shapix.cupy

shapix.cupy provides CuPy-native array aliases based on cupy.ndarray.

from shapix.cupy import (
  F32, I64, Int, Shaped,
  F32Like, I64Like,
  U8ScalarLike, make_scalar_like_type,
)

What it exports

Strict array aliases:

• concrete families such as Bool, I32, I64, F16, F32, F64, C64, C128
• category families such as Int, UInt, Integer, Float, Real, Complex, Inexact, Num, Shaped

Like aliases:

• BoolLike, I8Like through I64Like, U8Like through U64Like
• F16Like, F32Like, F64Like
• C64Like, C128Like
• category aliases such as IntLike, FloatLike, NumLike, ShapedLike

Other exports:

• NumPy-defined ScalarLike aliases re-exported for convenience
• make_scalar_like_type

Backend limits

shapix.cupy does not export:

• BF16
• F128
• C256
• V
• Str
• Bytes
• Obj
• DT64
• TD64

It also requires numpy alongside cupy at runtime.

Like behavior

CuPy Like aliases use cupy.asarray on the slow path, so they can accept:

• real CuPy arrays
• NumPy arrays
• Python scalars and nested sequences

Static type checkers still see the result as cupy.ndarray.

ScalarLike re-exports

ScalarLike aliases are re-exported from shapix.numpy. They validate Python and NumPy scalar values, not CuPy 0-D arrays.

For CuPy scalar arrays, prefer a Like alias with Scalar, for example F32Like[Scalar].