avewright/chess-soft-100m-disagreements
avewright/chess-soft-100m-disagreements Positions where the greedy policy of avewright/chess-transformer-100m-squares64 disagrees with a strong teacher best move (move_idx). Current upload: 1,782,505 rows in 90 shards. Mix split rows shards labels data/shard_*.parquet 1,768,622 83 teacher MultiPV from avewright/chess-soft-multipv-lichess data/sf19/*.parquet 13,883 7 Stockfish 19 max-Elo MultiPV from new games Stream rows are an argmax filter of… See the full description on the dataset page: https://huggingface.co/datasets/avewright/chess-soft-100m-disagreements.
avewright/chess-soft-100m-disagreements
Positions where the greedy policy of `avewright/chess-transformer-100m-squares64` disagrees with a strong teacher best move (move_idx).
Current upload: 1,782,505 rows in 90 shards.
Mix
Stream rows are an argmax filter of existing boards. SF19 rows are new games (100M vs unlimited SF19; hash-excluded against the stream). Soft targets on the SF19 split are live SF19 MultiPV, not the original teacher pack.
Columns
Same schema as avewright/chess-soft-multipv-lichess:
board_array[64],turn,castling,ep_squaremove_idx,cp,mate— teacher hard best + evalsoft_indices[8],soft_probs[8]— teacher soft policylabel_depth,phase,source(3 = harvest),cache_name
License
MIT. Stream labels inherit from the source MultiPV pack. SF19 labels are Stockfish 19 analysis from this harvest.
