train.log

[21:03:49] trial=exp276_lichess_endgame_stream arch=squares64 device=cuda vocab=1968
[21:03:49] train bs=528 accum=1 grad_ckpt=False compile=True elo_every=0 save_every=150
[21:03:49] model=Squares64RecurrentConfig(encoder_dim=256, hidden_dim=736, num_heads=8, ffn_ratio=4, dropout=0.05, use_swiglu=True, use_qk_norm=True, zero_init_out_proj=True, gradient_checkpointing=False, prefix_layers=4, recurrent_layers=7, recurrent_unrolls=3, suffix_layers=4, policy_head_dim=384, value_hidden=384, n_value_classes=3, use_history=False, use_threat_head=False)
[21:03:50] params=98.97M
[21:03:50] WEIGHTS-ONLY WARM START /root/transform/outputs/exp276_lichess_endgame_stream/init.pt from eval/SWA snapshot (optimizer/RNG re-initialized; step reset to 0)
[21:03:50] optimizer=PolarNorMuon (97.7M) + AdamW aux (1.3M) compile_polar=True
[21:03:50] force_lr muon=0.002 adam=3e-05
[21:03:51] torch.compile enabled (mode=default); will fall back if inductor fails
[21:03:51] external eval lichess_endgame n=8,192 /root/transform/outputs/exp276_lichess_endgame_stream/lichess_eval.pt
[21:03:51] soft train=479,716 val=0 blocked=0 targets empty=0 unnorm=0 external_eval
[21:04:54] batch probe ok bs=528 peak_vram=35.08GB compile=True
[21:04:54] average_recurrent_grads() divides bank .grad by unroll count. That does NOT divide optimizer updates by the same amount: Polar-NorMuon is scale-invariant after orthogonalization, and RMS-Adam (m/sqrt(v)) is also approximately scale-invariant for a stationary grad. What does scale is SGD-style updates and the pre-clip global grad norm (so clipping fires less often after averaging). Use explicit param-group LRs if different update sizes are intended.
[21:04:54] exposure deep/shallow odds=0.0x (mix=0.0 deep_n=0 shallow_n=479716)
[21:04:58] step 1/200 | loss=1.6934 | lr=2.00e-03,3.00e-05 | clip=1/25 | mix s/d/b=528/0/0 q=0 p=0 | 151 pos/s | vram=35.08GB
[21:05:19] step 25/200 | loss=1.5800 | lr=2.00e-03,3.00e-05 | clip=20/25 | mix s/d/b=13200/0/0 q=0 p=0 | 525 pos/s | vram=35.48GB
[21:05:42] step 50/200 | loss=1.5017 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=26400/0/0 q=0 p=0 | 553 pos/s | vram=35.48GB
[21:06:05] step 75/200 | loss=1.4767 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=39600/0/0 q=0 p=0 | 562 pos/s | vram=35.48GB
[21:06:27] step 100/200 | loss=1.4533 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=52800/0/0 q=0 p=0 | 567 pos/s | vram=35.48GB
[21:06:36] val/lichess_endgame hard_ce=1.3634 soft_ce=1.3634 soft_temp_ce=1.3634 teacher_entropy=0.0000 teacher_kl=1.3634 wdl_ce=0.5088 value_rows=8192.0000
[21:06:59] step 125/200 | loss=1.4420 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=66000/0/0 q=0 p=0 | 528 pos/s | vram=35.48GB
[21:07:22] step 150/200 | loss=1.4320 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=79200/0/0 q=0 p=0 | 536 pos/s | vram=35.48GB
[21:07:23] ckpt latest.pt (live weights) steps=150 status=mid
[21:07:46] step 175/200 | loss=1.4251 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=92400/0/0 q=0 p=0 | 539 pos/s | vram=35.48GB
[21:08:08] step 200/200 | loss=1.4273 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=105600/0/0 q=0 p=0 | 544 pos/s | vram=35.48GB
[21:08:17] val/lichess_endgame hard_ce=1.3480 soft_ce=1.3480 soft_temp_ce=1.3480 teacher_entropy=0.0000 teacher_kl=1.3480 wdl_ce=0.5050 value_rows=8192.0000
[21:08:18] timing wall=203s active=183s steps=200 status=trained
[21:08:18] done steps=200 pos_s=519.5 params=98.97M status=trained kind=full
[21:08:37] trial=exp276_lichess_endgame_stream arch=squares64 device=cuda vocab=1968
[21:08:37] train bs=528 accum=1 grad_ckpt=False compile=True elo_every=0 save_every=150
[21:08:37] model=Squares64RecurrentConfig(encoder_dim=256, hidden_dim=736, num_heads=8, ffn_ratio=4, dropout=0.05, use_swiglu=True, use_qk_norm=True, zero_init_out_proj=True, gradient_checkpointing=False, prefix_layers=4, recurrent_layers=7, recurrent_unrolls=3, suffix_layers=4, policy_head_dim=384, value_hidden=384, n_value_classes=3, use_history=False, use_threat_head=False)
[21:08:38] params=98.97M
[21:08:38] FULL RESUME /root/transform/outputs/exp276_lichess_endgame_stream/latest.pt steps=200
[21:08:38] optimizer=PolarNorMuon (97.7M) + AdamW aux (1.3M) compile_polar=True
[21:08:38] optimizer state restored
[21:08:38] force_lr muon=0.002 adam=3e-05
[21:08:38] torch.compile enabled (mode=default); will fall back if inductor fails
[21:08:38] external eval lichess_endgame n=8,192 /root/transform/outputs/exp276_lichess_endgame_stream/lichess_eval.pt
[21:08:41] soft train=3,998,779 val=0 blocked=0 targets empty=0 unnorm=0 external_eval
[21:09:44] batch probe ok bs=528 peak_vram=35.48GB compile=True
[21:09:44] RNG state restored
[21:09:44] average_recurrent_grads() divides bank .grad by unroll count. That does NOT divide optimizer updates by the same amount: Polar-NorMuon is scale-invariant after orthogonalization, and RMS-Adam (m/sqrt(v)) is also approximately scale-invariant for a stationary grad. What does scale is SGD-style updates and the pre-clip global grad norm (so clipping fires less often after averaging). Use explicit param-group LRs if different update sizes are intended.
[21:09:44] exposure deep/shallow odds=0.0x (mix=0.0 deep_n=0 shallow_n=3998779)
[21:09:45] timing wall=0s active=0s steps=200 status=trained
[21:09:45] done steps=200 pos_s=0.0 params=98.97M status=trained kind=full
[21:10:01] trial=exp276_lichess_endgame_stream arch=squares64 device=cuda vocab=1968
[21:10:01] train bs=528 accum=1 grad_ckpt=False compile=True elo_every=0 save_every=150
[21:10:01] model=Squares64RecurrentConfig(encoder_dim=256, hidden_dim=736, num_heads=8, ffn_ratio=4, dropout=0.05, use_swiglu=True, use_qk_norm=True, zero_init_out_proj=True, gradient_checkpointing=False, prefix_layers=4, recurrent_layers=7, recurrent_unrolls=3, suffix_layers=4, policy_head_dim=384, value_hidden=384, n_value_classes=3, use_history=False, use_threat_head=False)
[21:10:01] params=98.97M
[21:10:01] FULL RESUME /root/transform/outputs/exp276_lichess_endgame_stream/latest.pt steps=200
[21:10:01] optimizer=PolarNorMuon (97.7M) + AdamW aux (1.3M) compile_polar=True
[21:10:01] optimizer state restored
[21:10:01] force_lr muon=0.002 adam=3e-05
[21:10:01] torch.compile enabled (mode=default); will fall back if inductor fails
[21:10:01] external eval lichess_endgame n=8,192 /root/transform/outputs/exp276_lichess_endgame_stream/lichess_eval.pt
[21:10:08] soft train=5,038,409 val=0 blocked=0 targets empty=0 unnorm=0 external_eval
[21:11:12] batch probe ok bs=528 peak_vram=35.48GB compile=True
[21:11:12] RNG state restored
[21:11:12] average_recurrent_grads() divides bank .grad by unroll count. That does NOT divide optimizer updates by the same amount: Polar-NorMuon is scale-invariant after orthogonalization, and RMS-Adam (m/sqrt(v)) is also approximately scale-invariant for a stationary grad. What does scale is SGD-style updates and the pre-clip global grad norm (so clipping fires less often after averaging). Use explicit param-group LRs if different update sizes are intended.
[21:11:12] exposure deep/shallow odds=0.0x (mix=0.0 deep_n=0 shallow_n=5038409)
[21:11:13] timing wall=0s active=0s steps=200 status=trained
[21:11:13] done steps=200 pos_s=0.0 params=98.97M status=trained kind=full
[21:11:29] trial=exp276_lichess_endgame_stream arch=squares64 device=cuda vocab=1968
[21:11:29] train bs=528 accum=1 grad_ckpt=False compile=True elo_every=0 save_every=150
[21:11:29] model=Squares64RecurrentConfig(encoder_dim=256, hidden_dim=736, num_heads=8, ffn_ratio=4, dropout=0.05, use_swiglu=True, use_qk_norm=True, zero_init_out_proj=True, gradient_checkpointing=False, prefix_layers=4, recurrent_layers=7, recurrent_unrolls=3, suffix_layers=4, policy_head_dim=384, value_hidden=384, n_value_classes=3, use_history=False, use_threat_head=False)
[21:11:30] params=98.97M
[21:11:30] FULL RESUME /root/transform/outputs/exp276_lichess_endgame_stream/latest.pt steps=200
[21:11:30] optimizer=PolarNorMuon (97.7M) + AdamW aux (1.3M) compile_polar=True
[21:11:30] optimizer state restored
[21:11:30] force_lr muon=0.002 adam=3e-05
[21:11:30] torch.compile enabled (mode=default); will fall back if inductor fails
[21:11:30] external eval lichess_endgame n=8,192 /root/transform/outputs/exp276_lichess_endgame_stream/lichess_eval.pt
[21:11:38] soft train=6,157,859 val=0 blocked=0 targets empty=0 unnorm=0 external_eval
[21:12:42] batch probe ok bs=528 peak_vram=35.48GB compile=True
[21:12:42] RNG state restored
[21:12:42] average_recurrent_grads() divides bank .grad by unroll count. That does NOT divide optimizer updates by the same amount: Polar-NorMuon is scale-invariant after orthogonalization, and RMS-Adam (m/sqrt(v)) is also approximately scale-invariant for a stationary grad. What does scale is SGD-style updates and the pre-clip global grad norm (so clipping fires less often after averaging). Use explicit param-group LRs if different update sizes are intended.
[21:12:42] exposure deep/shallow odds=0.0x (mix=0.0 deep_n=0 shallow_n=6157859)
[21:12:43] timing wall=0s active=0s steps=200 status=trained
[21:12:43] done steps=200 pos_s=0.0 params=98.97M status=trained kind=full
[21:13:04] trial=exp276_lichess_endgame_stream arch=squares64 device=cuda vocab=1968
[21:13:04] train bs=528 accum=1 grad_ckpt=False compile=True elo_every=0 save_every=150
[21:13:04] model=Squares64RecurrentConfig(encoder_dim=256, hidden_dim=736, num_heads=8, ffn_ratio=4, dropout=0.05, use_swiglu=True, use_qk_norm=True, zero_init_out_proj=True, gradient_checkpointing=False, prefix_layers=4, recurrent_layers=7, recurrent_unrolls=3, suffix_layers=4, policy_head_dim=384, value_hidden=384, n_value_classes=3, use_history=False, use_threat_head=False)
[21:13:05] params=98.97M
[21:13:05] FULL RESUME /root/transform/outputs/exp276_lichess_endgame_stream/latest.pt steps=200
[21:13:05] optimizer=PolarNorMuon (97.7M) + AdamW aux (1.3M) compile_polar=True
[21:13:05] optimizer state restored
[21:13:05] force_lr muon=0.002 adam=3e-05
[21:13:05] torch.compile enabled (mode=default); will fall back if inductor fails
[21:13:05] external eval lichess_endgame n=8,192 /root/transform/outputs/exp276_lichess_endgame_stream/lichess_eval.pt
[21:13:15] soft train=7,277,812 val=0 blocked=0 targets empty=0 unnorm=0 external_eval
[21:14:21] batch probe ok bs=528 peak_vram=35.48GB compile=True
[21:14:21] RNG state restored
[21:14:21] average_recurrent_grads() divides bank .grad by unroll count. That does NOT divide optimizer updates by the same amount: Polar-NorMuon is scale-invariant after orthogonalization, and RMS-Adam (m/sqrt(v)) is also approximately scale-invariant for a stationary grad. What does scale is SGD-style updates and the pre-clip global grad norm (so clipping fires less often after averaging). Use explicit param-group LRs if different update sizes are intended.
[21:14:21] exposure deep/shallow odds=0.0x (mix=0.0 deep_n=0 shallow_n=7277812)
[21:14:22] timing wall=0s active=0s steps=200 status=trained
[21:14:22] done steps=200 pos_s=0.0 params=98.97M status=trained kind=full
[21:14:48] trial=exp276_lichess_endgame_stream arch=squares64 device=cuda vocab=1968
[21:14:48] train bs=528 accum=1 grad_ckpt=False compile=True elo_every=0 save_every=150
[21:14:48] model=Squares64RecurrentConfig(encoder_dim=256, hidden_dim=736, num_heads=8, ffn_ratio=4, dropout=0.05, use_swiglu=True, use_qk_norm=True, zero_init_out_proj=True, gradient_checkpointing=False, prefix_layers=4, recurrent_layers=7, recurrent_unrolls=3, suffix_layers=4, policy_head_dim=384, value_hidden=384, n_value_classes=3, use_history=False, use_threat_head=False)
[21:14:48] params=98.97M
[21:14:48] FULL RESUME /root/transform/outputs/exp276_lichess_endgame_stream/latest.pt steps=200
[21:14:48] optimizer=PolarNorMuon (97.7M) + AdamW aux (1.3M) compile_polar=True
[21:14:49] optimizer state restored
[21:14:49] force_lr muon=0.002 adam=3e-05
[21:14:49] torch.compile enabled (mode=default); will fall back if inductor fails
[21:14:49] external eval lichess_endgame n=8,192 /root/transform/outputs/exp276_lichess_endgame_stream/lichess_eval.pt
[21:15:00] soft train=8,397,800 val=0 blocked=0 targets empty=0 unnorm=0 external_eval
[21:15:57] trial=exp276_lichess_endgame_stream arch=squares64 device=cuda vocab=1968
[21:15:57] train bs=528 accum=1 grad_ckpt=False compile=True elo_every=0 save_every=150
[21:15:57] model=Squares64RecurrentConfig(encoder_dim=256, hidden_dim=736, num_heads=8, ffn_ratio=4, dropout=0.05, use_swiglu=True, use_qk_norm=True, zero_init_out_proj=True, gradient_checkpointing=False, prefix_layers=4, recurrent_layers=7, recurrent_unrolls=3, suffix_layers=4, policy_head_dim=384, value_hidden=384, n_value_classes=3, use_history=False, use_threat_head=False)
[21:15:58] params=98.97M
[21:15:58] FULL RESUME /root/transform/outputs/exp276_lichess_endgame_stream/latest.pt steps=200
[21:15:59] optimizer=PolarNorMuon (97.7M) + AdamW aux (1.3M) compile_polar=True
[21:15:59] optimizer state restored
[21:15:59] force_lr muon=0.002 adam=3e-05
[21:15:59] torch.compile enabled (mode=default); will fall back if inductor fails
[21:15:59] external eval lichess_endgame n=8,192 /root/transform/outputs/exp276_lichess_endgame_stream/lichess_eval.pt
[21:16:10] soft train=8,397,800 val=0 blocked=0 targets empty=0 unnorm=0 external_eval
[21:16:31] soft inbox +1,699,973 train_n=10,097,773 shards=shard_000105,shard_000106,shard_000107,shard_000108,shard_000109,shard_000110,shard_000111,shard_000112,shard_000113,shard_000114,shard_000115,shard_000116,shard_000117,shard_000118,shard_000119,shard_000120,shard_000121
[21:17:37] batch probe ok bs=528 peak_vram=35.48GB compile=True
[21:17:37] RNG state restored
[21:17:37] average_recurrent_grads() divides bank .grad by unroll count. That does NOT divide optimizer updates by the same amount: Polar-NorMuon is scale-invariant after orthogonalization, and RMS-Adam (m/sqrt(v)) is also approximately scale-invariant for a stationary grad. What does scale is SGD-style updates and the pre-clip global grad norm (so clipping fires less often after averaging). Use explicit param-group LRs if different update sizes are intended.
[21:17:37] exposure deep/shallow odds=0.0x (mix=0.0 deep_n=0 shallow_n=10097773)
[21:18:02] step 225/8000 | loss=1.5003 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=13200/0/0 q=0 p=0 | 520 pos/s | vram=35.48GB
[21:18:11] soft inbox +1,399,980 train_n=11,497,753 shards=shard_000122,shard_000123,shard_000124,shard_000125,shard_000126,shard_000127,shard_000128,shard_000129,shard_000130,shard_000131,shard_000132,shard_000133,shard_000134,shard_000135
[21:18:34] step 250/8000 | loss=1.4967 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=26400/0/0 q=0 p=0 | 461 pos/s | vram=35.48GB
[21:18:37] soft inbox +399,995 train_n=11,897,748 shards=shard_000136,shard_000137,shard_000138,shard_000139
[21:19:00] step 275/8000 | loss=1.4977 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=39600/0/0 q=0 p=0 | 475 pos/s | vram=35.48GB
[21:19:03] soft inbox +299,995 train_n=12,197,743 shards=shard_000140,shard_000141,shard_000142
[21:19:26] step 300/8000 | loss=1.4858 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=52800/0/0 q=0 p=0 | 482 pos/s | vram=35.48GB
[21:19:29] soft inbox +399,999 train_n=12,597,742 shards=shard_000143,shard_000144,shard_000145,shard_000146
[21:19:30] ckpt step_000300.pt + latest.pt (live) steps=300 status=mid kind=full
[21:19:40] val/lichess_endgame hard_ce=1.3408 soft_ce=1.3408 soft_temp_ce=1.3408 teacher_entropy=0.0000 teacher_kl=1.3408 wdl_ce=0.4994 value_rows=8192.0000
[21:20:02] step 325/8000 | loss=1.5038 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=66000/0/0 q=0 p=0 | 453 pos/s | vram=35.48GB
[21:20:07] soft inbox +499,996 train_n=13,097,738 shards=shard_000147,shard_000148,shard_000149,shard_000150,shard_000151
[21:20:30] step 350/8000 | loss=1.4888 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=79200/0/0 q=0 p=0 | 456 pos/s | vram=35.48GB
[21:20:34] soft inbox +299,983 train_n=13,397,721 shards=shard_000152,shard_000153,shard_000154
[21:20:57] step 375/8000 | loss=1.4981 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=92400/0/0 q=0 p=0 | 461 pos/s | vram=35.48GB
[21:21:01] soft inbox +399,991 train_n=13,797,712 shards=shard_000155,shard_000156,shard_000157,shard_000158
[21:21:24] step 400/8000 | loss=1.4966 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=105600/0/0 q=0 p=0 | 463 pos/s | vram=35.48GB
[21:21:28] soft inbox +399,997 train_n=14,197,709 shards=shard_000159,shard_000160,shard_000161,shard_000162
[21:21:38] val/lichess_endgame hard_ce=1.3393 soft_ce=1.3393 soft_temp_ce=1.3393 teacher_entropy=0.0000 teacher_kl=1.3393 wdl_ce=0.4999 value_rows=8192.0000
[21:22:01] step 425/8000 | loss=1.4800 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=118800/0/0 q=0 p=0 | 450 pos/s | vram=35.48GB
[21:22:05] soft inbox +399,995 train_n=14,597,704 shards=shard_000163,shard_000164,shard_000165,shard_000166
[21:22:28] step 450/8000 | loss=1.4744 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=132000/0/0 q=0 p=0 | 453 pos/s | vram=35.48GB
[21:22:33] soft inbox +399,994 train_n=14,997,698 shards=shard_000167,shard_000168,shard_000169,shard_000170
[21:22:33] ckpt latest.pt (live weights) steps=450 status=mid
[21:22:56] step 475/8000 | loss=1.4720 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=145200/0/0 q=0 p=0 | 454 pos/s | vram=35.48GB
[21:23:01] soft inbox +399,996 train_n=15,397,694 shards=shard_000171,shard_000172,shard_000173,shard_000174
[21:23:24] step 500/8000 | loss=1.4852 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=158400/0/0 q=0 p=0 | 456 pos/s | vram=35.48GB
[21:23:28] soft inbox +299,991 train_n=15,697,685 shards=shard_000175,shard_000176,shard_000177
[21:23:37] val/lichess_endgame hard_ce=1.3362 soft_ce=1.3362 soft_temp_ce=1.3362 teacher_entropy=0.0000 teacher_kl=1.3362 wdl_ce=0.4941 value_rows=8192.0000
[21:24:00] step 525/8000 | loss=1.4781 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=171600/0/0 q=0 p=0 | 447 pos/s | vram=35.48GB
[21:24:05] soft inbox +499,991 train_n=16,197,676 shards=shard_000178,shard_000179,shard_000180,shard_000181,shard_000182
[21:24:28] step 550/8000 | loss=1.4783 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=184800/0/0 q=0 p=0 | 449 pos/s | vram=35.48GB
[21:24:33] soft inbox +399,995 train_n=16,597,671 shards=shard_000183,shard_000184,shard_000185,shard_000186
[21:24:55] step 575/8000 | loss=1.4688 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=198000/0/0 q=0 p=0 | 451 pos/s | vram=35.48GB
[21:25:00] soft inbox +399,997 train_n=16,997,668 shards=shard_000187,shard_000188,shard_000189,shard_000190
[21:25:23] step 600/8000 | loss=1.4692 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=211200/0/0 q=0 p=0 | 453 pos/s | vram=35.48GB
[21:25:28] soft inbox +399,995 train_n=17,397,663 shards=shard_000191,shard_000192,shard_000193,shard_000194
[21:25:29] ckpt step_000600.pt + latest.pt (live) steps=600 status=mid kind=full
[21:25:38] val/lichess_endgame hard_ce=1.3364 soft_ce=1.3364 soft_temp_ce=1.3364 teacher_entropy=0.0000 teacher_kl=1.3364 wdl_ce=0.4928 value_rows=8192.0000
[21:26:01] step 625/8000 | loss=1.4948 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=224400/0/0 q=0 p=0 | 445 pos/s | vram=35.48GB
[21:26:06] soft inbox +499,996 train_n=17,897,659 shards=shard_000195,shard_000196,shard_000197,shard_000198,shard_000199
[21:26:29] step 650/8000 | loss=1.4796 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=237600/0/0 q=0 p=0 | 446 pos/s | vram=35.48GB
[21:26:34] soft inbox +399,996 train_n=18,297,655 shards=shard_000200,shard_000201,shard_000202,shard_000203
[21:26:57] step 675/8000 | loss=1.4832 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=250800/0/0 q=0 p=0 | 447 pos/s | vram=35.48GB
[21:27:03] soft inbox +399,994 train_n=18,697,649 shards=shard_000204,shard_000205,shard_000206,shard_000207
[21:27:26] step 700/8000 | loss=1.4729 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=264000/0/0 q=0 p=0 | 448 pos/s | vram=35.48GB
[21:27:30] soft inbox +299,995 train_n=18,997,644 shards=shard_000208,shard_000209,shard_000210
[21:27:39] val/lichess_endgame hard_ce=1.3283 soft_ce=1.3283 soft_temp_ce=1.3283 teacher_entropy=0.0000 teacher_kl=1.3283 wdl_ce=0.4939 value_rows=8192.0000
[21:28:02] step 725/8000 | loss=1.4744 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=277200/0/0 q=0 p=0 | 443 pos/s | vram=35.48GB
[21:28:08] soft inbox +499,995 train_n=19,497,639 shards=shard_000211,shard_000212,shard_000213,shard_000214,shard_000215
[21:28:31] step 750/8000 | loss=1.4761 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=290400/0/0 q=0 p=0 | 444 pos/s | vram=35.48GB
[21:28:36] soft inbox +299,990 train_n=19,797,629 shards=shard_000216,shard_000217,shard_000218
[21:28:37] ckpt latest.pt (live weights) steps=750 status=mid
[21:28:59] step 775/8000 | loss=1.4734 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=303600/0/0 q=0 p=0 | 445 pos/s | vram=35.48GB
[21:29:05] soft inbox +399,996 train_n=20,197,625 shards=shard_000219,shard_000220,shard_000221,shard_000222
[21:29:28] step 800/8000 | loss=1.4568 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=316800/0/0 q=0 p=0 | 445 pos/s | vram=35.48GB
[21:29:32] soft inbox +299,990 train_n=20,497,615 shards=shard_000223,shard_000224,shard_000225
[21:29:41] val/lichess_endgame hard_ce=1.3278 soft_ce=1.3278 soft_temp_ce=1.3278 teacher_entropy=0.0000 teacher_kl=1.3278 wdl_ce=0.4913 value_rows=8192.0000
[21:30:05] step 825/8000 | loss=1.4914 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=330000/0/0 q=0 p=0 | 441 pos/s | vram=35.48GB
[21:30:12] soft inbox +499,991 train_n=20,997,606 shards=shard_000226,shard_000227,shard_000228,shard_000229,shard_000230
[21:30:35] step 850/8000 | loss=1.4786 | lr=2.00e-03,3.00e-05 | clip=1/25 | mix s/d/b=343200/0/0 q=0 p=0 | 441 pos/s | vram=35.48GB
[21:30:40] soft inbox +399,989 train_n=21,397,595 shards=shard_000231,shard_000232,shard_000233,shard_000234
[21:31:03] step 875/8000 | loss=1.4866 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=356400/0/0 q=0 p=0 | 442 pos/s | vram=35.48GB
[21:31:09] soft inbox +299,994 train_n=21,697,589 shards=shard_000235,shard_000236,shard_000237
[21:31:32] step 900/8000 | loss=1.4688 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=369600/0/0 q=0 p=0 | 443 pos/s | vram=35.48GB
[21:31:37] soft inbox +399,989 train_n=22,097,578 shards=shard_000238,shard_000239,shard_000240,shard_000241
[21:31:39] ckpt step_000900.pt + latest.pt (live) steps=900 status=mid kind=full
[21:31:48] val/lichess_endgame hard_ce=1.3244 soft_ce=1.3244 soft_temp_ce=1.3244 teacher_entropy=0.0000 teacher_kl=1.3244 wdl_ce=0.4960 value_rows=8192.0000
[21:32:11] step 925/8000 | loss=1.4821 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=382800/0/0 q=0 p=0 | 438 pos/s | vram=35.48GB
[21:32:18] soft inbox +499,993 train_n=22,597,571 shards=shard_000242,shard_000243,shard_000244,shard_000245,shard_000246
[21:32:41] step 950/8000 | loss=1.4776 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=396000/0/0 q=0 p=0 | 438 pos/s | vram=35.48GB
[21:32:47] soft inbox +399,996 train_n=22,997,567 shards=shard_000247,shard_000248,shard_000249,shard_000250
[21:33:10] step 975/8000 | loss=1.4747 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=409200/0/0 q=0 p=0 | 438 pos/s | vram=35.48GB
[21:33:15] soft inbox +299,990 train_n=23,297,557 shards=shard_000251,shard_000252,shard_000253
[21:33:38] step 1000/8000 | loss=1.4823 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=422400/0/0 q=0 p=0 | 439 pos/s | vram=35.48GB
[21:33:44] soft inbox +399,993 train_n=23,697,550 shards=shard_000254,shard_000255,shard_000256,shard_000257
[21:33:53] val/lichess_endgame hard_ce=1.3237 soft_ce=1.3237 soft_temp_ce=1.3237 teacher_entropy=0.0000 teacher_kl=1.3237 wdl_ce=0.4906 value_rows=8192.0000
[21:34:16] step 1025/8000 | loss=1.4635 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=435600/0/0 q=0 p=0 | 436 pos/s | vram=35.48GB
[21:34:25] soft inbox +499,991 train_n=24,197,541 shards=shard_000258,shard_000259,shard_000260,shard_000261,shard_000262
[21:34:48] step 1050/8000 | loss=1.4874 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=448800/0/0 q=0 p=0 | 435 pos/s | vram=35.48GB
[21:34:55] soft inbox +399,994 train_n=24,597,535 shards=shard_000263,shard_000264,shard_000265,shard_000266
[21:34:56] ckpt latest.pt (live weights) steps=1050 status=mid
[21:35:19] step 1075/8000 | loss=1.4603 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=462000/0/0 q=0 p=0 | 435 pos/s | vram=35.48GB
[21:35:26] soft inbox +399,995 train_n=24,997,530 shards=shard_000267,shard_000268,shard_000269,shard_000270
[21:35:49] step 1100/8000 | loss=1.4514 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=475200/0/0 q=0 p=0 | 435 pos/s | vram=35.48GB
[21:35:55] soft inbox +399,988 train_n=25,397,518 shards=shard_000271,shard_000272,shard_000273,shard_000274
[21:36:04] val/lichess_endgame hard_ce=1.3243 soft_ce=1.3243 soft_temp_ce=1.3243 teacher_entropy=0.0000 teacher_kl=1.3243 wdl_ce=0.4970 value_rows=8192.0000
[21:36:27] step 1125/8000 | loss=1.4785 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=488400/0/0 q=0 p=0 | 432 pos/s | vram=35.48GB
[21:36:36] soft inbox +499,993 train_n=25,897,511 shards=shard_000275,shard_000276,shard_000277,shard_000278,shard_000279
[21:36:59] step 1150/8000 | loss=1.4581 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=501600/0/0 q=0 p=0 | 432 pos/s | vram=35.48GB
[21:37:05] soft inbox +399,981 train_n=26,297,492 shards=shard_000280,shard_000281,shard_000282,shard_000283
[21:37:28] step 1175/8000 | loss=1.4827 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=514800/0/0 q=0 p=0 | 432 pos/s | vram=35.48GB
[21:37:35] soft inbox +299,998 train_n=26,597,490 shards=shard_000284,shard_000285,shard_000286
[21:37:57] step 1200/8000 | loss=1.4782 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=528000/0/0 q=0 p=0 | 432 pos/s | vram=35.48GB
[21:38:04] soft inbox +399,995 train_n=26,997,485 shards=shard_000287,shard_000288,shard_000289,shard_000290
[21:38:05] ckpt step_001200.pt + latest.pt (live) steps=1200 status=mid kind=full
[21:38:14] val/lichess_endgame hard_ce=1.3185 soft_ce=1.3185 soft_temp_ce=1.3185 teacher_entropy=0.0000 teacher_kl=1.3185 wdl_ce=0.4910 value_rows=8192.0000
[21:38:37] step 1225/8000 | loss=1.4682 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=541200/0/0 q=0 p=0 | 429 pos/s | vram=35.48GB
[21:38:48] soft inbox +599,983 train_n=27,597,468 shards=shard_000291,shard_000292,shard_000293,shard_000294,shard_000295,shard_000296
[21:39:11] step 1250/8000 | loss=1.4693 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=554400/0/0 q=0 p=0 | 428 pos/s | vram=35.48GB
[21:39:17] soft inbox +299,997 train_n=27,897,465 shards=shard_000297,shard_000298,shard_000299
[21:39:40] step 1275/8000 | loss=1.4657 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=567600/0/0 q=0 p=0 | 429 pos/s | vram=35.48GB
[21:39:49] soft inbox +499,993 train_n=28,397,458 shards=shard_000300,shard_000301,shard_000302,shard_000303,shard_000304
[21:40:12] step 1300/8000 | loss=1.4687 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=580800/0/0 q=0 p=0 | 428 pos/s | vram=35.48GB
[21:40:21] soft inbox +499,989 train_n=28,897,447 shards=shard_000305,shard_000306,shard_000307,shard_000308,shard_000309
[21:40:30] val/lichess_endgame hard_ce=1.3203 soft_ce=1.3203 soft_temp_ce=1.3203 teacher_entropy=0.0000 teacher_kl=1.3203 wdl_ce=0.4962 value_rows=8192.0000
[21:40:53] step 1325/8000 | loss=1.4782 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=594000/0/0 q=0 p=0 | 425 pos/s | vram=35.48GB
[21:41:04] soft inbox +599,990 train_n=29,497,437 shards=shard_000310,shard_000311,shard_000312,shard_000313,shard_000314,shard_000315
[21:41:27] step 1350/8000 | loss=1.4682 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=607200/0/0 q=0 p=0 | 425 pos/s | vram=35.48GB
[21:41:36] soft inbox +499,990 train_n=29,997,427 shards=shard_000316,shard_000317,shard_000318,shard_000319,shard_000320
[21:41:36] ckpt latest.pt (live weights) steps=1350 status=mid
[21:41:59] step 1375/8000 | loss=1.4802 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=620400/0/0 q=0 p=0 | 424 pos/s | vram=35.48GB
[21:42:09] soft inbox +499,991 train_n=30,497,418 shards=shard_000321,shard_000322,shard_000323,shard_000324,shard_000325
[21:42:32] step 1400/8000 | loss=1.4638 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=633600/0/0 q=0 p=0 | 424 pos/s | vram=35.48GB
[21:42:41] soft inbox +499,988 train_n=30,997,406 shards=shard_000326,shard_000327,shard_000328,shard_000329,shard_000330
[21:42:50] val/lichess_endgame hard_ce=1.3201 soft_ce=1.3201 soft_temp_ce=1.3201 teacher_entropy=0.0000 teacher_kl=1.3201 wdl_ce=0.4905 value_rows=8192.0000
[21:43:13] step 1425/8000 | loss=1.4471 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=646800/0/0 q=0 p=0 | 421 pos/s | vram=35.48GB
[21:43:26] soft inbox +699,989 train_n=31,697,395 shards=shard_000331,shard_000332,shard_000333,shard_000334,shard_000335,shard_000336,shard_000337
[21:43:49] step 1450/8000 | loss=1.4802 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=660000/0/0 q=0 p=0 | 420 pos/s | vram=35.48GB
[21:43:59] soft inbox +499,990 train_n=32,197,385 shards=shard_000338,shard_000339,shard_000340,shard_000341,shard_000342
[21:44:22] step 1475/8000 | loss=1.4811 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=673200/0/0 q=0 p=0 | 419 pos/s | vram=35.48GB
[21:44:33] soft inbox +499,987 train_n=32,697,372 shards=shard_000343,shard_000344,shard_000345,shard_000346,shard_000347
[21:44:57] step 1500/8000 | loss=1.4744 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=686400/0/0 q=0 p=0 | 419 pos/s | vram=35.48GB
[21:45:07] soft inbox +499,993 train_n=33,197,365 shards=shard_000348,shard_000349,shard_000350,shard_000351,shard_000352
[21:45:08] ckpt step_001500.pt + latest.pt (live) steps=1500 status=mid kind=full
[21:45:17] val/lichess_endgame hard_ce=1.3159 soft_ce=1.3159 soft_temp_ce=1.3159 teacher_entropy=0.0000 teacher_kl=1.3159 wdl_ce=0.4938 value_rows=8192.0000
[21:45:40] step 1525/8000 | loss=1.4788 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=699600/0/0 q=0 p=0 | 416 pos/s | vram=35.48GB
[21:45:54] soft inbox +699,986 train_n=33,897,351 shards=shard_000353,shard_000354,shard_000355,shard_000356,shard_000357,shard_000358,shard_000359
[21:46:17] step 1550/8000 | loss=1.4834 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=712800/0/0 q=0 p=0 | 414 pos/s | vram=35.48GB
[21:46:29] soft inbox +499,994 train_n=34,397,345 shards=shard_000360,shard_000361,shard_000362,shard_000363,shard_000364
[21:46:52] step 1575/8000 | loss=1.4847 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=726000/0/0 q=0 p=0 | 414 pos/s | vram=35.48GB
[21:47:03] soft inbox +499,994 train_n=34,897,339 shards=shard_000365,shard_000366,shard_000367,shard_000368,shard_000369
[21:47:26] step 1600/8000 | loss=1.4977 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=739200/0/0 q=0 p=0 | 413 pos/s | vram=35.48GB
[21:47:37] soft inbox +499,990 train_n=35,397,329 shards=shard_000370,shard_000371,shard_000372,shard_000373,shard_000374
[21:47:46] val/lichess_endgame hard_ce=1.3136 soft_ce=1.3136 soft_temp_ce=1.3136 teacher_entropy=0.0000 teacher_kl=1.3136 wdl_ce=0.4966 value_rows=8192.0000
[21:48:09] step 1625/8000 | loss=1.4548 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=752400/0/0 q=0 p=0 | 411 pos/s | vram=35.48GB
[21:48:22] soft inbox +599,986 train_n=35,997,315 shards=shard_000375,shard_000376,shard_000377,shard_000378,shard_000379,shard_000380
[21:48:45] step 1650/8000 | loss=1.4725 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=765600/0/0 q=0 p=0 | 410 pos/s | vram=35.48GB
[21:48:55] soft inbox +399,987 train_n=36,397,302 shards=shard_000381,shard_000382,shard_000383,shard_000384
[21:48:55] ckpt latest.pt (live weights) steps=1650 status=mid
[21:49:18] step 1675/8000 | loss=1.4725 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=778800/0/0 q=0 p=0 | 410 pos/s | vram=35.48GB
[21:49:28] soft inbox +399,991 train_n=36,797,293 shards=shard_000385,shard_000386,shard_000387,shard_000388
[21:49:51] step 1700/8000 | loss=1.4556 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=792000/0/0 q=0 p=0 | 409 pos/s | vram=35.48GB
[21:50:01] soft inbox +399,992 train_n=37,197,285 shards=shard_000389,shard_000390,shard_000391,shard_000392
[21:50:10] val/lichess_endgame hard_ce=1.3109 soft_ce=1.3109 soft_temp_ce=1.3109 teacher_entropy=0.0000 teacher_kl=1.3109 wdl_ce=0.4899 value_rows=8192.0000
[21:50:33] step 1725/8000 | loss=1.4756 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=805200/0/0 q=0 p=0 | 407 pos/s | vram=35.48GB
[21:50:47] soft inbox +599,982 train_n=37,797,267 shards=shard_000393,shard_000394,shard_000395,shard_000396,shard_000397,shard_000398
[21:51:10] step 1750/8000 | loss=1.4636 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=818400/0/0 q=0 p=0 | 406 pos/s | vram=35.48GB
[21:51:21] soft inbox +399,993 train_n=38,197,260 shards=shard_000399,shard_000400,shard_000401,shard_000402
[21:51:44] step 1775/8000 | loss=1.4761 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=831600/0/0 q=0 p=0 | 406 pos/s | vram=35.48GB
[21:51:54] soft inbox +399,992 train_n=38,597,252 shards=shard_000403,shard_000404,shard_000405,shard_000406
[21:52:17] step 1800/8000 | loss=1.4820 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=844800/0/0 q=0 p=0 | 406 pos/s | vram=35.48GB
[21:52:26] soft inbox +299,997 train_n=38,897,249 shards=shard_000407,shard_000408,shard_000409
[21:52:27] ckpt step_001800.pt + latest.pt (live) steps=1800 status=mid kind=full
[21:52:36] val/lichess_endgame hard_ce=1.3087 soft_ce=1.3087 soft_temp_ce=1.3087 teacher_entropy=0.0000 teacher_kl=1.3087 wdl_ce=0.4917 value_rows=8192.0000
[21:52:59] step 1825/8000 | loss=1.4828 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=858000/0/0 q=0 p=0 | 404 pos/s | vram=35.48GB
[21:53:11] soft inbox +499,981 train_n=39,397,230 shards=shard_000410,shard_000411,shard_000412,shard_000413,shard_000414
[21:53:34] step 1850/8000 | loss=1.4755 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=871200/0/0 q=0 p=0 | 404 pos/s | vram=35.48GB
[21:53:44] soft inbox +399,987 train_n=39,797,217 shards=shard_000415,shard_000416,shard_000417,shard_000418
[21:54:08] step 1875/8000 | loss=1.4838 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=884400/0/0 q=0 p=0 | 404 pos/s | vram=35.48GB
[21:54:18] soft inbox +399,986 train_n=40,197,203 shards=shard_000419,shard_000420,shard_000421,shard_000422
[21:54:41] step 1900/8000 | loss=1.4666 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=897600/0/0 q=0 p=0 | 403 pos/s | vram=35.48GB
[21:54:50] soft inbox +299,994 train_n=40,497,197 shards=shard_000423,shard_000424,shard_000425
[21:54:59] val/lichess_endgame hard_ce=1.3101 soft_ce=1.3101 soft_temp_ce=1.3101 teacher_entropy=0.0000 teacher_kl=1.3101 wdl_ce=0.4907 value_rows=8192.0000
[21:55:22] step 1925/8000 | loss=1.4851 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=910800/0/0 q=0 p=0 | 402 pos/s | vram=35.48GB
[21:55:35] soft inbox +499,984 train_n=40,997,181 shards=shard_000426,shard_000427,shard_000428,shard_000429,shard_000430
[21:55:58] step 1950/8000 | loss=1.4773 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=924000/0/0 q=0 p=0 | 401 pos/s | vram=35.48GB
[21:56:09] soft inbox +399,993 train_n=41,397,174 shards=shard_000431,shard_000432,shard_000433,shard_000434
[21:56:10] ckpt latest.pt (live weights) steps=1950 status=mid
[21:56:33] step 1975/8000 | loss=1.4640 | lr=2.00e-03,3.00e-05 | clip=1/25 | mix s/d/b=937200/0/0 q=0 p=0 | 401 pos/s | vram=35.48GB
[21:56:43] soft inbox +299,995 train_n=41,697,169 shards=shard_000435,shard_000436,shard_000437
[21:57:06] step 2000/8000 | loss=1.4644 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=950400/0/0 q=0 p=0 | 401 pos/s | vram=35.48GB
[21:57:17] soft inbox +399,988 train_n=42,097,157 shards=shard_000438,shard_000439,shard_000440,shard_000441
[21:57:26] val/lichess_endgame hard_ce=1.3058 soft_ce=1.3058 soft_temp_ce=1.3058 teacher_entropy=0.0000 teacher_kl=1.3058 wdl_ce=0.4925 value_rows=8192.0000
[21:57:50] step 2025/8000 | loss=1.4829 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=963600/0/0 q=0 p=0 | 399 pos/s | vram=35.48GB
[21:58:01] soft inbox +399,989 train_n=42,497,146 shards=shard_000442,shard_000443,shard_000444,shard_000445
[21:58:24] step 2050/8000 | loss=1.4747 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=976800/0/0 q=0 p=0 | 399 pos/s | vram=35.48GB
[21:58:35] soft inbox +399,989 train_n=42,897,135 shards=shard_000446,shard_000447,shard_000448,shard_000449
[21:58:58] step 2075/8000 | loss=1.4873 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=990000/0/0 q=0 p=0 | 399 pos/s | vram=35.48GB
[21:59:08] soft inbox +299,995 train_n=43,197,130 shards=shard_000450,shard_000451,shard_000452
[21:59:31] step 2100/8000 | loss=1.4693 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1003200/0/0 q=0 p=0 | 399 pos/s | vram=35.48GB
[21:59:42] soft inbox +399,990 train_n=43,597,120 shards=shard_000453,shard_000454,shard_000455,shard_000456
[21:59:44] ckpt step_002100.pt + latest.pt (live) steps=2100 status=mid kind=full
[21:59:53] val/lichess_endgame hard_ce=1.3098 soft_ce=1.3098 soft_temp_ce=1.3098 teacher_entropy=0.0000 teacher_kl=1.3098 wdl_ce=0.4897 value_rows=8192.0000
[22:00:16] step 2125/8000 | loss=1.5057 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1016400/0/0 q=0 p=0 | 397 pos/s | vram=35.48GB
[22:00:28] soft inbox +399,991 train_n=43,997,111 shards=shard_000457,shard_000458,shard_000459,shard_000460
[22:00:51] step 2150/8000 | loss=1.4808 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1029600/0/0 q=0 p=0 | 397 pos/s | vram=35.48GB
[22:01:03] soft inbox +399,994 train_n=44,397,105 shards=shard_000461,shard_000462,shard_000463,shard_000464
[22:01:26] step 2175/8000 | loss=1.4826 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1042800/0/0 q=0 p=0 | 397 pos/s | vram=35.48GB
[22:01:38] soft inbox +399,983 train_n=44,797,088 shards=shard_000465,shard_000466,shard_000467,shard_000468
[22:02:01] step 2200/8000 | loss=1.4672 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1056000/0/0 q=0 p=0 | 396 pos/s | vram=35.48GB
[22:02:12] soft inbox +299,987 train_n=45,097,075 shards=shard_000469,shard_000470,shard_000471
[22:02:21] val/lichess_endgame hard_ce=1.3090 soft_ce=1.3090 soft_temp_ce=1.3090 teacher_entropy=0.0000 teacher_kl=1.3090 wdl_ce=0.4886 value_rows=8192.0000
[22:02:44] step 2225/8000 | loss=1.5002 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1069200/0/0 q=0 p=0 | 395 pos/s | vram=35.48GB
[22:02:59] soft inbox +499,977 train_n=45,597,052 shards=shard_000472,shard_000473,shard_000474,shard_000475,shard_000476
[22:03:22] step 2250/8000 | loss=1.4727 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1082400/0/0 q=0 p=0 | 394 pos/s | vram=35.48GB
[22:03:33] soft inbox +299,986 train_n=45,897,038 shards=shard_000477,shard_000478,shard_000479
[22:03:34] ckpt latest.pt (live weights) steps=2250 status=mid
[22:03:57] step 2275/8000 | loss=1.4772 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1095600/0/0 q=0 p=0 | 394 pos/s | vram=35.48GB
[22:04:10] soft inbox +399,989 train_n=46,297,027 shards=shard_000480,shard_000481,shard_000482,shard_000483
[22:04:33] step 2300/8000 | loss=1.4756 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1108800/0/0 q=0 p=0 | 394 pos/s | vram=35.48GB
[22:04:44] soft inbox +299,993 train_n=46,597,020 shards=shard_000484,shard_000485,shard_000486
[22:04:53] val/lichess_endgame hard_ce=1.3093 soft_ce=1.3093 soft_temp_ce=1.3093 teacher_entropy=0.0000 teacher_kl=1.3093 wdl_ce=0.4911 value_rows=8192.0000
[22:05:16] step 2325/8000 | loss=1.4662 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1122000/0/0 q=0 p=0 | 392 pos/s | vram=35.48GB
[22:05:31] soft inbox +499,984 train_n=47,097,004 shards=shard_000487,shard_000488,shard_000489,shard_000490,shard_000491
[22:05:54] step 2350/8000 | loss=1.4922 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1135200/0/0 q=0 p=0 | 392 pos/s | vram=35.48GB
[22:06:08] soft inbox +399,986 train_n=47,496,990 shards=shard_000492,shard_000493,shard_000494,shard_000495
[22:06:32] step 2375/8000 | loss=1.4846 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1148400/0/0 q=0 p=0 | 391 pos/s | vram=35.48GB
[22:06:43] soft inbox +299,982 train_n=47,796,972 shards=shard_000496,shard_000497,shard_000498
[22:07:06] step 2400/8000 | loss=1.4772 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1161600/0/0 q=0 p=0 | 391 pos/s | vram=35.48GB
[22:07:19] soft inbox +399,992 train_n=48,196,964 shards=shard_000499,shard_000500,shard_000501,shard_000502
[22:07:20] ckpt step_002400.pt + latest.pt (live) steps=2400 status=mid kind=full
[22:07:29] val/lichess_endgame hard_ce=1.3120 soft_ce=1.3120 soft_temp_ce=1.3120 teacher_entropy=0.0000 teacher_kl=1.3120 wdl_ce=0.4898 value_rows=8192.0000
[22:07:52] step 2425/8000 | loss=1.4758 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1174800/0/0 q=0 p=0 | 390 pos/s | vram=35.48GB
[22:08:07] soft inbox +499,989 train_n=48,696,953 shards=shard_000503,shard_000504,shard_000505,shard_000506,shard_000507
[22:08:31] step 2450/8000 | loss=1.4903 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1188000/0/0 q=0 p=0 | 389 pos/s | vram=35.48GB
[22:08:42] soft inbox +299,994 train_n=48,996,947 shards=shard_000508,shard_000509,shard_000510
[22:09:06] step 2475/8000 | loss=1.4764 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1201200/0/0 q=0 p=0 | 389 pos/s | vram=35.48GB
[22:09:20] soft inbox +399,986 train_n=49,396,933 shards=shard_000511,shard_000512,shard_000513,shard_000514
[22:09:43] step 2500/8000 | loss=1.5061 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1214400/0/0 q=0 p=0 | 388 pos/s | vram=35.48GB
[22:09:55] soft inbox +299,995 train_n=49,696,928 shards=shard_000515,shard_000516,shard_000517
[22:10:04] val/lichess_endgame hard_ce=1.3069 soft_ce=1.3069 soft_temp_ce=1.3069 teacher_entropy=0.0000 teacher_kl=1.3069 wdl_ce=0.4900 value_rows=8192.0000
[22:10:27] step 2525/8000 | loss=1.4698 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1227600/0/0 q=0 p=0 | 387 pos/s | vram=35.48GB
[22:10:41] soft inbox +399,983 train_n=50,096,911 shards=shard_000518,shard_000519,shard_000520,shard_000521
[22:11:04] step 2550/8000 | loss=1.4712 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1240800/0/0 q=0 p=0 | 387 pos/s | vram=35.48GB
[22:11:18] soft inbox +399,995 train_n=50,496,906 shards=shard_000522,shard_000523,shard_000524,shard_000525
[22:11:19] ckpt latest.pt (live weights) steps=2550 status=mid
[22:11:42] step 2575/8000 | loss=1.4717 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1254000/0/0 q=0 p=0 | 386 pos/s | vram=35.48GB
[22:11:56] soft inbox +399,985 train_n=50,896,891 shards=shard_000526,shard_000527,shard_000528,shard_000529
[22:12:19] step 2600/8000 | loss=1.4736 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1267200/0/0 q=0 p=0 | 386 pos/s | vram=35.48GB
[22:12:31] soft inbox +299,995 train_n=51,196,886 shards=shard_000530,shard_000531,shard_000532
[22:12:40] val/lichess_endgame hard_ce=1.3045 soft_ce=1.3045 soft_temp_ce=1.3045 teacher_entropy=0.0000 teacher_kl=1.3045 wdl_ce=0.4906 value_rows=8192.0000
[22:13:04] step 2625/8000 | loss=1.4944 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1280400/0/0 q=0 p=0 | 385 pos/s | vram=35.48GB
[22:13:19] soft inbox +399,990 train_n=51,596,876 shards=shard_000533,shard_000534,shard_000535,shard_000536
[22:13:42] step 2650/8000 | loss=1.4779 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1293600/0/0 q=0 p=0 | 384 pos/s | vram=35.48GB
[22:13:57] soft inbox +399,996 train_n=51,996,872 shards=shard_000537,shard_000538,shard_000539,shard_000540
[22:14:20] step 2675/8000 | loss=1.4849 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1306800/0/0 q=0 p=0 | 384 pos/s | vram=35.48GB
[22:14:33] soft inbox +299,991 train_n=52,296,863 shards=shard_000541,shard_000542,shard_000543
[22:14:56] step 2700/8000 | loss=1.4739 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1320000/0/0 q=0 p=0 | 384 pos/s | vram=35.48GB
[22:15:09] soft inbox +299,994 train_n=52,596,857 shards=shard_000544,shard_000545,shard_000546
[22:15:10] ckpt step_002700.pt + latest.pt (live) steps=2700 status=mid kind=full
[22:15:19] val/lichess_endgame hard_ce=1.3054 soft_ce=1.3054 soft_temp_ce=1.3054 teacher_entropy=0.0000 teacher_kl=1.3054 wdl_ce=0.4873 value_rows=8192.0000
[22:15:42] step 2725/8000 | loss=1.4860 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1333200/0/0 q=0 p=0 | 382 pos/s | vram=35.48GB
[22:15:59] soft inbox +499,985 train_n=53,096,842 shards=shard_000547,shard_000548,shard_000549,shard_000550,shard_000551
[22:16:23] step 2750/8000 | loss=1.4926 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1346400/0/0 q=0 p=0 | 382 pos/s | vram=35.48GB
[22:16:35] soft inbox +299,982 train_n=53,396,824 shards=shard_000552,shard_000553,shard_000554
[22:16:58] step 2775/8000 | loss=1.4805 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1359600/0/0 q=0 p=0 | 382 pos/s | vram=35.48GB
[22:17:13] soft inbox +399,984 train_n=53,796,808 shards=shard_000555,shard_000556,shard_000557,shard_000558
[22:17:36] step 2800/8000 | loss=1.4744 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1372800/0/0 q=0 p=0 | 381 pos/s | vram=35.48GB
[22:17:49] soft inbox +299,987 train_n=54,096,795 shards=shard_000559,shard_000560,shard_000561
[22:17:58] val/lichess_endgame hard_ce=1.3015 soft_ce=1.3015 soft_temp_ce=1.3015 teacher_entropy=0.0000 teacher_kl=1.3015 wdl_ce=0.4904 value_rows=8192.0000
[22:18:21] step 2825/8000 | loss=1.4732 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1386000/0/0 q=0 p=0 | 380 pos/s | vram=35.48GB
[22:18:36] soft inbox +399,988 train_n=54,496,783 shards=shard_000562,shard_000563,shard_000564,shard_000565
[22:18:59] step 2850/8000 | loss=1.4860 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1399200/0/0 q=0 p=0 | 380 pos/s | vram=35.48GB
[22:19:14] soft inbox +399,987 train_n=54,896,770 shards=shard_000566,shard_000567,shard_000568,shard_000569
[22:19:15] ckpt latest.pt (live weights) steps=2850 status=mid
[22:19:38] step 2875/8000 | loss=1.4877 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1412400/0/0 q=0 p=0 | 380 pos/s | vram=35.48GB
[22:19:51] soft inbox +299,992 train_n=55,196,762 shards=shard_000570,shard_000571,shard_000572
[22:20:14] step 2900/8000 | loss=1.4852 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1425600/0/0 q=0 p=0 | 379 pos/s | vram=35.48GB
[22:20:27] soft inbox +299,990 train_n=55,496,752 shards=shard_000573,shard_000574,shard_000575
[22:20:36] val/lichess_endgame hard_ce=1.2970 soft_ce=1.2970 soft_temp_ce=1.2970 teacher_entropy=0.0000 teacher_kl=1.2970 wdl_ce=0.4873 value_rows=8192.0000
[22:20:59] step 2925/8000 | loss=1.4791 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1438800/0/0 q=0 p=0 | 378 pos/s | vram=35.48GB
[22:21:15] soft inbox +399,989 train_n=55,896,741 shards=shard_000576,shard_000577,shard_000578,shard_000579
[22:21:38] step 2950/8000 | loss=1.4811 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1452000/0/0 q=0 p=0 | 378 pos/s | vram=35.48GB
[22:21:54] soft inbox +399,988 train_n=56,296,729 shards=shard_000580,shard_000581,shard_000582,shard_000583
[22:22:17] step 2975/8000 | loss=1.4688 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1465200/0/0 q=0 p=0 | 378 pos/s | vram=35.48GB
[22:22:30] soft inbox +299,987 train_n=56,596,716 shards=shard_000584,shard_000585,shard_000586
[22:22:53] step 3000/8000 | loss=1.4947 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1478400/0/0 q=0 p=0 | 377 pos/s | vram=35.48GB
[22:23:07] soft inbox +299,992 train_n=56,896,708 shards=shard_000587,shard_000588,shard_000589
[22:23:08] ckpt step_003000.pt + latest.pt (live) steps=3000 status=mid kind=full
[22:23:17] val/lichess_endgame hard_ce=1.2986 soft_ce=1.2986 soft_temp_ce=1.2986 teacher_entropy=0.0000 teacher_kl=1.2986 wdl_ce=0.4900 value_rows=8192.0000
[22:23:41] step 3025/8000 | loss=1.4750 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1491600/0/0 q=0 p=0 | 376 pos/s | vram=35.48GB
[22:23:59] soft inbox +499,985 train_n=57,396,693 shards=shard_000590,shard_000591,shard_000592,shard_000593,shard_000594
[22:24:22] step 3050/8000 | loss=1.4747 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1504800/0/0 q=0 p=0 | 376 pos/s | vram=35.48GB
[22:24:35] soft inbox +299,998 train_n=57,696,691 shards=shard_000595,shard_000596,shard_000597
[22:24:59] step 3075/8000 | loss=1.4765 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1518000/0/0 q=0 p=0 | 376 pos/s | vram=35.48GB
[22:25:15] soft inbox +399,988 train_n=58,096,679 shards=shard_000598,shard_000599,shard_000600,shard_000601
[22:25:38] step 3100/8000 | loss=1.4831 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1531200/0/0 q=0 p=0 | 375 pos/s | vram=35.48GB
[22:25:52] soft inbox +299,992 train_n=58,396,671 shards=shard_000602,shard_000603,shard_000604
[22:26:01] val/lichess_endgame hard_ce=1.2967 soft_ce=1.2967 soft_temp_ce=1.2967 teacher_entropy=0.0000 teacher_kl=1.2967 wdl_ce=0.4883 value_rows=8192.0000
[22:26:24] step 3125/8000 | loss=1.4715 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1544400/0/0 q=0 p=0 | 374 pos/s | vram=35.48GB
[22:26:40] soft inbox +399,990 train_n=58,796,661 shards=shard_000605,shard_000606,shard_000607,shard_000608
[22:27:03] step 3150/8000 | loss=1.4894 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1557600/0/0 q=0 p=0 | 374 pos/s | vram=35.48GB
[22:27:20] soft inbox +399,992 train_n=59,196,653 shards=shard_000609,shard_000610,shard_000611,shard_000612
[22:27:20] ckpt latest.pt (live weights) steps=3150 status=mid
[22:27:44] step 3175/8000 | loss=1.4908 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1570800/0/0 q=0 p=0 | 373 pos/s | vram=35.48GB
[22:27:57] soft inbox +299,988 train_n=59,496,641 shards=shard_000613,shard_000614,shard_000615
[22:28:21] step 3200/8000 | loss=1.4777 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1584000/0/0 q=0 p=0 | 373 pos/s | vram=35.48GB
[22:28:37] soft inbox +399,987 train_n=59,896,628 shards=shard_000616,shard_000617,shard_000618,shard_000619
[22:28:46] val/lichess_endgame hard_ce=1.2967 soft_ce=1.2967 soft_temp_ce=1.2967 teacher_entropy=0.0000 teacher_kl=1.2967 wdl_ce=0.4890 value_rows=8192.0000
[22:29:09] step 3225/8000 | loss=1.4718 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1597200/0/0 q=0 p=0 | 372 pos/s | vram=35.48GB
[22:29:25] soft inbox +399,991 train_n=60,296,619 shards=shard_000620,shard_000621,shard_000622,shard_000623
[22:29:48] step 3250/8000 | loss=1.4669 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1610400/0/0 q=0 p=0 | 372 pos/s | vram=35.48GB
[22:30:05] soft inbox +399,987 train_n=60,696,606 shards=shard_000624,shard_000625,shard_000626,shard_000627
[22:30:28] step 3275/8000 | loss=1.4858 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1623600/0/0 q=0 p=0 | 371 pos/s | vram=35.48GB
[22:30:42] soft inbox +299,986 train_n=60,996,592 shards=shard_000628,shard_000629,shard_000630
[22:31:05] step 3300/8000 | loss=1.4783 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1636800/0/0 q=0 p=0 | 371 pos/s | vram=35.48GB
[22:31:20] soft inbox +299,987 train_n=61,296,579 shards=shard_000631,shard_000632,shard_000633
[22:31:21] ckpt step_003300.pt + latest.pt (live) steps=3300 status=mid kind=full
[22:31:30] val/lichess_endgame hard_ce=1.2954 soft_ce=1.2954 soft_temp_ce=1.2954 teacher_entropy=0.0000 teacher_kl=1.2954 wdl_ce=0.4879 value_rows=8192.0000
[22:31:53] step 3325/8000 | loss=1.4780 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1650000/0/0 q=0 p=0 | 370 pos/s | vram=35.48GB
[22:32:13] soft inbox +499,972 train_n=61,796,551 shards=shard_000634,shard_000635,shard_000636,shard_000637,shard_000638
[22:32:36] step 3350/8000 | loss=1.4874 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1663200/0/0 q=0 p=0 | 370 pos/s | vram=35.48GB
[22:32:51] soft inbox +299,986 train_n=62,096,537 shards=shard_000639,shard_000640,shard_000641
[22:33:14] step 3375/8000 | loss=1.4743 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1676400/0/0 q=0 p=0 | 369 pos/s | vram=35.48GB
[22:33:32] soft inbox +399,986 train_n=62,496,523 shards=shard_000642,shard_000643,shard_000644,shard_000645
[22:33:55] step 3400/8000 | loss=1.4709 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1689600/0/0 q=0 p=0 | 369 pos/s | vram=35.48GB
[22:34:10] soft inbox +299,988 train_n=62,796,511 shards=shard_000646,shard_000647,shard_000648
[22:34:19] val/lichess_endgame hard_ce=1.2932 soft_ce=1.2932 soft_temp_ce=1.2932 teacher_entropy=0.0000 teacher_kl=1.2932 wdl_ce=0.4860 value_rows=8192.0000
[22:34:42] step 3425/8000 | loss=1.4900 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1702800/0/0 q=0 p=0 | 368 pos/s | vram=35.48GB
[22:35:00] soft inbox +399,989 train_n=63,196,500 shards=shard_000649,shard_000650,shard_000651,shard_000652
[22:35:23] step 3450/8000 | loss=1.4677 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1716000/0/0 q=0 p=0 | 368 pos/s | vram=35.48GB
[22:35:40] soft inbox +399,990 train_n=63,596,490 shards=shard_000653,shard_000654,shard_000655,shard_000656
[22:35:41] ckpt latest.pt (live weights) steps=3450 status=mid
[22:36:04] step 3475/8000 | loss=1.4707 | lr=2.00e-03,3.00e-05 | clip=1/25 | mix s/d/b=1729200/0/0 q=0 p=0 | 367 pos/s | vram=35.48GB
[22:36:18] soft inbox +299,993 train_n=63,896,483 shards=shard_000657,shard_000658,shard_000659
[22:36:42] step 3500/8000 | loss=1.4740 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1742400/0/0 q=0 p=0 | 367 pos/s | vram=35.48GB
[22:36:59] soft inbox +399,991 train_n=64,296,474 shards=shard_000660,shard_000661,shard_000662,shard_000663
[22:37:08] val/lichess_endgame hard_ce=1.2965 soft_ce=1.2965 soft_temp_ce=1.2965 teacher_entropy=0.0000 teacher_kl=1.2965 wdl_ce=0.4926 value_rows=8192.0000
[22:37:31] step 3525/8000 | loss=1.4735 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1755600/0/0 q=0 p=0 | 366 pos/s | vram=35.48GB
[22:37:49] soft inbox +399,975 train_n=64,696,449 shards=shard_000664,shard_000665,shard_000666,shard_000667
[22:38:12] step 3550/8000 | loss=1.4841 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1768800/0/0 q=0 p=0 | 366 pos/s | vram=35.48GB
[22:38:28] soft inbox +299,991 train_n=64,996,440 shards=shard_000668,shard_000669,shard_000670
[22:38:51] step 3575/8000 | loss=1.4858 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1782000/0/0 q=0 p=0 | 366 pos/s | vram=35.48GB
[22:39:06] soft inbox +299,991 train_n=65,296,431 shards=shard_000671,shard_000672,shard_000673
[22:39:29] step 3600/8000 | loss=1.4834 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1795200/0/0 q=0 p=0 | 365 pos/s | vram=35.48GB
[22:39:45] soft inbox +299,989 train_n=65,596,420 shards=shard_000674,shard_000675,shard_000676
[22:39:46] ckpt step_003600.pt + latest.pt (live) steps=3600 status=mid kind=full
[22:39:55] val/lichess_endgame hard_ce=1.2912 soft_ce=1.2912 soft_temp_ce=1.2912 teacher_entropy=0.0000 teacher_kl=1.2912 wdl_ce=0.4870 value_rows=8192.0000
[22:40:18] step 3625/8000 | loss=1.4885 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1808400/0/0 q=0 p=0 | 364 pos/s | vram=35.48GB
[22:40:36] soft inbox +399,983 train_n=65,996,403 shards=shard_000677,shard_000678,shard_000679,shard_000680
[22:40:59] step 3650/8000 | loss=1.4865 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1821600/0/0 q=0 p=0 | 364 pos/s | vram=35.48GB
[22:41:14] soft inbox +299,992 train_n=66,296,395 shards=shard_000681,shard_000682,shard_000683
[22:41:38] step 3675/8000 | loss=1.4790 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1834800/0/0 q=0 p=0 | 364 pos/s | vram=35.48GB
[22:41:52] soft inbox +299,992 train_n=66,596,387 shards=shard_000684,shard_000685,shard_000686
[22:42:15] step 3700/8000 | loss=1.4697 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1848000/0/0 q=0 p=0 | 364 pos/s | vram=35.48GB
[22:42:31] soft inbox +299,993 train_n=66,896,380 shards=shard_000687,shard_000688,shard_000689
[22:42:40] val/lichess_endgame hard_ce=1.2936 soft_ce=1.2936 soft_temp_ce=1.2936 teacher_entropy=0.0000 teacher_kl=1.2936 wdl_ce=0.4848 value_rows=8192.0000
[22:43:03] step 3725/8000 | loss=1.4682 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1861200/0/0 q=0 p=0 | 363 pos/s | vram=35.48GB
[22:43:21] soft inbox +399,989 train_n=67,296,369 shards=shard_000690,shard_000691,shard_000692,shard_000693
[22:43:44] step 3750/8000 | loss=1.4706 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1874400/0/0 q=0 p=0 | 363 pos/s | vram=35.48GB
[22:44:00] soft inbox +299,982 train_n=67,596,351 shards=shard_000694,shard_000695,shard_000696
[22:44:00] ckpt latest.pt (live weights) steps=3750 status=mid
[22:44:23] step 3775/8000 | loss=1.5026 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1887600/0/0 q=0 p=0 | 363 pos/s | vram=35.48GB
[22:44:39] soft inbox +299,992 train_n=67,896,343 shards=shard_000697,shard_000698,shard_000699
[22:45:02] step 3800/8000 | loss=1.4873 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1900800/0/0 q=0 p=0 | 362 pos/s | vram=35.48GB
[22:45:18] soft inbox +299,991 train_n=68,196,334 shards=shard_000700,shard_000701,shard_000702
[22:45:27] val/lichess_endgame hard_ce=1.2978 soft_ce=1.2978 soft_temp_ce=1.2978 teacher_entropy=0.0000 teacher_kl=1.2978 wdl_ce=0.4864 value_rows=8192.0000
[22:45:50] step 3825/8000 | loss=1.4895 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1914000/0/0 q=0 p=0 | 362 pos/s | vram=35.48GB
[22:46:09] soft inbox +399,994 train_n=68,596,328 shards=shard_000703,shard_000704,shard_000705,shard_000706
[22:46:32] step 3850/8000 | loss=1.4999 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1927200/0/0 q=0 p=0 | 361 pos/s | vram=35.48GB
[22:46:48] soft inbox +299,988 train_n=68,896,316 shards=shard_000707,shard_000708,shard_000709
[22:47:11] step 3875/8000 | loss=1.4950 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1940400/0/0 q=0 p=0 | 361 pos/s | vram=35.48GB
[22:47:26] soft inbox +299,987 train_n=69,196,303 shards=shard_000710,shard_000711,shard_000712
[22:47:50] step 3900/8000 | loss=1.4877 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1953600/0/0 q=0 p=0 | 361 pos/s | vram=35.48GB
[22:48:09] soft inbox +399,987 train_n=69,596,290 shards=shard_000713,shard_000714,shard_000715,shard_000716
[22:48:10] ckpt step_003900.pt + latest.pt (live) steps=3900 status=mid kind=full
[22:48:19] val/lichess_endgame hard_ce=1.2938 soft_ce=1.2938 soft_temp_ce=1.2938 teacher_entropy=0.0000 teacher_kl=1.2938 wdl_ce=0.4871 value_rows=8192.0000
[22:48:42] step 3925/8000 | loss=1.4879 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1966800/0/0 q=0 p=0 | 360 pos/s | vram=35.48GB
[22:49:01] soft inbox +399,981 train_n=69,996,271 shards=shard_000717,shard_000718,shard_000719,shard_000720
[22:49:25] step 3950/8000 | loss=1.4972 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=1980000/0/0 q=0 p=0 | 359 pos/s | vram=35.48GB
[22:49:41] soft inbox +299,983 train_n=70,296,254 shards=shard_000721,shard_000722,shard_000723
[22:50:04] step 3975/8000 | loss=1.4812 | lr=2.00e-03,3.00e-05 | clip=1/25 | mix s/d/b=1993200/0/0 q=0 p=0 | 359 pos/s | vram=35.48GB
[22:50:23] soft inbox +399,973 train_n=70,696,227 shards=shard_000724,shard_000725,shard_000726,shard_000727
[22:50:46] step 4000/8000 | loss=1.4944 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=2006400/0/0 q=0 p=0 | 359 pos/s | vram=35.48GB
[22:51:00] soft inbox +199,995 train_n=70,896,222 shards=shard_000728,shard_000729
[22:51:09] val/lichess_endgame hard_ce=1.2881 soft_ce=1.2881 soft_temp_ce=1.2881 teacher_entropy=0.0000 teacher_kl=1.2881 wdl_ce=0.4857 value_rows=8192.0000
[22:51:32] step 4025/8000 | loss=1.4966 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=2019600/0/0 q=0 p=0 | 358 pos/s | vram=35.48GB
[22:51:52] soft inbox +399,981 train_n=71,296,203 shards=shard_000730,shard_000731,shard_000732,shard_000733
[22:52:15] step 4050/8000 | loss=1.4766 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=2032800/0/0 q=0 p=0 | 358 pos/s | vram=35.48GB
[22:52:35] soft inbox +399,987 train_n=71,696,190 shards=shard_000734,shard_000735,shard_000736,shard_000737
[22:52:36] ckpt latest.pt (live weights) steps=4050 status=mid
[22:52:59] step 4075/8000 | loss=1.4833 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=2046000/0/0 q=0 p=0 | 358 pos/s | vram=35.48GB
[22:53:16] soft inbox +299,991 train_n=71,996,181 shards=shard_000738,shard_000739,shard_000740
[22:53:39] step 4100/8000 | loss=1.4781 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=2059200/0/0 q=0 p=0 | 357 pos/s | vram=35.48GB
[22:53:56] soft inbox +299,988 train_n=72,296,169 shards=shard_000741,shard_000742,shard_000743
[22:54:05] val/lichess_endgame hard_ce=1.2921 soft_ce=1.2921 soft_temp_ce=1.2921 teacher_entropy=0.0000 teacher_kl=1.2921 wdl_ce=0.4895 value_rows=8192.0000
[22:54:29] step 4125/8000 | loss=1.4776 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=2072400/0/0 q=0 p=0 | 357 pos/s | vram=35.48GB
[22:54:49] soft inbox +399,982 train_n=72,696,151 shards=shard_000744,shard_000745,shard_000746,shard_000747
[22:55:12] step 4150/8000 | loss=1.4830 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=2085600/0/0 q=0 p=0 | 356 pos/s | vram=35.48GB
[22:55:30] soft inbox +299,987 train_n=72,996,138 shards=shard_000748,shard_000749,shard_000750
[22:55:53] step 4175/8000 | loss=1.4754 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=2098800/0/0 q=0 p=0 | 356 pos/s | vram=35.48GB
[22:56:10] soft inbox +299,989 train_n=73,296,127 shards=shard_000751,shard_000752,shard_000753
[22:56:33] step 4200/8000 | loss=1.4911 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=2112000/0/0 q=0 p=0 | 356 pos/s | vram=35.48GB
[22:56:51] soft inbox +299,995 train_n=73,596,122 shards=shard_000754,shard_000755,shard_000756
[22:56:52] ckpt step_004200.pt + latest.pt (live) steps=4200 status=mid kind=full
[22:57:01] val/lichess_endgame hard_ce=1.2925 soft_ce=1.2925 soft_temp_ce=1.2925 teacher_entropy=0.0000 teacher_kl=1.2925 wdl_ce=0.4860 value_rows=8192.0000
[22:57:24] step 4225/8000 | loss=1.4831 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=2125200/0/0 q=0 p=0 | 355 pos/s | vram=35.48GB
[22:57:41] soft inbox +299,985 train_n=73,896,107 shards=shard_000757,shard_000758,shard_000759
[22:58:04] step 4250/8000 | loss=1.4841 | lr=2.00e-03,3.00e-05 | clip=1/25 | mix s/d/b=2138400/0/0 q=0 p=0 | 355 pos/s | vram=35.48GB
[22:58:25] soft inbox +399,988 train_n=74,296,095 shards=shard_000760,shard_000761,shard_000762,shard_000763
[22:58:49] step 4275/8000 | loss=1.4928 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=2151600/0/0 q=0 p=0 | 354 pos/s | vram=35.48GB
[22:59:06] soft inbox +299,990 train_n=74,596,085 shards=shard_000764,shard_000765,shard_000766
[22:59:29] step 4300/8000 | loss=1.4843 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=2164800/0/0 q=0 p=0 | 354 pos/s | vram=35.48GB
[22:59:47] soft inbox +299,988 train_n=74,896,073 shards=shard_000767,shard_000768,shard_000769
[22:59:56] val/lichess_endgame hard_ce=1.2928 soft_ce=1.2928 soft_temp_ce=1.2928 teacher_entropy=0.0000 teacher_kl=1.2928 wdl_ce=0.4849 value_rows=8192.0000
[23:00:20] step 4325/8000 | loss=1.4879 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=2178000/0/0 q=0 p=0 | 353 pos/s | vram=35.48GB
[23:00:38] soft inbox +299,988 train_n=75,196,061 shards=shard_000770,shard_000771,shard_000772
[23:01:02] step 4350/8000 | loss=1.4773 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=2191200/0/0 q=0 p=0 | 353 pos/s | vram=35.48GB
[23:01:20] soft inbox +299,986 train_n=75,496,047 shards=shard_000773,shard_000774,shard_000775
[23:01:20] ckpt latest.pt (live weights) steps=4350 status=mid
[23:01:44] step 4375/8000 | loss=1.4781 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=2204400/0/0 q=0 p=0 | 353 pos/s | vram=35.48GB
[23:02:01] soft inbox +299,990 train_n=75,796,037 shards=shard_000776,shard_000777,shard_000778
[23:02:25] step 4400/8000 | loss=1.4798 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=2217600/0/0 q=0 p=0 | 353 pos/s | vram=35.48GB
[23:02:43] soft inbox +299,994 train_n=76,096,031 shards=shard_000779,shard_000780,shard_000781
[23:02:52] val/lichess_endgame hard_ce=1.2914 soft_ce=1.2914 soft_temp_ce=1.2914 teacher_entropy=0.0000 teacher_kl=1.2914 wdl_ce=0.4848 value_rows=8192.0000
[23:03:15] step 4425/8000 | loss=1.4862 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=2230800/0/0 q=0 p=0 | 352 pos/s | vram=35.48GB
[23:03:37] soft inbox +399,988 train_n=76,496,019 shards=shard_000782,shard_000783,shard_000784,shard_000785
[23:04:00] step 4450/8000 | loss=1.4730 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=2244000/0/0 q=0 p=0 | 352 pos/s | vram=35.48GB
[23:04:23] soft inbox +399,987 train_n=76,896,006 shards=shard_000786,shard_000787,shard_000788,shard_000789
[23:04:46] step 4475/8000 | loss=1.4961 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=2257200/0/0 q=0 p=0 | 351 pos/s | vram=35.48GB
[23:05:04] soft inbox +299,989 train_n=77,195,995 shards=shard_000790,shard_000791,shard_000792
[23:05:27] step 4500/8000 | loss=1.4738 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=2270400/0/0 q=0 p=0 | 351 pos/s | vram=35.48GB
[23:05:49] soft inbox +399,984 train_n=77,595,979 shards=shard_000793,shard_000794,shard_000795,shard_000796
[23:05:51] ckpt step_004500.pt + latest.pt (live) steps=4500 status=mid kind=full
[23:06:00] val/lichess_endgame hard_ce=1.2928 soft_ce=1.2928 soft_temp_ce=1.2928 teacher_entropy=0.0000 teacher_kl=1.2928 wdl_ce=0.4853 value_rows=8192.0000
[23:06:23] step 4525/8000 | loss=1.4744 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=2283600/0/0 q=0 p=0 | 350 pos/s | vram=35.48GB
[23:06:44] soft inbox +399,980 train_n=77,995,959 shards=shard_000797,shard_000798,shard_000799,shard_000800
[23:07:08] step 4550/8000 | loss=1.4782 | lr=2.00e-03,3.00e-05 | clip=0/25 | mix s/d/b=2296800/0/0 q=0 p=0 | 350 pos/s | vram=35.48GB