For the complete documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /get-started.md).

Python function

bfloat16_as_uint16_to_float32

`bfloat16_as_uint16_to_float32()`

max.driver.bfloat16_as_uint16_to_float32(arr)

source

Converts bfloat16 bit patterns (stored as uint16) to float32.

The conversion is exact: a bfloat16 value is the top 16 bits of a float32.

For example, read a bfloat16 buffer back as float32:

import numpy as np
from max.driver import (
    Buffer,
    bfloat16_as_uint16_to_float32,
    float32_to_bfloat16_as_uint16,
)
from max.dtype import DType

bits = float32_to_bfloat16_as_uint16(np.array([1.5], np.float32))
buffer = Buffer.from_numpy(bits).view(DType.bfloat16)
values = bfloat16_as_uint16_to_float32(
    buffer.view(DType.uint16).to_numpy()
)

Parameters:: arr (ndarray[tuple[Any, ...], dtype[uint16]]) – A uint16 array holding bfloat16 bit patterns.
Returns:: A float32 array with the same shape as arr.
Return type:: ndarray[tuple[Any, …], dtype[float32]]

bfloat16_as_uint16_to_float32()​

`bfloat16_as_uint16_to_float32()`