get_autocast_cpu_dtype

paddle. get_autocast_cpu_dtype ( device_type: PlaceLike | None = None ) _DTypeLiteral [source]

Get the auto-mixed-precision dtype in the current context if autocast is enabled else default AMP dtype(float16).

Parameters

device_type (PlaceLike, optional) – The device type to check. This argument is ignored for all devices sharing the same AMP state in paddlepaddle.

Returns

The current AMP dtype.

Return type

_DTypeLiteral

Examples

>>> 
>>> # Demo1: Get default auto-mixed-precision dtype
>>> import paddle
>>> paddle.device.set_device('gpu')
>>> print(paddle.get_autocast_dtype())
float16

>>> # Demo2: Enable auto-mixed-precision and get the dtype
>>> with paddle.amp.auto_cast():
...     print(paddle.get_autocast_dtype())
float16