Data used for training recent (2022 - present) Stockfish default nets. Converted from Leela training data into the binpack data format for training NNUE with nnue-pytorch.
For converting Leela training data to unfiltered binpacks: lc0-data-converter
For filtering binpacks into subsets for Stockfish training: nnue-data
See the PR descriptions and stockfish commits for the dataset components used for the training run leading to the listed default net. Under each stockfish PR link is a list of dataset components unique to training that particular net.
As of November 2025, the datasets used in training the latest official networks (SFNNv10+ threat inputs) are documented in threats.yaml