Skip to content

v1.0.42

Choose a tag to compare

@github-actions github-actions released this 24 Apr 16:25
· 101 commits to main since this release

NNUE Net Trained on 3x More Data

+215 ± 17 Elo vs v1.0.40 (2000 games, 1+0.1 TC)

Changes

  • NNUE net retrained on 635M Stockfish depth-9 positions (up from 204M in v1.0.40)
  • Same architecture: Chess768 → 256×2 → 1, SCReLU, i16 quantization
  • 600 superbatches with AdamW, cosine LR decay, WDL proportion 0.75

Match Results

v1.0.42 vs v1.0.40: 1418W - 319L - 263D (2000 games)
Score: 0.775 | Elo: +215 ± 17 | Draw ratio: 13.2%

Cumulative Improvement

  • v1.0.40 was +65 Elo vs HCE (v1.0.39)
  • v1.0.42 is +215 Elo vs v1.0.40
  • Estimated ~+280 Elo over hand-crafted evaluation