TorchLean API

NN.Tensor.Internal.Check.Diagnostic

Shape-checking diagnostics #

Shape checking is separate from parsing, but uses the same source spans.

Stable categories for concrete shape-checking failures.

  • rankMismatch : DiagnosticCode

    The number of physical dimensions disagreed with the pattern rank.

  • invalidEllipsis : DiagnosticCode

    An ellipsis was placed or expanded in a way forbidden by the operation.

  • anonymousAxisInRearrange : DiagnosticCode

    A rearrangement used a numeric axis, which cannot identify input data.

  • axisMismatch : DiagnosticCode

    The input and output patterns referred to different logical axes.

  • duplicateAxisLength : DiagnosticCode

    The caller supplied more than one length for the same named axis.

  • unusedAxisLength : DiagnosticCode

    The caller supplied a named length that the pattern does not use.

  • cannotInferAxis : DiagnosticCode

    A composite input dimension did not determine one of its axis lengths.

  • dimensionMismatch : DiagnosticCode

    A known axis length disagreed with the corresponding tensor dimension.

  • missingAxisLength : DiagnosticCode

    An output-only axis had no length supplied by the caller.

  • einsumOperandCountMismatch : DiagnosticCode

    The number of einsum tensors disagreed with the number of input expressions.

  • unsupportedEinsumAxis : DiagnosticCode

    An einsum expression used grouping or an axis form outside its surface grammar.

  • unknownEinsumOutputAxis : DiagnosticCode

    An einsum output named an axis absent from every input.

  • duplicateEinsumOutputAxis : DiagnosticCode

    An einsum output named the same logical axis more than once.

  • tooManyEinsumAxes : DiagnosticCode

    The reference-compatible einsum front end exceeded its 52-label limit.

  • repeatedEinsumDimensionMismatch : DiagnosticCode

    Repeated occurrences of an einsum label in one operand had unequal lengths.

  • incompatibleEinsumBroadcast : DiagnosticCode

    Non-singleton occurrences of an einsum axis had incompatible lengths.

  • compositeParseShapeAxis : DiagnosticCode

    A parse_shape expression used a grouped physical axis.

  • duplicateParseShapeAxis : DiagnosticCode

    A parse_shape expression bound the same name more than once.

  • emptyPackInput : DiagnosticCode

    pack was asked to combine no tensors.

  • packRankMismatch : DiagnosticCode

    A tensor rank was too small for the fixed axes of a pack pattern.

  • packDimensionMismatch : DiagnosticCode

    Corresponding fixed axes of packed tensors had different lengths.

  • unpackRankMismatch : DiagnosticCode

    A packed tensor rank disagreed with the rank implied by its metadata.

  • invalidPackedShape : DiagnosticCode

    Unpack metadata did not describe a valid concrete packed shape.

  • multipleInferredDimensions : DiagnosticCode

    An unpack shape requested inference for more than one dimension.

  • cannotInferPackedDimension : DiagnosticCode

    The packed extent did not uniquely determine an inferred dimension.

  • packedAxisMismatch : DiagnosticCode

    Unpack metadata did not multiply to the packed tensor dimension.

  • internalInvariant : DiagnosticCode

    An implementation invariant failed after earlier checks had established it.

Instances For

    A structured concrete-checking error.

    • Stable machine-readable category of the checking failure.

    • message : String

      Human-readable explanation of the failed shape obligation.

    • Smallest useful range of the pattern responsible for the failure.

    Instances For
      @[reducible, inline]

      The result type used by concrete checkers.

      Instances For