size mismatch

[goandrei]

Hi,

I'm trying to do an inference on the latest model, but I get this error:
size mismatch for embedding.weight: copying a param with shape torch.Size([61, 256]) from checkpoint, the shape in current model is torch.Size([256, 256]).

I read all the related posts but nothing helped. I am running the notebook on fix_db7f3d3. Changing the num_chars to 61 works, but the result is quite bad(just some random noise).
Also I can't run all examples using num_chars=61 due to the lack of embedding for some characters.
What can I do in order to make it work? Thank you!

[erogol]

I don't think you use the right commit version of the repository for that checkpoint you use.
git checkout fix_db7f3d3 before you run anything.

[goandrei]

Actually I do. I figured out that the problem with the embedding layer mismatch was solved only on the server application. I was using the notebook from notebooks directory which used other functions for tts