Pius Friesch
Pius Friesch
Thanks for the answer! I am not working on the topic anymore, but at the time, I think I was also interested on the anecdotal performance you might have experienced....
that means that the phone number is not yet registered, try to signing up fist
In your alternative model, it seems like you don't use the dual softmax anymore. Did you experiment with it?
Minimal project that fails on osx-x64, but works on win-x64. [corerttest.tar.gz](https://github.com/dotnet/corert/files/4371543/corerttest.tar.gz)
I tried it again, but it does not compile to a binary. The same happens on the azure mac build environment. Local build environment: macOS: `10.15.4` ``` % clang --version...
If I am not mistaken, the linked example is a regularization scheme for the model parameters?! That is not the EMA to learn just the embeddings? Do you have an...
Well, the forum post you linked is not the EMA that is described in the VQVAE paper. Did you modify the implementation to follow the VQVAE paper?
I mean it is called both exponential moving average (EMA). The forum post which you posted and what they do in the VQVAE paper (or in the sonnet code) are...
Cool 👍 Looking forward to the samples 😄
Yes, sounds correct. You can look at [this notebook](https://github.com/zalandoresearch/pytorch-vq-vae/blob/master/vq-vae.ipynb) for a working example.