v0.9.3 Now with offline local GPU model training support, approx. 20x faster than CPU training.

Finally, I found a PyTorch and CUDA version, that did work on my Nvidia RTX 3080 card with 10GB RAM. Win 10.
This accelerates the model training times drastically, even compared to CPU rendering on an AMD Ryzen 5950X.

DO NOT INSTALL THIS VERSION IF YOU HAVE A SLOW GRAPHICS CARD OR LOW VIDEO RAM from Nvidia.
You will almost certainly run into problems while training the models.

A graphics card with above or similar specs to mine (Nvidia), or at least enough RAM, might run fine.

Stick with the previous release, if in doubt.

Installation:
Unpack the loose files in the source code archive below over the installed files of the binary release v0.9.0 (portable install) or the binary release v0.9.0 (portable install) + v0.9.2.
(If there is no overwrite warning to confirm, you did something wrong...)
Next release will contain a full installer again.

Start the .exe file of the portable release and re-/install the Build System from the GUI.
Have fun.