Commits

Commits on Dec 12, 2025

Commits on Dec 10, 2025

AdaMuon impl w/ a few other ideas based on recent reading
rwightman
committed

Commits on Dec 8, 2025

Add weights_only=False to avg_checkpoints, needs to load training state dict
rwightman
committed
Add dwee model def, and weights for dwee and dpwee
rwightman
committed
Trying out differential + parallel block, and fused out project option for ParallelScalingBlock to see if it's worthwhile now...
rwightman
committed

Commits on Dec 5, 2025

ParallelScalingBlock needs extra args even if not used
rwightman
committed
Implement diff attention in own layer file. Add support to vision_transformer.py and naflexivt.py ViTs
rwightman
committed

Commits on Dec 3, 2025

A simplification / fix for DropBlock2d. Handles even kernel sizes properly thanks to https://github.com/crutcher. For simplicty don't bother keeping drop blocks within feature maps. Remove 'batchwi…
rwightman
committed

Commits on Dec 2, 2025

Small improvement to map dataset retry logic
rwightman
committed

Commits on Nov 26, 2025

Add coord attn and some variants that I had lying around from old experiments.
rwightman
committed

Commits on Nov 6, 2025

Ensure final epoch always gets validated even if it doesn't line up with val interval. Add a few comments.
rwightman
committed
initial commit:

Tony
authored and
rwightman
committed

Commits on Nov 4, 2025

Release 1.0.22 for LayerScale regression
rwightman
committed
Refactor layer_scale to use init_values for gamma

Ilya-Fradlin
authored and
rwightman
committed

Commits on Nov 1, 2025

Update version.py to dev
rwightman
authored

Commits on Oct 31, 2025

Update README.md
rwightman
authored

Commits on Oct 30, 2025

Commits on Oct 24, 2025

Release 1.0.21
rwightman
committed

Commits on Oct 20, 2025

Rename 'simple' flag for Muon to 'fallback', add support for inverted 'use_muon' to be compat with other Muon impl. Add fallback_list arg to full optim factory call chain.
rwightman
committed

Commits on Oct 17, 2025