* allow loading from repo with dot in name
* put new arg at the end to avoid breaking compatibility
* add test for loading repo with dot in name
---------
Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>
* [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: hlky <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: Steven Liu <59462357+stevhliu@users.noreply.github.com>
---------
Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>
Co-authored-by: hlky <hlky@hlky.ac>
Co-authored-by: YiYi Xu <yixu310@gmail.com>
Co-authored-by: Steven Liu <59462357+stevhliu@users.noreply.github.com>
* Check correct model type is passed to `from_pretrained`
* Flax, skip scheduler
* test_wrong_model
* Fix for scheduler
* Update tests/pipelines/test_pipelines.py
Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>
* EnumMeta
* Flax
* scheduler in expected types
* make
* type object 'CLIPTokenizer' has no attribute '_PipelineFastTests__name'
* support union
* fix typing in kandinsky
* make
* add LCMScheduler
* 'LCMScheduler' object has no attribute 'sigmas'
* tests for wrong scheduler
* make
* update
* warning
* tests
* Update src/diffusers/pipelines/pipeline_utils.py
Co-authored-by: Dhruv Nair <dhruv.nair@gmail.com>
* import FlaxSchedulerMixin
* skip scheduler
---------
Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>
Co-authored-by: Dhruv Nair <dhruv.nair@gmail.com>
* feat: support saving a model in sharded checkpoints.
* feat: make loading of sharded checkpoints work.
* add tests
* cleanse the loading logic a bit more.
* more resilience while loading from the Hub.
* parallelize shard downloads by using snapshot_download()/
* default to a shard size.
* more fix
* Empty-Commit
* debug
* fix
* uality
* more debugging
* fix more
* initial comments from Benjamin
* move certain methods to loading_utils
* add test to check if the correct number of shards are present.
* add a test to check if loading of sharded checkpoints from the Hub is okay
* clarify the unit when passed as an int.
* use hf_hub for sharding.
* remove unnecessary code
* remove unnecessary function
* lucain's comments.
* fixes
* address high-level comments.
* fix test
* subfolder shenanigans./
* Update src/diffusers/utils/hub_utils.py
Co-authored-by: Lucain <lucainp@gmail.com>
* Apply suggestions from code review
Co-authored-by: Lucain <lucainp@gmail.com>
* remove _huggingface_hub_version as not needed.
* address more feedback.
* add a test for local_files_only=True/
* need hf hub to be at least 0.23.2
* style
* final comment.
* clean up subfolder.
* deal with suffixes in code.
* _add_variant default.
* use weights_name_pattern
* remove add_suffix_keyword
* clean up downloading of sharded ckpts.
* don't return something special when using index.json
* fix more
* don't use bare except
* remove comments and catch the errors better
* fix a couple of things when using is_file()
* empty
---------
Co-authored-by: Lucain <lucainp@gmail.com>
* Add properties and `IPAdapterTesterMixin` tests for `StableDiffusionPanoramaPipeline`
* Fix variable name typo and update comments
* Update deprecated `output_type="numpy"` to "np" in test files
* Discard changes to src/diffusers/pipelines/stable_diffusion_panorama/pipeline_stable_diffusion_panorama.py
* Update test_stable_diffusion_panorama.py
* Update numbers in README.md
* Update get_guidance_scale_embedding method to use timesteps instead of w
* Update number of checkpoints in README.md
* Add type hints and fix var name
* Fix PyTorch's convention for inplace functions
* Fix a typo
* Revert "Fix PyTorch's convention for inplace functions"
This reverts commit 74350cf65b.
* Fix typos
* Indent
* Refactor get_guidance_scale_embedding method in LEditsPPPipelineStableDiffusionXL class
* 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>
* iterate over unique tokens to avoid duplicate replacements
* added test for multiple references to multi embedding
* adhere to black formatting
* reorder test post-rebase
* Run ControlNet compile test in a separate subprocess
`torch.compile()` spawns several subprocesses and the GPU memory used
was not reclaimed after the test ran. This approach was taken from
`transformers`.
* Style
* Prepare a couple more compile tests to run in subprocess.
* Use require_torch_2 decorator.
* Test inpaint_compile in subprocess.
* Run img2img compile test in subprocess.
* Run stable diffusion compile test in subprocess.
* style
* Temporarily trigger on pr to test.
* Revert "Temporarily trigger on pr to test."
This reverts commit 82d76868dd.
* up
* fix more
* Apply suggestions from code review
* fix more
* fix more
* Check it
* Remove 16:8
* fix more
* fix more
* fix more
* up
* up
* Test only stable diffusion
* Test only two files
* up
* Try out spinning up processes that can be killed
* up
* Apply suggestions from code review
* up
* up
* enable deterministic pytorch and cuda operations.
* disable manual seeding.
* make style && make quality for unet_2d tests.
* enable determinism for the unet2dconditional model.
* add CUBLAS_WORKSPACE_CONFIG for better reproducibility.
* relax tolerance (very weird issue, though).
* revert to torch manual_seed() where needed.
* relax more tolerance.
* better placement of the cuda variable and relax more tolerance.
* enable determinism for 3d condition model.
* relax tolerance.
* add: determinism to alt_diffusion.
* relax tolerance for alt diffusion.
* dance diffusion.
* dance diffusion is flaky.
* test_dict_tuple_outputs_equivalent edit.
* fix two more tests.
* fix more ddim tests.
* fix: argument.
* change to diff in place of difference.
* fix: test_save_load call.
* test_save_load_float16 call.
* fix: expected_max_diff
* fix: paint by example.
* relax tolerance.
* add determinism to 1d unet model.
* torch 2.0 regressions seem to be brutal
* determinism to vae.
* add reason to skipping.
* up tolerance.
* determinism to vq.
* determinism to cuda.
* determinism to the generic test pipeline file.
* refactor general pipelines testing a bit.
* determinism to alt diffusion i2i
* up tolerance for alt diff i2i and audio diff
* up tolerance.
* determinism to audioldm
* increase tolerance for audioldm lms.
* increase tolerance for paint by paint.
* increase tolerance for repaint.
* determinism to cycle diffusion and sd 1.
* relax tol for cycle diffusion 🚲
* relax tol for sd 1.0
* relax tol for controlnet.
* determinism to img var.
* relax tol for img variation.
* tolerance to i2i sd
* make style
* determinism to inpaint.
* relax tolerance for inpaiting.
* determinism for inpainting legacy
* relax tolerance.
* determinism to instruct pix2pix
* determinism to model editing.
* model editing tolerance.
* panorama determinism
* determinism to pix2pix zero.
* determinism to sag.
* sd 2. determinism
* sd. tolerance
* disallow tf32 matmul.
* relax tolerance is all you need.
* make style and determinism to sd 2 depth
* relax tolerance for depth.
* tolerance to diffedit.
* tolerance to sd 2 inpaint.
* up tolerance.
* determinism in upscaling.
* tolerance in upscaler.
* more tolerance relaxation.
* determinism to v pred.
* up tol for v_pred
* unclip determinism
* determinism to unclip img2img
* determinism to text to video.
* determinism to last set of tests
* up tol.
* vq cumsum doesn't have a deterministic kernel
* relax tol
* relax tol
