HydrogenC / diffuser_training_samples

A couple of very simple samples about training models with the library diffusers.

Diffuser train samples

A series of samples demonstrating how to train simple samples with the lib diffusers.
May be a lot easier to read for beginners since the complicated conditioning stuff (e.g. CLIP) is removed. Can be a beginner's guide for the library diffusers.

• vae.py is a simple (unconditional) variational autoencoder
• vqvae.py is a simple (unconditional) vector-quantized VAE
• pixel_diffusion.py is a (unconditional) diffusion model that can generates images out of noise directly
• latent_diffusion.py is a (unconditional) latent diffusion model that generates latent vector, can work with either vae.py or vqvae.py to produce a final image