returnn.torch.frontend.bridge¶
Direct bridge between pure PyTorch code and modules and RF code and modules.
https://github.com/rwth-i6/returnn/issues/1287
- returnn.torch.frontend.bridge.pt_module_to_rf_module(pt_module: Module) Module[source]¶
- Parameters:
pt_module – torch module
- Returns:
RF module
- returnn.torch.frontend.bridge.wrapped_pt_module_to_rf_module(pt_module: Module) Module | None[source]¶
- Parameters:
pt_module – torch module
- Returns:
RF module if the torch module is a wrapped RF module, or None otherwise
- returnn.torch.frontend.bridge.rf_module_to_pt_module(rf_module: Module, *, aux_params_as_buffers: bool = True) Module[source]¶
- Parameters:
rf_module – RF module
aux_params_as_buffers – whether to map RF auxiliary parameters to PyTorch buffers, otherwise to normal parameters, i.e. they occur in model.named_parameters(). Note that even when they are part of model.named_parameters(), aux params usually don’t have a gradient, and then they are not updated by the optimizer. Historically, this was False. Now, this is True by default, as this is more reasonable. Note that the optimizer state dict will change if you change this, however, we will automatically convert such optimizer state dict.
- Returns:
torch module
- class returnn.torch.frontend.bridge.PTModuleAsRFModule(pt_module: Module)[source]¶
Wrapped module.
It is recommended to use
pt_module_to_rf_module()instead of using this directly.By convention, any options to the module are passed to __init__, and potential changing inputs (other tensors) are passed to
__call__().
- class returnn.torch.frontend.bridge.RFModuleAsPTModule(rf_module: Module, *, aux_params_as_buffers: bool = True)[source]¶
Wrapped module.
It is recommended to use
rf_module_to_pt_module()instead of using this directly.Initializes internal Module state, shared by both nn.Module and ScriptModule.