123456789101112131415161718192021 |
- """
- This global flag controls whether to assign new tensors to the parameters
- instead of changing the existing parameters in-place when converting an `nn.Module`
- using the following methods:
- 1. `module.cuda()` / `.cpu()` (for moving `module` between devices)
- 2. `module.float()` / `.double()` / `.half()` (for converting `module` to a different dtype)
- 3. `module.to()` / `.type()` (for changing `module`'s device or dtype)
- 4. `module._apply(fn)` (for generic functions applied to `module`)
- Default: False
- """
- _overwrite_module_params_on_conversion = False
- def set_overwrite_module_params_on_conversion(value):
- global _overwrite_module_params_on_conversion
- _overwrite_module_params_on_conversion = value
- def get_overwrite_module_params_on_conversion():
- return _overwrite_module_params_on_conversion
|