GitLab

* 2x speedup using memory efficient attention

* remove einops dependency

* Swap K, M in op instantiation

* Simplify code, remove unnecessary maybe_init call and function, remove unused self.scale parameter

* make xformers a soft dependency

* remove one-liner functions

* change one letter variable to appropriate names

* Remove Env variable dependency, remove MemoryEfficientCrossAttention class and use enable_xformers_memory_efficient_attention method

* Add memory efficient attention toggle to img2img and inpaint pipelines

* Clearer management of xformers' availability

* update optimizations markdown to add info about memory efficient attention

* add benchmarks for TITAN RTX

* More detailed explanation of how the mem eff benchmark were ran

* Removing autocast from optimization markdown

* import_utils: import torch only if is available
Co-authored-by: Nouamane Tazi <nouamane98@gmail.com>

Remove some unused parameter

The `downsample_padding` parameter does not seem to be used in `CrossAttnUpBlock2D` (or by any up block for that matter) so removing it.

* Remove nn sequential

* up

* Docs: refer to pre-RC version of PyTorch 1.13.0.

* Remove temporary workaround for unavailable op.

* Update comment to make it less ambiguous.

* Remove use of contiguous in mps.

It appears to not longer be necessary.

* Special case: use einsum for much better performance in mps

* Update mps docs.

* MPS: make pipeline work in half precision.

* fix `upsample_nearest_nhwc` for large bsz

* fix `upsample_nearest_nhwc` for large bsz

* add in fp16

* up

* start

* add more logic

* Update src/diffusers/models/unet_2d_condition_flax.py

* match weights

* up

* make model work

* making class more general, fixing missed file rename

* small fix

* make new conversion work

* up

* finalize conversion

* up

* first batch of variable renamings

* remove c and c_prev var names

* add mid and out block structure

* add pipeline

* up

* finish conversion

* finish

* upload

* more fixes

* Apply suggestions from code review

* add attr

* up

* uP

* up

* finish tests

* finish

* uP

* finish

* fix test

* up

* naming consistency in tests

* Apply suggestions from code review
Co-authored-by: Suraj Patil <surajp815@gmail.com>
Co-authored-by: Pedro Cuenca <pedro@huggingface.co>
Co-authored-by: Nathan Lambert <nathan@huggingface.co>
Co-authored-by: Anton Lozhkov <anton@huggingface.co>

* remove hardcoded 16

* Remove bogus

* fix some stuff

* finish

* improve logging

* docs

* upload
Co-authored-by: Nathan Lambert <nol@berkeley.edu>
Co-authored-by: Suraj Patil <surajp815@gmail.com>
Co-authored-by: Pedro Cuenca <pedro@huggingface.co>
Co-authored-by: Nathan Lambert <nathan@huggingface.co>
Co-authored-by: Anton Lozhkov <anton@huggingface.co>

* Docs: refer to pre-RC version of PyTorch 1.13.0.

* Remove temporary workaround for unavailable op.

* Update comment to make it less ambiguous.

* Remove use of contiguous in mps.

It appears to not longer be necessary.

* Special case: use einsum for much better performance in mps

* Update mps docs.

* Minor doc update.

* Accept suggestion
Co-authored-by: Anton Lozhkov <anton@huggingface.co>
Co-authored-by: Anton Lozhkov <anton@huggingface.co>

* add or fix license formatting

* fix quality

* pass norm_num_groups param and add tests

* set resnet_groups for FlaxUNetMidBlock2D

* fixed docstrings

* fixed typo

* using is_flax_available util and created require_flax decorator

* support bf16 for stable diffusion

* fix typo

* address review comments

fix typo docstring

* clean up resnet.py

* make style and quality

* minor formatting

* handle dtype in vae and image2image pipeline

* fix inpaint in fp16

* dtype should be handled in add_noise

* style

* address review comments

* add simple fast tests to check fp16

* fix test name

* put mask in fp16

This reverts commit 2e209c30.

Temporarily remove Flax modules from the public API

* Removing `autocast` for `35-25% speedup`.

* iQuality

* Adding a slow test.

* Fixing mps noise generation.

* Raising error on wrong device, instead of just casting on behalf of user.

* Quality.

* fix merge
Co-authored-by: Nouamane Tazi <nouamane98@gmail.com>

* renamed single letter variables

* renamed x to meaningful variable in resnet.py

Hello @patil-suraj can you verify it
Thanks

* Reformatted using black

* renamed x to meaningful variable in resnet.py

Hello @patil-suraj can you verify it
Thanks

* reformatted the files

* modified unboundlocalerror in line 374

* removed referenced before error

* renamed single variable x -> hidden_state, p-> pad_value
Co-authored-by: Nikhil A V <nikhilav@Nikhils-MacBook-Pro.local>
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
Co-authored-by: Suraj Patil <surajp815@gmail.com>

Remove comments no longer appropriate.

There were casting operations before, they are now gone.

fix docstring

fixes #709

* revert using baddbmm in attention
- to fix `test_stable_diffusion_memory_chunking` test

* styling

* Allow resolutions that are not multiples of 64

* ran black

* fix bug

* add test

* more explanation

* more comments
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* initial commit

* make UNet stream capturable

* try to fix noise_pred value

* remove cuda graph and keep NB

* non blocking unet with PNDMScheduler

* make timesteps np arrays for pndm scheduler
because lists don't get formatted to tensors in `self.set_format`

* make max async in pndm

* use channel last format in unet

* avoid moving timesteps device in each unet call

* avoid memcpy op in `get_timestep_embedding`

* add `channels_last` kwarg to `DiffusionPipeline.from_pretrained`

* update TODO

* replace `channels_last` kwarg with `memory_format` for more generality

* revert the channels_last changes to leave it for another PR

* remove non_blocking when moving input ids to device

* remove blocking from all .to() operations at beginning of pipeline

* fix merging

* fix merging

* model can run in other precisions without autocast

* attn refactoring

* Revert "attn refactoring"

This reverts commit 0c70c0e189cd2c4d8768274c9fcf5b940ee310fb.

* remove restriction to run conv_norm in fp32

* use `baddbmm` instead of `matmul`for better in attention for better perf

* removing all reshapes to test perf

* Revert "removing all reshapes to test perf"

This reverts commit 006ccb8a8c6bc7eb7e512392e692a29d9b1553cd.

* add shapes comments

* hardcore whats needed for jitting

* Revert "hardcore whats needed for jitting"

This reverts commit 2fa9c698eae2890ac5f8e367ca80532ecf94df9a.

* Revert "remove restriction to run conv_norm in fp32"

This reverts commit cec592890c32da3d1b78d38b49e4307aedf459b9.

* revert using baddmm in attention's forward

* cleanup comment

* remove restriction to run conv_norm in fp32. no quality loss was noticed

This reverts commit cc9bc1339c998ebe9e7d733f910c6d72d9792213.

* add more optimizations techniques to docs

* Revert "add shapes comments"

This reverts commit 31c58eadb8892f95478cdf05229adf678678c5f4.

* apply suggestions

* make quality

* apply suggestions

* styling

* `scheduler.timesteps` are now arrays so we dont need .to()

* remove useless .type()

* use mean instead of max in `test_stable_diffusion_inpaint_pipeline_k_lms`

* move scheduler timestamps to correct device if tensors

* add device to `set_timesteps` in LMSD scheduler

* `self.scheduler.set_timesteps` now uses device arg for schedulers that accept it

* quick fix

* styling

* remove kwargs from schedulers `set_timesteps`

* revert to using max in K-LMS inpaint pipeline test

* Revert "`self.scheduler.set_timesteps` now uses device arg for schedulers that accept it"

This reverts commit 00d5a51e5c20d8d445c8664407ef29608106d899.

* move timesteps to correct device before loop in SD pipeline

* apply previous fix to other SD pipelines

* UNet now accepts tensor timesteps even on wrong device, to avoid errors
- it shouldnt affect performance if timesteps are alrdy on correct device
- it does slow down performance if they're on the wrong device

* fix pipeline when timesteps are arrays with strides

renamed x to hidden_states

* Fix SpatialTransformer

* Fix SpatialTransformer
Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>

* documenting `attention_flax.py` file

* documenting `embeddings_flax.py`

* documenting `unet_blocks_flax.py`

* Add new objs to doc page

* document `vae_flax.py`

* Apply suggestions from code review

* modify `unet_2d_condition_flax.py`

* make style

* Apply suggestions from code review

* make style

* Apply suggestions from code review

* fix indent

* fix typo

* fix indent unet

* Update src/diffusers/models/vae_flax.py

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

* add grad ckpt to downsample blocks

* make it work

* don't pass gradient_checkpointing to upsample block

* add tests for UNet2DConditionModel

* add test_gradient_checkpointing

* add gradient_checkpointing for up and down blocks

* add functions to enable and disable grad ckpt

* remove the forward argument

* better naming

* make supports_gradient_checkpointing private

replace `dropout_prob` by `dropout` in `vae`

* [Flax] Fix unet and ddim scheduler

* correct

* finish

* WIP: flax FlaxDiffusionPipeline & FlaxStableDiffusionPipeline

* todo comment

* Fix imports

* Fix imports

* add dummies

* Fix empty init

* make pipeline work

* up

* Use Flax schedulers (typing, docstring)

* Wrap model imports inside availability checks.

* more updates

* make sure flax is not broken

* make style

* more fixes

* up
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
Co-authored-by: Pedro Cuenca <pedro@latenitesoft.com>

* rename weights to align with PT

* DiagonalGaussianDistribution => FlaxDiagonalGaussianDistribution

* fix name

* first commit:

- add `from_pt` argument in `from_pretrained` function
- add `modeling_flax_pytorch_utils.py` file

* small nit

- fix a small nit - to not enter in the second if condition

* major changes

- modify FlaxUnet modules
- first conversion script
- more keys to be matched

* keys match

- now all keys match
- change module names for correct matching
- upsample module name changed

* working v1

- test pass with atol and rtol= `4e-02`

* replace unsued arg

* make quality

* add small docstring

* add more comments

- add TODO for embedding layers

* small change

- use `jnp.expand_dims` for converting `timesteps` in case it is a 0-dimensional array

* add more conditions on conversion

- add better test to check for keys conversion

* make shapes consistent

- output `img_w x img_h x n_channels` from the VAE

* Revert "make shapes consistent"

This reverts commit 4cad1aeb4aeb224402dad13c018a5d42e96267f6.

* fix unet shape

- channels first!

* Fix CrossAttention._sliced_attention
Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>

* [Flax] Add Vae

* correct

* Apply suggestions from code review
Co-authored-by: Suraj Patil <surajp815@gmail.com>

* Finish
Co-authored-by: Suraj Patil <surajp815@gmail.com>

* Fix _upsample_2d
Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>