Hyunjik Kim

Apologies for the late reply. Hmm you shouldn't be seeing this error. I've just tried following the instructions from scratch in a new venv and the test runs fine for...

Hi there, apologies for the late reply! We'll be trying to open source the core code for spatial functa soon, but there are some steps for authorization that we need...

Hi, thanks! The TLDR is unfortunately we don't have plans to open source code for classification/generation, but here below is some explanation of why this is: So the code for...

When running the code on pytorch 0.4.1 (as listed in the Dependencies section of README), the above NaN behaviour did not happen and learning seems stable (although I haven't checked...

**A side note for compatibility of code with pytorch 1.0 and 1.1**: it seems like using in-place operations for tensors that lead to the training loss `obj_per_accum` can give rise...