diff --git "a/trainer_state.json" "b/trainer_state.json" new file mode 100644--- /dev/null +++ "b/trainer_state.json" @@ -0,0 +1,7479 @@ +{ + "best_metric": null, + "best_model_checkpoint": null, + "epoch": 10.0, + "eval_steps": 500, + "global_step": 5250, + "is_hyper_param_search": false, + "is_local_process_zero": true, + "is_world_process_zero": true, + "log_history": [ + { + "epoch": 0.0019047619047619048, + "grad_norm": 4.547202110290527, + "learning_rate": 3.8095238095238096e-07, + "loss": 1.5025, + "step": 1 + }, + { + "epoch": 0.009523809523809525, + "grad_norm": 4.337673187255859, + "learning_rate": 1.9047619047619051e-06, + "loss": 1.5226, + "step": 5 + }, + { + "epoch": 0.01904761904761905, + "grad_norm": 4.153797149658203, + "learning_rate": 3.8095238095238102e-06, + "loss": 1.5223, + "step": 10 + }, + { + "epoch": 0.02857142857142857, + "grad_norm": 4.489221096038818, + "learning_rate": 5.7142857142857145e-06, + "loss": 1.5005, + "step": 15 + }, + { + "epoch": 0.0380952380952381, + "grad_norm": 4.049315452575684, + "learning_rate": 7.6190476190476205e-06, + "loss": 1.4699, + "step": 20 + }, + { + "epoch": 0.047619047619047616, + "grad_norm": 3.1139485836029053, + "learning_rate": 9.523809523809523e-06, + "loss": 1.4301, + "step": 25 + }, + { + "epoch": 0.05714285714285714, + "grad_norm": 2.510863780975342, + "learning_rate": 1.1428571428571429e-05, + "loss": 1.4343, + "step": 30 + }, + { + "epoch": 0.06666666666666667, + "grad_norm": 1.4479591846466064, + "learning_rate": 1.3333333333333333e-05, + "loss": 1.3446, + "step": 35 + }, + { + "epoch": 0.0761904761904762, + "grad_norm": 0.9517801403999329, + "learning_rate": 1.5238095238095241e-05, + "loss": 1.3221, + "step": 40 + }, + { + "epoch": 0.08571428571428572, + "grad_norm": 0.5813843011856079, + "learning_rate": 1.7142857142857145e-05, + "loss": 1.2988, + "step": 45 + }, + { + "epoch": 0.09523809523809523, + "grad_norm": 0.4426492154598236, + "learning_rate": 1.9047619047619046e-05, + "loss": 1.2713, + "step": 50 + }, + { + "epoch": 0.10476190476190476, + "grad_norm": 0.38477158546447754, + "learning_rate": 2.0952380952380954e-05, + "loss": 1.2046, + "step": 55 + }, + { + "epoch": 0.11428571428571428, + "grad_norm": 0.44873887300491333, + "learning_rate": 2.2857142857142858e-05, + "loss": 1.1903, + "step": 60 + }, + { + "epoch": 0.12380952380952381, + "grad_norm": 0.3480363190174103, + "learning_rate": 2.4761904761904762e-05, + "loss": 1.1401, + "step": 65 + }, + { + "epoch": 0.13333333333333333, + "grad_norm": 0.40956687927246094, + "learning_rate": 2.6666666666666667e-05, + "loss": 1.1247, + "step": 70 + }, + { + "epoch": 0.14285714285714285, + "grad_norm": 0.3006480038166046, + "learning_rate": 2.857142857142857e-05, + "loss": 1.0727, + "step": 75 + }, + { + "epoch": 0.1523809523809524, + "grad_norm": 0.32749098539352417, + "learning_rate": 3.0476190476190482e-05, + "loss": 1.0581, + "step": 80 + }, + { + "epoch": 0.1619047619047619, + "grad_norm": 0.3036528527736664, + "learning_rate": 3.2380952380952386e-05, + "loss": 1.0331, + "step": 85 + }, + { + "epoch": 0.17142857142857143, + "grad_norm": 0.2610386908054352, + "learning_rate": 3.428571428571429e-05, + "loss": 0.9981, + "step": 90 + }, + { + "epoch": 0.18095238095238095, + "grad_norm": 0.29137253761291504, + "learning_rate": 3.619047619047619e-05, + "loss": 0.9739, + "step": 95 + }, + { + "epoch": 0.19047619047619047, + "grad_norm": 0.24456577003002167, + "learning_rate": 3.809523809523809e-05, + "loss": 0.9688, + "step": 100 + }, + { + "epoch": 0.2, + "grad_norm": 0.2633320689201355, + "learning_rate": 4e-05, + "loss": 0.9689, + "step": 105 + }, + { + "epoch": 0.20952380952380953, + "grad_norm": 0.25056907534599304, + "learning_rate": 4.190476190476191e-05, + "loss": 0.9563, + "step": 110 + }, + { + "epoch": 0.21904761904761905, + "grad_norm": 0.24942700564861298, + "learning_rate": 4.380952380952381e-05, + "loss": 0.9452, + "step": 115 + }, + { + "epoch": 0.22857142857142856, + "grad_norm": 0.23507161438465118, + "learning_rate": 4.5714285714285716e-05, + "loss": 0.922, + "step": 120 + }, + { + "epoch": 0.23809523809523808, + "grad_norm": 0.24385438859462738, + "learning_rate": 4.761904761904762e-05, + "loss": 0.9187, + "step": 125 + }, + { + "epoch": 0.24761904761904763, + "grad_norm": 0.285789430141449, + "learning_rate": 4.9523809523809525e-05, + "loss": 0.912, + "step": 130 + }, + { + "epoch": 0.2571428571428571, + "grad_norm": 0.27400535345077515, + "learning_rate": 5.142857142857143e-05, + "loss": 0.9069, + "step": 135 + }, + { + "epoch": 0.26666666666666666, + "grad_norm": 0.2844380736351013, + "learning_rate": 5.333333333333333e-05, + "loss": 0.8973, + "step": 140 + }, + { + "epoch": 0.2761904761904762, + "grad_norm": 0.26373493671417236, + "learning_rate": 5.5238095238095244e-05, + "loss": 0.9018, + "step": 145 + }, + { + "epoch": 0.2857142857142857, + "grad_norm": 0.28538423776626587, + "learning_rate": 5.714285714285714e-05, + "loss": 0.9021, + "step": 150 + }, + { + "epoch": 0.29523809523809524, + "grad_norm": 0.2729125916957855, + "learning_rate": 5.904761904761905e-05, + "loss": 0.8798, + "step": 155 + }, + { + "epoch": 0.3047619047619048, + "grad_norm": 0.3309907019138336, + "learning_rate": 6.0952380952380964e-05, + "loss": 0.8709, + "step": 160 + }, + { + "epoch": 0.3142857142857143, + "grad_norm": 0.2918560802936554, + "learning_rate": 6.285714285714286e-05, + "loss": 0.8643, + "step": 165 + }, + { + "epoch": 0.3238095238095238, + "grad_norm": 0.3260324001312256, + "learning_rate": 6.476190476190477e-05, + "loss": 0.8742, + "step": 170 + }, + { + "epoch": 0.3333333333333333, + "grad_norm": 0.2992309033870697, + "learning_rate": 6.666666666666667e-05, + "loss": 0.8893, + "step": 175 + }, + { + "epoch": 0.34285714285714286, + "grad_norm": 0.3110353350639343, + "learning_rate": 6.857142857142858e-05, + "loss": 0.8452, + "step": 180 + }, + { + "epoch": 0.3523809523809524, + "grad_norm": 0.3042120635509491, + "learning_rate": 7.047619047619048e-05, + "loss": 0.8585, + "step": 185 + }, + { + "epoch": 0.3619047619047619, + "grad_norm": 0.31247013807296753, + "learning_rate": 7.238095238095238e-05, + "loss": 0.8443, + "step": 190 + }, + { + "epoch": 0.37142857142857144, + "grad_norm": 0.29538431763648987, + "learning_rate": 7.428571428571429e-05, + "loss": 0.8431, + "step": 195 + }, + { + "epoch": 0.38095238095238093, + "grad_norm": 0.34301429986953735, + "learning_rate": 7.619047619047618e-05, + "loss": 0.8555, + "step": 200 + }, + { + "epoch": 0.3904761904761905, + "grad_norm": 0.3382505476474762, + "learning_rate": 7.80952380952381e-05, + "loss": 0.8392, + "step": 205 + }, + { + "epoch": 0.4, + "grad_norm": 0.2907902002334595, + "learning_rate": 8e-05, + "loss": 0.8232, + "step": 210 + }, + { + "epoch": 0.4095238095238095, + "grad_norm": 0.29543590545654297, + "learning_rate": 8.19047619047619e-05, + "loss": 0.8215, + "step": 215 + }, + { + "epoch": 0.41904761904761906, + "grad_norm": 0.33882054686546326, + "learning_rate": 8.380952380952382e-05, + "loss": 0.8142, + "step": 220 + }, + { + "epoch": 0.42857142857142855, + "grad_norm": 0.3038513958454132, + "learning_rate": 8.571428571428571e-05, + "loss": 0.8162, + "step": 225 + }, + { + "epoch": 0.4380952380952381, + "grad_norm": 0.30865398049354553, + "learning_rate": 8.761904761904762e-05, + "loss": 0.8218, + "step": 230 + }, + { + "epoch": 0.44761904761904764, + "grad_norm": 0.2758966088294983, + "learning_rate": 8.952380952380953e-05, + "loss": 0.8131, + "step": 235 + }, + { + "epoch": 0.45714285714285713, + "grad_norm": 0.29004842042922974, + "learning_rate": 9.142857142857143e-05, + "loss": 0.7924, + "step": 240 + }, + { + "epoch": 0.4666666666666667, + "grad_norm": 0.3257029354572296, + "learning_rate": 9.333333333333334e-05, + "loss": 0.8138, + "step": 245 + }, + { + "epoch": 0.47619047619047616, + "grad_norm": 0.32075703144073486, + "learning_rate": 9.523809523809524e-05, + "loss": 0.8137, + "step": 250 + }, + { + "epoch": 0.4857142857142857, + "grad_norm": 0.3217829763889313, + "learning_rate": 9.714285714285715e-05, + "loss": 0.8051, + "step": 255 + }, + { + "epoch": 0.49523809523809526, + "grad_norm": 0.35465100407600403, + "learning_rate": 9.904761904761905e-05, + "loss": 0.8048, + "step": 260 + }, + { + "epoch": 0.5047619047619047, + "grad_norm": 0.30548200011253357, + "learning_rate": 0.00010095238095238096, + "loss": 0.798, + "step": 265 + }, + { + "epoch": 0.5142857142857142, + "grad_norm": 0.30726566910743713, + "learning_rate": 0.00010285714285714286, + "loss": 0.8034, + "step": 270 + }, + { + "epoch": 0.5238095238095238, + "grad_norm": 0.31289350986480713, + "learning_rate": 0.00010476190476190477, + "loss": 0.8003, + "step": 275 + }, + { + "epoch": 0.5333333333333333, + "grad_norm": 0.2915325164794922, + "learning_rate": 0.00010666666666666667, + "loss": 0.7923, + "step": 280 + }, + { + "epoch": 0.5428571428571428, + "grad_norm": 0.29550987482070923, + "learning_rate": 0.00010857142857142856, + "loss": 0.8084, + "step": 285 + }, + { + "epoch": 0.5523809523809524, + "grad_norm": 0.28710678219795227, + "learning_rate": 0.00011047619047619049, + "loss": 0.7715, + "step": 290 + }, + { + "epoch": 0.5619047619047619, + "grad_norm": 0.35985657572746277, + "learning_rate": 0.00011238095238095239, + "loss": 0.8112, + "step": 295 + }, + { + "epoch": 0.5714285714285714, + "grad_norm": 0.2753393352031708, + "learning_rate": 0.00011428571428571428, + "loss": 0.7816, + "step": 300 + }, + { + "epoch": 0.580952380952381, + "grad_norm": 0.2836596667766571, + "learning_rate": 0.00011619047619047621, + "loss": 0.7905, + "step": 305 + }, + { + "epoch": 0.5904761904761905, + "grad_norm": 0.31738153100013733, + "learning_rate": 0.0001180952380952381, + "loss": 0.7929, + "step": 310 + }, + { + "epoch": 0.6, + "grad_norm": 0.28743502497673035, + "learning_rate": 0.00012, + "loss": 0.7838, + "step": 315 + }, + { + "epoch": 0.6095238095238096, + "grad_norm": 0.2714017629623413, + "learning_rate": 0.00012190476190476193, + "loss": 0.7833, + "step": 320 + }, + { + "epoch": 0.6190476190476191, + "grad_norm": 0.3143031895160675, + "learning_rate": 0.0001238095238095238, + "loss": 0.789, + "step": 325 + }, + { + "epoch": 0.6285714285714286, + "grad_norm": 0.3043210804462433, + "learning_rate": 0.00012571428571428572, + "loss": 0.7845, + "step": 330 + }, + { + "epoch": 0.638095238095238, + "grad_norm": 0.3003801703453064, + "learning_rate": 0.0001276190476190476, + "loss": 0.7717, + "step": 335 + }, + { + "epoch": 0.6476190476190476, + "grad_norm": 0.29364722967147827, + "learning_rate": 0.00012952380952380954, + "loss": 0.784, + "step": 340 + }, + { + "epoch": 0.6571428571428571, + "grad_norm": 0.29768991470336914, + "learning_rate": 0.00013142857142857143, + "loss": 0.7856, + "step": 345 + }, + { + "epoch": 0.6666666666666666, + "grad_norm": 0.3462805151939392, + "learning_rate": 0.00013333333333333334, + "loss": 0.7826, + "step": 350 + }, + { + "epoch": 0.6761904761904762, + "grad_norm": 0.3385864794254303, + "learning_rate": 0.00013523809523809525, + "loss": 0.7879, + "step": 355 + }, + { + "epoch": 0.6857142857142857, + "grad_norm": 0.2977157533168793, + "learning_rate": 0.00013714285714285716, + "loss": 0.7764, + "step": 360 + }, + { + "epoch": 0.6952380952380952, + "grad_norm": 0.3157060444355011, + "learning_rate": 0.00013904761904761905, + "loss": 0.7707, + "step": 365 + }, + { + "epoch": 0.7047619047619048, + "grad_norm": 0.2835277020931244, + "learning_rate": 0.00014095238095238096, + "loss": 0.7956, + "step": 370 + }, + { + "epoch": 0.7142857142857143, + "grad_norm": 0.3145689070224762, + "learning_rate": 0.00014285714285714287, + "loss": 0.775, + "step": 375 + }, + { + "epoch": 0.7238095238095238, + "grad_norm": 0.28412529826164246, + "learning_rate": 0.00014476190476190475, + "loss": 0.786, + "step": 380 + }, + { + "epoch": 0.7333333333333333, + "grad_norm": 0.27334657311439514, + "learning_rate": 0.00014666666666666666, + "loss": 0.7635, + "step": 385 + }, + { + "epoch": 0.7428571428571429, + "grad_norm": 0.274685263633728, + "learning_rate": 0.00014857142857142857, + "loss": 0.7684, + "step": 390 + }, + { + "epoch": 0.7523809523809524, + "grad_norm": 0.27724364399909973, + "learning_rate": 0.00015047619047619048, + "loss": 0.7631, + "step": 395 + }, + { + "epoch": 0.7619047619047619, + "grad_norm": 0.28986790776252747, + "learning_rate": 0.00015238095238095237, + "loss": 0.7644, + "step": 400 + }, + { + "epoch": 0.7714285714285715, + "grad_norm": 0.28879615664482117, + "learning_rate": 0.0001542857142857143, + "loss": 0.7611, + "step": 405 + }, + { + "epoch": 0.780952380952381, + "grad_norm": 0.27653956413269043, + "learning_rate": 0.0001561904761904762, + "loss": 0.7796, + "step": 410 + }, + { + "epoch": 0.7904761904761904, + "grad_norm": 0.281981498003006, + "learning_rate": 0.0001580952380952381, + "loss": 0.7664, + "step": 415 + }, + { + "epoch": 0.8, + "grad_norm": 0.2850152850151062, + "learning_rate": 0.00016, + "loss": 0.7677, + "step": 420 + }, + { + "epoch": 0.8095238095238095, + "grad_norm": 0.254380464553833, + "learning_rate": 0.00016190476190476192, + "loss": 0.7669, + "step": 425 + }, + { + "epoch": 0.819047619047619, + "grad_norm": 0.28639087080955505, + "learning_rate": 0.0001638095238095238, + "loss": 0.7781, + "step": 430 + }, + { + "epoch": 0.8285714285714286, + "grad_norm": 0.27872613072395325, + "learning_rate": 0.00016571428571428575, + "loss": 0.7654, + "step": 435 + }, + { + "epoch": 0.8380952380952381, + "grad_norm": 0.27628201246261597, + "learning_rate": 0.00016761904761904763, + "loss": 0.772, + "step": 440 + }, + { + "epoch": 0.8476190476190476, + "grad_norm": 0.2789876461029053, + "learning_rate": 0.00016952380952380954, + "loss": 0.7657, + "step": 445 + }, + { + "epoch": 0.8571428571428571, + "grad_norm": 0.2589787542819977, + "learning_rate": 0.00017142857142857143, + "loss": 0.7608, + "step": 450 + }, + { + "epoch": 0.8666666666666667, + "grad_norm": 0.25564825534820557, + "learning_rate": 0.00017333333333333334, + "loss": 0.7701, + "step": 455 + }, + { + "epoch": 0.8761904761904762, + "grad_norm": 0.26511332392692566, + "learning_rate": 0.00017523809523809525, + "loss": 0.7608, + "step": 460 + }, + { + "epoch": 0.8857142857142857, + "grad_norm": 0.254036545753479, + "learning_rate": 0.00017714285714285713, + "loss": 0.7547, + "step": 465 + }, + { + "epoch": 0.8952380952380953, + "grad_norm": 0.2760804295539856, + "learning_rate": 0.00017904761904761907, + "loss": 0.7676, + "step": 470 + }, + { + "epoch": 0.9047619047619048, + "grad_norm": 0.27230989933013916, + "learning_rate": 0.00018095238095238095, + "loss": 0.7666, + "step": 475 + }, + { + "epoch": 0.9142857142857143, + "grad_norm": 0.27643173933029175, + "learning_rate": 0.00018285714285714286, + "loss": 0.7658, + "step": 480 + }, + { + "epoch": 0.9238095238095239, + "grad_norm": 0.253610759973526, + "learning_rate": 0.00018476190476190478, + "loss": 0.7485, + "step": 485 + }, + { + "epoch": 0.9333333333333333, + "grad_norm": 0.2513977289199829, + "learning_rate": 0.0001866666666666667, + "loss": 0.7615, + "step": 490 + }, + { + "epoch": 0.9428571428571428, + "grad_norm": 0.24851179122924805, + "learning_rate": 0.00018857142857142857, + "loss": 0.761, + "step": 495 + }, + { + "epoch": 0.9523809523809523, + "grad_norm": 0.2603645920753479, + "learning_rate": 0.00019047619047619048, + "loss": 0.767, + "step": 500 + }, + { + "epoch": 0.9619047619047619, + "grad_norm": 0.24542096257209778, + "learning_rate": 0.0001923809523809524, + "loss": 0.7535, + "step": 505 + }, + { + "epoch": 0.9714285714285714, + "grad_norm": 0.265168696641922, + "learning_rate": 0.0001942857142857143, + "loss": 0.7583, + "step": 510 + }, + { + "epoch": 0.9809523809523809, + "grad_norm": 0.2615059018135071, + "learning_rate": 0.0001961904761904762, + "loss": 0.7683, + "step": 515 + }, + { + "epoch": 0.9904761904761905, + "grad_norm": 0.23228660225868225, + "learning_rate": 0.0001980952380952381, + "loss": 0.7632, + "step": 520 + }, + { + "epoch": 1.0, + "grad_norm": 0.2468753159046173, + "learning_rate": 0.0002, + "loss": 0.741, + "step": 525 + }, + { + "epoch": 1.0, + "eval_loss": 1.3567252159118652, + "eval_runtime": 0.3239, + "eval_samples_per_second": 12.351, + "eval_steps_per_second": 3.088, + "step": 525 + }, + { + "epoch": 1.0095238095238095, + "grad_norm": 0.25585678219795227, + "learning_rate": 0.00019999944740655014, + "loss": 0.7468, + "step": 530 + }, + { + "epoch": 1.019047619047619, + "grad_norm": 0.2317453920841217, + "learning_rate": 0.00019999778963230775, + "loss": 0.7413, + "step": 535 + }, + { + "epoch": 1.0285714285714285, + "grad_norm": 0.2543046772480011, + "learning_rate": 0.00019999502669559432, + "loss": 0.7451, + "step": 540 + }, + { + "epoch": 1.0380952380952382, + "grad_norm": 0.2301589846611023, + "learning_rate": 0.00019999115862694546, + "loss": 0.7481, + "step": 545 + }, + { + "epoch": 1.0476190476190477, + "grad_norm": 0.3147750794887543, + "learning_rate": 0.00019998618546911056, + "loss": 0.749, + "step": 550 + }, + { + "epoch": 1.0571428571428572, + "grad_norm": 0.27922147512435913, + "learning_rate": 0.00019998010727705236, + "loss": 0.7626, + "step": 555 + }, + { + "epoch": 1.0666666666666667, + "grad_norm": 0.22671197354793549, + "learning_rate": 0.00019997292411794618, + "loss": 0.7553, + "step": 560 + }, + { + "epoch": 1.0761904761904761, + "grad_norm": 0.23539680242538452, + "learning_rate": 0.00019996463607117935, + "loss": 0.7339, + "step": 565 + }, + { + "epoch": 1.0857142857142856, + "grad_norm": 0.2762920558452606, + "learning_rate": 0.00019995524322835034, + "loss": 0.7574, + "step": 570 + }, + { + "epoch": 1.0952380952380953, + "grad_norm": 0.23747943341732025, + "learning_rate": 0.00019994474569326757, + "loss": 0.7561, + "step": 575 + }, + { + "epoch": 1.1047619047619048, + "grad_norm": 0.24105724692344666, + "learning_rate": 0.00019993314358194843, + "loss": 0.7553, + "step": 580 + }, + { + "epoch": 1.1142857142857143, + "grad_norm": 0.2368220090866089, + "learning_rate": 0.00019992043702261793, + "loss": 0.74, + "step": 585 + }, + { + "epoch": 1.1238095238095238, + "grad_norm": 0.24352073669433594, + "learning_rate": 0.0001999066261557073, + "loss": 0.7399, + "step": 590 + }, + { + "epoch": 1.1333333333333333, + "grad_norm": 0.22486181557178497, + "learning_rate": 0.0001998917111338525, + "loss": 0.7385, + "step": 595 + }, + { + "epoch": 1.1428571428571428, + "grad_norm": 0.22657416760921478, + "learning_rate": 0.00019987569212189224, + "loss": 0.7475, + "step": 600 + }, + { + "epoch": 1.1523809523809523, + "grad_norm": 0.2311800867319107, + "learning_rate": 0.00019985856929686667, + "loss": 0.751, + "step": 605 + }, + { + "epoch": 1.161904761904762, + "grad_norm": 0.2517586946487427, + "learning_rate": 0.00019984034284801502, + "loss": 0.7407, + "step": 610 + }, + { + "epoch": 1.1714285714285715, + "grad_norm": 0.22927944362163544, + "learning_rate": 0.0001998210129767735, + "loss": 0.7343, + "step": 615 + }, + { + "epoch": 1.180952380952381, + "grad_norm": 0.22523367404937744, + "learning_rate": 0.00019980057989677345, + "loss": 0.7676, + "step": 620 + }, + { + "epoch": 1.1904761904761905, + "grad_norm": 0.22527702152729034, + "learning_rate": 0.0001997790438338385, + "loss": 0.7318, + "step": 625 + }, + { + "epoch": 1.2, + "grad_norm": 0.3828093409538269, + "learning_rate": 0.00019975640502598244, + "loss": 0.7502, + "step": 630 + }, + { + "epoch": 1.2095238095238094, + "grad_norm": 0.22656366229057312, + "learning_rate": 0.00019973266372340639, + "loss": 0.7466, + "step": 635 + }, + { + "epoch": 1.2190476190476192, + "grad_norm": 0.26040101051330566, + "learning_rate": 0.0001997078201884961, + "loss": 0.7327, + "step": 640 + }, + { + "epoch": 1.2285714285714286, + "grad_norm": 0.22036729753017426, + "learning_rate": 0.0001996818746958191, + "loss": 0.7548, + "step": 645 + }, + { + "epoch": 1.2380952380952381, + "grad_norm": 0.2276628315448761, + "learning_rate": 0.00019965482753212156, + "loss": 0.7364, + "step": 650 + }, + { + "epoch": 1.2476190476190476, + "grad_norm": 0.24642202258110046, + "learning_rate": 0.00019962667899632518, + "loss": 0.7406, + "step": 655 + }, + { + "epoch": 1.2571428571428571, + "grad_norm": 0.22199499607086182, + "learning_rate": 0.00019959742939952392, + "loss": 0.7488, + "step": 660 + }, + { + "epoch": 1.2666666666666666, + "grad_norm": 0.22032096982002258, + "learning_rate": 0.00019956707906498044, + "loss": 0.7436, + "step": 665 + }, + { + "epoch": 1.276190476190476, + "grad_norm": 0.22946354746818542, + "learning_rate": 0.00019953562832812272, + "loss": 0.7498, + "step": 670 + }, + { + "epoch": 1.2857142857142856, + "grad_norm": 0.22464948892593384, + "learning_rate": 0.00019950307753654017, + "loss": 0.7385, + "step": 675 + }, + { + "epoch": 1.2952380952380953, + "grad_norm": 0.2189594954252243, + "learning_rate": 0.00019946942704997982, + "loss": 0.742, + "step": 680 + }, + { + "epoch": 1.3047619047619048, + "grad_norm": 0.24271249771118164, + "learning_rate": 0.00019943467724034252, + "loss": 0.7282, + "step": 685 + }, + { + "epoch": 1.3142857142857143, + "grad_norm": 0.21535710990428925, + "learning_rate": 0.00019939882849167852, + "loss": 0.7428, + "step": 690 + }, + { + "epoch": 1.3238095238095238, + "grad_norm": 0.21307721734046936, + "learning_rate": 0.0001993618812001836, + "loss": 0.7448, + "step": 695 + }, + { + "epoch": 1.3333333333333333, + "grad_norm": 0.2116234302520752, + "learning_rate": 0.00019932383577419432, + "loss": 0.7315, + "step": 700 + }, + { + "epoch": 1.342857142857143, + "grad_norm": 0.21006180346012115, + "learning_rate": 0.00019928469263418374, + "loss": 0.7288, + "step": 705 + }, + { + "epoch": 1.3523809523809525, + "grad_norm": 0.2520844638347626, + "learning_rate": 0.00019924445221275675, + "loss": 0.7345, + "step": 710 + }, + { + "epoch": 1.361904761904762, + "grad_norm": 0.2356862872838974, + "learning_rate": 0.00019920311495464518, + "loss": 0.7299, + "step": 715 + }, + { + "epoch": 1.3714285714285714, + "grad_norm": 0.22165431082248688, + "learning_rate": 0.00019916068131670302, + "loss": 0.7289, + "step": 720 + }, + { + "epoch": 1.380952380952381, + "grad_norm": 0.21119216084480286, + "learning_rate": 0.0001991171517679013, + "loss": 0.7436, + "step": 725 + }, + { + "epoch": 1.3904761904761904, + "grad_norm": 0.2150939702987671, + "learning_rate": 0.0001990725267893228, + "loss": 0.7343, + "step": 730 + }, + { + "epoch": 1.4, + "grad_norm": 0.21897323429584503, + "learning_rate": 0.00019902680687415705, + "loss": 0.7314, + "step": 735 + }, + { + "epoch": 1.4095238095238094, + "grad_norm": 0.22644034028053284, + "learning_rate": 0.00019897999252769448, + "loss": 0.7392, + "step": 740 + }, + { + "epoch": 1.4190476190476191, + "grad_norm": 0.2236684113740921, + "learning_rate": 0.00019893208426732115, + "loss": 0.7223, + "step": 745 + }, + { + "epoch": 1.4285714285714286, + "grad_norm": 0.22646702826023102, + "learning_rate": 0.00019888308262251285, + "loss": 0.7378, + "step": 750 + }, + { + "epoch": 1.438095238095238, + "grad_norm": 0.19864632189273834, + "learning_rate": 0.00019883298813482938, + "loss": 0.7224, + "step": 755 + }, + { + "epoch": 1.4476190476190476, + "grad_norm": 0.20702700316905975, + "learning_rate": 0.00019878180135790845, + "loss": 0.7228, + "step": 760 + }, + { + "epoch": 1.457142857142857, + "grad_norm": 0.21697306632995605, + "learning_rate": 0.00019872952285745959, + "loss": 0.7221, + "step": 765 + }, + { + "epoch": 1.4666666666666668, + "grad_norm": 0.209822416305542, + "learning_rate": 0.00019867615321125795, + "loss": 0.7186, + "step": 770 + }, + { + "epoch": 1.4761904761904763, + "grad_norm": 0.20457518100738525, + "learning_rate": 0.00019862169300913785, + "loss": 0.7318, + "step": 775 + }, + { + "epoch": 1.4857142857142858, + "grad_norm": 0.21236108243465424, + "learning_rate": 0.0001985661428529863, + "loss": 0.7339, + "step": 780 + }, + { + "epoch": 1.4952380952380953, + "grad_norm": 0.19768942892551422, + "learning_rate": 0.00019850950335673643, + "loss": 0.7266, + "step": 785 + }, + { + "epoch": 1.5047619047619047, + "grad_norm": 0.2030116766691208, + "learning_rate": 0.00019845177514636042, + "loss": 0.7331, + "step": 790 + }, + { + "epoch": 1.5142857142857142, + "grad_norm": 0.22577205300331116, + "learning_rate": 0.00019839295885986296, + "loss": 0.7199, + "step": 795 + }, + { + "epoch": 1.5238095238095237, + "grad_norm": 0.23508670926094055, + "learning_rate": 0.00019833305514727395, + "loss": 0.7315, + "step": 800 + }, + { + "epoch": 1.5333333333333332, + "grad_norm": 0.2087986171245575, + "learning_rate": 0.00019827206467064133, + "loss": 0.7369, + "step": 805 + }, + { + "epoch": 1.5428571428571427, + "grad_norm": 0.2076413631439209, + "learning_rate": 0.0001982099881040239, + "loss": 0.7253, + "step": 810 + }, + { + "epoch": 1.5523809523809524, + "grad_norm": 0.2177678495645523, + "learning_rate": 0.0001981468261334837, + "loss": 0.7252, + "step": 815 + }, + { + "epoch": 1.561904761904762, + "grad_norm": 0.20691955089569092, + "learning_rate": 0.0001980825794570786, + "loss": 0.7231, + "step": 820 + }, + { + "epoch": 1.5714285714285714, + "grad_norm": 0.2094532549381256, + "learning_rate": 0.00019801724878485438, + "loss": 0.7346, + "step": 825 + }, + { + "epoch": 1.580952380952381, + "grad_norm": 0.2252875119447708, + "learning_rate": 0.00019795083483883715, + "loss": 0.7352, + "step": 830 + }, + { + "epoch": 1.5904761904761906, + "grad_norm": 0.20988918840885162, + "learning_rate": 0.0001978833383530251, + "loss": 0.7201, + "step": 835 + }, + { + "epoch": 1.6, + "grad_norm": 0.22274397313594818, + "learning_rate": 0.00019781476007338058, + "loss": 0.7457, + "step": 840 + }, + { + "epoch": 1.6095238095238096, + "grad_norm": 0.21066966652870178, + "learning_rate": 0.00019774510075782172, + "loss": 0.7431, + "step": 845 + }, + { + "epoch": 1.619047619047619, + "grad_norm": 0.21072280406951904, + "learning_rate": 0.00019767436117621413, + "loss": 0.719, + "step": 850 + }, + { + "epoch": 1.6285714285714286, + "grad_norm": 0.20725810527801514, + "learning_rate": 0.00019760254211036244, + "loss": 0.743, + "step": 855 + }, + { + "epoch": 1.638095238095238, + "grad_norm": 0.21349456906318665, + "learning_rate": 0.00019752964435400155, + "loss": 0.7367, + "step": 860 + }, + { + "epoch": 1.6476190476190475, + "grad_norm": 0.22150424122810364, + "learning_rate": 0.00019745566871278794, + "loss": 0.7249, + "step": 865 + }, + { + "epoch": 1.657142857142857, + "grad_norm": 0.22179411351680756, + "learning_rate": 0.00019738061600429064, + "loss": 0.722, + "step": 870 + }, + { + "epoch": 1.6666666666666665, + "grad_norm": 0.20809294283390045, + "learning_rate": 0.00019730448705798239, + "loss": 0.7271, + "step": 875 + }, + { + "epoch": 1.6761904761904762, + "grad_norm": 0.19328345358371735, + "learning_rate": 0.00019722728271523034, + "loss": 0.7191, + "step": 880 + }, + { + "epoch": 1.6857142857142857, + "grad_norm": 0.20828521251678467, + "learning_rate": 0.00019714900382928675, + "loss": 0.7193, + "step": 885 + }, + { + "epoch": 1.6952380952380952, + "grad_norm": 0.20919948816299438, + "learning_rate": 0.00019706965126527963, + "loss": 0.742, + "step": 890 + }, + { + "epoch": 1.704761904761905, + "grad_norm": 0.21169473230838776, + "learning_rate": 0.00019698922590020312, + "loss": 0.7152, + "step": 895 + }, + { + "epoch": 1.7142857142857144, + "grad_norm": 0.2047414630651474, + "learning_rate": 0.0001969077286229078, + "loss": 0.7139, + "step": 900 + }, + { + "epoch": 1.723809523809524, + "grad_norm": 0.2073826789855957, + "learning_rate": 0.00019682516033409092, + "loss": 0.7158, + "step": 905 + }, + { + "epoch": 1.7333333333333334, + "grad_norm": 0.21763114631175995, + "learning_rate": 0.00019674152194628638, + "loss": 0.7258, + "step": 910 + }, + { + "epoch": 1.7428571428571429, + "grad_norm": 0.20960763096809387, + "learning_rate": 0.00019665681438385473, + "loss": 0.7255, + "step": 915 + }, + { + "epoch": 1.7523809523809524, + "grad_norm": 0.19667227566242218, + "learning_rate": 0.0001965710385829728, + "loss": 0.7242, + "step": 920 + }, + { + "epoch": 1.7619047619047619, + "grad_norm": 0.20691533386707306, + "learning_rate": 0.00019648419549162348, + "loss": 0.7243, + "step": 925 + }, + { + "epoch": 1.7714285714285714, + "grad_norm": 0.22077150642871857, + "learning_rate": 0.00019639628606958533, + "loss": 0.7244, + "step": 930 + }, + { + "epoch": 1.7809523809523808, + "grad_norm": 0.20273786783218384, + "learning_rate": 0.0001963073112884217, + "loss": 0.7294, + "step": 935 + }, + { + "epoch": 1.7904761904761903, + "grad_norm": 0.19783906638622284, + "learning_rate": 0.00019621727213147027, + "loss": 0.7242, + "step": 940 + }, + { + "epoch": 1.8, + "grad_norm": 0.21224017441272736, + "learning_rate": 0.0001961261695938319, + "loss": 0.7213, + "step": 945 + }, + { + "epoch": 1.8095238095238095, + "grad_norm": 0.21960359811782837, + "learning_rate": 0.00019603400468235998, + "loss": 0.7171, + "step": 950 + }, + { + "epoch": 1.819047619047619, + "grad_norm": 0.20761699974536896, + "learning_rate": 0.00019594077841564907, + "loss": 0.7187, + "step": 955 + }, + { + "epoch": 1.8285714285714287, + "grad_norm": 0.20614556968212128, + "learning_rate": 0.00019584649182402357, + "loss": 0.7153, + "step": 960 + }, + { + "epoch": 1.8380952380952382, + "grad_norm": 0.19090235233306885, + "learning_rate": 0.0001957511459495266, + "loss": 0.7202, + "step": 965 + }, + { + "epoch": 1.8476190476190477, + "grad_norm": 0.19847603142261505, + "learning_rate": 0.00019565474184590826, + "loss": 0.7073, + "step": 970 + }, + { + "epoch": 1.8571428571428572, + "grad_norm": 0.211056649684906, + "learning_rate": 0.0001955572805786141, + "loss": 0.7144, + "step": 975 + }, + { + "epoch": 1.8666666666666667, + "grad_norm": 0.2638969421386719, + "learning_rate": 0.0001954587632247732, + "loss": 0.7315, + "step": 980 + }, + { + "epoch": 1.8761904761904762, + "grad_norm": 0.21314914524555206, + "learning_rate": 0.00019535919087318652, + "loss": 0.7267, + "step": 985 + }, + { + "epoch": 1.8857142857142857, + "grad_norm": 0.19470569491386414, + "learning_rate": 0.0001952585646243146, + "loss": 0.7205, + "step": 990 + }, + { + "epoch": 1.8952380952380952, + "grad_norm": 0.21095521748065948, + "learning_rate": 0.00019515688559026563, + "loss": 0.7285, + "step": 995 + }, + { + "epoch": 1.9047619047619047, + "grad_norm": 0.20666122436523438, + "learning_rate": 0.0001950541548947829, + "loss": 0.7123, + "step": 1000 + }, + { + "epoch": 1.9142857142857141, + "grad_norm": 0.19656658172607422, + "learning_rate": 0.00019495037367323262, + "loss": 0.709, + "step": 1005 + }, + { + "epoch": 1.9238095238095239, + "grad_norm": 0.21879138052463531, + "learning_rate": 0.0001948455430725913, + "loss": 0.729, + "step": 1010 + }, + { + "epoch": 1.9333333333333333, + "grad_norm": 0.1921882927417755, + "learning_rate": 0.00019473966425143292, + "loss": 0.721, + "step": 1015 + }, + { + "epoch": 1.9428571428571428, + "grad_norm": 0.21759168803691864, + "learning_rate": 0.00019463273837991643, + "loss": 0.7274, + "step": 1020 + }, + { + "epoch": 1.9523809523809523, + "grad_norm": 0.222857266664505, + "learning_rate": 0.00019452476663977248, + "loss": 0.7275, + "step": 1025 + }, + { + "epoch": 1.961904761904762, + "grad_norm": 0.2145960032939911, + "learning_rate": 0.00019441575022429065, + "loss": 0.7234, + "step": 1030 + }, + { + "epoch": 1.9714285714285715, + "grad_norm": 0.5522246360778809, + "learning_rate": 0.00019430569033830605, + "loss": 0.7188, + "step": 1035 + }, + { + "epoch": 1.980952380952381, + "grad_norm": 0.23135340213775635, + "learning_rate": 0.00019419458819818614, + "loss": 0.7352, + "step": 1040 + }, + { + "epoch": 1.9904761904761905, + "grad_norm": 0.19514022767543793, + "learning_rate": 0.00019408244503181724, + "loss": 0.7397, + "step": 1045 + }, + { + "epoch": 2.0, + "grad_norm": 0.19880671799182892, + "learning_rate": 0.00019396926207859084, + "loss": 0.7109, + "step": 1050 + }, + { + "epoch": 2.0, + "eval_loss": 1.3157501220703125, + "eval_runtime": 0.3231, + "eval_samples_per_second": 12.38, + "eval_steps_per_second": 3.095, + "step": 1050 + }, + { + "epoch": 2.0095238095238095, + "grad_norm": 0.22727364301681519, + "learning_rate": 0.00019385504058939024, + "loss": 0.7229, + "step": 1055 + }, + { + "epoch": 2.019047619047619, + "grad_norm": 0.20397067070007324, + "learning_rate": 0.00019373978182657625, + "loss": 0.7093, + "step": 1060 + }, + { + "epoch": 2.0285714285714285, + "grad_norm": 0.23505325615406036, + "learning_rate": 0.00019362348706397373, + "loss": 0.71, + "step": 1065 + }, + { + "epoch": 2.038095238095238, + "grad_norm": 0.21519462764263153, + "learning_rate": 0.00019350615758685708, + "loss": 0.7154, + "step": 1070 + }, + { + "epoch": 2.0476190476190474, + "grad_norm": 0.21023060381412506, + "learning_rate": 0.00019338779469193639, + "loss": 0.7081, + "step": 1075 + }, + { + "epoch": 2.057142857142857, + "grad_norm": 0.21591557562351227, + "learning_rate": 0.00019326839968734279, + "loss": 0.7125, + "step": 1080 + }, + { + "epoch": 2.066666666666667, + "grad_norm": 0.21711720526218414, + "learning_rate": 0.00019314797389261424, + "loss": 0.6959, + "step": 1085 + }, + { + "epoch": 2.0761904761904764, + "grad_norm": 0.21242506802082062, + "learning_rate": 0.00019302651863868092, + "loss": 0.7166, + "step": 1090 + }, + { + "epoch": 2.085714285714286, + "grad_norm": 0.2289026975631714, + "learning_rate": 0.00019290403526785025, + "loss": 0.6892, + "step": 1095 + }, + { + "epoch": 2.0952380952380953, + "grad_norm": 0.2135750651359558, + "learning_rate": 0.00019278052513379255, + "loss": 0.7101, + "step": 1100 + }, + { + "epoch": 2.104761904761905, + "grad_norm": 0.21561966836452484, + "learning_rate": 0.00019265598960152555, + "loss": 0.7063, + "step": 1105 + }, + { + "epoch": 2.1142857142857143, + "grad_norm": 0.2171257734298706, + "learning_rate": 0.00019253043004739968, + "loss": 0.6985, + "step": 1110 + }, + { + "epoch": 2.123809523809524, + "grad_norm": 0.20231732726097107, + "learning_rate": 0.00019240384785908265, + "loss": 0.6996, + "step": 1115 + }, + { + "epoch": 2.1333333333333333, + "grad_norm": 0.19682158529758453, + "learning_rate": 0.00019227624443554425, + "loss": 0.7008, + "step": 1120 + }, + { + "epoch": 2.142857142857143, + "grad_norm": 0.21168991923332214, + "learning_rate": 0.00019214762118704076, + "loss": 0.7115, + "step": 1125 + }, + { + "epoch": 2.1523809523809523, + "grad_norm": 0.2140636444091797, + "learning_rate": 0.00019201797953509955, + "loss": 0.706, + "step": 1130 + }, + { + "epoch": 2.1619047619047618, + "grad_norm": 0.19730402529239655, + "learning_rate": 0.00019188732091250307, + "loss": 0.7092, + "step": 1135 + }, + { + "epoch": 2.1714285714285713, + "grad_norm": 0.20363454520702362, + "learning_rate": 0.00019175564676327339, + "loss": 0.7051, + "step": 1140 + }, + { + "epoch": 2.1809523809523808, + "grad_norm": 0.1948300451040268, + "learning_rate": 0.00019162295854265594, + "loss": 0.7114, + "step": 1145 + }, + { + "epoch": 2.1904761904761907, + "grad_norm": 0.20962835848331451, + "learning_rate": 0.00019148925771710347, + "loss": 0.7053, + "step": 1150 + }, + { + "epoch": 2.2, + "grad_norm": 0.221847265958786, + "learning_rate": 0.0001913545457642601, + "loss": 0.7077, + "step": 1155 + }, + { + "epoch": 2.2095238095238097, + "grad_norm": 0.20366375148296356, + "learning_rate": 0.00019121882417294462, + "loss": 0.7064, + "step": 1160 + }, + { + "epoch": 2.219047619047619, + "grad_norm": 0.20227287709712982, + "learning_rate": 0.00019108209444313433, + "loss": 0.7035, + "step": 1165 + }, + { + "epoch": 2.2285714285714286, + "grad_norm": 0.22311125695705414, + "learning_rate": 0.00019094435808594823, + "loss": 0.7088, + "step": 1170 + }, + { + "epoch": 2.238095238095238, + "grad_norm": 0.20252376794815063, + "learning_rate": 0.0001908056166236305, + "loss": 0.6998, + "step": 1175 + }, + { + "epoch": 2.2476190476190476, + "grad_norm": 0.19497165083885193, + "learning_rate": 0.00019066587158953366, + "loss": 0.7102, + "step": 1180 + }, + { + "epoch": 2.257142857142857, + "grad_norm": 0.21889087557792664, + "learning_rate": 0.0001905251245281015, + "loss": 0.6957, + "step": 1185 + }, + { + "epoch": 2.2666666666666666, + "grad_norm": 0.20565317571163177, + "learning_rate": 0.00019038337699485208, + "loss": 0.7151, + "step": 1190 + }, + { + "epoch": 2.276190476190476, + "grad_norm": 0.19329777359962463, + "learning_rate": 0.00019024063055636057, + "loss": 0.7063, + "step": 1195 + }, + { + "epoch": 2.2857142857142856, + "grad_norm": 0.22063401341438293, + "learning_rate": 0.0001900968867902419, + "loss": 0.71, + "step": 1200 + }, + { + "epoch": 2.295238095238095, + "grad_norm": 0.20509225130081177, + "learning_rate": 0.00018995214728513343, + "loss": 0.7075, + "step": 1205 + }, + { + "epoch": 2.3047619047619046, + "grad_norm": 0.20319826900959015, + "learning_rate": 0.0001898064136406771, + "loss": 0.6987, + "step": 1210 + }, + { + "epoch": 2.314285714285714, + "grad_norm": 0.19624005258083344, + "learning_rate": 0.0001896596874675021, + "loss": 0.7062, + "step": 1215 + }, + { + "epoch": 2.323809523809524, + "grad_norm": 0.20942609012126923, + "learning_rate": 0.00018951197038720688, + "loss": 0.7122, + "step": 1220 + }, + { + "epoch": 2.3333333333333335, + "grad_norm": 0.20319467782974243, + "learning_rate": 0.00018936326403234125, + "loss": 0.7012, + "step": 1225 + }, + { + "epoch": 2.342857142857143, + "grad_norm": 0.19727739691734314, + "learning_rate": 0.00018921357004638835, + "loss": 0.7056, + "step": 1230 + }, + { + "epoch": 2.3523809523809525, + "grad_norm": 0.20682814717292786, + "learning_rate": 0.00018906289008374655, + "loss": 0.702, + "step": 1235 + }, + { + "epoch": 2.361904761904762, + "grad_norm": 0.19265605509281158, + "learning_rate": 0.00018891122580971098, + "loss": 0.6957, + "step": 1240 + }, + { + "epoch": 2.3714285714285714, + "grad_norm": 0.1969050019979477, + "learning_rate": 0.00018875857890045543, + "loss": 0.6915, + "step": 1245 + }, + { + "epoch": 2.380952380952381, + "grad_norm": 0.19742543995380402, + "learning_rate": 0.00018860495104301345, + "loss": 0.7074, + "step": 1250 + }, + { + "epoch": 2.3904761904761904, + "grad_norm": 0.1971733719110489, + "learning_rate": 0.00018845034393526005, + "loss": 0.7133, + "step": 1255 + }, + { + "epoch": 2.4, + "grad_norm": 0.204182431101799, + "learning_rate": 0.00018829475928589271, + "loss": 0.7091, + "step": 1260 + }, + { + "epoch": 2.4095238095238094, + "grad_norm": 0.20317105948925018, + "learning_rate": 0.0001881381988144126, + "loss": 0.7056, + "step": 1265 + }, + { + "epoch": 2.419047619047619, + "grad_norm": 0.19810158014297485, + "learning_rate": 0.0001879806642511055, + "loss": 0.7128, + "step": 1270 + }, + { + "epoch": 2.4285714285714284, + "grad_norm": 0.20122739672660828, + "learning_rate": 0.00018782215733702286, + "loss": 0.7047, + "step": 1275 + }, + { + "epoch": 2.4380952380952383, + "grad_norm": 0.20654700696468353, + "learning_rate": 0.00018766267982396224, + "loss": 0.7044, + "step": 1280 + }, + { + "epoch": 2.447619047619048, + "grad_norm": 0.21352505683898926, + "learning_rate": 0.00018750223347444828, + "loss": 0.7102, + "step": 1285 + }, + { + "epoch": 2.4571428571428573, + "grad_norm": 0.1927068829536438, + "learning_rate": 0.00018734082006171299, + "loss": 0.6973, + "step": 1290 + }, + { + "epoch": 2.466666666666667, + "grad_norm": 0.19341471791267395, + "learning_rate": 0.00018717844136967624, + "loss": 0.707, + "step": 1295 + }, + { + "epoch": 2.4761904761904763, + "grad_norm": 0.20617789030075073, + "learning_rate": 0.00018701509919292613, + "loss": 0.6977, + "step": 1300 + }, + { + "epoch": 2.4857142857142858, + "grad_norm": 0.2034485936164856, + "learning_rate": 0.0001868507953366989, + "loss": 0.7044, + "step": 1305 + }, + { + "epoch": 2.4952380952380953, + "grad_norm": 0.211845263838768, + "learning_rate": 0.00018668553161685933, + "loss": 0.7057, + "step": 1310 + }, + { + "epoch": 2.5047619047619047, + "grad_norm": 0.19829222559928894, + "learning_rate": 0.00018651930985988036, + "loss": 0.7129, + "step": 1315 + }, + { + "epoch": 2.5142857142857142, + "grad_norm": 0.20161347091197968, + "learning_rate": 0.0001863521319028231, + "loss": 0.7088, + "step": 1320 + }, + { + "epoch": 2.5238095238095237, + "grad_norm": 0.19120529294013977, + "learning_rate": 0.0001861839995933164, + "loss": 0.6949, + "step": 1325 + }, + { + "epoch": 2.533333333333333, + "grad_norm": 0.1997355818748474, + "learning_rate": 0.00018601491478953657, + "loss": 0.7107, + "step": 1330 + }, + { + "epoch": 2.5428571428571427, + "grad_norm": 0.19840583205223083, + "learning_rate": 0.00018584487936018661, + "loss": 0.7073, + "step": 1335 + }, + { + "epoch": 2.552380952380952, + "grad_norm": 0.2014908641576767, + "learning_rate": 0.0001856738951844759, + "loss": 0.7045, + "step": 1340 + }, + { + "epoch": 2.5619047619047617, + "grad_norm": 0.18779955804347992, + "learning_rate": 0.00018550196415209914, + "loss": 0.7076, + "step": 1345 + }, + { + "epoch": 2.571428571428571, + "grad_norm": 0.196727454662323, + "learning_rate": 0.00018532908816321558, + "loss": 0.708, + "step": 1350 + }, + { + "epoch": 2.580952380952381, + "grad_norm": 0.2060198038816452, + "learning_rate": 0.00018515526912842796, + "loss": 0.6914, + "step": 1355 + }, + { + "epoch": 2.5904761904761906, + "grad_norm": 0.2006414532661438, + "learning_rate": 0.0001849805089687615, + "loss": 0.7079, + "step": 1360 + }, + { + "epoch": 2.6, + "grad_norm": 0.20242097973823547, + "learning_rate": 0.0001848048096156426, + "loss": 0.7092, + "step": 1365 + }, + { + "epoch": 2.6095238095238096, + "grad_norm": 0.2044093757867813, + "learning_rate": 0.00018462817301087748, + "loss": 0.7065, + "step": 1370 + }, + { + "epoch": 2.619047619047619, + "grad_norm": 0.19251036643981934, + "learning_rate": 0.0001844506011066308, + "loss": 0.707, + "step": 1375 + }, + { + "epoch": 2.6285714285714286, + "grad_norm": 0.2017953246831894, + "learning_rate": 0.0001842720958654039, + "loss": 0.7098, + "step": 1380 + }, + { + "epoch": 2.638095238095238, + "grad_norm": 0.19051124155521393, + "learning_rate": 0.00018409265926001343, + "loss": 0.703, + "step": 1385 + }, + { + "epoch": 2.6476190476190475, + "grad_norm": 0.20295244455337524, + "learning_rate": 0.00018391229327356916, + "loss": 0.7007, + "step": 1390 + }, + { + "epoch": 2.657142857142857, + "grad_norm": 0.1967683732509613, + "learning_rate": 0.00018373099989945236, + "loss": 0.7099, + "step": 1395 + }, + { + "epoch": 2.6666666666666665, + "grad_norm": 0.19073644280433655, + "learning_rate": 0.00018354878114129367, + "loss": 0.6934, + "step": 1400 + }, + { + "epoch": 2.6761904761904765, + "grad_norm": 0.19312947988510132, + "learning_rate": 0.0001833656390129509, + "loss": 0.6955, + "step": 1405 + }, + { + "epoch": 2.685714285714286, + "grad_norm": 0.19207774102687836, + "learning_rate": 0.0001831815755384869, + "loss": 0.6975, + "step": 1410 + }, + { + "epoch": 2.6952380952380954, + "grad_norm": 0.21370422840118408, + "learning_rate": 0.00018299659275214706, + "loss": 0.6936, + "step": 1415 + }, + { + "epoch": 2.704761904761905, + "grad_norm": 0.20860233902931213, + "learning_rate": 0.00018281069269833692, + "loss": 0.709, + "step": 1420 + }, + { + "epoch": 2.7142857142857144, + "grad_norm": 0.19530543684959412, + "learning_rate": 0.0001826238774315995, + "loss": 0.6969, + "step": 1425 + }, + { + "epoch": 2.723809523809524, + "grad_norm": 0.20660221576690674, + "learning_rate": 0.00018243614901659264, + "loss": 0.7213, + "step": 1430 + }, + { + "epoch": 2.7333333333333334, + "grad_norm": 0.19530445337295532, + "learning_rate": 0.00018224750952806624, + "loss": 0.7118, + "step": 1435 + }, + { + "epoch": 2.742857142857143, + "grad_norm": 0.20584052801132202, + "learning_rate": 0.00018205796105083915, + "loss": 0.712, + "step": 1440 + }, + { + "epoch": 2.7523809523809524, + "grad_norm": 0.19385926425457, + "learning_rate": 0.00018186750567977637, + "loss": 0.711, + "step": 1445 + }, + { + "epoch": 2.761904761904762, + "grad_norm": 0.1934896558523178, + "learning_rate": 0.00018167614551976567, + "loss": 0.7109, + "step": 1450 + }, + { + "epoch": 2.7714285714285714, + "grad_norm": 0.2056296020746231, + "learning_rate": 0.00018148388268569453, + "loss": 0.7139, + "step": 1455 + }, + { + "epoch": 2.780952380952381, + "grad_norm": 0.25994202494621277, + "learning_rate": 0.00018129071930242648, + "loss": 0.7144, + "step": 1460 + }, + { + "epoch": 2.7904761904761903, + "grad_norm": 0.19467397034168243, + "learning_rate": 0.00018109665750477806, + "loss": 0.7075, + "step": 1465 + }, + { + "epoch": 2.8, + "grad_norm": 0.2021484375, + "learning_rate": 0.00018090169943749476, + "loss": 0.7032, + "step": 1470 + }, + { + "epoch": 2.8095238095238093, + "grad_norm": 0.19192054867744446, + "learning_rate": 0.00018070584725522762, + "loss": 0.6992, + "step": 1475 + }, + { + "epoch": 2.819047619047619, + "grad_norm": 0.19371001422405243, + "learning_rate": 0.00018050910312250931, + "loss": 0.6881, + "step": 1480 + }, + { + "epoch": 2.8285714285714287, + "grad_norm": 0.20279107987880707, + "learning_rate": 0.00018031146921373018, + "loss": 0.6949, + "step": 1485 + }, + { + "epoch": 2.8380952380952382, + "grad_norm": 0.1979864537715912, + "learning_rate": 0.00018011294771311435, + "loss": 0.7011, + "step": 1490 + }, + { + "epoch": 2.8476190476190477, + "grad_norm": 0.18850283324718475, + "learning_rate": 0.00017991354081469538, + "loss": 0.6992, + "step": 1495 + }, + { + "epoch": 2.857142857142857, + "grad_norm": 0.1913542002439499, + "learning_rate": 0.00017971325072229226, + "loss": 0.7061, + "step": 1500 + }, + { + "epoch": 2.8666666666666667, + "grad_norm": 0.1887340098619461, + "learning_rate": 0.0001795120796494848, + "loss": 0.7039, + "step": 1505 + }, + { + "epoch": 2.876190476190476, + "grad_norm": 0.1913866549730301, + "learning_rate": 0.00017931002981958933, + "loss": 0.6995, + "step": 1510 + }, + { + "epoch": 2.8857142857142857, + "grad_norm": 0.201620414853096, + "learning_rate": 0.00017910710346563416, + "loss": 0.6939, + "step": 1515 + }, + { + "epoch": 2.895238095238095, + "grad_norm": 0.20726467669010162, + "learning_rate": 0.00017890330283033468, + "loss": 0.7071, + "step": 1520 + }, + { + "epoch": 2.9047619047619047, + "grad_norm": 0.2361179143190384, + "learning_rate": 0.0001786986301660689, + "loss": 0.6983, + "step": 1525 + }, + { + "epoch": 2.914285714285714, + "grad_norm": 0.19285356998443604, + "learning_rate": 0.00017849308773485226, + "loss": 0.6997, + "step": 1530 + }, + { + "epoch": 2.923809523809524, + "grad_norm": 0.19193626940250397, + "learning_rate": 0.00017828667780831278, + "loss": 0.6892, + "step": 1535 + }, + { + "epoch": 2.9333333333333336, + "grad_norm": 0.19089607894420624, + "learning_rate": 0.00017807940266766593, + "loss": 0.7065, + "step": 1540 + }, + { + "epoch": 2.942857142857143, + "grad_norm": 0.19156192243099213, + "learning_rate": 0.0001778712646036894, + "loss": 0.7004, + "step": 1545 + }, + { + "epoch": 2.9523809523809526, + "grad_norm": 0.1920241266489029, + "learning_rate": 0.00017766226591669785, + "loss": 0.6856, + "step": 1550 + }, + { + "epoch": 2.961904761904762, + "grad_norm": 0.1933414787054062, + "learning_rate": 0.00017745240891651735, + "loss": 0.6941, + "step": 1555 + }, + { + "epoch": 2.9714285714285715, + "grad_norm": 0.19441571831703186, + "learning_rate": 0.00017724169592245995, + "loss": 0.7047, + "step": 1560 + }, + { + "epoch": 2.980952380952381, + "grad_norm": 0.19140082597732544, + "learning_rate": 0.00017703012926329815, + "loss": 0.6935, + "step": 1565 + }, + { + "epoch": 2.9904761904761905, + "grad_norm": 0.18173792958259583, + "learning_rate": 0.0001768177112772388, + "loss": 0.6973, + "step": 1570 + }, + { + "epoch": 3.0, + "grad_norm": 0.1952088624238968, + "learning_rate": 0.0001766044443118978, + "loss": 0.7026, + "step": 1575 + }, + { + "epoch": 3.0, + "eval_loss": 1.3116141557693481, + "eval_runtime": 0.3239, + "eval_samples_per_second": 12.349, + "eval_steps_per_second": 3.087, + "step": 1575 + }, + { + "epoch": 3.0095238095238095, + "grad_norm": 0.19785767793655396, + "learning_rate": 0.00017639033072427366, + "loss": 0.6929, + "step": 1580 + }, + { + "epoch": 3.019047619047619, + "grad_norm": 0.2005433589220047, + "learning_rate": 0.0001761753728807217, + "loss": 0.6741, + "step": 1585 + }, + { + "epoch": 3.0285714285714285, + "grad_norm": 0.2056550532579422, + "learning_rate": 0.00017595957315692782, + "loss": 0.688, + "step": 1590 + }, + { + "epoch": 3.038095238095238, + "grad_norm": 0.674697756767273, + "learning_rate": 0.00017574293393788235, + "loss": 0.6673, + "step": 1595 + }, + { + "epoch": 3.0476190476190474, + "grad_norm": 0.20014941692352295, + "learning_rate": 0.0001755254576178535, + "loss": 0.682, + "step": 1600 + }, + { + "epoch": 3.057142857142857, + "grad_norm": 0.27740803360939026, + "learning_rate": 0.00017530714660036112, + "loss": 0.6746, + "step": 1605 + }, + { + "epoch": 3.066666666666667, + "grad_norm": 0.1996471881866455, + "learning_rate": 0.00017508800329814995, + "loss": 0.6926, + "step": 1610 + }, + { + "epoch": 3.0761904761904764, + "grad_norm": 0.21270829439163208, + "learning_rate": 0.000174868030133163, + "loss": 0.6846, + "step": 1615 + }, + { + "epoch": 3.085714285714286, + "grad_norm": 0.21131370961666107, + "learning_rate": 0.00017464722953651504, + "loss": 0.684, + "step": 1620 + }, + { + "epoch": 3.0952380952380953, + "grad_norm": 0.19949893653392792, + "learning_rate": 0.00017442560394846516, + "loss": 0.6814, + "step": 1625 + }, + { + "epoch": 3.104761904761905, + "grad_norm": 0.20186954736709595, + "learning_rate": 0.00017420315581839044, + "loss": 0.6892, + "step": 1630 + }, + { + "epoch": 3.1142857142857143, + "grad_norm": 0.19453243911266327, + "learning_rate": 0.0001739798876047584, + "loss": 0.6695, + "step": 1635 + }, + { + "epoch": 3.123809523809524, + "grad_norm": 0.19434688985347748, + "learning_rate": 0.00017375580177510016, + "loss": 0.68, + "step": 1640 + }, + { + "epoch": 3.1333333333333333, + "grad_norm": 0.8618267178535461, + "learning_rate": 0.0001735309008059829, + "loss": 0.6898, + "step": 1645 + }, + { + "epoch": 3.142857142857143, + "grad_norm": 0.2208489179611206, + "learning_rate": 0.00017330518718298264, + "loss": 0.6874, + "step": 1650 + }, + { + "epoch": 3.1523809523809523, + "grad_norm": 0.2234090119600296, + "learning_rate": 0.00017307866340065685, + "loss": 0.6937, + "step": 1655 + }, + { + "epoch": 3.1619047619047618, + "grad_norm": 0.1979946345090866, + "learning_rate": 0.00017285133196251663, + "loss": 0.6724, + "step": 1660 + }, + { + "epoch": 3.1714285714285713, + "grad_norm": 0.19695109128952026, + "learning_rate": 0.0001726231953809993, + "loss": 0.6995, + "step": 1665 + }, + { + "epoch": 3.1809523809523808, + "grad_norm": 0.21705491840839386, + "learning_rate": 0.00017239425617744048, + "loss": 0.6741, + "step": 1670 + }, + { + "epoch": 3.1904761904761907, + "grad_norm": 0.20663681626319885, + "learning_rate": 0.0001721645168820462, + "loss": 0.6822, + "step": 1675 + }, + { + "epoch": 3.2, + "grad_norm": 0.20704145729541779, + "learning_rate": 0.0001719339800338651, + "loss": 0.691, + "step": 1680 + }, + { + "epoch": 3.2095238095238097, + "grad_norm": 0.20241037011146545, + "learning_rate": 0.00017170264818076026, + "loss": 0.695, + "step": 1685 + }, + { + "epoch": 3.219047619047619, + "grad_norm": 0.20969533920288086, + "learning_rate": 0.0001714705238793809, + "loss": 0.6893, + "step": 1690 + }, + { + "epoch": 3.2285714285714286, + "grad_norm": 0.19700659811496735, + "learning_rate": 0.0001712376096951345, + "loss": 0.6773, + "step": 1695 + }, + { + "epoch": 3.238095238095238, + "grad_norm": 0.21133695542812347, + "learning_rate": 0.00017100390820215804, + "loss": 0.6976, + "step": 1700 + }, + { + "epoch": 3.2476190476190476, + "grad_norm": 0.19604110717773438, + "learning_rate": 0.00017076942198328987, + "loss": 0.6872, + "step": 1705 + }, + { + "epoch": 3.257142857142857, + "grad_norm": 0.19546429812908173, + "learning_rate": 0.0001705341536300409, + "loss": 0.6919, + "step": 1710 + }, + { + "epoch": 3.2666666666666666, + "grad_norm": 0.1990043967962265, + "learning_rate": 0.0001702981057425662, + "loss": 0.6802, + "step": 1715 + }, + { + "epoch": 3.276190476190476, + "grad_norm": 0.20021270215511322, + "learning_rate": 0.00017006128092963605, + "loss": 0.6834, + "step": 1720 + }, + { + "epoch": 3.2857142857142856, + "grad_norm": 0.2077239453792572, + "learning_rate": 0.00016982368180860728, + "loss": 0.6876, + "step": 1725 + }, + { + "epoch": 3.295238095238095, + "grad_norm": 0.2189949005842209, + "learning_rate": 0.00016958531100539427, + "loss": 0.6837, + "step": 1730 + }, + { + "epoch": 3.3047619047619046, + "grad_norm": 0.2016359120607376, + "learning_rate": 0.00016934617115443992, + "loss": 0.6905, + "step": 1735 + }, + { + "epoch": 3.314285714285714, + "grad_norm": 0.18884477019309998, + "learning_rate": 0.00016910626489868649, + "loss": 0.7001, + "step": 1740 + }, + { + "epoch": 3.323809523809524, + "grad_norm": 0.1908182054758072, + "learning_rate": 0.00016886559488954648, + "loss": 0.6843, + "step": 1745 + }, + { + "epoch": 3.3333333333333335, + "grad_norm": 0.20577913522720337, + "learning_rate": 0.0001686241637868734, + "loss": 0.6845, + "step": 1750 + }, + { + "epoch": 3.342857142857143, + "grad_norm": 0.19659753143787384, + "learning_rate": 0.00016838197425893202, + "loss": 0.6744, + "step": 1755 + }, + { + "epoch": 3.3523809523809525, + "grad_norm": 0.2035914957523346, + "learning_rate": 0.00016813902898236939, + "loss": 0.6851, + "step": 1760 + }, + { + "epoch": 3.361904761904762, + "grad_norm": 0.2145099639892578, + "learning_rate": 0.00016789533064218485, + "loss": 0.6848, + "step": 1765 + }, + { + "epoch": 3.3714285714285714, + "grad_norm": 0.1982029229402542, + "learning_rate": 0.00016765088193170053, + "loss": 0.694, + "step": 1770 + }, + { + "epoch": 3.380952380952381, + "grad_norm": 0.2153388410806656, + "learning_rate": 0.00016740568555253155, + "loss": 0.6788, + "step": 1775 + }, + { + "epoch": 3.3904761904761904, + "grad_norm": 0.2058364748954773, + "learning_rate": 0.00016715974421455617, + "loss": 0.6855, + "step": 1780 + }, + { + "epoch": 3.4, + "grad_norm": 0.20370839536190033, + "learning_rate": 0.00016691306063588583, + "loss": 0.6989, + "step": 1785 + }, + { + "epoch": 3.4095238095238094, + "grad_norm": 0.19771145284175873, + "learning_rate": 0.00016666563754283515, + "loss": 0.6849, + "step": 1790 + }, + { + "epoch": 3.419047619047619, + "grad_norm": 0.20471984148025513, + "learning_rate": 0.0001664174776698917, + "loss": 0.6843, + "step": 1795 + }, + { + "epoch": 3.4285714285714284, + "grad_norm": 0.1989559680223465, + "learning_rate": 0.00016616858375968595, + "loss": 0.6805, + "step": 1800 + }, + { + "epoch": 3.4380952380952383, + "grad_norm": 0.20598559081554413, + "learning_rate": 0.00016591895856296073, + "loss": 0.6648, + "step": 1805 + }, + { + "epoch": 3.447619047619048, + "grad_norm": 0.22308282554149628, + "learning_rate": 0.00016566860483854104, + "loss": 0.703, + "step": 1810 + }, + { + "epoch": 3.4571428571428573, + "grad_norm": 0.20057585835456848, + "learning_rate": 0.00016541752535330345, + "loss": 0.6776, + "step": 1815 + }, + { + "epoch": 3.466666666666667, + "grad_norm": 0.19391490519046783, + "learning_rate": 0.00016516572288214552, + "loss": 0.6869, + "step": 1820 + }, + { + "epoch": 3.4761904761904763, + "grad_norm": 0.2048014998435974, + "learning_rate": 0.0001649132002079552, + "loss": 0.6825, + "step": 1825 + }, + { + "epoch": 3.4857142857142858, + "grad_norm": 0.19976018369197845, + "learning_rate": 0.00016465996012157995, + "loss": 0.6886, + "step": 1830 + }, + { + "epoch": 3.4952380952380953, + "grad_norm": 0.19274355471134186, + "learning_rate": 0.00016440600542179615, + "loss": 0.6869, + "step": 1835 + }, + { + "epoch": 3.5047619047619047, + "grad_norm": 0.20784519612789154, + "learning_rate": 0.0001641513389152777, + "loss": 0.6921, + "step": 1840 + }, + { + "epoch": 3.5142857142857142, + "grad_norm": 0.26407596468925476, + "learning_rate": 0.0001638959634165656, + "loss": 0.6869, + "step": 1845 + }, + { + "epoch": 3.5238095238095237, + "grad_norm": 0.20616909861564636, + "learning_rate": 0.00016363988174803638, + "loss": 0.6916, + "step": 1850 + }, + { + "epoch": 3.533333333333333, + "grad_norm": 0.21633301675319672, + "learning_rate": 0.00016338309673987101, + "loss": 0.6761, + "step": 1855 + }, + { + "epoch": 3.5428571428571427, + "grad_norm": 0.19825606048107147, + "learning_rate": 0.0001631256112300239, + "loss": 0.6871, + "step": 1860 + }, + { + "epoch": 3.552380952380952, + "grad_norm": 0.21330222487449646, + "learning_rate": 0.00016286742806419108, + "loss": 0.6774, + "step": 1865 + }, + { + "epoch": 3.5619047619047617, + "grad_norm": 0.20014986395835876, + "learning_rate": 0.0001626085500957791, + "loss": 0.6758, + "step": 1870 + }, + { + "epoch": 3.571428571428571, + "grad_norm": 0.21541140973567963, + "learning_rate": 0.00016234898018587337, + "loss": 0.6792, + "step": 1875 + }, + { + "epoch": 3.580952380952381, + "grad_norm": 0.19707003235816956, + "learning_rate": 0.0001620887212032065, + "loss": 0.6761, + "step": 1880 + }, + { + "epoch": 3.5904761904761906, + "grad_norm": 0.2032250314950943, + "learning_rate": 0.00016182777602412665, + "loss": 0.6836, + "step": 1885 + }, + { + "epoch": 3.6, + "grad_norm": 0.20881584286689758, + "learning_rate": 0.0001615661475325658, + "loss": 0.6894, + "step": 1890 + }, + { + "epoch": 3.6095238095238096, + "grad_norm": 0.20583228766918182, + "learning_rate": 0.0001613038386200078, + "loss": 0.6872, + "step": 1895 + }, + { + "epoch": 3.619047619047619, + "grad_norm": 0.1901794672012329, + "learning_rate": 0.00016104085218545633, + "loss": 0.6958, + "step": 1900 + }, + { + "epoch": 3.6285714285714286, + "grad_norm": 0.20203644037246704, + "learning_rate": 0.00016077719113540302, + "loss": 0.6973, + "step": 1905 + }, + { + "epoch": 3.638095238095238, + "grad_norm": 0.19529937207698822, + "learning_rate": 0.00016051285838379525, + "loss": 0.6917, + "step": 1910 + }, + { + "epoch": 3.6476190476190475, + "grad_norm": 0.19104109704494476, + "learning_rate": 0.00016024785685200395, + "loss": 0.6851, + "step": 1915 + }, + { + "epoch": 3.657142857142857, + "grad_norm": 0.20573295652866364, + "learning_rate": 0.00015998218946879138, + "loss": 0.6788, + "step": 1920 + }, + { + "epoch": 3.6666666666666665, + "grad_norm": 0.19037216901779175, + "learning_rate": 0.00015971585917027862, + "loss": 0.6834, + "step": 1925 + }, + { + "epoch": 3.6761904761904765, + "grad_norm": 0.21168257296085358, + "learning_rate": 0.00015944886889991325, + "loss": 0.6858, + "step": 1930 + }, + { + "epoch": 3.685714285714286, + "grad_norm": 0.19083018600940704, + "learning_rate": 0.00015918122160843678, + "loss": 0.6828, + "step": 1935 + }, + { + "epoch": 3.6952380952380954, + "grad_norm": 0.20215772092342377, + "learning_rate": 0.000158912920253852, + "loss": 0.6886, + "step": 1940 + }, + { + "epoch": 3.704761904761905, + "grad_norm": 0.19086819887161255, + "learning_rate": 0.0001586439678013903, + "loss": 0.6848, + "step": 1945 + }, + { + "epoch": 3.7142857142857144, + "grad_norm": 0.1868395358324051, + "learning_rate": 0.000158374367223479, + "loss": 0.676, + "step": 1950 + }, + { + "epoch": 3.723809523809524, + "grad_norm": 0.20506930351257324, + "learning_rate": 0.00015810412149970833, + "loss": 0.6799, + "step": 1955 + }, + { + "epoch": 3.7333333333333334, + "grad_norm": 0.19333495199680328, + "learning_rate": 0.00015783323361679864, + "loss": 0.694, + "step": 1960 + }, + { + "epoch": 3.742857142857143, + "grad_norm": 0.22014223039150238, + "learning_rate": 0.00015756170656856737, + "loss": 0.7038, + "step": 1965 + }, + { + "epoch": 3.7523809523809524, + "grad_norm": 0.1922711282968521, + "learning_rate": 0.0001572895433558958, + "loss": 0.6888, + "step": 1970 + }, + { + "epoch": 3.761904761904762, + "grad_norm": 0.19480884075164795, + "learning_rate": 0.0001570167469866962, + "loss": 0.6851, + "step": 1975 + }, + { + "epoch": 3.7714285714285714, + "grad_norm": 0.19552657008171082, + "learning_rate": 0.0001567433204758782, + "loss": 0.683, + "step": 1980 + }, + { + "epoch": 3.780952380952381, + "grad_norm": 0.19440296292304993, + "learning_rate": 0.00015646926684531585, + "loss": 0.6828, + "step": 1985 + }, + { + "epoch": 3.7904761904761903, + "grad_norm": 0.19595205783843994, + "learning_rate": 0.00015619458912381396, + "loss": 0.6838, + "step": 1990 + }, + { + "epoch": 3.8, + "grad_norm": 0.19198662042617798, + "learning_rate": 0.0001559192903470747, + "loss": 0.6851, + "step": 1995 + }, + { + "epoch": 3.8095238095238093, + "grad_norm": 0.21291592717170715, + "learning_rate": 0.00015564337355766412, + "loss": 0.6873, + "step": 2000 + }, + { + "epoch": 3.819047619047619, + "grad_norm": 0.19576378166675568, + "learning_rate": 0.0001553668418049784, + "loss": 0.6793, + "step": 2005 + }, + { + "epoch": 3.8285714285714287, + "grad_norm": 0.23415517807006836, + "learning_rate": 0.00015508969814521025, + "loss": 0.6866, + "step": 2010 + }, + { + "epoch": 3.8380952380952382, + "grad_norm": 0.20917555689811707, + "learning_rate": 0.00015481194564131512, + "loss": 0.6802, + "step": 2015 + }, + { + "epoch": 3.8476190476190477, + "grad_norm": 0.19065144658088684, + "learning_rate": 0.00015453358736297729, + "loss": 0.6795, + "step": 2020 + }, + { + "epoch": 3.857142857142857, + "grad_norm": 0.2099611908197403, + "learning_rate": 0.00015425462638657595, + "loss": 0.6907, + "step": 2025 + }, + { + "epoch": 3.8666666666666667, + "grad_norm": 0.2239813506603241, + "learning_rate": 0.0001539750657951513, + "loss": 0.6878, + "step": 2030 + }, + { + "epoch": 3.876190476190476, + "grad_norm": 0.19681893289089203, + "learning_rate": 0.00015369490867837035, + "loss": 0.681, + "step": 2035 + }, + { + "epoch": 3.8857142857142857, + "grad_norm": 0.20367826521396637, + "learning_rate": 0.00015341415813249288, + "loss": 0.6942, + "step": 2040 + }, + { + "epoch": 3.895238095238095, + "grad_norm": 0.21106091141700745, + "learning_rate": 0.00015313281726033715, + "loss": 0.6846, + "step": 2045 + }, + { + "epoch": 3.9047619047619047, + "grad_norm": 0.2056264877319336, + "learning_rate": 0.00015285088917124556, + "loss": 0.6911, + "step": 2050 + }, + { + "epoch": 3.914285714285714, + "grad_norm": 0.19982090592384338, + "learning_rate": 0.00015256837698105047, + "loss": 0.6846, + "step": 2055 + }, + { + "epoch": 3.923809523809524, + "grad_norm": 0.2040010392665863, + "learning_rate": 0.00015228528381203962, + "loss": 0.6866, + "step": 2060 + }, + { + "epoch": 3.9333333333333336, + "grad_norm": 0.19176547229290009, + "learning_rate": 0.00015200161279292155, + "loss": 0.6765, + "step": 2065 + }, + { + "epoch": 3.942857142857143, + "grad_norm": 0.1905089169740677, + "learning_rate": 0.00015171736705879126, + "loss": 0.6947, + "step": 2070 + }, + { + "epoch": 3.9523809523809526, + "grad_norm": 0.20202653110027313, + "learning_rate": 0.00015143254975109538, + "loss": 0.6977, + "step": 2075 + }, + { + "epoch": 3.961904761904762, + "grad_norm": 0.200669065117836, + "learning_rate": 0.0001511471640175974, + "loss": 0.6864, + "step": 2080 + }, + { + "epoch": 3.9714285714285715, + "grad_norm": 0.19808419048786163, + "learning_rate": 0.00015086121301234316, + "loss": 0.6777, + "step": 2085 + }, + { + "epoch": 3.980952380952381, + "grad_norm": 0.1924077272415161, + "learning_rate": 0.00015057469989562567, + "loss": 0.6869, + "step": 2090 + }, + { + "epoch": 3.9904761904761905, + "grad_norm": 0.21015213429927826, + "learning_rate": 0.00015028762783395034, + "loss": 0.6822, + "step": 2095 + }, + { + "epoch": 4.0, + "grad_norm": 0.20121638476848602, + "learning_rate": 0.00015000000000000001, + "loss": 0.6682, + "step": 2100 + }, + { + "epoch": 4.0, + "eval_loss": 1.3089699745178223, + "eval_runtime": 0.3257, + "eval_samples_per_second": 12.281, + "eval_steps_per_second": 3.07, + "step": 2100 + }, + { + "epoch": 4.0095238095238095, + "grad_norm": 0.28249043226242065, + "learning_rate": 0.0001497118195725998, + "loss": 0.6699, + "step": 2105 + }, + { + "epoch": 4.019047619047619, + "grad_norm": 0.19913609325885773, + "learning_rate": 0.0001494230897366821, + "loss": 0.6576, + "step": 2110 + }, + { + "epoch": 4.0285714285714285, + "grad_norm": 0.2098485380411148, + "learning_rate": 0.00014913381368325115, + "loss": 0.6695, + "step": 2115 + }, + { + "epoch": 4.038095238095238, + "grad_norm": 0.20271775126457214, + "learning_rate": 0.00014884399460934805, + "loss": 0.6792, + "step": 2120 + }, + { + "epoch": 4.0476190476190474, + "grad_norm": 0.1934160739183426, + "learning_rate": 0.00014855363571801523, + "loss": 0.6703, + "step": 2125 + }, + { + "epoch": 4.057142857142857, + "grad_norm": 0.205865740776062, + "learning_rate": 0.0001482627402182611, + "loss": 0.6614, + "step": 2130 + }, + { + "epoch": 4.066666666666666, + "grad_norm": 0.19480110704898834, + "learning_rate": 0.00014797131132502465, + "loss": 0.6589, + "step": 2135 + }, + { + "epoch": 4.076190476190476, + "grad_norm": 0.19697819650173187, + "learning_rate": 0.00014767935225913975, + "loss": 0.6762, + "step": 2140 + }, + { + "epoch": 4.085714285714285, + "grad_norm": 0.20213504135608673, + "learning_rate": 0.00014738686624729986, + "loss": 0.6664, + "step": 2145 + }, + { + "epoch": 4.095238095238095, + "grad_norm": 0.19599653780460358, + "learning_rate": 0.00014709385652202203, + "loss": 0.6727, + "step": 2150 + }, + { + "epoch": 4.104761904761904, + "grad_norm": 0.201912984251976, + "learning_rate": 0.0001468003263216113, + "loss": 0.6554, + "step": 2155 + }, + { + "epoch": 4.114285714285714, + "grad_norm": 0.19933155179023743, + "learning_rate": 0.00014650627889012507, + "loss": 0.6704, + "step": 2160 + }, + { + "epoch": 4.123809523809523, + "grad_norm": 0.19774621725082397, + "learning_rate": 0.00014621171747733697, + "loss": 0.6582, + "step": 2165 + }, + { + "epoch": 4.133333333333334, + "grad_norm": 0.1958772838115692, + "learning_rate": 0.00014591664533870118, + "loss": 0.6744, + "step": 2170 + }, + { + "epoch": 4.142857142857143, + "grad_norm": 0.20352153480052948, + "learning_rate": 0.0001456210657353163, + "loss": 0.6655, + "step": 2175 + }, + { + "epoch": 4.152380952380953, + "grad_norm": 0.20768842101097107, + "learning_rate": 0.0001453249819338894, + "loss": 0.68, + "step": 2180 + }, + { + "epoch": 4.161904761904762, + "grad_norm": 0.21102721989154816, + "learning_rate": 0.00014502839720669989, + "loss": 0.651, + "step": 2185 + }, + { + "epoch": 4.171428571428572, + "grad_norm": 0.2110520899295807, + "learning_rate": 0.00014473131483156327, + "loss": 0.6647, + "step": 2190 + }, + { + "epoch": 4.180952380952381, + "grad_norm": 0.2165745049715042, + "learning_rate": 0.00014443373809179508, + "loss": 0.6762, + "step": 2195 + }, + { + "epoch": 4.190476190476191, + "grad_norm": 0.21253971755504608, + "learning_rate": 0.0001441356702761744, + "loss": 0.668, + "step": 2200 + }, + { + "epoch": 4.2, + "grad_norm": 0.21589303016662598, + "learning_rate": 0.00014383711467890774, + "loss": 0.6643, + "step": 2205 + }, + { + "epoch": 4.20952380952381, + "grad_norm": 0.19887423515319824, + "learning_rate": 0.00014353807459959242, + "loss": 0.665, + "step": 2210 + }, + { + "epoch": 4.219047619047619, + "grad_norm": 0.21460965275764465, + "learning_rate": 0.00014323855334318026, + "loss": 0.6661, + "step": 2215 + }, + { + "epoch": 4.228571428571429, + "grad_norm": 0.20307758450508118, + "learning_rate": 0.00014293855421994094, + "loss": 0.6564, + "step": 2220 + }, + { + "epoch": 4.238095238095238, + "grad_norm": 0.20516367256641388, + "learning_rate": 0.0001426380805454254, + "loss": 0.6694, + "step": 2225 + }, + { + "epoch": 4.247619047619048, + "grad_norm": 0.20536041259765625, + "learning_rate": 0.00014233713564042937, + "loss": 0.6724, + "step": 2230 + }, + { + "epoch": 4.257142857142857, + "grad_norm": 0.25682541728019714, + "learning_rate": 0.00014203572283095657, + "loss": 0.6835, + "step": 2235 + }, + { + "epoch": 4.266666666666667, + "grad_norm": 0.20103533565998077, + "learning_rate": 0.0001417338454481818, + "loss": 0.66, + "step": 2240 + }, + { + "epoch": 4.276190476190476, + "grad_norm": 0.22302131354808807, + "learning_rate": 0.00014143150682841438, + "loss": 0.6786, + "step": 2245 + }, + { + "epoch": 4.285714285714286, + "grad_norm": 0.20563514530658722, + "learning_rate": 0.00014112871031306119, + "loss": 0.6642, + "step": 2250 + }, + { + "epoch": 4.295238095238095, + "grad_norm": 0.20527109503746033, + "learning_rate": 0.00014082545924858954, + "loss": 0.6601, + "step": 2255 + }, + { + "epoch": 4.304761904761905, + "grad_norm": 0.20536208152770996, + "learning_rate": 0.00014052175698649053, + "loss": 0.6609, + "step": 2260 + }, + { + "epoch": 4.314285714285714, + "grad_norm": 0.19894732534885406, + "learning_rate": 0.00014021760688324176, + "loss": 0.6655, + "step": 2265 + }, + { + "epoch": 4.3238095238095235, + "grad_norm": 0.20036424696445465, + "learning_rate": 0.0001399130123002703, + "loss": 0.6615, + "step": 2270 + }, + { + "epoch": 4.333333333333333, + "grad_norm": 0.19823205471038818, + "learning_rate": 0.0001396079766039157, + "loss": 0.6608, + "step": 2275 + }, + { + "epoch": 4.3428571428571425, + "grad_norm": 0.1884753406047821, + "learning_rate": 0.00013930250316539238, + "loss": 0.6716, + "step": 2280 + }, + { + "epoch": 4.352380952380952, + "grad_norm": 0.21051181852817535, + "learning_rate": 0.0001389965953607528, + "loss": 0.6822, + "step": 2285 + }, + { + "epoch": 4.3619047619047615, + "grad_norm": 0.2015211582183838, + "learning_rate": 0.00013869025657084995, + "loss": 0.673, + "step": 2290 + }, + { + "epoch": 4.371428571428572, + "grad_norm": 0.19995340704917908, + "learning_rate": 0.00013838349018130007, + "loss": 0.6659, + "step": 2295 + }, + { + "epoch": 4.380952380952381, + "grad_norm": 0.20316636562347412, + "learning_rate": 0.00013807629958244498, + "loss": 0.6639, + "step": 2300 + }, + { + "epoch": 4.390476190476191, + "grad_norm": 0.2030089646577835, + "learning_rate": 0.00013776868816931502, + "loss": 0.6668, + "step": 2305 + }, + { + "epoch": 4.4, + "grad_norm": 0.19299709796905518, + "learning_rate": 0.00013746065934159123, + "loss": 0.646, + "step": 2310 + }, + { + "epoch": 4.40952380952381, + "grad_norm": 0.20314723253250122, + "learning_rate": 0.0001371522165035678, + "loss": 0.6961, + "step": 2315 + }, + { + "epoch": 4.419047619047619, + "grad_norm": 0.20023757219314575, + "learning_rate": 0.00013684336306411468, + "loss": 0.6673, + "step": 2320 + }, + { + "epoch": 4.428571428571429, + "grad_norm": 0.20633894205093384, + "learning_rate": 0.00013653410243663952, + "loss": 0.6707, + "step": 2325 + }, + { + "epoch": 4.438095238095238, + "grad_norm": 0.1962614208459854, + "learning_rate": 0.00013622443803905027, + "loss": 0.6777, + "step": 2330 + }, + { + "epoch": 4.447619047619048, + "grad_norm": 0.1993137001991272, + "learning_rate": 0.00013591437329371736, + "loss": 0.6661, + "step": 2335 + }, + { + "epoch": 4.457142857142857, + "grad_norm": 0.22276465594768524, + "learning_rate": 0.00013560391162743569, + "loss": 0.6728, + "step": 2340 + }, + { + "epoch": 4.466666666666667, + "grad_norm": 0.20663803815841675, + "learning_rate": 0.00013529305647138687, + "loss": 0.661, + "step": 2345 + }, + { + "epoch": 4.476190476190476, + "grad_norm": 0.20806623995304108, + "learning_rate": 0.0001349818112611015, + "loss": 0.6642, + "step": 2350 + }, + { + "epoch": 4.485714285714286, + "grad_norm": 0.20528347790241241, + "learning_rate": 0.00013467017943642073, + "loss": 0.6891, + "step": 2355 + }, + { + "epoch": 4.495238095238095, + "grad_norm": 0.20824523270130157, + "learning_rate": 0.0001343581644414587, + "loss": 0.6698, + "step": 2360 + }, + { + "epoch": 4.504761904761905, + "grad_norm": 0.19966593384742737, + "learning_rate": 0.00013404576972456431, + "loss": 0.6745, + "step": 2365 + }, + { + "epoch": 4.514285714285714, + "grad_norm": 0.20751707255840302, + "learning_rate": 0.00013373299873828303, + "loss": 0.6647, + "step": 2370 + }, + { + "epoch": 4.523809523809524, + "grad_norm": 0.21567291021347046, + "learning_rate": 0.00013341985493931877, + "loss": 0.6636, + "step": 2375 + }, + { + "epoch": 4.533333333333333, + "grad_norm": 0.2077573835849762, + "learning_rate": 0.0001331063417884958, + "loss": 0.6649, + "step": 2380 + }, + { + "epoch": 4.542857142857143, + "grad_norm": 0.1955225169658661, + "learning_rate": 0.00013279246275072046, + "loss": 0.6729, + "step": 2385 + }, + { + "epoch": 4.552380952380952, + "grad_norm": 0.19223564863204956, + "learning_rate": 0.00013247822129494266, + "loss": 0.665, + "step": 2390 + }, + { + "epoch": 4.561904761904762, + "grad_norm": 0.2114327996969223, + "learning_rate": 0.00013216362089411783, + "loss": 0.6588, + "step": 2395 + }, + { + "epoch": 4.571428571428571, + "grad_norm": 0.20199786126613617, + "learning_rate": 0.00013184866502516845, + "loss": 0.6597, + "step": 2400 + }, + { + "epoch": 4.580952380952381, + "grad_norm": 0.21868473291397095, + "learning_rate": 0.00013153335716894544, + "loss": 0.6785, + "step": 2405 + }, + { + "epoch": 4.59047619047619, + "grad_norm": 0.20326976478099823, + "learning_rate": 0.00013121770081018998, + "loss": 0.679, + "step": 2410 + }, + { + "epoch": 4.6, + "grad_norm": 0.2070162296295166, + "learning_rate": 0.00013090169943749476, + "loss": 0.6778, + "step": 2415 + }, + { + "epoch": 4.609523809523809, + "grad_norm": 0.20076832175254822, + "learning_rate": 0.00013058535654326554, + "loss": 0.6707, + "step": 2420 + }, + { + "epoch": 4.619047619047619, + "grad_norm": 0.20883376896381378, + "learning_rate": 0.0001302686756236826, + "loss": 0.68, + "step": 2425 + }, + { + "epoch": 4.628571428571428, + "grad_norm": 0.20020318031311035, + "learning_rate": 0.00012995166017866193, + "loss": 0.6657, + "step": 2430 + }, + { + "epoch": 4.638095238095238, + "grad_norm": 0.21386635303497314, + "learning_rate": 0.00012963431371181672, + "loss": 0.6802, + "step": 2435 + }, + { + "epoch": 4.647619047619048, + "grad_norm": 0.1988811492919922, + "learning_rate": 0.00012931663973041855, + "loss": 0.6674, + "step": 2440 + }, + { + "epoch": 4.6571428571428575, + "grad_norm": 0.20621821284294128, + "learning_rate": 0.00012899864174535864, + "loss": 0.6794, + "step": 2445 + }, + { + "epoch": 4.666666666666667, + "grad_norm": 0.20699866116046906, + "learning_rate": 0.00012868032327110904, + "loss": 0.6751, + "step": 2450 + }, + { + "epoch": 4.6761904761904765, + "grad_norm": 0.2192639857530594, + "learning_rate": 0.00012836168782568385, + "loss": 0.6654, + "step": 2455 + }, + { + "epoch": 4.685714285714286, + "grad_norm": 0.19943931698799133, + "learning_rate": 0.00012804273893060028, + "loss": 0.6686, + "step": 2460 + }, + { + "epoch": 4.695238095238095, + "grad_norm": 0.20029397308826447, + "learning_rate": 0.00012772348011083973, + "loss": 0.6742, + "step": 2465 + }, + { + "epoch": 4.704761904761905, + "grad_norm": 0.20713527500629425, + "learning_rate": 0.00012740391489480884, + "loss": 0.6718, + "step": 2470 + }, + { + "epoch": 4.714285714285714, + "grad_norm": 0.20206685364246368, + "learning_rate": 0.00012708404681430053, + "loss": 0.6717, + "step": 2475 + }, + { + "epoch": 4.723809523809524, + "grad_norm": 0.2047334611415863, + "learning_rate": 0.0001267638794044549, + "loss": 0.666, + "step": 2480 + }, + { + "epoch": 4.733333333333333, + "grad_norm": 0.20825915038585663, + "learning_rate": 0.00012644341620372023, + "loss": 0.6734, + "step": 2485 + }, + { + "epoch": 4.742857142857143, + "grad_norm": 0.2174612432718277, + "learning_rate": 0.00012612266075381386, + "loss": 0.6808, + "step": 2490 + }, + { + "epoch": 4.752380952380952, + "grad_norm": 0.2005445957183838, + "learning_rate": 0.00012580161659968294, + "loss": 0.6694, + "step": 2495 + }, + { + "epoch": 4.761904761904762, + "grad_norm": 0.20034897327423096, + "learning_rate": 0.0001254802872894655, + "loss": 0.6857, + "step": 2500 + }, + { + "epoch": 4.771428571428571, + "grad_norm": 0.2014506757259369, + "learning_rate": 0.00012515867637445086, + "loss": 0.6832, + "step": 2505 + }, + { + "epoch": 4.780952380952381, + "grad_norm": 0.21148435771465302, + "learning_rate": 0.00012483678740904082, + "loss": 0.6684, + "step": 2510 + }, + { + "epoch": 4.79047619047619, + "grad_norm": 0.21509744226932526, + "learning_rate": 0.00012451462395071, + "loss": 0.6692, + "step": 2515 + }, + { + "epoch": 4.8, + "grad_norm": 0.20884525775909424, + "learning_rate": 0.00012419218955996676, + "loss": 0.6753, + "step": 2520 + }, + { + "epoch": 4.809523809523809, + "grad_norm": 0.19881895184516907, + "learning_rate": 0.0001238694878003138, + "loss": 0.6713, + "step": 2525 + }, + { + "epoch": 4.819047619047619, + "grad_norm": 0.2082035094499588, + "learning_rate": 0.00012354652223820858, + "loss": 0.6762, + "step": 2530 + }, + { + "epoch": 4.828571428571428, + "grad_norm": 0.20466946065425873, + "learning_rate": 0.00012322329644302426, + "loss": 0.6581, + "step": 2535 + }, + { + "epoch": 4.838095238095238, + "grad_norm": 0.2181474268436432, + "learning_rate": 0.00012289981398700995, + "loss": 0.6621, + "step": 2540 + }, + { + "epoch": 4.847619047619047, + "grad_norm": 0.20615698397159576, + "learning_rate": 0.00012257607844525146, + "loss": 0.6689, + "step": 2545 + }, + { + "epoch": 4.857142857142857, + "grad_norm": 0.20270757377147675, + "learning_rate": 0.00012225209339563145, + "loss": 0.6753, + "step": 2550 + }, + { + "epoch": 4.866666666666667, + "grad_norm": 0.19730758666992188, + "learning_rate": 0.00012192786241879033, + "loss": 0.6691, + "step": 2555 + }, + { + "epoch": 4.876190476190477, + "grad_norm": 0.19855643808841705, + "learning_rate": 0.0001216033890980864, + "loss": 0.6686, + "step": 2560 + }, + { + "epoch": 4.885714285714286, + "grad_norm": 0.2075609266757965, + "learning_rate": 0.00012127867701955622, + "loss": 0.6698, + "step": 2565 + }, + { + "epoch": 4.895238095238096, + "grad_norm": 0.1980564445257187, + "learning_rate": 0.0001209537297718752, + "loss": 0.6738, + "step": 2570 + }, + { + "epoch": 4.904761904761905, + "grad_norm": 0.20312833786010742, + "learning_rate": 0.00012062855094631778, + "loss": 0.6615, + "step": 2575 + }, + { + "epoch": 4.914285714285715, + "grad_norm": 0.20446398854255676, + "learning_rate": 0.00012030314413671762, + "loss": 0.6674, + "step": 2580 + }, + { + "epoch": 4.923809523809524, + "grad_norm": 0.19080471992492676, + "learning_rate": 0.00011997751293942827, + "loss": 0.6662, + "step": 2585 + }, + { + "epoch": 4.933333333333334, + "grad_norm": 0.19706608355045319, + "learning_rate": 0.00011965166095328301, + "loss": 0.6728, + "step": 2590 + }, + { + "epoch": 4.942857142857143, + "grad_norm": 0.1980735808610916, + "learning_rate": 0.00011932559177955533, + "loss": 0.682, + "step": 2595 + }, + { + "epoch": 4.9523809523809526, + "grad_norm": 0.2065589874982834, + "learning_rate": 0.00011899930902191902, + "loss": 0.6628, + "step": 2600 + }, + { + "epoch": 4.961904761904762, + "grad_norm": 0.20613136887550354, + "learning_rate": 0.00011867281628640835, + "loss": 0.6709, + "step": 2605 + }, + { + "epoch": 4.9714285714285715, + "grad_norm": 0.20417103171348572, + "learning_rate": 0.00011834611718137824, + "loss": 0.6821, + "step": 2610 + }, + { + "epoch": 4.980952380952381, + "grad_norm": 0.1957087367773056, + "learning_rate": 0.00011801921531746444, + "loss": 0.671, + "step": 2615 + }, + { + "epoch": 4.9904761904761905, + "grad_norm": 0.20346994698047638, + "learning_rate": 0.00011769211430754357, + "loss": 0.6656, + "step": 2620 + }, + { + "epoch": 5.0, + "grad_norm": 0.20766137540340424, + "learning_rate": 0.00011736481776669306, + "loss": 0.6825, + "step": 2625 + }, + { + "epoch": 5.0, + "eval_loss": 1.312631368637085, + "eval_runtime": 0.3238, + "eval_samples_per_second": 12.354, + "eval_steps_per_second": 3.089, + "step": 2625 + }, + { + "epoch": 5.0095238095238095, + "grad_norm": 0.20680904388427734, + "learning_rate": 0.00011703732931215141, + "loss": 0.6432, + "step": 2630 + }, + { + "epoch": 5.019047619047619, + "grad_norm": 0.21646055579185486, + "learning_rate": 0.00011670965256327818, + "loss": 0.6509, + "step": 2635 + }, + { + "epoch": 5.0285714285714285, + "grad_norm": 0.19928839802742004, + "learning_rate": 0.00011638179114151377, + "loss": 0.6454, + "step": 2640 + }, + { + "epoch": 5.038095238095238, + "grad_norm": 0.21113866567611694, + "learning_rate": 0.00011605374867033977, + "loss": 0.652, + "step": 2645 + }, + { + "epoch": 5.0476190476190474, + "grad_norm": 0.21099436283111572, + "learning_rate": 0.00011572552877523854, + "loss": 0.6464, + "step": 2650 + }, + { + "epoch": 5.057142857142857, + "grad_norm": 0.21483275294303894, + "learning_rate": 0.00011539713508365335, + "loss": 0.6557, + "step": 2655 + }, + { + "epoch": 5.066666666666666, + "grad_norm": 0.24092718958854675, + "learning_rate": 0.00011506857122494831, + "loss": 0.6628, + "step": 2660 + }, + { + "epoch": 5.076190476190476, + "grad_norm": 0.2057563066482544, + "learning_rate": 0.00011473984083036813, + "loss": 0.6475, + "step": 2665 + }, + { + "epoch": 5.085714285714285, + "grad_norm": 0.20357713103294373, + "learning_rate": 0.00011441094753299801, + "loss": 0.6508, + "step": 2670 + }, + { + "epoch": 5.095238095238095, + "grad_norm": 0.21014833450317383, + "learning_rate": 0.00011408189496772368, + "loss": 0.6538, + "step": 2675 + }, + { + "epoch": 5.104761904761904, + "grad_norm": 0.22002901136875153, + "learning_rate": 0.00011375268677119089, + "loss": 0.6547, + "step": 2680 + }, + { + "epoch": 5.114285714285714, + "grad_norm": 0.2036370486021042, + "learning_rate": 0.00011342332658176555, + "loss": 0.6635, + "step": 2685 + }, + { + "epoch": 5.123809523809523, + "grad_norm": 0.22806309163570404, + "learning_rate": 0.00011309381803949333, + "loss": 0.6556, + "step": 2690 + }, + { + "epoch": 5.133333333333334, + "grad_norm": 0.20940910279750824, + "learning_rate": 0.00011276416478605949, + "loss": 0.6602, + "step": 2695 + }, + { + "epoch": 5.142857142857143, + "grad_norm": 0.21158607304096222, + "learning_rate": 0.00011243437046474853, + "loss": 0.6563, + "step": 2700 + }, + { + "epoch": 5.152380952380953, + "grad_norm": 0.2098589837551117, + "learning_rate": 0.00011210443872040414, + "loss": 0.6555, + "step": 2705 + }, + { + "epoch": 5.161904761904762, + "grad_norm": 0.2072463035583496, + "learning_rate": 0.00011177437319938875, + "loss": 0.6586, + "step": 2710 + }, + { + "epoch": 5.171428571428572, + "grad_norm": 0.21415144205093384, + "learning_rate": 0.0001114441775495432, + "loss": 0.6483, + "step": 2715 + }, + { + "epoch": 5.180952380952381, + "grad_norm": 0.21063564717769623, + "learning_rate": 0.00011111385542014663, + "loss": 0.6535, + "step": 2720 + }, + { + "epoch": 5.190476190476191, + "grad_norm": 0.225935161113739, + "learning_rate": 0.00011078341046187589, + "loss": 0.6597, + "step": 2725 + }, + { + "epoch": 5.2, + "grad_norm": 0.2182355672121048, + "learning_rate": 0.00011045284632676536, + "loss": 0.6437, + "step": 2730 + }, + { + "epoch": 5.20952380952381, + "grad_norm": 0.20297573506832123, + "learning_rate": 0.00011012216666816659, + "loss": 0.6479, + "step": 2735 + }, + { + "epoch": 5.219047619047619, + "grad_norm": 0.22021973133087158, + "learning_rate": 0.00010979137514070782, + "loss": 0.6556, + "step": 2740 + }, + { + "epoch": 5.228571428571429, + "grad_norm": 0.21546654403209686, + "learning_rate": 0.00010946047540025372, + "loss": 0.666, + "step": 2745 + }, + { + "epoch": 5.238095238095238, + "grad_norm": 0.21640990674495697, + "learning_rate": 0.00010912947110386484, + "loss": 0.6568, + "step": 2750 + }, + { + "epoch": 5.247619047619048, + "grad_norm": 0.21457038819789886, + "learning_rate": 0.00010879836590975731, + "loss": 0.6552, + "step": 2755 + }, + { + "epoch": 5.257142857142857, + "grad_norm": 0.21117344498634338, + "learning_rate": 0.00010846716347726233, + "loss": 0.6649, + "step": 2760 + }, + { + "epoch": 5.266666666666667, + "grad_norm": 0.21039853990077972, + "learning_rate": 0.00010813586746678583, + "loss": 0.6627, + "step": 2765 + }, + { + "epoch": 5.276190476190476, + "grad_norm": 0.21392405033111572, + "learning_rate": 0.00010780448153976793, + "loss": 0.6563, + "step": 2770 + }, + { + "epoch": 5.285714285714286, + "grad_norm": 0.21208111941814423, + "learning_rate": 0.00010747300935864243, + "loss": 0.666, + "step": 2775 + }, + { + "epoch": 5.295238095238095, + "grad_norm": 0.20485153794288635, + "learning_rate": 0.00010714145458679649, + "loss": 0.6537, + "step": 2780 + }, + { + "epoch": 5.304761904761905, + "grad_norm": 0.224151611328125, + "learning_rate": 0.00010680982088853002, + "loss": 0.6574, + "step": 2785 + }, + { + "epoch": 5.314285714285714, + "grad_norm": 0.2293432652950287, + "learning_rate": 0.00010647811192901518, + "loss": 0.6521, + "step": 2790 + }, + { + "epoch": 5.3238095238095235, + "grad_norm": 0.21431459486484528, + "learning_rate": 0.00010614633137425598, + "loss": 0.6594, + "step": 2795 + }, + { + "epoch": 5.333333333333333, + "grad_norm": 0.2100796401500702, + "learning_rate": 0.00010581448289104758, + "loss": 0.6493, + "step": 2800 + }, + { + "epoch": 5.3428571428571425, + "grad_norm": 0.20995594561100006, + "learning_rate": 0.00010548257014693601, + "loss": 0.657, + "step": 2805 + }, + { + "epoch": 5.352380952380952, + "grad_norm": 0.20966996252536774, + "learning_rate": 0.0001051505968101774, + "loss": 0.6603, + "step": 2810 + }, + { + "epoch": 5.3619047619047615, + "grad_norm": 0.22000014781951904, + "learning_rate": 0.00010481856654969758, + "loss": 0.6525, + "step": 2815 + }, + { + "epoch": 5.371428571428572, + "grad_norm": 0.2085055559873581, + "learning_rate": 0.00010448648303505151, + "loss": 0.6539, + "step": 2820 + }, + { + "epoch": 5.380952380952381, + "grad_norm": 0.2067158818244934, + "learning_rate": 0.00010415434993638269, + "loss": 0.655, + "step": 2825 + }, + { + "epoch": 5.390476190476191, + "grad_norm": 0.22521977126598358, + "learning_rate": 0.00010382217092438255, + "loss": 0.6621, + "step": 2830 + }, + { + "epoch": 5.4, + "grad_norm": 0.22613851726055145, + "learning_rate": 0.00010348994967025012, + "loss": 0.6542, + "step": 2835 + }, + { + "epoch": 5.40952380952381, + "grad_norm": 0.2127508819103241, + "learning_rate": 0.0001031576898456511, + "loss": 0.6455, + "step": 2840 + }, + { + "epoch": 5.419047619047619, + "grad_norm": 0.21902593970298767, + "learning_rate": 0.00010282539512267757, + "loss": 0.6449, + "step": 2845 + }, + { + "epoch": 5.428571428571429, + "grad_norm": 0.21509668231010437, + "learning_rate": 0.0001024930691738073, + "loss": 0.663, + "step": 2850 + }, + { + "epoch": 5.438095238095238, + "grad_norm": 0.2239295095205307, + "learning_rate": 0.00010216071567186312, + "loss": 0.6648, + "step": 2855 + }, + { + "epoch": 5.447619047619048, + "grad_norm": 0.20996320247650146, + "learning_rate": 0.00010182833828997238, + "loss": 0.6614, + "step": 2860 + }, + { + "epoch": 5.457142857142857, + "grad_norm": 0.2127327024936676, + "learning_rate": 0.00010149594070152638, + "loss": 0.6431, + "step": 2865 + }, + { + "epoch": 5.466666666666667, + "grad_norm": 0.20127449929714203, + "learning_rate": 0.00010116352658013973, + "loss": 0.6504, + "step": 2870 + }, + { + "epoch": 5.476190476190476, + "grad_norm": 0.21320253610610962, + "learning_rate": 0.00010083109959960973, + "loss": 0.6619, + "step": 2875 + }, + { + "epoch": 5.485714285714286, + "grad_norm": 0.2200571596622467, + "learning_rate": 0.00010049866343387581, + "loss": 0.6638, + "step": 2880 + }, + { + "epoch": 5.495238095238095, + "grad_norm": 0.21618418395519257, + "learning_rate": 0.00010016622175697898, + "loss": 0.6496, + "step": 2885 + }, + { + "epoch": 5.504761904761905, + "grad_norm": 0.2114275097846985, + "learning_rate": 9.983377824302106e-05, + "loss": 0.657, + "step": 2890 + }, + { + "epoch": 5.514285714285714, + "grad_norm": 0.21886934340000153, + "learning_rate": 9.950133656612421e-05, + "loss": 0.6493, + "step": 2895 + }, + { + "epoch": 5.523809523809524, + "grad_norm": 0.2145605832338333, + "learning_rate": 9.916890040039031e-05, + "loss": 0.6619, + "step": 2900 + }, + { + "epoch": 5.533333333333333, + "grad_norm": 0.21204130351543427, + "learning_rate": 9.883647341986032e-05, + "loss": 0.6664, + "step": 2905 + }, + { + "epoch": 5.542857142857143, + "grad_norm": 0.20686715841293335, + "learning_rate": 9.850405929847366e-05, + "loss": 0.6507, + "step": 2910 + }, + { + "epoch": 5.552380952380952, + "grad_norm": 0.21932560205459595, + "learning_rate": 9.817166171002765e-05, + "loss": 0.6579, + "step": 2915 + }, + { + "epoch": 5.561904761904762, + "grad_norm": 0.21961069107055664, + "learning_rate": 9.783928432813688e-05, + "loss": 0.65, + "step": 2920 + }, + { + "epoch": 5.571428571428571, + "grad_norm": 0.23916806280612946, + "learning_rate": 9.750693082619273e-05, + "loss": 0.6525, + "step": 2925 + }, + { + "epoch": 5.580952380952381, + "grad_norm": 0.20841637253761292, + "learning_rate": 9.717460487732245e-05, + "loss": 0.6553, + "step": 2930 + }, + { + "epoch": 5.59047619047619, + "grad_norm": 0.20813757181167603, + "learning_rate": 9.68423101543489e-05, + "loss": 0.6548, + "step": 2935 + }, + { + "epoch": 5.6, + "grad_norm": 0.21043892204761505, + "learning_rate": 9.651005032974994e-05, + "loss": 0.6486, + "step": 2940 + }, + { + "epoch": 5.609523809523809, + "grad_norm": 0.20587663352489471, + "learning_rate": 9.617782907561748e-05, + "loss": 0.6553, + "step": 2945 + }, + { + "epoch": 5.619047619047619, + "grad_norm": 0.21101590991020203, + "learning_rate": 9.584565006361734e-05, + "loss": 0.6454, + "step": 2950 + }, + { + "epoch": 5.628571428571428, + "grad_norm": 0.21065913140773773, + "learning_rate": 9.551351696494854e-05, + "loss": 0.6697, + "step": 2955 + }, + { + "epoch": 5.638095238095238, + "grad_norm": 0.21288540959358215, + "learning_rate": 9.518143345030246e-05, + "loss": 0.6571, + "step": 2960 + }, + { + "epoch": 5.647619047619048, + "grad_norm": 0.20005963742733002, + "learning_rate": 9.48494031898226e-05, + "loss": 0.67, + "step": 2965 + }, + { + "epoch": 5.6571428571428575, + "grad_norm": 0.20527806878089905, + "learning_rate": 9.451742985306398e-05, + "loss": 0.6589, + "step": 2970 + }, + { + "epoch": 5.666666666666667, + "grad_norm": 0.21453768014907837, + "learning_rate": 9.418551710895243e-05, + "loss": 0.6603, + "step": 2975 + }, + { + "epoch": 5.6761904761904765, + "grad_norm": 0.20562489330768585, + "learning_rate": 9.385366862574404e-05, + "loss": 0.6515, + "step": 2980 + }, + { + "epoch": 5.685714285714286, + "grad_norm": 0.21697136759757996, + "learning_rate": 9.352188807098481e-05, + "loss": 0.6558, + "step": 2985 + }, + { + "epoch": 5.695238095238095, + "grad_norm": 0.2106858789920807, + "learning_rate": 9.319017911147e-05, + "loss": 0.6527, + "step": 2990 + }, + { + "epoch": 5.704761904761905, + "grad_norm": 0.21161264181137085, + "learning_rate": 9.285854541320352e-05, + "loss": 0.6621, + "step": 2995 + }, + { + "epoch": 5.714285714285714, + "grad_norm": 0.2115710973739624, + "learning_rate": 9.252699064135758e-05, + "loss": 0.6545, + "step": 3000 + }, + { + "epoch": 5.723809523809524, + "grad_norm": 0.21325847506523132, + "learning_rate": 9.219551846023211e-05, + "loss": 0.6548, + "step": 3005 + }, + { + "epoch": 5.733333333333333, + "grad_norm": 0.2185753583908081, + "learning_rate": 9.186413253321418e-05, + "loss": 0.6563, + "step": 3010 + }, + { + "epoch": 5.742857142857143, + "grad_norm": 0.21847587823867798, + "learning_rate": 9.153283652273768e-05, + "loss": 0.6597, + "step": 3015 + }, + { + "epoch": 5.752380952380952, + "grad_norm": 0.22264257073402405, + "learning_rate": 9.120163409024271e-05, + "loss": 0.6555, + "step": 3020 + }, + { + "epoch": 5.761904761904762, + "grad_norm": 0.21518436074256897, + "learning_rate": 9.087052889613518e-05, + "loss": 0.6676, + "step": 3025 + }, + { + "epoch": 5.771428571428571, + "grad_norm": 0.21668776869773865, + "learning_rate": 9.05395245997463e-05, + "loss": 0.6638, + "step": 3030 + }, + { + "epoch": 5.780952380952381, + "grad_norm": 0.23866741359233856, + "learning_rate": 9.020862485929219e-05, + "loss": 0.6614, + "step": 3035 + }, + { + "epoch": 5.79047619047619, + "grad_norm": 0.21523581445217133, + "learning_rate": 8.987783333183344e-05, + "loss": 0.6421, + "step": 3040 + }, + { + "epoch": 5.8, + "grad_norm": 0.22735083103179932, + "learning_rate": 8.954715367323468e-05, + "loss": 0.6432, + "step": 3045 + }, + { + "epoch": 5.809523809523809, + "grad_norm": 0.20661625266075134, + "learning_rate": 8.921658953812415e-05, + "loss": 0.6475, + "step": 3050 + }, + { + "epoch": 5.819047619047619, + "grad_norm": 0.21183475852012634, + "learning_rate": 8.888614457985341e-05, + "loss": 0.6693, + "step": 3055 + }, + { + "epoch": 5.828571428571428, + "grad_norm": 0.2123451828956604, + "learning_rate": 8.855582245045683e-05, + "loss": 0.6406, + "step": 3060 + }, + { + "epoch": 5.838095238095238, + "grad_norm": 0.2733854651451111, + "learning_rate": 8.822562680061125e-05, + "loss": 0.6548, + "step": 3065 + }, + { + "epoch": 5.847619047619047, + "grad_norm": 0.2196865677833557, + "learning_rate": 8.789556127959585e-05, + "loss": 0.6513, + "step": 3070 + }, + { + "epoch": 5.857142857142857, + "grad_norm": 0.20598062872886658, + "learning_rate": 8.756562953525152e-05, + "loss": 0.663, + "step": 3075 + }, + { + "epoch": 5.866666666666667, + "grad_norm": 0.21008974313735962, + "learning_rate": 8.723583521394054e-05, + "loss": 0.6511, + "step": 3080 + }, + { + "epoch": 5.876190476190477, + "grad_norm": 0.21377860009670258, + "learning_rate": 8.690618196050666e-05, + "loss": 0.6601, + "step": 3085 + }, + { + "epoch": 5.885714285714286, + "grad_norm": 0.22736646234989166, + "learning_rate": 8.657667341823448e-05, + "loss": 0.6644, + "step": 3090 + }, + { + "epoch": 5.895238095238096, + "grad_norm": 0.20148470997810364, + "learning_rate": 8.624731322880912e-05, + "loss": 0.6517, + "step": 3095 + }, + { + "epoch": 5.904761904761905, + "grad_norm": 0.20768018066883087, + "learning_rate": 8.591810503227635e-05, + "loss": 0.6382, + "step": 3100 + }, + { + "epoch": 5.914285714285715, + "grad_norm": 0.21072714030742645, + "learning_rate": 8.558905246700201e-05, + "loss": 0.6539, + "step": 3105 + }, + { + "epoch": 5.923809523809524, + "grad_norm": 0.2142985463142395, + "learning_rate": 8.526015916963191e-05, + "loss": 0.6655, + "step": 3110 + }, + { + "epoch": 5.933333333333334, + "grad_norm": 0.2132679671049118, + "learning_rate": 8.49314287750517e-05, + "loss": 0.6733, + "step": 3115 + }, + { + "epoch": 5.942857142857143, + "grad_norm": 0.2185792326927185, + "learning_rate": 8.460286491634663e-05, + "loss": 0.6632, + "step": 3120 + }, + { + "epoch": 5.9523809523809526, + "grad_norm": 0.212264746427536, + "learning_rate": 8.427447122476148e-05, + "loss": 0.6484, + "step": 3125 + }, + { + "epoch": 5.961904761904762, + "grad_norm": 0.21196219325065613, + "learning_rate": 8.394625132966025e-05, + "loss": 0.6574, + "step": 3130 + }, + { + "epoch": 5.9714285714285715, + "grad_norm": 0.206856831908226, + "learning_rate": 8.361820885848624e-05, + "loss": 0.6597, + "step": 3135 + }, + { + "epoch": 5.980952380952381, + "grad_norm": 0.214055135846138, + "learning_rate": 8.329034743672187e-05, + "loss": 0.6464, + "step": 3140 + }, + { + "epoch": 5.9904761904761905, + "grad_norm": 0.21056272089481354, + "learning_rate": 8.296267068784862e-05, + "loss": 0.6465, + "step": 3145 + }, + { + "epoch": 6.0, + "grad_norm": 0.2108917087316513, + "learning_rate": 8.263518223330697e-05, + "loss": 0.6429, + "step": 3150 + }, + { + "epoch": 6.0, + "eval_loss": 1.3227609395980835, + "eval_runtime": 0.3235, + "eval_samples_per_second": 12.366, + "eval_steps_per_second": 3.092, + "step": 3150 + }, + { + "epoch": 6.0095238095238095, + "grad_norm": 0.21242249011993408, + "learning_rate": 8.230788569245648e-05, + "loss": 0.6373, + "step": 3155 + }, + { + "epoch": 6.019047619047619, + "grad_norm": 0.21833695471286774, + "learning_rate": 8.198078468253557e-05, + "loss": 0.6458, + "step": 3160 + }, + { + "epoch": 6.0285714285714285, + "grad_norm": 0.21860025823116302, + "learning_rate": 8.165388281862178e-05, + "loss": 0.6412, + "step": 3165 + }, + { + "epoch": 6.038095238095238, + "grad_norm": 0.2336193323135376, + "learning_rate": 8.132718371359166e-05, + "loss": 0.6468, + "step": 3170 + }, + { + "epoch": 6.0476190476190474, + "grad_norm": 0.2238347828388214, + "learning_rate": 8.100069097808103e-05, + "loss": 0.6318, + "step": 3175 + }, + { + "epoch": 6.057142857142857, + "grad_norm": 0.21204425394535065, + "learning_rate": 8.067440822044469e-05, + "loss": 0.6458, + "step": 3180 + }, + { + "epoch": 6.066666666666666, + "grad_norm": 0.22270914912223816, + "learning_rate": 8.034833904671698e-05, + "loss": 0.6386, + "step": 3185 + }, + { + "epoch": 6.076190476190476, + "grad_norm": 0.2162003070116043, + "learning_rate": 8.002248706057177e-05, + "loss": 0.6432, + "step": 3190 + }, + { + "epoch": 6.085714285714285, + "grad_norm": 0.21775895357131958, + "learning_rate": 7.96968558632824e-05, + "loss": 0.6471, + "step": 3195 + }, + { + "epoch": 6.095238095238095, + "grad_norm": 0.2277059406042099, + "learning_rate": 7.937144905368226e-05, + "loss": 0.6523, + "step": 3200 + }, + { + "epoch": 6.104761904761904, + "grad_norm": 0.21727009117603302, + "learning_rate": 7.904627022812483e-05, + "loss": 0.641, + "step": 3205 + }, + { + "epoch": 6.114285714285714, + "grad_norm": 0.21867170929908752, + "learning_rate": 7.872132298044382e-05, + "loss": 0.6504, + "step": 3210 + }, + { + "epoch": 6.123809523809523, + "grad_norm": 0.21738098561763763, + "learning_rate": 7.839661090191362e-05, + "loss": 0.6381, + "step": 3215 + }, + { + "epoch": 6.133333333333334, + "grad_norm": 0.21628010272979736, + "learning_rate": 7.807213758120966e-05, + "loss": 0.6503, + "step": 3220 + }, + { + "epoch": 6.142857142857143, + "grad_norm": 0.21854370832443237, + "learning_rate": 7.774790660436858e-05, + "loss": 0.6434, + "step": 3225 + }, + { + "epoch": 6.152380952380953, + "grad_norm": 0.22102175652980804, + "learning_rate": 7.742392155474858e-05, + "loss": 0.6506, + "step": 3230 + }, + { + "epoch": 6.161904761904762, + "grad_norm": 0.21102498471736908, + "learning_rate": 7.710018601299004e-05, + "loss": 0.6416, + "step": 3235 + }, + { + "epoch": 6.171428571428572, + "grad_norm": 0.2162449210882187, + "learning_rate": 7.677670355697577e-05, + "loss": 0.6511, + "step": 3240 + }, + { + "epoch": 6.180952380952381, + "grad_norm": 0.23095406591892242, + "learning_rate": 7.645347776179144e-05, + "loss": 0.6337, + "step": 3245 + }, + { + "epoch": 6.190476190476191, + "grad_norm": 0.22487138211727142, + "learning_rate": 7.613051219968623e-05, + "loss": 0.6332, + "step": 3250 + }, + { + "epoch": 6.2, + "grad_norm": 0.21796458959579468, + "learning_rate": 7.580781044003324e-05, + "loss": 0.6472, + "step": 3255 + }, + { + "epoch": 6.20952380952381, + "grad_norm": 0.24705886840820312, + "learning_rate": 7.548537604929001e-05, + "loss": 0.6506, + "step": 3260 + }, + { + "epoch": 6.219047619047619, + "grad_norm": 0.23071187734603882, + "learning_rate": 7.516321259095921e-05, + "loss": 0.6386, + "step": 3265 + }, + { + "epoch": 6.228571428571429, + "grad_norm": 0.2173680067062378, + "learning_rate": 7.484132362554915e-05, + "loss": 0.64, + "step": 3270 + }, + { + "epoch": 6.238095238095238, + "grad_norm": 1.5498923063278198, + "learning_rate": 7.451971271053455e-05, + "loss": 0.6438, + "step": 3275 + }, + { + "epoch": 6.247619047619048, + "grad_norm": 0.6907296180725098, + "learning_rate": 7.419838340031708e-05, + "loss": 0.65, + "step": 3280 + }, + { + "epoch": 6.257142857142857, + "grad_norm": 0.23223426938056946, + "learning_rate": 7.387733924618617e-05, + "loss": 0.636, + "step": 3285 + }, + { + "epoch": 6.266666666666667, + "grad_norm": 0.2620706856250763, + "learning_rate": 7.35565837962798e-05, + "loss": 0.6509, + "step": 3290 + }, + { + "epoch": 6.276190476190476, + "grad_norm": 0.26130273938179016, + "learning_rate": 7.323612059554513e-05, + "loss": 0.6511, + "step": 3295 + }, + { + "epoch": 6.285714285714286, + "grad_norm": 0.22935442626476288, + "learning_rate": 7.291595318569951e-05, + "loss": 0.6357, + "step": 3300 + }, + { + "epoch": 6.295238095238095, + "grad_norm": 0.3201715350151062, + "learning_rate": 7.25960851051912e-05, + "loss": 0.6442, + "step": 3305 + }, + { + "epoch": 6.304761904761905, + "grad_norm": 0.22515980899333954, + "learning_rate": 7.227651988916031e-05, + "loss": 0.6467, + "step": 3310 + }, + { + "epoch": 6.314285714285714, + "grad_norm": 0.22413288056850433, + "learning_rate": 7.195726106939974e-05, + "loss": 0.6425, + "step": 3315 + }, + { + "epoch": 6.3238095238095235, + "grad_norm": 0.22419866919517517, + "learning_rate": 7.163831217431615e-05, + "loss": 0.6572, + "step": 3320 + }, + { + "epoch": 6.333333333333333, + "grad_norm": 0.23005759716033936, + "learning_rate": 7.131967672889101e-05, + "loss": 0.6473, + "step": 3325 + }, + { + "epoch": 6.3428571428571425, + "grad_norm": 0.2409326583147049, + "learning_rate": 7.100135825464139e-05, + "loss": 0.6342, + "step": 3330 + }, + { + "epoch": 6.352380952380952, + "grad_norm": 0.2287633717060089, + "learning_rate": 7.068336026958146e-05, + "loss": 0.6445, + "step": 3335 + }, + { + "epoch": 6.3619047619047615, + "grad_norm": 0.22215120494365692, + "learning_rate": 7.036568628818331e-05, + "loss": 0.6404, + "step": 3340 + }, + { + "epoch": 6.371428571428572, + "grad_norm": 0.2332102507352829, + "learning_rate": 7.004833982133808e-05, + "loss": 0.6309, + "step": 3345 + }, + { + "epoch": 6.380952380952381, + "grad_norm": 0.21468514204025269, + "learning_rate": 6.973132437631742e-05, + "loss": 0.6544, + "step": 3350 + }, + { + "epoch": 6.390476190476191, + "grad_norm": 0.23263736069202423, + "learning_rate": 6.941464345673449e-05, + "loss": 0.6378, + "step": 3355 + }, + { + "epoch": 6.4, + "grad_norm": 0.24136883020401, + "learning_rate": 6.909830056250527e-05, + "loss": 0.6567, + "step": 3360 + }, + { + "epoch": 6.40952380952381, + "grad_norm": 0.21449565887451172, + "learning_rate": 6.878229918981003e-05, + "loss": 0.6455, + "step": 3365 + }, + { + "epoch": 6.419047619047619, + "grad_norm": 0.24325934052467346, + "learning_rate": 6.846664283105455e-05, + "loss": 0.6362, + "step": 3370 + }, + { + "epoch": 6.428571428571429, + "grad_norm": 0.22252559661865234, + "learning_rate": 6.815133497483157e-05, + "loss": 0.6426, + "step": 3375 + }, + { + "epoch": 6.438095238095238, + "grad_norm": 0.21811635792255402, + "learning_rate": 6.783637910588216e-05, + "loss": 0.6419, + "step": 3380 + }, + { + "epoch": 6.447619047619048, + "grad_norm": 0.22843381762504578, + "learning_rate": 6.752177870505736e-05, + "loss": 0.6408, + "step": 3385 + }, + { + "epoch": 6.457142857142857, + "grad_norm": 0.23752233386039734, + "learning_rate": 6.720753724927958e-05, + "loss": 0.6687, + "step": 3390 + }, + { + "epoch": 6.466666666666667, + "grad_norm": 0.21157948672771454, + "learning_rate": 6.68936582115042e-05, + "loss": 0.6307, + "step": 3395 + }, + { + "epoch": 6.476190476190476, + "grad_norm": 0.23496676981449127, + "learning_rate": 6.658014506068126e-05, + "loss": 0.6453, + "step": 3400 + }, + { + "epoch": 6.485714285714286, + "grad_norm": 0.22223064303398132, + "learning_rate": 6.626700126171702e-05, + "loss": 0.65, + "step": 3405 + }, + { + "epoch": 6.495238095238095, + "grad_norm": 0.22827011346817017, + "learning_rate": 6.595423027543571e-05, + "loss": 0.6496, + "step": 3410 + }, + { + "epoch": 6.504761904761905, + "grad_norm": 0.22207534313201904, + "learning_rate": 6.56418355585413e-05, + "loss": 0.6375, + "step": 3415 + }, + { + "epoch": 6.514285714285714, + "grad_norm": 0.2754807770252228, + "learning_rate": 6.532982056357928e-05, + "loss": 0.6477, + "step": 3420 + }, + { + "epoch": 6.523809523809524, + "grad_norm": 0.21566829085350037, + "learning_rate": 6.501818873889855e-05, + "loss": 0.6349, + "step": 3425 + }, + { + "epoch": 6.533333333333333, + "grad_norm": 0.22195175290107727, + "learning_rate": 6.470694352861312e-05, + "loss": 0.6492, + "step": 3430 + }, + { + "epoch": 6.542857142857143, + "grad_norm": 0.2323753833770752, + "learning_rate": 6.439608837256432e-05, + "loss": 0.6388, + "step": 3435 + }, + { + "epoch": 6.552380952380952, + "grad_norm": 0.23055404424667358, + "learning_rate": 6.408562670628266e-05, + "loss": 0.6385, + "step": 3440 + }, + { + "epoch": 6.561904761904762, + "grad_norm": 0.21605758368968964, + "learning_rate": 6.377556196094973e-05, + "loss": 0.6473, + "step": 3445 + }, + { + "epoch": 6.571428571428571, + "grad_norm": 0.22067958116531372, + "learning_rate": 6.34658975633605e-05, + "loss": 0.6369, + "step": 3450 + }, + { + "epoch": 6.580952380952381, + "grad_norm": 0.22081324458122253, + "learning_rate": 6.315663693588534e-05, + "loss": 0.638, + "step": 3455 + }, + { + "epoch": 6.59047619047619, + "grad_norm": 0.224380224943161, + "learning_rate": 6.28477834964322e-05, + "loss": 0.6333, + "step": 3460 + }, + { + "epoch": 6.6, + "grad_norm": 0.2965473532676697, + "learning_rate": 6.25393406584088e-05, + "loss": 0.6405, + "step": 3465 + }, + { + "epoch": 6.609523809523809, + "grad_norm": 0.23683632910251617, + "learning_rate": 6.223131183068499e-05, + "loss": 0.645, + "step": 3470 + }, + { + "epoch": 6.619047619047619, + "grad_norm": 0.2217981517314911, + "learning_rate": 6.192370041755505e-05, + "loss": 0.6384, + "step": 3475 + }, + { + "epoch": 6.628571428571428, + "grad_norm": 0.21454942226409912, + "learning_rate": 6.161650981869998e-05, + "loss": 0.6418, + "step": 3480 + }, + { + "epoch": 6.638095238095238, + "grad_norm": 0.22158564627170563, + "learning_rate": 6.130974342915005e-05, + "loss": 0.6503, + "step": 3485 + }, + { + "epoch": 6.647619047619048, + "grad_norm": 0.22903573513031006, + "learning_rate": 6.100340463924723e-05, + "loss": 0.6491, + "step": 3490 + }, + { + "epoch": 6.6571428571428575, + "grad_norm": 0.22213099896907806, + "learning_rate": 6.069749683460765e-05, + "loss": 0.6368, + "step": 3495 + }, + { + "epoch": 6.666666666666667, + "grad_norm": 0.2271195501089096, + "learning_rate": 6.039202339608432e-05, + "loss": 0.6427, + "step": 3500 + }, + { + "epoch": 6.6761904761904765, + "grad_norm": 0.22851118445396423, + "learning_rate": 6.008698769972967e-05, + "loss": 0.6484, + "step": 3505 + }, + { + "epoch": 6.685714285714286, + "grad_norm": 0.22598107159137726, + "learning_rate": 5.978239311675826e-05, + "loss": 0.6433, + "step": 3510 + }, + { + "epoch": 6.695238095238095, + "grad_norm": 0.2218712866306305, + "learning_rate": 5.9478243013509505e-05, + "loss": 0.6384, + "step": 3515 + }, + { + "epoch": 6.704761904761905, + "grad_norm": 0.2214185744524002, + "learning_rate": 5.9174540751410487e-05, + "loss": 0.6518, + "step": 3520 + }, + { + "epoch": 6.714285714285714, + "grad_norm": 0.2286664992570877, + "learning_rate": 5.887128968693887e-05, + "loss": 0.6418, + "step": 3525 + }, + { + "epoch": 6.723809523809524, + "grad_norm": 0.2170267552137375, + "learning_rate": 5.856849317158563e-05, + "loss": 0.6486, + "step": 3530 + }, + { + "epoch": 6.733333333333333, + "grad_norm": 0.21755385398864746, + "learning_rate": 5.8266154551818216e-05, + "loss": 0.6285, + "step": 3535 + }, + { + "epoch": 6.742857142857143, + "grad_norm": 0.2144262194633484, + "learning_rate": 5.796427716904347e-05, + "loss": 0.6431, + "step": 3540 + }, + { + "epoch": 6.752380952380952, + "grad_norm": 0.21666479110717773, + "learning_rate": 5.7662864359570624e-05, + "loss": 0.6369, + "step": 3545 + }, + { + "epoch": 6.761904761904762, + "grad_norm": 0.2362881898880005, + "learning_rate": 5.736191945457463e-05, + "loss": 0.6449, + "step": 3550 + }, + { + "epoch": 6.771428571428571, + "grad_norm": 0.22996211051940918, + "learning_rate": 5.7061445780059074e-05, + "loss": 0.6356, + "step": 3555 + }, + { + "epoch": 6.780952380952381, + "grad_norm": 0.22610628604888916, + "learning_rate": 5.676144665681974e-05, + "loss": 0.65, + "step": 3560 + }, + { + "epoch": 6.79047619047619, + "grad_norm": 0.22624777257442474, + "learning_rate": 5.6461925400407576e-05, + "loss": 0.6434, + "step": 3565 + }, + { + "epoch": 6.8, + "grad_norm": 0.22202523052692413, + "learning_rate": 5.616288532109225e-05, + "loss": 0.6369, + "step": 3570 + }, + { + "epoch": 6.809523809523809, + "grad_norm": 0.22538915276527405, + "learning_rate": 5.58643297238256e-05, + "loss": 0.6407, + "step": 3575 + }, + { + "epoch": 6.819047619047619, + "grad_norm": 0.22355104982852936, + "learning_rate": 5.5566261908204966e-05, + "loss": 0.6484, + "step": 3580 + }, + { + "epoch": 6.828571428571428, + "grad_norm": 0.22738416492938995, + "learning_rate": 5.526868516843673e-05, + "loss": 0.6377, + "step": 3585 + }, + { + "epoch": 6.838095238095238, + "grad_norm": 0.22791539132595062, + "learning_rate": 5.497160279330014e-05, + "loss": 0.6384, + "step": 3590 + }, + { + "epoch": 6.847619047619047, + "grad_norm": 0.22362259030342102, + "learning_rate": 5.467501806611062e-05, + "loss": 0.6445, + "step": 3595 + }, + { + "epoch": 6.857142857142857, + "grad_norm": 0.23865683376789093, + "learning_rate": 5.43789342646837e-05, + "loss": 0.6421, + "step": 3600 + }, + { + "epoch": 6.866666666666667, + "grad_norm": 0.2309015691280365, + "learning_rate": 5.4083354661298814e-05, + "loss": 0.6562, + "step": 3605 + }, + { + "epoch": 6.876190476190477, + "grad_norm": 0.2259485423564911, + "learning_rate": 5.378828252266308e-05, + "loss": 0.6313, + "step": 3610 + }, + { + "epoch": 6.885714285714286, + "grad_norm": 0.2315201759338379, + "learning_rate": 5.349372110987496e-05, + "loss": 0.6583, + "step": 3615 + }, + { + "epoch": 6.895238095238096, + "grad_norm": 0.21792897582054138, + "learning_rate": 5.3199673678388685e-05, + "loss": 0.6534, + "step": 3620 + }, + { + "epoch": 6.904761904761905, + "grad_norm": 0.22640301287174225, + "learning_rate": 5.290614347797802e-05, + "loss": 0.6454, + "step": 3625 + }, + { + "epoch": 6.914285714285715, + "grad_norm": 0.2275838702917099, + "learning_rate": 5.261313375270014e-05, + "loss": 0.6471, + "step": 3630 + }, + { + "epoch": 6.923809523809524, + "grad_norm": 0.22041475772857666, + "learning_rate": 5.232064774086022e-05, + "loss": 0.6573, + "step": 3635 + }, + { + "epoch": 6.933333333333334, + "grad_norm": 0.2304404079914093, + "learning_rate": 5.2028688674975415e-05, + "loss": 0.6377, + "step": 3640 + }, + { + "epoch": 6.942857142857143, + "grad_norm": 0.22729218006134033, + "learning_rate": 5.1737259781738936e-05, + "loss": 0.6455, + "step": 3645 + }, + { + "epoch": 6.9523809523809526, + "grad_norm": 0.22103598713874817, + "learning_rate": 5.1446364281984774e-05, + "loss": 0.6503, + "step": 3650 + }, + { + "epoch": 6.961904761904762, + "grad_norm": 0.2224428504705429, + "learning_rate": 5.115600539065197e-05, + "loss": 0.6443, + "step": 3655 + }, + { + "epoch": 6.9714285714285715, + "grad_norm": 0.22665762901306152, + "learning_rate": 5.086618631674888e-05, + "loss": 0.6415, + "step": 3660 + }, + { + "epoch": 6.980952380952381, + "grad_norm": 0.2255854308605194, + "learning_rate": 5.057691026331792e-05, + "loss": 0.6288, + "step": 3665 + }, + { + "epoch": 6.9904761904761905, + "grad_norm": 0.21485072374343872, + "learning_rate": 5.02881804274002e-05, + "loss": 0.6466, + "step": 3670 + }, + { + "epoch": 7.0, + "grad_norm": 0.2187153398990631, + "learning_rate": 5.000000000000002e-05, + "loss": 0.6334, + "step": 3675 + }, + { + "epoch": 7.0, + "eval_loss": 1.3276076316833496, + "eval_runtime": 0.3234, + "eval_samples_per_second": 12.37, + "eval_steps_per_second": 3.093, + "step": 3675 + }, + { + "epoch": 7.0095238095238095, + "grad_norm": 0.21199530363082886, + "learning_rate": 4.971237216604967e-05, + "loss": 0.6287, + "step": 3680 + }, + { + "epoch": 7.019047619047619, + "grad_norm": 0.23007389903068542, + "learning_rate": 4.942530010437435e-05, + "loss": 0.6213, + "step": 3685 + }, + { + "epoch": 7.0285714285714285, + "grad_norm": 0.23050346970558167, + "learning_rate": 4.913878698765686e-05, + "loss": 0.6315, + "step": 3690 + }, + { + "epoch": 7.038095238095238, + "grad_norm": 0.22807465493679047, + "learning_rate": 4.885283598240259e-05, + "loss": 0.6316, + "step": 3695 + }, + { + "epoch": 7.0476190476190474, + "grad_norm": 0.23785021901130676, + "learning_rate": 4.856745024890466e-05, + "loss": 0.6464, + "step": 3700 + }, + { + "epoch": 7.057142857142857, + "grad_norm": 0.23706254363059998, + "learning_rate": 4.8282632941208725e-05, + "loss": 0.6321, + "step": 3705 + }, + { + "epoch": 7.066666666666666, + "grad_norm": 0.228632390499115, + "learning_rate": 4.799838720707846e-05, + "loss": 0.6293, + "step": 3710 + }, + { + "epoch": 7.076190476190476, + "grad_norm": 0.22894619405269623, + "learning_rate": 4.771471618796043e-05, + "loss": 0.631, + "step": 3715 + }, + { + "epoch": 7.085714285714285, + "grad_norm": 0.2303706556558609, + "learning_rate": 4.743162301894952e-05, + "loss": 0.6231, + "step": 3720 + }, + { + "epoch": 7.095238095238095, + "grad_norm": 0.22928594052791595, + "learning_rate": 4.7149110828754464e-05, + "loss": 0.6379, + "step": 3725 + }, + { + "epoch": 7.104761904761904, + "grad_norm": 0.22976277768611908, + "learning_rate": 4.686718273966291e-05, + "loss": 0.6253, + "step": 3730 + }, + { + "epoch": 7.114285714285714, + "grad_norm": 0.22555364668369293, + "learning_rate": 4.658584186750713e-05, + "loss": 0.629, + "step": 3735 + }, + { + "epoch": 7.123809523809523, + "grad_norm": 0.22725003957748413, + "learning_rate": 4.6305091321629666e-05, + "loss": 0.622, + "step": 3740 + }, + { + "epoch": 7.133333333333334, + "grad_norm": 0.2352193295955658, + "learning_rate": 4.6024934204848745e-05, + "loss": 0.6296, + "step": 3745 + }, + { + "epoch": 7.142857142857143, + "grad_norm": 0.22691534459590912, + "learning_rate": 4.574537361342407e-05, + "loss": 0.6505, + "step": 3750 + }, + { + "epoch": 7.152380952380953, + "grad_norm": 0.22884990274906158, + "learning_rate": 4.5466412637022704e-05, + "loss": 0.6438, + "step": 3755 + }, + { + "epoch": 7.161904761904762, + "grad_norm": 0.22874386608600616, + "learning_rate": 4.518805435868492e-05, + "loss": 0.6395, + "step": 3760 + }, + { + "epoch": 7.171428571428572, + "grad_norm": 0.23881952464580536, + "learning_rate": 4.491030185478976e-05, + "loss": 0.6338, + "step": 3765 + }, + { + "epoch": 7.180952380952381, + "grad_norm": 0.22728921473026276, + "learning_rate": 4.4633158195021594e-05, + "loss": 0.6424, + "step": 3770 + }, + { + "epoch": 7.190476190476191, + "grad_norm": 0.2279832810163498, + "learning_rate": 4.435662644233594e-05, + "loss": 0.6337, + "step": 3775 + }, + { + "epoch": 7.2, + "grad_norm": 0.22659078240394592, + "learning_rate": 4.4080709652925336e-05, + "loss": 0.6354, + "step": 3780 + }, + { + "epoch": 7.20952380952381, + "grad_norm": 0.2384444922208786, + "learning_rate": 4.380541087618606e-05, + "loss": 0.6262, + "step": 3785 + }, + { + "epoch": 7.219047619047619, + "grad_norm": 0.2297607809305191, + "learning_rate": 4.3530733154684164e-05, + "loss": 0.6331, + "step": 3790 + }, + { + "epoch": 7.228571428571429, + "grad_norm": 0.23174314200878143, + "learning_rate": 4.3256679524121834e-05, + "loss": 0.6279, + "step": 3795 + }, + { + "epoch": 7.238095238095238, + "grad_norm": 0.240789994597435, + "learning_rate": 4.298325301330383e-05, + "loss": 0.6279, + "step": 3800 + }, + { + "epoch": 7.247619047619048, + "grad_norm": 0.22910235822200775, + "learning_rate": 4.27104566441042e-05, + "loss": 0.6329, + "step": 3805 + }, + { + "epoch": 7.257142857142857, + "grad_norm": 0.22811570763587952, + "learning_rate": 4.2438293431432665e-05, + "loss": 0.628, + "step": 3810 + }, + { + "epoch": 7.266666666666667, + "grad_norm": 0.23355652391910553, + "learning_rate": 4.216676638320135e-05, + "loss": 0.6418, + "step": 3815 + }, + { + "epoch": 7.276190476190476, + "grad_norm": 0.2335626780986786, + "learning_rate": 4.189587850029169e-05, + "loss": 0.6349, + "step": 3820 + }, + { + "epoch": 7.285714285714286, + "grad_norm": 0.23812420666217804, + "learning_rate": 4.1625632776521037e-05, + "loss": 0.6326, + "step": 3825 + }, + { + "epoch": 7.295238095238095, + "grad_norm": 0.24018298089504242, + "learning_rate": 4.1356032198609706e-05, + "loss": 0.6249, + "step": 3830 + }, + { + "epoch": 7.304761904761905, + "grad_norm": 0.22917360067367554, + "learning_rate": 4.108707974614804e-05, + "loss": 0.6397, + "step": 3835 + }, + { + "epoch": 7.314285714285714, + "grad_norm": 0.23437809944152832, + "learning_rate": 4.081877839156325e-05, + "loss": 0.6334, + "step": 3840 + }, + { + "epoch": 7.3238095238095235, + "grad_norm": 0.23028285801410675, + "learning_rate": 4.0551131100086745e-05, + "loss": 0.6394, + "step": 3845 + }, + { + "epoch": 7.333333333333333, + "grad_norm": 0.22391854226589203, + "learning_rate": 4.028414082972141e-05, + "loss": 0.6291, + "step": 3850 + }, + { + "epoch": 7.3428571428571425, + "grad_norm": 0.22412873804569244, + "learning_rate": 4.001781053120863e-05, + "loss": 0.6377, + "step": 3855 + }, + { + "epoch": 7.352380952380952, + "grad_norm": 0.24517199397087097, + "learning_rate": 3.975214314799607e-05, + "loss": 0.6311, + "step": 3860 + }, + { + "epoch": 7.3619047619047615, + "grad_norm": 0.2389526069164276, + "learning_rate": 3.94871416162048e-05, + "loss": 0.6321, + "step": 3865 + }, + { + "epoch": 7.371428571428572, + "grad_norm": 0.2269456386566162, + "learning_rate": 3.9222808864597004e-05, + "loss": 0.6372, + "step": 3870 + }, + { + "epoch": 7.380952380952381, + "grad_norm": 0.2337706834077835, + "learning_rate": 3.89591478145437e-05, + "loss": 0.6336, + "step": 3875 + }, + { + "epoch": 7.390476190476191, + "grad_norm": 0.2252926528453827, + "learning_rate": 3.8696161379992225e-05, + "loss": 0.6329, + "step": 3880 + }, + { + "epoch": 7.4, + "grad_norm": 0.22934380173683167, + "learning_rate": 3.843385246743417e-05, + "loss": 0.6374, + "step": 3885 + }, + { + "epoch": 7.40952380952381, + "grad_norm": 0.22791922092437744, + "learning_rate": 3.817222397587336e-05, + "loss": 0.6326, + "step": 3890 + }, + { + "epoch": 7.419047619047619, + "grad_norm": 0.2295857071876526, + "learning_rate": 3.7911278796793516e-05, + "loss": 0.6353, + "step": 3895 + }, + { + "epoch": 7.428571428571429, + "grad_norm": 0.23317788541316986, + "learning_rate": 3.7651019814126654e-05, + "loss": 0.6317, + "step": 3900 + }, + { + "epoch": 7.438095238095238, + "grad_norm": 0.2250727415084839, + "learning_rate": 3.739144990422089e-05, + "loss": 0.6281, + "step": 3905 + }, + { + "epoch": 7.447619047619048, + "grad_norm": 0.22562646865844727, + "learning_rate": 3.7132571935808924e-05, + "loss": 0.6454, + "step": 3910 + }, + { + "epoch": 7.457142857142857, + "grad_norm": 0.2368139922618866, + "learning_rate": 3.687438876997612e-05, + "loss": 0.6444, + "step": 3915 + }, + { + "epoch": 7.466666666666667, + "grad_norm": 0.23156170547008514, + "learning_rate": 3.661690326012897e-05, + "loss": 0.6249, + "step": 3920 + }, + { + "epoch": 7.476190476190476, + "grad_norm": 0.23131678998470306, + "learning_rate": 3.6360118251963645e-05, + "loss": 0.6224, + "step": 3925 + }, + { + "epoch": 7.485714285714286, + "grad_norm": 0.23170669376850128, + "learning_rate": 3.610403658343443e-05, + "loss": 0.6355, + "step": 3930 + }, + { + "epoch": 7.495238095238095, + "grad_norm": 0.22911617159843445, + "learning_rate": 3.58486610847223e-05, + "loss": 0.6344, + "step": 3935 + }, + { + "epoch": 7.504761904761905, + "grad_norm": 0.23085066676139832, + "learning_rate": 3.5593994578203896e-05, + "loss": 0.6332, + "step": 3940 + }, + { + "epoch": 7.514285714285714, + "grad_norm": 0.23891034722328186, + "learning_rate": 3.534003987842005e-05, + "loss": 0.6165, + "step": 3945 + }, + { + "epoch": 7.523809523809524, + "grad_norm": 0.23303653299808502, + "learning_rate": 3.508679979204481e-05, + "loss": 0.6275, + "step": 3950 + }, + { + "epoch": 7.533333333333333, + "grad_norm": 0.2292027771472931, + "learning_rate": 3.483427711785449e-05, + "loss": 0.6424, + "step": 3955 + }, + { + "epoch": 7.542857142857143, + "grad_norm": 0.22289618849754333, + "learning_rate": 3.458247464669657e-05, + "loss": 0.6396, + "step": 3960 + }, + { + "epoch": 7.552380952380952, + "grad_norm": 0.22831101715564728, + "learning_rate": 3.4331395161458955e-05, + "loss": 0.6293, + "step": 3965 + }, + { + "epoch": 7.561904761904762, + "grad_norm": 0.23642738163471222, + "learning_rate": 3.408104143703929e-05, + "loss": 0.6394, + "step": 3970 + }, + { + "epoch": 7.571428571428571, + "grad_norm": 0.23651988804340363, + "learning_rate": 3.383141624031408e-05, + "loss": 0.65, + "step": 3975 + }, + { + "epoch": 7.580952380952381, + "grad_norm": 0.229757159948349, + "learning_rate": 3.35825223301083e-05, + "loss": 0.6244, + "step": 3980 + }, + { + "epoch": 7.59047619047619, + "grad_norm": 0.2353103756904602, + "learning_rate": 3.333436245716488e-05, + "loss": 0.6218, + "step": 3985 + }, + { + "epoch": 7.6, + "grad_norm": 0.24895426630973816, + "learning_rate": 3.308693936411421e-05, + "loss": 0.6323, + "step": 3990 + }, + { + "epoch": 7.609523809523809, + "grad_norm": 0.24273917078971863, + "learning_rate": 3.2840255785443855e-05, + "loss": 0.6307, + "step": 3995 + }, + { + "epoch": 7.619047619047619, + "grad_norm": 0.24038879573345184, + "learning_rate": 3.259431444746846e-05, + "loss": 0.6326, + "step": 4000 + }, + { + "epoch": 7.628571428571428, + "grad_norm": 0.23644651472568512, + "learning_rate": 3.234911806829948e-05, + "loss": 0.6284, + "step": 4005 + }, + { + "epoch": 7.638095238095238, + "grad_norm": 0.238076314330101, + "learning_rate": 3.210466935781516e-05, + "loss": 0.6317, + "step": 4010 + }, + { + "epoch": 7.647619047619048, + "grad_norm": 0.23139876127243042, + "learning_rate": 3.1860971017630604e-05, + "loss": 0.6214, + "step": 4015 + }, + { + "epoch": 7.6571428571428575, + "grad_norm": 0.23607772588729858, + "learning_rate": 3.161802574106799e-05, + "loss": 0.6337, + "step": 4020 + }, + { + "epoch": 7.666666666666667, + "grad_norm": 0.23559069633483887, + "learning_rate": 3.137583621312665e-05, + "loss": 0.6278, + "step": 4025 + }, + { + "epoch": 7.6761904761904765, + "grad_norm": 0.2418035864830017, + "learning_rate": 3.1134405110453515e-05, + "loss": 0.6302, + "step": 4030 + }, + { + "epoch": 7.685714285714286, + "grad_norm": 0.25391387939453125, + "learning_rate": 3.089373510131354e-05, + "loss": 0.647, + "step": 4035 + }, + { + "epoch": 7.695238095238095, + "grad_norm": 0.2297278344631195, + "learning_rate": 3.065382884556012e-05, + "loss": 0.6296, + "step": 4040 + }, + { + "epoch": 7.704761904761905, + "grad_norm": 0.22566130757331848, + "learning_rate": 3.0414688994605723e-05, + "loss": 0.6324, + "step": 4045 + }, + { + "epoch": 7.714285714285714, + "grad_norm": 0.23526619374752045, + "learning_rate": 3.0176318191392726e-05, + "loss": 0.6403, + "step": 4050 + }, + { + "epoch": 7.723809523809524, + "grad_norm": 0.230539932847023, + "learning_rate": 2.9938719070363952e-05, + "loss": 0.6397, + "step": 4055 + }, + { + "epoch": 7.733333333333333, + "grad_norm": 0.23514920473098755, + "learning_rate": 2.9701894257433826e-05, + "loss": 0.626, + "step": 4060 + }, + { + "epoch": 7.742857142857143, + "grad_norm": 0.24501506984233856, + "learning_rate": 2.9465846369959127e-05, + "loss": 0.6322, + "step": 4065 + }, + { + "epoch": 7.752380952380952, + "grad_norm": 0.22882947325706482, + "learning_rate": 2.923057801671015e-05, + "loss": 0.624, + "step": 4070 + }, + { + "epoch": 7.761904761904762, + "grad_norm": 0.23827053606510162, + "learning_rate": 2.8996091797841973e-05, + "loss": 0.6377, + "step": 4075 + }, + { + "epoch": 7.771428571428571, + "grad_norm": 0.23343512415885925, + "learning_rate": 2.876239030486554e-05, + "loss": 0.6388, + "step": 4080 + }, + { + "epoch": 7.780952380952381, + "grad_norm": 0.25341862440109253, + "learning_rate": 2.8529476120619104e-05, + "loss": 0.622, + "step": 4085 + }, + { + "epoch": 7.79047619047619, + "grad_norm": 0.23422300815582275, + "learning_rate": 2.829735181923978e-05, + "loss": 0.6281, + "step": 4090 + }, + { + "epoch": 7.8, + "grad_norm": 0.22450385987758636, + "learning_rate": 2.8066019966134904e-05, + "loss": 0.6316, + "step": 4095 + }, + { + "epoch": 7.809523809523809, + "grad_norm": 0.2333240658044815, + "learning_rate": 2.7835483117953788e-05, + "loss": 0.6406, + "step": 4100 + }, + { + "epoch": 7.819047619047619, + "grad_norm": 0.23980993032455444, + "learning_rate": 2.7605743822559506e-05, + "loss": 0.6351, + "step": 4105 + }, + { + "epoch": 7.828571428571428, + "grad_norm": 0.2439979910850525, + "learning_rate": 2.7376804619000707e-05, + "loss": 0.638, + "step": 4110 + }, + { + "epoch": 7.838095238095238, + "grad_norm": 0.2424413561820984, + "learning_rate": 2.7148668037483372e-05, + "loss": 0.6357, + "step": 4115 + }, + { + "epoch": 7.847619047619047, + "grad_norm": 0.231084942817688, + "learning_rate": 2.692133659934315e-05, + "loss": 0.6217, + "step": 4120 + }, + { + "epoch": 7.857142857142857, + "grad_norm": 0.23306681215763092, + "learning_rate": 2.669481281701739e-05, + "loss": 0.6293, + "step": 4125 + }, + { + "epoch": 7.866666666666667, + "grad_norm": 0.2443678230047226, + "learning_rate": 2.6469099194017143e-05, + "loss": 0.6348, + "step": 4130 + }, + { + "epoch": 7.876190476190477, + "grad_norm": 0.2363799512386322, + "learning_rate": 2.624419822489985e-05, + "loss": 0.6295, + "step": 4135 + }, + { + "epoch": 7.885714285714286, + "grad_norm": 0.2323981076478958, + "learning_rate": 2.6020112395241624e-05, + "loss": 0.6375, + "step": 4140 + }, + { + "epoch": 7.895238095238096, + "grad_norm": 0.24826794862747192, + "learning_rate": 2.579684418160958e-05, + "loss": 0.6389, + "step": 4145 + }, + { + "epoch": 7.904761904761905, + "grad_norm": 0.223325714468956, + "learning_rate": 2.5574396051534832e-05, + "loss": 0.636, + "step": 4150 + }, + { + "epoch": 7.914285714285715, + "grad_norm": 0.22773179411888123, + "learning_rate": 2.5352770463484987e-05, + "loss": 0.6256, + "step": 4155 + }, + { + "epoch": 7.923809523809524, + "grad_norm": 0.23265981674194336, + "learning_rate": 2.5131969866836992e-05, + "loss": 0.6241, + "step": 4160 + }, + { + "epoch": 7.933333333333334, + "grad_norm": 0.23152920603752136, + "learning_rate": 2.491199670185008e-05, + "loss": 0.6366, + "step": 4165 + }, + { + "epoch": 7.942857142857143, + "grad_norm": 0.24560460448265076, + "learning_rate": 2.4692853399638917e-05, + "loss": 0.6378, + "step": 4170 + }, + { + "epoch": 7.9523809523809526, + "grad_norm": 0.22926020622253418, + "learning_rate": 2.4474542382146537e-05, + "loss": 0.6402, + "step": 4175 + }, + { + "epoch": 7.961904761904762, + "grad_norm": 0.2434161901473999, + "learning_rate": 2.425706606211767e-05, + "loss": 0.6338, + "step": 4180 + }, + { + "epoch": 7.9714285714285715, + "grad_norm": 0.23746095597743988, + "learning_rate": 2.4040426843072206e-05, + "loss": 0.6484, + "step": 4185 + }, + { + "epoch": 7.980952380952381, + "grad_norm": 0.2342091053724289, + "learning_rate": 2.3824627119278342e-05, + "loss": 0.6273, + "step": 4190 + }, + { + "epoch": 7.9904761904761905, + "grad_norm": 0.2437712550163269, + "learning_rate": 2.3609669275726355e-05, + "loss": 0.6365, + "step": 4195 + }, + { + "epoch": 8.0, + "grad_norm": 0.23214805126190186, + "learning_rate": 2.339555568810221e-05, + "loss": 0.6257, + "step": 4200 + }, + { + "epoch": 8.0, + "eval_loss": 1.3403503894805908, + "eval_runtime": 0.3249, + "eval_samples_per_second": 12.312, + "eval_steps_per_second": 3.078, + "step": 4200 + }, + { + "epoch": 8.00952380952381, + "grad_norm": 0.23136037588119507, + "learning_rate": 2.318228872276118e-05, + "loss": 0.6216, + "step": 4205 + }, + { + "epoch": 8.019047619047619, + "grad_norm": 0.23694069683551788, + "learning_rate": 2.2969870736701895e-05, + "loss": 0.6208, + "step": 4210 + }, + { + "epoch": 8.028571428571428, + "grad_norm": 0.23975005745887756, + "learning_rate": 2.275830407754006e-05, + "loss": 0.6298, + "step": 4215 + }, + { + "epoch": 8.038095238095238, + "grad_norm": 0.23182031512260437, + "learning_rate": 2.2547591083482665e-05, + "loss": 0.627, + "step": 4220 + }, + { + "epoch": 8.047619047619047, + "grad_norm": 0.24627025425434113, + "learning_rate": 2.2337734083302164e-05, + "loss": 0.6128, + "step": 4225 + }, + { + "epoch": 8.057142857142857, + "grad_norm": 0.23380368947982788, + "learning_rate": 2.212873539631061e-05, + "loss": 0.6315, + "step": 4230 + }, + { + "epoch": 8.066666666666666, + "grad_norm": 0.23186807334423065, + "learning_rate": 2.192059733233408e-05, + "loss": 0.626, + "step": 4235 + }, + { + "epoch": 8.076190476190476, + "grad_norm": 0.23652274906635284, + "learning_rate": 2.1713322191687237e-05, + "loss": 0.6358, + "step": 4240 + }, + { + "epoch": 8.085714285714285, + "grad_norm": 0.24210993945598602, + "learning_rate": 2.1506912265147772e-05, + "loss": 0.6315, + "step": 4245 + }, + { + "epoch": 8.095238095238095, + "grad_norm": 0.23931226134300232, + "learning_rate": 2.1301369833931117e-05, + "loss": 0.6093, + "step": 4250 + }, + { + "epoch": 8.104761904761904, + "grad_norm": 0.2417602837085724, + "learning_rate": 2.1096697169665313e-05, + "loss": 0.6271, + "step": 4255 + }, + { + "epoch": 8.114285714285714, + "grad_norm": 0.23360390961170197, + "learning_rate": 2.0892896534365904e-05, + "loss": 0.6216, + "step": 4260 + }, + { + "epoch": 8.123809523809523, + "grad_norm": 0.2413237988948822, + "learning_rate": 2.068997018041069e-05, + "loss": 0.6184, + "step": 4265 + }, + { + "epoch": 8.133333333333333, + "grad_norm": 0.2482381910085678, + "learning_rate": 2.0487920350515212e-05, + "loss": 0.624, + "step": 4270 + }, + { + "epoch": 8.142857142857142, + "grad_norm": 0.24830950796604156, + "learning_rate": 2.0286749277707782e-05, + "loss": 0.6253, + "step": 4275 + }, + { + "epoch": 8.152380952380952, + "grad_norm": 0.24451254308223724, + "learning_rate": 2.0086459185304618e-05, + "loss": 0.6213, + "step": 4280 + }, + { + "epoch": 8.161904761904761, + "grad_norm": 0.23981262743473053, + "learning_rate": 1.9887052286885655e-05, + "loss": 0.6237, + "step": 4285 + }, + { + "epoch": 8.17142857142857, + "grad_norm": 0.24213843047618866, + "learning_rate": 1.9688530786269855e-05, + "loss": 0.6297, + "step": 4290 + }, + { + "epoch": 8.18095238095238, + "grad_norm": 0.23491358757019043, + "learning_rate": 1.9490896877490716e-05, + "loss": 0.6309, + "step": 4295 + }, + { + "epoch": 8.19047619047619, + "grad_norm": 0.24114489555358887, + "learning_rate": 1.929415274477239e-05, + "loss": 0.6273, + "step": 4300 + }, + { + "epoch": 8.2, + "grad_norm": 0.24128074944019318, + "learning_rate": 1.9098300562505266e-05, + "loss": 0.6283, + "step": 4305 + }, + { + "epoch": 8.209523809523809, + "grad_norm": 0.23723888397216797, + "learning_rate": 1.8903342495221977e-05, + "loss": 0.6221, + "step": 4310 + }, + { + "epoch": 8.219047619047618, + "grad_norm": 0.23763377964496613, + "learning_rate": 1.870928069757353e-05, + "loss": 0.6309, + "step": 4315 + }, + { + "epoch": 8.228571428571428, + "grad_norm": 0.2403092235326767, + "learning_rate": 1.8516117314305524e-05, + "loss": 0.6221, + "step": 4320 + }, + { + "epoch": 8.238095238095237, + "grad_norm": 0.23820042610168457, + "learning_rate": 1.832385448023435e-05, + "loss": 0.6254, + "step": 4325 + }, + { + "epoch": 8.247619047619047, + "grad_norm": 0.23579280078411102, + "learning_rate": 1.8132494320223638e-05, + "loss": 0.6225, + "step": 4330 + }, + { + "epoch": 8.257142857142856, + "grad_norm": 0.24339617788791656, + "learning_rate": 1.7942038949160854e-05, + "loss": 0.6292, + "step": 4335 + }, + { + "epoch": 8.266666666666667, + "grad_norm": 0.24123533070087433, + "learning_rate": 1.775249047193377e-05, + "loss": 0.6241, + "step": 4340 + }, + { + "epoch": 8.276190476190477, + "grad_norm": 0.2402196228504181, + "learning_rate": 1.756385098340736e-05, + "loss": 0.629, + "step": 4345 + }, + { + "epoch": 8.285714285714286, + "grad_norm": 0.22874706983566284, + "learning_rate": 1.7376122568400532e-05, + "loss": 0.6161, + "step": 4350 + }, + { + "epoch": 8.295238095238096, + "grad_norm": 0.23019570112228394, + "learning_rate": 1.7189307301663084e-05, + "loss": 0.6144, + "step": 4355 + }, + { + "epoch": 8.304761904761905, + "grad_norm": 0.2534734010696411, + "learning_rate": 1.7003407247852943e-05, + "loss": 0.6275, + "step": 4360 + }, + { + "epoch": 8.314285714285715, + "grad_norm": 0.2459252029657364, + "learning_rate": 1.681842446151313e-05, + "loss": 0.6095, + "step": 4365 + }, + { + "epoch": 8.323809523809524, + "grad_norm": 0.23832803964614868, + "learning_rate": 1.6634360987049115e-05, + "loss": 0.6295, + "step": 4370 + }, + { + "epoch": 8.333333333333334, + "grad_norm": 0.22661805152893066, + "learning_rate": 1.6451218858706374e-05, + "loss": 0.6322, + "step": 4375 + }, + { + "epoch": 8.342857142857143, + "grad_norm": 0.246062234044075, + "learning_rate": 1.6269000100547683e-05, + "loss": 0.63, + "step": 4380 + }, + { + "epoch": 8.352380952380953, + "grad_norm": 0.24505558609962463, + "learning_rate": 1.6087706726430873e-05, + "loss": 0.6271, + "step": 4385 + }, + { + "epoch": 8.361904761904762, + "grad_norm": 0.2376982420682907, + "learning_rate": 1.5907340739986575e-05, + "loss": 0.6346, + "step": 4390 + }, + { + "epoch": 8.371428571428572, + "grad_norm": 0.23987463116645813, + "learning_rate": 1.5727904134596083e-05, + "loss": 0.6239, + "step": 4395 + }, + { + "epoch": 8.380952380952381, + "grad_norm": 0.24543854594230652, + "learning_rate": 1.5549398893369216e-05, + "loss": 0.634, + "step": 4400 + }, + { + "epoch": 8.39047619047619, + "grad_norm": 0.23671837151050568, + "learning_rate": 1.5371826989122506e-05, + "loss": 0.6288, + "step": 4405 + }, + { + "epoch": 8.4, + "grad_norm": 0.23273004591464996, + "learning_rate": 1.5195190384357404e-05, + "loss": 0.6288, + "step": 4410 + }, + { + "epoch": 8.40952380952381, + "grad_norm": 0.24326454102993011, + "learning_rate": 1.501949103123852e-05, + "loss": 0.6386, + "step": 4415 + }, + { + "epoch": 8.41904761904762, + "grad_norm": 0.2386908084154129, + "learning_rate": 1.4844730871572043e-05, + "loss": 0.6312, + "step": 4420 + }, + { + "epoch": 8.428571428571429, + "grad_norm": 0.23784001171588898, + "learning_rate": 1.467091183678444e-05, + "loss": 0.6158, + "step": 4425 + }, + { + "epoch": 8.438095238095238, + "grad_norm": 0.24026724696159363, + "learning_rate": 1.449803584790086e-05, + "loss": 0.6275, + "step": 4430 + }, + { + "epoch": 8.447619047619048, + "grad_norm": 0.23797613382339478, + "learning_rate": 1.4326104815524088e-05, + "loss": 0.6254, + "step": 4435 + }, + { + "epoch": 8.457142857142857, + "grad_norm": 0.2374219447374344, + "learning_rate": 1.415512063981339e-05, + "loss": 0.6308, + "step": 4440 + }, + { + "epoch": 8.466666666666667, + "grad_norm": 0.26075613498687744, + "learning_rate": 1.3985085210463477e-05, + "loss": 0.6279, + "step": 4445 + }, + { + "epoch": 8.476190476190476, + "grad_norm": 0.23775546252727509, + "learning_rate": 1.3816000406683604e-05, + "loss": 0.6369, + "step": 4450 + }, + { + "epoch": 8.485714285714286, + "grad_norm": 0.25033655762672424, + "learning_rate": 1.364786809717692e-05, + "loss": 0.6317, + "step": 4455 + }, + { + "epoch": 8.495238095238095, + "grad_norm": 0.23537905514240265, + "learning_rate": 1.3480690140119657e-05, + "loss": 0.6252, + "step": 4460 + }, + { + "epoch": 8.504761904761905, + "grad_norm": 0.24787379801273346, + "learning_rate": 1.3314468383140688e-05, + "loss": 0.6241, + "step": 4465 + }, + { + "epoch": 8.514285714285714, + "grad_norm": 0.24332831799983978, + "learning_rate": 1.3149204663301118e-05, + "loss": 0.6369, + "step": 4470 + }, + { + "epoch": 8.523809523809524, + "grad_norm": 0.23973241448402405, + "learning_rate": 1.2984900807073919e-05, + "loss": 0.6275, + "step": 4475 + }, + { + "epoch": 8.533333333333333, + "grad_norm": 0.24601459503173828, + "learning_rate": 1.2821558630323772e-05, + "loss": 0.6347, + "step": 4480 + }, + { + "epoch": 8.542857142857143, + "grad_norm": 0.23905131220817566, + "learning_rate": 1.2659179938287035e-05, + "loss": 0.6208, + "step": 4485 + }, + { + "epoch": 8.552380952380952, + "grad_norm": 0.23523162305355072, + "learning_rate": 1.2497766525551724e-05, + "loss": 0.6155, + "step": 4490 + }, + { + "epoch": 8.561904761904762, + "grad_norm": 0.2428881824016571, + "learning_rate": 1.2337320176037759e-05, + "loss": 0.6283, + "step": 4495 + }, + { + "epoch": 8.571428571428571, + "grad_norm": 0.24822403490543365, + "learning_rate": 1.2177842662977135e-05, + "loss": 0.6298, + "step": 4500 + }, + { + "epoch": 8.58095238095238, + "grad_norm": 0.23006483912467957, + "learning_rate": 1.201933574889449e-05, + "loss": 0.6264, + "step": 4505 + }, + { + "epoch": 8.59047619047619, + "grad_norm": 0.23746708035469055, + "learning_rate": 1.186180118558743e-05, + "loss": 0.6143, + "step": 4510 + }, + { + "epoch": 8.6, + "grad_norm": 0.23508812487125397, + "learning_rate": 1.1705240714107302e-05, + "loss": 0.621, + "step": 4515 + }, + { + "epoch": 8.60952380952381, + "grad_norm": 0.2398723065853119, + "learning_rate": 1.1549656064739967e-05, + "loss": 0.6246, + "step": 4520 + }, + { + "epoch": 8.619047619047619, + "grad_norm": 0.2376699149608612, + "learning_rate": 1.1395048956986575e-05, + "loss": 0.6205, + "step": 4525 + }, + { + "epoch": 8.628571428571428, + "grad_norm": 0.2414540946483612, + "learning_rate": 1.124142109954459e-05, + "loss": 0.6204, + "step": 4530 + }, + { + "epoch": 8.638095238095238, + "grad_norm": 0.2395474761724472, + "learning_rate": 1.108877419028902e-05, + "loss": 0.6255, + "step": 4535 + }, + { + "epoch": 8.647619047619047, + "grad_norm": 0.24437417089939117, + "learning_rate": 1.0937109916253474e-05, + "loss": 0.6231, + "step": 4540 + }, + { + "epoch": 8.657142857142857, + "grad_norm": 0.25323808193206787, + "learning_rate": 1.0786429953611666e-05, + "loss": 0.6291, + "step": 4545 + }, + { + "epoch": 8.666666666666666, + "grad_norm": 0.23473167419433594, + "learning_rate": 1.0636735967658784e-05, + "loss": 0.6207, + "step": 4550 + }, + { + "epoch": 8.676190476190476, + "grad_norm": 0.23271633684635162, + "learning_rate": 1.0488029612793138e-05, + "loss": 0.6226, + "step": 4555 + }, + { + "epoch": 8.685714285714285, + "grad_norm": 0.24144107103347778, + "learning_rate": 1.034031253249792e-05, + "loss": 0.6329, + "step": 4560 + }, + { + "epoch": 8.695238095238095, + "grad_norm": 0.2525869905948639, + "learning_rate": 1.0193586359322927e-05, + "loss": 0.6192, + "step": 4565 + }, + { + "epoch": 8.704761904761904, + "grad_norm": 0.24347785115242004, + "learning_rate": 1.004785271486659e-05, + "loss": 0.6284, + "step": 4570 + }, + { + "epoch": 8.714285714285714, + "grad_norm": 0.24428115785121918, + "learning_rate": 9.903113209758096e-06, + "loss": 0.6346, + "step": 4575 + }, + { + "epoch": 8.723809523809523, + "grad_norm": 0.2366633117198944, + "learning_rate": 9.759369443639454e-06, + "loss": 0.6257, + "step": 4580 + }, + { + "epoch": 8.733333333333333, + "grad_norm": 0.24603548645973206, + "learning_rate": 9.616623005147951e-06, + "loss": 0.6286, + "step": 4585 + }, + { + "epoch": 8.742857142857144, + "grad_norm": 0.240610271692276, + "learning_rate": 9.474875471898526e-06, + "loss": 0.6288, + "step": 4590 + }, + { + "epoch": 8.752380952380953, + "grad_norm": 0.2394147515296936, + "learning_rate": 9.334128410466358e-06, + "loss": 0.6304, + "step": 4595 + }, + { + "epoch": 8.761904761904763, + "grad_norm": 0.23732179403305054, + "learning_rate": 9.194383376369508e-06, + "loss": 0.6256, + "step": 4600 + }, + { + "epoch": 8.771428571428572, + "grad_norm": 0.23427680134773254, + "learning_rate": 9.055641914051782e-06, + "loss": 0.6154, + "step": 4605 + }, + { + "epoch": 8.780952380952382, + "grad_norm": 0.24093270301818848, + "learning_rate": 8.917905556865713e-06, + "loss": 0.6189, + "step": 4610 + }, + { + "epoch": 8.790476190476191, + "grad_norm": 0.24400576949119568, + "learning_rate": 8.781175827055389e-06, + "loss": 0.6301, + "step": 4615 + }, + { + "epoch": 8.8, + "grad_norm": 0.24120011925697327, + "learning_rate": 8.645454235739903e-06, + "loss": 0.6239, + "step": 4620 + }, + { + "epoch": 8.80952380952381, + "grad_norm": 0.2423393577337265, + "learning_rate": 8.510742282896544e-06, + "loss": 0.6383, + "step": 4625 + }, + { + "epoch": 8.81904761904762, + "grad_norm": 0.24667339026927948, + "learning_rate": 8.377041457344103e-06, + "loss": 0.628, + "step": 4630 + }, + { + "epoch": 8.82857142857143, + "grad_norm": 0.23888331651687622, + "learning_rate": 8.24435323672661e-06, + "loss": 0.6236, + "step": 4635 + }, + { + "epoch": 8.838095238095239, + "grad_norm": 0.2347164750099182, + "learning_rate": 8.112679087496933e-06, + "loss": 0.621, + "step": 4640 + }, + { + "epoch": 8.847619047619048, + "grad_norm": 0.249232217669487, + "learning_rate": 7.982020464900486e-06, + "loss": 0.6296, + "step": 4645 + }, + { + "epoch": 8.857142857142858, + "grad_norm": 0.2342270165681839, + "learning_rate": 7.852378812959227e-06, + "loss": 0.6261, + "step": 4650 + }, + { + "epoch": 8.866666666666667, + "grad_norm": 0.24151065945625305, + "learning_rate": 7.72375556445577e-06, + "loss": 0.6207, + "step": 4655 + }, + { + "epoch": 8.876190476190477, + "grad_norm": 0.24049705266952515, + "learning_rate": 7.596152140917368e-06, + "loss": 0.6294, + "step": 4660 + }, + { + "epoch": 8.885714285714286, + "grad_norm": 0.24534159898757935, + "learning_rate": 7.46956995260033e-06, + "loss": 0.6227, + "step": 4665 + }, + { + "epoch": 8.895238095238096, + "grad_norm": 0.2385033518075943, + "learning_rate": 7.344010398474455e-06, + "loss": 0.6309, + "step": 4670 + }, + { + "epoch": 8.904761904761905, + "grad_norm": 0.24666495621204376, + "learning_rate": 7.219474866207465e-06, + "loss": 0.6197, + "step": 4675 + }, + { + "epoch": 8.914285714285715, + "grad_norm": 0.23790596425533295, + "learning_rate": 7.09596473214974e-06, + "loss": 0.6161, + "step": 4680 + }, + { + "epoch": 8.923809523809524, + "grad_norm": 0.23783983290195465, + "learning_rate": 6.973481361319123e-06, + "loss": 0.6148, + "step": 4685 + }, + { + "epoch": 8.933333333333334, + "grad_norm": 0.2483368068933487, + "learning_rate": 6.852026107385756e-06, + "loss": 0.6311, + "step": 4690 + }, + { + "epoch": 8.942857142857143, + "grad_norm": 0.24231784045696259, + "learning_rate": 6.731600312657238e-06, + "loss": 0.631, + "step": 4695 + }, + { + "epoch": 8.952380952380953, + "grad_norm": 0.24928317964076996, + "learning_rate": 6.612205308063646e-06, + "loss": 0.6235, + "step": 4700 + }, + { + "epoch": 8.961904761904762, + "grad_norm": 0.23735187947750092, + "learning_rate": 6.493842413142914e-06, + "loss": 0.6259, + "step": 4705 + }, + { + "epoch": 8.971428571428572, + "grad_norm": 0.2443312555551529, + "learning_rate": 6.37651293602628e-06, + "loss": 0.6372, + "step": 4710 + }, + { + "epoch": 8.980952380952381, + "grad_norm": 0.24989666044712067, + "learning_rate": 6.260218173423749e-06, + "loss": 0.6253, + "step": 4715 + }, + { + "epoch": 8.99047619047619, + "grad_norm": 0.24133141338825226, + "learning_rate": 6.144959410609785e-06, + "loss": 0.6334, + "step": 4720 + }, + { + "epoch": 9.0, + "grad_norm": 0.24009673297405243, + "learning_rate": 6.030737921409169e-06, + "loss": 0.6314, + "step": 4725 + }, + { + "epoch": 9.0, + "eval_loss": 1.3410394191741943, + "eval_runtime": 0.3239, + "eval_samples_per_second": 12.35, + "eval_steps_per_second": 3.088, + "step": 4725 + }, + { + "epoch": 9.00952380952381, + "grad_norm": 0.2452923059463501, + "learning_rate": 5.917554968182803e-06, + "loss": 0.6214, + "step": 4730 + }, + { + "epoch": 9.019047619047619, + "grad_norm": 0.2375546544790268, + "learning_rate": 5.805411801813865e-06, + "loss": 0.6109, + "step": 4735 + }, + { + "epoch": 9.028571428571428, + "grad_norm": 0.2484077662229538, + "learning_rate": 5.694309661693942e-06, + "loss": 0.6179, + "step": 4740 + }, + { + "epoch": 9.038095238095238, + "grad_norm": 0.23496288061141968, + "learning_rate": 5.584249775709371e-06, + "loss": 0.6322, + "step": 4745 + }, + { + "epoch": 9.047619047619047, + "grad_norm": 0.25102752447128296, + "learning_rate": 5.475233360227516e-06, + "loss": 0.6257, + "step": 4750 + }, + { + "epoch": 9.057142857142857, + "grad_norm": 0.24880193173885345, + "learning_rate": 5.367261620083575e-06, + "loss": 0.6333, + "step": 4755 + }, + { + "epoch": 9.066666666666666, + "grad_norm": 0.24257133901119232, + "learning_rate": 5.26033574856708e-06, + "loss": 0.6297, + "step": 4760 + }, + { + "epoch": 9.076190476190476, + "grad_norm": 0.24016863107681274, + "learning_rate": 5.1544569274087125e-06, + "loss": 0.6263, + "step": 4765 + }, + { + "epoch": 9.085714285714285, + "grad_norm": 0.24308805167675018, + "learning_rate": 5.049626326767365e-06, + "loss": 0.6202, + "step": 4770 + }, + { + "epoch": 9.095238095238095, + "grad_norm": 0.25135210156440735, + "learning_rate": 4.945845105217117e-06, + "loss": 0.6202, + "step": 4775 + }, + { + "epoch": 9.104761904761904, + "grad_norm": 0.24231822788715363, + "learning_rate": 4.843114409734384e-06, + "loss": 0.6275, + "step": 4780 + }, + { + "epoch": 9.114285714285714, + "grad_norm": 0.24114476144313812, + "learning_rate": 4.741435375685377e-06, + "loss": 0.6163, + "step": 4785 + }, + { + "epoch": 9.123809523809523, + "grad_norm": 0.25297802686691284, + "learning_rate": 4.640809126813484e-06, + "loss": 0.6278, + "step": 4790 + }, + { + "epoch": 9.133333333333333, + "grad_norm": 0.24216073751449585, + "learning_rate": 4.541236775226809e-06, + "loss": 0.6208, + "step": 4795 + }, + { + "epoch": 9.142857142857142, + "grad_norm": 0.23923125863075256, + "learning_rate": 4.442719421385922e-06, + "loss": 0.6148, + "step": 4800 + }, + { + "epoch": 9.152380952380952, + "grad_norm": 0.25497567653656006, + "learning_rate": 4.3452581540917465e-06, + "loss": 0.6215, + "step": 4805 + }, + { + "epoch": 9.161904761904761, + "grad_norm": 0.2442941963672638, + "learning_rate": 4.248854050473405e-06, + "loss": 0.6197, + "step": 4810 + }, + { + "epoch": 9.17142857142857, + "grad_norm": 0.2433978170156479, + "learning_rate": 4.153508175976428e-06, + "loss": 0.6216, + "step": 4815 + }, + { + "epoch": 9.18095238095238, + "grad_norm": 0.24015413224697113, + "learning_rate": 4.059221584350958e-06, + "loss": 0.6261, + "step": 4820 + }, + { + "epoch": 9.19047619047619, + "grad_norm": 0.24067072570323944, + "learning_rate": 3.965995317640025e-06, + "loss": 0.6264, + "step": 4825 + }, + { + "epoch": 9.2, + "grad_norm": 0.24210909008979797, + "learning_rate": 3.873830406168111e-06, + "loss": 0.6106, + "step": 4830 + }, + { + "epoch": 9.209523809523809, + "grad_norm": 0.24473240971565247, + "learning_rate": 3.7827278685297785e-06, + "loss": 0.6207, + "step": 4835 + }, + { + "epoch": 9.219047619047618, + "grad_norm": 0.24637968838214874, + "learning_rate": 3.692688711578296e-06, + "loss": 0.6292, + "step": 4840 + }, + { + "epoch": 9.228571428571428, + "grad_norm": 0.24839821457862854, + "learning_rate": 3.6037139304146762e-06, + "loss": 0.6222, + "step": 4845 + }, + { + "epoch": 9.238095238095237, + "grad_norm": 0.2498600333929062, + "learning_rate": 3.515804508376508e-06, + "loss": 0.6266, + "step": 4850 + }, + { + "epoch": 9.247619047619047, + "grad_norm": 0.2570066750049591, + "learning_rate": 3.428961417027221e-06, + "loss": 0.6265, + "step": 4855 + }, + { + "epoch": 9.257142857142856, + "grad_norm": 0.24512985348701477, + "learning_rate": 3.3431856161452835e-06, + "loss": 0.6185, + "step": 4860 + }, + { + "epoch": 9.266666666666667, + "grad_norm": 0.2434176504611969, + "learning_rate": 3.2584780537136207e-06, + "loss": 0.6322, + "step": 4865 + }, + { + "epoch": 9.276190476190477, + "grad_norm": 0.24330537021160126, + "learning_rate": 3.1748396659090797e-06, + "loss": 0.6241, + "step": 4870 + }, + { + "epoch": 9.285714285714286, + "grad_norm": 0.24642570316791534, + "learning_rate": 3.092271377092215e-06, + "loss": 0.6274, + "step": 4875 + }, + { + "epoch": 9.295238095238096, + "grad_norm": 0.23440000414848328, + "learning_rate": 3.010774099796898e-06, + "loss": 0.6163, + "step": 4880 + }, + { + "epoch": 9.304761904761905, + "grad_norm": 0.24107569456100464, + "learning_rate": 2.9303487347203783e-06, + "loss": 0.6226, + "step": 4885 + }, + { + "epoch": 9.314285714285715, + "grad_norm": 0.23449283838272095, + "learning_rate": 2.8509961707132494e-06, + "loss": 0.6318, + "step": 4890 + }, + { + "epoch": 9.323809523809524, + "grad_norm": 0.25119608640670776, + "learning_rate": 2.772717284769677e-06, + "loss": 0.6264, + "step": 4895 + }, + { + "epoch": 9.333333333333334, + "grad_norm": 0.23543590307235718, + "learning_rate": 2.6955129420176196e-06, + "loss": 0.6128, + "step": 4900 + }, + { + "epoch": 9.342857142857143, + "grad_norm": 0.23934414982795715, + "learning_rate": 2.619383995709368e-06, + "loss": 0.6163, + "step": 4905 + }, + { + "epoch": 9.352380952380953, + "grad_norm": 0.25171783566474915, + "learning_rate": 2.5443312872120763e-06, + "loss": 0.6229, + "step": 4910 + }, + { + "epoch": 9.361904761904762, + "grad_norm": 0.23963721096515656, + "learning_rate": 2.4703556459984456e-06, + "loss": 0.6269, + "step": 4915 + }, + { + "epoch": 9.371428571428572, + "grad_norm": 0.25053393840789795, + "learning_rate": 2.3974578896375553e-06, + "loss": 0.6227, + "step": 4920 + }, + { + "epoch": 9.380952380952381, + "grad_norm": 0.24071863293647766, + "learning_rate": 2.3256388237858807e-06, + "loss": 0.6276, + "step": 4925 + }, + { + "epoch": 9.39047619047619, + "grad_norm": 0.2486407309770584, + "learning_rate": 2.25489924217831e-06, + "loss": 0.6228, + "step": 4930 + }, + { + "epoch": 9.4, + "grad_norm": 0.2375013381242752, + "learning_rate": 2.1852399266194314e-06, + "loss": 0.6139, + "step": 4935 + }, + { + "epoch": 9.40952380952381, + "grad_norm": 0.2476935237646103, + "learning_rate": 2.1166616469749044e-06, + "loss": 0.6162, + "step": 4940 + }, + { + "epoch": 9.41904761904762, + "grad_norm": 0.24889154732227325, + "learning_rate": 2.049165161162858e-06, + "loss": 0.613, + "step": 4945 + }, + { + "epoch": 9.428571428571429, + "grad_norm": 0.240982323884964, + "learning_rate": 1.9827512151456173e-06, + "loss": 0.63, + "step": 4950 + }, + { + "epoch": 9.438095238095238, + "grad_norm": 0.2406960278749466, + "learning_rate": 1.917420542921433e-06, + "loss": 0.6204, + "step": 4955 + }, + { + "epoch": 9.447619047619048, + "grad_norm": 0.2408684343099594, + "learning_rate": 1.8531738665163112e-06, + "loss": 0.6126, + "step": 4960 + }, + { + "epoch": 9.457142857142857, + "grad_norm": 0.2530766725540161, + "learning_rate": 1.790011895976118e-06, + "loss": 0.6303, + "step": 4965 + }, + { + "epoch": 9.466666666666667, + "grad_norm": 0.2396114617586136, + "learning_rate": 1.7279353293586765e-06, + "loss": 0.6274, + "step": 4970 + }, + { + "epoch": 9.476190476190476, + "grad_norm": 0.23766325414180756, + "learning_rate": 1.66694485272606e-06, + "loss": 0.6246, + "step": 4975 + }, + { + "epoch": 9.485714285714286, + "grad_norm": 0.24630650877952576, + "learning_rate": 1.6070411401370334e-06, + "loss": 0.6217, + "step": 4980 + }, + { + "epoch": 9.495238095238095, + "grad_norm": 0.25678691267967224, + "learning_rate": 1.5482248536395905e-06, + "loss": 0.626, + "step": 4985 + }, + { + "epoch": 9.504761904761905, + "grad_norm": 0.24416512250900269, + "learning_rate": 1.4904966432635947e-06, + "loss": 0.6299, + "step": 4990 + }, + { + "epoch": 9.514285714285714, + "grad_norm": 0.2470226287841797, + "learning_rate": 1.4338571470137063e-06, + "loss": 0.6229, + "step": 4995 + }, + { + "epoch": 9.523809523809524, + "grad_norm": 0.24221652746200562, + "learning_rate": 1.378306990862177e-06, + "loss": 0.621, + "step": 5000 + }, + { + "epoch": 9.533333333333333, + "grad_norm": 0.24181492626667023, + "learning_rate": 1.323846788742078e-06, + "loss": 0.6048, + "step": 5005 + }, + { + "epoch": 9.542857142857143, + "grad_norm": 0.24268200993537903, + "learning_rate": 1.2704771425404382e-06, + "loss": 0.6382, + "step": 5010 + }, + { + "epoch": 9.552380952380952, + "grad_norm": 0.24577122926712036, + "learning_rate": 1.2181986420915615e-06, + "loss": 0.6284, + "step": 5015 + }, + { + "epoch": 9.561904761904762, + "grad_norm": 0.24282944202423096, + "learning_rate": 1.1670118651706197e-06, + "loss": 0.6201, + "step": 5020 + }, + { + "epoch": 9.571428571428571, + "grad_norm": 0.24517573416233063, + "learning_rate": 1.1169173774871478e-06, + "loss": 0.6234, + "step": 5025 + }, + { + "epoch": 9.58095238095238, + "grad_norm": 0.23740537464618683, + "learning_rate": 1.067915732678859e-06, + "loss": 0.6202, + "step": 5030 + }, + { + "epoch": 9.59047619047619, + "grad_norm": 0.24459858238697052, + "learning_rate": 1.0200074723055398e-06, + "loss": 0.6273, + "step": 5035 + }, + { + "epoch": 9.6, + "grad_norm": 0.24076281487941742, + "learning_rate": 9.731931258429638e-07, + "loss": 0.6337, + "step": 5040 + }, + { + "epoch": 9.60952380952381, + "grad_norm": 0.24906916916370392, + "learning_rate": 9.274732106771988e-07, + "loss": 0.6172, + "step": 5045 + }, + { + "epoch": 9.619047619047619, + "grad_norm": 0.23475927114486694, + "learning_rate": 8.828482320987319e-07, + "loss": 0.617, + "step": 5050 + }, + { + "epoch": 9.628571428571428, + "grad_norm": 0.2411186695098877, + "learning_rate": 8.393186832969746e-07, + "loss": 0.6235, + "step": 5055 + }, + { + "epoch": 9.638095238095238, + "grad_norm": 0.24285031855106354, + "learning_rate": 7.968850453548226e-07, + "loss": 0.6293, + "step": 5060 + }, + { + "epoch": 9.647619047619047, + "grad_norm": 0.24694561958312988, + "learning_rate": 7.555477872432715e-07, + "loss": 0.6242, + "step": 5065 + }, + { + "epoch": 9.657142857142857, + "grad_norm": 0.24676960706710815, + "learning_rate": 7.153073658162646e-07, + "loss": 0.6236, + "step": 5070 + }, + { + "epoch": 9.666666666666666, + "grad_norm": 0.25139757990837097, + "learning_rate": 6.761642258056978e-07, + "loss": 0.635, + "step": 5075 + }, + { + "epoch": 9.676190476190476, + "grad_norm": 0.2367604374885559, + "learning_rate": 6.381187998164229e-07, + "loss": 0.6248, + "step": 5080 + }, + { + "epoch": 9.685714285714285, + "grad_norm": 0.24842996895313263, + "learning_rate": 6.011715083214741e-07, + "loss": 0.6203, + "step": 5085 + }, + { + "epoch": 9.695238095238095, + "grad_norm": 0.25487616658210754, + "learning_rate": 5.653227596575161e-07, + "loss": 0.6235, + "step": 5090 + }, + { + "epoch": 9.704761904761904, + "grad_norm": 0.23450079560279846, + "learning_rate": 5.305729500201917e-07, + "loss": 0.6179, + "step": 5095 + }, + { + "epoch": 9.714285714285714, + "grad_norm": 0.24460509419441223, + "learning_rate": 4.969224634598591e-07, + "loss": 0.6174, + "step": 5100 + }, + { + "epoch": 9.723809523809523, + "grad_norm": 0.2516217827796936, + "learning_rate": 4.6437167187728393e-07, + "loss": 0.6149, + "step": 5105 + }, + { + "epoch": 9.733333333333333, + "grad_norm": 0.24219274520874023, + "learning_rate": 4.329209350195651e-07, + "loss": 0.6235, + "step": 5110 + }, + { + "epoch": 9.742857142857144, + "grad_norm": 0.24507369101047516, + "learning_rate": 4.025706004760932e-07, + "loss": 0.6322, + "step": 5115 + }, + { + "epoch": 9.752380952380953, + "grad_norm": 0.24700859189033508, + "learning_rate": 3.7332100367482024e-07, + "loss": 0.619, + "step": 5120 + }, + { + "epoch": 9.761904761904763, + "grad_norm": 0.23718251287937164, + "learning_rate": 3.451724678784518e-07, + "loss": 0.6156, + "step": 5125 + }, + { + "epoch": 9.771428571428572, + "grad_norm": 0.2427196353673935, + "learning_rate": 3.1812530418090513e-07, + "loss": 0.6186, + "step": 5130 + }, + { + "epoch": 9.780952380952382, + "grad_norm": 0.23827961087226868, + "learning_rate": 2.921798115039009e-07, + "loss": 0.608, + "step": 5135 + }, + { + "epoch": 9.790476190476191, + "grad_norm": 0.23600171506404877, + "learning_rate": 2.673362765936327e-07, + "loss": 0.6194, + "step": 5140 + }, + { + "epoch": 9.8, + "grad_norm": 0.2432773858308792, + "learning_rate": 2.4359497401758024e-07, + "loss": 0.6259, + "step": 5145 + }, + { + "epoch": 9.80952380952381, + "grad_norm": 0.24990539252758026, + "learning_rate": 2.2095616616150115e-07, + "loss": 0.6227, + "step": 5150 + }, + { + "epoch": 9.81904761904762, + "grad_norm": 0.23904049396514893, + "learning_rate": 1.9942010322655524e-07, + "loss": 0.6174, + "step": 5155 + }, + { + "epoch": 9.82857142857143, + "grad_norm": 0.23718209564685822, + "learning_rate": 1.7898702322648453e-07, + "loss": 0.6279, + "step": 5160 + }, + { + "epoch": 9.838095238095239, + "grad_norm": 0.24559420347213745, + "learning_rate": 1.596571519850043e-07, + "loss": 0.6209, + "step": 5165 + }, + { + "epoch": 9.847619047619048, + "grad_norm": 0.23925460875034332, + "learning_rate": 1.414307031333273e-07, + "loss": 0.6195, + "step": 5170 + }, + { + "epoch": 9.857142857142858, + "grad_norm": 0.2313607931137085, + "learning_rate": 1.2430787810776555e-07, + "loss": 0.614, + "step": 5175 + }, + { + "epoch": 9.866666666666667, + "grad_norm": 0.2343767285346985, + "learning_rate": 1.0828886614754341e-07, + "loss": 0.6238, + "step": 5180 + }, + { + "epoch": 9.876190476190477, + "grad_norm": 0.24600039422512054, + "learning_rate": 9.337384429269901e-08, + "loss": 0.6264, + "step": 5185 + }, + { + "epoch": 9.885714285714286, + "grad_norm": 0.24011261761188507, + "learning_rate": 7.956297738207497e-08, + "loss": 0.6147, + "step": 5190 + }, + { + "epoch": 9.895238095238096, + "grad_norm": 0.23987029492855072, + "learning_rate": 6.685641805158627e-08, + "loss": 0.6204, + "step": 5195 + }, + { + "epoch": 9.904761904761905, + "grad_norm": 0.23944221436977386, + "learning_rate": 5.5254306732444025e-08, + "loss": 0.6315, + "step": 5200 + }, + { + "epoch": 9.914285714285715, + "grad_norm": 0.24449606239795685, + "learning_rate": 4.475677164966774e-08, + "loss": 0.6149, + "step": 5205 + }, + { + "epoch": 9.923809523809524, + "grad_norm": 0.2435421347618103, + "learning_rate": 3.536392882064199e-08, + "loss": 0.6352, + "step": 5210 + }, + { + "epoch": 9.933333333333334, + "grad_norm": 0.24883785843849182, + "learning_rate": 2.7075882053828605e-08, + "loss": 0.6303, + "step": 5215 + }, + { + "epoch": 9.942857142857143, + "grad_norm": 0.25345560908317566, + "learning_rate": 1.9892722947645326e-08, + "loss": 0.6168, + "step": 5220 + }, + { + "epoch": 9.952380952380953, + "grad_norm": 0.24040359258651733, + "learning_rate": 1.3814530889433296e-08, + "loss": 0.6159, + "step": 5225 + }, + { + "epoch": 9.961904761904762, + "grad_norm": 0.24196043610572815, + "learning_rate": 8.841373054546686e-09, + "loss": 0.6198, + "step": 5230 + }, + { + "epoch": 9.971428571428572, + "grad_norm": 0.2445971667766571, + "learning_rate": 4.973304405697654e-09, + "loss": 0.6088, + "step": 5235 + }, + { + "epoch": 9.980952380952381, + "grad_norm": 0.245397686958313, + "learning_rate": 2.2103676922680117e-09, + "loss": 0.6265, + "step": 5240 + }, + { + "epoch": 9.99047619047619, + "grad_norm": 0.24436482787132263, + "learning_rate": 5.525934498651352e-10, + "loss": 0.6103, + "step": 5245 + }, + { + "epoch": 10.0, + "grad_norm": 0.2441110908985138, + "learning_rate": 0.0, + "loss": 0.6205, + "step": 5250 + }, + { + "epoch": 10.0, + "eval_loss": 1.3443903923034668, + "eval_runtime": 0.326, + "eval_samples_per_second": 12.27, + "eval_steps_per_second": 3.068, + "step": 5250 + }, + { + "epoch": 10.0, + "step": 5250, + "total_flos": 7.760593801491513e+18, + "train_loss": 0.6862981784684318, + "train_runtime": 15610.7374, + "train_samples_per_second": 10.757, + "train_steps_per_second": 0.336 + } + ], + "logging_steps": 5, + "max_steps": 5250, + "num_input_tokens_seen": 0, + "num_train_epochs": 10, + "save_steps": 100, + "stateful_callbacks": { + "TrainerControl": { + "args": { + "should_epoch_stop": false, + "should_evaluate": false, + "should_log": false, + "should_save": true, + "should_training_stop": true + }, + "attributes": {} + } + }, + "total_flos": 7.760593801491513e+18, + "train_batch_size": 4, + "trial_name": null, + "trial_params": null +}