* fix: training resume from fp16.
* add: comment
* remove residue from another branch.
* remove more residues.
* thanks to Younes; no hacks.
* style.
* clean things a bit and modularize _set_state_dict_into_text_encoder
* add comment about the fix detailed.
* support compile
* make style
* move unwrap_model inside function
* change unwrap call
* run make style
* Update examples/dreambooth/train_dreambooth.py
Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>
* Revert "Update examples/dreambooth/train_dreambooth.py"
This reverts commit 70ab09732e.
---------
Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>
* remove validation args from textual onverson tests
* reduce number of train steps in textual inversion tests
* fix: directories.
* debig
* fix: directories.
* remove validation tests from textual onversion
* try reducing the time of test_text_to_image_checkpointing_use_ema
* fix: directories
* speed up test_text_to_image_checkpointing
* speed up test_text_to_image_checkpointing_checkpoints_total_limit_removes_multiple_checkpoints
* fix
* speed up test_instruct_pix2pix_checkpointing_checkpoints_total_limit_removes_multiple_checkpoints
* set checkpoints_total_limit to 2.
* test_text_to_image_lora_checkpointing_checkpoints_total_limit_removes_multiple_checkpoints speed up
* speed up test_unconditional_checkpointing_checkpoints_total_limit_removes_multiple_checkpoints
* debug
* fix: directories.
* speed up test_instruct_pix2pix_checkpointing_checkpoints_total_limit
* speed up: test_controlnet_checkpointing_checkpoints_total_limit_removes_multiple_checkpoints
* speed up test_controlnet_sdxl
* speed up dreambooth tests
* speed up test_dreambooth_lora_checkpointing_checkpoints_total_limit_removes_multiple_checkpoints
* speed up test_custom_diffusion_checkpointing_checkpoints_total_limit_removes_multiple_checkpoints
* speed up test_text_to_image_lora_sdxl_text_encoder_checkpointing_checkpoints_total_limit
* speed up # checkpoint-2 should have been deleted
* speed up examples/text_to_image/test_text_to_image.py::TextToImage::test_text_to_image_checkpointing_checkpoints_total_limit
* additional speed ups
* style
* fix: unscale fp16 gradient problem
* fix for dreambooth lora sdxl
* make the type-casting conditional.
* Apply suggestions from code review
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
---------
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* Additions:
- support for different lr for text encoder
- support for Prodigy optimizer
- support for min snr gamma
- support for custom captions and dataset loading from the hub
* adjusted --caption_column behaviour (to -not- use the second column of the dataset by default if --caption_column is not provided)
* fixed --output_dir / --model_dir_name confusion
* added --repeats, --adam_weight_decay_text_encoder
+ some fixes
* Update examples/dreambooth/train_dreambooth_lora_sdxl.py
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* Update examples/dreambooth/train_dreambooth_lora_sdxl.py
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* Update examples/dreambooth/train_dreambooth_lora_sdxl.py
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* - import compute_snr from diffusers/training_utils.py
- cluster adamw together
- when using 'prodigy', if --train_text_encoder == True and --text_encoder_lr != --learning rate, changes the lr of the text encoders optimization params to be --learning_rate (otherwise errors)
* shape fixes when custom captions are used
* formatting and a little cleanup
* code styling
* --repeats default value fixed, changed to 1
* bug fix - removed redundant lines of embedding concatenation when using prior_preservation (that duplicated class_prompt embeddings)
* changed dataset loading logic according to the following usecases (to avoid unnecessary dependency on datasets)-
1. user provides --dataset_name
2. user provides local dir --instance_data_dir that contains a metadata .jsonl file
3. user provides local dir --instance_data_dir that contains only images
in cases [1,2] we import datasets and use load_dataset method, in case [3] we process the data same as in the original script setting
* styling fix
* arg name fix
* adjusted the --repeats logic
* -removed redundant arg and 'if' when loading local folder with prompts
-updated readme template
-some default val fixes
-custom caption tests
* image path fix for readme
* code style
* bug fix
* --caption_column arg
* readme fix
---------
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
Co-authored-by: Linoy Tsaban <linoy@huggingface.co>
* fix error reported 'find_unused_parameters' running in mutiple GPUs or NPUs
* fix code check of importing module by its alphabetic order
---------
Co-authored-by: jiaqiw <wangjiaqi50@huawei.com>
Co-authored-by: Dhruv Nair <dhruv.nair@gmail.com>
* min-SNR gamma for Dreambooth training
* Align the mse_loss_weights style with SDXL training example
---------
Co-authored-by: bghira <bghira@users.github.com>
Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>
* dreambooth training
* train_dreambooth validation scheduler
* set a particular scheduler via a string
* modify readme after setting a particular scheduler via a string
* modify readme after setting a particular scheduler
* use importlib to set a particular scheduler
* import with correct sort
* make safetensors default
* set default save method as safetensors
* update tests
* update to support saving safetensors
* update test to account for safetensors default
* update example tests to use safetensors
* update example to support safetensors
* update unet tests for safetensors
* fix failing loader tests
* fix qc issues
* fix pipeline tests
* fix example test
---------
Co-authored-by: Dhruv Nair <dhruv.nair@gmail.com>
