hf-internal-testing/dummy_image_class_data
+                --model_config_name_or_path diffusers/ddpm_dummy
+                --resolution 64
+                --output_dir {tmpdir}
+                --train_batch_size 1
+                --num_epochs 1
+                --gradient_accumulation_steps 1
+                --ddpm_num_inference_steps 2
+                --learning_rate 1e-3
+                --lr_warmup_steps 5
+                --checkpointing_steps=2
+                --checkpoints_total_limit=2
+                """.split()
+
+            run_command(self._launch_args + initial_run_args)
+
+            # check checkpoint directories exist
+            self.assertEqual(
+                {x for x in os.listdir(tmpdir) if "checkpoint" in x},
+                # checkpoint-2 should have been deleted
+                {"checkpoint-4", "checkpoint-6"},
+            )
+
+    def test_unconditional_checkpointing_checkpoints_total_limit_removes_multiple_checkpoints(self):
+        with tempfile.TemporaryDirectory() as tmpdir:
+            initial_run_args = f"""
+                examples/unconditional_image_generation/train_unconditional.py
+                --dataset_name hf-internal-testing/dummy_image_class_data
+                --model_config_name_or_path diffusers/ddpm_dummy
+                --resolution 64
+                --output_dir {tmpdir}
+                --train_batch_size 1
+                --num_epochs 1
+                --gradient_accumulation_steps 1
+                --ddpm_num_inference_steps 1
+                --learning_rate 1e-3
+                --lr_warmup_steps 5
+                --checkpointing_steps=2
+                """.split()
+
+            run_command(self._launch_args + initial_run_args)
+
+            # check checkpoint directories exist
+            self.assertEqual(
+                {x for x in os.listdir(tmpdir) if "checkpoint" in x},
+                {"checkpoint-2", "checkpoint-4", "checkpoint-6"},
+            )
+
+            resume_run_args = f"""
+                examples/unconditional_image_generation/train_unconditional.py
+                --dataset_name hf-internal-testing/dummy_image_class_data
+                --model_config_name_or_path diffusers/ddpm_dummy
+                --resolution 64
+                --output_dir {tmpdir}
+                --train_batch_size 1
+                --num_epochs 2
+                --gradient_accumulation_steps 1
+                --ddpm_num_inference_steps 1
+                --learning_rate 1e-3
+                --lr_warmup_steps 5
+                --resume_from_checkpoint=checkpoint-6
+                --checkpointing_steps=2
+                --checkpoints_total_limit=2
+                """.split()
+
+            run_command(self._launch_args + resume_run_args)
+
+            # check checkpoint directories exist
+            self.assertEqual(
+                {x for x in os.listdir(tmpdir) if "checkpoint" in x},
+                {"checkpoint-10", "checkpoint-12"},
+            )

diffusers-0.27.0/examples/unconditional_image_generation/train_unconditional.py

+import argparse
+import inspect
+import logging
+import math
+import os
+import shutil
+from datetime import timedelta
+from pathlib import Path
+
+import accelerate
+import datasets
+import torch
+import torch.nn.functional as F
+from accelerate import Accelerator, InitProcessGroupKwargs
+from accelerate.logging import get_logger
+from accelerate.utils import ProjectConfiguration
+from datasets import load_dataset
+from huggingface_hub import create_repo, upload_folder
+from packaging import version
+from torchvision import transforms
+from tqdm.auto import tqdm
+
+import diffusers
+from diffusers import DDPMPipeline, DDPMScheduler, UNet2DModel
+from diffusers.optimization import get_scheduler
+from diffusers.training_utils import EMAModel
+from diffusers.utils import check_min_version, is_accelerate_version, is_tensorboard_available, is_wandb_available
+from diffusers.utils.import_utils import is_xformers_available
+
+
+# Will error if the minimal version of diffusers is not installed. Remove at your own risks.
+check_min_version("0.27.0")
+
+logger = get_logger(__name__, log_level="INFO")
+
+
+def _extract_into_tensor(arr, timesteps, broadcast_shape):
+    """
+    Extract values from a 1-D numpy array for a batch of indices.
+
+    :param arr: the 1-D numpy array.
+    :param timesteps: a tensor of indices into the array to extract.
+    :param broadcast_shape: a larger shape of K dimensions with the batch
+                            dimension equal to the length of timesteps.
+    :return: a tensor of shape [batch_size, 1, ...] where the shape has K dims.
+    """
+    if not isinstance(arr, torch.Tensor):
+        arr = torch.from_numpy(arr)
+    res = arr[timesteps].float().to(timesteps.device)
+    while len(res.shape) < len(broadcast_shape):
+        res = res[..., None]
+    return res.expand(broadcast_shape)
+
+
+def parse_args():
+    parser = argparse.ArgumentParser(description="Simple example of a training script.")
+    parser.add_argument(
+        "--dataset_name",
+        type=str,
+        default=None,
+        help=(
+            "The name of the Dataset (from the HuggingFace hub) to train on (could be your own, possibly private,"
+            " dataset). It can also be a path pointing to a local copy of a dataset in your filesystem,"
+            " or to a folder containing files that HF Datasets can understand."
+        ),
+    )
+    parser.add_argument(
+        "--dataset_config_name",
+        type=str,
+        default=None,
+        help="The config of the Dataset, leave as None if there's only one config.",
+    )
+    parser.add_argument(
+        "--model_config_name_or_path",
+        type=str,
+        default=None,
+        help="The config of the UNet model to train, leave as None to use standard DDPM configuration.",
+    )
+    parser.add_argument(
+        "--train_data_dir",
+        type=str,
+        default=None,
+        help=(
+            "A folder containing the training data. Folder contents must follow the structure described in"
+            " https://huggingface.co/docs/datasets/image_dataset#imagefolder. In particular, a `metadata.jsonl` file"
+            " must exist to provide the captions for the images. Ignored if `dataset_name` is specified."
+        ),
+    )
+    parser.add_argument(
+        "--output_dir",
+        type=str,
+        default="ddpm-model-64",
+        help="The output directory where the model predictions and checkpoints will be written.",
+    )
+    parser.add_argument("--overwrite_output_dir", action="store_true")
+    parser.add_argument(
+        "--cache_dir",
+        type=str,
+        default=None,
+        help="The directory where the downloaded models and datasets will be stored.",
+    )
+    parser.add_argument(
+        "--resolution",
+        type=int,
+        default=64,
+        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",
+        default=False,
+        action="store_true",
+        help=(
+            "Whether to center crop the input images to the resolution. If not set, the images will be randomly"
+            " cropped. The images will be resized to the resolution first before cropping."
+        ),
+    )
+    parser.add_argument(
+        "--random_flip",
+        default=False,
+        action="store_true",
+        help="whether to randomly flip images horizontally",
+    )
+    parser.add_argument(
+        "--train_batch_size", type=int, default=16, help="Batch size (per device) for the training dataloader."
+    )
+    parser.add_argument(
+        "--eval_batch_size", type=int, default=16, help="The number of images to generate for evaluation."
+    )
+    parser.add_argument(
+        "--dataloader_num_workers",
+        type=int,
+        default=0,
+        help=(
+            "The number of subprocesses to use for data loading. 0 means that the data will be loaded in the main"
+            " process."
+        ),
+    )
+    parser.add_argument("--num_epochs", type=int, default=100)
+    parser.add_argument("--save_images_epochs", type=int, default=10, help="How often to save images during training.")
+    parser.add_argument(
+        "--save_model_epochs", type=int, default=10, help="How often to save the model during training."
+    )
+    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(
+        "--lr_scheduler",
+        type=str,
+        default="cosine",
+        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=0.95, 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-6, help="Weight decay magnitude for the Adam optimizer."
+    )
+    parser.add_argument("--adam_epsilon", type=float, default=1e-08, help="Epsilon value for the Adam optimizer.")
+    parser.add_argument(
+        "--use_ema",
+        action="store_true",
+        help="Whether to use Exponential Moving Average for the final model weights.",
+    )
+    parser.add_argument("--ema_inv_gamma", type=float, default=1.0, help="The inverse gamma value for the EMA decay.")
+    parser.add_argument("--ema_power", type=float, default=3 / 4, help="The power value for the EMA decay.")
+    parser.add_argument("--ema_max_decay", type=float, default=0.9999, help="The maximum decay magnitude for EMA.")
+    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(
+        "--hub_private_repo", action="store_true", help="Whether or not to create a private repository."
+    )
+    parser.add_argument(
+        "--logger",
+        type=str,
+        default="tensorboard",
+        choices=["tensorboard", "wandb"],
+        help=(
+            "Whether to use [tensorboard](https://www.tensorflow.org/tensorboard) or [wandb](https://www.wandb.ai)"
+            " for experiment tracking and logging of model metrics and model checkpoints"
+        ),
+    )
+    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("--local_rank", type=int, default=-1, help="For distributed training: local_rank")
+    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(
+        "--prediction_type",
+        type=str,
+        default="epsilon",
+        choices=["epsilon", "sample"],
+        help="Whether the model should predict the 'epsilon'/noise error or directly the reconstructed image 'x0'.",
+    )
+    parser.add_argument("--ddpm_num_steps", type=int, default=1000)
+    parser.add_argument("--ddpm_num_inference_steps", type=int, default=1000)
+    parser.add_argument("--ddpm_beta_schedule", type=str, default="linear")
+    parser.add_argument(
+        "--checkpointing_steps",
+        type=int,
+        default=500,
+        help=(
+            "Save a checkpoint of the training state every X updates. These checkpoints are only suitable for resuming"
+            " training using `--resume_from_checkpoint`."
+        ),
+    )
+    parser.add_argument(
+        "--checkpoints_total_limit",
+        type=int,
+        default=None,
+        help=("Max number of checkpoints to store."),
+    )
+    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(
+        "--enable_xformers_memory_efficient_attention", action="store_true", help="Whether or not to use xformers."
+    )
+
+    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.dataset_name is None and args.train_data_dir is None:
+        raise ValueError("You must specify either a dataset name from the hub or a train data directory.")
+
+    return args
+
+
+def main(args):
+    logging_dir = os.path.join(args.output_dir, args.logging_dir)
+    accelerator_project_config = ProjectConfiguration(project_dir=args.output_dir, logging_dir=logging_dir)
+
+    kwargs = InitProcessGroupKwargs(timeout=timedelta(seconds=7200))  # a big number for high resolution or big dataset
+    accelerator = Accelerator(
+        gradient_accumulation_steps=args.gradient_accumulation_steps,
+        mixed_precision=args.mixed_precision,
+        log_with=args.logger,
+        project_config=accelerator_project_config,
+        kwargs_handlers=[kwargs],
+    )
+
+    if args.logger == "tensorboard":
+        if not is_tensorboard_available():
+            raise ImportError("Make sure to install tensorboard if you want to use it for logging during training.")
+
+    elif args.logger == "wandb":
+        if not is_wandb_available():
+            raise ImportError("Make sure to install wandb if you want to use it for logging during training.")
+        import wandb
+
+    # `accelerate` 0.16.0 will have better support for customized saving
+    if version.parse(accelerate.__version__) >= version.parse("0.16.0"):
+        # create custom saving & loading hooks so that `accelerator.save_state(...)` serializes in a nice format
+        def save_model_hook(models, weights, output_dir):
+            if accelerator.is_main_process:
+                if args.use_ema:
+                    ema_model.save_pretrained(os.path.join(output_dir, "unet_ema"))
+
+                for i, model in enumerate(models):
+                    model.save_pretrained(os.path.join(output_dir, "unet"))
+
+                    # make sure to pop weight so that corresponding model is not saved again
+                    weights.pop()
+
+        def load_model_hook(models, input_dir):
+            if args.use_ema:
+                load_model = EMAModel.from_pretrained(os.path.join(input_dir, "unet_ema"), UNet2DModel)
+                ema_model.load_state_dict(load_model.state_dict())
+                ema_model.to(accelerator.device)
+                del load_model
+
+            for i in range(len(models)):
+                # pop models so that they are not loaded again
+                model = models.pop()
+
+                # load diffusers style into model
+                load_model = UNet2DModel.from_pretrained(input_dir, subfolder="unet")
+                model.register_to_config(**load_model.config)
+
+                model.load_state_dict(load_model.state_dict())
+                del load_model
+
+        accelerator.register_save_state_pre_hook(save_model_hook)
+        accelerator.register_load_state_pre_hook(load_model_hook)
+
+    # 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()
+        diffusers.utils.logging.set_verbosity_info()
+    else:
+        datasets.utils.logging.set_verbosity_error()
+        diffusers.utils.logging.set_verbosity_error()
+
+    # Handle the repository creation
+    if accelerator.is_main_process:
+        if args.output_dir is not None:
+            os.makedirs(args.output_dir, exist_ok=True)
+
+        if args.push_to_hub:
+            repo_id = create_repo(
+                repo_id=args.hub_model_id or Path(args.output_dir).name, exist_ok=True, token=args.hub_token
+            ).repo_id
+
+    # Initialize the model
+    if args.model_config_name_or_path is None:
+        model = UNet2DModel(
+            sample_size=args.resolution,
+            in_channels=3,
+            out_channels=3,
+            layers_per_block=2,
+            block_out_channels=(128, 128, 256, 256, 512, 512),
+            down_block_types=(
+                "DownBlock2D",
+                "DownBlock2D",
+                "DownBlock2D",
+                "DownBlock2D",
+                "AttnDownBlock2D",
+                "DownBlock2D",
+            ),
+            up_block_types=(
+                "UpBlock2D",
+                "AttnUpBlock2D",
+                "UpBlock2D",
+                "UpBlock2D",
+                "UpBlock2D",
+                "UpBlock2D",
+            ),
+        )
+    else:
+        config = UNet2DModel.load_config(args.model_config_name_or_path)
+        model = UNet2DModel.from_config(config)
+
+    # Create EMA for the model.
+    if args.use_ema:
+        ema_model = EMAModel(
+            model.parameters(),
+            decay=args.ema_max_decay,
+            use_ema_warmup=True,
+            inv_gamma=args.ema_inv_gamma,
+            power=args.ema_power,
+            model_cls=UNet2DModel,
+            model_config=model.config,
+        )
+
+    weight_dtype = torch.float32
+    if accelerator.mixed_precision == "fp16":
+        weight_dtype = torch.float16
+        args.mixed_precision = accelerator.mixed_precision
+    elif accelerator.mixed_precision == "bf16":
+        weight_dtype = torch.bfloat16
+        args.mixed_precision = accelerator.mixed_precision
+
+    if args.enable_xformers_memory_efficient_attention:
+        if is_xformers_available():
+            import xformers
+
+            xformers_version = version.parse(xformers.__version__)
+            if xformers_version == version.parse("0.0.16"):
+                logger.warning(
+                    "xFormers 0.0.16 cannot be used for training in some GPUs. If you observe problems during training, please update xFormers to at least 0.0.17. See https://huggingface.co/docs/diffusers/main/en/optimization/xformers for more details."
+                )
+            model.enable_xformers_memory_efficient_attention()
+        else:
+            raise ValueError("xformers is not available. Make sure it is installed correctly")
+
+    # Initialize the scheduler
+    accepts_prediction_type = "prediction_type" in set(inspect.signature(DDPMScheduler.__init__).parameters.keys())
+    if accepts_prediction_type:
+        noise_scheduler = DDPMScheduler(
+            num_train_timesteps=args.ddpm_num_steps,
+            beta_schedule=args.ddpm_beta_schedule,
+            prediction_type=args.prediction_type,
+        )
+    else:
+        noise_scheduler = DDPMScheduler(num_train_timesteps=args.ddpm_num_steps, beta_schedule=args.ddpm_beta_schedule)
+
+    # Initialize the optimizer
+    optimizer = torch.optim.AdamW(
+        model.parameters(),
+        lr=args.learning_rate,
+        betas=(args.adam_beta1, args.adam_beta2),
+        weight_decay=args.adam_weight_decay,
+        eps=args.adam_epsilon,
+    )
+
+    # Get the datasets: you can either provide your own training and evaluation files (see below)
+    # or specify a Dataset from the hub (the dataset will be downloaded automatically from the datasets Hub).
+
+    # In distributed training, the load_dataset function guarantees that only one local process can concurrently
+    # download the dataset.
+    if args.dataset_name is not None:
+        dataset = load_dataset(
+            args.dataset_name,
+            args.dataset_config_name,
+            cache_dir=args.cache_dir,
+            split="train",
+        )
+    else:
+        dataset = load_dataset("imagefolder", data_dir=args.train_data_dir, cache_dir=args.cache_dir, split="train")
+        # See more about loading custom images at
+        # https://huggingface.co/docs/datasets/v2.4.0/en/image_load#imagefolder
+
+    # Preprocessing the datasets and DataLoaders creation.
+    augmentations = transforms.Compose(
+        [
+            transforms.Resize(args.resolution, interpolation=transforms.InterpolationMode.BILINEAR),
+            transforms.CenterCrop(args.resolution) if args.center_crop else transforms.RandomCrop(args.resolution),
+            transforms.RandomHorizontalFlip() if args.random_flip else transforms.Lambda(lambda x: x),
+            transforms.ToTensor(),
+            transforms.Normalize([0.5], [0.5]),
+        ]
+    )
+
+    def transform_images(examples):
+        images = [augmentations(image.convert("RGB")) for image in examples["image"]]
+        return {"input": images}
+
+    logger.info(f"Dataset size: {len(dataset)}")
+
+    dataset.set_transform(transform_images)
+    train_dataloader = torch.utils.data.DataLoader(
+        dataset, batch_size=args.train_batch_size, shuffle=True, num_workers=args.dataloader_num_workers
+    )
+
+    # Initialize the learning rate scheduler
+    lr_scheduler = get_scheduler(
+        args.lr_scheduler,
+        optimizer=optimizer,
+        num_warmup_steps=args.lr_warmup_steps * args.gradient_accumulation_steps,
+        num_training_steps=(len(train_dataloader) * args.num_epochs),
+    )
+
+    # Prepare everything with our `accelerator`.
+    model, optimizer, train_dataloader, lr_scheduler = accelerator.prepare(
+        model, optimizer, train_dataloader, lr_scheduler
+    )
+
+    if args.use_ema:
+        ema_model.to(accelerator.device)
+
+    # 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:
+        run = os.path.split(__file__)[-1].split(".")[0]
+        accelerator.init_trackers(run)
+
+    total_batch_size = args.train_batch_size * accelerator.num_processes * args.gradient_accumulation_steps
+    num_update_steps_per_epoch = math.ceil(len(train_dataloader) / args.gradient_accumulation_steps)
+    max_train_steps = args.num_epochs * num_update_steps_per_epoch
+
+    logger.info("***** Running training *****")
+    logger.info(f"  Num examples = {len(dataset)}")
+    logger.info(f"  Num Epochs = {args.num_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 = {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 most 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] if len(dirs) > 0 else None
+
+        if path is None:
+            accelerator.print(
+                f"Checkpoint '{args.resume_from_checkpoint}' does not exist. Starting a new training run."
+            )
+            args.resume_from_checkpoint = None
+        else:
+            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 = global_step // num_update_steps_per_epoch
+            resume_step = resume_global_step % (num_update_steps_per_epoch * args.gradient_accumulation_steps)
+
+    # Train!
+    for epoch in range(first_epoch, args.num_epochs):
+        model.train()
+        progress_bar = tqdm(total=num_update_steps_per_epoch, disable=not accelerator.is_local_main_process)
+        progress_bar.set_description(f"Epoch {epoch}")
+        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
+
+            clean_images = batch["input"].to(weight_dtype)
+            # Sample noise that we'll add to the images
+            noise = torch.randn(clean_images.shape, dtype=weight_dtype, device=clean_images.device)
+            bsz = clean_images.shape[0]
+            # Sample a random timestep for each image
+            timesteps = torch.randint(
+                0, noise_scheduler.config.num_train_timesteps, (bsz,), device=clean_images.device
+            ).long()
+
+            # Add noise to the clean images according to the noise magnitude at each timestep
+            # (this is the forward diffusion process)
+            noisy_images = noise_scheduler.add_noise(clean_images, noise, timesteps)
+
+            with accelerator.accumulate(model):
+                # Predict the noise residual
+                model_output = model(noisy_images, timesteps).sample
+
+                if args.prediction_type == "epsilon":
+                    loss = F.mse_loss(model_output.float(), noise.float())  # this could have different weights!
+                elif args.prediction_type == "sample":
+                    alpha_t = _extract_into_tensor(
+                        noise_scheduler.alphas_cumprod, timesteps, (clean_images.shape[0], 1, 1, 1)
+                    )
+                    snr_weights = alpha_t / (1 - alpha_t)
+                    # use SNR weighting from distillation paper
+                    loss = snr_weights * F.mse_loss(model_output.float(), clean_images.float(), reduction="none")
+                    loss = loss.mean()
+                else:
+                    raise ValueError(f"Unsupported prediction type: {args.prediction_type}")
+
+                accelerator.backward(loss)
+
+                if accelerator.sync_gradients:
+                    accelerator.clip_grad_norm_(model.parameters(), 1.0)
+                optimizer.step()
+                lr_scheduler.step()
+                optimizer.zero_grad()
+
+            # Checks if the accelerator has performed an optimization step behind the scenes
+            if accelerator.sync_gradients:
+                if args.use_ema:
+                    ema_model.step(model.parameters())
+                progress_bar.update(1)
+                global_step += 1
+
+                if accelerator.is_main_process:
+                    if global_step % args.checkpointing_steps == 0:
+                        # _before_ saving state, check if this save would set us over the `checkpoints_total_limit`
+                        if args.checkpoints_total_limit is not None:
+                            checkpoints = os.listdir(args.output_dir)
+                            checkpoints = [d for d in checkpoints if d.startswith("checkpoint")]
+                            checkpoints = sorted(checkpoints, key=lambda x: int(x.split("-")[1]))
+
+                            # before we save the new checkpoint, we need to have at _most_ `checkpoints_total_limit - 1` checkpoints
+                            if len(checkpoints) >= args.checkpoints_total_limit:
+                                num_to_remove = len(checkpoints) - args.checkpoints_total_limit + 1
+                                removing_checkpoints = checkpoints[0:num_to_remove]
+
+                                logger.info(
+                                    f"{len(checkpoints)} checkpoints already exist, removing {len(removing_checkpoints)} checkpoints"
+                                )
+                                logger.info(f"removing checkpoints: {', '.join(removing_checkpoints)}")
+
+                                for removing_checkpoint in removing_checkpoints:
+                                    removing_checkpoint = os.path.join(args.output_dir, removing_checkpoint)
+                                    shutil.rmtree(removing_checkpoint)
+
+                        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], "step": global_step}
+            if args.use_ema:
+                logs["ema_decay"] = ema_model.cur_decay_value
+            progress_bar.set_postfix(**logs)
+            accelerator.log(logs, step=global_step)
+        progress_bar.close()
+
+        accelerator.wait_for_everyone()
+
+        # Generate sample images for visual inspection
+        if accelerator.is_main_process:
+            if epoch % args.save_images_epochs == 0 or epoch == args.num_epochs - 1:
+                unet = accelerator.unwrap_model(model)
+
+                if args.use_ema:
+                    ema_model.store(unet.parameters())
+                    ema_model.copy_to(unet.parameters())
+
+                pipeline = DDPMPipeline(
+                    unet=unet,
+                    scheduler=noise_scheduler,
+                )
+
+                generator = torch.Generator(device=pipeline.device).manual_seed(0)
+                # run pipeline in inference (sample random noise and denoise)
+                images = pipeline(
+                    generator=generator,
+                    batch_size=args.eval_batch_size,
+                    num_inference_steps=args.ddpm_num_inference_steps,
+                    output_type="numpy",
+                ).images
+
+                if args.use_ema:
+                    ema_model.restore(unet.parameters())
+
+                # denormalize the images and save to tensorboard
+                images_processed = (images * 255).round().astype("uint8")
+
+                if args.logger == "tensorboard":
+                    if is_accelerate_version(">=", "0.17.0.dev0"):
+                        tracker = accelerator.get_tracker("tensorboard", unwrap=True)
+                    else:
+                        tracker = accelerator.get_tracker("tensorboard")
+                    tracker.add_images("test_samples", images_processed.transpose(0, 3, 1, 2), epoch)
+                elif args.logger == "wandb":
+                    # Upcoming `log_images` helper coming in https://github.com/huggingface/accelerate/pull/962/files
+                    accelerator.get_tracker("wandb").log(
+                        {"test_samples": [wandb.Image(img) for img in images_processed], "epoch": epoch},
+                        step=global_step,
+                    )
+
+            if epoch % args.save_model_epochs == 0 or epoch == args.num_epochs - 1:
+                # save the model
+                unet = accelerator.unwrap_model(model)
+
+                if args.use_ema:
+                    ema_model.store(unet.parameters())
+                    ema_model.copy_to(unet.parameters())
+
+                pipeline = DDPMPipeline(
+                    unet=unet,
+                    scheduler=noise_scheduler,
+                )
+
+                pipeline.save_pretrained(args.output_dir)
+
+                if args.use_ema:
+                    ema_model.restore(unet.parameters())
+
+                if args.push_to_hub:
+                    upload_folder(
+                        repo_id=repo_id,
+                        folder_path=args.output_dir,
+                        commit_message=f"Epoch {epoch}",
+                        ignore_patterns=["step_*", "epoch_*"],
+                    )
+
+    accelerator.end_training()
+
+
+if __name__ == "__main__":
+    args = parse_args()
+    main(args)

diffusers-0.27.0/examples/wuerstchen/text_to_image/README.md

+# Würstchen text-to-image fine-tuning
+
+## Running locally with PyTorch
+
+Before running the scripts, make sure to install the library's training dependencies:
+
+**Important**
+
+To make sure you can successfully run the latest versions of the example scripts, we highly recommend **installing from source** and keeping the install up to date. To do this, execute the following steps in a new virtual environment:
+```bash
+git clone https://github.com/huggingface/diffusers
+cd diffusers
+pip install .
+```
+
+Then cd into the example folder and run
+```bash
+cd examples/wuerstchen/text_to_image
+pip install -r requirements.txt
+```
+
+And initialize an [🤗Accelerate](https://github.com/huggingface/accelerate/) environment with:
+
+```bash
+accelerate config
+```
+For this example we want to directly store the trained LoRA embeddings on the Hub, so we need to be logged in and add the `--push_to_hub` flag to the training script. To log in, run:
+```bash
+huggingface-cli login
+```
+
+## Prior training
+
+You can fine-tune the Würstchen prior model with the `train_text_to_image_prior.py` script. Note that we currently support `--gradient_checkpointing` for prior model fine-tuning so you can use it for more GPU memory constrained setups.
+
+<br>
+
+<!-- accelerate_snippet_start -->
+```bash
+export DATASET_NAME="lambdalabs/pokemon-blip-captions"
+
+accelerate launch  train_text_to_image_prior.py \
+  --mixed_precision="fp16" \
+  --dataset_name=$DATASET_NAME \
+  --resolution=768 \
+  --train_batch_size=4 \
+  --gradient_accumulation_steps=4 \
+  --gradient_checkpointing \
+  --dataloader_num_workers=4 \
+  --max_train_steps=15000 \
+  --learning_rate=1e-05 \
+  --max_grad_norm=1 \
+  --checkpoints_total_limit=3 \
+  --lr_scheduler="constant" --lr_warmup_steps=0 \
+  --validation_prompts="A robot pokemon, 4k photo" \
+  --report_to="wandb" \
+  --push_to_hub \
+  --output_dir="wuerstchen-prior-pokemon-model"
+```
+<!-- accelerate_snippet_end -->
+
+## Training with LoRA
+
+Low-Rank Adaption of Large Language Models (or LoRA) was first introduced by Microsoft in [LoRA: Low-Rank Adaptation of Large Language Models](https://arxiv.org/abs/2106.09685) by *Edward J. Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, Weizhu Chen*.
+
+In a nutshell, LoRA allows adapting pretrained models by adding pairs of rank-decomposition matrices to existing weights and **only** training those newly added weights. This has a couple of advantages:
+
+- Previous pretrained weights are kept frozen so that the model is not prone to [catastrophic forgetting](https://www.pnas.org/doi/10.1073/pnas.1611835114).
+- Rank-decomposition matrices have significantly fewer parameters than original model, which means that trained LoRA weights are easily portable.
+- LoRA attention layers allow to control to which extent the model is adapted toward new training images via a `scale` parameter.
+
+
+### Prior Training
+
+First, you need to set up your development environment as explained in the [installation](#Running-locally-with-PyTorch) section. Make sure to set the `DATASET_NAME` environment variable. Here, we will use the [Pokemon captions dataset](https://huggingface.co/datasets/lambdalabs/pokemon-blip-captions).
+
+```bash
+export DATASET_NAME="lambdalabs/pokemon-blip-captions"
+
+accelerate launch train_text_to_image_lora_prior.py \
+  --mixed_precision="fp16" \
+  --dataset_name=$DATASET_NAME --caption_column="text" \
+  --resolution=768 \
+  --train_batch_size=8 \
+  --num_train_epochs=100 --checkpointing_steps=5000 \
+  --learning_rate=1e-04 --lr_scheduler="constant" --lr_warmup_steps=0 \
+  --seed=42 \
+  --rank=4 \
+  --validation_prompt="cute dragon creature" \
+  --report_to="wandb" \
+  --push_to_hub \
+  --output_dir="wuerstchen-prior-pokemon-lora"
+```

diffusers-0.27.0/examples/wuerstchen/text_to_image/modeling_efficient_net_encoder.py

+import torch.nn as nn
+from torchvision.models import efficientnet_v2_l, efficientnet_v2_s
+
+from diffusers.configuration_utils import ConfigMixin, register_to_config
+from diffusers.models.modeling_utils import ModelMixin
+
+
+class EfficientNetEncoder(ModelMixin, ConfigMixin):
+    @register_to_config
+    def __init__(self, c_latent=16, c_cond=1280, effnet="efficientnet_v2_s"):
+        super().__init__()
+
+        if effnet == "efficientnet_v2_s":
+            self.backbone = efficientnet_v2_s(weights="DEFAULT").features
+        else:
+            self.backbone = efficientnet_v2_l(weights="DEFAULT").features
+        self.mapper = nn.Sequential(
+            nn.Conv2d(c_cond, c_latent, kernel_size=1, bias=False),
+            nn.BatchNorm2d(c_latent),  # then normalize them to have mean 0 and std 1
+        )
+
+    def forward(self, x):
+        return self.mapper(self.backbone(x))