Add LGATr model#22

Open

spinjo wants to merge 29 commits intohqucms:dev/custom_train_evalfrom

Commits on Feb 13, 2025

Commits on Feb 14, 2025

Commits on Feb 16, 2025

Move spurions to device (oopsie)
spinjo
committed
Enforce autocast to float32 for equi_layer_norm (required for mixed-precision training)
spinjo
committed
Add support for full Lorentz group (default is fully connected subgroup)
spinjo
committed

Commits on Feb 17, 2025

Commits on Feb 18, 2025

Prepend spurions (instead of appending them) -> easier to keep track
spinjo
committed

Commits on Mar 3, 2025

Commits on Mar 5, 2025

Add stabilizing LayerNorm
spinjo
committed

Commits on Mar 15, 2025

Commits on Mar 17, 2025

Add 'use_flex_attention' option (uses torch_geometric)
spinjo
committed

Commits on Mar 18, 2025

Minor change
spinjo
committed

Commits on Mar 19, 2025

Commits on Jun 1, 2025