dynamic_learning_rate, extra checks

rwth-i6 · Nov 3, 2023 · 4e76da6 · 4e76da6
1 parent 5c82ab3
commit 4e76da6
Show file tree

Hide file tree

Showing 2 changed files with 4 additions and 0 deletions.
diff --git a/returnn/tf/updater.py b/returnn/tf/updater.py
@@ -222,6 +222,8 @@ def get_current_step_learning_rate(self):
             assert any(
                 [arg.kind == inspect.Parameter.VAR_KEYWORD for arg in signature.parameters.values()]
             ), "please specify **kwargs in dynamic_learning_rate for future compatibility"
+            if "epoch" in signature.parameters:
+                raise NotImplementedError("TF updater: dynamic_learning_rate with epoch not supported currently")
             lr = learning_rate_function(
                 network=self.network, global_train_step=self.global_train_step, learning_rate=lr
             )

diff --git a/returnn/torch/updater.py b/returnn/torch/updater.py
@@ -105,6 +105,8 @@ def __init__(self, *, config, network, device, initial_learning_rate=1.0):
                 assert any(
                     [arg.kind == inspect.Parameter.VAR_KEYWORD for arg in signature.parameters.values()]
                 ), "please specify **kwargs in dynamic_learning_rate for future compatibility"
+                if "network" in signature.parameters:
+                    raise ValueError("Torch updater: dynamic_learning_rate network is TF specific")
             else:
                 raise NotImplementedError("not implemented for not callable dynamic_learning_rate")