Closed swords123 closed 3 years ago
Hi @swords123
I am sorry for the late reply, I missed this issue.
Between v2 and v3, I redesigned the training framework. It's much more modular (handles different kind of architectures, multi-gpu, different templates...)
Where do you see that it improves performance? It should not change much.
Thanks, Thibault
I see that V3.0 code has been greatly improved in performance. What changes have you made?