diffusers

diffusers copied to clipboard

Fix IP Adapter Support for SAG Pipeline

2

# What does this PR do? Addresses part of #7064 by updating the IP adapter support for the SAG pipeline ## Before submitting - [ ] This PR fixes a...

Stepheni12

Change step_offset scheduler docstrings

9

See https://github.com/huggingface/diffusers/discussions/6931 and https://github.com/huggingface/diffusers/issues/6068 for background Right now it changes all the docstrings from > An offset added to the inference steps. You can use a combination of `offset=1` and...

Beinsezii

train_controlnet.py broken local data loading

1

### Describe the bug The script `examples/controlnet/train_controlnet.py` is designed to either take a HF dataset with `--dataset_name` (as illustrated [here](https://huggingface.co/blog/train-your-controlnet)) or local data `--train_data_dir`, but the latter case fails out-of-the-box....

morrisalp

Fixing implementation of ControlNet-XS

4

# What does this PR do? Implements ControlNet-XS with the feedback from [PR #5827](https://github.com/huggingface/diffusers/pull/5827) incorporated. See original PR for details. ## Who can review? Core library: - Pipelines: @patrickvonplaten and...

UmerHA

'StableDiffusionXLPipeline' object has no attribute 'named_parameters' when using accelerate.load_checkpoint_and_dispatch

4

### Describe the bug How can I make the stable diffusion xl model to work with the `init_empty_weights` API? I've followed https://huggingface.co/docs/accelerate/v0.11.0/en/big_modeling but still got an unexpected error: ### Reproduction...

thiagocrepaldi

[Refactor] How attention is set in 3D UNet blocks

7

# What does this PR do? There's some confusion around how `num_attention_heads` is being used in the 3D UNet + Block. This PR attempts to fix the issue by correcting...

DN6

cads support

5

https://arxiv.org/abs/2310.17347

loboere

Applying Blockwise Weights to LoRA

2

It would be wonderful if I could apply stepwise or blockwise weights to LoRA, just like the webui extension below. https://github.com/hako-mikan/sd-webui-lora-block-weight Does anyone know if it's also possible in diffusers??

dain5832

ValueError: Query/Key/Value should either all have the same dtype training LORA

11

### Describe the bug LORA training don't work with mixed precision enabled: File "/home/imgen/miniconda3/envs/py32/lib/python3.11/site-packages/xformers/ops/fmha/__init__.py", line 348, in _memory_efficient_at tention_forward_requires_grad inp.validate_inputs() File "/home/imgen/miniconda3/envs/py32/lib/python3.11/site-packages/xformers/ops/fmha/common.py", line 121, in validate_inputs raise ValueError( ValueError: Query/Key/Value...

noskill

Allow more arguments to be passed to convert_from_ckpt

2

# What does this PR do? Allows management of safety checker and feature extractor from outside of the convert ckpt class. Without this one must rely on the function to...

w4ffl35