Update smallnet to nn-baff1ede1f90.nnue with wider eval range #4919
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
this is meant to be a placeholder draft PR for now while the other PR is undergoing changes. however i can't seem to change this into a draft PR. |
linrock
added a commit
to linrock/Stockfish
that referenced
this pull request
Dec 28, 2023
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: official-stockfish#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 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 Ptnml(0-2): 153, 30175, 71254, 30877, 204 This was additionally tested with the latest DualNNUE and passed SPRTs: Passed STC vs. official-stockfish#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 Ptnml(0-2): 1223, 35768, 73617, 35979, 1477 Passed LTC vs. official-stockfish#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 Ptnml(0-2): 45, 8420, 20497, 8779, 68 bench 1049162
Disservin
pushed a commit
that referenced
this pull request
Dec 30, 2023
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: #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 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 Ptnml(0-2): 153, 30175, 71254, 30877, 204 This was additionally tested with the latest DualNNUE and passed SPRTs: Passed STC vs. #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 Ptnml(0-2): 1223, 35768, 73617, 35979, 1477 Passed LTC vs. #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 Ptnml(0-2): 45, 8420, 20497, 8779, 68 closes #4942 Bench: 1304666
Credit goes to @mstembera for: - writing the code enabling dual NNUE: official-stockfish#4898 - the idea of trying L1-128 trained exclusively on high simple eval positions The L1-128 smallnet is: - epoch 399 of a single-stage training from scratch - trained only on positions from filtered data with high material difference - defined by abs(simple_eval) > 1000 ```yaml experiment-name: 128--S1-only-hse-v2 training-dataset: - /data/hse/S3/dfrc99-16tb7p-eval-filt-v2.min.high-simple-eval-1k.binpack - /data/hse/S3/leela96-filt-v2.min.high-simple-eval-1k.binpack - /data/hse/S3/test80-apr2022-16tb7p.min.high-simple-eval-1k.binpack - /data/hse/S7/test60-2020-2tb7p.v6-3072.high-simple-eval-1k.binpack - /data/hse/S7/test60-novdec2021-12tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-1k.binpack - /data/hse/S7/test77-nov2021-2tb7p.v6-3072.min.high-simple-eval-1k.binpack - /data/hse/S7/test77-dec2021-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-1k.binpack - /data/hse/S7/test77-jan2022-2tb7p.high-simple-eval-1k.binpack - /data/hse/S7/test78-jantomay2022-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-1k.binpack - /data/hse/S7/test78-juntosep2022-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-1k.binpack - /data/hse/S7/test79-apr2022-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-1k.binpack - /data/hse/S7/test79-may2022-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-1k.binpack # T80 2022 - /data/hse/S7/test80-may2022-16tb7p.high-simple-eval-1k.binpack - /data/hse/S7/test80-jun2022-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-1k.binpack - /data/hse/S7/test80-jul2022-16tb7p.v6-dd.min.high-simple-eval-1k.binpack - /data/hse/S7/test80-aug2022-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-1k.binpack - /data/hse/S7/test80-sep2022-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-1k.binpack - /data/hse/S7/test80-oct2022-16tb7p.v6-dd.high-simple-eval-1k.binpack - /data/hse/S7/test80-nov2022-16tb7p-v6-dd.min.high-simple-eval-1k.binpack # T80 2023 - /data/hse/S7/test80-jan2023-3of3-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-1k.binpack - /data/hse/S7/test80-feb2023-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-1k.binpack - /data/hse/S7/test80-mar2023-2tb7p.v6-sk16.min.high-simple-eval-1k.binpack - /data/hse/S7/test80-apr2023-2tb7p-filter-v6-sk16.min.high-simple-eval-1k.binpack - /data/hse/S7/test80-may2023-2tb7p.v6.min.high-simple-eval-1k.binpack - /data/hse/S7/test80-jun2023-2tb7p.v6-3072.min.high-simple-eval-1k.binpack - /data/hse/S7/test80-jul2023-2tb7p.v6-3072.min.high-simple-eval-1k.binpack - /data/hse/S7/test80-aug2023-2tb7p.v6.min.high-simple-eval-1k.binpack - /data/hse/S7/test80-sep2023-2tb7p.high-simple-eval-1k.binpack - /data/hse/S7/test80-oct2023-2tb7p.high-simple-eval-1k.binpack start-from-engine-test-net: False nnue-pytorch-branch: linrock/nnue-pytorch/L1-128 engine-test-branch: linrock/Stockfish/L1-128-nolazy engine-base-branch: linrock/Stockfish/L1-128 num-epochs: 500 lambda: 1.0 ``` Experiment yaml configs converted to easy_train.sh commands with: https://github.com/linrock/nnue-tools/blob/4339954/yaml_easy_train.py Binpacks interleaved at training time with: official-stockfish/nnue-pytorch#259 Data filtered for high simple eval positions with: https://github.com/linrock/nnue-data/blob/32d6a68/filter_high_simple_eval_plain.py https://github.com/linrock/Stockfish/blob/61dbfe/src/tools/transform.cpp#L626-L655 Training data can be found at: https://robotmoon.com/nnue-training-data/ Local elo at 25k nodes per move of L1-128 smallnet (nnue-only eval) vs. L1-128 trained on standard S1 data: nn-epoch399.nnue : -318.1 +/- 2.1 Passed STC: https://tests.stockfishchess.org/tests/view/6574cb9d95ea6ba1fcd49e3b LLR: 2.93 (-2.94,2.94) <0.00,2.00> Total: 62432 W: 15875 L: 15521 D: 31036 Ptnml(0-2): 177, 7331, 15872, 7633, 203 Passed LTC: https://tests.stockfishchess.org/tests/view/6575da2d4d789acf40aaac6e LLR: 2.94 (-2.94,2.94) <0.50,2.50> Total: 64830 W: 16118 L: 15738 D: 32974 Ptnml(0-2): 43, 7129, 17697, 7497, 49 Bench: 1330050 Co-Authored-By: mstembera <[email protected]>
Created by training an L1-128 net from scratch with a wider range of evals in the training data and wld-fen-skipping disabled during training. The differences in this training data compared to the first dual nnue PR are: - removal of all positions with 3 pieces - when piece count >= 16, keep positions with simple eval above 750 - when piece count < 16, remove positions with simple eval above 3000 The asymmetric data filtering was meant to flatten the training data piece count distribution, which was previously heavily skewed towards positions with low piece counts. Additionally, the simple eval range where the smallnet is used was widened to cover more positions previously evaluated by the big net and simple eval. ```yaml experiment-name: 128--S1-hse-S7-v4-S3-v1-no-wld-skip training-dataset: - /data/hse/S3/leela96-filt-v2.min.high-simple-eval-1k.binpack - /data/hse/S3/dfrc99-16tb7p-eval-filt-v2.min.high-simple-eval-1k.binpack - /data/hse/S3/test80-apr2022-16tb7p.min.high-simple-eval-1k.binpack - /data/hse/S7/test60-2020-2tb7p.v6-3072.high-simple-eval-v4.binpack - /data/hse/S7/test60-novdec2021-12tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test77-nov2021-2tb7p.v6-3072.min.high-simple-eval-v4.binpack - /data/hse/S7/test77-dec2021-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test77-jan2022-2tb7p.high-simple-eval-v4.binpack - /data/hse/S7/test78-jantomay2022-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test78-juntosep2022-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test79-apr2022-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test79-may2022-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test80-may2022-16tb7p.high-simple-eval-v4.binpack - /data/hse/S7/test80-jun2022-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test80-jul2022-16tb7p.v6-dd.min.high-simple-eval-v4.binpack - /data/hse/S7/test80-aug2022-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test80-sep2022-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test80-oct2022-16tb7p.v6-dd.high-simple-eval-v4.binpack - /data/hse/S7/test80-nov2022-16tb7p-v6-dd.min.high-simple-eval-v4.binpack - /data/hse/S7/test80-jan2023-3of3-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test80-feb2023-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test80-mar2023-2tb7p.v6-sk16.min.high-simple-eval-v4.binpack - /data/hse/S7/test80-apr2023-2tb7p-filter-v6-sk16.min.high-simple-eval-v4.binpack - /data/hse/S7/test80-may2023-2tb7p.v6.min.high-simple-eval-v4.binpack - /data/hse/S7/test80-jun2023-2tb7p.v6-3072.min.high-simple-eval-v4.binpack - /data/hse/S7/test80-jul2023-2tb7p.v6-3072.min.high-simple-eval-v4.binpack - /data/hse/S7/test80-aug2023-2tb7p.v6.min.high-simple-eval-v4.binpack - /data/hse/S7/test80-sep2023-2tb7p.high-simple-eval-v4.binpack - /data/hse/S7/test80-oct2023-2tb7p.high-simple-eval-v4.binpack wld-fen-skipping: False start-from-engine-test-net: False nnue-pytorch-branch: linrock/nnue-pytorch/L1-128 engine-test-branch: linrock/Stockfish/L1-128-nolazy engine-base-branch: linrock/Stockfish/L1-128 num-epochs: 500 start-lambda: 1.0 end-lambda: 1.0 ``` Experiment yaml configs converted to easy_train.sh commands with: https://github.com/linrock/nnue-tools/blob/4339954/yaml_easy_train.py Binpacks interleaved at training time with: official-stockfish/nnue-pytorch#259 FT weights permuted with 10k positions from fishpack32.binpack with: official-stockfish/nnue-pytorch#254 Data filtered for high simple eval positions (v4) with: https://github.com/linrock/Stockfish/blob/b9c8440/src/tools/transform.cpp#L640-L675 Training data can be found at: https://robotmoon.com/nnue-training-data/ Local elo at 25k nodes per move of L1-128 smallnet (nnue-only eval) vs. L1-128 trained on standard S1 data: nn-epoch319.nnue : -241.7 +/- 3.2 Passed STC vs. 36db936: https://tests.stockfishchess.org/tests/view/6576b3484d789acf40aabbfe LLR: 2.94 (-2.94,2.94) <0.00,2.00> Total: 21920 W: 5680 L: 5381 D: 10859 Ptnml(0-2): 82, 2488, 5520, 2789, 81 Passed LTC vs. DualNNUE official-stockfish#4915: https://tests.stockfishchess.org/tests/view/65775c034d789acf40aac7e3 LLR: 2.95 (-2.94,2.94) <0.50,2.50> Total: 147606 W: 36619 L: 36063 D: 74924 Ptnml(0-2): 98, 16591, 39891, 17103, 120 Bench: 1438336
linrock
force-pushed
the
n2-lazy-1050-2550-pr
branch
from
595e24e to
14f4d95
Compare
linrock
changed the title
Disservin
pushed a commit
to Disservin/Stockfish
that referenced
this pull request
Jan 8, 2024
Created by training an L1-128 net from scratch with a wider range of evals in the training data and wld-fen-skipping disabled during training. The differences in this training data compared to the first dual nnue PR are: - removal of all positions with 3 pieces - when piece count >= 16, keep positions with simple eval above 750 - when piece count < 16, remove positions with simple eval above 3000 The asymmetric data filtering was meant to flatten the training data piece count distribution, which was previously heavily skewed towards positions with low piece counts. Additionally, the simple eval range where the smallnet is used was widened to cover more positions previously evaluated by the big net and simple eval. ```yaml experiment-name: 128--S1-hse-S7-v4-S3-v1-no-wld-skip training-dataset: - /data/hse/S3/leela96-filt-v2.min.high-simple-eval-1k.binpack - /data/hse/S3/dfrc99-16tb7p-eval-filt-v2.min.high-simple-eval-1k.binpack - /data/hse/S3/test80-apr2022-16tb7p.min.high-simple-eval-1k.binpack - /data/hse/S7/test60-2020-2tb7p.v6-3072.high-simple-eval-v4.binpack - /data/hse/S7/test60-novdec2021-12tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test77-nov2021-2tb7p.v6-3072.min.high-simple-eval-v4.binpack - /data/hse/S7/test77-dec2021-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test77-jan2022-2tb7p.high-simple-eval-v4.binpack - /data/hse/S7/test78-jantomay2022-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test78-juntosep2022-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test79-apr2022-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test79-may2022-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test80-may2022-16tb7p.high-simple-eval-v4.binpack - /data/hse/S7/test80-jun2022-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test80-jul2022-16tb7p.v6-dd.min.high-simple-eval-v4.binpack - /data/hse/S7/test80-aug2022-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test80-sep2022-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test80-oct2022-16tb7p.v6-dd.high-simple-eval-v4.binpack - /data/hse/S7/test80-nov2022-16tb7p-v6-dd.min.high-simple-eval-v4.binpack - /data/hse/S7/test80-jan2023-3of3-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test80-feb2023-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test80-mar2023-2tb7p.v6-sk16.min.high-simple-eval-v4.binpack - /data/hse/S7/test80-apr2023-2tb7p-filter-v6-sk16.min.high-simple-eval-v4.binpack - /data/hse/S7/test80-may2023-2tb7p.v6.min.high-simple-eval-v4.binpack - /data/hse/S7/test80-jun2023-2tb7p.v6-3072.min.high-simple-eval-v4.binpack - /data/hse/S7/test80-jul2023-2tb7p.v6-3072.min.high-simple-eval-v4.binpack - /data/hse/S7/test80-aug2023-2tb7p.v6.min.high-simple-eval-v4.binpack - /data/hse/S7/test80-sep2023-2tb7p.high-simple-eval-v4.binpack - /data/hse/S7/test80-oct2023-2tb7p.high-simple-eval-v4.binpack wld-fen-skipping: False start-from-engine-test-net: False nnue-pytorch-branch: linrock/nnue-pytorch/L1-128 engine-test-branch: linrock/Stockfish/L1-128-nolazy engine-base-branch: linrock/Stockfish/L1-128 num-epochs: 500 start-lambda: 1.0 end-lambda: 1.0 ``` Experiment yaml configs converted to easy_train.sh commands with: https://github.com/linrock/nnue-tools/blob/4339954/yaml_easy_train.py Binpacks interleaved at training time with: official-stockfish/nnue-pytorch#259 FT weights permuted with 10k positions from fishpack32.binpack with: official-stockfish/nnue-pytorch#254 Data filtered for high simple eval positions (v4) with: https://github.com/linrock/Stockfish/blob/b9c8440/src/tools/transform.cpp#L640-L675 Training data can be found at: https://robotmoon.com/nnue-training-data/ Local elo at 25k nodes per move of L1-128 smallnet (nnue-only eval) vs. L1-128 trained on standard S1 data: nn-epoch319.nnue : -241.7 +/- 3.2 Passed STC vs. 36db936: https://tests.stockfishchess.org/tests/view/6576b3484d789acf40aabbfe LLR: 2.94 (-2.94,2.94) <0.00,2.00> Total: 21920 W: 5680 L: 5381 D: 10859 Ptnml(0-2): 82, 2488, 5520, 2789, 81 Passed LTC vs. DualNNUE official-stockfish#4915: https://tests.stockfishchess.org/tests/view/65775c034d789acf40aac7e3 LLR: 2.95 (-2.94,2.94) <0.50,2.50> Total: 147606 W: 36619 L: 36063 D: 74924 Ptnml(0-2): 98, 16591, 39891, 17103, 120 closes official-stockfish#4919 Bench: 1438336
Joachim26
pushed a commit
to Joachim26/StockfishNPS
that referenced
this pull request
Jan 14, 2024
Created by training an L1-128 net from scratch with a wider range of evals in the training data and wld-fen-skipping disabled during training. The differences in this training data compared to the first dual nnue PR are: - removal of all positions with 3 pieces - when piece count >= 16, keep positions with simple eval above 750 - when piece count < 16, remove positions with simple eval above 3000 The asymmetric data filtering was meant to flatten the training data piece count distribution, which was previously heavily skewed towards positions with low piece counts. Additionally, the simple eval range where the smallnet is used was widened to cover more positions previously evaluated by the big net and simple eval. ```yaml experiment-name: 128--S1-hse-S7-v4-S3-v1-no-wld-skip training-dataset: - /data/hse/S3/leela96-filt-v2.min.high-simple-eval-1k.binpack - /data/hse/S3/dfrc99-16tb7p-eval-filt-v2.min.high-simple-eval-1k.binpack - /data/hse/S3/test80-apr2022-16tb7p.min.high-simple-eval-1k.binpack - /data/hse/S7/test60-2020-2tb7p.v6-3072.high-simple-eval-v4.binpack - /data/hse/S7/test60-novdec2021-12tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test77-nov2021-2tb7p.v6-3072.min.high-simple-eval-v4.binpack - /data/hse/S7/test77-dec2021-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test77-jan2022-2tb7p.high-simple-eval-v4.binpack - /data/hse/S7/test78-jantomay2022-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test78-juntosep2022-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test79-apr2022-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test79-may2022-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test80-may2022-16tb7p.high-simple-eval-v4.binpack - /data/hse/S7/test80-jun2022-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test80-jul2022-16tb7p.v6-dd.min.high-simple-eval-v4.binpack - /data/hse/S7/test80-aug2022-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test80-sep2022-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test80-oct2022-16tb7p.v6-dd.high-simple-eval-v4.binpack - /data/hse/S7/test80-nov2022-16tb7p-v6-dd.min.high-simple-eval-v4.binpack - /data/hse/S7/test80-jan2023-3of3-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test80-feb2023-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test80-mar2023-2tb7p.v6-sk16.min.high-simple-eval-v4.binpack - /data/hse/S7/test80-apr2023-2tb7p-filter-v6-sk16.min.high-simple-eval-v4.binpack - /data/hse/S7/test80-may2023-2tb7p.v6.min.high-simple-eval-v4.binpack - /data/hse/S7/test80-jun2023-2tb7p.v6-3072.min.high-simple-eval-v4.binpack - /data/hse/S7/test80-jul2023-2tb7p.v6-3072.min.high-simple-eval-v4.binpack - /data/hse/S7/test80-aug2023-2tb7p.v6.min.high-simple-eval-v4.binpack - /data/hse/S7/test80-sep2023-2tb7p.high-simple-eval-v4.binpack - /data/hse/S7/test80-oct2023-2tb7p.high-simple-eval-v4.binpack wld-fen-skipping: False start-from-engine-test-net: False nnue-pytorch-branch: linrock/nnue-pytorch/L1-128 engine-test-branch: linrock/Stockfish/L1-128-nolazy engine-base-branch: linrock/Stockfish/L1-128 num-epochs: 500 start-lambda: 1.0 end-lambda: 1.0 ``` Experiment yaml configs converted to easy_train.sh commands with: https://github.com/linrock/nnue-tools/blob/4339954/yaml_easy_train.py Binpacks interleaved at training time with: official-stockfish/nnue-pytorch#259 FT weights permuted with 10k positions from fishpack32.binpack with: official-stockfish/nnue-pytorch#254 Data filtered for high simple eval positions (v4) with: https://github.com/linrock/Stockfish/blob/b9c8440/src/tools/transform.cpp#L640-L675 Training data can be found at: https://robotmoon.com/nnue-training-data/ Local elo at 25k nodes per move of L1-128 smallnet (nnue-only eval) vs. L1-128 trained on standard S1 data: nn-epoch319.nnue : -241.7 +/- 3.2 Passed STC vs. 36db936: https://tests.stockfishchess.org/tests/view/6576b3484d789acf40aabbfe LLR: 2.94 (-2.94,2.94) <0.00,2.00> Total: 21920 W: 5680 L: 5381 D: 10859 Ptnml(0-2): 82, 2488, 5520, 2789, 81 Passed LTC vs. DualNNUE official-stockfish#4915: https://tests.stockfishchess.org/tests/view/65775c034d789acf40aac7e3 LLR: 2.95 (-2.94,2.94) <0.50,2.50> Total: 147606 W: 36619 L: 36063 D: 74924 Ptnml(0-2): 98, 16591, 39891, 17103, 120 closes official-stockfish#4919 Bench: 1438336
rn5f107s2
pushed a commit
to rn5f107s2/Stockfish
that referenced
this pull request
Jan 14, 2024
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: official-stockfish#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 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 Ptnml(0-2): 153, 30175, 71254, 30877, 204 This was additionally tested with the latest DualNNUE and passed SPRTs: Passed STC vs. official-stockfish#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 Ptnml(0-2): 1223, 35768, 73617, 35979, 1477 Passed LTC vs. official-stockfish#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 Ptnml(0-2): 45, 8420, 20497, 8779, 68 closes official-stockfish#4942 Bench: 1304666
rn5f107s2
pushed a commit
to rn5f107s2/Stockfish
that referenced
this pull request
Jan 14, 2024
Created by training an L1-128 net from scratch with a wider range of evals in the training data and wld-fen-skipping disabled during training. The differences in this training data compared to the first dual nnue PR are: - removal of all positions with 3 pieces - when piece count >= 16, keep positions with simple eval above 750 - when piece count < 16, remove positions with simple eval above 3000 The asymmetric data filtering was meant to flatten the training data piece count distribution, which was previously heavily skewed towards positions with low piece counts. Additionally, the simple eval range where the smallnet is used was widened to cover more positions previously evaluated by the big net and simple eval. ```yaml experiment-name: 128--S1-hse-S7-v4-S3-v1-no-wld-skip training-dataset: - /data/hse/S3/leela96-filt-v2.min.high-simple-eval-1k.binpack - /data/hse/S3/dfrc99-16tb7p-eval-filt-v2.min.high-simple-eval-1k.binpack - /data/hse/S3/test80-apr2022-16tb7p.min.high-simple-eval-1k.binpack - /data/hse/S7/test60-2020-2tb7p.v6-3072.high-simple-eval-v4.binpack - /data/hse/S7/test60-novdec2021-12tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test77-nov2021-2tb7p.v6-3072.min.high-simple-eval-v4.binpack - /data/hse/S7/test77-dec2021-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test77-jan2022-2tb7p.high-simple-eval-v4.binpack - /data/hse/S7/test78-jantomay2022-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test78-juntosep2022-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test79-apr2022-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test79-may2022-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test80-may2022-16tb7p.high-simple-eval-v4.binpack - /data/hse/S7/test80-jun2022-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test80-jul2022-16tb7p.v6-dd.min.high-simple-eval-v4.binpack - /data/hse/S7/test80-aug2022-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test80-sep2022-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test80-oct2022-16tb7p.v6-dd.high-simple-eval-v4.binpack - /data/hse/S7/test80-nov2022-16tb7p-v6-dd.min.high-simple-eval-v4.binpack - /data/hse/S7/test80-jan2023-3of3-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test80-feb2023-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test80-mar2023-2tb7p.v6-sk16.min.high-simple-eval-v4.binpack - /data/hse/S7/test80-apr2023-2tb7p-filter-v6-sk16.min.high-simple-eval-v4.binpack - /data/hse/S7/test80-may2023-2tb7p.v6.min.high-simple-eval-v4.binpack - /data/hse/S7/test80-jun2023-2tb7p.v6-3072.min.high-simple-eval-v4.binpack - /data/hse/S7/test80-jul2023-2tb7p.v6-3072.min.high-simple-eval-v4.binpack - /data/hse/S7/test80-aug2023-2tb7p.v6.min.high-simple-eval-v4.binpack - /data/hse/S7/test80-sep2023-2tb7p.high-simple-eval-v4.binpack - /data/hse/S7/test80-oct2023-2tb7p.high-simple-eval-v4.binpack wld-fen-skipping: False start-from-engine-test-net: False nnue-pytorch-branch: linrock/nnue-pytorch/L1-128 engine-test-branch: linrock/Stockfish/L1-128-nolazy engine-base-branch: linrock/Stockfish/L1-128 num-epochs: 500 start-lambda: 1.0 end-lambda: 1.0 ``` Experiment yaml configs converted to easy_train.sh commands with: https://github.com/linrock/nnue-tools/blob/4339954/yaml_easy_train.py Binpacks interleaved at training time with: official-stockfish/nnue-pytorch#259 FT weights permuted with 10k positions from fishpack32.binpack with: official-stockfish/nnue-pytorch#254 Data filtered for high simple eval positions (v4) with: https://github.com/linrock/Stockfish/blob/b9c8440/src/tools/transform.cpp#L640-L675 Training data can be found at: https://robotmoon.com/nnue-training-data/ Local elo at 25k nodes per move of L1-128 smallnet (nnue-only eval) vs. L1-128 trained on standard S1 data: nn-epoch319.nnue : -241.7 +/- 3.2 Passed STC vs. 36db936: https://tests.stockfishchess.org/tests/view/6576b3484d789acf40aabbfe LLR: 2.94 (-2.94,2.94) <0.00,2.00> Total: 21920 W: 5680 L: 5381 D: 10859 Ptnml(0-2): 82, 2488, 5520, 2789, 81 Passed LTC vs. DualNNUE official-stockfish#4915: https://tests.stockfishchess.org/tests/view/65775c034d789acf40aac7e3 LLR: 2.95 (-2.94,2.94) <0.50,2.50> Total: 147606 W: 36619 L: 36063 D: 74924 Ptnml(0-2): 98, 16591, 39891, 17103, 120 closes official-stockfish#4919 Bench: 1438336
windfishballad
pushed a commit
to windfishballad/Stockfish
that referenced
this pull request
Jan 23, 2024
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: official-stockfish#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 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 Ptnml(0-2): 153, 30175, 71254, 30877, 204 This was additionally tested with the latest DualNNUE and passed SPRTs: Passed STC vs. official-stockfish#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 Ptnml(0-2): 1223, 35768, 73617, 35979, 1477 Passed LTC vs. official-stockfish#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 Ptnml(0-2): 45, 8420, 20497, 8779, 68 closes official-stockfish#4942 Bench: 1304666
windfishballad
pushed a commit
to windfishballad/Stockfish
that referenced
this pull request
Jan 23, 2024
Created by training an L1-128 net from scratch with a wider range of evals in the training data and wld-fen-skipping disabled during training. The differences in this training data compared to the first dual nnue PR are: - removal of all positions with 3 pieces - when piece count >= 16, keep positions with simple eval above 750 - when piece count < 16, remove positions with simple eval above 3000 The asymmetric data filtering was meant to flatten the training data piece count distribution, which was previously heavily skewed towards positions with low piece counts. Additionally, the simple eval range where the smallnet is used was widened to cover more positions previously evaluated by the big net and simple eval. ```yaml experiment-name: 128--S1-hse-S7-v4-S3-v1-no-wld-skip training-dataset: - /data/hse/S3/leela96-filt-v2.min.high-simple-eval-1k.binpack - /data/hse/S3/dfrc99-16tb7p-eval-filt-v2.min.high-simple-eval-1k.binpack - /data/hse/S3/test80-apr2022-16tb7p.min.high-simple-eval-1k.binpack - /data/hse/S7/test60-2020-2tb7p.v6-3072.high-simple-eval-v4.binpack - /data/hse/S7/test60-novdec2021-12tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test77-nov2021-2tb7p.v6-3072.min.high-simple-eval-v4.binpack - /data/hse/S7/test77-dec2021-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test77-jan2022-2tb7p.high-simple-eval-v4.binpack - /data/hse/S7/test78-jantomay2022-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test78-juntosep2022-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test79-apr2022-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test79-may2022-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test80-may2022-16tb7p.high-simple-eval-v4.binpack - /data/hse/S7/test80-jun2022-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test80-jul2022-16tb7p.v6-dd.min.high-simple-eval-v4.binpack - /data/hse/S7/test80-aug2022-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test80-sep2022-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test80-oct2022-16tb7p.v6-dd.high-simple-eval-v4.binpack - /data/hse/S7/test80-nov2022-16tb7p-v6-dd.min.high-simple-eval-v4.binpack - /data/hse/S7/test80-jan2023-3of3-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test80-feb2023-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-v4.binpack - /data/hse/S7/test80-mar2023-2tb7p.v6-sk16.min.high-simple-eval-v4.binpack - /data/hse/S7/test80-apr2023-2tb7p-filter-v6-sk16.min.high-simple-eval-v4.binpack - /data/hse/S7/test80-may2023-2tb7p.v6.min.high-simple-eval-v4.binpack - /data/hse/S7/test80-jun2023-2tb7p.v6-3072.min.high-simple-eval-v4.binpack - /data/hse/S7/test80-jul2023-2tb7p.v6-3072.min.high-simple-eval-v4.binpack - /data/hse/S7/test80-aug2023-2tb7p.v6.min.high-simple-eval-v4.binpack - /data/hse/S7/test80-sep2023-2tb7p.high-simple-eval-v4.binpack - /data/hse/S7/test80-oct2023-2tb7p.high-simple-eval-v4.binpack wld-fen-skipping: False start-from-engine-test-net: False nnue-pytorch-branch: linrock/nnue-pytorch/L1-128 engine-test-branch: linrock/Stockfish/L1-128-nolazy engine-base-branch: linrock/Stockfish/L1-128 num-epochs: 500 start-lambda: 1.0 end-lambda: 1.0 ``` Experiment yaml configs converted to easy_train.sh commands with: https://github.com/linrock/nnue-tools/blob/4339954/yaml_easy_train.py Binpacks interleaved at training time with: official-stockfish/nnue-pytorch#259 FT weights permuted with 10k positions from fishpack32.binpack with: official-stockfish/nnue-pytorch#254 Data filtered for high simple eval positions (v4) with: https://github.com/linrock/Stockfish/blob/b9c8440/src/tools/transform.cpp#L640-L675 Training data can be found at: https://robotmoon.com/nnue-training-data/ Local elo at 25k nodes per move of L1-128 smallnet (nnue-only eval) vs. L1-128 trained on standard S1 data: nn-epoch319.nnue : -241.7 +/- 3.2 Passed STC vs. 36db936: https://tests.stockfishchess.org/tests/view/6576b3484d789acf40aabbfe LLR: 2.94 (-2.94,2.94) <0.00,2.00> Total: 21920 W: 5680 L: 5381 D: 10859 Ptnml(0-2): 82, 2488, 5520, 2789, 81 Passed LTC vs. DualNNUE official-stockfish#4915: https://tests.stockfishchess.org/tests/view/65775c034d789acf40aac7e3 LLR: 2.95 (-2.94,2.94) <0.50,2.50> Total: 147606 W: 36619 L: 36063 D: 74924 Ptnml(0-2): 98, 16591, 39891, 17103, 120 closes official-stockfish#4919 Bench: 1438336
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Created by training an L1-128 net from scratch with a wider range of evals in the training data and wld-fen-skipping disabled during training. The differences in this training data compared to the first dual nnue PR are:
The asymmetric data filtering was meant to flatten the training data piece count distribution, which was previously heavily skewed towards positions with low piece counts.
Additionally, the simple eval range where the smallnet is used was widened to cover more positions previously evaluated by the big net and simple eval.
Experiment yaml configs converted to easy_train.sh commands with:
https://github.com/linrock/nnue-tools/blob/4339954/yaml_easy_train.py
Binpacks interleaved at training time with:
official-stockfish/nnue-pytorch#259
FT weights permuted with 10k positions from fishpack32.binpack with:
official-stockfish/nnue-pytorch#254
Data filtered for high simple eval positions (v4) with:
https://github.com/linrock/Stockfish/blob/b9c8440/src/tools/transform.cpp#L640-L675
Training data can be found at:
https://robotmoon.com/nnue-training-data/
Local elo at 25k nodes per move of
L1-128 smallnet (nnue-only eval) vs. L1-128 trained on standard S1 data:
nn-epoch319.nnue : -241.7 +/- 3.2
Passed STC vs. 36db936:
https://tests.stockfishchess.org/tests/view/6576b3484d789acf40aabbfe
LLR: 2.94 (-2.94,2.94) <0.00,2.00>
Total: 21920 W: 5680 L: 5381 D: 10859
Ptnml(0-2): 82, 2488, 5520, 2789, 81
Passed LTC vs. DualNNUE #4915:
https://tests.stockfishchess.org/tests/view/65775c034d789acf40aac7e3
LLR: 2.95 (-2.94,2.94) <0.50,2.50>
Total: 147606 W: 36619 L: 36063 D: 74924
Ptnml(0-2): 98, 16591, 39891, 17103, 120
Bench: 1438336