training / scripts /fetch_raw.sh
cazyundee's picture
tinychess: self-play research substrate (phase 1-4)
3495881 verified
Raw History Blame Contribute Delete
785 Bytes
#!/usr/bin/env bash
# Fetch raw research data. Idempotent.
set -euo pipefail
RAW=${1:-/home/user/tinychess/data/raw}
mkdir -p "$RAW"
PGN_URL="https://database.lichess.org/standard/lichess_db_standard_rated_2014-07.pgn.zst"
if [ ! -f "$RAW/games.pgn.zst" ]; then
echo "[fetch] lichess games ..."
curl -sL -o "$RAW/games.pgn.zst.part" "$PGN_URL"
mv "$RAW/games.pgn.zst.part" "$RAW/games.pgn.zst"
fi
echo "[fetch] games: $(du -h "$RAW/games.pgn.zst" | cut -f1)"
for i in 0 1 2; do
f="$RAW/puzzles-$i.parquet"
if [ ! -f "$f" ]; then
echo "[fetch] puzzles shard $i ..."
curl -sL -o "$f.part" \
"https://huggingface.co/datasets/Lichess/chess-puzzles/resolve/main/data/train-0000$i-of-00003.parquet"
mv "$f.part" "$f"
fi
done
echo "[fetch] done"
ls -lh "$RAW"