Hi,
I’m trying to run a Text2Speech model found here : https://github.com/mozilla/TTS.
I cloned the latest version and downloaded their latest trained model, but when trying to load the model I get different shapes for the weights.
cp = torch.load(MODEL_PATH)
model.load_state_dict(cp['model'])
For example:
Layer name : embedding.weight
Loaded model size : torch.Size([61, 256])
Model size : torch.Size([256, 256])
The vocabulary size should be 256 since we are working with ASCII characters, but in the loaded model is 61. I was thinking that maybe something can interfere with the load of the model and it lost some data. There are a few more problems like this one. Thank you!