michaelwaves/diffusion-practice-training-artifacts

Diffusion practice training artifacts

Model and optimizer checkpoints and generated sample grids from Michael Yu's diffusion / flow-matching practice experiments, recovered from the MATS cluster on 2026-11-10 UTC. The associated code is at https://github.com/michaelwaves/diffusion_practice . The experiments use MNIST and synthetic teaching examples; no personal conversation data or credentials are included.

Original run folder names and filenames are preserved under runs/ and mine/runs/. MANIFEST.json records the SHA-256 and size of each recovered file. Two identical artifacts already present in the GitHub repository were omitted. The existing michaelwaves Hugging Face repositories were checked for matching file hashes before this upload.

These are experimental training artifacts, not a production model release. PyTorch .pt files contain training state and require the corresponding architecture and code to interpret. No raw MNIST download is included.