Package-level declarations

Brain Float 16 (BFloat16) type. 16-bit floating point format with same exponent range as FP32 but reduced mantissa. Commonly used in machine learning workloads.

64-bit floating point type (Double precision). Highest precision floating point type in SKaiNET.

16-bit signed integer type.

64-bit signed integer type (Long).

Custom data type holding 3 values -1,0,1 stored in 2 bits. Used e.g. with BitNet models https://huggingface.co/microsoft/bitnet-b1.58-2B-4T

Utility functions and extensions for the DType type system.

16-bit unsigned integer type.

32-bit unsigned integer type.

64-bit unsigned integer type.

8-bit unsigned integer type.

Returns the common precision type for mixed operations

Checks if conversion between two DTypes is supported at compile time

Checks if a precision chain is valid (e.g., for multi-stage networks).

Validates that two DTypes are compatible for operations.

Validates that a conversion path exists between two DTypes.

Validates an entire precision chain, throwing an exception if invalid.

Validates that a promotion between two DTypes will produce the expected result.