val[frozen kmeans-tiered (before)] CE = 2.0644 (21s) step 0 loss 2.0509 lr 1.00e-05 14s step 10 loss 1.7405 lr 6.00e-05 138s step 20 loss 1.6460 lr 9.88e-05 228s step 30 loss 1.6519 lr 9.74e-05 318s step 40 loss 1.6056 lr 9.55e-05 409s step 50 loss 1.6589 lr 9.30e-05 499s step 60 loss 1.5835 lr 9.01e-05 589s step 70 loss 1.4230 lr 8.68e-05 679s step 80 loss 1.6415 lr 8.31e-05 769s step 90 loss 1.5090 lr 7.90e-05 3965s step 100 loss 1.6062 lr 7.45e-05 4059s val[kmeans-tiered @ step 100] CE = 1.5337 (19s) step 110 loss 1.6061 lr 6.99e-05 4168s step 120 loss 1.4446 lr 6.50e-05 4261s step 130 loss 1.3781 lr 5.99e-05 4359s step 140 loss 1.3375 lr 5.47e-05 4459s step 150 loss 1.6208 lr 4.95e-05 4557s step 160 loss 1.3232 lr 4.43e-05 4655s step 170 loss 1.3663 lr 3.91e-05 4755s step 180 loss 1.6057 lr 3.41e-05 4853s step 190 loss 1.5667 lr 2.92e-05 5681s step 200 loss 1.3921 lr 2.45e-05 5777s val[kmeans-tiered @ step 200] CE = 1.4669 (21s) step 210 loss 1.5067 lr 2.02e-05 5890s step 220 loss 1.2962 lr 1.62e-05 5982s step 230 loss 1.5839 lr 1.25e-05 6076s step 240 loss 1.4582 lr 9.24e-06 6175s step 250 loss 1.4328 lr 6.44e-06 6272s step 260 loss 1.5160 lr 4.11e-06 6369s step 270 loss 1.5190 lr 2.29e-06 6477s step 280 loss 1.5186 lr 9.86e-07 6581s step 290 loss 1.3888 lr 2.22e-07 6689s val[tuned kmeans-tiered (final)] CE = 1.4438 (21s) exported ../models/stories15M-tiered.bin