torchtyping

Allows specifying tensor dimensions with named placeholders (e.g., 'batch'), integers, and ellipses for batch dimensions, as demonstrated in the batch_outer_product example, making shape contracts explicit.

Seamlessly integrates with typeguard to enforce annotations at runtime, catching shape mismatches early, like inconsistent batch sizes, without invasive asserts.

Supports checking dtype, layout, named dimensions, and custom extensions via the details flag, offering fine-grained control over tensor properties in a single annotation.

Replaces verbose inline comments with structured type hints, as shown in the README example, turning '# x has shape (batch, hidden_state)' into machine-checkable annotations.