Prepared-tensor cache for behavior-cloning training in the
pokebot RL workstream: the full hashed
cache directory (manifest.json + games.jsonl + tensor shards) produced
from the v3 human-replay corpus (rated replays, >=1300 Elo), archived as
a single tar piped through zstd.
archive: human-full-v3-prep-cache.tar.zst (tar | zstd -10)
raw size: 187,257,951,668 bytes (174.4 GiB)
compressed: 1,646,749,271 bytes (1.53 GiB)
sha256 (archive):… See the full description on the dataset page:
https://huggingface.co/datasets/Looyyd/pokebot-human-prepcache-full-v3.
