peras-v4

Network for Peras v3.4.0 and later.

Architecture (768x16hm + threats + pawn pairs -> 1024)x2 -> (32 -> 32 -> 1) x 8. Each side's accumulator has its two halves clipped and multiplied together before anything else sees them, and the single output layer is replaced by three layers per material bucket, the second of which is fed both its inputs and their squares. King buckets go from 10 to 16.

Piece-square weights i16 and threat/pawn-pair weights i8, both quantised by 255; the first affine layer is i8 quantised by 128, and the two layers after it i32.

Trained from scratch over 1200 superbatches on about 30 billion positions, a third more data than peras-v3 had: 100 superbatches ramping the learning rate up and back down, then 1100 annealing it from 1e-3 to 1e-6 while the training target shifts from the search score towards the game result.

Worth +6.4 +/- 10.1 Elo over peras-v3 at 10s+0.1s on one thread, and +31.4 +/- 30.8 at a fixed 40000 nodes.

Score of peras-v4 vs peras-v3: 460 - 423 - 1117  [0.509] 2000
Elo: +6.4 +/- 10.1  (LOS 89.3%), 0 time losses
10s+0.1s, one thread, 64 MB hash

peras-v3

Network for Peras v3.0.0 and later.

Architecture (768x10hm + threats + pawn pairs -> 1024)x2 -> 8, SCReLU. Piece-square weights i16 and threat/pawn-pair weights i8, both quantised by 255; output layer i16 quantised by 64.
Trained on six months of Leela Chess Zero test80 data over two cosine cycles, 620 superbatches totalling 62 billion positions.

Worth +110 Elo over peras-v2 at 10s+0.1s on one thread, and +143 Elo at a fixed 40k nodes.

sha256  a668c98ee350...  peras-v3.nnue
size    81676352 bytes

peras-v2

Network for Peras v2.0.0 through v2.1.0.

Architecture (768x10hm -> 1024)x2 -> 8, SCReLU, i16 weights quantised by 255 and 64.
Trained on two months of Leela Chess Zero test80 data, 800 superbatches.

sha256  62819b022c5b...  peras-v2.nnue
size    15763520 bytes