torch.future — PyTorch 2.7 documentation (original) (raw)

torch.__future__.set_overwrite_module_params_on_conversion(value)[source][source]

Sets whether to assign new tensors to the parameters instead of changing the existing parameters in-place when converting an nn.Module.

When enabled, the following methods will assign new parameters to the module:

  1. module.{device}() (e.g. nn.Module.cuda()) for moving a module between devices
  2. module.{dtype}() (e.g. nn.Module.float()) for converting a module to a different dtype
  3. nn.Module.to()
  4. nn.Module.to_empty()

Parameters

value (bool) – Whether to assign new tensors or not.

torch.__future__.get_overwrite_module_params_on_conversion()[source][source]

Returns whether to assign new tensors to the parameters instead of changing the existing parameters in-place when converting an torch.nn.Module. Defaults to False.

See set_overwrite_module_params_on_conversion() for more information.

Return type

bool

torch.__future__.set_swap_module_params_on_conversion(value)[source][source]

Sets whether to use swap_tensors() instead of setting .data to change the existing parameters in-place when converting an nn.Module and instead of param.copy_(state_dict[key]) when loading a state dict into an nn.Module.

Note

This function takes precedence over get_overwrite_module_params_on_conversion()

When enabled, the following methods will swap the existing parameters in-place:

  1. module.{device}() (e.g. nn.Module.cuda()) for moving a module between devices
  2. module.{dtype}() (e.g. nn.Module.float()) for converting a module to a different dtype
  3. nn.Module.to()
  4. nn.Module.to_empty()
  5. nn.Module.load_state_dict()

The semantics for load_state_dict() when this is set are as follows:

  1. For each parameter/buffer, its corresponding state_dict['key'] is transformed viamodule_load() (i.e. res = param.module_load(state_dict['key']))
  2. If necessary, res will be wrapped in an Parameter
  3. The parameter/buffer in the module will be swapped via swap_tensors()with res

Parameters

value (bool) – Whether to use swap_tensors() or not.

torch.__future__.get_swap_module_params_on_conversion()[source][source]

Returns whether to use swap_tensors() instead of setting .data to change the existing parameters in-place when converting an nn.Module. Defaults to False.

See set_swap_module_params_on_conversion() for more information.

Return type

bool