| Finished | ||||||
|---|---|---|---|---|---|---|
| Styxdo | Reckless | 896_test_branch | diff | 8.0+0.08 | LLR: 2.91 (-2.25, 2.89) [0.00, 4.00] Games: 36144 W: 9296 L: 9049 D: 17799 Ptnml(0-2): 205, 4231, 8969, 4446, 221 | Consistent S1 WDL Scheduler with previous training |
| Styxdo | Reckless | 896_test_branch | diff | 8.0+0.08 | LLR: 2.99 (-2.25, 2.89) [0.00, 4.00] Games: 103882 W: 26251 L: 25789 D: 51842 Ptnml(0-2): 559, 12548, 25322, 12896, 616 | Update nnue.rs |
| Styxdo | Reckless | 896_test_branch | diff | N=25000 | LLR: 3.06 (-2.25, 2.89) [0.00, 4.00] Games: 4828 W: 1754 L: 1538 D: 1536 Ptnml(0-2): 158, 486, 992, 538, 240 | Update nnue.rs |
| Styxdo | Reckless | pre_batches_s1_s2 | diff | 8.0+0.08 | LLR: -2.27 (-2.25, 2.89) [0.00, 4.00] Games: 11382 W: 2916 L: 2988 D: 5478 Ptnml(0-2): 93, 1380, 2799, 1344, 75 | Added a few pre-s1 and pre-s2 batches. Model branch: styx_warmup_expt1 |
| Styxdo | Reckless | tb-support-3 | diff | 40.0+0.40 | LLR: 2.90 (-2.25, 2.89) [0.00, 4.00] Games: 11862 W: 2805 L: 2659 D: 6398 Ptnml(0-2): 10, 1126, 3508, 1282, 5 | UHO LTC? |
| Styxdo | Reckless | reduced_training_time_expt4 | diff | 8.0+0.08 | LLR: -1.04 (-2.25, 2.89) [0.00, 4.00] Games: 430 W: 89 L: 142 D: 199 Ptnml(0-2): 8, 72, 103, 29, 3 | Since partial blending seemed to work, this is full blend on same slim model that was -1 elo. |
| Styxdo | Reckless | s0_lite_test2 | diff | 8.0+0.08 | LLR: -2.26 (-2.25, 2.89) [0.00, 4.00] Games: 4366 W: 1051 L: 1143 D: 2172 Ptnml(0-2): 24, 578, 1072, 484, 25 | Lowered the S0 batches to 15. |
| Styxdo | Reckless | s0_lite_test1 | diff | 8.0+0.08 | LLR: -2.26 (-2.25, 2.89) [0.00, 4.00] Games: 3682 W: 885 L: 981 D: 1816 Ptnml(0-2): 26, 501, 877, 417, 20 | Introduction of S0 as discussed. This is 'lite' version, full training on 3080 to follow. |
| Styxdo | Reckless | reduced_training_time_expt3 | diff | 8.0+0.08 | LLR: -2.32 (-2.25, 2.89) [-4.00, 0.00] Games: 42342 W: 10402 L: 10643 D: 21297 Ptnml(0-2): 268, 5319, 10181, 5192, 211 | Adjused S1 Epoch/Batch_size balance but maintained overall training time. S2 Epochs reduced. |
| Styxdo | Reckless | reduced_training_time_expt2 | diff | 8.0+0.08 | LLR: -2.29 (-2.25, 2.89) [0.00, 4.00] Games: 2542 W: 563 L: 662 D: 1317 Ptnml(0-2): 18, 349, 630, 262, 12 | Similar to last time, this time keeps more epoch in S2 |
| Styxdo | Reckless | reduced_training_time_expt1 | diff | 8.0+0.08 | LLR: -2.26 (-2.25, 2.89) [-4.00, 0.00] Games: 4236 W: 1009 L: 1128 D: 2099 Ptnml(0-2): 27, 572, 1033, 465, 21 | Reduced Batchsize, retrains full network on reduced batchsizes attempting to go back to lower training times without losing elo |
| Styxdo | Reckless | material-output-buckets | diff | 8.0+0.08 | LLR: 2.23 (-2.25, 2.89) [0.00, 4.00] Games: 51622 W: 13016 L: 12753 D: 25853 Ptnml(0-2): 258, 6172, 12760, 6291, 330 | STC |
| Styxdo | Reckless | v20-0e6ecbd4 | diff | N=25000 | LLR: -2.42 (-2.25, 2.89) [0.00, 4.00] Games: 39120 W: 12806 L: 12795 D: 13519 Ptnml(0-2): 1452, 4309, 7955, 4464, 1380 | Increased Early Ply Skipping in S2, no reset Optimizer, single output bucket |
| Styxdo | Reckless | v20-0e6ecbd4 | diff | 8.0+0.08 | LLR: 2.94 (-2.25, 2.89) [0.00, 4.00] Games: 56510 W: 14109 L: 13797 D: 28604 Ptnml(0-2): 288, 6806, 13803, 7022, 336 | STC |
| Styxdo | Reckless | material-output-buckets | diff | N=25000 | LLR: 2.93 (-2.25, 2.89) [0.00, 4.00] Games: 7374 W: 2492 L: 2281 D: 2601 Ptnml(0-2): 234, 786, 1469, 931, 267 | S2 Early ply skipping to 28, passed STC on Slim, not resetting optimizer this time. |
| Styxdo | Reckless | early_ply_skipping_full | diff | 8.0+0.08 | LLR: -1.93 (-2.25, 2.89) [0.00, 4.00] Games: 4352 W: 1071 L: 1149 D: 2132 Ptnml(0-2): 31, 576, 1033, 512, 24 | Full S2 training with early ply skipping. Slim passed STC. |
| Styxdo | Reckless | blend_both_s1_s2 | diff | 8.0+0.08 | LLR: -2.26 (-2.25, 2.89) [0.00, 4.00] Games: 15240 W: 3826 L: 3884 D: 7530 Ptnml(0-2): 97, 1833, 3789, 1833, 68 | Blend 640 and 1536 data in both stages |
| Styxdo | Reckless | s2_sigmoid_expt_1 | diff | 8.0+0.08 | LLR: -2.30 (-2.25, 2.89) [0.00, 4.00] Games: 2244 W: 519 L: 621 D: 1104 Ptnml(0-2): 26, 304, 546, 238, 8 | S2 Loss SigmoidMPE modification test. |
| Styxdo | Reckless | blend_s1_pure_1536_s2 | diff | 8.0+0.08 | LLR: -2.29 (-2.25, 2.89) [0.00, 4.00] Games: 35482 W: 8954 L: 8951 D: 17577 Ptnml(0-2): 223, 4342, 8569, 4423, 184 | Blend 640+1536 data in S1, Pure 1536 in S2 |
| Styxdo | Reckless | sigmoid_expt_1 | diff | 8.0+0.08 | LLR: -2.29 (-2.25, 2.89) [0.00, 4.00] Games: 18078 W: 4468 L: 4520 D: 9090 Ptnml(0-2): 75, 2339, 4352, 2109, 164 | Loss sigmoid expt |
| Styxdo | Reckless | s2_lr_ub_increase | diff | 40.0+0.40 | LLR: -2.28 (-2.25, 2.89) [0.00, 4.00] Games: 24380 W: 5778 L: 5807 D: 12795 Ptnml(0-2): 35, 2930, 6289, 2901, 35 | LTC |
| Styxdo | Reckless | s2_lr_ub_increase | diff | 8.0+0.08 | LLR: -2.32 (-2.25, 2.89) [0.00, 4.00] Games: 10394 W: 2529 L: 2606 D: 5259 Ptnml(0-2): 68, 1312, 2506, 1251, 60 | Matching increase in S2 LR |
| Styxdo | Reckless | half_batchsize_double_epoch | diff | 40.0+0.40 | LLR: -1.64 (-2.25, 2.89) [0.00, 4.00] Games: 3156 W: 725 L: 786 D: 1645 Ptnml(0-2): 2, 403, 828, 344, 1 | Extending the experiment to S1 too. Half the batchsize as v18, double the epochs. |
| Styxdo | Reckless | test_s2_400_epoch | diff | 40.0+0.40 | LLR: -2.26 (-2.25, 2.89) [0.00, 4.00] Games: 5554 W: 1234 L: 1316 D: 3004 Ptnml(0-2): 9, 695, 1449, 617, 7 | Test changing to smaller batches but more epochs to match. |
| Styxdo | Reckless | styx_gut_feel_experiment | diff | 8.0+0.08 | LLR: 2.97 (-2.25, 2.89) [0.00, 4.00] Games: 11604 W: 2974 L: 2797 D: 5833 Ptnml(0-2): 38, 1417, 2791, 1442, 114 | STC Test of it |