HalfPrecision

class lightning.pytorch.plugins.precision.HalfPrecision(precision='16-true')

Bases: Precision

Plugin for training with half precision.

Parameters:

precision (Literal['bf16-true', '16-true']) – Whether to use torch.float16 ('16-true') or torch.bfloat16 ('bf16-true').

convert_input(data)

Convert model inputs (forward) to the floating point precision type of this plugin.

This is a no-op in the base precision plugin, since we assume the data already has the desired type (default is torch.float32).

Return type:

Any

convert_module(module)

Convert the module parameters to the precision type this plugin handles.

This is optional and depends on the precision limitations during optimization.

Return type:

Module

forward_context()

A context manager to change the default tensor type when tensors get created during the module’s forward.

See: torch.set_default_tensor_type()

Return type:

Generator[None, None, None]

module_init_context()

Instantiate module parameters or tensors in the precision type this plugin handles.

This is optional and depends on the precision limitations during optimization.

Return type:

AbstractContextManager

tensor_init_context()

Controls how tensors get created (device, dtype).

Return type:

AbstractContextManager