Training setup claimed to have more than 8,000 worlds
A user says renderer optimizations let them keep a “near indefinite” number of worlds ready to swap in for each training episode.
TLDR
The user says they have more than 8,000 worlds and can keep a “near indefinite” number ready to swap in per training episode thanks to renderer optimizations. They say 8,000 was the batch size that worked best in their experiments.
Combined views
2.2K
1 Source, first seen 5h ago
29 likes