datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
lichess-elite
Lichess Elite Database
A curated collection of high-level chess games from Lichess.org, extracted from the Lichess Elite Database by nikonoel.
Dataset Description
This dataset contains 26.3 million games played by strong players on Lichess, filtered to include only:
White player: rated 2400+ (2500+ from Dec 2021)
Black player: rated 2200+ (2300+ from Dec 2021)
Time controls: Rapid and Blitz (bullet excluded)
Variant: Standard chess only
Source
Original data… See the full description on the dataset page: https://huggingface.co/datasets/nuriyev/lichess-elite.lichess-chess-1000-1800
Lichess Chess Games (1000-1800 ELO) — Pre-tokenized
Pre-tokenized chess game data from Lichess, filtered to 1000-1800 ELO rated games. Ready for training self-supervised game models.
Format
Files are in .somabin binary format — a self-describing, indexed, memory-mappable format created by ailed-soulsteal.
Each game contains:
token_ids (u16) — tokenized UCI moves with BOS/EOS markers
turn_ids (u8) — 0=white, 1=black per token
category_ids (u8) — piece type per move… See the full description on the dataset page: https://huggingface.co/datasets/Ailed-AI/lichess-chess-1000-1800.