Module tensor_diff 

TensorDiffEngine — structural and numeric diff between tensor snapshots.

Used for change-detection in federated learning checkpoints: given two sets of named tensors (old and new), compute per-tensor diffs covering additions, removals, shape changes, and numeric value changes.

Structs

DiffSummary

Aggregated statistics over a slice of TensorDiff values.

TensorDiff

Diff result for one named tensor.

TensorDiffEngine

Engine that computes structural and numeric diffs between tensor snapshots.

TensorSnapshot

A named, flat snapshot of a single tensor.

Enums

DiffKind

Describes what changed (or didn’t) for a single named tensor.