Documentation
¶
Overview ¶
Package half converts the 16-bit floating-point storage encodings used by model files — IEEE 754 half-precision (binary16) and bfloat16 — to float32. Born has no native half-precision dtype, so format readers widen on load.
Index ¶
Constants ¶
This section is empty.
Variables ¶
This section is empty.
Functions ¶
func BFloat16ToFloat32 ¶ added in v0.9.19
BFloat16ToFloat32 converts a bfloat16 value to float32. bfloat16 is the top 16 bits of the IEEE 754 float32 representation, so widening is a left shift.
func Float16ToFloat32 ¶
Float16ToFloat32 converts an IEEE 754 half-precision (binary16) value to float32, handling subnormals, infinities, and NaNs.
Types ¶
This section is empty.
Source Files
Click to show internal directories.
Click to hide internal directories.