diffusers

diffusers copied to clipboard

[Dreambooth] Multi-GPU training with accelerate is magnitudes slower than single GPU (non-flax)

1

### Describe the bug I have access to a machine with several NVIDIA A100s. Initially I was using multiple GPUs for Dreambooth training under the idea that it would speed...

jkcarney

[happy to fix this] Wrong a weighted mean squared error loss

1

https://github.com/huggingface/diffusers/blob/acd317810bc138b3a78fa30e1b3da1006c1d60ad/examples/unconditional_image_generation/train_unconditional.py#L421-L428 1. Big issue: `alpha_t / (1 - alpha_t)` is the **signal-to-noise ratio (SNR)**. The weighting scheme `snr_weights` is different from the paper and doesn't make sense because the weight...

aengusng8

[Community Pipeline] Superresolution with Transformers

1

:hugs: Transformers now also includes an image super resolution/restoration model: Swin2SR (based on Swinv2) :fire: models on the hub: https://huggingface.co/caidas docs: https://huggingface.co/docs/transformers/main/en/model_doc/swin2sr We could combine this very nicely with `diffusers`...

patrickvonplaten

The wheels will have PyTorch pinned to the latest stable version: https://github.com/facebookresearch/xformers/issues/533#issuecomment-1355162407 For older versions, we can recommend to install from source.

pcuenca

storing original images from dataset

4

Script to store original images from unconditional training datasets as a collection of images. This script can be used before computing FID scores with tools like torch-fidelity.

lukovnikov

Question about noise schedule parametrization & samplers

2

Thanks for all the work first of all, I appreciate that we have all these pretrained models freely available open-source. I'm a bit confused by the relationship between sampler implementations...

jaywhang

Best practices / example results and run settings for Dreambooth inpainting

3

Hi @thedarkzeno @patil-suraj I'm trying to run the script here, https://github.com/huggingface/diffusers/tree/main/examples/research_projects/dreambooth_inpaint (ty for this) with similar setup as with training regular text=>image dreambooth. I'm finding that the quality of the...

aaronlohwy

Examples contain hardcoded model names

5

### Describe the bug Commit [c18941b](https://github.com/huggingface/diffusers/commit/c18941b01ad0ea6b07d020f353d81153c632a374) introduced hardcoded model names in the example scripts. Most of them should probably be replaced by `args.pretrained_model_name_or_path`. ### Reproduction Run one of the scripts...

allo-

NCSNPP checkpoint conversion script

2

I tried to convert the cifar10 ncsnpp original checkpoint [here](https://drive.google.com/drive/folders/1b0gy_LLgO_DaQBgoWXwlVnL_rcAUgREh) to the diffusers model using `convert_ncsnpp_original_checkpoint_to_diffusers.py` but found an error: `Traceback (most recent call last): File "convert_ncsnpp_original_checkpoint_to_diffusers.py", line 171, in...

karimul

Small performance improvements to some schedulers

4

Small performance improvements to `dpmsolver_multistep` and `euler_ancestral_discrete` by avoiding redundant calculations. Vectorized the loop in `lms_discrete`. Some example benchmarks with code below: **DPMSolverMultistepScheduler** > 100 steps: > before: 3.6375 sec...

timlueg