ut3_masking

The plain-ut3 mask layer: build, validate, apply.

ut3_make_masks and ut3_apply_masks. (The jit tracer guard require_concrete_masks is infrastructure and lives in common, beside the ValueHashedMasks mixin – the two halves of the uniform mask-representation contract, shared by every uniform object’s masks.) Masks are boolean prefix vectors, static structure, and ALWAYS host numpy (np, never xnp) – do not “fix” that (docs/uniform_masks_vs_ranks.md, docs/contributor/uniform_rank_masks_rationale.md).

Functions

ut3_make_masks(tucker_ranks, tt_ranks, n, r)	Build the prefix RANK edge masks for a uniform Tucker tensor train.
ut3_apply_masks(x)	Zero the padded ("garbage") regions of the supercores by multiplying through the edge masks.