crowsonkb/k-diffusion
A PyTorch implementation of diffusion-based generative models from the Karras et al. (2022) paper with transformer-based enhancements.
Not currently ranked — collecting fresh signals.
star history
This library implements the Elucidating the Design Space of Diffusion-Based Generative Models paper by Karras et al. (2022) for PyTorch. It includes enhancements such as improved sampling algorithms and transformer-based diffusion models like image_transformer_v2, which combines ideas from Hourglass Transformer and DiT. The project supports training on image datasets like Oxford Flowers and requires custom CUDA kernels (NATTEN, FlashAttention-2) for optimal performance.
Frequently asked
- What is crowsonkb/k-diffusion?
- A PyTorch implementation of diffusion-based generative models from the Karras et al. (2022) paper with transformer-based enhancements.
- Is k-diffusion open source?
- Yes — crowsonkb/k-diffusion is open source, released under the MIT license.
- What language is k-diffusion written in?
- crowsonkb/k-diffusion is primarily written in Python.
- How popular is k-diffusion?
- crowsonkb/k-diffusion has 2.6k stars on GitHub.
- Where can I find k-diffusion?
- crowsonkb/k-diffusion is on GitHub at https://github.com/crowsonkb/k-diffusion.