Skip to main content

dtype

The type of the dtype objects (clika_runtime.float32, clika_runtime.int8, ...). A dtype object is what dtype= takes and what Tensor.dtype returns; compare with ==, read itemsize / is_floating_point / is_signed.

bits (property)​

Bits per element, exact for the sub-byte types (int4 -> 4).

is_complex (property)​

False: no complex dtype exists.

is_floating_point (property)​

True for every floating format (float16 through the float8/6/4 families).

is_signed (property)​

True when the format carries a sign (signed integers and the signed floating formats).

itemsize (property)​

Bytes per element (1 for the sub-byte types; see bits).

name (property)​

The canonical name: "float32", "bfloat16", "int8", ...

__init__​

__init____init__(self, name: str) -> None

init(self, name: str) -> None

dtype(name) -> dtype

The dtype object named by a string: the canonical names ("float32", "bfloat16", "uint4", ...) and the torch aliases ("float", "half", "double", "short", "int", "long").