Rerun of depth=4 NNUE training after increasing batch size
from 512 to 8192. The small batch size was a huge problem
for supervized learning too, so hopefully this run will be a success.