GitLab

[examples/advanced_diffusion_training] bug fixes and improvements for LoRA Dreambooth SDXL advanced training script (#5935)

* imports and readme bug fixes

* bug fix - ensures text_encoder params are dtype==float32 (when using pivotal tuning) even if the rest of the model is loaded in fp16

* added pivotal tuning to readme

* mapping token identifier to new inserted token in validation prompt (if used)

* correct default value of --train_text_encoder_frac

* change default value of  --adam_weight_decay_text_encoder

* validation prompt generations when using pivotal tuning bug fix

* style fix

* textual inversion embeddings name change

* style fix

* bug fix - stopping text encoder optimization halfway

* readme - will include token abstraction and new inserted tokens when using pivotal tuning
- added type to --num_new_tokens_per_abstraction

* style fix

---------
Co-authored-by: Linoy Tsaban <linoy@huggingface.co>

Update __init__.py

* make `requires_safety_checker` a kwarg instead of a positional argument as it's more future-proof

* apply `make style` formatting edits

* add image_encoder to arguments and pass to super constructor

Update svd.md

fix url

* add diffusers example

* add diffusers example

* Comment about making it faster

* Apply suggestions from code review
Co-authored-by: Pedro Cuenca <pedro@huggingface.co>

---------
Co-authored-by: Pedro Cuenca <pedro@huggingface.co>

* begin model

* finish blocks

* add_embedding

* addition_time_embed_dim

* use TimestepEmbedding

* fix temporal res block

* fix time_pos_embed

* fix add_embedding

* add conversion script

* fix model

* up

* add new resnet blocks

* make forward work

* return sample in original shape

* fix temb shape in TemporalResnetBlock

* add spatio temporal transformers

* add vae blocks

* fix blocks

* update

* update

* fix shapes in Alphablender and add time activation in res blcok

* use new blocks

* style

* fix temb shape

* fix SpatioTemporalResBlock

* reuse TemporalBasicTransformerBlock

* fix TemporalBasicTransformerBlock

* use TransformerSpatioTemporalModel

* fix TransformerSpatioTemporalModel

* fix time_context dim

* clean up

* make temb optional

* add blocks

* rename model

* update conversion script

* remove UNetMidBlockSpatioTemporal

* add in init

* remove unused arg

* remove unused arg
...

* Fixed custom module importing on Windows

Windows use back slash and `os.path.join()` follows that convention.

* Apply suggestions from code review
Co-authored-by: Lucain <lucainp@gmail.com>

* Update pipeline_utils.py

---------
Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>
Co-authored-by: Lucain <lucainp@gmail.com>

* Adapt lora example scripts to use PEFT

* add to_out.0

* integrated sdxl for the text2video-zero pipeline

* make fix-copies

* fixed CI issues

* make fix-copies

* added docs and `copied from` statements

* added fast tests

* made a small change in docs

* quality+style check fix

* updated docs. added controlnet inference with sdxl

* added device compatibility for fast tests

* fixed docstrings

* changing vae upcasting

* remove torch.empty_cache to speed up inference
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* made fast tests to run on dummy models only, fixed copied from statements

* fixed testing utils imports

* Added bullet points for SDXL support

* fixed formatting & quality

* Update tests/pipelines/text_to_video/test_text_to_video_zero_sdxl.py
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* Update tests/pipelines/text_to_video/test_text_to_video_zero_sdxl.py
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* fixed minor error for merging

* fixed updates of sdxl

* made fast tests inherit from `PipelineTesterMixin` and run in 3-4secs on CPU

* make style && make quality

* reimplemented fast tests w/o default attn processor

* make style & make quality

* make fix-copies

* make fix-copies

* fixed docs

* make style & make quality & make fix-copies

* bug fix in cross attention

* make style && make quality

* make fix-copies

* fix gpu issues

* make fix-copies

* updated pipeline signature

---------
Co-authored-by: Vahram <vahram.tadevosyan@lambda-loginnode02.cm.cluster>
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
Co-authored-by: Dhruv Nair <dhruv.nair@gmail.com>

* Add SSD-1B support for controlnet model

* Add conditioning_channels into ControlNet init from unet

* Fix black formatting

* Isort fixes

* Adds SSD-1B controlnet pipeline test with UNetMidBlock2D as mid block

* Overrides failing ssd-1b tests

* Fixes tests after main branch update

* Fixes code quality checks

---------
Co-authored-by: Marko Kostiv <marko@linearity.io>
Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>

* first draft

* feedback

* Change pipeline_controlnet_inpaint.py to add ip-adapter support. Changes are similar to those in pipeline_controlnet

* Change tests for the StableDiffusionControlNetInpaintPipeline by adding image_encoder: None

* Update src/diffusers/pipelines/controlnet/pipeline_controlnet_inpaint.py
Co-authored-by: YiYi Xu <yixu310@gmail.com>

---------
Co-authored-by: YiYi Xu <yixu310@gmail.com>

* move several state dict conversion utils out of lora.py

* check

* check

* check

* check

* check

* check

* check

* revert back

* check

* check

* again check

* maybe fix?

* 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>

Fix typo in output_dir argument: "text-inversion-model" → "dreambooth-model"

* bug in MultiAdapter for Inpainting

* adapter_input is a list for MultiAdapter

---------
Co-authored-by: andres <andres@hax.ai>
Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>

* added en doc for Kandinsky3.0

* required changes

* Update docs/source/en/api/pipelines/kandinsky3.md

* Update docs/source/en/api/pipelines/kandinsky3.md

* Update docs/source/en/api/pipelines/kandinsky3.md

---------
Co-authored-by: YiYi Xu <yixu310@gmail.com>
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* control and custom diffusion

* dreambooth

* instructpix2pix and dreambooth ckpting

* t2i adapters.

* text to image ft

* textual inversion

* unconditional

* workflows

* import fix

* fix import

add to list

---------
Co-authored-by: Aflalo <estellea@isl-gpu27.rr.intel.com>
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
Co-authored-by: YiYi Xu <yixu310@gmail.com>

* fix

* minor edits

Co-authored-by: yiyixuxu <yixu310@gmail,com>

* [community pipeline] dps impl

* add type checking

* pass ruff check

* ruff formatter

* [Tests] Make sure that we don't run tests mulitple times

* [Tests] Make sure that we don't run tests mulitple times

* [Tests] Make sure that we don't run tests mulitple times

* add comments to explain the code better

* add comments to explain the code better

* add comments to explain the code better

* add comments to explain the code better

* add comments to explain the code better

* fix more

* fix more

* fix more

* fix more

* fix more

* fix more

fix: error on device for `lpw_stable_diffusion_xl` pipeline if `pipe.enable_sequential_cpu_offload()` enabled (#5885)

fix: set device for pipe.enable_sequential_cpu_offload()

* I added a new doc string to the class. This is more flexible to understanding other developers what are doing and where it's using.

* Update src/diffusers/models/unet_2d_blocks.py

This changes suggest by maintener.
Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>

* Update src/diffusers/models/unet_2d_blocks.py

Add suggested text
Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>

* Update unet_2d_blocks.py

I changed the Parameter to Args text.

* Update unet_2d_blocks.py

proper indentation set in this file.

* Update unet_2d_blocks.py

a little bit of change in the act_fun argument line.

* I run the black command to reformat style in the code

* Update unet_2d_blocks.py

similar doc-string add to have in the original diffusion repository.

* I enhanced the code by replacing multiple redundant variables with a single variable, as they all served the same purpose. Additionally, I utilized the get_activation function for improved flexibility in choosing activation functions.

* Using as black package to reformated my file

* reverte some changes

* Remove conv_out_padding variables and using as conv_in_padding

* conv_out_padding create and add them into the code.

* run black command to solving styling problem

* add little bit space between comment and import statement

* I am utilizing the ruff library to address the style issues in my Makefile.

---------
Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>
Co-authored-by: YiYi Xu <yixu310@gmail.com>
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

add IPEX pipeline LoRA weights loading support

* Fix bug related to parsing unet_time_cond_proj_dim.

* Fix analogous bug in the SD-XL LCM distillation script.

Avoid computing min() that is expensive when do_normalize is False

Avoid extra computing when do_normalize is False

Signed-off-by: Wang, Yi A <yi.a.wang@intel.com>

* Add custom timesteps support to LCMScheduler.

* Add custom timesteps support to StableDiffusionPipeline.

* Add custom timesteps support to StableDiffusionXLPipeline.

* Add custom timesteps support to remaining Stable Diffusion pipelines which support LCMScheduler (img2img, inpaint).

* Add custom timesteps support to remaining Stable Diffusion XL pipelines which support LCMScheduler (img2img, inpaint).

* Add custom timesteps support to StableDiffusionControlNetPipeline.

* Add custom timesteps support to T21 Stable Diffusion (XL) Adapters.

* Clean up Stable Diffusion inpaint tests.

* Manually add support for custom timesteps to AltDiffusion pipelines since make fix-copies doesn't appear to work correctly (it deletes the whole pipeline).

* make style

* Refactor pipeline timestep handling into the retrieve_timesteps function.

* deprecated: KarrasVeScheduler, ScoreSdeVpScheduler

* delete tests relevant to deprecated schedulers

* chore: run make style

* fix: import error caused due to incorrect _import_structure after deprecation

* fix: ScoreSdeVpScheduler was not importable from diffusers

* remove import added by assumption

* Update src/diffusers/schedulers/__init__.py as suggested by @patrickvonplaten
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* make it a part deprecated

* Apply suggestions from code review
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* Fix

* fix

* fix doc

* fix doc....again.......

* remove karras_ve test folder
Co-Authored-By: YiYi Xu <yixu310@gmail.com>

---------
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>
Co-authored-by: YiYi Xu <yixu310@gmail.com>
Co-authored-by: yiyixuxu <yixu310@gmail,com>

add support for gradient checkpointing in transformer_2d

* [Fix: pixart-alpha]
add ASPECT_RATIO_512_BIN in use_resolution_binning for random 512px image generation.

* add slow test file for 512px generation without resolution binning

* fix: slow tests for resolution binning.

---------
Co-authored-by: jschen <chenjunsong4@h-partners.com>
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>

* add variant

* add variant

* Apply suggestions from code review

* reformat

* fix: textual_inversion.py

* fix: variant in model_info

---------
Co-authored-by: sayakpaul <spsayakpaul@gmail.com>

* finalize

* finalize

* finalize

* add slow test

* add slow test

* add slow test

* Fix more

* add slow test

* fix more

* fix more

* fix more

* fix more

* fix more

* fix more

* fix more

* fix more

* fix more

* Better

* Fix more

* Fix more

* add slow test

* Add auto pipelines

* add slow test

* Add all

* add slow test

* add slow test

* add slow test

* add slow test

* add slow test

* Apply suggestions from code review

* add slow test

* add slow test