Set up the entire training using docker #44
Description
From discord discussion here:
"Ubdip: I was also considering to set up the entire training using docker (bookgen, datagen, trainer, variantfishtest, engine), the only extra part there would be to set up the nvidia container toolkit to forward the GPU access to the container. But haven't done anything in that direction yet, was just a consideration, since it would make things easier to automate cross platform."
Also relevant discussion here:
"Ubdip: I actually also created a private repo to test all 3 (copilot, codex, Claude) on the NNUE trainer in docker idea. Codex was close, especially because I selected 3x, but Claude was slightly better with just a single attempt."
//
Documenting it here to keep track, I think setting up the training stack + libraries is basically the trickiest part of training a NNUE network, especially for non-programmers, so this would be a great unlocker.