Stockfish 2023123010 NNUE for Windows and Linux, interesting compilation by Linmiao Xu (Elo +2.50)

Since you’re here...

... we have a small favour to ask. More people, like you, are reading and supporting our blog: "Chess Engines Diary". And unlike many other sites and blogs, we made the choice to keep our articles open for all, regardless of where they live or what they can afford to pay.

We hope you will consider supporting us today. We need your support to continue to exist, because good entries are more and more work time. Every reader contribution, however big or small, is so valuable. Support "Chess Engines Diary" even a small amount– and it only takes a minute. Thank you.
============================== My email: jotes@go2.pl

Stockfish 2023123010 NNUE for Windows and Linux, interesting compilation by Linmiao Xu (Elo +2.50)

Stockfish 2023123010 - UCI chess engine, compiled by Linmiao Xu

Timestamp: 1703930883

Update default net to nn-b1e55edbea57.nnue

Created by retraining the master big net `nn-0000000000a0.nnue` on the same dataset with the ranger21 optimizer and more WDL skipping at training time.

More WDL skipping is meant to increase lambda accuracy and train on fewer misevaluated positions where position scores are unlikely to correlate with game outcomes. Inspired by:

- repeated reports in discord #events-discuss about SF misplaying due to wrong endgame

evals, possibly due to Leela's endgame weaknesses reflected in training data

- an attempt to reduce the skewed dataset piece count distribution where there are much more positions with less than 16 pieces, since the target piece count distribution in the trainer is symmetric around 16

The faster convergence seen with ranger21 is meant to:

- prune experiment ideas more quickly since fewer epochs are needed to reach elo maxima

- research faster potential trainings by shortening each run

```yaml

experiment-name: 2560-S7-Re-514G-ranger21-more-wdl-skip

training-dataset: /data/S6-514G.binpack

early-fen-skipping: 28

start-from-engine-test-net: True

nnue-pytorch-branch: linrock/nnue-pytorch/r21-more-wdl-skip

num-epochs: 1200

lr: 4.375e-4

gamma: 0.995

start-lambda: 1.0

end-lambda: 0.7

```

Experiment yaml configs converted to easy_train.sh commands with:

https://github.com/linrock/nnue-tools/blob/4339954/yaml_easy_train.py

Implementations based off of Sopel's NNUE training & experimentation log:

https://docs.google.com/document/d/1gTlrr02qSNKiXNZ_SuO4-RjK4MXBiFlLE6jvNqqMkAY

- Experiment 336 - ranger21 https://github.com/Sopel97/nnue-pytorch/tree/experiment_336

- Experiment 351 - more WDL skipping

The version of the ranger21 optimizer used is:

https://github.com/lessw2020/Ranger21/blob/b507df6/ranger21/ranger21.py

The dataset is the exact same as in:

https://github.com/official-stockfish/Stockfish/pull/4782

Local elo at 25k nodes per move:

nn-epoch619.nnue : 6.2 +/- 4.2

Passed STC:

https://tests.stockfishchess.org/tests/view/658a029779aa8af82b94fbe6

LLR: 2.93 (-2.94,2.94) <0.00,2.00>

Total: 46528 W: 11985 L: 11650 D: 22893 Elo +2.50

Ptnml(0-2): 154, 5489, 11688, 5734, 199

Passed LTC:

https://tests.stockfishchess.org/tests/view/658a448979aa8af82b95010f

LLR: 2.94 (-2.94,2.94) <0.50,2.50>

Total: 265326 W: 66378 L: 65574 D: 133374 Elo +1.05

Ptnml(0-2): 153, 30175, 71254, 30877, 204

This was additionally tested with the latest DualNNUE and passed SPRTs:

Passed STC vs. https://github.com/official-stockfish/Stockfish/pull/4919

https://tests.stockfishchess.org/tests/view/658bcd5c79aa8af82b951846

LLR: 2.93 (-2.94,2.94) <0.00,2.00>

Total: 296128 W: 76273 L: 75554 D: 144301 Elo +0.84

Ptnml(0-2): 1223, 35768, 73617, 35979, 1477

Passed LTC vs. https://github.com/official-stockfish/Stockfish/pull/4919

https://tests.stockfishchess.org/tests/view/658c988d79aa8af82b95240f

LLR: 2.95 (-2.94,2.94) <0.50,2.50>

Total: 75618 W: 19085 L: 18680 D: 37853 Elo +1.86

Ptnml(0-2): 45, 8420, 20497, 8779, 68

closes https://github.com/official-stockfish/Stockfish/pull/4942

Bench: 1304666

Stockfish rating CEDR progress:

Pl	Engine	Rating	Score	Games	%
20	Stockfish 20230227	3752.9	92.00	134	68.66
39	Stockfish ba43fcaff995	3745.4	112.50	159	70.75
40	Stockfish 241022	3745.3	64.00	112	57.14
42	Stockfish 031022	3742.8	110.50	158	69.94
54	Stockfish 070922	3740.2	123.00	164	75.00
58	Stockfish 20230911	3739.7	190.00	324	58.64
64	Stockfish 20230325	3739.1	109.00	169	64.50
67	Stockfish 20221107	3739.0	106.50	186	57.26
73	Stockfish 301022	3738.4	112.00	175	64.00
78	Stockfish 20230218	3737.5	110.50	164	67.38
81	Stockfish 20231008	3737.1	144.50	246	58.74
99	Stockfish 20230128	3735.2	69.00	106	65.09
101	Stockfish 20230813	3734.9	164.50	238	69.12
103	Stockfish 20231014	3734.8	146.00	256	57.03
118	Stockfish 20230525	3733.2	71.00	122	58.20
132	Stockfish 20230903	3731.9	308.50	457	67.51
138	Stockfish 20230711	3731.4	80.50	122	65.98
159	Stockfish 161022	3730.4	147.00	240	61.25
160	Stockfish 20221123	3730.4	247.50	356	69.52
165	Stockfish 20230520	3730.2	106.50	174	61.21
187	Stockfish 20221224	3728.8	65.50	104	62.98
198	Stockfish 20231021	3728.2	90.00	126	71.43
206	Stockfish 20230531	3727.9	61.00	116	52.59
220	Stockfish 23092217	3727.2	91.00	113	80.53
223	Stockfish 16	3727.0	1717.00	2877	59.68
232	Stockfish 20230114	3726.4	295.00	410	71.95