> I am comfortable with autograd/computation graphs, PyTorch, "classic" neural nets and ones used for vision-type applications, as well as the basics of Transfo

> I am comfortable with autograd/computation graphs, PyTorch, "classic" neural nets and ones used for vision-type applications, as well as the basics of Transformer networks (I've trained a few smaller ones myself) and RNNs. It sounds like you're into it already. And you already know which new papers are interesting to you.