diffusers

diffusers copied to clipboard

Fix object has no attribute 'flush' when using without a console

1

# What does this PR do? Fix a recurring problem when using diffusers without console, e.g. using `pythonw.exe` in windows, since there's no `sys.stderr`. I can make a more advanced...

asomoza

[`Cont'd`] Add the SDE variant of DPM-Solver and DPM-Solver++ to DPM Single Step

7

Thanks for the opportunity to continue #4251 @yiyixuxu. There was uncertainty about whether `sigma_s0` or `sigma_s1` should be used in the second-order update in `sde-dpmsolver++` when @cmdr2 was implementing. I...

tolgacangoz

[Core] refactor `embeddings`

5

# What does this PR do? Extension of https://github.com/huggingface/diffusers/pull/7995. Some comments are in line. ## Todos - [ ] Add licensing

sayakpaul

[LoRA] introduce LoraBaseMixin to promote reusability.

4

# What does this PR do? It is basically a mirror of https://github.com/huggingface/diffusers/pull/8670. I had accidentally merged it but I have reverted it in https://github.com/huggingface/diffusers/pull/8773. Apologies for this. I have...

sayakpaul

[Tests] add a test suite to check if loading from subfolder works as expected.

2

# What does this PR do? To prevent things like this from happening: https://github.com/huggingface/diffusers/pull/8416

sayakpaul

[Tests] use the `output_shape` property in the tests.

1

# What does this PR do? Refer to https://github.com/huggingface/diffusers/pull/7647/#discussion_r1621374787

sayakpaul

[Core] be more specific when doing accelerate imports.

6

# What does this PR do? By being more specific on the imports, we can cut down the overall import time in our library. Follow this thread: https://huggingface.slack.com/archives/C021H1P1HKR/p1719924173632779. We should...

sayakpaul

[Core] introduce `ContinuousTransformer2DModelBlock` to replace intermediate `Transformer2DModel`

1

# What does this PR do? This PR introduces a `ContinuousTransformer2DModelBlock` to replace the intermediate `Transformer2DModel` from model classes like `UNet2DConditionModel`. Currently, all our UNets that have cross-attention (`UNetMotionModel`, `UNet2DConditionModel`,...

sayakpaul

from diffusers.loaders.autoencoder import FromOriginalVAEMixin fails with ImportError

4

On 0.28.2, this from diffusers.loaders.autoencoder import FromOriginalVAEMixin will fail with this error ImportError: cannot import name 'create_diffusers_vae_model_from_ldm' from 'diffusers.loaders.single_file_utils' (/usr/local/lib/python3.10/dist-packages/diffusers/loaders/single_file_utils.py) If you look at the source code, `create_diffusers_vae_model_from_ldm` imported here:...

juancamilog

[Single File] Allow loading T5 encoder in mixed precision

1

# What does this PR do? When loading the T5 encoder in SD3 with `from_pretrained` certain layers are are kept in fp32. Currently this is not the case with single...

DN6