Upload folder using huggingface_hub

README.md

@@ -0,0 +1,339 @@
+# Stable Diffusion text-to-image fine-tuning
+
+The `train_text_to_image.py` script shows how to fine-tune stable diffusion model on your own dataset.
+
+___Note___:
+
+___This script is experimental. The script fine-tunes the whole model and often times the model overfits and runs into issues like catastrophic forgetting. It's recommended to try different hyperparameters to get the best result on your dataset.___
+
+
+## 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
+```
+
+Note also that we use PEFT library as backend for LoRA training, make sure to have `peft>=0.6.0` installed in your environment.
+
+### Naruto example
+
+You need to accept the model license before downloading or using the weights. In this example we'll use model version `v1-4`, so you'll need to visit [its card](https://huggingface.co/CompVis/stable-diffusion-v1-4), read the license and tick the checkbox if you agree.
+
+You have to be a registered user in 🤗 Hugging Face Hub, and you'll also need to use an access token for the code to work. For more information on access tokens, please refer to [this section of the documentation](https://huggingface.co/docs/hub/security-tokens).
+
+Run the following command to authenticate your token
+
+```bash
+huggingface-cli login
+```
+
+If you have already cloned the repo, then you won't need to go through these steps.
+
+<br>
+
+#### Hardware
+With `gradient_checkpointing` and `mixed_precision` it should be possible to fine tune the model on a single 24GB GPU. For higher `batch_size` and faster training it's better to use GPUs with >30GB memory.
+
+**___Note: Change the `resolution` to 768 if you are using the [stable-diffusion-2](https://huggingface.co/stabilityai/stable-diffusion-2) 768x768 model.___**
+<!-- accelerate_snippet_start -->
+```bash
+export MODEL_NAME="CompVis/stable-diffusion-v1-4"
+export DATASET_NAME="lambdalabs/naruto-blip-captions"
+
+accelerate launch --mixed_precision="fp16"  train_text_to_image.py \
+  --pretrained_model_name_or_path=$MODEL_NAME \
+  --dataset_name=$DATASET_NAME \
+  --use_ema \
+  --resolution=512 --center_crop --random_flip \
+  --train_batch_size=1 \
+  --gradient_accumulation_steps=4 \
+  --gradient_checkpointing \
+  --max_train_steps=15000 \
+  --learning_rate=1e-05 \
+  --max_grad_norm=1 \
+  --lr_scheduler="constant" --lr_warmup_steps=0 \
+  --output_dir="sd-naruto-model"
+```
+<!-- accelerate_snippet_end -->
+
+
+To run on your own training files prepare the dataset according to the format required by `datasets`, you can find the instructions for how to do that in this [document](https://huggingface.co/docs/datasets/v2.4.0/en/image_load#imagefolder-with-metadata).
+If you wish to use custom loading logic, you should modify the script, we have left pointers for that in the training script.
+
+```bash
+export MODEL_NAME="CompVis/stable-diffusion-v1-4"
+export TRAIN_DIR="path_to_your_dataset"
+
+accelerate launch --mixed_precision="fp16" train_text_to_image.py \
+  --pretrained_model_name_or_path=$MODEL_NAME \
+  --train_data_dir=$TRAIN_DIR \
+  --use_ema \
+  --resolution=512 --center_crop --random_flip \
+  --train_batch_size=1 \
+  --gradient_accumulation_steps=4 \
+  --gradient_checkpointing \
+  --max_train_steps=15000 \
+  --learning_rate=1e-05 \
+  --max_grad_norm=1 \
+  --lr_scheduler="constant" --lr_warmup_steps=0 \
+  --output_dir="sd-naruto-model"
+```
+
+
+Once the training is finished the model will be saved in the `output_dir` specified in the command. In this example it's `sd-naruto-model`. To load the fine-tuned model for inference just pass that path to `StableDiffusionPipeline`
+
+```python
+import torch
+from diffusers import StableDiffusionPipeline
+
+model_path = "path_to_saved_model"
+pipe = StableDiffusionPipeline.from_pretrained(model_path, torch_dtype=torch.float16)
+pipe.to("cuda")
+
+image = pipe(prompt="yoda").images[0]
+image.save("yoda-naruto.png")
+```
+
+Checkpoints only save the unet, so to run inference from a checkpoint, just load the unet
+
+```python
+import torch
+from diffusers import StableDiffusionPipeline, UNet2DConditionModel
+
+model_path = "path_to_saved_model"
+unet = UNet2DConditionModel.from_pretrained(model_path + "/checkpoint-<N>/unet", torch_dtype=torch.float16)
+
+pipe = StableDiffusionPipeline.from_pretrained("<initial model>", unet=unet, torch_dtype=torch.float16)
+pipe.to("cuda")
+
+image = pipe(prompt="yoda").images[0]
+image.save("yoda-naruto.png")
+```
+
+#### 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
+export MODEL_NAME="CompVis/stable-diffusion-v1-4"
+export DATASET_NAME="lambdalabs/naruto-blip-captions"
+
+accelerate launch --mixed_precision="fp16" --multi_gpu  train_text_to_image.py \
+  --pretrained_model_name_or_path=$MODEL_NAME \
+  --dataset_name=$DATASET_NAME \
+  --use_ema \
+  --resolution=512 --center_crop --random_flip \
+  --train_batch_size=1 \
+  --gradient_accumulation_steps=4 \
+  --gradient_checkpointing \
+  --max_train_steps=15000 \
+  --learning_rate=1e-05 \
+  --max_grad_norm=1 \
+  --lr_scheduler="constant" --lr_warmup_steps=0 \
+  --output_dir="sd-naruto-model"
+```
+
+
+#### Training with Min-SNR weighting
+
+We support training with the Min-SNR weighting strategy proposed in [Efficient Diffusion Training via Min-SNR Weighting Strategy](https://arxiv.org/abs/2303.09556) which helps to achieve faster convergence
+by rebalancing the loss. In order to use it, one needs to set the `--snr_gamma` argument. The recommended
+value when using it is 5.0.
+
+You can find [this project on Weights and Biases](https://wandb.ai/sayakpaul/text2image-finetune-minsnr) that compares the loss surfaces of the following setups:
+
+* Training without the Min-SNR weighting strategy
+* Training with the Min-SNR weighting strategy (`snr_gamma` set to 5.0)
+* Training with the Min-SNR weighting strategy (`snr_gamma` set to 1.0)
+
+For our small Narutos dataset, the effects of Min-SNR weighting strategy might not appear to be pronounced, but for larger datasets, we believe the effects will be more pronounced.
+
+Also, note that in this example, we either predict `epsilon` (i.e., the noise) or the `v_prediction`. For both of these cases, the formulation of the Min-SNR weighting strategy that we have used holds.
+
+
+#### Training with EMA weights
+
+Through the `EMAModel` class, we support a convenient method of tracking an exponential moving average of model parameters.  This helps to smooth out noise in model parameter updates and generally improves model performance.  If enabled with the `--use_ema` argument, the final model checkpoint that is saved at the end of training will use the EMA weights.
+
+EMA weights require an additional full-precision copy of the model parameters to be stored in memory, but otherwise have very little performance overhead.  `--foreach_ema` can be used to further reduce the overhead.  If you are short on VRAM and still want to use EMA weights, you can store them in CPU RAM by using the `--offload_ema` argument.  This will keep the EMA weights in pinned CPU memory during the training step.  Then, once every model parameter update, it will transfer the EMA weights back to the GPU which can then update the parameters on the GPU, before sending them back to the CPU.  Both of these transfers are set up as non-blocking, so CUDA devices should be able to overlap this transfer with other computations.  With sufficient bandwidth between the host and device and a sufficiently long gap between model parameter updates, storing EMA weights in CPU RAM should have no additional performance overhead, as long as no other calls force synchronization.
+
+#### Training with DREAM
+
+We support training epsilon (noise) prediction models using the [DREAM (Diffusion Rectification and Estimation-Adaptive Models) strategy](https://arxiv.org/abs/2312.00210). DREAM claims to increase model fidelity for the performance cost of an extra grad-less unet `forward` step in the training loop.  You can turn on DREAM training by using the `--dream_training` argument. The `--dream_detail_preservation` argument controls the detail preservation variable p and is the default of 1 from the paper.
+
+
+
+## Training with LoRA
+
+Low-Rank Adaption of Large Language Models 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 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.
+
+[cloneofsimo](https://github.com/cloneofsimo) was the first to try out LoRA training for Stable Diffusion in the popular [lora](https://github.com/cloneofsimo/lora) GitHub repository.
+
+With LoRA, it's possible to fine-tune Stable Diffusion on a custom image-caption pair dataset
+on consumer GPUs like Tesla T4, Tesla V100.
+
+### Training
+
+First, you need to set up your development environment as is explained in the [installation section](#installing-the-dependencies). Make sure to set the `MODEL_NAME` and `DATASET_NAME` environment variables. Here, we will use [Stable Diffusion v1-4](https://hf.co/CompVis/stable-diffusion-v1-4) and the [Narutos dataset](https://huggingface.co/datasets/lambdalabs/naruto-blip-captions).
+
+**___Note: Change the `resolution` to 768 if you are using the [stable-diffusion-2](https://huggingface.co/stabilityai/stable-diffusion-2) 768x768 model.___**
+
+**___Note: It is quite useful to monitor the training progress by regularly generating sample images during training. [Weights and Biases](https://docs.wandb.ai/quickstart) is a nice solution to easily see generating images during training. All you need to do is to run `pip install wandb` before training to automatically log images.___**
+
+```bash
+export MODEL_NAME="CompVis/stable-diffusion-v1-4"
+export DATASET_NAME="lambdalabs/naruto-blip-captions"
+```
+
+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.
+
+```bash
+huggingface-cli login
+```
+
+Now we can start training!
+
+```bash
+accelerate launch --mixed_precision="fp16" train_text_to_image_lora.py \
+  --pretrained_model_name_or_path=$MODEL_NAME \
+  --dataset_name=$DATASET_NAME --caption_column="text" \
+  --resolution=512 --random_flip \
+  --train_batch_size=1 \
+  --num_train_epochs=100 --checkpointing_steps=5000 \
+  --learning_rate=1e-04 --lr_scheduler="constant" --lr_warmup_steps=0 \
+  --seed=42 \
+  --output_dir="sd-naruto-model-lora" \
+  --validation_prompt="cute dragon creature" --report_to="wandb"
+```
+
+The above command will also run inference as fine-tuning progresses and log the results to Weights and Biases.
+
+**___Note: When using LoRA we can use a much higher learning rate compared to non-LoRA fine-tuning. Here we use *1e-4* instead of the usual *1e-5*. Also, by using LoRA, it's possible to run `train_text_to_image_lora.py` in consumer GPUs like T4 or V100.___**
+
+The final LoRA embedding weights have been uploaded to [sayakpaul/sd-model-finetuned-lora-t4](https://huggingface.co/sayakpaul/sd-model-finetuned-lora-t4). **___Note: [The final weights](https://huggingface.co/sayakpaul/sd-model-finetuned-lora-t4/blob/main/pytorch_lora_weights.bin) are only 3 MB in size, which is orders of magnitudes smaller than the original model.___**
+
+You can check some inference samples that were logged during the course of the fine-tuning process [here](https://wandb.ai/sayakpaul/text2image-fine-tune/runs/q4lc0xsw).
+
+### Inference
+
+Once you have trained a model using above command, the inference can be done simply using the `StableDiffusionPipeline` after loading the trained LoRA weights.  You
+need to pass the `output_dir` for loading the LoRA weights which, in this case, is `sd-naruto-model-lora`.
+
+```python
+from diffusers import StableDiffusionPipeline
+import torch
+
+model_path = "sayakpaul/sd-model-finetuned-lora-t4"
+pipe = StableDiffusionPipeline.from_pretrained("CompVis/stable-diffusion-v1-4", torch_dtype=torch.float16)
+pipe.unet.load_attn_procs(model_path)
+pipe.to("cuda")
+
+prompt = "A naruto with green eyes and red legs."
+image = pipe(prompt, num_inference_steps=30, guidance_scale=7.5).images[0]
+image.save("naruto.png")
+```
+
+If you are loading the LoRA parameters from the Hub and if the Hub repository has
+a `base_model` tag (such as [this](https://huggingface.co/sayakpaul/sd-model-finetuned-lora-t4/blob/main/README.md?code=true#L4)), then
+you can do:
+
+```py
+from huggingface_hub.repocard import RepoCard
+
+lora_model_id = "sayakpaul/sd-model-finetuned-lora-t4"
+card = RepoCard.load(lora_model_id)
+base_model_id = card.data.to_dict()["base_model"]
+
+pipe = StableDiffusionPipeline.from_pretrained(base_model_id, torch_dtype=torch.float16)
+...
+```
+
+## 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.
+
+**___Note: The flax example doesn't yet support features like gradient checkpoint, gradient accumulation etc, so to use flax for faster training we will need >30GB cards or TPU v3.___**
+
+
+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 DATASET_NAME="lambdalabs/naruto-blip-captions"
+
+python train_text_to_image_flax.py \
+  --pretrained_model_name_or_path=$MODEL_NAME \
+  --dataset_name=$DATASET_NAME \
+  --resolution=512 --center_crop --random_flip \
+  --train_batch_size=1 \
+  --mixed_precision="fp16" \
+  --max_train_steps=15000 \
+  --learning_rate=1e-05 \
+  --max_grad_norm=1 \
+  --output_dir="sd-naruto-model"
+```
+
+To run on your own training files prepare the dataset according to the format required by `datasets`, you can find the instructions for how to do that in this [document](https://huggingface.co/docs/datasets/v2.4.0/en/image_load#imagefolder-with-metadata).
+If you wish to use custom loading logic, you should modify the script, we have left pointers for that in the training script.
+
+```bash
+export MODEL_NAME="duongna/stable-diffusion-v1-4-flax"
+export TRAIN_DIR="path_to_your_dataset"
+
+python train_text_to_image_flax.py \
+  --pretrained_model_name_or_path=$MODEL_NAME \
+  --train_data_dir=$TRAIN_DIR \
+  --resolution=512 --center_crop --random_flip \
+  --train_batch_size=1 \
+  --mixed_precision="fp16" \
+  --max_train_steps=15000 \
+  --learning_rate=1e-05 \
+  --max_grad_norm=1 \
+  --output_dir="sd-naruto-model"
+```
+
+### Training with xFormers:
+
+You can enable memory efficient attention by [installing xFormers](https://huggingface.co/docs/diffusers/main/en/optimization/xformers) and passing the `--enable_xformers_memory_efficient_attention` argument to the script.
+
+xFormers training is not available for Flax/JAX.
+
+**Note**:
+
+According to [this issue](https://github.com/huggingface/diffusers/issues/2234#issuecomment-1416931212), xFormers `v0.0.16` cannot be used for training in some GPUs. If you observe that problem, please install a development version as indicated in that comment.
+
+## Stable Diffusion XL
+
+* We support fine-tuning the UNet shipped in [Stable Diffusion XL](https://huggingface.co/papers/2307.01952) via the `train_text_to_image_sdxl.py` script. Please refer to the docs [here](./README_sdxl.md).
+* We also support fine-tuning of the UNet and Text Encoder shipped in [Stable Diffusion XL](https://huggingface.co/papers/2307.01952) with LoRA via the `train_text_to_image_lora_sdxl.py` script. Please refer to the docs [here](./README_sdxl.md).

README_sdxl.md

@@ -0,0 +1,285 @@
+# Stable Diffusion XL text-to-image fine-tuning
+
+The `train_text_to_image_sdxl.py` script shows how to fine-tune Stable Diffusion XL (SDXL) on your own dataset.
+
+🚨 This script is experimental. The script fine-tunes the whole model and often times the model overfits and runs into issues like catastrophic forgetting. It's recommended to try different hyperparameters to get the best result on your dataset. 🚨
+
+## 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 -e .
+```
+
+Then cd in the `examples/text_to_image` folder and run
+```bash
+pip install -r requirements_sdxl.txt
+```
+
+And initialize an [🤗Accelerate](https://github.com/huggingface/accelerate/) environment with:
+
+```bash
+accelerate config
+```
+
+Or for a default accelerate configuration without answering questions about your environment
+
+```bash
+accelerate config default
+```
+
+Or if your environment doesn't support an interactive shell (e.g., a notebook)
+
+```python
+from accelerate.utils import write_basic_config
+write_basic_config()
+```
+
+When running `accelerate config`, if we specify torch compile mode to True there can be dramatic speedups.
+Note also that we use PEFT library as backend for LoRA training, make sure to have `peft>=0.6.0` installed in your environment.
+
+### Training
+
+```bash
+export MODEL_NAME="stabilityai/stable-diffusion-xl-base-1.0"
+export VAE_NAME="madebyollin/sdxl-vae-fp16-fix"
+export DATASET_NAME="lambdalabs/naruto-blip-captions"
+
+accelerate launch train_text_to_image_sdxl.py \
+  --pretrained_model_name_or_path=$MODEL_NAME \
+  --pretrained_vae_model_name_or_path=$VAE_NAME \
+  --dataset_name=$DATASET_NAME \
+  --enable_xformers_memory_efficient_attention \
+  --resolution=512 --center_crop --random_flip \
+  --proportion_empty_prompts=0.2 \
+  --train_batch_size=1 \
+  --gradient_accumulation_steps=4 --gradient_checkpointing \
+  --max_train_steps=10000 \
+  --use_8bit_adam \
+  --learning_rate=1e-06 --lr_scheduler="constant" --lr_warmup_steps=0 \
+  --mixed_precision="fp16" \
+  --report_to="wandb" \
+  --validation_prompt="a cute Sundar Pichai creature" --validation_epochs 5 \
+  --checkpointing_steps=5000 \
+  --output_dir="sdxl-naruto-model" \
+  --push_to_hub
+```
+
+**Notes**:
+
+*  The `train_text_to_image_sdxl.py` script pre-computes text embeddings and the VAE encodings and keeps them in memory. While for smaller datasets like [`lambdalabs/naruto-blip-captions`](https://hf.co/datasets/lambdalabs/naruto-blip-captions), it might not be a problem, it can definitely lead to memory problems when the script is used on a larger dataset. For those purposes, you would want to serialize these pre-computed representations to disk separately and load them during the fine-tuning process. Refer to [this PR](https://github.com/huggingface/diffusers/pull/4505) for a more in-depth discussion.
+* The training script is compute-intensive and may not run on a consumer GPU like Tesla T4.
+* The training command shown above performs intermediate quality validation in between the training epochs and logs the results to Weights and Biases. `--report_to`, `--validation_prompt`, and `--validation_epochs` are the relevant CLI arguments here.
+* SDXL's VAE is known to suffer from numerical instability issues. This is why we also expose a CLI argument namely `--pretrained_vae_model_name_or_path` that lets you specify the location of a better VAE (such as [this one](https://huggingface.co/madebyollin/sdxl-vae-fp16-fix)).
+
+### Inference
+
+```python
+from diffusers import DiffusionPipeline
+import torch
+
+model_path = "you-model-id-goes-here" # <-- change this
+pipe = DiffusionPipeline.from_pretrained(model_path, torch_dtype=torch.float16)
+pipe.to("cuda")
+
+prompt = "A naruto with green eyes and red legs."
+image = pipe(prompt, num_inference_steps=30, guidance_scale=7.5).images[0]
+image.save("naruto.png")
+```
+
+### Inference in Pytorch XLA
+```python
+from diffusers import DiffusionPipeline
+import torch
+import torch_xla.core.xla_model as xm
+
+model_id = "stabilityai/stable-diffusion-xl-base-1.0"
+pipe = DiffusionPipeline.from_pretrained(model_id)
+
+device = xm.xla_device()
+pipe.to(device)
+
+prompt = "A naruto with green eyes and red legs."
+start = time()
+image = pipe(prompt, num_inference_steps=inference_steps).images[0]
+print(f'Compilation time is {time()-start} sec')
+image.save("naruto.png")
+
+start = time()
+image = pipe(prompt, num_inference_steps=inference_steps).images[0]
+print(f'Inference time is {time()-start} sec after compilation')
+```
+
+Note: There is a warmup step in PyTorch XLA. This takes longer because of
+compilation and optimization. To see the real benefits of Pytorch XLA and
+speedup, we need to call the pipe again on the input with the same length
+as the original prompt to reuse the optimized graph and get the performance
+boost.
+
+## LoRA training example for Stable Diffusion XL (SDXL)
+
+Low-Rank Adaption of Large Language Models 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 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.
+
+[cloneofsimo](https://github.com/cloneofsimo) was the first to try out LoRA training for Stable Diffusion in the popular [lora](https://github.com/cloneofsimo/lora) GitHub repository.
+
+With LoRA, it's possible to fine-tune Stable Diffusion on a custom image-caption pair dataset
+on consumer GPUs like Tesla T4, Tesla V100.
+
+### Training
+
+First, you need to set up your development environment as is explained in the [installation section](#installing-the-dependencies). Make sure to set the `MODEL_NAME` and `DATASET_NAME` environment variables and, optionally, the `VAE_NAME` variable. Here, we will use [Stable Diffusion XL 1.0-base](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0) and the [Narutos dataset](https://huggingface.co/datasets/lambdalabs/naruto-blip-captions).
+
+**___Note: It is quite useful to monitor the training progress by regularly generating sample images during training. [Weights and Biases](https://docs.wandb.ai/quickstart) is a nice solution to easily see generating images during training. All you need to do is to run `pip install wandb` before training to automatically log images.___**
+
+```bash
+export MODEL_NAME="stabilityai/stable-diffusion-xl-base-1.0"
+export VAE_NAME="madebyollin/sdxl-vae-fp16-fix"
+export DATASET_NAME="lambdalabs/naruto-blip-captions"
+```
+
+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.
+
+```bash
+huggingface-cli login
+```
+
+Now we can start training!
+
+```bash
+accelerate launch train_text_to_image_lora_sdxl.py \
+  --pretrained_model_name_or_path=$MODEL_NAME \
+  --pretrained_vae_model_name_or_path=$VAE_NAME \
+  --dataset_name=$DATASET_NAME --caption_column="text" \
+  --resolution=1024 --random_flip \
+  --train_batch_size=1 \
+  --num_train_epochs=2 --checkpointing_steps=500 \
+  --learning_rate=1e-04 --lr_scheduler="constant" --lr_warmup_steps=0 \
+  --mixed_precision="fp16" \
+  --seed=42 \
+  --output_dir="sd-naruto-model-lora-sdxl" \
+  --validation_prompt="cute dragon creature" --report_to="wandb" \
+  --push_to_hub
+```
+
+The above command will also run inference as fine-tuning progresses and log the results to Weights and Biases.
+
+**Notes**:
+
+* SDXL's VAE is known to suffer from numerical instability issues. This is why we also expose a CLI argument namely `--pretrained_vae_model_name_or_path` that lets you specify the location of a better VAE (such as [this one](https://huggingface.co/madebyollin/sdxl-vae-fp16-fix)).
+
+
+### Using DeepSpeed
+Using DeepSpeed one can reduce the consumption of GPU memory, enabling the training of models on GPUs with smaller memory sizes. DeepSpeed is capable of offloading model parameters to the machine's memory, or it can distribute parameters, gradients, and optimizer states across multiple GPUs. This allows for the training of larger models under the same hardware configuration.
+
+First, you need to use the `accelerate config` command to choose to use DeepSpeed, or manually use the accelerate config file to set up DeepSpeed.
+
+Here is an example of a config file for using DeepSpeed. For more detailed explanations of the configuration, you can refer to this [link](https://huggingface.co/docs/accelerate/usage_guides/deepspeed).
+```yaml
+compute_environment: LOCAL_MACHINE
+debug: true
+deepspeed_config:
+  gradient_accumulation_steps: 1
+  gradient_clipping: 1.0
+  offload_optimizer_device: none
+  offload_param_device: none
+  zero3_init_flag: false
+  zero_stage: 2
+distributed_type: DEEPSPEED
+downcast_bf16: 'no'
+machine_rank: 0
+main_training_function: main
+mixed_precision: fp16
+num_machines: 1
+num_processes: 1
+rdzv_backend: static
+same_network: true
+tpu_env: []
+tpu_use_cluster: false
+tpu_use_sudo: false
+use_cpu: false
+```
+You need to save the mentioned configuration as an `accelerate_config.yaml` file. Then, you need to input the path of your `accelerate_config.yaml` file into the `ACCELERATE_CONFIG_FILE` parameter. This way you can use DeepSpeed to train your SDXL model in LoRA. Additionally, you can use DeepSpeed to train other SD models in this way.
+
+```shell
+export MODEL_NAME="stabilityai/stable-diffusion-xl-base-1.0"
+export VAE_NAME="madebyollin/sdxl-vae-fp16-fix"
+export DATASET_NAME="lambdalabs/naruto-blip-captions"
+export ACCELERATE_CONFIG_FILE="your accelerate_config.yaml"
+
+accelerate launch  --config_file $ACCELERATE_CONFIG_FILE train_text_to_image_lora_sdxl.py \
+  --pretrained_model_name_or_path=$MODEL_NAME \
+  --pretrained_vae_model_name_or_path=$VAE_NAME \
+  --dataset_name=$DATASET_NAME --caption_column="text" \
+  --resolution=1024  \
+  --train_batch_size=1 \
+  --num_train_epochs=2 \
+  --checkpointing_steps=2 \
+  --learning_rate=1e-04 \
+  --lr_scheduler="constant" \
+  --lr_warmup_steps=0 \
+  --mixed_precision="fp16" \
+  --max_train_steps=20 \
+  --validation_epochs=20 \
+  --seed=1234 \
+  --output_dir="sd-naruto-model-lora-sdxl" \
+  --validation_prompt="cute dragon creature"
+```
+
+
+### Finetuning the text encoder and UNet
+
+The script also allows you to finetune the `text_encoder` along with the `unet`.
+
+🚨 Training the text encoder requires additional memory.
+
+Pass the `--train_text_encoder` argument to the training script to enable finetuning the `text_encoder` and `unet`:
+
+```bash
+accelerate launch train_text_to_image_lora_sdxl.py \
+  --pretrained_model_name_or_path=$MODEL_NAME \
+  --dataset_name=$DATASET_NAME --caption_column="text" \
+  --resolution=1024 --random_flip \
+  --train_batch_size=1 \
+  --num_train_epochs=2 --checkpointing_steps=500 \
+  --learning_rate=1e-04 --lr_scheduler="constant" --lr_warmup_steps=0 \
+  --seed=42 \
+  --output_dir="sd-naruto-model-lora-sdxl-txt" \
+  --train_text_encoder \
+  --validation_prompt="cute dragon creature" --report_to="wandb" \
+  --push_to_hub
+```
+
+### Inference
+
+Once you have trained a model using above command, the inference can be done simply using the `DiffusionPipeline` after loading the trained LoRA weights.  You
+need to pass the `output_dir` for loading the LoRA weights which, in this case, is `sd-naruto-model-lora-sdxl`.
+
+```python
+from diffusers import DiffusionPipeline
+import torch
+
+model_path = "takuoko/sd-naruto-model-lora-sdxl"
+pipe = DiffusionPipeline.from_pretrained("stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16)
+pipe.to("cuda")
+pipe.load_lora_weights(model_path)
+
+prompt = "A naruto with green eyes and red legs."
+image = pipe(prompt, num_inference_steps=30, guidance_scale=7.5).images[0]
+image.save("naruto.png")
+```

requirements.txt

@@ -0,0 +1,8 @@
+accelerate>=0.16.0
+torchvision
+transformers>=4.25.1
+datasets>=2.19.1
+ftfy
+tensorboard
+Jinja2
+peft==0.7.0

requirements_flax.txt

@@ -0,0 +1,9 @@
+transformers>=4.25.1
+datasets
+flax
+optax
+torch
+torchvision
+ftfy
+tensorboard
+Jinja2

requirements_sdxl.txt

@@ -0,0 +1,8 @@
+accelerate>=0.22.0
+torchvision
+transformers>=4.25.1
+ftfy
+tensorboard
+Jinja2
+datasets
+peft==0.7.0

test_text_to_image.py

@@ -0,0 +1,365 @@
+#!/usr/bin/env python
+# coding=utf-8
+# Copyright 2025 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
+# limitations under the License.
+
+import logging
+import os
+import shutil
+import sys
+import tempfile
+
+from diffusers import DiffusionPipeline, UNet2DConditionModel  # noqa: E402
+
+
+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 TextToImage(ExamplesTestsAccelerate):
+    def test_text_to_image(self):
+        with tempfile.TemporaryDirectory() as tmpdir:
+            test_args = f"""
+                examples/text_to_image/train_text_to_image.py
+                --pretrained_model_name_or_path hf-internal-testing/tiny-stable-diffusion-pipe
+                --dataset_name hf-internal-testing/dummy_image_text_data
+                --resolution 64
+                --center_crop
+                --random_flip
+                --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, "unet", "diffusion_pytorch_model.safetensors")))
+            self.assertTrue(os.path.isfile(os.path.join(tmpdir, "scheduler", "scheduler_config.json")))
+
+    def test_text_to_image_checkpointing(self):
+        pretrained_model_name_or_path = "hf-internal-testing/tiny-stable-diffusion-pipe"
+        prompt = "a prompt"
+
+        with tempfile.TemporaryDirectory() as tmpdir:
+            # Run training script with checkpointing
+            # max_train_steps == 4, checkpointing_steps == 2
+            # Should create checkpoints at steps 2, 4
+
+            initial_run_args = f"""
+                examples/text_to_image/train_text_to_image.py
+                --pretrained_model_name_or_path {pretrained_model_name_or_path}
+                --dataset_name hf-internal-testing/dummy_image_text_data
+                --resolution 64
+                --center_crop
+                --random_flip
+                --train_batch_size 1
+                --gradient_accumulation_steps 1
+                --max_train_steps 4
+                --learning_rate 5.0e-04
+                --scale_lr
+                --lr_scheduler constant
+                --lr_warmup_steps 0
+                --output_dir {tmpdir}
+                --checkpointing_steps=2
+                --seed=0
+                """.split()
+
+            run_command(self._launch_args + initial_run_args)
+
+            pipe = DiffusionPipeline.from_pretrained(tmpdir, safety_checker=None)
+            pipe(prompt, num_inference_steps=1)
+
+            # check checkpoint directories exist
+            self.assertEqual(
+                {x for x in os.listdir(tmpdir) if "checkpoint" in x},
+                {"checkpoint-2", "checkpoint-4"},
+            )
+
+            # check can run an intermediate checkpoint
+            unet = UNet2DConditionModel.from_pretrained(tmpdir, subfolder="checkpoint-2/unet")
+            pipe = DiffusionPipeline.from_pretrained(pretrained_model_name_or_path, unet=unet, safety_checker=None)
+            pipe(prompt, num_inference_steps=1)
+
+            # Remove checkpoint 2 so that we can check only later checkpoints exist after resuming
+            shutil.rmtree(os.path.join(tmpdir, "checkpoint-2"))
+
+            # Run training script for 2 total steps resuming from checkpoint 4
+
+            resume_run_args = f"""
+                examples/text_to_image/train_text_to_image.py
+                --pretrained_model_name_or_path {pretrained_model_name_or_path}
+                --dataset_name hf-internal-testing/dummy_image_text_data
+                --resolution 64
+                --center_crop
+                --random_flip
+                --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-4
+                --seed=0
+                """.split()
+
+            run_command(self._launch_args + resume_run_args)
+
+            # check can run new fully trained pipeline
+            pipe = DiffusionPipeline.from_pretrained(tmpdir, safety_checker=None)
+            pipe(prompt, num_inference_steps=1)
+
+            # no checkpoint-2 -> check old checkpoints do not exist
+            # check new checkpoints exist
+            self.assertEqual(
+                {x for x in os.listdir(tmpdir) if "checkpoint" in x},
+                {"checkpoint-4", "checkpoint-5"},
+            )
+
+    def test_text_to_image_checkpointing_use_ema(self):
+        pretrained_model_name_or_path = "hf-internal-testing/tiny-stable-diffusion-pipe"
+        prompt = "a prompt"
+
+        with tempfile.TemporaryDirectory() as tmpdir:
+            # Run training script with checkpointing
+            # max_train_steps == 4, checkpointing_steps == 2
+            # Should create checkpoints at steps 2, 4
+
+            initial_run_args = f"""
+                examples/text_to_image/train_text_to_image.py
+                --pretrained_model_name_or_path {pretrained_model_name_or_path}
+                --dataset_name hf-internal-testing/dummy_image_text_data
+                --resolution 64
+                --center_crop
+                --random_flip
+                --train_batch_size 1
+                --gradient_accumulation_steps 1
+                --max_train_steps 4
+                --learning_rate 5.0e-04
+                --scale_lr
+                --lr_scheduler constant
+                --lr_warmup_steps 0
+                --output_dir {tmpdir}
+                --checkpointing_steps=2
+                --use_ema
+                --seed=0
+                """.split()
+
+            run_command(self._launch_args + initial_run_args)
+
+            pipe = DiffusionPipeline.from_pretrained(tmpdir, safety_checker=None)
+            pipe(prompt, num_inference_steps=2)
+
+            # check checkpoint directories exist
+            self.assertEqual(
+                {x for x in os.listdir(tmpdir) if "checkpoint" in x},
+                {"checkpoint-2", "checkpoint-4"},
+            )
+
+            # check can run an intermediate checkpoint
+            unet = UNet2DConditionModel.from_pretrained(tmpdir, subfolder="checkpoint-2/unet")
+            pipe = DiffusionPipeline.from_pretrained(pretrained_model_name_or_path, unet=unet, safety_checker=None)
+            pipe(prompt, num_inference_steps=1)
+
+            # Remove checkpoint 2 so that we can check only later checkpoints exist after resuming
+            shutil.rmtree(os.path.join(tmpdir, "checkpoint-2"))
+
+            # Run training script for 2 total steps resuming from checkpoint 4
+
+            resume_run_args = f"""
+                examples/text_to_image/train_text_to_image.py
+                --pretrained_model_name_or_path {pretrained_model_name_or_path}
+                --dataset_name hf-internal-testing/dummy_image_text_data
+                --resolution 64
+                --center_crop
+                --random_flip
+                --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-4
+                --use_ema
+                --seed=0
+                """.split()
+
+            run_command(self._launch_args + resume_run_args)
+
+            # check can run new fully trained pipeline
+            pipe = DiffusionPipeline.from_pretrained(tmpdir, safety_checker=None)
+            pipe(prompt, num_inference_steps=1)
+
+            # no checkpoint-2 -> check old checkpoints do not exist
+            # check new checkpoints exist
+            self.assertEqual(
+                {x for x in os.listdir(tmpdir) if "checkpoint" in x},
+                {"checkpoint-4", "checkpoint-5"},
+            )
+
+    def test_text_to_image_checkpointing_checkpoints_total_limit(self):
+        pretrained_model_name_or_path = "hf-internal-testing/tiny-stable-diffusion-pipe"
+        prompt = "a prompt"
+
+        with tempfile.TemporaryDirectory() as tmpdir:
+            # Run training script with checkpointing
+            # max_train_steps == 6, checkpointing_steps == 2, checkpoints_total_limit == 2
+            # Should create checkpoints at steps 2, 4, 6
+            # with checkpoint at step 2 deleted
+
+            initial_run_args = f"""
+                examples/text_to_image/train_text_to_image.py
+                --pretrained_model_name_or_path {pretrained_model_name_or_path}
+                --dataset_name hf-internal-testing/dummy_image_text_data
+                --resolution 64
+                --center_crop
+                --random_flip
+                --train_batch_size 1
+                --gradient_accumulation_steps 1
+                --max_train_steps 6
+                --learning_rate 5.0e-04
+                --scale_lr
+                --lr_scheduler constant
+                --lr_warmup_steps 0
+                --output_dir {tmpdir}
+                --checkpointing_steps=2
+                --checkpoints_total_limit=2
+                --seed=0
+                """.split()
+
+            run_command(self._launch_args + initial_run_args)
+
+            pipe = DiffusionPipeline.from_pretrained(tmpdir, safety_checker=None)
+            pipe(prompt, num_inference_steps=1)
+
+            # check checkpoint directories exist
+            # checkpoint-2 should have been deleted
+            self.assertEqual({x for x in os.listdir(tmpdir) if "checkpoint" in x}, {"checkpoint-4", "checkpoint-6"})
+
+    def test_text_to_image_checkpointing_checkpoints_total_limit_removes_multiple_checkpoints(self):
+        pretrained_model_name_or_path = "hf-internal-testing/tiny-stable-diffusion-pipe"
+        prompt = "a prompt"
+
+        with tempfile.TemporaryDirectory() as tmpdir:
+            # Run training script with checkpointing
+            # max_train_steps == 4, checkpointing_steps == 2
+            # Should create checkpoints at steps 2, 4
+
+            initial_run_args = f"""
+                examples/text_to_image/train_text_to_image.py
+                --pretrained_model_name_or_path {pretrained_model_name_or_path}
+                --dataset_name hf-internal-testing/dummy_image_text_data
+                --resolution 64
+                --center_crop
+                --random_flip
+                --train_batch_size 1
+                --gradient_accumulation_steps 1
+                --max_train_steps 4
+                --learning_rate 5.0e-04
+                --scale_lr
+                --lr_scheduler constant
+                --lr_warmup_steps 0
+                --output_dir {tmpdir}
+                --checkpointing_steps=2
+                --seed=0
+                """.split()
+
+            run_command(self._launch_args + initial_run_args)
+
+            pipe = DiffusionPipeline.from_pretrained(tmpdir, safety_checker=None)
+            pipe(prompt, num_inference_steps=1)
+
+            # check checkpoint directories exist
+            self.assertEqual(
+                {x for x in os.listdir(tmpdir) if "checkpoint" in x},
+                {"checkpoint-2", "checkpoint-4"},
+            )
+
+            # resume and we should try to checkpoint at 6, where we'll have to remove
+            # checkpoint-2 and checkpoint-4 instead of just a single previous checkpoint
+
+            resume_run_args = f"""
+                examples/text_to_image/train_text_to_image.py
+                --pretrained_model_name_or_path {pretrained_model_name_or_path}
+                --dataset_name hf-internal-testing/dummy_image_text_data
+                --resolution 64
+                --center_crop
+                --random_flip
+                --train_batch_size 1
+                --gradient_accumulation_steps 1
+                --max_train_steps 8
+                --learning_rate 5.0e-04
+                --scale_lr
+                --lr_scheduler constant
+                --lr_warmup_steps 0
+                --output_dir {tmpdir}
+                --checkpointing_steps=2
+                --resume_from_checkpoint=checkpoint-4
+                --checkpoints_total_limit=2
+                --seed=0
+                """.split()
+
+            run_command(self._launch_args + resume_run_args)
+
+            pipe = DiffusionPipeline.from_pretrained(tmpdir, safety_checker=None)
+            pipe(prompt, num_inference_steps=1)
+
+            # check checkpoint directories exist
+            self.assertEqual(
+                {x for x in os.listdir(tmpdir) if "checkpoint" in x},
+                {"checkpoint-6", "checkpoint-8"},
+            )
+
+
+class TextToImageSDXL(ExamplesTestsAccelerate):
+    def test_text_to_image_sdxl(self):
+        with tempfile.TemporaryDirectory() as tmpdir:
+            test_args = f"""
+                examples/text_to_image/train_text_to_image_sdxl.py
+                --pretrained_model_name_or_path hf-internal-testing/tiny-stable-diffusion-xl-pipe
+                --dataset_name hf-internal-testing/dummy_image_text_data
+                --resolution 64
+                --center_crop
+                --random_flip
+                --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, "unet", "diffusion_pytorch_model.safetensors")))
+            self.assertTrue(os.path.isfile(os.path.join(tmpdir, "scheduler", "scheduler_config.json")))

test_text_to_image_lora.py

@@ -0,0 +1,300 @@
+#!/usr/bin/env python
+# coding=utf-8
+# Copyright 2025 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
+# limitations under the License.
+
+import logging
+import os
+import sys
+import tempfile
+
+import safetensors
+
+from diffusers import DiffusionPipeline  # noqa: E402
+
+
+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 TextToImageLoRA(ExamplesTestsAccelerate):
+    def test_text_to_image_lora_sdxl_checkpointing_checkpoints_total_limit(self):
+        prompt = "a prompt"
+        pipeline_path = "hf-internal-testing/tiny-stable-diffusion-xl-pipe"
+
+        with tempfile.TemporaryDirectory() as tmpdir:
+            # Run training script with checkpointing
+            # max_train_steps == 6, checkpointing_steps == 2, checkpoints_total_limit == 2
+            # Should create checkpoints at steps 2, 4, 6
+            # with checkpoint at step 2 deleted
+
+            initial_run_args = f"""
+                examples/text_to_image/train_text_to_image_lora_sdxl.py
+                --pretrained_model_name_or_path {pipeline_path}
+                --dataset_name hf-internal-testing/dummy_image_text_data
+                --resolution 64
+                --train_batch_size 1
+                --gradient_accumulation_steps 1
+                --max_train_steps 6
+                --learning_rate 5.0e-04
+                --scale_lr
+                --lr_scheduler constant
+                --lr_warmup_steps 0
+                --output_dir {tmpdir}
+                --checkpointing_steps=2
+                --checkpoints_total_limit=2
+                """.split()
+
+            run_command(self._launch_args + initial_run_args)
+
+            pipe = DiffusionPipeline.from_pretrained(pipeline_path)
+            pipe.load_lora_weights(tmpdir)
+            pipe(prompt, num_inference_steps=1)
+
+            # check checkpoint directories exist
+            # checkpoint-2 should have been deleted
+            self.assertEqual({x for x in os.listdir(tmpdir) if "checkpoint" in x}, {"checkpoint-4", "checkpoint-6"})
+
+    def test_text_to_image_lora_checkpointing_checkpoints_total_limit(self):
+        pretrained_model_name_or_path = "hf-internal-testing/tiny-stable-diffusion-pipe"
+        prompt = "a prompt"
+
+        with tempfile.TemporaryDirectory() as tmpdir:
+            # Run training script with checkpointing
+            # max_train_steps == 6, checkpointing_steps == 2, checkpoints_total_limit == 2
+            # Should create checkpoints at steps 2, 4, 6
+            # with checkpoint at step 2 deleted
+
+            initial_run_args = f"""
+                examples/text_to_image/train_text_to_image_lora.py
+                --pretrained_model_name_or_path {pretrained_model_name_or_path}
+                --dataset_name hf-internal-testing/dummy_image_text_data
+                --resolution 64
+                --center_crop
+                --random_flip
+                --train_batch_size 1
+                --gradient_accumulation_steps 1
+                --max_train_steps 6
+                --learning_rate 5.0e-04
+                --scale_lr
+                --lr_scheduler constant
+                --lr_warmup_steps 0
+                --output_dir {tmpdir}
+                --checkpointing_steps=2
+                --checkpoints_total_limit=2
+                --seed=0
+                --num_validation_images=0
+                """.split()
+
+            run_command(self._launch_args + initial_run_args)
+
+            pipe = DiffusionPipeline.from_pretrained(
+                "hf-internal-testing/tiny-stable-diffusion-pipe", safety_checker=None
+            )
+            pipe.load_lora_weights(tmpdir)
+            pipe(prompt, num_inference_steps=1)
+
+            # check checkpoint directories exist
+            # checkpoint-2 should have been deleted
+            self.assertEqual({x for x in os.listdir(tmpdir) if "checkpoint" in x}, {"checkpoint-4", "checkpoint-6"})
+
+    def test_text_to_image_lora_checkpointing_checkpoints_total_limit_removes_multiple_checkpoints(self):
+        pretrained_model_name_or_path = "hf-internal-testing/tiny-stable-diffusion-pipe"
+        prompt = "a prompt"
+
+        with tempfile.TemporaryDirectory() as tmpdir:
+            # Run training script with checkpointing
+            # max_train_steps == 4, checkpointing_steps == 2
+            # Should create checkpoints at steps 2, 4
+
+            initial_run_args = f"""
+                examples/text_to_image/train_text_to_image_lora.py
+                --pretrained_model_name_or_path {pretrained_model_name_or_path}
+                --dataset_name hf-internal-testing/dummy_image_text_data
+                --resolution 64
+                --center_crop
+                --random_flip
+                --train_batch_size 1
+                --gradient_accumulation_steps 1
+                --max_train_steps 4
+                --learning_rate 5.0e-04
+                --scale_lr
+                --lr_scheduler constant
+                --lr_warmup_steps 0
+                --output_dir {tmpdir}
+                --checkpointing_steps=2
+                --seed=0
+                --num_validation_images=0
+                """.split()
+
+            run_command(self._launch_args + initial_run_args)
+
+            pipe = DiffusionPipeline.from_pretrained(
+                "hf-internal-testing/tiny-stable-diffusion-pipe", safety_checker=None
+            )
+            pipe.load_lora_weights(tmpdir)
+            pipe(prompt, num_inference_steps=1)
+
+            # check checkpoint directories exist
+            self.assertEqual(
+                {x for x in os.listdir(tmpdir) if "checkpoint" in x},
+                {"checkpoint-2", "checkpoint-4"},
+            )
+
+            # resume and we should try to checkpoint at 6, where we'll have to remove
+            # checkpoint-2 and checkpoint-4 instead of just a single previous checkpoint
+
+            resume_run_args = f"""
+                examples/text_to_image/train_text_to_image_lora.py
+                --pretrained_model_name_or_path {pretrained_model_name_or_path}
+                --dataset_name hf-internal-testing/dummy_image_text_data
+                --resolution 64
+                --center_crop
+                --random_flip
+                --train_batch_size 1
+                --gradient_accumulation_steps 1
+                --max_train_steps 8
+                --learning_rate 5.0e-04
+                --scale_lr
+                --lr_scheduler constant
+                --lr_warmup_steps 0
+                --output_dir {tmpdir}
+                --checkpointing_steps=2
+                --resume_from_checkpoint=checkpoint-4
+                --checkpoints_total_limit=2
+                --seed=0
+                --num_validation_images=0
+                """.split()
+
+            run_command(self._launch_args + resume_run_args)
+
+            pipe = DiffusionPipeline.from_pretrained(
+                "hf-internal-testing/tiny-stable-diffusion-pipe", safety_checker=None
+            )
+            pipe.load_lora_weights(tmpdir)
+            pipe(prompt, num_inference_steps=1)
+
+            # check checkpoint directories exist
+            self.assertEqual(
+                {x for x in os.listdir(tmpdir) if "checkpoint" in x},
+                {"checkpoint-6", "checkpoint-8"},
+            )
+
+
+class TextToImageLoRASDXL(ExamplesTestsAccelerate):
+    def test_text_to_image_lora_sdxl(self):
+        with tempfile.TemporaryDirectory() as tmpdir:
+            test_args = f"""
+                examples/text_to_image/train_text_to_image_lora_sdxl.py
+                --pretrained_model_name_or_path hf-internal-testing/tiny-stable-diffusion-xl-pipe
+                --dataset_name hf-internal-testing/dummy_image_text_data
+                --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, "pytorch_lora_weights.safetensors")))
+
+            # make sure the state_dict has the correct naming in the parameters.
+            lora_state_dict = safetensors.torch.load_file(os.path.join(tmpdir, "pytorch_lora_weights.safetensors"))
+            is_lora = all("lora" in k for k in lora_state_dict.keys())
+            self.assertTrue(is_lora)
+
+    def test_text_to_image_lora_sdxl_with_text_encoder(self):
+        with tempfile.TemporaryDirectory() as tmpdir:
+            test_args = f"""
+                examples/text_to_image/train_text_to_image_lora_sdxl.py
+                --pretrained_model_name_or_path hf-internal-testing/tiny-stable-diffusion-xl-pipe
+                --dataset_name hf-internal-testing/dummy_image_text_data
+                --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}
+                --train_text_encoder
+                """.split()
+
+            run_command(self._launch_args + test_args)
+            # save_pretrained smoke test
+            self.assertTrue(os.path.isfile(os.path.join(tmpdir, "pytorch_lora_weights.safetensors")))
+
+            # make sure the state_dict has the correct naming in the parameters.
+            lora_state_dict = safetensors.torch.load_file(os.path.join(tmpdir, "pytorch_lora_weights.safetensors"))
+            is_lora = all("lora" in k for k in lora_state_dict.keys())
+            self.assertTrue(is_lora)
+
+            # when not training the text encoder, all the parameters in the state dict should start
+            # with `"unet"` or `"text_encoder"` or `"text_encoder_2"` in their names.
+            keys = lora_state_dict.keys()
+            starts_with_unet = all(
+                k.startswith("unet") or k.startswith("text_encoder") or k.startswith("text_encoder_2") for k in keys
+            )
+            self.assertTrue(starts_with_unet)
+
+    def test_text_to_image_lora_sdxl_text_encoder_checkpointing_checkpoints_total_limit(self):
+        prompt = "a prompt"
+        pipeline_path = "hf-internal-testing/tiny-stable-diffusion-xl-pipe"
+
+        with tempfile.TemporaryDirectory() as tmpdir:
+            # Run training script with checkpointing
+            # max_train_steps == 6, checkpointing_steps == 2, checkpoints_total_limit == 2
+            # Should create checkpoints at steps 2, 4, 6
+            # with checkpoint at step 2 deleted
+
+            initial_run_args = f"""
+                examples/text_to_image/train_text_to_image_lora_sdxl.py
+                --pretrained_model_name_or_path {pipeline_path}
+                --dataset_name hf-internal-testing/dummy_image_text_data
+                --resolution 64
+                --train_batch_size 1
+                --gradient_accumulation_steps 1
+                --max_train_steps 6
+                --learning_rate 5.0e-04
+                --scale_lr
+                --lr_scheduler constant
+                --train_text_encoder
+                --lr_warmup_steps 0
+                --output_dir {tmpdir}
+                --checkpointing_steps=2
+                --checkpoints_total_limit=2
+                """.split()
+
+            run_command(self._launch_args + initial_run_args)
+
+            pipe = DiffusionPipeline.from_pretrained(pipeline_path)
+            pipe.load_lora_weights(tmpdir)
+            pipe(prompt, num_inference_steps=1)
+
+            # check checkpoint directories exist
+            # checkpoint-2 should have been deleted
+            self.assertEqual({x for x in os.listdir(tmpdir) if "checkpoint" in x}, {"checkpoint-4", "checkpoint-6"})

train_text_to_image.py

@@ -0,0 +1,1153 @@
+#!/usr/bin/env python
+# coding=utf-8
+# Copyright 2025 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
+# limitations under the License.
+
+import argparse
+import logging
+import math
+import os
+import random
+import shutil
+from contextlib import nullcontext
+from pathlib import Path
+
+import accelerate
+import datasets
+import numpy as np
+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.state import AcceleratorState
+from accelerate.utils import ProjectConfiguration, set_seed
+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
+from transformers import CLIPTextModel, CLIPTokenizer
+from transformers.utils import ContextManagers
+
+import diffusers
+from diffusers import AutoencoderKL, DDPMScheduler, StableDiffusionPipeline, UNet2DConditionModel
+from diffusers.optimization import get_scheduler
+from diffusers.training_utils import EMAModel, compute_dream_and_update_latents, compute_snr
+from diffusers.utils import check_min_version, deprecate, is_wandb_available, make_image_grid
+from diffusers.utils.hub_utils import load_or_create_model_card, populate_model_card
+from diffusers.utils.import_utils import is_xformers_available
+from diffusers.utils.torch_utils import is_compiled_module
+
+
+if is_wandb_available():
+    import wandb
+
+
+# Will error if the minimal version of diffusers is not installed. Remove at your own risks.
+check_min_version("0.33.0.dev0")
+
+logger = get_logger(__name__, log_level="INFO")
+
+DATASET_NAME_MAPPING = {
+    "lambdalabs/naruto-blip-captions": ("image", "text"),
+}
+
+
+def save_model_card(
+    args,
+    repo_id: str,
+    images: list = None,
+    repo_folder: str = None,
+):
+    img_str = ""
+    if len(images) > 0:
+        image_grid = make_image_grid(images, 1, len(args.validation_prompts))
+        image_grid.save(os.path.join(repo_folder, "val_imgs_grid.png"))
+        img_str += "![val_imgs_grid](./val_imgs_grid.png)\n"
+
+    model_description = f"""
+# Text-to-image finetuning - {repo_id}
+
+This pipeline was finetuned from **{args.pretrained_model_name_or_path}** on the **{args.dataset_name}** dataset. Below are some example images generated with the finetuned pipeline using the following prompts: {args.validation_prompts}: \n
+{img_str}
+
+## Pipeline usage
+
+You can use the pipeline like so:
+
+```python
+from diffusers import DiffusionPipeline
+import torch
+
+pipeline = DiffusionPipeline.from_pretrained("{repo_id}", torch_dtype=torch.float16)
+prompt = "{args.validation_prompts[0]}"
+image = pipeline(prompt).images[0]
+image.save("my_image.png")
+```
+
+## Training info
+
+These are the key hyperparameters used during training:
+
+* Epochs: {args.num_train_epochs}
+* Learning rate: {args.learning_rate}
+* Batch size: {args.train_batch_size}
+* Gradient accumulation steps: {args.gradient_accumulation_steps}
+* Image resolution: {args.resolution}
+* Mixed-precision: {args.mixed_precision}
+
+"""
+    wandb_info = ""
+    if is_wandb_available():
+        wandb_run_url = None
+        if wandb.run is not None:
+            wandb_run_url = wandb.run.url
+
+    if wandb_run_url is not None:
+        wandb_info = f"""
+More information on all the CLI arguments and the environment are available on your [`wandb` run page]({wandb_run_url}).
+"""
+
+    model_description += wandb_info
+
+    model_card = load_or_create_model_card(
+        repo_id_or_path=repo_id,
+        from_training=True,
+        license="creativeml-openrail-m",
+        base_model=args.pretrained_model_name_or_path,
+        model_description=model_description,
+        inference=True,
+    )
+
+    tags = ["stable-diffusion", "stable-diffusion-diffusers", "text-to-image", "diffusers", "diffusers-training"]
+    model_card = populate_model_card(model_card, tags=tags)
+
+    model_card.save(os.path.join(repo_folder, "README.md"))
+
+
+def log_validation(vae, text_encoder, tokenizer, unet, args, accelerator, weight_dtype, epoch):
+    logger.info("Running validation... ")
+
+    pipeline = StableDiffusionPipeline.from_pretrained(
+        args.pretrained_model_name_or_path,
+        vae=accelerator.unwrap_model(vae),
+        text_encoder=accelerator.unwrap_model(text_encoder),
+        tokenizer=tokenizer,
+        unet=accelerator.unwrap_model(unet),
+        safety_checker=None,
+        revision=args.revision,
+        variant=args.variant,
+        torch_dtype=weight_dtype,
+    )
+    pipeline = pipeline.to(accelerator.device)
+    pipeline.set_progress_bar_config(disable=True)
+
+    if args.enable_xformers_memory_efficient_attention:
+        pipeline.enable_xformers_memory_efficient_attention()
+
+    if args.seed is None:
+        generator = None
+    else:
+        generator = torch.Generator(device=accelerator.device).manual_seed(args.seed)
+
+    images = []
+    for i in range(len(args.validation_prompts)):
+        if torch.backends.mps.is_available():
+            autocast_ctx = nullcontext()
+        else:
+            autocast_ctx = torch.autocast(accelerator.device.type)
+
+        with autocast_ctx:
+            image = pipeline(args.validation_prompts[i], num_inference_steps=20, 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")
+        elif tracker.name == "wandb":
+            tracker.log(
+                {
+                    "validation": [
+                        wandb.Image(image, caption=f"{i}: {args.validation_prompts[i]}")
+                        for i, image in enumerate(images)
+                    ]
+                }
+            )
+        else:
+            logger.warning(f"image logging not implemented for {tracker.name}")
+
+    del pipeline
+    torch.cuda.empty_cache()
+
+    return images
+
+
+def parse_args():
+    parser = argparse.ArgumentParser(description="Simple example of a training script.")
+    parser.add_argument(
+        "--input_perturbation", type=float, default=0, help="The scale of input perturbation. Recommended 0.1."
+    )
+    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(
+        "--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 🤗 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(
+        "--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(
+        "--image_column", type=str, default="image", help="The column of the dataset containing an image."
+    )
+    parser.add_argument(
+        "--caption_column",
+        type=str,
+        default="text",
+        help="The column of the dataset containing a caption or a list of captions.",
+    )
+    parser.add_argument(
+        "--max_train_samples",
+        type=int,
+        default=None,
+        help=(
+            "For debugging purposes or quicker training, truncate the number of training examples to this "
+            "value if set."
+        ),
+    )
+    parser.add_argument(
+        "--validation_prompts",
+        type=str,
+        default=None,
+        nargs="+",
+        help=("A set of prompts evaluated every `--validation_epochs` and logged to `--report_to`."),
+    )
+    parser.add_argument(
+        "--output_dir",
+        type=str,
+        default="sd-model-finetuned",
+        help="The output directory where the model predictions and checkpoints will be written.",
+    )
+    parser.add_argument(
+        "--cache_dir",
+        type=str,
+        default=None,
+        help="The directory where the downloaded models and datasets will be stored.",
+    )
+    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",
+        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",
+        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("--num_train_epochs", type=int, default=100)
+    parser.add_argument(
+        "--max_train_steps",
+        type=int,
+        default=None,
+        help="Total number of training steps to perform.  If provided, overrides num_train_epochs.",
+    )
+    parser.add_argument(
+        "--gradient_accumulation_steps",
+        type=int,
+        default=1,
+        help="Number of updates steps to accumulate before performing a backward/update pass.",
+    )
+    parser.add_argument(
+        "--gradient_checkpointing",
+        action="store_true",
+        help="Whether or not to use gradient checkpointing to save memory at the expense of slower backward pass.",
+    )
+    parser.add_argument(
+        "--learning_rate",
+        type=float,
+        default=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(
+        "--snr_gamma",
+        type=float,
+        default=None,
+        help="SNR weighting gamma to be used if rebalancing the loss. Recommended value is 5.0. "
+        "More details here: https://arxiv.org/abs/2303.09556.",
+    )
+    parser.add_argument(
+        "--dream_training",
+        action="store_true",
+        help=(
+            "Use the DREAM training method, which makes training more efficient and accurate at the "
+            "expense of doing an extra forward pass. See: https://arxiv.org/abs/2312.00210"
+        ),
+    )
+    parser.add_argument(
+        "--dream_detail_preservation",
+        type=float,
+        default=1.0,
+        help="Dream detail preservation factor p (should be greater than 0; default=1.0, as suggested in the paper)",
+    )
+    parser.add_argument(
+        "--use_8bit_adam", action="store_true", help="Whether or not to use 8-bit Adam from bitsandbytes."
+    )
+    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("--use_ema", action="store_true", help="Whether to use EMA model.")
+    parser.add_argument("--offload_ema", action="store_true", help="Offload EMA model to CPU during training step.")
+    parser.add_argument("--foreach_ema", action="store_true", help="Use faster foreach implementation of EMAModel.")
+    parser.add_argument(
+        "--non_ema_revision",
+        type=str,
+        default=None,
+        required=False,
+        help=(
+            "Revision of pretrained non-ema model identifier. Must be a branch, tag or git identifier of the local or"
+            " remote repository specified with --pretrained_model_name_or_path."
+        ),
+    )
+    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("--max_grad_norm", default=1.0, type=float, help="Max gradient norm.")
+    parser.add_argument("--push_to_hub", action="store_true", help="Whether or not to push the model to the Hub.")
+    parser.add_argument("--hub_token", type=str, default=None, help="The token to use to push to the Model Hub.")
+    parser.add_argument(
+        "--prediction_type",
+        type=str,
+        default=None,
+        help="The prediction_type that shall be used for training. Choose between 'epsilon' or 'v_prediction' or leave `None`. If left to `None` the default prediction type of the scheduler: `noise_scheduler.config.prediction_type` is chosen.",
+    )
+    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=None,
+        choices=["no", "fp16", "bf16"],
+        help=(
+            "Whether to use mixed precision. Choose between fp16 and bf16 (bfloat16). Bf16 requires PyTorch >="
+            " 1.10.and an Nvidia Ampere GPU.  Default to the value of accelerate config of the current system or the"
+            " flag passed with the `accelerate.launch` command. Use this argument to override the accelerate config."
+        ),
+    )
+    parser.add_argument(
+        "--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("--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("--noise_offset", type=float, default=0, help="The scale of noise offset.")
+    parser.add_argument(
+        "--validation_epochs",
+        type=int,
+        default=5,
+        help="Run validation every X epochs.",
+    )
+    parser.add_argument(
+        "--tracker_project_name",
+        type=str,
+        default="text2image-fine-tune",
+        help=(
+            "The `project_name` argument passed to Accelerator.init_trackers for"
+            " more information see https://huggingface.co/docs/accelerate/v0.17.0/en/package_reference/accelerator#accelerate.Accelerator"
+        ),
+    )
+
+    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
+
+    # Sanity checks
+    if args.dataset_name is None and args.train_data_dir is None:
+        raise ValueError("Need either a dataset name or a training folder.")
+
+    # default to using the same revision for the non-ema model if not specified
+    if args.non_ema_revision is None:
+        args.non_ema_revision = args.revision
+
+    return args
+
+
+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."
+        )
+
+    if args.non_ema_revision is not None:
+        deprecate(
+            "non_ema_revision!=None",
+            "0.15.0",
+            message=(
+                "Downloading 'non_ema' weights from revision branches of the Hub is deprecated. Please make sure to"
+                " use `--variant=non_ema` instead."
+            ),
+        )
+    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,
+    )
+
+    # Disable AMP for MPS.
+    if torch.backends.mps.is_available():
+        accelerator.native_amp = False
+
+    # Make one log on every process with the configuration for debugging.
+    logging.basicConfig(
+        format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
+        datefmt="%m/%d/%Y %H:%M:%S",
+        level=logging.INFO,
+    )
+    logger.info(accelerator.state, main_process_only=False)
+    if accelerator.is_local_main_process:
+        datasets.utils.logging.set_verbosity_warning()
+        transformers.utils.logging.set_verbosity_warning()
+        diffusers.utils.logging.set_verbosity_info()
+    else:
+        datasets.utils.logging.set_verbosity_error()
+        transformers.utils.logging.set_verbosity_error()
+        diffusers.utils.logging.set_verbosity_error()
+
+    # If passed along, set the training seed now.
+    if args.seed is not None:
+        set_seed(args.seed)
+
+    # 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 scheduler, tokenizer and models.
+    noise_scheduler = DDPMScheduler.from_pretrained(args.pretrained_model_name_or_path, subfolder="scheduler")
+    tokenizer = CLIPTokenizer.from_pretrained(
+        args.pretrained_model_name_or_path, subfolder="tokenizer", revision=args.revision
+    )
+
+    def deepspeed_zero_init_disabled_context_manager():
+        """
+        returns either a context list that includes one that will disable zero.Init or an empty context list
+        """
+        deepspeed_plugin = AcceleratorState().deepspeed_plugin if accelerate.state.is_initialized() else None
+        if deepspeed_plugin is None:
+            return []
+
+        return [deepspeed_plugin.zero3_init_context_manager(enable=False)]
+
+    # Currently Accelerate doesn't know how to handle multiple models under Deepspeed ZeRO stage 3.
+    # For this to work properly all models must be run through `accelerate.prepare`. But accelerate
+    # will try to assign the same optimizer with the same weights to all models during
+    # `deepspeed.initialize`, which of course doesn't work.
+    #
+    # For now the following workaround will partially support Deepspeed ZeRO-3, by excluding the 2
+    # frozen models from being partitioned during `zero.Init` which gets called during
+    # `from_pretrained` So CLIPTextModel and AutoencoderKL will not enjoy the parameter sharding
+    # across multiple gpus and only UNet2DConditionModel will get ZeRO sharded.
+    with ContextManagers(deepspeed_zero_init_disabled_context_manager()):
+        text_encoder = CLIPTextModel.from_pretrained(
+            args.pretrained_model_name_or_path, subfolder="text_encoder", revision=args.revision, variant=args.variant
+        )
+        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.non_ema_revision
+    )
+
+    # Freeze vae and text_encoder and set unet to trainable
+    vae.requires_grad_(False)
+    text_encoder.requires_grad_(False)
+    unet.train()
+
+    # Create EMA for the unet.
+    if args.use_ema:
+        ema_unet = UNet2DConditionModel.from_pretrained(
+            args.pretrained_model_name_or_path, subfolder="unet", revision=args.revision, variant=args.variant
+        )
+        ema_unet = EMAModel(
+            ema_unet.parameters(),
+            model_cls=UNet2DConditionModel,
+            model_config=ema_unet.config,
+            foreach=args.foreach_ema,
+        )
+
+    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")
+
+    # `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_unet.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"), UNet2DConditionModel, foreach=args.foreach_ema
+                )
+                ema_unet.load_state_dict(load_model.state_dict())
+                if args.offload_ema:
+                    ema_unet.pin_memory()
+                else:
+                    ema_unet.to(accelerator.device)
+                del load_model
+
+            for _ in range(len(models)):
+                # pop models so that they are not loaded again
+                model = models.pop()
+
+                # load diffusers style into model
+                load_model = UNet2DConditionModel.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)
+
+    if args.gradient_checkpointing:
+        unet.enable_gradient_checkpointing()
+
+    # 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(
+                "Please install bitsandbytes to use 8-bit Adam. You can do so by running `pip install bitsandbytes`"
+            )
+
+        optimizer_cls = bnb.optim.AdamW8bit
+    else:
+        optimizer_cls = torch.optim.AdamW
+
+    optimizer = optimizer_cls(
+        unet.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:
+        # Downloading and loading a dataset from the hub.
+        dataset = load_dataset(
+            args.dataset_name,
+            args.dataset_config_name,
+            cache_dir=args.cache_dir,
+            data_dir=args.train_data_dir,
+        )
+    else:
+        data_files = {}
+        if args.train_data_dir is not None:
+            data_files["train"] = os.path.join(args.train_data_dir, "**")
+        dataset = load_dataset(
+            "imagefolder",
+            data_files=data_files,
+            cache_dir=args.cache_dir,
+        )
+        # See more about loading custom images at
+        # https://huggingface.co/docs/datasets/v2.4.0/en/image_load#imagefolder
+
+    # Preprocessing the datasets.
+    # We need to tokenize inputs and targets.
+    column_names = dataset["train"].column_names
+
+    # 6. Get the column names for input/target.
+    dataset_columns = DATASET_NAME_MAPPING.get(args.dataset_name, None)
+    if args.image_column is None:
+        image_column = dataset_columns[0] if dataset_columns is not None else column_names[0]
+    else:
+        image_column = args.image_column
+        if image_column not in column_names:
+            raise ValueError(
+                f"--image_column' value '{args.image_column}' needs to be one of: {', '.join(column_names)}"
+            )
+    if args.caption_column is None:
+        caption_column = dataset_columns[1] if dataset_columns is not None else column_names[1]
+    else:
+        caption_column = args.caption_column
+        if caption_column not in column_names:
+            raise ValueError(
+                f"--caption_column' value '{args.caption_column}' needs to be one of: {', '.join(column_names)}"
+            )
+
+    # Preprocessing the datasets.
+    # We need to tokenize input captions and transform the images.
+    def tokenize_captions(examples, is_train=True):
+        captions = []
+        for caption in examples[caption_column]:
+            if isinstance(caption, str):
+                captions.append(caption)
+            elif isinstance(caption, (list, np.ndarray)):
+                # take a random caption if there are multiple
+                captions.append(random.choice(caption) if is_train else caption[0])
+            else:
+                raise ValueError(
+                    f"Caption column `{caption_column}` should contain either strings or lists of strings."
+                )
+        inputs = tokenizer(
+            captions, max_length=tokenizer.model_max_length, padding="max_length", truncation=True, return_tensors="pt"
+        )
+        return inputs.input_ids
+
+    # Preprocessing the datasets.
+    train_transforms = 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 preprocess_train(examples):
+        images = [image.convert("RGB") for image in examples[image_column]]
+        examples["pixel_values"] = [train_transforms(image) for image in images]
+        examples["input_ids"] = tokenize_captions(examples)
+        return examples
+
+    with accelerator.main_process_first():
+        if args.max_train_samples is not None:
+            dataset["train"] = dataset["train"].shuffle(seed=args.seed).select(range(args.max_train_samples))
+        # Set the training transforms
+        train_dataset = dataset["train"].with_transform(preprocess_train)
+
+    def collate_fn(examples):
+        pixel_values = torch.stack([example["pixel_values"] for example in examples])
+        pixel_values = pixel_values.to(memory_format=torch.contiguous_format).float()
+        input_ids = torch.stack([example["input_ids"] for example in examples])
+        return {"pixel_values": pixel_values, "input_ids": input_ids}
+
+    # DataLoaders creation:
+    train_dataloader = torch.utils.data.DataLoader(
+        train_dataset,
+        shuffle=True,
+        collate_fn=collate_fn,
+        batch_size=args.train_batch_size,
+        num_workers=args.dataloader_num_workers,
+    )
+
+    # Scheduler and math around the number of training steps.
+    # Check the PR https://github.com/huggingface/diffusers/pull/8312 for detailed explanation.
+    num_warmup_steps_for_scheduler = args.lr_warmup_steps * accelerator.num_processes
+    if args.max_train_steps is None:
+        len_train_dataloader_after_sharding = math.ceil(len(train_dataloader) / accelerator.num_processes)
+        num_update_steps_per_epoch = math.ceil(len_train_dataloader_after_sharding / args.gradient_accumulation_steps)
+        num_training_steps_for_scheduler = (
+            args.num_train_epochs * num_update_steps_per_epoch * accelerator.num_processes
+        )
+    else:
+        num_training_steps_for_scheduler = args.max_train_steps * accelerator.num_processes
+
+    lr_scheduler = get_scheduler(
+        args.lr_scheduler,
+        optimizer=optimizer,
+        num_warmup_steps=num_warmup_steps_for_scheduler,
+        num_training_steps=num_training_steps_for_scheduler,
+    )
+
+    # Prepare everything with our `accelerator`.
+    unet, optimizer, train_dataloader, lr_scheduler = accelerator.prepare(
+        unet, optimizer, train_dataloader, lr_scheduler
+    )
+
+    if args.use_ema:
+        if args.offload_ema:
+            ema_unet.pin_memory()
+        else:
+            ema_unet.to(accelerator.device)
+
+    # For mixed precision training we cast all non-trainable weights (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
+        args.mixed_precision = accelerator.mixed_precision
+    elif accelerator.mixed_precision == "bf16":
+        weight_dtype = torch.bfloat16
+        args.mixed_precision = accelerator.mixed_precision
+
+    # Move text_encode and vae to gpu and cast to weight_dtype
+    text_encoder.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 args.max_train_steps is None:
+        args.max_train_steps = args.num_train_epochs * num_update_steps_per_epoch
+        if num_training_steps_for_scheduler != args.max_train_steps * accelerator.num_processes:
+            logger.warning(
+                f"The length of the 'train_dataloader' after 'accelerator.prepare' ({len(train_dataloader)}) does not match "
+                f"the expected length ({len_train_dataloader_after_sharding}) when the learning rate scheduler was created. "
+                f"This inconsistency may result in the learning rate scheduler not functioning properly."
+            )
+    # 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:
+        tracker_config = dict(vars(args))
+        tracker_config.pop("validation_prompts")
+        accelerator.init_trackers(args.tracker_project_name, tracker_config)
+
+    # Function for unwrapping if model was compiled with `torch.compile`.
+    def unwrap_model(model):
+        model = accelerator.unwrap_model(model)
+        model = model._orig_mod if is_compiled_module(model) else model
+        return model
+
+    # 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,
+    )
+
+    for epoch in range(first_epoch, args.num_train_epochs):
+        train_loss = 0.0
+        for step, batch in enumerate(train_dataloader):
+            with accelerator.accumulate(unet):
+                # Convert images to latent space
+                latents = vae.encode(batch["pixel_values"].to(weight_dtype)).latent_dist.sample()
+                latents = latents * vae.config.scaling_factor
+
+                # Sample noise that we'll add to the latents
+                noise = torch.randn_like(latents)
+                if args.noise_offset:
+                    # https://www.crosslabs.org//blog/diffusion-with-offset-noise
+                    noise += args.noise_offset * torch.randn(
+                        (latents.shape[0], latents.shape[1], 1, 1), device=latents.device
+                    )
+                if args.input_perturbation:
+                    new_noise = noise + args.input_perturbation * torch.randn_like(noise)
+                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)
+                if args.input_perturbation:
+                    noisy_latents = noise_scheduler.add_noise(latents, new_noise, timesteps)
+                else:
+                    noisy_latents = noise_scheduler.add_noise(latents, noise, timesteps)
+
+                # Get the text embedding for conditioning
+                encoder_hidden_states = text_encoder(batch["input_ids"], return_dict=False)[0]
+
+                # Get the target for loss depending on the prediction type
+                if args.prediction_type is not None:
+                    # set prediction_type of scheduler if defined
+                    noise_scheduler.register_to_config(prediction_type=args.prediction_type)
+
+                if noise_scheduler.config.prediction_type == "epsilon":
+                    target = noise
+                elif noise_scheduler.config.prediction_type == "v_prediction":
+                    target = noise_scheduler.get_velocity(latents, noise, timesteps)
+                else:
+                    raise ValueError(f"Unknown prediction type {noise_scheduler.config.prediction_type}")
+
+                if args.dream_training:
+                    noisy_latents, target = compute_dream_and_update_latents(
+                        unet,
+                        noise_scheduler,
+                        timesteps,
+                        noise,
+                        noisy_latents,
+                        target,
+                        encoder_hidden_states,
+                        args.dream_detail_preservation,
+                    )
+
+                # Predict the noise residual and compute loss
+                model_pred = unet(noisy_latents, timesteps, encoder_hidden_states, return_dict=False)[0]
+
+                if args.snr_gamma is None:
+                    loss = F.mse_loss(model_pred.float(), target.float(), reduction="mean")
+                else:
+                    # Compute loss-weights as per Section 3.4 of https://arxiv.org/abs/2303.09556.
+                    # Since we predict the noise instead of x_0, the original formulation is slightly changed.
+                    # This is discussed in Section 4.2 of the same paper.
+                    snr = compute_snr(noise_scheduler, timesteps)
+                    mse_loss_weights = torch.stack([snr, args.snr_gamma * torch.ones_like(timesteps)], dim=1).min(
+                        dim=1
+                    )[0]
+                    if noise_scheduler.config.prediction_type == "epsilon":
+                        mse_loss_weights = mse_loss_weights / snr
+                    elif noise_scheduler.config.prediction_type == "v_prediction":
+                        mse_loss_weights = mse_loss_weights / (snr + 1)
+
+                    loss = F.mse_loss(model_pred.float(), target.float(), reduction="none")
+                    loss = loss.mean(dim=list(range(1, len(loss.shape)))) * mse_loss_weights
+                    loss = loss.mean()
+
+                # Gather the losses across all processes for logging (if we use distributed training).
+                avg_loss = accelerator.gather(loss.repeat(args.train_batch_size)).mean()
+                train_loss += avg_loss.item() / args.gradient_accumulation_steps
+
+                # Backpropagate
+                accelerator.backward(loss)
+                if accelerator.sync_gradients:
+                    accelerator.clip_grad_norm_(unet.parameters(), args.max_grad_norm)
+                optimizer.step()
+                lr_scheduler.step()
+                optimizer.zero_grad()
+
+            # Checks if the accelerator has performed an optimization step behind the scenes
+            if accelerator.sync_gradients:
+                if args.use_ema:
+                    if args.offload_ema:
+                        ema_unet.to(device="cuda", non_blocking=True)
+                    ema_unet.step(unet.parameters())
+                    if args.offload_ema:
+                        ema_unet.to(device="cpu", non_blocking=True)
+                progress_bar.update(1)
+                global_step += 1
+                accelerator.log({"train_loss": train_loss}, step=global_step)
+                train_loss = 0.0
+
+                if global_step % args.checkpointing_steps == 0:
+                    if accelerator.is_main_process:
+                        # _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 = {"step_loss": loss.detach().item(), "lr": lr_scheduler.get_last_lr()[0]}
+            progress_bar.set_postfix(**logs)
+
+            if global_step >= args.max_train_steps:
+                break
+
+        if accelerator.is_main_process:
+            if args.validation_prompts is not None and epoch % args.validation_epochs == 0:
+                if args.use_ema:
+                    # Store the UNet parameters temporarily and load the EMA parameters to perform inference.
+                    ema_unet.store(unet.parameters())
+                    ema_unet.copy_to(unet.parameters())
+                log_validation(
+                    vae,
+                    text_encoder,
+                    tokenizer,
+                    unet,
+                    args,
+                    accelerator,
+                    weight_dtype,
+                    global_step,
+                )
+                if args.use_ema:
+                    # Switch back to the original UNet parameters.
+                    ema_unet.restore(unet.parameters())
+
+    # Create the pipeline using the trained modules and save it.
+    accelerator.wait_for_everyone()
+    if accelerator.is_main_process:
+        unet = unwrap_model(unet)
+        if args.use_ema:
+            ema_unet.copy_to(unet.parameters())
+
+        pipeline = StableDiffusionPipeline.from_pretrained(
+            args.pretrained_model_name_or_path,
+            text_encoder=text_encoder,
+            vae=vae,
+            unet=unet,
+            revision=args.revision,
+            variant=args.variant,
+        )
+        pipeline.save_pretrained(args.output_dir)
+
+        # Run a final round of inference.
+        images = []
+        if args.validation_prompts is not None:
+            logger.info("Running inference for collecting generated images...")
+            pipeline = pipeline.to(accelerator.device)
+            pipeline.torch_dtype = weight_dtype
+            pipeline.set_progress_bar_config(disable=True)
+
+            if args.enable_xformers_memory_efficient_attention:
+                pipeline.enable_xformers_memory_efficient_attention()
+
+            if args.seed is None:
+                generator = None
+            else:
+                generator = torch.Generator(device=accelerator.device).manual_seed(args.seed)
+
+            for i in range(len(args.validation_prompts)):
+                with torch.autocast("cuda"):
+                    image = pipeline(args.validation_prompts[i], num_inference_steps=20, generator=generator).images[0]
+                images.append(image)
+
+        if args.push_to_hub:
+            save_model_card(args, repo_id, images, 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()

train_text_to_image_flax.py

@@ -0,0 +1,620 @@
+#!/usr/bin/env python
+# coding=utf-8
+# Copyright 2025 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
+# limitations under the License.
+
+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 torch
+import torch.utils.checkpoint
+import transformers
+from datasets import load_dataset
+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
+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
+
+
+# Will error if the minimal version of diffusers is not installed. Remove at your own risks.
+check_min_version("0.33.0.dev0")
+
+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(
+        "--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(
+        "--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 🤗 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(
+        "--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(
+        "--image_column", type=str, default="image", help="The column of the dataset containing an image."
+    )
+    parser.add_argument(
+        "--caption_column",
+        type=str,
+        default="text",
+        help="The column of the dataset containing a caption or a list of captions.",
+    )
+    parser.add_argument(
+        "--max_train_samples",
+        type=int,
+        default=None,
+        help=(
+            "For debugging purposes or quicker training, truncate the number of training examples to this "
+            "value if set."
+        ),
+    )
+    parser.add_argument(
+        "--output_dir",
+        type=str,
+        default="sd-model-finetuned",
+        help="The output directory where the model predictions and checkpoints will be written.",
+    )
+    parser.add_argument(
+        "--cache_dir",
+        type=str,
+        default=None,
+        help="The directory where the downloaded models and datasets will be stored.",
+    )
+    parser.add_argument("--seed", type=int, default=0, 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",
+        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",
+        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("--num_train_epochs", type=int, default=100)
+    parser.add_argument(
+        "--max_train_steps",
+        type=int,
+        default=None,
+        help="Total number of training steps to perform.  If provided, overrides num_train_epochs.",
+    )
+    parser.add_argument(
+        "--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("--adam_beta1", type=float, default=0.9, help="The beta1 parameter for the Adam optimizer.")
+    parser.add_argument("--adam_beta2", type=float, default=0.999, help="The beta2 parameter for the Adam optimizer.")
+    parser.add_argument("--adam_weight_decay", type=float, default=1e-2, help="Weight decay to use.")
+    parser.add_argument("--adam_epsilon", type=float, default=1e-08, help="Epsilon value for the Adam optimizer")
+    parser.add_argument("--max_grad_norm", default=1.0, type=float, help="Max gradient norm.")
+    parser.add_argument("--push_to_hub", action="store_true", help="Whether or not to push the model to the Hub.")
+    parser.add_argument("--hub_token", type=str, default=None, help="The token to use to push to the Model Hub.")
+    parser.add_argument(
+        "--hub_model_id",
+        type=str,
+        default=None,
+        help="The name of the repository to keep in sync with the local `output_dir`.",
+    )
+    parser.add_argument(
+        "--logging_dir",
+        type=str,
+        default="logs",
+        help=(
+            "[TensorBoard](https://www.tensorflow.org/tensorboard) log directory. Will default to"
+            " *output_dir/runs/**CURRENT_DATETIME_HOSTNAME***."
+        ),
+    )
+    parser.add_argument(
+        "--report_to",
+        type=str,
+        default="tensorboard",
+        help=(
+            'The integration to report the results and logs to. Supported platforms are `"tensorboard"`'
+            ' (default), `"wandb"` and `"comet_ml"`. Use `"all"` to report to all integrations.'
+        ),
+    )
+    parser.add_argument(
+        "--mixed_precision",
+        type=str,
+        default="no",
+        choices=["no", "fp16", "bf16"],
+        help=(
+            "Whether to use mixed precision. Choose"
+            "between fp16 and bf16 (bfloat16). Bf16 requires PyTorch >= 1.10."
+            "and an Nvidia Ampere GPU."
+        ),
+    )
+    parser.add_argument("--local_rank", type=int, default=-1, help="For distributed training: local_rank")
+    parser.add_argument(
+        "--from_pt",
+        action="store_true",
+        default=False,
+        help="Flag to indicate whether to convert models from PyTorch.",
+    )
+
+    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
+
+    # Sanity checks
+    if args.dataset_name is None and args.train_data_dir is None:
+        raise ValueError("Need either a dataset name or a training folder.")
+
+    return args
+
+
+dataset_name_mapping = {
+    "lambdalabs/naruto-blip-captions": ("image", "text"),
+}
+
+
+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.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.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()
+
+    if args.seed is not None:
+        set_seed(args.seed)
+
+    # Handle the repository creation
+    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
+
+    # 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:
+        # Downloading and loading a dataset from the hub.
+        dataset = load_dataset(
+            args.dataset_name, args.dataset_config_name, cache_dir=args.cache_dir, data_dir=args.train_data_dir
+        )
+    else:
+        data_files = {}
+        if args.train_data_dir is not None:
+            data_files["train"] = os.path.join(args.train_data_dir, "**")
+        dataset = load_dataset(
+            "imagefolder",
+            data_files=data_files,
+            cache_dir=args.cache_dir,
+        )
+        # See more about loading custom images at
+        # https://huggingface.co/docs/datasets/v2.4.0/en/image_load#imagefolder
+
+    # Preprocessing the datasets.
+    # We need to tokenize inputs and targets.
+    column_names = dataset["train"].column_names
+
+    # 6. Get the column names for input/target.
+    dataset_columns = dataset_name_mapping.get(args.dataset_name, None)
+    if args.image_column is None:
+        image_column = dataset_columns[0] if dataset_columns is not None else column_names[0]
+    else:
+        image_column = args.image_column
+        if image_column not in column_names:
+            raise ValueError(
+                f"--image_column' value '{args.image_column}' needs to be one of: {', '.join(column_names)}"
+            )
+    if args.caption_column is None:
+        caption_column = dataset_columns[1] if dataset_columns is not None else column_names[1]
+    else:
+        caption_column = args.caption_column
+        if caption_column not in column_names:
+            raise ValueError(
+                f"--caption_column' value '{args.caption_column}' needs to be one of: {', '.join(column_names)}"
+            )
+
+    # Preprocessing the datasets.
+    # We need to tokenize input captions and transform the images.
+    def tokenize_captions(examples, is_train=True):
+        captions = []
+        for caption in examples[caption_column]:
+            if isinstance(caption, str):
+                captions.append(caption)
+            elif isinstance(caption, (list, np.ndarray)):
+                # take a random caption if there are multiple
+                captions.append(random.choice(caption) if is_train else caption[0])
+            else:
+                raise ValueError(
+                    f"Caption column `{caption_column}` should contain either strings or lists of strings."
+                )
+        inputs = tokenizer(captions, max_length=tokenizer.model_max_length, padding="do_not_pad", truncation=True)
+        input_ids = inputs.input_ids
+        return input_ids
+
+    train_transforms = 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 preprocess_train(examples):
+        images = [image.convert("RGB") for image in examples[image_column]]
+        examples["pixel_values"] = [train_transforms(image) for image in images]
+        examples["input_ids"] = tokenize_captions(examples)
+
+        return examples
+
+    if args.max_train_samples is not None:
+        dataset["train"] = dataset["train"].shuffle(seed=args.seed).select(range(args.max_train_samples))
+        # Set the training transforms
+    train_dataset = dataset["train"].with_transform(preprocess_train)
+
+    def collate_fn(examples):
+        pixel_values = torch.stack([example["pixel_values"] for example in examples])
+        pixel_values = pixel_values.to(memory_format=torch.contiguous_format).float()
+        input_ids = [example["input_ids"] for example in examples]
+
+        padded_tokens = tokenizer.pad(
+            {"input_ids": input_ids}, padding="max_length", max_length=tokenizer.model_max_length, return_tensors="pt"
+        )
+        batch = {
+            "pixel_values": pixel_values,
+            "input_ids": padded_tokens.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, shuffle=True, collate_fn=collate_fn, batch_size=total_train_batch_size, drop_last=True
+    )
+
+    weight_dtype = jnp.float32
+    if args.mixed_precision == "fp16":
+        weight_dtype = jnp.float16
+    elif args.mixed_precision == "bf16":
+        weight_dtype = jnp.bfloat16
+
+    # Load models and create wrapper for stable diffusion
+    tokenizer = CLIPTokenizer.from_pretrained(
+        args.pretrained_model_name_or_path,
+        from_pt=args.from_pt,
+        revision=args.revision,
+        subfolder="tokenizer",
+    )
+    text_encoder = FlaxCLIPTextModel.from_pretrained(
+        args.pretrained_model_name_or_path,
+        from_pt=args.from_pt,
+        revision=args.revision,
+        subfolder="text_encoder",
+        dtype=weight_dtype,
+    )
+    vae, vae_params = FlaxAutoencoderKL.from_pretrained(
+        args.pretrained_model_name_or_path,
+        from_pt=args.from_pt,
+        revision=args.revision,
+        subfolder="vae",
+        dtype=weight_dtype,
+    )
+    unet, unet_params = FlaxUNet2DConditionModel.from_pretrained(
+        args.pretrained_model_name_or_path,
+        from_pt=args.from_pt,
+        revision=args.revision,
+        subfolder="unet",
+        dtype=weight_dtype,
+    )
+
+    # Optimization
+    if args.scale_lr:
+        args.learning_rate = args.learning_rate * total_train_batch_size
+
+    constant_scheduler = optax.constant_schedule(args.learning_rate)
+
+    adamw = optax.adamw(
+        learning_rate=constant_scheduler,
+        b1=args.adam_beta1,
+        b2=args.adam_beta2,
+        eps=args.adam_epsilon,
+        weight_decay=args.adam_weight_decay,
+    )
+
+    optimizer = optax.chain(
+        optax.clip_by_global_norm(args.max_grad_norm),
+        adamw,
+    )
+
+    state = train_state.TrainState.create(apply_fn=unet.__call__, params=unet_params, tx=optimizer)
+
+    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
+    rng = jax.random.PRNGKey(args.seed)
+    train_rngs = jax.random.split(rng, jax.local_device_count())
+
+    def train_step(state, text_encoder_params, vae_params, batch, train_rng):
+        dropout_rng, sample_rng, new_train_rng = jax.random.split(train_rng, 3)
+
+        def compute_loss(params):
+            # Convert images to latent space
+            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
+
+            # Sample noise that we'll add to the latents
+            noise_rng, timestep_rng = jax.random.split(sample_rng)
+            noise = jax.random.normal(noise_rng, latents.shape)
+            # Sample a random timestep for each image
+            bsz = latents.shape[0]
+            timesteps = jax.random.randint(
+                timestep_rng,
+                (bsz,),
+                0,
+                noise_scheduler.config.num_train_timesteps,
+            )
+
+            # 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(noise_scheduler_state, latents, noise, timesteps)
+
+            # Get the text embedding for conditioning
+            encoder_hidden_states = text_encoder(
+                batch["input_ids"],
+                params=text_encoder_params,
+                train=False,
+            )[0]
+
+            # Predict the noise residual and compute loss
+            model_pred = unet.apply(
+                {"params": params}, noisy_latents, timesteps, encoder_hidden_states, train=True
+            ).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)
+
+        metrics = {"loss": loss}
+        metrics = jax.lax.pmean(metrics, axis_name="batch")
+
+        return new_state, metrics, new_train_rng
+
+    # Create parallel version of the train step
+    p_train_step = jax.pmap(train_step, "batch", donate_argnums=(0,))
+
+    # Replicate the train state on each device
+    state = jax_utils.replicate(state)
+    text_encoder_params = jax_utils.replicate(text_encoder.params)
+    vae_params = jax_utils.replicate(vae_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="Epoch ... ", 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, text_encoder_params, vae_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
+
+        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(text_encoder_params),
+                "vae": get_params_to_save(vae_params),
+                "unet": get_params_to_save(state.params),
+                "safety_checker": safety_checker.params,
+            },
+        )
+
+        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()

train_text_to_image_lora.py

@@ -0,0 +1,975 @@
+#!/usr/bin/env python
+# coding=utf-8
+# Copyright 2025 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
+# limitations under the License.
+"""Fine-tuning script for Stable Diffusion for text2image with support for LoRA."""
+
+import argparse
+import logging
+import math
+import os
+import random
+import shutil
+from contextlib import nullcontext
+from pathlib import Path
+
+import datasets
+import numpy as np
+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 datasets import load_dataset
+from huggingface_hub import create_repo, upload_folder
+from packaging import version
+from peft import LoraConfig
+from peft.utils import get_peft_model_state_dict
+from torchvision import transforms
+from tqdm.auto import tqdm
+from transformers import CLIPTextModel, CLIPTokenizer
+
+import diffusers
+from diffusers import AutoencoderKL, DDPMScheduler, DiffusionPipeline, StableDiffusionPipeline, UNet2DConditionModel
+from diffusers.optimization import get_scheduler
+from diffusers.training_utils import cast_training_params, compute_snr
+from diffusers.utils import check_min_version, convert_state_dict_to_diffusers, 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
+from diffusers.utils.torch_utils import is_compiled_module
+
+
+if is_wandb_available():
+    import wandb
+
+# Will error if the minimal version of diffusers is not installed. Remove at your own risks.
+check_min_version("0.33.0.dev0")
+
+logger = get_logger(__name__, log_level="INFO")
+
+
+def save_model_card(
+    repo_id: str,
+    images: list = None,
+    base_model: str = None,
+    dataset_name: 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"""
+# LoRA text2image fine-tuning - {repo_id}
+These are LoRA adaption weights for {base_model}. The weights were fine-tuned on the {dataset_name} dataset. 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",
+        "diffusers-training",
+        "lora",
+    ]
+    model_card = populate_model_card(model_card, tags=tags)
+
+    model_card.save(os.path.join(repo_folder, "README.md"))
+
+
+def log_validation(
+    pipeline,
+    args,
+    accelerator,
+    epoch,
+    is_final_validation=False,
+):
+    logger.info(
+        f"Running validation... \n Generating {args.num_validation_images} images with prompt:"
+        f" {args.validation_prompt}."
+    )
+    pipeline = pipeline.to(accelerator.device)
+    pipeline.set_progress_bar_config(disable=True)
+    generator = torch.Generator(device=accelerator.device)
+    if args.seed is not None:
+        generator = generator.manual_seed(args.seed)
+    images = []
+    if torch.backends.mps.is_available():
+        autocast_ctx = nullcontext()
+    else:
+        autocast_ctx = torch.autocast(accelerator.device.type)
+
+    with autocast_ctx:
+        for _ in range(args.num_validation_images):
+            images.append(pipeline(args.validation_prompt, num_inference_steps=30, generator=generator).images[0])
+
+    for tracker in accelerator.trackers:
+        phase_name = "test" if is_final_validation else "validation"
+        if tracker.name == "tensorboard":
+            np_images = np.stack([np.asarray(img) for img in images])
+            tracker.writer.add_images(phase_name, np_images, epoch, dataformats="NHWC")
+        if tracker.name == "wandb":
+            tracker.log(
+                {
+                    phase_name: [
+                        wandb.Image(image, caption=f"{i}: {args.validation_prompt}") for i, image in enumerate(images)
+                    ]
+                }
+            )
+    return images
+
+
+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(
+        "--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(
+        "--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 🤗 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(
+        "--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(
+        "--image_column", type=str, default="image", help="The column of the dataset containing an image."
+    )
+    parser.add_argument(
+        "--caption_column",
+        type=str,
+        default="text",
+        help="The column of the dataset containing a caption or a list of captions.",
+    )
+    parser.add_argument(
+        "--validation_prompt", type=str, default=None, help="A prompt that is sampled during training for inference."
+    )
+    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_epochs",
+        type=int,
+        default=1,
+        help=(
+            "Run fine-tuning validation every X epochs. The validation process consists of running the prompt"
+            " `args.validation_prompt` multiple times: `args.num_validation_images`."
+        ),
+    )
+    parser.add_argument(
+        "--max_train_samples",
+        type=int,
+        default=None,
+        help=(
+            "For debugging purposes or quicker training, truncate the number of training examples to this "
+            "value if set."
+        ),
+    )
+    parser.add_argument(
+        "--output_dir",
+        type=str,
+        default="sd-model-finetuned-lora",
+        help="The output directory where the model predictions and checkpoints will be written.",
+    )
+    parser.add_argument(
+        "--cache_dir",
+        type=str,
+        default=None,
+        help="The directory where the downloaded models and datasets will be stored.",
+    )
+    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",
+        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",
+        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("--num_train_epochs", type=int, default=100)
+    parser.add_argument(
+        "--max_train_steps",
+        type=int,
+        default=None,
+        help="Total number of training steps to perform.  If provided, overrides num_train_epochs.",
+    )
+    parser.add_argument(
+        "--gradient_accumulation_steps",
+        type=int,
+        default=1,
+        help="Number of updates steps to accumulate before performing a backward/update pass.",
+    )
+    parser.add_argument(
+        "--gradient_checkpointing",
+        action="store_true",
+        help="Whether or not to use gradient checkpointing to save memory at the expense of slower backward pass.",
+    )
+    parser.add_argument(
+        "--learning_rate",
+        type=float,
+        default=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(
+        "--snr_gamma",
+        type=float,
+        default=None,
+        help="SNR weighting gamma to be used if rebalancing the loss. Recommended value is 5.0. "
+        "More details here: https://arxiv.org/abs/2303.09556.",
+    )
+    parser.add_argument(
+        "--use_8bit_adam", action="store_true", help="Whether or not to use 8-bit Adam from bitsandbytes."
+    )
+    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(
+        "--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("--max_grad_norm", default=1.0, type=float, help="Max gradient norm.")
+    parser.add_argument("--push_to_hub", action="store_true", help="Whether or not to push the model to the Hub.")
+    parser.add_argument("--hub_token", type=str, default=None, help="The token to use to push to the Model Hub.")
+    parser.add_argument(
+        "--prediction_type",
+        type=str,
+        default=None,
+        help="The prediction_type that shall be used for training. Choose between 'epsilon' or 'v_prediction' or leave `None`. If left to `None` the default prediction type of the scheduler: `noise_scheduler.config.prediction_type` is chosen.",
+    )
+    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=None,
+        choices=["no", "fp16", "bf16"],
+        help=(
+            "Whether to use mixed precision. Choose between fp16 and bf16 (bfloat16). Bf16 requires PyTorch >="
+            " 1.10.and an Nvidia Ampere GPU.  Default to the value of accelerate config of the current system or the"
+            " flag passed with the `accelerate.launch` command. Use this argument to override the accelerate config."
+        ),
+    )
+    parser.add_argument(
+        "--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("--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("--noise_offset", type=float, default=0, help="The scale of noise offset.")
+    parser.add_argument(
+        "--rank",
+        type=int,
+        default=4,
+        help=("The dimension of the LoRA update matrices."),
+    )
+
+    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
+
+    # Sanity checks
+    if args.dataset_name is None and args.train_data_dir is None:
+        raise ValueError("Need either a dataset name or a training folder.")
+
+    return args
+
+
+DATASET_NAME_MAPPING = {
+    "lambdalabs/naruto-blip-captions": ("image", "text"),
+}
+
+
+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 = Path(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,
+    )
+
+    # Disable AMP for MPS.
+    if torch.backends.mps.is_available():
+        accelerator.native_amp = False
+
+    # Make one log on every process with the configuration for debugging.
+    logging.basicConfig(
+        format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
+        datefmt="%m/%d/%Y %H:%M:%S",
+        level=logging.INFO,
+    )
+    logger.info(accelerator.state, main_process_only=False)
+    if accelerator.is_local_main_process:
+        datasets.utils.logging.set_verbosity_warning()
+        transformers.utils.logging.set_verbosity_warning()
+        diffusers.utils.logging.set_verbosity_info()
+    else:
+        datasets.utils.logging.set_verbosity_error()
+        transformers.utils.logging.set_verbosity_error()
+        diffusers.utils.logging.set_verbosity_error()
+
+    # If passed along, set the training seed now.
+    if args.seed is not None:
+        set_seed(args.seed)
+
+    # 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 scheduler, tokenizer and models.
+    noise_scheduler = DDPMScheduler.from_pretrained(args.pretrained_model_name_or_path, subfolder="scheduler")
+    tokenizer = CLIPTokenizer.from_pretrained(
+        args.pretrained_model_name_or_path, subfolder="tokenizer", revision=args.revision
+    )
+    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
+    )
+    # freeze parameters of models to save more memory
+    unet.requires_grad_(False)
+    vae.requires_grad_(False)
+    text_encoder.requires_grad_(False)
+
+    # For mixed precision training we cast all non-trainable weights (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
+
+    unet_lora_config = LoraConfig(
+        r=args.rank,
+        lora_alpha=args.rank,
+        init_lora_weights="gaussian",
+        target_modules=["to_k", "to_q", "to_v", "to_out.0"],
+    )
+
+    # Move unet, vae and text_encoder to device and cast to weight_dtype
+    unet.to(accelerator.device, dtype=weight_dtype)
+    vae.to(accelerator.device, dtype=weight_dtype)
+    text_encoder.to(accelerator.device, dtype=weight_dtype)
+
+    # Add adapter and make sure the trainable params are in float32.
+    unet.add_adapter(unet_lora_config)
+    if args.mixed_precision == "fp16":
+        # only upcast trainable parameters (LoRA) into fp32
+        cast_training_params(unet, dtype=torch.float32)
+
+    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")
+
+    lora_layers = filter(lambda p: p.requires_grad, unet.parameters())
+
+    if args.gradient_checkpointing:
+        unet.enable_gradient_checkpointing()
+
+    # 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(
+                "Please install bitsandbytes to use 8-bit Adam. You can do so by running `pip install bitsandbytes`"
+            )
+
+        optimizer_cls = bnb.optim.AdamW8bit
+    else:
+        optimizer_cls = torch.optim.AdamW
+
+    optimizer = optimizer_cls(
+        lora_layers,
+        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:
+        # Downloading and loading a dataset from the hub.
+        dataset = load_dataset(
+            args.dataset_name,
+            args.dataset_config_name,
+            cache_dir=args.cache_dir,
+            data_dir=args.train_data_dir,
+        )
+    else:
+        data_files = {}
+        if args.train_data_dir is not None:
+            data_files["train"] = os.path.join(args.train_data_dir, "**")
+        dataset = load_dataset(
+            "imagefolder",
+            data_files=data_files,
+            cache_dir=args.cache_dir,
+        )
+        # See more about loading custom images at
+        # https://huggingface.co/docs/datasets/v2.4.0/en/image_load#imagefolder
+
+    # Preprocessing the datasets.
+    # We need to tokenize inputs and targets.
+    column_names = dataset["train"].column_names
+
+    # 6. Get the column names for input/target.
+    dataset_columns = DATASET_NAME_MAPPING.get(args.dataset_name, None)
+    if args.image_column is None:
+        image_column = dataset_columns[0] if dataset_columns is not None else column_names[0]
+    else:
+        image_column = args.image_column
+        if image_column not in column_names:
+            raise ValueError(
+                f"--image_column' value '{args.image_column}' needs to be one of: {', '.join(column_names)}"
+            )
+    if args.caption_column is None:
+        caption_column = dataset_columns[1] if dataset_columns is not None else column_names[1]
+    else:
+        caption_column = args.caption_column
+        if caption_column not in column_names:
+            raise ValueError(
+                f"--caption_column' value '{args.caption_column}' needs to be one of: {', '.join(column_names)}"
+            )
+
+    # Preprocessing the datasets.
+    # We need to tokenize input captions and transform the images.
+    def tokenize_captions(examples, is_train=True):
+        captions = []
+        for caption in examples[caption_column]:
+            if isinstance(caption, str):
+                captions.append(caption)
+            elif isinstance(caption, (list, np.ndarray)):
+                # take a random caption if there are multiple
+                captions.append(random.choice(caption) if is_train else caption[0])
+            else:
+                raise ValueError(
+                    f"Caption column `{caption_column}` should contain either strings or lists of strings."
+                )
+        inputs = tokenizer(
+            captions, max_length=tokenizer.model_max_length, padding="max_length", truncation=True, return_tensors="pt"
+        )
+        return inputs.input_ids
+
+    # Preprocessing the datasets.
+    train_transforms = 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 unwrap_model(model):
+        model = accelerator.unwrap_model(model)
+        model = model._orig_mod if is_compiled_module(model) else model
+        return model
+
+    def preprocess_train(examples):
+        images = [image.convert("RGB") for image in examples[image_column]]
+        examples["pixel_values"] = [train_transforms(image) for image in images]
+        examples["input_ids"] = tokenize_captions(examples)
+        return examples
+
+    with accelerator.main_process_first():
+        if args.max_train_samples is not None:
+            dataset["train"] = dataset["train"].shuffle(seed=args.seed).select(range(args.max_train_samples))
+        # Set the training transforms
+        train_dataset = dataset["train"].with_transform(preprocess_train)
+
+    def collate_fn(examples):
+        pixel_values = torch.stack([example["pixel_values"] for example in examples])
+        pixel_values = pixel_values.to(memory_format=torch.contiguous_format).float()
+        input_ids = torch.stack([example["input_ids"] for example in examples])
+        return {"pixel_values": pixel_values, "input_ids": input_ids}
+
+    # DataLoaders creation:
+    train_dataloader = torch.utils.data.DataLoader(
+        train_dataset,
+        shuffle=True,
+        collate_fn=collate_fn,
+        batch_size=args.train_batch_size,
+        num_workers=args.dataloader_num_workers,
+    )
+
+    # Scheduler and math around the number of training steps.
+    # Check the PR https://github.com/huggingface/diffusers/pull/8312 for detailed explanation.
+    num_warmup_steps_for_scheduler = args.lr_warmup_steps * accelerator.num_processes
+    if args.max_train_steps is None:
+        len_train_dataloader_after_sharding = math.ceil(len(train_dataloader) / accelerator.num_processes)
+        num_update_steps_per_epoch = math.ceil(len_train_dataloader_after_sharding / args.gradient_accumulation_steps)
+        num_training_steps_for_scheduler = (
+            args.num_train_epochs * num_update_steps_per_epoch * accelerator.num_processes
+        )
+    else:
+        num_training_steps_for_scheduler = args.max_train_steps * accelerator.num_processes
+
+    lr_scheduler = get_scheduler(
+        args.lr_scheduler,
+        optimizer=optimizer,
+        num_warmup_steps=num_warmup_steps_for_scheduler,
+        num_training_steps=num_training_steps_for_scheduler,
+    )
+
+    # Prepare everything with our `accelerator`.
+    unet, optimizer, train_dataloader, lr_scheduler = accelerator.prepare(
+        unet, optimizer, train_dataloader, lr_scheduler
+    )
+
+    # We need to recalculate our total training steps as the size of the training dataloader may have changed.
+    num_update_steps_per_epoch = math.ceil(len(train_dataloader) / args.gradient_accumulation_steps)
+    if args.max_train_steps is None:
+        args.max_train_steps = args.num_train_epochs * num_update_steps_per_epoch
+        if num_training_steps_for_scheduler != args.max_train_steps * accelerator.num_processes:
+            logger.warning(
+                f"The length of the 'train_dataloader' after 'accelerator.prepare' ({len(train_dataloader)}) does not match "
+                f"the expected length ({len_train_dataloader_after_sharding}) when the learning rate scheduler was created. "
+                f"This inconsistency may result in the learning rate scheduler not functioning properly."
+            )
+    # 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("text2image-fine-tune", 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,
+    )
+
+    for epoch in range(first_epoch, args.num_train_epochs):
+        unet.train()
+        train_loss = 0.0
+        for step, batch in enumerate(train_dataloader):
+            with accelerator.accumulate(unet):
+                # Convert images to latent space
+                latents = vae.encode(batch["pixel_values"].to(dtype=weight_dtype)).latent_dist.sample()
+                latents = latents * vae.config.scaling_factor
+
+                # Sample noise that we'll add to the latents
+                noise = torch.randn_like(latents)
+                if args.noise_offset:
+                    # https://www.crosslabs.org//blog/diffusion-with-offset-noise
+                    noise += args.noise_offset * torch.randn(
+                        (latents.shape[0], latents.shape[1], 1, 1), device=latents.device
+                    )
+
+                bsz = latents.shape[0]
+                # Sample a random timestep for each image
+                timesteps = torch.randint(0, noise_scheduler.config.num_train_timesteps, (bsz,), device=latents.device)
+                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"], return_dict=False)[0]
+
+                # Get the target for loss depending on the prediction type
+                if args.prediction_type is not None:
+                    # set prediction_type of scheduler if defined
+                    noise_scheduler.register_to_config(prediction_type=args.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}")
+
+                # Predict the noise residual and compute loss
+                model_pred = unet(noisy_latents, timesteps, encoder_hidden_states, return_dict=False)[0]
+
+                if args.snr_gamma is None:
+                    loss = F.mse_loss(model_pred.float(), target.float(), reduction="mean")
+                else:
+                    # Compute loss-weights as per Section 3.4 of https://arxiv.org/abs/2303.09556.
+                    # Since we predict the noise instead of x_0, the original formulation is slightly changed.
+                    # This is discussed in Section 4.2 of the same paper.
+                    snr = compute_snr(noise_scheduler, timesteps)
+                    mse_loss_weights = torch.stack([snr, args.snr_gamma * torch.ones_like(timesteps)], dim=1).min(
+                        dim=1
+                    )[0]
+                    if noise_scheduler.config.prediction_type == "epsilon":
+                        mse_loss_weights = mse_loss_weights / snr
+                    elif noise_scheduler.config.prediction_type == "v_prediction":
+                        mse_loss_weights = mse_loss_weights / (snr + 1)
+
+                    loss = F.mse_loss(model_pred.float(), target.float(), reduction="none")
+                    loss = loss.mean(dim=list(range(1, len(loss.shape)))) * mse_loss_weights
+                    loss = loss.mean()
+
+                # Gather the losses across all processes for logging (if we use distributed training).
+                avg_loss = accelerator.gather(loss.repeat(args.train_batch_size)).mean()
+                train_loss += avg_loss.item() / args.gradient_accumulation_steps
+
+                # Backpropagate
+                accelerator.backward(loss)
+                if accelerator.sync_gradients:
+                    params_to_clip = lora_layers
+                    accelerator.clip_grad_norm_(params_to_clip, args.max_grad_norm)
+                optimizer.step()
+                lr_scheduler.step()
+                optimizer.zero_grad()
+
+            # Checks if the accelerator has performed an optimization step behind the scenes
+            if accelerator.sync_gradients:
+                progress_bar.update(1)
+                global_step += 1
+                accelerator.log({"train_loss": train_loss}, step=global_step)
+                train_loss = 0.0
+
+                if global_step % args.checkpointing_steps == 0:
+                    if accelerator.is_main_process:
+                        # _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)
+
+                        unwrapped_unet = unwrap_model(unet)
+                        unet_lora_state_dict = convert_state_dict_to_diffusers(
+                            get_peft_model_state_dict(unwrapped_unet)
+                        )
+
+                        StableDiffusionPipeline.save_lora_weights(
+                            save_directory=save_path,
+                            unet_lora_layers=unet_lora_state_dict,
+                            safe_serialization=True,
+                        )
+
+                        logger.info(f"Saved state to {save_path}")
+
+            logs = {"step_loss": loss.detach().item(), "lr": lr_scheduler.get_last_lr()[0]}
+            progress_bar.set_postfix(**logs)
+
+            if global_step >= args.max_train_steps:
+                break
+
+        if accelerator.is_main_process:
+            if args.validation_prompt is not None and epoch % args.validation_epochs == 0:
+                # create pipeline
+                pipeline = DiffusionPipeline.from_pretrained(
+                    args.pretrained_model_name_or_path,
+                    unet=unwrap_model(unet),
+                    revision=args.revision,
+                    variant=args.variant,
+                    torch_dtype=weight_dtype,
+                )
+                images = log_validation(pipeline, args, accelerator, epoch)
+
+                del pipeline
+                torch.cuda.empty_cache()
+
+    # Save the lora layers
+    accelerator.wait_for_everyone()
+    if accelerator.is_main_process:
+        unet = unet.to(torch.float32)
+
+        unwrapped_unet = unwrap_model(unet)
+        unet_lora_state_dict = convert_state_dict_to_diffusers(get_peft_model_state_dict(unwrapped_unet))
+        StableDiffusionPipeline.save_lora_weights(
+            save_directory=args.output_dir,
+            unet_lora_layers=unet_lora_state_dict,
+            safe_serialization=True,
+        )
+
+        # Final inference
+        # Load previous pipeline
+        if args.validation_prompt is not None:
+            pipeline = DiffusionPipeline.from_pretrained(
+                args.pretrained_model_name_or_path,
+                revision=args.revision,
+                variant=args.variant,
+                torch_dtype=weight_dtype,
+            )
+
+            # load attention processors
+            pipeline.load_lora_weights(args.output_dir)
+
+            # run inference
+            images = log_validation(pipeline, args, accelerator, epoch, is_final_validation=True)
+
+        if args.push_to_hub:
+            save_model_card(
+                repo_id,
+                images=images,
+                base_model=args.pretrained_model_name_or_path,
+                dataset_name=args.dataset_name,
+                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()

train_text_to_image_lora_sdxl.py

@@ -0,0 +1,1327 @@
+#!/usr/bin/env python
+# coding=utf-8
+# Copyright 2025 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
+# limitations under the License.
+"""Fine-tuning script for Stable Diffusion XL for text2image with support for LoRA."""
+
+import argparse
+import logging
+import math
+import os
+import random
+import shutil
+from contextlib import nullcontext
+from pathlib import Path
+
+import datasets
+import numpy as np
+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 DistributedDataParallelKwargs, DistributedType, ProjectConfiguration, set_seed
+from datasets import load_dataset
+from huggingface_hub import create_repo, upload_folder
+from packaging import version
+from peft import LoraConfig, set_peft_model_state_dict
+from peft.utils import get_peft_model_state_dict
+from torchvision import transforms
+from torchvision.transforms.functional import crop
+from tqdm.auto import tqdm
+from transformers import AutoTokenizer, PretrainedConfig
+
+import diffusers
+from diffusers import (
+    AutoencoderKL,
+    DDPMScheduler,
+    StableDiffusionXLPipeline,
+    UNet2DConditionModel,
+)
+from diffusers.loaders import StableDiffusionLoraLoaderMixin
+from diffusers.optimization import get_scheduler
+from diffusers.training_utils import _set_state_dict_into_text_encoder, cast_training_params, compute_snr
+from diffusers.utils import (
+    check_min_version,
+    convert_state_dict_to_diffusers,
+    convert_unet_state_dict_to_peft,
+    is_wandb_available,
+)
+from diffusers.utils.hub_utils import load_or_create_model_card, populate_model_card
+from diffusers.utils.import_utils import is_torch_npu_available, is_xformers_available
+from diffusers.utils.torch_utils import is_compiled_module
+
+
+if is_wandb_available():
+    import wandb
+
+# Will error if the minimal version of diffusers is not installed. Remove at your own risks.
+check_min_version("0.33.0.dev0")
+
+logger = get_logger(__name__)
+if is_torch_npu_available():
+    torch.npu.config.allow_internal_format = False
+
+
+def save_model_card(
+    repo_id: str,
+    images: list = None,
+    base_model: str = None,
+    dataset_name: str = None,
+    train_text_encoder: bool = False,
+    repo_folder: str = None,
+    vae_path: 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"""
+# LoRA text2image fine-tuning - {repo_id}
+
+These are LoRA adaption weights for {base_model}. The weights were fine-tuned on the {dataset_name} dataset. You can find some example images in the following. \n
+{img_str}
+
+LoRA for the text encoder was enabled: {train_text_encoder}.
+
+Special VAE used for training: {vae_path}.
+"""
+    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",
+        "lora",
+    ]
+    model_card = populate_model_card(model_card, tags=tags)
+
+    model_card.save(os.path.join(repo_folder, "README.md"))
+
+
+def log_validation(
+    pipeline,
+    args,
+    accelerator,
+    epoch,
+    is_final_validation=False,
+):
+    logger.info(
+        f"Running validation... \n Generating {args.num_validation_images} images with prompt:"
+        f" {args.validation_prompt}."
+    )
+    pipeline = pipeline.to(accelerator.device)
+    pipeline.set_progress_bar_config(disable=True)
+
+    # run inference
+    generator = torch.Generator(device=accelerator.device).manual_seed(args.seed) if args.seed is not None else None
+    pipeline_args = {"prompt": args.validation_prompt}
+    if torch.backends.mps.is_available():
+        autocast_ctx = nullcontext()
+    else:
+        autocast_ctx = torch.autocast(accelerator.device.type)
+
+    with autocast_ctx:
+        images = [pipeline(**pipeline_args, generator=generator).images[0] for _ in range(args.num_validation_images)]
+
+    for tracker in accelerator.trackers:
+        phase_name = "test" if is_final_validation else "validation"
+        if tracker.name == "tensorboard":
+            np_images = np.stack([np.asarray(img) for img in images])
+            tracker.writer.add_images(phase_name, np_images, epoch, dataformats="NHWC")
+        if tracker.name == "wandb":
+            tracker.log(
+                {
+                    phase_name: [
+                        wandb.Image(image, caption=f"{i}: {args.validation_prompt}") for i, image in enumerate(images)
+                    ]
+                }
+            )
+    return images
+
+
+def import_model_class_from_model_name_or_path(
+    pretrained_model_name_or_path: str, revision: str, subfolder: str = "text_encoder"
+):
+    text_encoder_config = PretrainedConfig.from_pretrained(
+        pretrained_model_name_or_path, subfolder=subfolder, revision=revision
+    )
+    model_class = text_encoder_config.architectures[0]
+
+    if model_class == "CLIPTextModel":
+        from transformers import CLIPTextModel
+
+        return CLIPTextModel
+    elif model_class == "CLIPTextModelWithProjection":
+        from transformers import CLIPTextModelWithProjection
+
+        return CLIPTextModelWithProjection
+    else:
+        raise ValueError(f"{model_class} is not supported.")
+
+
+def parse_args(input_args=None):
+    parser = argparse.ArgumentParser(description="Simple example of a training script.")
+    parser.add_argument(
+        "--pretrained_model_name_or_path",
+        type=str,
+        default=None,
+        required=True,
+        help="Path to pretrained model or model identifier from huggingface.co/models.",
+    )
+    parser.add_argument(
+        "--pretrained_vae_model_name_or_path",
+        type=str,
+        default=None,
+        help="Path to pretrained VAE model with better numerical stability. More details: https://github.com/huggingface/diffusers/pull/4038.",
+    )
+    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(
+        "--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 🤗 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(
+        "--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(
+        "--image_column", type=str, default="image", help="The column of the dataset containing an image."
+    )
+    parser.add_argument(
+        "--caption_column",
+        type=str,
+        default="text",
+        help="The column of the dataset containing a caption or a list of captions.",
+    )
+    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_epochs",
+        type=int,
+        default=1,
+        help=(
+            "Run fine-tuning validation every X epochs. The validation process consists of running the prompt"
+            " `args.validation_prompt` multiple times: `args.num_validation_images`."
+        ),
+    )
+    parser.add_argument(
+        "--max_train_samples",
+        type=int,
+        default=None,
+        help=(
+            "For debugging purposes or quicker training, truncate the number of training examples to this "
+            "value if set."
+        ),
+    )
+    parser.add_argument(
+        "--output_dir",
+        type=str,
+        default="sd-model-finetuned-lora",
+        help="The output directory where the model predictions and checkpoints will be written.",
+    )
+    parser.add_argument(
+        "--cache_dir",
+        type=str,
+        default=None,
+        help="The directory where the downloaded models and datasets will be stored.",
+    )
+    parser.add_argument("--seed", type=int, default=None, help="A seed for reproducible training.")
+    parser.add_argument(
+        "--resolution",
+        type=int,
+        default=1024,
+        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",
+        action="store_true",
+        help="whether to randomly flip images horizontally",
+    )
+    parser.add_argument(
+        "--train_text_encoder",
+        action="store_true",
+        help="Whether to train the text encoder. If set, the text encoder should be float32 precision.",
+    )
+    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=None,
+        help="Total number of training steps to perform.  If provided, overrides num_train_epochs.",
+    )
+    parser.add_argument(
+        "--checkpointing_steps",
+        type=int,
+        default=500,
+        help=(
+            "Save a checkpoint of the training state every X updates. These checkpoints can be used both as final"
+            " checkpoints in case they are better than the last checkpoint, and are also suitable for resuming"
+            " training using `--resume_from_checkpoint`."
+        ),
+    )
+    parser.add_argument(
+        "--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(
+        "--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(
+        "--snr_gamma",
+        type=float,
+        default=None,
+        help="SNR weighting gamma to be used if rebalancing the loss. Recommended value is 5.0. "
+        "More details here: https://arxiv.org/abs/2303.09556.",
+    )
+    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(
+        "--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("--max_grad_norm", default=1.0, type=float, help="Max gradient norm.")
+    parser.add_argument("--push_to_hub", action="store_true", help="Whether or not to push the model to the Hub.")
+    parser.add_argument("--hub_token", type=str, default=None, help="The token to use to push to the Model Hub.")
+    parser.add_argument(
+        "--prediction_type",
+        type=str,
+        default=None,
+        help="The prediction_type that shall be used for training. Choose between 'epsilon' or 'v_prediction' or leave `None`. If left to `None` the default prediction type of the scheduler: `noise_scheduler.config.prediction_type` is chosen.",
+    )
+    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(
+        "--report_to",
+        type=str,
+        default="tensorboard",
+        help=(
+            'The integration to report the results and logs to. Supported platforms are `"tensorboard"`'
+            ' (default), `"wandb"` and `"comet_ml"`. Use `"all"` to report to all integrations.'
+        ),
+    )
+    parser.add_argument(
+        "--mixed_precision",
+        type=str,
+        default=None,
+        choices=["no", "fp16", "bf16"],
+        help=(
+            "Whether to use mixed precision. Choose between fp16 and bf16 (bfloat16). Bf16 requires PyTorch >="
+            " 1.10.and an Nvidia Ampere GPU.  Default to the value of accelerate config of the current system or the"
+            " flag passed with the `accelerate.launch` command. Use this argument to override the accelerate config."
+        ),
+    )
+    parser.add_argument("--local_rank", type=int, default=-1, help="For distributed training: local_rank")
+    parser.add_argument(
+        "--enable_xformers_memory_efficient_attention", action="store_true", help="Whether or not to use xformers."
+    )
+    parser.add_argument(
+        "--enable_npu_flash_attention", action="store_true", help="Whether or not to use npu flash attention."
+    )
+    parser.add_argument("--noise_offset", type=float, default=0, help="The scale of noise offset.")
+    parser.add_argument(
+        "--rank",
+        type=int,
+        default=4,
+        help=("The dimension of the LoRA update matrices."),
+    )
+    parser.add_argument(
+        "--debug_loss",
+        action="store_true",
+        help="debug loss for each image, if filenames are available in the dataset",
+    )
+
+    if input_args is not None:
+        args = parser.parse_args(input_args)
+    else:
+        args = parser.parse_args()
+
+    env_local_rank = int(os.environ.get("LOCAL_RANK", -1))
+    if env_local_rank != -1 and env_local_rank != args.local_rank:
+        args.local_rank = env_local_rank
+
+    # Sanity checks
+    if args.dataset_name is None and args.train_data_dir is None:
+        raise ValueError("Need either a dataset name or a training folder.")
+
+    return args
+
+
+DATASET_NAME_MAPPING = {
+    "lambdalabs/naruto-blip-captions": ("image", "text"),
+}
+
+
+def tokenize_prompt(tokenizer, prompt):
+    text_inputs = tokenizer(
+        prompt,
+        padding="max_length",
+        max_length=tokenizer.model_max_length,
+        truncation=True,
+        return_tensors="pt",
+    )
+    text_input_ids = text_inputs.input_ids
+    return text_input_ids
+
+
+# Adapted from pipelines.StableDiffusionXLPipeline.encode_prompt
+def encode_prompt(text_encoders, tokenizers, prompt, text_input_ids_list=None):
+    prompt_embeds_list = []
+
+    for i, text_encoder in enumerate(text_encoders):
+        if tokenizers is not None:
+            tokenizer = tokenizers[i]
+            text_input_ids = tokenize_prompt(tokenizer, prompt)
+        else:
+            assert text_input_ids_list is not None
+            text_input_ids = text_input_ids_list[i]
+
+        prompt_embeds = text_encoder(
+            text_input_ids.to(text_encoder.device), output_hidden_states=True, return_dict=False
+        )
+
+        # We are only ALWAYS interested in the pooled output of the final text encoder
+        pooled_prompt_embeds = prompt_embeds[0]
+        prompt_embeds = prompt_embeds[-1][-2]
+        bs_embed, seq_len, _ = prompt_embeds.shape
+        prompt_embeds = prompt_embeds.view(bs_embed, seq_len, -1)
+        prompt_embeds_list.append(prompt_embeds)
+
+    prompt_embeds = torch.concat(prompt_embeds_list, dim=-1)
+    pooled_prompt_embeds = pooled_prompt_embeds.view(bs_embed, -1)
+    return prompt_embeds, pooled_prompt_embeds
+
+
+def main(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 = Path(args.output_dir, args.logging_dir)
+
+    if torch.backends.mps.is_available() and args.mixed_precision == "bf16":
+        # due to pytorch#99272, MPS does not yet support bfloat16.
+        raise ValueError(
+            "Mixed precision training with bfloat16 is not supported on MPS. Please use fp16 (recommended) or fp32 instead."
+        )
+
+    accelerator_project_config = ProjectConfiguration(project_dir=args.output_dir, logging_dir=logging_dir)
+    kwargs = DistributedDataParallelKwargs(find_unused_parameters=True)
+    accelerator = Accelerator(
+        gradient_accumulation_steps=args.gradient_accumulation_steps,
+        mixed_precision=args.mixed_precision,
+        log_with=args.report_to,
+        project_config=accelerator_project_config,
+        kwargs_handlers=[kwargs],
+    )
+
+    # Make one log on every process with the configuration for debugging.
+    logging.basicConfig(
+        format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
+        datefmt="%m/%d/%Y %H:%M:%S",
+        level=logging.INFO,
+    )
+    logger.info(accelerator.state, main_process_only=False)
+    if accelerator.is_local_main_process:
+        datasets.utils.logging.set_verbosity_warning()
+        transformers.utils.logging.set_verbosity_warning()
+        diffusers.utils.logging.set_verbosity_info()
+    else:
+        datasets.utils.logging.set_verbosity_error()
+        transformers.utils.logging.set_verbosity_error()
+        diffusers.utils.logging.set_verbosity_error()
+
+    # If passed along, set the training seed now.
+    if args.seed is not None:
+        set_seed(args.seed)
+
+    # 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 the tokenizers
+    tokenizer_one = AutoTokenizer.from_pretrained(
+        args.pretrained_model_name_or_path,
+        subfolder="tokenizer",
+        revision=args.revision,
+        use_fast=False,
+    )
+    tokenizer_two = AutoTokenizer.from_pretrained(
+        args.pretrained_model_name_or_path,
+        subfolder="tokenizer_2",
+        revision=args.revision,
+        use_fast=False,
+    )
+
+    # import correct text encoder classes
+    text_encoder_cls_one = import_model_class_from_model_name_or_path(
+        args.pretrained_model_name_or_path, args.revision
+    )
+    text_encoder_cls_two = import_model_class_from_model_name_or_path(
+        args.pretrained_model_name_or_path, args.revision, subfolder="text_encoder_2"
+    )
+
+    # Load scheduler and models
+    noise_scheduler = DDPMScheduler.from_pretrained(args.pretrained_model_name_or_path, subfolder="scheduler")
+    text_encoder_one = text_encoder_cls_one.from_pretrained(
+        args.pretrained_model_name_or_path, subfolder="text_encoder", revision=args.revision, variant=args.variant
+    )
+    text_encoder_two = text_encoder_cls_two.from_pretrained(
+        args.pretrained_model_name_or_path, subfolder="text_encoder_2", revision=args.revision, variant=args.variant
+    )
+    vae_path = (
+        args.pretrained_model_name_or_path
+        if args.pretrained_vae_model_name_or_path is None
+        else args.pretrained_vae_model_name_or_path
+    )
+    vae = AutoencoderKL.from_pretrained(
+        vae_path,
+        subfolder="vae" if args.pretrained_vae_model_name_or_path is None else None,
+        revision=args.revision,
+        variant=args.variant,
+    )
+    unet = UNet2DConditionModel.from_pretrained(
+        args.pretrained_model_name_or_path, subfolder="unet", revision=args.revision, variant=args.variant
+    )
+
+    # We only train the additional adapter LoRA layers
+    vae.requires_grad_(False)
+    text_encoder_one.requires_grad_(False)
+    text_encoder_two.requires_grad_(False)
+    unet.requires_grad_(False)
+
+    # For mixed precision training we cast all non-trainable weights (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 unet, vae and text_encoder to device and cast to weight_dtype
+    # The VAE is in float32 to avoid NaN losses.
+    unet.to(accelerator.device, dtype=weight_dtype)
+
+    if args.pretrained_vae_model_name_or_path is None:
+        vae.to(accelerator.device, dtype=torch.float32)
+    else:
+        vae.to(accelerator.device, dtype=weight_dtype)
+    text_encoder_one.to(accelerator.device, dtype=weight_dtype)
+    text_encoder_two.to(accelerator.device, dtype=weight_dtype)
+
+    if args.enable_npu_flash_attention:
+        if is_torch_npu_available():
+            logger.info("npu flash attention enabled.")
+            unet.enable_npu_flash_attention()
+        else:
+            raise ValueError("npu flash attention requires torch_npu extensions and is supported only on npu devices.")
+
+    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")
+
+    # now we will add new LoRA weights to the attention layers
+    # Set correct lora layers
+    unet_lora_config = LoraConfig(
+        r=args.rank,
+        lora_alpha=args.rank,
+        init_lora_weights="gaussian",
+        target_modules=["to_k", "to_q", "to_v", "to_out.0"],
+    )
+
+    unet.add_adapter(unet_lora_config)
+
+    # The text encoder comes from 🤗 transformers, we will also attach adapters to it.
+    if args.train_text_encoder:
+        # ensure that dtype is float32, even if rest of the model that isn't trained is loaded in fp16
+        text_lora_config = LoraConfig(
+            r=args.rank,
+            lora_alpha=args.rank,
+            init_lora_weights="gaussian",
+            target_modules=["q_proj", "k_proj", "v_proj", "out_proj"],
+        )
+        text_encoder_one.add_adapter(text_lora_config)
+        text_encoder_two.add_adapter(text_lora_config)
+
+    def unwrap_model(model):
+        model = accelerator.unwrap_model(model)
+        model = model._orig_mod if is_compiled_module(model) else model
+        return model
+
+    # 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:
+            # there are only two options here. Either are just the unet attn processor layers
+            # or there are the unet and text encoder attn layers
+            unet_lora_layers_to_save = None
+            text_encoder_one_lora_layers_to_save = None
+            text_encoder_two_lora_layers_to_save = None
+
+            for model in models:
+                if isinstance(unwrap_model(model), type(unwrap_model(unet))):
+                    unet_lora_layers_to_save = convert_state_dict_to_diffusers(get_peft_model_state_dict(model))
+                elif isinstance(unwrap_model(model), type(unwrap_model(text_encoder_one))):
+                    text_encoder_one_lora_layers_to_save = convert_state_dict_to_diffusers(
+                        get_peft_model_state_dict(model)
+                    )
+                elif isinstance(unwrap_model(model), type(unwrap_model(text_encoder_two))):
+                    text_encoder_two_lora_layers_to_save = convert_state_dict_to_diffusers(
+                        get_peft_model_state_dict(model)
+                    )
+                else:
+                    raise ValueError(f"unexpected save model: {model.__class__}")
+
+                # make sure to pop weight so that corresponding model is not saved again
+                if weights:
+                    weights.pop()
+
+            StableDiffusionXLPipeline.save_lora_weights(
+                output_dir,
+                unet_lora_layers=unet_lora_layers_to_save,
+                text_encoder_lora_layers=text_encoder_one_lora_layers_to_save,
+                text_encoder_2_lora_layers=text_encoder_two_lora_layers_to_save,
+            )
+
+    def load_model_hook(models, input_dir):
+        unet_ = None
+        text_encoder_one_ = None
+        text_encoder_two_ = None
+
+        while len(models) > 0:
+            model = models.pop()
+
+            if isinstance(model, type(unwrap_model(unet))):
+                unet_ = model
+            elif isinstance(model, type(unwrap_model(text_encoder_one))):
+                text_encoder_one_ = model
+            elif isinstance(model, type(unwrap_model(text_encoder_two))):
+                text_encoder_two_ = model
+            else:
+                raise ValueError(f"unexpected save model: {model.__class__}")
+
+        lora_state_dict, _ = StableDiffusionLoraLoaderMixin.lora_state_dict(input_dir)
+        unet_state_dict = {f"{k.replace('unet.', '')}": v for k, v in lora_state_dict.items() if k.startswith("unet.")}
+        unet_state_dict = convert_unet_state_dict_to_peft(unet_state_dict)
+        incompatible_keys = set_peft_model_state_dict(unet_, unet_state_dict, adapter_name="default")
+        if incompatible_keys is not None:
+            # check only for unexpected keys
+            unexpected_keys = getattr(incompatible_keys, "unexpected_keys", None)
+            if unexpected_keys:
+                logger.warning(
+                    f"Loading adapter weights from state_dict led to unexpected keys not found in the model: "
+                    f" {unexpected_keys}. "
+                )
+
+        if args.train_text_encoder:
+            _set_state_dict_into_text_encoder(lora_state_dict, prefix="text_encoder.", text_encoder=text_encoder_one_)
+
+            _set_state_dict_into_text_encoder(
+                lora_state_dict, prefix="text_encoder_2.", text_encoder=text_encoder_two_
+            )
+
+        # Make sure the trainable params are in float32. This is again needed since the base models
+        # are in `weight_dtype`. More details:
+        # https://github.com/huggingface/diffusers/pull/6514#discussion_r1449796804
+        if args.mixed_precision == "fp16":
+            models = [unet_]
+            if args.train_text_encoder:
+                models.extend([text_encoder_one_, text_encoder_two_])
+            cast_training_params(models, dtype=torch.float32)
+
+    accelerator.register_save_state_pre_hook(save_model_hook)
+    accelerator.register_load_state_pre_hook(load_model_hook)
+
+    if args.gradient_checkpointing:
+        unet.enable_gradient_checkpointing()
+        if args.train_text_encoder:
+            text_encoder_one.gradient_checkpointing_enable()
+            text_encoder_two.gradient_checkpointing_enable()
+
+    # Enable TF32 for faster training on Ampere GPUs,
+    # cf https://pytorch.org/docs/stable/notes/cuda.html#tensorfloat-32-tf32-on-ampere-devices
+    if args.allow_tf32:
+        torch.backends.cuda.matmul.allow_tf32 = True
+
+    if args.scale_lr:
+        args.learning_rate = (
+            args.learning_rate * args.gradient_accumulation_steps * args.train_batch_size * accelerator.num_processes
+        )
+
+    # Make sure the trainable params are in float32.
+    if args.mixed_precision == "fp16":
+        models = [unet]
+        if args.train_text_encoder:
+            models.extend([text_encoder_one, text_encoder_two])
+        cast_training_params(models, dtype=torch.float32)
+
+    # Use 8-bit Adam for lower memory usage or to fine-tune the model in 16GB GPUs
+    if args.use_8bit_adam:
+        try:
+            import bitsandbytes as bnb
+        except ImportError:
+            raise ImportError(
+                "To use 8-bit Adam, please install the bitsandbytes library: `pip install bitsandbytes`."
+            )
+
+        optimizer_class = bnb.optim.AdamW8bit
+    else:
+        optimizer_class = torch.optim.AdamW
+
+    # Optimizer creation
+    params_to_optimize = list(filter(lambda p: p.requires_grad, unet.parameters()))
+    if args.train_text_encoder:
+        params_to_optimize = (
+            params_to_optimize
+            + list(filter(lambda p: p.requires_grad, text_encoder_one.parameters()))
+            + list(filter(lambda p: p.requires_grad, text_encoder_two.parameters()))
+        )
+    optimizer = optimizer_class(
+        params_to_optimize,
+        lr=args.learning_rate,
+        betas=(args.adam_beta1, args.adam_beta2),
+        weight_decay=args.adam_weight_decay,
+        eps=args.adam_epsilon,
+    )
+
+    # 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:
+        # Downloading and loading a dataset from the hub.
+        dataset = load_dataset(
+            args.dataset_name, args.dataset_config_name, cache_dir=args.cache_dir, data_dir=args.train_data_dir
+        )
+    else:
+        data_files = {}
+        if args.train_data_dir is not None:
+            data_files["train"] = os.path.join(args.train_data_dir, "**")
+        dataset = load_dataset(
+            "imagefolder",
+            data_files=data_files,
+            cache_dir=args.cache_dir,
+        )
+        # See more about loading custom images at
+        # https://huggingface.co/docs/datasets/v2.4.0/en/image_load#imagefolder
+
+    # Preprocessing the datasets.
+    # We need to tokenize inputs and targets.
+    column_names = dataset["train"].column_names
+
+    # 6. Get the column names for input/target.
+    dataset_columns = DATASET_NAME_MAPPING.get(args.dataset_name, None)
+    if args.image_column is None:
+        image_column = dataset_columns[0] if dataset_columns is not None else column_names[0]
+    else:
+        image_column = args.image_column
+        if image_column not in column_names:
+            raise ValueError(
+                f"--image_column' value '{args.image_column}' needs to be one of: {', '.join(column_names)}"
+            )
+    if args.caption_column is None:
+        caption_column = dataset_columns[1] if dataset_columns is not None else column_names[1]
+    else:
+        caption_column = args.caption_column
+        if caption_column not in column_names:
+            raise ValueError(
+                f"--caption_column' value '{args.caption_column}' needs to be one of: {', '.join(column_names)}"
+            )
+
+    # Preprocessing the datasets.
+    # We need to tokenize input captions and transform the images.
+    def tokenize_captions(examples, is_train=True):
+        captions = []
+        for caption in examples[caption_column]:
+            if isinstance(caption, str):
+                captions.append(caption)
+            elif isinstance(caption, (list, np.ndarray)):
+                # take a random caption if there are multiple
+                captions.append(random.choice(caption) if is_train else caption[0])
+            else:
+                raise ValueError(
+                    f"Caption column `{caption_column}` should contain either strings or lists of strings."
+                )
+        tokens_one = tokenize_prompt(tokenizer_one, captions)
+        tokens_two = tokenize_prompt(tokenizer_two, captions)
+        return tokens_one, tokens_two
+
+    # Preprocessing the datasets.
+    train_resize = transforms.Resize(args.resolution, interpolation=transforms.InterpolationMode.BILINEAR)
+    train_crop = transforms.CenterCrop(args.resolution) if args.center_crop else transforms.RandomCrop(args.resolution)
+    train_flip = transforms.RandomHorizontalFlip(p=1.0)
+    train_transforms = transforms.Compose(
+        [
+            transforms.ToTensor(),
+            transforms.Normalize([0.5], [0.5]),
+        ]
+    )
+
+    def preprocess_train(examples):
+        images = [image.convert("RGB") for image in examples[image_column]]
+        # image aug
+        original_sizes = []
+        all_images = []
+        crop_top_lefts = []
+        for image in images:
+            original_sizes.append((image.height, image.width))
+            image = train_resize(image)
+            if args.random_flip and random.random() < 0.5:
+                # flip
+                image = train_flip(image)
+            if args.center_crop:
+                y1 = max(0, int(round((image.height - args.resolution) / 2.0)))
+                x1 = max(0, int(round((image.width - args.resolution) / 2.0)))
+                image = train_crop(image)
+            else:
+                y1, x1, h, w = train_crop.get_params(image, (args.resolution, args.resolution))
+                image = crop(image, y1, x1, h, w)
+            crop_top_left = (y1, x1)
+            crop_top_lefts.append(crop_top_left)
+            image = train_transforms(image)
+            all_images.append(image)
+
+        examples["original_sizes"] = original_sizes
+        examples["crop_top_lefts"] = crop_top_lefts
+        examples["pixel_values"] = all_images
+        tokens_one, tokens_two = tokenize_captions(examples)
+        examples["input_ids_one"] = tokens_one
+        examples["input_ids_two"] = tokens_two
+        if args.debug_loss:
+            fnames = [os.path.basename(image.filename) for image in examples[image_column] if image.filename]
+            if fnames:
+                examples["filenames"] = fnames
+        return examples
+
+    with accelerator.main_process_first():
+        if args.max_train_samples is not None:
+            dataset["train"] = dataset["train"].shuffle(seed=args.seed).select(range(args.max_train_samples))
+        # Set the training transforms
+        train_dataset = dataset["train"].with_transform(preprocess_train, output_all_columns=True)
+
+    def collate_fn(examples):
+        pixel_values = torch.stack([example["pixel_values"] for example in examples])
+        pixel_values = pixel_values.to(memory_format=torch.contiguous_format).float()
+        original_sizes = [example["original_sizes"] for example in examples]
+        crop_top_lefts = [example["crop_top_lefts"] for example in examples]
+        input_ids_one = torch.stack([example["input_ids_one"] for example in examples])
+        input_ids_two = torch.stack([example["input_ids_two"] for example in examples])
+        result = {
+            "pixel_values": pixel_values,
+            "input_ids_one": input_ids_one,
+            "input_ids_two": input_ids_two,
+            "original_sizes": original_sizes,
+            "crop_top_lefts": crop_top_lefts,
+        }
+
+        filenames = [example["filenames"] for example in examples if "filenames" in example]
+        if filenames:
+            result["filenames"] = filenames
+        return result
+
+    # DataLoaders creation:
+    train_dataloader = torch.utils.data.DataLoader(
+        train_dataset,
+        shuffle=True,
+        collate_fn=collate_fn,
+        batch_size=args.train_batch_size,
+        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 * args.gradient_accumulation_steps,
+        num_training_steps=args.max_train_steps * args.gradient_accumulation_steps,
+    )
+
+    # Prepare everything with our `accelerator`.
+    if args.train_text_encoder:
+        unet, text_encoder_one, text_encoder_two, optimizer, train_dataloader, lr_scheduler = accelerator.prepare(
+            unet, text_encoder_one, text_encoder_two, optimizer, train_dataloader, lr_scheduler
+        )
+    else:
+        unet, optimizer, train_dataloader, lr_scheduler = accelerator.prepare(
+            unet, optimizer, train_dataloader, lr_scheduler
+        )
+
+    # We need to recalculate our total training steps as the size of the training dataloader may have changed.
+    num_update_steps_per_epoch = math.ceil(len(train_dataloader) / args.gradient_accumulation_steps)
+    if overrode_max_train_steps:
+        args.max_train_steps = args.num_train_epochs * num_update_steps_per_epoch
+    # Afterwards we recalculate our number of training epochs
+    args.num_train_epochs = math.ceil(args.max_train_steps / num_update_steps_per_epoch)
+
+    # We need to initialize the trackers we use, and also store our configuration.
+    # The trackers initializes automatically on the main process.
+    if accelerator.is_main_process:
+        accelerator.init_trackers("text2image-fine-tune", 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,
+    )
+
+    for epoch in range(first_epoch, args.num_train_epochs):
+        unet.train()
+        if args.train_text_encoder:
+            text_encoder_one.train()
+            text_encoder_two.train()
+        train_loss = 0.0
+        for step, batch in enumerate(train_dataloader):
+            with accelerator.accumulate(unet):
+                # Convert images to latent space
+                if args.pretrained_vae_model_name_or_path is not None:
+                    pixel_values = batch["pixel_values"].to(dtype=weight_dtype)
+                else:
+                    pixel_values = batch["pixel_values"]
+
+                model_input = vae.encode(pixel_values).latent_dist.sample()
+                model_input = model_input * vae.config.scaling_factor
+                if args.pretrained_vae_model_name_or_path is None:
+                    model_input = model_input.to(weight_dtype)
+
+                # Sample noise that we'll add to the latents
+                noise = torch.randn_like(model_input)
+                if args.noise_offset:
+                    # https://www.crosslabs.org//blog/diffusion-with-offset-noise
+                    noise += args.noise_offset * torch.randn(
+                        (model_input.shape[0], model_input.shape[1], 1, 1), device=model_input.device
+                    )
+
+                bsz = model_input.shape[0]
+                # Sample a random timestep for each image
+                timesteps = torch.randint(
+                    0, noise_scheduler.config.num_train_timesteps, (bsz,), device=model_input.device
+                )
+                timesteps = timesteps.long()
+
+                # Add noise to the model input according to the noise magnitude at each timestep
+                # (this is the forward diffusion process)
+                noisy_model_input = noise_scheduler.add_noise(model_input, noise, timesteps)
+
+                # time ids
+                def compute_time_ids(original_size, crops_coords_top_left):
+                    # Adapted from pipeline.StableDiffusionXLPipeline._get_add_time_ids
+                    target_size = (args.resolution, args.resolution)
+                    add_time_ids = list(original_size + crops_coords_top_left + target_size)
+                    add_time_ids = torch.tensor([add_time_ids])
+                    add_time_ids = add_time_ids.to(accelerator.device, dtype=weight_dtype)
+                    return add_time_ids
+
+                add_time_ids = torch.cat(
+                    [compute_time_ids(s, c) for s, c in zip(batch["original_sizes"], batch["crop_top_lefts"])]
+                )
+
+                # Predict the noise residual
+                unet_added_conditions = {"time_ids": add_time_ids}
+                prompt_embeds, pooled_prompt_embeds = encode_prompt(
+                    text_encoders=[text_encoder_one, text_encoder_two],
+                    tokenizers=None,
+                    prompt=None,
+                    text_input_ids_list=[batch["input_ids_one"], batch["input_ids_two"]],
+                )
+                unet_added_conditions.update({"text_embeds": pooled_prompt_embeds})
+                model_pred = unet(
+                    noisy_model_input,
+                    timesteps,
+                    prompt_embeds,
+                    added_cond_kwargs=unet_added_conditions,
+                    return_dict=False,
+                )[0]
+
+                # Get the target for loss depending on the prediction type
+                if args.prediction_type is not None:
+                    # set prediction_type of scheduler if defined
+                    noise_scheduler.register_to_config(prediction_type=args.prediction_type)
+
+                if noise_scheduler.config.prediction_type == "epsilon":
+                    target = noise
+                elif noise_scheduler.config.prediction_type == "v_prediction":
+                    target = noise_scheduler.get_velocity(model_input, noise, timesteps)
+                else:
+                    raise ValueError(f"Unknown prediction type {noise_scheduler.config.prediction_type}")
+
+                if args.snr_gamma is None:
+                    loss = F.mse_loss(model_pred.float(), target.float(), reduction="mean")
+                else:
+                    # Compute loss-weights as per Section 3.4 of https://arxiv.org/abs/2303.09556.
+                    # Since we predict the noise instead of x_0, the original formulation is slightly changed.
+                    # This is discussed in Section 4.2 of the same paper.
+                    snr = compute_snr(noise_scheduler, timesteps)
+                    mse_loss_weights = torch.stack([snr, args.snr_gamma * torch.ones_like(timesteps)], dim=1).min(
+                        dim=1
+                    )[0]
+                    if noise_scheduler.config.prediction_type == "epsilon":
+                        mse_loss_weights = mse_loss_weights / snr
+                    elif noise_scheduler.config.prediction_type == "v_prediction":
+                        mse_loss_weights = mse_loss_weights / (snr + 1)
+
+                    loss = F.mse_loss(model_pred.float(), target.float(), reduction="none")
+                    loss = loss.mean(dim=list(range(1, len(loss.shape)))) * mse_loss_weights
+                    loss = loss.mean()
+                if args.debug_loss and "filenames" in batch:
+                    for fname in batch["filenames"]:
+                        accelerator.log({"loss_for_" + fname: loss}, step=global_step)
+                # Gather the losses across all processes for logging (if we use distributed training).
+                avg_loss = accelerator.gather(loss.repeat(args.train_batch_size)).mean()
+                train_loss += avg_loss.item() / args.gradient_accumulation_steps
+
+                # Backpropagate
+                accelerator.backward(loss)
+                if accelerator.sync_gradients:
+                    accelerator.clip_grad_norm_(params_to_optimize, args.max_grad_norm)
+                optimizer.step()
+                lr_scheduler.step()
+                optimizer.zero_grad()
+
+            # Checks if the accelerator has performed an optimization step behind the scenes
+            if accelerator.sync_gradients:
+                progress_bar.update(1)
+                global_step += 1
+                accelerator.log({"train_loss": train_loss}, step=global_step)
+                train_loss = 0.0
+
+                # DeepSpeed requires saving weights on every device; saving weights only on the main process would cause issues.
+                if accelerator.distributed_type == DistributedType.DEEPSPEED or 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 = {"step_loss": loss.detach().item(), "lr": lr_scheduler.get_last_lr()[0]}
+            progress_bar.set_postfix(**logs)
+
+            if global_step >= args.max_train_steps:
+                break
+
+        if accelerator.is_main_process:
+            if args.validation_prompt is not None and epoch % args.validation_epochs == 0:
+                # create pipeline
+                pipeline = StableDiffusionXLPipeline.from_pretrained(
+                    args.pretrained_model_name_or_path,
+                    vae=vae,
+                    text_encoder=unwrap_model(text_encoder_one),
+                    text_encoder_2=unwrap_model(text_encoder_two),
+                    unet=unwrap_model(unet),
+                    revision=args.revision,
+                    variant=args.variant,
+                    torch_dtype=weight_dtype,
+                )
+
+                images = log_validation(pipeline, args, accelerator, epoch)
+
+                del pipeline
+                torch.cuda.empty_cache()
+
+    # Save the lora layers
+    accelerator.wait_for_everyone()
+    if accelerator.is_main_process:
+        unet = unwrap_model(unet)
+        unet_lora_state_dict = convert_state_dict_to_diffusers(get_peft_model_state_dict(unet))
+
+        if args.train_text_encoder:
+            text_encoder_one = unwrap_model(text_encoder_one)
+            text_encoder_two = unwrap_model(text_encoder_two)
+
+            text_encoder_lora_layers = convert_state_dict_to_diffusers(get_peft_model_state_dict(text_encoder_one))
+            text_encoder_2_lora_layers = convert_state_dict_to_diffusers(get_peft_model_state_dict(text_encoder_two))
+        else:
+            text_encoder_lora_layers = None
+            text_encoder_2_lora_layers = None
+
+        StableDiffusionXLPipeline.save_lora_weights(
+            save_directory=args.output_dir,
+            unet_lora_layers=unet_lora_state_dict,
+            text_encoder_lora_layers=text_encoder_lora_layers,
+            text_encoder_2_lora_layers=text_encoder_2_lora_layers,
+        )
+
+        del unet
+        del text_encoder_one
+        del text_encoder_two
+        del text_encoder_lora_layers
+        del text_encoder_2_lora_layers
+        torch.cuda.empty_cache()
+
+        # Final inference
+        # Make sure vae.dtype is consistent with the unet.dtype
+        if args.mixed_precision == "fp16":
+            vae.to(weight_dtype)
+        # Load previous pipeline
+        pipeline = StableDiffusionXLPipeline.from_pretrained(
+            args.pretrained_model_name_or_path,
+            vae=vae,
+            revision=args.revision,
+            variant=args.variant,
+            torch_dtype=weight_dtype,
+        )
+
+        # load attention processors
+        pipeline.load_lora_weights(args.output_dir)
+
+        # run inference
+        if args.validation_prompt and args.num_validation_images > 0:
+            images = log_validation(pipeline, args, accelerator, epoch, is_final_validation=True)
+
+        if args.push_to_hub:
+            save_model_card(
+                repo_id,
+                images=images,
+                base_model=args.pretrained_model_name_or_path,
+                dataset_name=args.dataset_name,
+                train_text_encoder=args.train_text_encoder,
+                repo_folder=args.output_dir,
+                vae_path=args.pretrained_vae_model_name_or_path,
+            )
+            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__":
+    args = parse_args()
+    main(args)

train_text_to_image_sdxl.py

@@ -0,0 +1,1358 @@
+#!/usr/bin/env python
+# coding=utf-8
+# Copyright 2025 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
+# limitations under the License.
+"""Fine-tuning script for Stable Diffusion XL for text2image."""
+
+import argparse
+import functools
+import gc
+import logging
+import math
+import os
+import random
+import shutil
+from contextlib import nullcontext
+from pathlib import Path
+
+import accelerate
+import datasets
+import numpy as np
+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 DistributedType, ProjectConfiguration, set_seed
+from datasets import concatenate_datasets, load_dataset
+from huggingface_hub import create_repo, upload_folder
+from packaging import version
+from torchvision import transforms
+from torchvision.transforms.functional import crop
+from tqdm.auto import tqdm
+from transformers import AutoTokenizer, PretrainedConfig
+
+import diffusers
+from diffusers import AutoencoderKL, DDPMScheduler, StableDiffusionXLPipeline, UNet2DConditionModel
+from diffusers.optimization import get_scheduler
+from diffusers.training_utils import EMAModel, compute_snr
+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_torch_npu_available, is_xformers_available
+from diffusers.utils.torch_utils import is_compiled_module
+
+
+# Will error if the minimal version of diffusers is not installed. Remove at your own risks.
+check_min_version("0.33.0.dev0")
+
+logger = get_logger(__name__)
+if is_torch_npu_available():
+    import torch_npu
+
+    torch.npu.config.allow_internal_format = False
+
+DATASET_NAME_MAPPING = {
+    "lambdalabs/naruto-blip-captions": ("image", "text"),
+}
+
+
+def save_model_card(
+    repo_id: str,
+    images: list = None,
+    validation_prompt: str = None,
+    base_model: str = None,
+    dataset_name: str = None,
+    repo_folder: str = None,
+    vae_path: 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"""
+# Text-to-image finetuning - {repo_id}
+
+This pipeline was finetuned from **{base_model}** on the **{dataset_name}** dataset. Below are some example images generated with the finetuned pipeline using the following prompt: {validation_prompt}: \n
+{img_str}
+
+Special VAE used for training: {vae_path}.
+"""
+
+    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-training",
+        "diffusers",
+    ]
+    model_card = populate_model_card(model_card, tags=tags)
+
+    model_card.save(os.path.join(repo_folder, "README.md"))
+
+
+def import_model_class_from_model_name_or_path(
+    pretrained_model_name_or_path: str, revision: str, subfolder: str = "text_encoder"
+):
+    text_encoder_config = PretrainedConfig.from_pretrained(
+        pretrained_model_name_or_path, subfolder=subfolder, revision=revision
+    )
+    model_class = text_encoder_config.architectures[0]
+
+    if model_class == "CLIPTextModel":
+        from transformers import CLIPTextModel
+
+        return CLIPTextModel
+    elif model_class == "CLIPTextModelWithProjection":
+        from transformers import CLIPTextModelWithProjection
+
+        return CLIPTextModelWithProjection
+    else:
+        raise ValueError(f"{model_class} is not supported.")
+
+
+def parse_args(input_args=None):
+    parser = argparse.ArgumentParser(description="Simple example of a training script.")
+    parser.add_argument(
+        "--pretrained_model_name_or_path",
+        type=str,
+        default=None,
+        required=True,
+        help="Path to pretrained model or model identifier from huggingface.co/models.",
+    )
+    parser.add_argument(
+        "--pretrained_vae_model_name_or_path",
+        type=str,
+        default=None,
+        help="Path to pretrained VAE model with better numerical stability. More details: https://github.com/huggingface/diffusers/pull/4038.",
+    )
+    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(
+        "--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 🤗 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(
+        "--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(
+        "--image_column", type=str, default="image", help="The column of the dataset containing an image."
+    )
+    parser.add_argument(
+        "--caption_column",
+        type=str,
+        default="text",
+        help="The column of the dataset containing a caption or a list of captions.",
+    )
+    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_epochs",
+        type=int,
+        default=1,
+        help=(
+            "Run fine-tuning validation every X epochs. The validation process consists of running the prompt"
+            " `args.validation_prompt` multiple times: `args.num_validation_images`."
+        ),
+    )
+    parser.add_argument(
+        "--max_train_samples",
+        type=int,
+        default=None,
+        help=(
+            "For debugging purposes or quicker training, truncate the number of training examples to this "
+            "value if set."
+        ),
+    )
+    parser.add_argument(
+        "--proportion_empty_prompts",
+        type=float,
+        default=0,
+        help="Proportion of image prompts to be replaced with empty strings. Defaults to 0 (no prompt replacement).",
+    )
+    parser.add_argument(
+        "--output_dir",
+        type=str,
+        default="sdxl-model-finetuned",
+        help="The output directory where the model predictions and checkpoints will be written.",
+    )
+    parser.add_argument(
+        "--cache_dir",
+        type=str,
+        default=None,
+        help="The directory where the downloaded models and datasets will be stored.",
+    )
+    parser.add_argument("--seed", type=int, default=None, help="A seed for reproducible training.")
+    parser.add_argument(
+        "--resolution",
+        type=int,
+        default=1024,
+        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",
+        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("--num_train_epochs", type=int, default=100)
+    parser.add_argument(
+        "--max_train_steps",
+        type=int,
+        default=None,
+        help="Total number of training steps to perform.  If provided, overrides num_train_epochs.",
+    )
+    parser.add_argument(
+        "--checkpointing_steps",
+        type=int,
+        default=500,
+        help=(
+            "Save a checkpoint of the training state every X updates. These checkpoints can be used both as final"
+            " checkpoints in case they are better than the last checkpoint, and are also suitable for resuming"
+            " training using `--resume_from_checkpoint`."
+        ),
+    )
+    parser.add_argument(
+        "--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(
+        "--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(
+        "--timestep_bias_strategy",
+        type=str,
+        default="none",
+        choices=["earlier", "later", "range", "none"],
+        help=(
+            "The timestep bias strategy, which may help direct the model toward learning low or high frequency details."
+            " Choices: ['earlier', 'later', 'range', 'none']."
+            " The default is 'none', which means no bias is applied, and training proceeds normally."
+            " The value of 'later' will increase the frequency of the model's final training timesteps."
+        ),
+    )
+    parser.add_argument(
+        "--timestep_bias_multiplier",
+        type=float,
+        default=1.0,
+        help=(
+            "The multiplier for the bias. Defaults to 1.0, which means no bias is applied."
+            " A value of 2.0 will double the weight of the bias, and a value of 0.5 will halve it."
+        ),
+    )
+    parser.add_argument(
+        "--timestep_bias_begin",
+        type=int,
+        default=0,
+        help=(
+            "When using `--timestep_bias_strategy=range`, the beginning (inclusive) timestep to bias."
+            " Defaults to zero, which equates to having no specific bias."
+        ),
+    )
+    parser.add_argument(
+        "--timestep_bias_end",
+        type=int,
+        default=1000,
+        help=(
+            "When using `--timestep_bias_strategy=range`, the final timestep (inclusive) to bias."
+            " Defaults to 1000, which is the number of timesteps that Stable Diffusion is trained on."
+        ),
+    )
+    parser.add_argument(
+        "--timestep_bias_portion",
+        type=float,
+        default=0.25,
+        help=(
+            "The portion of timesteps to bias. Defaults to 0.25, which 25% of timesteps will be biased."
+            " A value of 0.5 will bias one half of the timesteps. The value provided for `--timestep_bias_strategy` determines"
+            " whether the biased portions are in the earlier or later timesteps."
+        ),
+    )
+    parser.add_argument(
+        "--snr_gamma",
+        type=float,
+        default=None,
+        help="SNR weighting gamma to be used if rebalancing the loss. Recommended value is 5.0. "
+        "More details here: https://arxiv.org/abs/2303.09556.",
+    )
+    parser.add_argument("--use_ema", action="store_true", help="Whether to use EMA model.")
+    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(
+        "--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("--max_grad_norm", default=1.0, type=float, help="Max gradient norm.")
+    parser.add_argument("--push_to_hub", action="store_true", help="Whether or not to push the model to the Hub.")
+    parser.add_argument("--hub_token", type=str, default=None, help="The token to use to push to the Model Hub.")
+    parser.add_argument(
+        "--prediction_type",
+        type=str,
+        default=None,
+        help="The prediction_type that shall be used for training. Choose between 'epsilon' or 'v_prediction' or leave `None`. If left to `None` the default prediction type of the scheduler: `noise_scheduler.config.prediction_type` is chosen.",
+    )
+    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(
+        "--report_to",
+        type=str,
+        default="tensorboard",
+        help=(
+            'The integration to report the results and logs to. Supported platforms are `"tensorboard"`'
+            ' (default), `"wandb"` and `"comet_ml"`. Use `"all"` to report to all integrations.'
+        ),
+    )
+    parser.add_argument(
+        "--mixed_precision",
+        type=str,
+        default=None,
+        choices=["no", "fp16", "bf16"],
+        help=(
+            "Whether to use mixed precision. Choose between fp16 and bf16 (bfloat16). Bf16 requires PyTorch >="
+            " 1.10.and an Nvidia Ampere GPU.  Default to the value of accelerate config of the current system or the"
+            " flag passed with the `accelerate.launch` command. Use this argument to override the accelerate config."
+        ),
+    )
+    parser.add_argument("--local_rank", type=int, default=-1, help="For distributed training: local_rank")
+    parser.add_argument(
+        "--enable_npu_flash_attention", action="store_true", help="Whether or not to use npu flash attention."
+    )
+    parser.add_argument(
+        "--enable_xformers_memory_efficient_attention", action="store_true", help="Whether or not to use xformers."
+    )
+    parser.add_argument("--noise_offset", type=float, default=0, help="The scale of noise offset.")
+
+    if input_args is not None:
+        args = parser.parse_args(input_args)
+    else:
+        args = parser.parse_args()
+
+    env_local_rank = int(os.environ.get("LOCAL_RANK", -1))
+    if env_local_rank != -1 and env_local_rank != args.local_rank:
+        args.local_rank = env_local_rank
+
+    # Sanity checks
+    if args.dataset_name is None and args.train_data_dir is None:
+        raise ValueError("Need either a dataset name or a training folder.")
+    if args.proportion_empty_prompts < 0 or args.proportion_empty_prompts > 1:
+        raise ValueError("`--proportion_empty_prompts` must be in the range [0, 1].")
+
+    return args
+
+
+# Adapted from pipelines.StableDiffusionXLPipeline.encode_prompt
+def encode_prompt(batch, text_encoders, tokenizers, proportion_empty_prompts, caption_column, is_train=True):
+    prompt_embeds_list = []
+    prompt_batch = batch[caption_column]
+
+    captions = []
+    for caption in prompt_batch:
+        if random.random() < proportion_empty_prompts:
+            captions.append("")
+        elif isinstance(caption, str):
+            captions.append(caption)
+        elif isinstance(caption, (list, np.ndarray)):
+            # take a random caption if there are multiple
+            captions.append(random.choice(caption) if is_train else caption[0])
+
+    with torch.no_grad():
+        for tokenizer, text_encoder in zip(tokenizers, text_encoders):
+            text_inputs = tokenizer(
+                captions,
+                padding="max_length",
+                max_length=tokenizer.model_max_length,
+                truncation=True,
+                return_tensors="pt",
+            )
+            text_input_ids = text_inputs.input_ids
+            prompt_embeds = text_encoder(
+                text_input_ids.to(text_encoder.device),
+                output_hidden_states=True,
+                return_dict=False,
+            )
+
+            # We are only ALWAYS interested in the pooled output of the final text encoder
+            pooled_prompt_embeds = prompt_embeds[0]
+            prompt_embeds = prompt_embeds[-1][-2]
+            bs_embed, seq_len, _ = prompt_embeds.shape
+            prompt_embeds = prompt_embeds.view(bs_embed, seq_len, -1)
+            prompt_embeds_list.append(prompt_embeds)
+
+    prompt_embeds = torch.concat(prompt_embeds_list, dim=-1)
+    pooled_prompt_embeds = pooled_prompt_embeds.view(bs_embed, -1)
+    return {"prompt_embeds": prompt_embeds.cpu(), "pooled_prompt_embeds": pooled_prompt_embeds.cpu()}
+
+
+def compute_vae_encodings(batch, vae):
+    images = batch.pop("pixel_values")
+    pixel_values = torch.stack(list(images))
+    pixel_values = pixel_values.to(memory_format=torch.contiguous_format).float()
+    pixel_values = pixel_values.to(vae.device, dtype=vae.dtype)
+
+    with torch.no_grad():
+        model_input = vae.encode(pixel_values).latent_dist.sample()
+    model_input = model_input * vae.config.scaling_factor
+
+    # There might have slightly performance improvement
+    # by changing model_input.cpu() to accelerator.gather(model_input)
+    return {"model_input": model_input.cpu()}
+
+
+def generate_timestep_weights(args, num_timesteps):
+    weights = torch.ones(num_timesteps)
+
+    # Determine the indices to bias
+    num_to_bias = int(args.timestep_bias_portion * num_timesteps)
+
+    if args.timestep_bias_strategy == "later":
+        bias_indices = slice(-num_to_bias, None)
+    elif args.timestep_bias_strategy == "earlier":
+        bias_indices = slice(0, num_to_bias)
+    elif args.timestep_bias_strategy == "range":
+        # Out of the possible 1000 timesteps, we might want to focus on eg. 200-500.
+        range_begin = args.timestep_bias_begin
+        range_end = args.timestep_bias_end
+        if range_begin < 0:
+            raise ValueError(
+                "When using the range strategy for timestep bias, you must provide a beginning timestep greater or equal to zero."
+            )
+        if range_end > num_timesteps:
+            raise ValueError(
+                "When using the range strategy for timestep bias, you must provide an ending timestep smaller than the number of timesteps."
+            )
+        bias_indices = slice(range_begin, range_end)
+    else:  # 'none' or any other string
+        return weights
+    if args.timestep_bias_multiplier <= 0:
+        return ValueError(
+            "The parameter --timestep_bias_multiplier is not intended to be used to disable the training of specific timesteps."
+            " If it was intended to disable timestep bias, use `--timestep_bias_strategy none` instead."
+            " A timestep bias multiplier less than or equal to 0 is not allowed."
+        )
+
+    # Apply the bias
+    weights[bias_indices] *= args.timestep_bias_multiplier
+
+    # Normalize
+    weights /= weights.sum()
+
+    return weights
+
+
+def main(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 = Path(args.output_dir, args.logging_dir)
+
+    accelerator_project_config = ProjectConfiguration(project_dir=args.output_dir, logging_dir=logging_dir)
+
+    if torch.backends.mps.is_available() and args.mixed_precision == "bf16":
+        # due to pytorch#99272, MPS does not yet support bfloat16.
+        raise ValueError(
+            "Mixed precision training with bfloat16 is not supported on MPS. Please use fp16 (recommended) or fp32 instead."
+        )
+
+    accelerator = Accelerator(
+        gradient_accumulation_steps=args.gradient_accumulation_steps,
+        mixed_precision=args.mixed_precision,
+        log_with=args.report_to,
+        project_config=accelerator_project_config,
+    )
+
+    # Disable AMP for MPS.
+    if torch.backends.mps.is_available():
+        accelerator.native_amp = False
+
+    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.")
+        import wandb
+
+    # Make one log on every process with the configuration for debugging.
+    logging.basicConfig(
+        format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
+        datefmt="%m/%d/%Y %H:%M:%S",
+        level=logging.INFO,
+    )
+    logger.info(accelerator.state, main_process_only=False)
+    if accelerator.is_local_main_process:
+        datasets.utils.logging.set_verbosity_warning()
+        transformers.utils.logging.set_verbosity_warning()
+        diffusers.utils.logging.set_verbosity_info()
+    else:
+        datasets.utils.logging.set_verbosity_error()
+        transformers.utils.logging.set_verbosity_error()
+        diffusers.utils.logging.set_verbosity_error()
+
+    # If passed along, set the training seed now.
+    if args.seed is not None:
+        set_seed(args.seed)
+
+    # 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 the tokenizers
+    tokenizer_one = AutoTokenizer.from_pretrained(
+        args.pretrained_model_name_or_path,
+        subfolder="tokenizer",
+        revision=args.revision,
+        use_fast=False,
+    )
+    tokenizer_two = AutoTokenizer.from_pretrained(
+        args.pretrained_model_name_or_path,
+        subfolder="tokenizer_2",
+        revision=args.revision,
+        use_fast=False,
+    )
+
+    # import correct text encoder classes
+    text_encoder_cls_one = import_model_class_from_model_name_or_path(
+        args.pretrained_model_name_or_path, args.revision
+    )
+    text_encoder_cls_two = import_model_class_from_model_name_or_path(
+        args.pretrained_model_name_or_path, args.revision, subfolder="text_encoder_2"
+    )
+
+    # Load scheduler and models
+    noise_scheduler = DDPMScheduler.from_pretrained(args.pretrained_model_name_or_path, subfolder="scheduler")
+    # Check for terminal SNR in combination with SNR Gamma
+    text_encoder_one = text_encoder_cls_one.from_pretrained(
+        args.pretrained_model_name_or_path, subfolder="text_encoder", revision=args.revision, variant=args.variant
+    )
+    text_encoder_two = text_encoder_cls_two.from_pretrained(
+        args.pretrained_model_name_or_path, subfolder="text_encoder_2", revision=args.revision, variant=args.variant
+    )
+    vae_path = (
+        args.pretrained_model_name_or_path
+        if args.pretrained_vae_model_name_or_path is None
+        else args.pretrained_vae_model_name_or_path
+    )
+    vae = AutoencoderKL.from_pretrained(
+        vae_path,
+        subfolder="vae" if args.pretrained_vae_model_name_or_path is None else None,
+        revision=args.revision,
+        variant=args.variant,
+    )
+    unet = UNet2DConditionModel.from_pretrained(
+        args.pretrained_model_name_or_path, subfolder="unet", revision=args.revision, variant=args.variant
+    )
+
+    # Freeze vae and text encoders.
+    vae.requires_grad_(False)
+    text_encoder_one.requires_grad_(False)
+    text_encoder_two.requires_grad_(False)
+    # Set unet as trainable.
+    unet.train()
+
+    # For mixed precision training we cast all non-trainable weights 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 unet, vae and text_encoder to device and cast to weight_dtype
+    # The VAE is in float32 to avoid NaN losses.
+    vae.to(accelerator.device, dtype=torch.float32)
+    text_encoder_one.to(accelerator.device, dtype=weight_dtype)
+    text_encoder_two.to(accelerator.device, dtype=weight_dtype)
+
+    # Create EMA for the unet.
+    if args.use_ema:
+        ema_unet = UNet2DConditionModel.from_pretrained(
+            args.pretrained_model_name_or_path, subfolder="unet", revision=args.revision, variant=args.variant
+        )
+        ema_unet = EMAModel(ema_unet.parameters(), model_cls=UNet2DConditionModel, model_config=ema_unet.config)
+    if args.enable_npu_flash_attention:
+        if is_torch_npu_available():
+            logger.info("npu flash attention enabled.")
+            unet.enable_npu_flash_attention()
+        else:
+            raise ValueError("npu flash attention requires torch_npu extensions and is supported only on npu devices.")
+    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")
+
+    # `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_unet.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
+                    if weights:
+                        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"), UNet2DConditionModel)
+                ema_unet.load_state_dict(load_model.state_dict())
+                ema_unet.to(accelerator.device)
+                del load_model
+
+            for _ in range(len(models)):
+                # pop models so that they are not loaded again
+                model = models.pop()
+
+                # load diffusers style into model
+                load_model = UNet2DConditionModel.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)
+
+    if args.gradient_checkpointing:
+        unet.enable_gradient_checkpointing()
+
+    # Enable TF32 for faster training on Ampere GPUs,
+    # cf https://pytorch.org/docs/stable/notes/cuda.html#tensorfloat-32-tf32-on-ampere-devices
+    if args.allow_tf32:
+        torch.backends.cuda.matmul.allow_tf32 = True
+
+    if args.scale_lr:
+        args.learning_rate = (
+            args.learning_rate * args.gradient_accumulation_steps * args.train_batch_size * accelerator.num_processes
+        )
+
+    # Use 8-bit Adam for lower memory usage or to fine-tune the model in 16GB GPUs
+    if args.use_8bit_adam:
+        try:
+            import bitsandbytes as bnb
+        except ImportError:
+            raise ImportError(
+                "To use 8-bit Adam, please install the bitsandbytes library: `pip install bitsandbytes`."
+            )
+
+        optimizer_class = bnb.optim.AdamW8bit
+    else:
+        optimizer_class = torch.optim.AdamW
+
+    # Optimizer creation
+    params_to_optimize = unet.parameters()
+    optimizer = optimizer_class(
+        params_to_optimize,
+        lr=args.learning_rate,
+        betas=(args.adam_beta1, args.adam_beta2),
+        weight_decay=args.adam_weight_decay,
+        eps=args.adam_epsilon,
+    )
+
+    # 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:
+        # Downloading and loading a dataset from the hub.
+        dataset = load_dataset(
+            args.dataset_name, args.dataset_config_name, cache_dir=args.cache_dir, data_dir=args.train_data_dir
+        )
+    else:
+        data_files = {}
+        if args.train_data_dir is not None:
+            data_files["train"] = os.path.join(args.train_data_dir, "**")
+        dataset = load_dataset(
+            "imagefolder",
+            data_files=data_files,
+            cache_dir=args.cache_dir,
+        )
+        # See more about loading custom images at
+        # https://huggingface.co/docs/datasets/v2.4.0/en/image_load#imagefolder
+
+    # Preprocessing the datasets.
+    # We need to tokenize inputs and targets.
+    column_names = dataset["train"].column_names
+
+    # 6. Get the column names for input/target.
+    dataset_columns = DATASET_NAME_MAPPING.get(args.dataset_name, None)
+    if args.image_column is None:
+        image_column = dataset_columns[0] if dataset_columns is not None else column_names[0]
+    else:
+        image_column = args.image_column
+        if image_column not in column_names:
+            raise ValueError(
+                f"--image_column' value '{args.image_column}' needs to be one of: {', '.join(column_names)}"
+            )
+    if args.caption_column is None:
+        caption_column = dataset_columns[1] if dataset_columns is not None else column_names[1]
+    else:
+        caption_column = args.caption_column
+        if caption_column not in column_names:
+            raise ValueError(
+                f"--caption_column' value '{args.caption_column}' needs to be one of: {', '.join(column_names)}"
+            )
+
+    # Preprocessing the datasets.
+    train_resize = transforms.Resize(args.resolution, interpolation=transforms.InterpolationMode.BILINEAR)
+    train_crop = transforms.CenterCrop(args.resolution) if args.center_crop else transforms.RandomCrop(args.resolution)
+    train_flip = transforms.RandomHorizontalFlip(p=1.0)
+    train_transforms = transforms.Compose([transforms.ToTensor(), transforms.Normalize([0.5], [0.5])])
+
+    def preprocess_train(examples):
+        images = [image.convert("RGB") for image in examples[image_column]]
+        # image aug
+        original_sizes = []
+        all_images = []
+        crop_top_lefts = []
+        for image in images:
+            original_sizes.append((image.height, image.width))
+            image = train_resize(image)
+            if args.random_flip and random.random() < 0.5:
+                # flip
+                image = train_flip(image)
+            if args.center_crop:
+                y1 = max(0, int(round((image.height - args.resolution) / 2.0)))
+                x1 = max(0, int(round((image.width - args.resolution) / 2.0)))
+                image = train_crop(image)
+            else:
+                y1, x1, h, w = train_crop.get_params(image, (args.resolution, args.resolution))
+                image = crop(image, y1, x1, h, w)
+            crop_top_left = (y1, x1)
+            crop_top_lefts.append(crop_top_left)
+            image = train_transforms(image)
+            all_images.append(image)
+
+        examples["original_sizes"] = original_sizes
+        examples["crop_top_lefts"] = crop_top_lefts
+        examples["pixel_values"] = all_images
+        return examples
+
+    with accelerator.main_process_first():
+        if args.max_train_samples is not None:
+            dataset["train"] = dataset["train"].shuffle(seed=args.seed).select(range(args.max_train_samples))
+        # Set the training transforms
+        train_dataset = dataset["train"].with_transform(preprocess_train)
+
+    # Let's first compute all the embeddings so that we can free up the text encoders
+    # from memory. We will pre-compute the VAE encodings too.
+    text_encoders = [text_encoder_one, text_encoder_two]
+    tokenizers = [tokenizer_one, tokenizer_two]
+    compute_embeddings_fn = functools.partial(
+        encode_prompt,
+        text_encoders=text_encoders,
+        tokenizers=tokenizers,
+        proportion_empty_prompts=args.proportion_empty_prompts,
+        caption_column=args.caption_column,
+    )
+    compute_vae_encodings_fn = functools.partial(compute_vae_encodings, vae=vae)
+    with accelerator.main_process_first():
+        from datasets.fingerprint import Hasher
+
+        # fingerprint used by the cache for the other processes to load the result
+        # details: https://github.com/huggingface/diffusers/pull/4038#discussion_r1266078401
+        new_fingerprint = Hasher.hash(args)
+        new_fingerprint_for_vae = Hasher.hash((vae_path, args))
+        train_dataset_with_embeddings = train_dataset.map(
+            compute_embeddings_fn, batched=True, new_fingerprint=new_fingerprint
+        )
+        train_dataset_with_vae = train_dataset.map(
+            compute_vae_encodings_fn,
+            batched=True,
+            batch_size=args.train_batch_size,
+            new_fingerprint=new_fingerprint_for_vae,
+        )
+        precomputed_dataset = concatenate_datasets(
+            [train_dataset_with_embeddings, train_dataset_with_vae.remove_columns(["image", "text"])], axis=1
+        )
+        precomputed_dataset = precomputed_dataset.with_transform(preprocess_train)
+
+    del compute_vae_encodings_fn, compute_embeddings_fn, text_encoder_one, text_encoder_two
+    del text_encoders, tokenizers, vae
+    gc.collect()
+    if is_torch_npu_available():
+        torch_npu.npu.empty_cache()
+    elif torch.cuda.is_available():
+        torch.cuda.empty_cache()
+
+    def collate_fn(examples):
+        model_input = torch.stack([torch.tensor(example["model_input"]) for example in examples])
+        original_sizes = [example["original_sizes"] for example in examples]
+        crop_top_lefts = [example["crop_top_lefts"] for example in examples]
+        prompt_embeds = torch.stack([torch.tensor(example["prompt_embeds"]) for example in examples])
+        pooled_prompt_embeds = torch.stack([torch.tensor(example["pooled_prompt_embeds"]) for example in examples])
+
+        return {
+            "model_input": model_input,
+            "prompt_embeds": prompt_embeds,
+            "pooled_prompt_embeds": pooled_prompt_embeds,
+            "original_sizes": original_sizes,
+            "crop_top_lefts": crop_top_lefts,
+        }
+
+    # DataLoaders creation:
+    train_dataloader = torch.utils.data.DataLoader(
+        precomputed_dataset,
+        shuffle=True,
+        collate_fn=collate_fn,
+        batch_size=args.train_batch_size,
+        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 * args.gradient_accumulation_steps,
+        num_training_steps=args.max_train_steps * args.gradient_accumulation_steps,
+    )
+
+    # Prepare everything with our `accelerator`.
+    unet, optimizer, train_dataloader, lr_scheduler = accelerator.prepare(
+        unet, optimizer, train_dataloader, lr_scheduler
+    )
+
+    if args.use_ema:
+        ema_unet.to(accelerator.device)
+
+    # 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("text2image-fine-tune-sdxl", config=vars(args))
+
+    # Function for unwrapping if torch.compile() was used in accelerate.
+    def unwrap_model(model):
+        model = accelerator.unwrap_model(model)
+        model = model._orig_mod if is_compiled_module(model) else model
+        return model
+
+    if torch.backends.mps.is_available() or "playground" in args.pretrained_model_name_or_path:
+        autocast_ctx = nullcontext()
+    else:
+        autocast_ctx = torch.autocast(accelerator.device.type)
+
+    # 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(precomputed_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,
+    )
+
+    for epoch in range(first_epoch, args.num_train_epochs):
+        train_loss = 0.0
+        for step, batch in enumerate(train_dataloader):
+            with accelerator.accumulate(unet):
+                # Sample noise that we'll add to the latents
+                model_input = batch["model_input"].to(accelerator.device)
+                noise = torch.randn_like(model_input)
+                if args.noise_offset:
+                    # https://www.crosslabs.org//blog/diffusion-with-offset-noise
+                    noise += args.noise_offset * torch.randn(
+                        (model_input.shape[0], model_input.shape[1], 1, 1), device=model_input.device
+                    )
+
+                bsz = model_input.shape[0]
+                if args.timestep_bias_strategy == "none":
+                    # Sample a random timestep for each image without bias.
+                    timesteps = torch.randint(
+                        0, noise_scheduler.config.num_train_timesteps, (bsz,), device=model_input.device
+                    )
+                else:
+                    # Sample a random timestep for each image, potentially biased by the timestep weights.
+                    # Biasing the timestep weights allows us to spend less time training irrelevant timesteps.
+                    weights = generate_timestep_weights(args, noise_scheduler.config.num_train_timesteps).to(
+                        model_input.device
+                    )
+                    timesteps = torch.multinomial(weights, bsz, replacement=True).long()
+
+                # Add noise to the model input according to the noise magnitude at each timestep
+                # (this is the forward diffusion process)
+                noisy_model_input = noise_scheduler.add_noise(model_input, noise, timesteps).to(dtype=weight_dtype)
+
+                # time ids
+                def compute_time_ids(original_size, crops_coords_top_left):
+                    # Adapted from pipeline.StableDiffusionXLPipeline._get_add_time_ids
+                    target_size = (args.resolution, args.resolution)
+                    add_time_ids = list(original_size + crops_coords_top_left + target_size)
+                    add_time_ids = torch.tensor([add_time_ids], device=accelerator.device, dtype=weight_dtype)
+                    return add_time_ids
+
+                add_time_ids = torch.cat(
+                    [compute_time_ids(s, c) for s, c in zip(batch["original_sizes"], batch["crop_top_lefts"])]
+                )
+
+                # Predict the noise residual
+                unet_added_conditions = {"time_ids": add_time_ids}
+                prompt_embeds = batch["prompt_embeds"].to(accelerator.device, dtype=weight_dtype)
+                pooled_prompt_embeds = batch["pooled_prompt_embeds"].to(accelerator.device)
+                unet_added_conditions.update({"text_embeds": pooled_prompt_embeds})
+                model_pred = unet(
+                    noisy_model_input,
+                    timesteps,
+                    prompt_embeds,
+                    added_cond_kwargs=unet_added_conditions,
+                    return_dict=False,
+                )[0]
+
+                # Get the target for loss depending on the prediction type
+                if args.prediction_type is not None:
+                    # set prediction_type of scheduler if defined
+                    noise_scheduler.register_to_config(prediction_type=args.prediction_type)
+
+                if noise_scheduler.config.prediction_type == "epsilon":
+                    target = noise
+                elif noise_scheduler.config.prediction_type == "v_prediction":
+                    target = noise_scheduler.get_velocity(model_input, noise, timesteps)
+                elif noise_scheduler.config.prediction_type == "sample":
+                    # We set the target to latents here, but the model_pred will return the noise sample prediction.
+                    target = model_input
+                    # We will have to subtract the noise residual from the prediction to get the target sample.
+                    model_pred = model_pred - noise
+                else:
+                    raise ValueError(f"Unknown prediction type {noise_scheduler.config.prediction_type}")
+
+                if args.snr_gamma is None:
+                    loss = F.mse_loss(model_pred.float(), target.float(), reduction="mean")
+                else:
+                    # Compute loss-weights as per Section 3.4 of https://arxiv.org/abs/2303.09556.
+                    # Since we predict the noise instead of x_0, the original formulation is slightly changed.
+                    # This is discussed in Section 4.2 of the same paper.
+                    snr = compute_snr(noise_scheduler, timesteps)
+                    mse_loss_weights = torch.stack([snr, args.snr_gamma * torch.ones_like(timesteps)], dim=1).min(
+                        dim=1
+                    )[0]
+                    if noise_scheduler.config.prediction_type == "epsilon":
+                        mse_loss_weights = mse_loss_weights / snr
+                    elif noise_scheduler.config.prediction_type == "v_prediction":
+                        mse_loss_weights = mse_loss_weights / (snr + 1)
+
+                    loss = F.mse_loss(model_pred.float(), target.float(), reduction="none")
+                    loss = loss.mean(dim=list(range(1, len(loss.shape)))) * mse_loss_weights
+                    loss = loss.mean()
+
+                # Gather the losses across all processes for logging (if we use distributed training).
+                avg_loss = accelerator.gather(loss.repeat(args.train_batch_size)).mean()
+                train_loss += avg_loss.item() / args.gradient_accumulation_steps
+
+                # Backpropagate
+                accelerator.backward(loss)
+                if accelerator.sync_gradients:
+                    params_to_clip = unet.parameters()
+                    accelerator.clip_grad_norm_(params_to_clip, args.max_grad_norm)
+                optimizer.step()
+                lr_scheduler.step()
+                optimizer.zero_grad()
+
+            # Checks if the accelerator has performed an optimization step behind the scenes
+            if accelerator.sync_gradients:
+                if args.use_ema:
+                    ema_unet.step(unet.parameters())
+                progress_bar.update(1)
+                global_step += 1
+                accelerator.log({"train_loss": train_loss}, step=global_step)
+                train_loss = 0.0
+
+                # DeepSpeed requires saving weights on every device; saving weights only on the main process would cause issues.
+                if accelerator.distributed_type == DistributedType.DEEPSPEED or 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 = {"step_loss": loss.detach().item(), "lr": lr_scheduler.get_last_lr()[0]}
+            progress_bar.set_postfix(**logs)
+
+            if global_step >= args.max_train_steps:
+                break
+
+        if accelerator.is_main_process:
+            if args.validation_prompt is not None and epoch % args.validation_epochs == 0:
+                logger.info(
+                    f"Running validation... \n Generating {args.num_validation_images} images with prompt:"
+                    f" {args.validation_prompt}."
+                )
+                if args.use_ema:
+                    # Store the UNet parameters temporarily and load the EMA parameters to perform inference.
+                    ema_unet.store(unet.parameters())
+                    ema_unet.copy_to(unet.parameters())
+
+                # create pipeline
+                vae = AutoencoderKL.from_pretrained(
+                    vae_path,
+                    subfolder="vae" if args.pretrained_vae_model_name_or_path is None else None,
+                    revision=args.revision,
+                    variant=args.variant,
+                )
+                pipeline = StableDiffusionXLPipeline.from_pretrained(
+                    args.pretrained_model_name_or_path,
+                    vae=vae,
+                    unet=accelerator.unwrap_model(unet),
+                    revision=args.revision,
+                    variant=args.variant,
+                    torch_dtype=weight_dtype,
+                )
+                if args.prediction_type is not None:
+                    scheduler_args = {"prediction_type": args.prediction_type}
+                    pipeline.scheduler = pipeline.scheduler.from_config(pipeline.scheduler.config, **scheduler_args)
+
+                pipeline = pipeline.to(accelerator.device)
+                pipeline.set_progress_bar_config(disable=True)
+
+                # run inference
+                generator = (
+                    torch.Generator(device=accelerator.device).manual_seed(args.seed)
+                    if args.seed is not None
+                    else None
+                )
+                pipeline_args = {"prompt": args.validation_prompt}
+
+                with autocast_ctx:
+                    images = [
+                        pipeline(**pipeline_args, generator=generator, num_inference_steps=25).images[0]
+                        for _ in range(args.num_validation_images)
+                    ]
+
+                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
+                if is_torch_npu_available():
+                    torch_npu.npu.empty_cache()
+                elif torch.cuda.is_available():
+                    torch.cuda.empty_cache()
+
+                if args.use_ema:
+                    # Switch back to the original UNet parameters.
+                    ema_unet.restore(unet.parameters())
+
+    accelerator.wait_for_everyone()
+    if accelerator.is_main_process:
+        unet = unwrap_model(unet)
+        if args.use_ema:
+            ema_unet.copy_to(unet.parameters())
+
+        # Serialize pipeline.
+        vae = AutoencoderKL.from_pretrained(
+            vae_path,
+            subfolder="vae" if args.pretrained_vae_model_name_or_path is None else None,
+            revision=args.revision,
+            variant=args.variant,
+            torch_dtype=weight_dtype,
+        )
+        pipeline = StableDiffusionXLPipeline.from_pretrained(
+            args.pretrained_model_name_or_path,
+            unet=unet,
+            vae=vae,
+            revision=args.revision,
+            variant=args.variant,
+            torch_dtype=weight_dtype,
+        )
+        if args.prediction_type is not None:
+            scheduler_args = {"prediction_type": args.prediction_type}
+            pipeline.scheduler = pipeline.scheduler.from_config(pipeline.scheduler.config, **scheduler_args)
+        pipeline.save_pretrained(args.output_dir)
+
+        # run inference
+        images = []
+        if args.validation_prompt and args.num_validation_images > 0:
+            pipeline = pipeline.to(accelerator.device)
+            generator = (
+                torch.Generator(device=accelerator.device).manual_seed(args.seed) if args.seed is not None else None
+            )
+
+            with autocast_ctx:
+                images = [
+                    pipeline(args.validation_prompt, num_inference_steps=25, generator=generator).images[0]
+                    for _ in range(args.num_validation_images)
+                ]
+
+            for tracker in accelerator.trackers:
+                if tracker.name == "tensorboard":
+                    np_images = np.stack([np.asarray(img) for img in images])
+                    tracker.writer.add_images("test", np_images, epoch, dataformats="NHWC")
+                if tracker.name == "wandb":
+                    tracker.log(
+                        {
+                            "test": [
+                                wandb.Image(image, caption=f"{i}: {args.validation_prompt}")
+                                for i, image in enumerate(images)
+                            ]
+                        }
+                    )
+
+        if args.push_to_hub:
+            save_model_card(
+                repo_id=repo_id,
+                images=images,
+                validation_prompt=args.validation_prompt,
+                base_model=args.pretrained_model_name_or_path,
+                dataset_name=args.dataset_name,
+                repo_folder=args.output_dir,
+                vae_path=args.pretrained_vae_model_name_or_path,
+            )
+            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__":
+    args = parse_args()
+    main(args)