tdrussell/diffusion-pipe
A pipeline parallel training implementation for diffusion models including SDXL, Flux, and many other image/video generative models.
Not currently ranked — collecting fresh signals.
star history
A distributed training pipeline for diffusion-based generative models with pipeline parallelism support. It handles large-scale multi-GPU training with efficient latent and text embedding caching, checkpointing, and evaluation metrics. The framework is modular, allowing easy addition of new model architectures through subclassing, and currently supports a wide range of image and video generation models.
Frequently asked
- What is tdrussell/diffusion-pipe?
- A pipeline parallel training implementation for diffusion models including SDXL, Flux, and many other image/video generative models.
- Is diffusion-pipe open source?
- Yes — tdrussell/diffusion-pipe is open source, released under the GPL-3.0 license.
- What language is diffusion-pipe written in?
- tdrussell/diffusion-pipe is primarily written in Python.
- How popular is diffusion-pipe?
- tdrussell/diffusion-pipe has 2k stars on GitHub.
- Where can I find diffusion-pipe?
- tdrussell/diffusion-pipe is on GitHub at https://github.com/tdrussell/diffusion-pipe.