| """ |
| 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 |
