Python function

float32_to_bfloat16_as_uint16

`float32_to_bfloat16_as_uint16()`

max.pipelines.lib.float32_to_bfloat16_as_uint16(arr)

source

Converts a float32 array to bfloat16 representation stored as uint16.

BFloat16 is the upper 16 bits of float32 with proper rounding. This allows us to halve memory usage while maintaining the exponent range.

Parameters:: arr (ndarray[tuple[Any, ...], dtype[float32]]) – Float32 numpy array
Returns:: Uint16 array containing bfloat16 bit representation with same shape
Return type:: ndarray[tuple[Any, …], dtype[uint16]]