* Fixes#6418 Advanced Dreambooth LoRa Training
* change order of import to fix nit
* fix nit, use cast_training_params
* remove torch.compile fix, will move to a new PR
* remove unnecessary import
* Enable image resizing to adjust its height and width in StableDiffusionXLInstructPix2PixPipeline
* Ensure that validation is performed at every 'validation_step', not at every step
* 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>
* base template file - train_instruct_pix2pix.py
* additional import and parser argument requried for lora
* finetune only instructpix2pix model -- no need to include these layers
* inject lora layers
* freeze unet model -- only lora layers are trained
* training modifications to train only lora parameters
* store only lora parameters
* move train script to research project
* run quality and style code checks
* move train script to a new folder
* add README
* update README
* update references in README
---------
Co-authored-by: Rahul Raman <rahulraman@gmail.com>
Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>
* enable stable-xl textual inversion
* check if optimizer_2 exists
* check text_encoder_2 before using
* add textual inversion for sdxl in a single file
* fix style
* fix example style
* reset for error changes
* add readme for sdxl
* fix style
* disable autocast as it will cause cast error when weight_dtype=bf16
* fix spelling error
* fix style and readme and 8bit optimizer
* add README_sdxl.md link
* add tracker key on log_validation
* run style
* rm the second center crop
---------
Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>
* add: experimental script for diffusion dpo training.
* random_crop cli.
* fix: caption tokenization.
* fix: pixel_values index.
* fix: grad?
* debug
* fix: reduction.
* fixes in the loss calculation.
* style
* fix: unwrap call.
* fix: validation inference.
* add: initial sdxl script
* debug
* make sure images in the tuple are of same res
* fix model_max_length
* report print
* boom
* fix: numerical issues.
* fix: resolution
* comment about resize.
* change the order of the training transformation.
* save call.
* debug
* remove print
* manually detaching necessary?
* use the same vae for validation.
* add: readme.
* unwrap text encoder when saving hook only for full text encoder tuning
* unwrap text encoder when saving hook only for full text encoder tuning
* save embeddings in each checkpoint as well
* save embeddings in each checkpoint as well
* save embeddings in each checkpoint as well
* Update examples/advanced_diffusion_training/train_dreambooth_lora_sdxl_advanced.py
Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>
---------
Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>
* Make WDS pipeline interpolation type configurable.
* Make the VAE encoding batch size configurable.
* Make lora_alpha and lora_dropout configurable for LCM LoRA scripts.
* Generalize scalings_for_boundary_conditions function and make the timestep scaling configurable.
* Make LoRA target modules configurable for LCM-LoRA scripts.
* Move resolve_interpolation_mode to src/diffusers/training_utils.py and make interpolation type configurable in non-WDS script.
* apply suggestions from review
* change timesteps used to calculate snr when --with_prior_preservation is enabled
* change timesteps used to calculate snr when --with_prior_preservation is enabled (canonical script)
* style
* revert canonical script to before snr gamma change
---------
Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>
* Fix gradient-checkpointing option is ignored in SDXL+LoRA training. (#6388)
* Fix gradient-checkpointing option is ignored in SD+LoRA training.
* Fix gradient checkpoint is not applied to text encoders. (SDXL+LoRA)
---------
Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>
