Skip to content

GitLab

Switch branch/tag
  1. 16 Apr, 2025 2 commits
  3. 15 Apr, 2025 1 commit
    • Hameer Abbasi's avatar
      [LoRA] Add LoRA support to AuraFlow (#10216) · 9352a5ca
      Hameer Abbasi authored 
      

* Add AuraFlowLoraLoaderMixin

* Add comments, remove qkv fusion

* Add Tests

* Add AuraFlowLoraLoaderMixin to documentation

* Add Suggested changes

* Change attention_kwargs->joint_attention_kwargs

* Rebasing derp.

* fix

* fix

* Quality fixes.

* make style

* `make fix-copies`

* `ruff check --fix`

* Attept 1 to fix tests.

* Attept 2 to fix tests.

* Attept 3 to fix tests.

* Address review comments.

* Rebasing derp.

* Get more tests passing by copying from Flux. Address review comments.

* `joint_attention_kwargs`->`attention_kwargs`

* Add `lora_scale` property for te LoRAs.

* Make test better.

* Remove useless property.

* Skip TE-only tests for AuraFlow.

* Support LoRA for non-CLIP TEs.

* Restore LoRA tests.

* Undo adding LoRA support for non-CLIP TEs.

* Undo support for TE in AuraFlow LoRA.

* `make fix-copies`

* Sync with upstream changes.

* Remove unneeded stuff.

* Mirror `Lumina2`.

* Skip for MPS.

* Address review comments.

* Remove duplicated code.

* Remove unnecessary code.

* Remove repeated docs.

* Propagate attention.

* Fix TE target modules.

* MPS fix for LoRA tests.

* Unrelated TE LoRA tests fix.

* Fix AuraFlow LoRA tests by applying to the right denoiser layers.
Co-authored-by: default avatarAstraliteHeart <81396681+AstraliteHeart@users.noreply.github.com>

* Apply style fixes

* empty commit

* Fix the repo consistency issues.

* Remove unrelated changes.

* Style.

* Fix `test_lora_fuse_nan`.

* fix quality issues.

* `pytest.xfail` -> `ValueError`.

* Add back `skip_mps`.

* Apply style fixes

* `make fix-copies`

---------
Co-authored-by: default avatarWarlord-K <warlordk28@gmail.com>
Co-authored-by: default avatarhlky <hlky@hlky.ac>
Co-authored-by: default avatarSayak Paul <spsayakpaul@gmail.com>
Co-authored-by: default avatarAstraliteHeart <81396681+AstraliteHeart@users.noreply.github.com>
Co-authored-by: default avatargithub-actions[bot] <github-actions[bot]@users.noreply.github.com>
      9352a5ca
  5. 14 Apr, 2025 1 commit
  7. 10 Apr, 2025 2 commits
  9. 09 Apr, 2025 2 commits
  11. 08 Apr, 2025 2 commits
    • hlky's avatar
      Flux quantized with lora (#10990) · 5d49b3e8
      hlky authored 
      

* Flux quantized with lora

* fix

* changes

* Apply suggestions from code review
Co-authored-by: default avatarSayak Paul <spsayakpaul@gmail.com>

* Apply style fixes

* enable model cpu offload()

* Update src/diffusers/loaders/lora_pipeline.py
Co-authored-by: default avatarhlky <hlky@hlky.ac>

* update

* Apply suggestions from code review

* update

* add peft as an additional dependency for gguf

---------
Co-authored-by: default avatarSayak Paul <spsayakpaul@gmail.com>
Co-authored-by: default avatargithub-actions[bot] <github-actions[bot]@users.noreply.github.com>
Co-authored-by: default avatarDhruv Nair <dhruv.nair@gmail.com>
      5d49b3e8
    • Benjamin Bossan's avatar
      [LoRA] Implement hot-swapping of LoRA (#9453) · fb544996
      Benjamin Bossan authored 
      * [WIP][LoRA] Implement hot-swapping of LoRA

This PR adds the possibility to hot-swap LoRA adapters. It is WIP.

Description

As of now, users can already load multiple LoRA adapters. They can
offload existing adapters or they can unload them (i.e. delete them).
However, they cannot "hotswap" adapters yet, i.e. substitute the weights
from one LoRA adapter with the weights of another, without the need to
create a separate LoRA adapter.

Generally, hot-swapping may not appear not super useful but when the
model is compiled, it is necessary to prevent recompilation. See #9279
for more context.

Caveats

To hot-swap a LoRA adapter for another, these two adapters should target
exactly the same layers and the "hyper-parameters" of the two adapters
should be identical. For instance, the LoRA alpha has to be the same:
Given that we keep the alpha from the first adapter, the LoRA scaling
would be incorrect for the second adapter otherwise.

Theoretically, we could override the scaling dict with the alpha values
derived from the second adapter's config, but changing the dict will
trigger a guard for recompilation, defeating the main purpose of the
feature.

I also found that compilation flags can have an impact on whether this
works or not. E.g. when passing "reduce-overhead", there will be errors
of the type:

> input name: arg861_1. data pointer changed from 139647332027392 to
139647331054592

I don't know enough about compilation to determine whether this is
problematic or not.

Current state

This is obviously WIP right now to collect feedback and discuss which
direction to take this. If this PR turns out to be useful, the
hot-swapping functions will be added to PEFT itself and can be imported
here (or there is a separate copy in diffusers to avoid the need for a
min PEFT version to use this feature).

Moreover, more tests need to be added to better cover this feature,
although we don't necessarily need tests for the hot-swapping
functionality itself, since those tests will be added to PEFT.

Furthermore, as of now, this is only implemented for the unet. Other
pipeline components have yet to implement this feature.

Finally, it should be properly documented.

I would like to collect feedback on the current state of the PR before
putting more time into finalizing it.

* Reviewer feedback

* Reviewer feedback, adjust test

* Fix, doc

* Make fix

* Fix for possible g++ error

* Add test for recompilation w/o hotswapping

* Make hotswap work

Requires https://github.com/huggingface/peft/pull/2366

More changes to make hotswapping work. Together with the mentioned PEFT
PR, the tests pass for me locally.

List of changes:

- docstring for hotswap
- remove code copied from PEFT, import from PEFT now
- adjustments to PeftAdapterMixin.load_lora_adapter (unfortunately, some
  state dict renaming was necessary, LMK if there is a better solution)
- adjustments to UNet2DConditionLoadersMixin._process_lora: LMK if this
  is even necessary or not, I'm unsure what the overall relationship is
  between this and PeftAdapterMixin.load_lora_adapter
- also in UNet2DConditionLoadersMixin._process_lora, I saw that there is
  no LoRA unloading when loading the adapter fails, so I added it
  there (in line with what happens in PeftAdapterMixin.load_lora_adapter)
- rewritten tests to avoid shelling out, make the test more precise by
  making sure that the outputs align, parametrize it
- also checked the pipeline code mentioned in this comment:
  https://github.com/huggingface/diffusers/pull/9453#issuecomment-2418508871;


  when running this inside the with
  torch._dynamo.config.patch(error_on_recompile=True) context, there is
  no error, so I think hotswapping is now working with pipelines.

* Address reviewer feedback:

- Revert deprecated method
- Fix PEFT doc link to main
- Don't use private function
- Clarify magic numbers
- Add pipeline test

Moreover:
- Extend docstrings
- Extend existing test for outputs != 0
- Extend existing test for wrong adapter name

* Change order of test decorators

parameterized.expand seems to ignore skip decorators if added in last
place (i.e. innermost decorator).

* Split model and pipeline tests

Also increase test coverage by also targeting conv2d layers (support of
which was added recently on the PEFT PR).

* Reviewer feedback: Move decorator to test classes

... instead of having them on each test method.

* Apply suggestions from code review
Co-authored-by: default avatarhlky <hlky@hlky.ac>

* Reviewer feedback: version check, TODO comment

* Add enable_lora_hotswap method

* Reviewer feedback: check _lora_loadable_modules

* Revert changes in unet.py

* Add possibility to ignore enabled at wrong time

* Fix docstrings

* Log possible PEFT error, test

* Raise helpful error if hotswap not supported

I.e. for the text encoder

* Formatting

* More linter

* More ruff

* Doc-builder complaint

* Update docstring:

- mention no text encoder support yet
- make it clear that LoRA is meant
- mention that same adapter name should be passed

* Fix error in docstring

* Update more methods with hotswap argument

- SDXL
- SD3
- Flux

No changes were made to load_lora_into_transformer.

* Add hotswap argument to load_lora_into_transformer

For SD3 and Flux. Use shorter docstring for brevity.

* Extend docstrings

* Add version guards to tests

* Formatting

* Fix LoRA loading call to add prefix=None

See:
https://github.com/huggingface/diffusers/pull/10187#issuecomment-2717571064



* Run make fix-copies

* Add hot swap documentation to the docs

* Apply suggestions from code review
Co-authored-by: default avatarSteven Liu <59462357+stevhliu@users.noreply.github.com>

---------
Co-authored-by: default avatarSayak Paul <spsayakpaul@gmail.com>
Co-authored-by: default avatarhlky <hlky@hlky.ac>
Co-authored-by: default avatarYiYi Xu <yixu310@gmail.com>
Co-authored-by: default avatarSteven Liu <59462357+stevhliu@users.noreply.github.com>
      fb544996
  13. 04 Apr, 2025 2 commits
  15. 26 Mar, 2025 2 commits
  17. 21 Mar, 2025 1 commit
  19. 20 Mar, 2025 1 commit
  21. 19 Mar, 2025 1 commit
  23. 14 Mar, 2025 1 commit
  25. 12 Mar, 2025 1 commit
  27. 11 Mar, 2025 3 commits
  29. 10 Mar, 2025 3 commits
  31. 08 Mar, 2025 1 commit
  33. 07 Mar, 2025 2 commits
  35. 06 Mar, 2025 2 commits
  37. 04 Mar, 2025 3 commits
  39. 03 Mar, 2025 1 commit
    • Teriks's avatar
      Fix SD2.X clip single file load projection_dim (#10770) · 9e910c46
      Teriks authored 
      

* Fix SD2.X clip single file load projection_dim

Infer projection_dim from the checkpoint before loading
from pretrained, override any incorrect hub config.

Hub configuration for SD2.X specifies projection_dim=512
which is incorrect for SD2.X checkpoints loaded from civitai
and similar.

Exception was previously thrown upon attempting to
load_model_dict_into_meta for SD2.X single file checkpoints.

Such LDM models usually require projection_dim=1024

* convert_open_clip_checkpoint use hidden_size for text_proj_dim

* convert_open_clip_checkpoint, revert checkpoint[text_proj_key].shape[1] -> [0]

values are identical

---------
Co-authored-by: default avatarTeriks <Teriks@users.noreply.github.com>
Co-authored-by: default avatarDhruv Nair <dhruv.nair@gmail.com>
      9e910c46
  41. 25 Feb, 2025 1 commit
  43. 24 Feb, 2025 2 commits
  45. 21 Feb, 2025 1 commit
  47. 20 Feb, 2025 2 commits