add warning in Conv1DCell and synthesis.py for wavenet and deepvoice 3(auto-regressive models)

examples/deepvoice3/synthesis.py

@@ -101,6 +101,8 @@ if __name__ == "__main__":
         state, _ = dg.load_dygraph(args.checkpoint)
         dv3.set_dict(state)
 
+        # WARNING: don't forget to remove weight norm to re-compute each wrapped layer's weight
+        # removing weight norm also speeds up computation
         for layer in dv3.sublayers():
             if isinstance(layer, WeightNormWrapper):
                 layer.remove_weight_norm()

examples/wavenet/synthesis.py

@@ -115,6 +115,8 @@ if __name__ == "__main__":
         print("Loading from {}.pdparams".format(args.checkpoint))
         model.set_dict(model_dict)
 
+        # WARNING: don't forget to remove weight norm to re-compute each wrapped layer's weight
+        # removing weight norm also speeds up computation
         for layer in model.sublayers():
             if isinstance(layer, WeightNormWrapper):
                 layer.remove_weight_norm()

parakeet/modules/customized.py

@@ -225,6 +225,12 @@ class Conv1DCell(Conv1D):
 
     def start_sequence(self):
         """Prepare the Conv1DCell to generate a new sequence, this method should be called before calling add_input multiple times.
+
+        WARNING: 
+            This method accesses `self.weight` directly. If a `Conv1DCell` object is wrapped in a `WeightNormWrapper`, make sure this method is called only after the `WeightNormWrapper`'s hook is called. 
+            `WeightNormWrapper` removes the wrapped layer's `weight`, add has a `weight_v` and `weight_g` to re-compute the wrapped layer's weight as $weight = weight_g * weight_v / ||weight_v||$. (Recomputing the `weight` is a hook before calling the wrapped layer's `forward` method.)
+            Whenever a `WeightNormWrapper`'s `forward` method is called, the wrapped layer's weight is updated. But when loading from a checkpoint, `weight_v` and `weight_g` are updated but the wrapped layer's weight is not, since it is no longer a `Parameter`. You should manually call `remove_weight_norm` or `hook` to re-compute the wrapped layer's weight before calling this method if you don't call `forward` first.
+            So when loading a model which uses `Conv1DCell` objects wrapped in `WeightNormWrapper`s, remember to call `remove_weight_norm` for all `WeightNormWrapper`s before synthesizing. Also, removing weight norm speeds up computation.
         """
         if not self.causal:
             raise ValueError(