diff --git "a/trainer_state.json" "b/trainer_state.json" --- "a/trainer_state.json" +++ "b/trainer_state.json" @@ -10,2998 +10,2998 @@ "log_history": [ { "epoch": 0.0, - "grad_norm": 9.972599255002624, - "learning_rate": 4.934477379095164e-07, - "loss": 0.4175, + "grad_norm": 11.189342496234838, + "learning_rate": 4.2000000000000006e-07, + "loss": 0.6334, "step": 20 }, { "epoch": 0.0, - "grad_norm": 9.535056632903434, - "learning_rate": 8.023400936037442e-07, - "loss": 0.4234, + "grad_norm": 12.102946252189938, + "learning_rate": 6.515789473684211e-07, + "loss": 0.6334, "step": 40 }, { "epoch": 0.01, - "grad_norm": 8.69996199325503, - "learning_rate": 1.1112324492979721e-06, - "loss": 0.4021, + "grad_norm": 9.23868714422811, + "learning_rate": 8.831578947368421e-07, + "loss": 0.5919, "step": 60 }, { "epoch": 0.01, - "grad_norm": 7.56775274786591, - "learning_rate": 1.4201248049921997e-06, - "loss": 0.393, + "grad_norm": 10.163427827153294, + "learning_rate": 1.1147368421052632e-06, + "loss": 0.591, "step": 80 }, { "epoch": 0.01, - "grad_norm": 9.287484589226363, - "learning_rate": 1.7290171606864275e-06, - "loss": 0.4086, + "grad_norm": 7.703301789458378, + "learning_rate": 1.3463157894736842e-06, + "loss": 0.5663, "step": 100 }, { "epoch": 0.01, - "grad_norm": 8.052654553706383, - "learning_rate": 2.037909516380655e-06, - "loss": 0.3761, + "grad_norm": 10.292384648986614, + "learning_rate": 1.5778947368421053e-06, + "loss": 0.5649, "step": 120 }, { "epoch": 0.02, - "grad_norm": 12.124856535072642, - "learning_rate": 2.346801872074883e-06, - "loss": 0.3431, + "grad_norm": 9.622475032798782, + "learning_rate": 1.8094736842105263e-06, + "loss": 0.5215, "step": 140 }, { "epoch": 0.02, - "grad_norm": 7.602396496836849, - "learning_rate": 2.655694227769111e-06, - "loss": 0.3907, + "grad_norm": 12.80974448742081, + "learning_rate": 2.0410526315789475e-06, + "loss": 0.5166, "step": 160 }, { "epoch": 0.02, - "grad_norm": 7.226437129402364, - "learning_rate": 2.964586583463339e-06, - "loss": 0.3281, + "grad_norm": 7.2494201428270495, + "learning_rate": 2.2726315789473687e-06, + "loss": 0.4708, "step": 180 }, { "epoch": 0.02, - "grad_norm": 23.40605051442028, - "learning_rate": 3.2734789391575668e-06, - "loss": 0.3253, + "grad_norm": 7.525527996410342, + "learning_rate": 2.5042105263157895e-06, + "loss": 0.4809, "step": 200 }, { "epoch": 0.03, - "grad_norm": 7.708682270572112, - "learning_rate": 3.5823712948517946e-06, - "loss": 0.336, + "grad_norm": 9.553244905277726, + "learning_rate": 2.735789473684211e-06, + "loss": 0.4149, "step": 220 }, { "epoch": 0.03, - "grad_norm": 7.7105187144857865, - "learning_rate": 3.891263650546022e-06, - "loss": 0.3283, + "grad_norm": 10.08934392514418, + "learning_rate": 2.967368421052632e-06, + "loss": 0.4709, "step": 240 }, { "epoch": 0.03, - "grad_norm": 6.988204554760109, - "learning_rate": 4.200156006240251e-06, - "loss": 0.3512, + "grad_norm": 8.77781615537256, + "learning_rate": 3.198947368421053e-06, + "loss": 0.4706, "step": 260 }, { "epoch": 0.03, - "grad_norm": 8.836133501086515, - "learning_rate": 4.509048361934478e-06, - "loss": 0.3266, + "grad_norm": 8.842231539252408, + "learning_rate": 3.430526315789474e-06, + "loss": 0.4033, "step": 280 }, { "epoch": 0.04, - "grad_norm": 7.405216965880714, - "learning_rate": 4.817940717628705e-06, - "loss": 0.3259, + "grad_norm": 10.007575408247385, + "learning_rate": 3.662105263157895e-06, + "loss": 0.4411, "step": 300 }, { "epoch": 0.04, - "grad_norm": 8.288061025740957, - "learning_rate": 5.126833073322934e-06, - "loss": 0.2918, + "grad_norm": 8.652379461692668, + "learning_rate": 3.893684210526316e-06, + "loss": 0.4011, "step": 320 }, { "epoch": 0.04, - "grad_norm": 9.310442601222784, - "learning_rate": 5.435725429017161e-06, - "loss": 0.3229, + "grad_norm": 9.03023367578369, + "learning_rate": 4.125263157894737e-06, + "loss": 0.4252, "step": 340 }, { "epoch": 0.04, - "grad_norm": 6.570232957864239, - "learning_rate": 5.744617784711389e-06, - "loss": 0.3063, + "grad_norm": 8.141636176102654, + "learning_rate": 4.356842105263159e-06, + "loss": 0.4364, "step": 360 }, { "epoch": 0.04, - "grad_norm": 8.212480249482766, - "learning_rate": 6.053510140405617e-06, - "loss": 0.2867, + "grad_norm": 8.361423248626743, + "learning_rate": 4.58842105263158e-06, + "loss": 0.4193, "step": 380 }, { "epoch": 0.05, - "grad_norm": 7.688921541109132, - "learning_rate": 6.362402496099844e-06, - "loss": 0.2987, + "grad_norm": 9.373028764113052, + "learning_rate": 4.8200000000000004e-06, + "loss": 0.4078, "step": 400 }, { "epoch": 0.05, - "grad_norm": 6.800952048131668, - "learning_rate": 6.671294851794072e-06, - "loss": 0.3047, + "grad_norm": 8.799844515033747, + "learning_rate": 5.051578947368422e-06, + "loss": 0.4291, "step": 420 }, { "epoch": 0.05, - "grad_norm": 6.193094209969253, - "learning_rate": 6.9801872074883005e-06, - "loss": 0.316, + "grad_norm": 8.642174028335415, + "learning_rate": 5.283157894736842e-06, + "loss": 0.382, "step": 440 }, { "epoch": 0.05, - "grad_norm": 7.878186262498564, - "learning_rate": 7.289079563182527e-06, - "loss": 0.331, + "grad_norm": 9.451813409785364, + "learning_rate": 5.514736842105264e-06, + "loss": 0.4089, "step": 460 }, { "epoch": 0.06, - "grad_norm": 9.807655645103456, - "learning_rate": 7.597971918876755e-06, - "loss": 0.2761, + "grad_norm": 11.329024075313082, + "learning_rate": 5.7463157894736845e-06, + "loss": 0.4015, "step": 480 }, { "epoch": 0.06, - "grad_norm": 10.286905976546715, - "learning_rate": 7.906864274570983e-06, - "loss": 0.2978, + "grad_norm": 7.271736970582471, + "learning_rate": 5.977894736842106e-06, + "loss": 0.4028, "step": 500 }, { "epoch": 0.06, - "grad_norm": 6.140392036969283, - "learning_rate": 8.215756630265212e-06, - "loss": 0.3164, + "grad_norm": 10.856360235187129, + "learning_rate": 6.209473684210528e-06, + "loss": 0.3814, "step": 520 }, { "epoch": 0.06, - "grad_norm": 6.714849999088133, - "learning_rate": 8.52464898595944e-06, - "loss": 0.2659, + "grad_norm": 10.40735773829138, + "learning_rate": 6.441052631578948e-06, + "loss": 0.3868, "step": 540 }, { "epoch": 0.07, - "grad_norm": 6.369031618010756, - "learning_rate": 8.833541341653666e-06, - "loss": 0.2994, + "grad_norm": 7.176143531188922, + "learning_rate": 6.672631578947369e-06, + "loss": 0.4089, "step": 560 }, { "epoch": 0.07, - "grad_norm": 7.515486432421921, - "learning_rate": 9.142433697347895e-06, - "loss": 0.295, + "grad_norm": 9.628282656024224, + "learning_rate": 6.90421052631579e-06, + "loss": 0.3771, "step": 580 }, { "epoch": 0.07, - "grad_norm": 6.180107864376804, - "learning_rate": 9.451326053042123e-06, - "loss": 0.3205, + "grad_norm": 11.199839686200068, + "learning_rate": 7.135789473684212e-06, + "loss": 0.383, "step": 600 }, { "epoch": 0.07, - "grad_norm": 8.619489153176001, - "learning_rate": 9.760218408736351e-06, - "loss": 0.2835, + "grad_norm": 8.042449860326105, + "learning_rate": 7.367368421052632e-06, + "loss": 0.3875, "step": 620 }, { "epoch": 0.07, - "grad_norm": 5.689814505344143, - "learning_rate": 1.0069110764430576e-05, - "loss": 0.2983, + "grad_norm": 6.256777636127039, + "learning_rate": 7.598947368421053e-06, + "loss": 0.399, "step": 640 }, { "epoch": 0.08, - "grad_norm": 7.447069695249161, - "learning_rate": 1.0378003120124806e-05, - "loss": 0.2736, + "grad_norm": 12.140400652487944, + "learning_rate": 7.830526315789473e-06, + "loss": 0.399, "step": 660 }, { "epoch": 0.08, - "grad_norm": 6.997342558926417, - "learning_rate": 1.0686895475819034e-05, - "loss": 0.2697, + "grad_norm": 9.31740323216306, + "learning_rate": 8.062105263157896e-06, + "loss": 0.383, "step": 680 }, { "epoch": 0.08, - "grad_norm": 9.33003253698042, - "learning_rate": 1.099578783151326e-05, - "loss": 0.2897, + "grad_norm": 11.518507949844242, + "learning_rate": 8.293684210526317e-06, + "loss": 0.3673, "step": 700 }, { "epoch": 0.08, - "grad_norm": 7.585581715153782, - "learning_rate": 1.1304680187207489e-05, - "loss": 0.2864, + "grad_norm": 12.23485079993539, + "learning_rate": 8.525263157894737e-06, + "loss": 0.4105, "step": 720 }, { "epoch": 0.09, - "grad_norm": 8.595084235914209, - "learning_rate": 1.1613572542901717e-05, - "loss": 0.273, + "grad_norm": 7.506176757455166, + "learning_rate": 8.756842105263158e-06, + "loss": 0.425, "step": 740 }, { "epoch": 0.09, - "grad_norm": 6.3292154008480415, - "learning_rate": 1.1922464898595945e-05, - "loss": 0.2817, + "grad_norm": 28.3327677352401, + "learning_rate": 8.988421052631579e-06, + "loss": 0.3874, "step": 760 }, { "epoch": 0.09, - "grad_norm": 14.605530389011415, - "learning_rate": 1.2231357254290172e-05, - "loss": 0.2553, + "grad_norm": 10.006030362522525, + "learning_rate": 9.220000000000002e-06, + "loss": 0.4055, "step": 780 }, { "epoch": 0.09, - "grad_norm": 6.541062863744024, - "learning_rate": 1.25402496099844e-05, - "loss": 0.2841, + "grad_norm": 10.010446813711976, + "learning_rate": 9.451578947368422e-06, + "loss": 0.3599, "step": 800 }, { "epoch": 0.1, - "grad_norm": 5.8850560260656515, - "learning_rate": 1.2849141965678628e-05, - "loss": 0.2807, + "grad_norm": 7.043688000572858, + "learning_rate": 9.683157894736843e-06, + "loss": 0.3624, "step": 820 }, { "epoch": 0.1, - "grad_norm": 6.8846024470910585, - "learning_rate": 1.3158034321372857e-05, - "loss": 0.2676, + "grad_norm": 9.607021831412599, + "learning_rate": 9.914736842105264e-06, + "loss": 0.3952, "step": 840 }, { "epoch": 0.1, - "grad_norm": 8.570829697532297, - "learning_rate": 1.3466926677067085e-05, - "loss": 0.2927, + "grad_norm": 9.147395412600936, + "learning_rate": 1.0146315789473686e-05, + "loss": 0.3819, "step": 860 }, { "epoch": 0.1, - "grad_norm": 5.270646324408607, - "learning_rate": 1.3775819032761311e-05, - "loss": 0.2602, + "grad_norm": 7.993375419567414, + "learning_rate": 1.0377894736842107e-05, + "loss": 0.3678, "step": 880 }, { "epoch": 0.11, - "grad_norm": 9.01642423951394, - "learning_rate": 1.4084711388455538e-05, - "loss": 0.3052, + "grad_norm": 7.52609874523302, + "learning_rate": 1.0609473684210528e-05, + "loss": 0.375, "step": 900 }, { "epoch": 0.11, - "grad_norm": 7.50913242007014, - "learning_rate": 1.4393603744149766e-05, - "loss": 0.2528, + "grad_norm": 9.52738539095041, + "learning_rate": 1.084105263157895e-05, + "loss": 0.3681, "step": 920 }, { "epoch": 0.11, - "grad_norm": 9.734458861747513, - "learning_rate": 1.4702496099843994e-05, - "loss": 0.2873, + "grad_norm": 11.370588516129013, + "learning_rate": 1.107263157894737e-05, + "loss": 0.3911, "step": 940 }, { "epoch": 0.11, - "grad_norm": 7.558759284363097, - "learning_rate": 1.5011388455538223e-05, - "loss": 0.2422, + "grad_norm": 9.391054368953709, + "learning_rate": 1.130421052631579e-05, + "loss": 0.329, "step": 960 }, { "epoch": 0.11, - "grad_norm": 6.004819523801505, - "learning_rate": 1.532028081123245e-05, - "loss": 0.286, + "grad_norm": 7.997549873904863, + "learning_rate": 1.1535789473684211e-05, + "loss": 0.359, "step": 980 }, { "epoch": 0.12, - "grad_norm": 6.551216816169416, - "learning_rate": 1.562917316692668e-05, - "loss": 0.2656, + "grad_norm": 10.755576173466588, + "learning_rate": 1.1767368421052634e-05, + "loss": 0.393, "step": 1000 }, { "epoch": 0.12, - "grad_norm": 8.811195222833438, - "learning_rate": 1.5938065522620907e-05, - "loss": 0.2875, + "grad_norm": 6.458117606549104, + "learning_rate": 1.1998947368421054e-05, + "loss": 0.363, "step": 1020 }, { "epoch": 0.12, - "grad_norm": 7.722575980788583, - "learning_rate": 1.6246957878315132e-05, - "loss": 0.3019, + "grad_norm": 9.937055469862932, + "learning_rate": 1.2230526315789475e-05, + "loss": 0.3766, "step": 1040 }, { "epoch": 0.12, - "grad_norm": 7.541514386348537, - "learning_rate": 1.655585023400936e-05, - "loss": 0.2558, + "grad_norm": 7.485816768913991, + "learning_rate": 1.2462105263157898e-05, + "loss": 0.3845, "step": 1060 }, { "epoch": 0.13, - "grad_norm": 5.48873197937876, - "learning_rate": 1.686474258970359e-05, - "loss": 0.2688, + "grad_norm": 8.01333127284645, + "learning_rate": 1.2693684210526319e-05, + "loss": 0.3617, "step": 1080 }, { "epoch": 0.13, - "grad_norm": 6.341266315002328, - "learning_rate": 1.7173634945397817e-05, - "loss": 0.2737, + "grad_norm": 8.911089094498925, + "learning_rate": 1.292526315789474e-05, + "loss": 0.3599, "step": 1100 }, { "epoch": 0.13, - "grad_norm": 5.9824355844807, - "learning_rate": 1.748252730109205e-05, - "loss": 0.2894, + "grad_norm": 8.32079434857686, + "learning_rate": 1.3156842105263158e-05, + "loss": 0.3722, "step": 1120 }, { "epoch": 0.13, - "grad_norm": 7.806450783046352, - "learning_rate": 1.7791419656786273e-05, - "loss": 0.2642, + "grad_norm": 9.806231592184165, + "learning_rate": 1.3388421052631581e-05, + "loss": 0.3755, "step": 1140 }, { "epoch": 0.14, - "grad_norm": 6.224028987297279, - "learning_rate": 1.81003120124805e-05, - "loss": 0.2963, + "grad_norm": 8.798076540805445, + "learning_rate": 1.3620000000000002e-05, + "loss": 0.3616, "step": 1160 }, { "epoch": 0.14, - "grad_norm": 6.9210132897774, - "learning_rate": 1.840920436817473e-05, - "loss": 0.267, + "grad_norm": 8.343613755617508, + "learning_rate": 1.3851578947368423e-05, + "loss": 0.3631, "step": 1180 }, { "epoch": 0.14, - "grad_norm": 7.506216143453127, - "learning_rate": 1.8718096723868958e-05, - "loss": 0.2855, + "grad_norm": 8.35814690526204, + "learning_rate": 1.4083157894736843e-05, + "loss": 0.3638, "step": 1200 }, { "epoch": 0.14, - "grad_norm": 7.678726275837568, - "learning_rate": 1.9026989079563183e-05, - "loss": 0.257, + "grad_norm": 7.706542458286787, + "learning_rate": 1.4314736842105266e-05, + "loss": 0.375, "step": 1220 }, { "epoch": 0.15, - "grad_norm": 11.103121729216026, - "learning_rate": 1.933588143525741e-05, - "loss": 0.2615, + "grad_norm": 10.89212756963322, + "learning_rate": 1.4546315789473687e-05, + "loss": 0.4049, "step": 1240 }, { "epoch": 0.15, - "grad_norm": 7.821432637451867, - "learning_rate": 1.964477379095164e-05, - "loss": 0.2802, + "grad_norm": 11.46198790559678, + "learning_rate": 1.4777894736842107e-05, + "loss": 0.332, "step": 1260 }, { "epoch": 0.15, - "grad_norm": 6.305517886474544, - "learning_rate": 1.9953666146645868e-05, - "loss": 0.2799, + "grad_norm": 8.365156167551179, + "learning_rate": 1.500947368421053e-05, + "loss": 0.3586, "step": 1280 }, { "epoch": 0.15, - "grad_norm": 6.67203594790867, - "learning_rate": 1.9999973049973136e-05, - "loss": 0.2708, + "grad_norm": 7.841036992292662, + "learning_rate": 1.5241052631578949e-05, + "loss": 0.368, "step": 1300 }, { "epoch": 0.15, - "grad_norm": 5.618638234623427, - "learning_rate": 1.999987988957295e-05, - "loss": 0.2964, + "grad_norm": 9.84275168548854, + "learning_rate": 1.547263157894737e-05, + "loss": 0.3767, "step": 1320 }, { "epoch": 0.16, - "grad_norm": 7.048756147457391, - "learning_rate": 1.9999720186702903e-05, - "loss": 0.2883, + "grad_norm": 10.506254240871392, + "learning_rate": 1.5704210526315792e-05, + "loss": 0.378, "step": 1340 }, { "epoch": 0.16, - "grad_norm": 7.697343680221464, - "learning_rate": 1.999949394242581e-05, - "loss": 0.3157, + "grad_norm": 7.844581104033147, + "learning_rate": 1.5935789473684213e-05, + "loss": 0.3453, "step": 1360 }, { "epoch": 0.16, - "grad_norm": 5.891830361487415, - "learning_rate": 1.9999201158247326e-05, - "loss": 0.2823, + "grad_norm": 8.193758138016792, + "learning_rate": 1.6167368421052634e-05, + "loss": 0.3752, "step": 1380 }, { "epoch": 0.16, - "grad_norm": 6.585748092805747, - "learning_rate": 1.9998841836115927e-05, - "loss": 0.2284, + "grad_norm": 7.475620318034102, + "learning_rate": 1.6398947368421055e-05, + "loss": 0.3642, "step": 1400 }, { "epoch": 0.17, - "grad_norm": 5.437814286894138, - "learning_rate": 1.99984159784229e-05, - "loss": 0.2702, + "grad_norm": 7.4964850772334355, + "learning_rate": 1.6630526315789475e-05, + "loss": 0.3377, "step": 1420 }, { "epoch": 0.17, - "grad_norm": 6.817107156388952, - "learning_rate": 1.9997923588002327e-05, - "loss": 0.2801, + "grad_norm": 7.903753436810423, + "learning_rate": 1.6862105263157896e-05, + "loss": 0.3586, "step": 1440 }, { "epoch": 0.17, - "grad_norm": 18.42457670763344, - "learning_rate": 1.9997364668131065e-05, - "loss": 0.3078, + "grad_norm": 10.49594728277505, + "learning_rate": 1.7093684210526317e-05, + "loss": 0.3468, "step": 1460 }, { "epoch": 0.17, - "grad_norm": 9.998003569634188, - "learning_rate": 1.9996739222528713e-05, - "loss": 0.2839, + "grad_norm": 8.027713457118994, + "learning_rate": 1.7325263157894738e-05, + "loss": 0.3491, "step": 1480 }, { "epoch": 0.18, - "grad_norm": 6.684087435193273, - "learning_rate": 1.999604725535761e-05, - "loss": 0.3049, + "grad_norm": 7.2748681125287415, + "learning_rate": 1.755684210526316e-05, + "loss": 0.3581, "step": 1500 }, { "epoch": 0.18, - "grad_norm": 6.376706960696693, - "learning_rate": 1.9995288771222792e-05, - "loss": 0.2556, + "grad_norm": 6.077326940397276, + "learning_rate": 1.7788421052631583e-05, + "loss": 0.3356, "step": 1520 }, { "epoch": 0.18, - "grad_norm": 8.261169271029942, - "learning_rate": 1.999446377517197e-05, - "loss": 0.2766, + "grad_norm": 6.724980645701413, + "learning_rate": 1.802e-05, + "loss": 0.3656, "step": 1540 }, { "epoch": 0.18, - "grad_norm": 6.805041788997332, - "learning_rate": 1.999357227269548e-05, - "loss": 0.2727, + "grad_norm": 8.592315221136559, + "learning_rate": 1.825157894736842e-05, + "loss": 0.3467, "step": 1560 }, { "epoch": 0.18, - "grad_norm": 6.8738954523752795, - "learning_rate": 1.999261426972627e-05, - "loss": 0.2616, + "grad_norm": 9.036253074638084, + "learning_rate": 1.8483157894736845e-05, + "loss": 0.3414, "step": 1580 }, { "epoch": 0.19, - "grad_norm": 5.946552322782328, - "learning_rate": 1.9991589772639836e-05, - "loss": 0.2782, + "grad_norm": 11.349080832480995, + "learning_rate": 1.8714736842105266e-05, + "loss": 0.3655, "step": 1600 }, { "epoch": 0.19, - "grad_norm": 8.221627199562976, - "learning_rate": 1.9990498788254206e-05, - "loss": 0.2675, + "grad_norm": 9.361475025905081, + "learning_rate": 1.8946315789473687e-05, + "loss": 0.3564, "step": 1620 }, { "epoch": 0.19, - "grad_norm": 8.064560737732597, - "learning_rate": 1.9989341323829865e-05, - "loss": 0.2947, + "grad_norm": 6.53643855564322, + "learning_rate": 1.9177894736842108e-05, + "loss": 0.3714, "step": 1640 }, { "epoch": 0.19, - "grad_norm": 15.364919443710688, - "learning_rate": 1.9988117387069733e-05, - "loss": 0.2752, + "grad_norm": 6.65699298275954, + "learning_rate": 1.940947368421053e-05, + "loss": 0.3546, "step": 1660 }, { "epoch": 0.2, - "grad_norm": 5.108588948577897, - "learning_rate": 1.9986826986119102e-05, - "loss": 0.2588, + "grad_norm": 9.856127510930383, + "learning_rate": 1.964105263157895e-05, + "loss": 0.3308, "step": 1680 }, { "epoch": 0.2, - "grad_norm": 7.676958998385404, - "learning_rate": 1.998547012956558e-05, - "loss": 0.2639, + "grad_norm": 8.532146476622113, + "learning_rate": 1.987263157894737e-05, + "loss": 0.366, "step": 1700 }, { "epoch": 0.2, - "grad_norm": 6.998387014654503, - "learning_rate": 1.9984046826439038e-05, - "loss": 0.2687, + "grad_norm": 8.327666937449752, + "learning_rate": 1.999999532098205e-05, + "loss": 0.3247, "step": 1720 }, { "epoch": 0.2, - "grad_norm": 5.572858958673421, - "learning_rate": 1.9982557086211542e-05, - "loss": 0.2903, + "grad_norm": 10.876035443701728, + "learning_rate": 1.99999578888647e-05, + "loss": 0.3543, "step": 1740 }, { "epoch": 0.21, - "grad_norm": 8.209097528142939, - "learning_rate": 1.9981000918797306e-05, - "loss": 0.2608, + "grad_norm": 8.263711805253918, + "learning_rate": 1.999988302477013e-05, + "loss": 0.3664, "step": 1760 }, { "epoch": 0.21, - "grad_norm": 7.106558508035025, - "learning_rate": 1.9979378334552614e-05, - "loss": 0.2847, + "grad_norm": 7.575419338189409, + "learning_rate": 1.9999770728978607e-05, + "loss": 0.345, "step": 1780 }, { "epoch": 0.21, - "grad_norm": 4.4357068374392234, - "learning_rate": 1.997768934427575e-05, - "loss": 0.2348, + "grad_norm": 7.814279338388738, + "learning_rate": 1.9999621001910508e-05, + "loss": 0.3573, "step": 1800 }, { "epoch": 0.21, - "grad_norm": 6.012720927979585, - "learning_rate": 1.9975933959206926e-05, - "loss": 0.2597, + "grad_norm": 13.040107479471628, + "learning_rate": 1.999943384412636e-05, + "loss": 0.3344, "step": 1820 }, { "epoch": 0.22, - "grad_norm": 7.101130876339373, - "learning_rate": 1.997411219102823e-05, - "loss": 0.2804, + "grad_norm": 7.535441012066771, + "learning_rate": 1.99992092563268e-05, + "loss": 0.3552, "step": 1840 }, { "epoch": 0.22, - "grad_norm": 6.753118571675995, - "learning_rate": 1.99722240518635e-05, - "loss": 0.2635, + "grad_norm": 8.059491683038381, + "learning_rate": 1.9998947239352593e-05, + "loss": 0.3442, "step": 1860 }, { "epoch": 0.22, - "grad_norm": 5.548898247312993, - "learning_rate": 1.9970269554278296e-05, - "loss": 0.2312, + "grad_norm": 6.700609139607897, + "learning_rate": 1.999864779418462e-05, + "loss": 0.3584, "step": 1880 }, { "epoch": 0.22, - "grad_norm": 6.659248070869921, - "learning_rate": 1.9968248711279784e-05, - "loss": 0.2457, + "grad_norm": 6.831136707088909, + "learning_rate": 1.999831092194389e-05, + "loss": 0.3341, "step": 1900 }, { "epoch": 0.22, - "grad_norm": 7.084690577306244, - "learning_rate": 1.996616153631666e-05, - "loss": 0.2654, + "grad_norm": 10.30262237054047, + "learning_rate": 1.9997936623891498e-05, + "loss": 0.3704, "step": 1920 }, { "epoch": 0.23, - "grad_norm": 5.444008275594034, - "learning_rate": 1.9964008043279056e-05, - "loss": 0.24, + "grad_norm": 6.703691792033561, + "learning_rate": 1.9997524901428677e-05, + "loss": 0.3769, "step": 1940 }, { "epoch": 0.23, - "grad_norm": 5.36653730738069, - "learning_rate": 1.996178824649845e-05, - "loss": 0.2841, + "grad_norm": 7.251026212477321, + "learning_rate": 1.9997075756096734e-05, + "loss": 0.3234, "step": 1960 }, { "epoch": 0.23, - "grad_norm": 5.808350425195198, - "learning_rate": 1.9959502160747575e-05, - "loss": 0.2706, + "grad_norm": 6.14471190830255, + "learning_rate": 1.9996589189577094e-05, + "loss": 0.3443, "step": 1980 }, { "epoch": 0.23, - "grad_norm": 5.156993650505354, - "learning_rate": 1.9957149801240313e-05, - "loss": 0.2696, + "grad_norm": 7.254885638733633, + "learning_rate": 1.999606520369125e-05, + "loss": 0.3178, "step": 2000 }, { "epoch": 0.24, - "grad_norm": 5.642951633427938, - "learning_rate": 1.99547311836316e-05, - "loss": 0.2644, + "grad_norm": 7.330977870771951, + "learning_rate": 1.9995503800400805e-05, + "loss": 0.3309, "step": 2020 }, { "epoch": 0.24, - "grad_norm": 3.4649157253216916, - "learning_rate": 1.995224632401732e-05, - "loss": 0.2567, + "grad_norm": 7.291895603037368, + "learning_rate": 1.9994904981807407e-05, + "loss": 0.3372, "step": 2040 }, { "epoch": 0.24, - "grad_norm": 10.299176851731511, - "learning_rate": 1.9949695238934187e-05, - "loss": 0.264, + "grad_norm": 12.772894555548028, + "learning_rate": 1.9994268750152793e-05, + "loss": 0.3747, "step": 2060 }, { "epoch": 0.24, - "grad_norm": 9.341352786786667, - "learning_rate": 1.994707794535966e-05, - "loss": 0.288, + "grad_norm": 9.497788021161657, + "learning_rate": 1.999359510781875e-05, + "loss": 0.3529, "step": 2080 }, { "epoch": 0.25, - "grad_norm": 5.416770268941113, - "learning_rate": 1.9944394460711806e-05, - "loss": 0.2342, + "grad_norm": 6.581748417513793, + "learning_rate": 1.9992884057327118e-05, + "loss": 0.3395, "step": 2100 }, { "epoch": 0.25, - "grad_norm": 6.406230051893485, - "learning_rate": 1.9941644802849197e-05, - "loss": 0.2684, + "grad_norm": 9.523030924078585, + "learning_rate": 1.9992135601339776e-05, + "loss": 0.3469, "step": 2120 }, { "epoch": 0.25, - "grad_norm": 5.161284052745025, - "learning_rate": 1.993882899007078e-05, - "loss": 0.2733, + "grad_norm": 17.640537636796516, + "learning_rate": 1.999134974265864e-05, + "loss": 0.3365, "step": 2140 }, { "epoch": 0.25, - "grad_norm": 5.193421993616942, - "learning_rate": 1.9935947041115774e-05, - "loss": 0.2623, + "grad_norm": 5.4385511763432985, + "learning_rate": 1.999052648422563e-05, + "loss": 0.3845, "step": 2160 }, { "epoch": 0.26, - "grad_norm": 13.024715697248084, - "learning_rate": 1.9932998975163525e-05, - "loss": 0.2702, + "grad_norm": 7.585921156892153, + "learning_rate": 1.9989665829122704e-05, + "loss": 0.3464, "step": 2180 }, { "epoch": 0.26, - "grad_norm": 6.8241910797414125, - "learning_rate": 1.9929984811833385e-05, - "loss": 0.2566, + "grad_norm": 12.163408304702841, + "learning_rate": 1.9988767780571778e-05, + "loss": 0.3127, "step": 2200 }, { "epoch": 0.26, - "grad_norm": 6.730073692519145, - "learning_rate": 1.9926904571184594e-05, - "loss": 0.242, + "grad_norm": 8.246994395835456, + "learning_rate": 1.9987832341934793e-05, + "loss": 0.3424, "step": 2220 }, { "epoch": 0.26, - "grad_norm": 9.451222932996513, - "learning_rate": 1.992375827371613e-05, - "loss": 0.278, + "grad_norm": 9.628684787640838, + "learning_rate": 1.9986859516713636e-05, + "loss": 0.3472, "step": 2240 }, { "epoch": 0.26, - "grad_norm": 8.019858508300759, - "learning_rate": 1.9920545940366576e-05, - "loss": 0.2341, + "grad_norm": 10.912326639203211, + "learning_rate": 1.9985849308550167e-05, + "loss": 0.3326, "step": 2260 }, { "epoch": 0.27, - "grad_norm": 5.705986412954751, - "learning_rate": 1.9917267592513983e-05, - "loss": 0.2598, + "grad_norm": 6.419410251137725, + "learning_rate": 1.9984801721226188e-05, + "loss": 0.3297, "step": 2280 }, { "epoch": 0.27, - "grad_norm": 6.536565196669616, - "learning_rate": 1.991392325197574e-05, - "loss": 0.3296, + "grad_norm": 7.798212276898864, + "learning_rate": 1.998371675866344e-05, + "loss": 0.3389, "step": 2300 }, { "epoch": 0.27, - "grad_norm": 6.445428798666668, - "learning_rate": 1.99105129410084e-05, - "loss": 0.26, + "grad_norm": 7.811483576458649, + "learning_rate": 1.9982594424923566e-05, + "loss": 0.332, "step": 2320 }, { "epoch": 0.27, - "grad_norm": 7.151819181763656, - "learning_rate": 1.9907036682307563e-05, - "loss": 0.2571, + "grad_norm": 6.967736687994363, + "learning_rate": 1.9981434724208126e-05, + "loss": 0.3099, "step": 2340 }, { "epoch": 0.28, - "grad_norm": 5.051575919032304, - "learning_rate": 1.99034944990077e-05, - "loss": 0.2623, + "grad_norm": 9.635115819134949, + "learning_rate": 1.9980237660858565e-05, + "loss": 0.3285, "step": 2360 }, { "epoch": 0.28, - "grad_norm": 4.396965905437974, - "learning_rate": 1.9899886414682012e-05, - "loss": 0.222, + "grad_norm": 6.720069081541748, + "learning_rate": 1.997900323935619e-05, + "loss": 0.343, "step": 2380 }, { "epoch": 0.28, - "grad_norm": 5.550275936655984, - "learning_rate": 1.989621245334228e-05, - "loss": 0.2472, + "grad_norm": 10.300889386045979, + "learning_rate": 1.9977731464322162e-05, + "loss": 0.3667, "step": 2400 }, { "epoch": 0.28, - "grad_norm": 7.484173415620986, - "learning_rate": 1.9892472639438684e-05, - "loss": 0.2572, + "grad_norm": 7.888854177455323, + "learning_rate": 1.9976422340517495e-05, + "loss": 0.3397, "step": 2420 }, { "epoch": 0.29, - "grad_norm": 6.330013944541192, - "learning_rate": 1.9888666997859657e-05, - "loss": 0.256, + "grad_norm": 9.643051064776678, + "learning_rate": 1.9975075872843e-05, + "loss": 0.3371, "step": 2440 }, { "epoch": 0.29, - "grad_norm": 6.839909560291623, - "learning_rate": 1.9884795553931713e-05, - "loss": 0.2505, + "grad_norm": 6.166262467342929, + "learning_rate": 1.9973692066339298e-05, + "loss": 0.3424, "step": 2460 }, { "epoch": 0.29, - "grad_norm": 3.7368937022928987, - "learning_rate": 1.988085833341929e-05, - "loss": 0.2232, + "grad_norm": 7.520601841401043, + "learning_rate": 1.9972270926186794e-05, + "loss": 0.3083, "step": 2480 }, { "epoch": 0.29, - "grad_norm": 7.700733239077536, - "learning_rate": 1.9876855362524555e-05, - "loss": 0.2513, + "grad_norm": 6.095472200047568, + "learning_rate": 1.9970812457705642e-05, + "loss": 0.3512, "step": 2500 }, { "epoch": 0.29, - "grad_norm": 9.003459689305725, - "learning_rate": 1.9872786667887246e-05, - "loss": 0.2701, + "grad_norm": 7.259492119293066, + "learning_rate": 1.9969316666355757e-05, + "loss": 0.3309, "step": 2520 }, { "epoch": 0.3, - "grad_norm": 6.173066538462164, - "learning_rate": 1.9868652276584498e-05, - "loss": 0.2641, + "grad_norm": 8.355978245335697, + "learning_rate": 1.996778355773676e-05, + "loss": 0.3296, "step": 2540 }, { "epoch": 0.3, - "grad_norm": 7.517799617986063, - "learning_rate": 1.9864452216130656e-05, - "loss": 0.2602, + "grad_norm": 9.612502269967832, + "learning_rate": 1.996621313758798e-05, + "loss": 0.3391, "step": 2560 }, { "epoch": 0.3, - "grad_norm": 5.206716769795917, - "learning_rate": 1.986018651447709e-05, - "loss": 0.2702, + "grad_norm": 8.148285817739712, + "learning_rate": 1.9964605411788424e-05, + "loss": 0.3416, "step": 2580 }, { "epoch": 0.3, - "grad_norm": 5.352469307581187, - "learning_rate": 1.9855855200012012e-05, - "loss": 0.2551, + "grad_norm": 7.645206131186397, + "learning_rate": 1.9962960386356748e-05, + "loss": 0.352, "step": 2600 }, { "epoch": 0.31, - "grad_norm": 4.307397892605768, - "learning_rate": 1.985145830156028e-05, - "loss": 0.2641, + "grad_norm": 5.406075006199623, + "learning_rate": 1.9961278067451258e-05, + "loss": 0.3194, "step": 2620 }, { "epoch": 0.31, - "grad_norm": 4.264075140901128, - "learning_rate": 1.9846995848383227e-05, - "loss": 0.2531, + "grad_norm": 7.23674984010716, + "learning_rate": 1.995955846136986e-05, + "loss": 0.3216, "step": 2640 }, { "epoch": 0.31, - "grad_norm": 7.436472589390343, - "learning_rate": 1.9842467870178446e-05, - "loss": 0.2338, + "grad_norm": 8.044071872202302, + "learning_rate": 1.9957801574550044e-05, + "loss": 0.321, "step": 2660 }, { "epoch": 0.31, - "grad_norm": 7.224638151723576, - "learning_rate": 1.9837874397079597e-05, - "loss": 0.2704, + "grad_norm": 7.2624583573058485, + "learning_rate": 1.9956007413568874e-05, + "loss": 0.3352, "step": 2680 }, { "epoch": 0.32, - "grad_norm": 5.457029152871823, - "learning_rate": 1.9833215459656204e-05, - "loss": 0.2268, + "grad_norm": 9.277277549105909, + "learning_rate": 1.9954175985142955e-05, + "loss": 0.3391, "step": 2700 }, { "epoch": 0.32, - "grad_norm": 8.049908882715352, - "learning_rate": 1.9828491088913466e-05, - "loss": 0.2761, + "grad_norm": 10.975275864316648, + "learning_rate": 1.9952307296128395e-05, + "loss": 0.3298, "step": 2720 }, { "epoch": 0.32, - "grad_norm": 5.589531357284561, - "learning_rate": 1.9823701316292027e-05, - "loss": 0.2226, + "grad_norm": 11.416959735607907, + "learning_rate": 1.9950401353520795e-05, + "loss": 0.3382, "step": 2740 }, { "epoch": 0.32, - "grad_norm": 5.958210692666308, - "learning_rate": 1.9818846173667795e-05, - "loss": 0.2351, + "grad_norm": 7.384993942351236, + "learning_rate": 1.994845816445522e-05, + "loss": 0.3422, "step": 2760 }, { "epoch": 0.33, - "grad_norm": 4.946607085809116, - "learning_rate": 1.9813925693351706e-05, - "loss": 0.2428, + "grad_norm": 6.775839902047113, + "learning_rate": 1.9946477736206167e-05, + "loss": 0.3694, "step": 2780 }, { "epoch": 0.33, - "grad_norm": 5.0976059249790815, - "learning_rate": 1.9808939908089514e-05, - "loss": 0.2694, + "grad_norm": 8.246414330131564, + "learning_rate": 1.9944460076187546e-05, + "loss": 0.3458, "step": 2800 }, { "epoch": 0.33, - "grad_norm": 3.8642265929763013, - "learning_rate": 1.9803888851061584e-05, - "loss": 0.2507, + "grad_norm": 7.152660285783071, + "learning_rate": 1.9942405191952646e-05, + "loss": 0.3426, "step": 2820 }, { "epoch": 0.33, - "grad_norm": 5.4071017790539, - "learning_rate": 1.979877255588267e-05, - "loss": 0.2443, + "grad_norm": 6.9408668390055315, + "learning_rate": 1.9940313091194096e-05, + "loss": 0.3097, "step": 2840 }, { "epoch": 0.33, - "grad_norm": 5.291228849466517, - "learning_rate": 1.979359105660167e-05, - "loss": 0.2449, + "grad_norm": 5.797513942002627, + "learning_rate": 1.9938183781743873e-05, + "loss": 0.3102, "step": 2860 }, { "epoch": 0.34, - "grad_norm": 18.084778243085054, - "learning_rate": 1.9788344387701422e-05, - "loss": 0.2587, + "grad_norm": 7.361935180345544, + "learning_rate": 1.9936017271573223e-05, + "loss": 0.3102, "step": 2880 }, { "epoch": 0.34, - "grad_norm": 7.733243732371791, - "learning_rate": 1.9783032584098475e-05, - "loss": 0.2534, + "grad_norm": 7.4796257480282, + "learning_rate": 1.9933813568792678e-05, + "loss": 0.3256, "step": 2900 }, { "epoch": 0.34, - "grad_norm": 5.9419057461797875, - "learning_rate": 1.9777655681142828e-05, - "loss": 0.2887, + "grad_norm": 5.500192812227557, + "learning_rate": 1.9931572681651984e-05, + "loss": 0.3107, "step": 2920 }, { "epoch": 0.34, - "grad_norm": 6.194967060632705, - "learning_rate": 1.9772213714617746e-05, - "loss": 0.271, + "grad_norm": 13.84776864376109, + "learning_rate": 1.9929294618540103e-05, + "loss": 0.3062, "step": 2940 }, { "epoch": 0.35, - "grad_norm": 5.416199101539707, - "learning_rate": 1.9766706720739462e-05, - "loss": 0.2479, + "grad_norm": 7.937410128828084, + "learning_rate": 1.992697938798516e-05, + "loss": 0.3293, "step": 2960 }, { "epoch": 0.35, - "grad_norm": 7.381640790370163, - "learning_rate": 1.9761134736156984e-05, - "loss": 0.253, + "grad_norm": 5.447364190742796, + "learning_rate": 1.9924626998654433e-05, + "loss": 0.3318, "step": 2980 }, { "epoch": 0.35, - "grad_norm": 5.289607190437299, - "learning_rate": 1.975549779795183e-05, - "loss": 0.2398, + "grad_norm": 7.825239369201605, + "learning_rate": 1.9922237459354287e-05, + "loss": 0.3238, "step": 3000 }, { "epoch": 0.35, - "grad_norm": 6.393511952492927, - "learning_rate": 1.9749795943637778e-05, - "loss": 0.2527, + "grad_norm": 6.466926354506395, + "learning_rate": 1.9919810779030177e-05, + "loss": 0.3136, "step": 3020 }, { "epoch": 0.36, - "grad_norm": 5.786789008808832, - "learning_rate": 1.9744029211160623e-05, - "loss": 0.264, + "grad_norm": 10.115420483963225, + "learning_rate": 1.9917346966766596e-05, + "loss": 0.3329, "step": 3040 }, { "epoch": 0.36, - "grad_norm": 6.325619802149641, - "learning_rate": 1.9738197638897923e-05, - "loss": 0.2604, + "grad_norm": 6.08100630613984, + "learning_rate": 1.991484603178704e-05, + "loss": 0.2999, "step": 3060 }, { "epoch": 0.36, - "grad_norm": 5.138427821373369, - "learning_rate": 1.973230126565876e-05, - "loss": 0.2501, + "grad_norm": 6.153474908414458, + "learning_rate": 1.991230798345398e-05, + "loss": 0.3344, "step": 3080 }, { "epoch": 0.36, - "grad_norm": 5.754983152469265, - "learning_rate": 1.972634013068344e-05, - "loss": 0.2563, + "grad_norm": 6.956842076140817, + "learning_rate": 1.990973283126882e-05, + "loss": 0.3356, "step": 3100 }, { "epoch": 0.37, - "grad_norm": 8.479269441048931, - "learning_rate": 1.9720314273643277e-05, - "loss": 0.2428, + "grad_norm": 7.977473013654998, + "learning_rate": 1.990712058487187e-05, + "loss": 0.2984, "step": 3120 }, { "epoch": 0.37, - "grad_norm": 5.213630952429399, - "learning_rate": 1.9714223734640303e-05, - "loss": 0.2161, + "grad_norm": 7.472067890046691, + "learning_rate": 1.9904471254042304e-05, + "loss": 0.308, "step": 3140 }, { "epoch": 0.37, - "grad_norm": 4.644334537071628, - "learning_rate": 1.970806855420701e-05, - "loss": 0.2529, + "grad_norm": 6.98090482864238, + "learning_rate": 1.9901784848698125e-05, + "loss": 0.3036, "step": 3160 }, { "epoch": 0.37, - "grad_norm": 6.1305106149806665, - "learning_rate": 1.9701848773306074e-05, - "loss": 0.236, + "grad_norm": 7.229553593404789, + "learning_rate": 1.989906137889613e-05, + "loss": 0.3027, "step": 3180 }, { "epoch": 0.37, - "grad_norm": 5.353262482329062, - "learning_rate": 1.9695564433330096e-05, - "loss": 0.2549, + "grad_norm": 5.9700766482137295, + "learning_rate": 1.9896300854831862e-05, + "loss": 0.3329, "step": 3200 }, { "epoch": 0.38, - "grad_norm": 5.060708753381077, - "learning_rate": 1.9689215576101306e-05, - "loss": 0.215, + "grad_norm": 7.864131148182889, + "learning_rate": 1.989350328683959e-05, + "loss": 0.308, "step": 3220 }, { "epoch": 0.38, - "grad_norm": 6.674948634260756, - "learning_rate": 1.9682802243871302e-05, - "loss": 0.2383, + "grad_norm": 20.04577205534678, + "learning_rate": 1.989066868539225e-05, + "loss": 0.3217, "step": 3240 }, { "epoch": 0.38, - "grad_norm": 7.099655314930459, - "learning_rate": 1.9676324479320758e-05, - "loss": 0.2343, + "grad_norm": 8.46751929768339, + "learning_rate": 1.9887797061101424e-05, + "loss": 0.3036, "step": 3260 }, { "epoch": 0.38, - "grad_norm": 6.495667113827656, - "learning_rate": 1.9669782325559156e-05, - "loss": 0.271, + "grad_norm": 7.667111535917501, + "learning_rate": 1.9884888424717296e-05, + "loss": 0.3084, "step": 3280 }, { "epoch": 0.39, - "grad_norm": 8.167355047783982, - "learning_rate": 1.9663175826124467e-05, - "loss": 0.275, + "grad_norm": 6.615069450077278, + "learning_rate": 1.9881942787128594e-05, + "loss": 0.3177, "step": 3300 }, { "epoch": 0.39, - "grad_norm": 5.097460933758238, - "learning_rate": 1.9656505024982906e-05, - "loss": 0.2525, + "grad_norm": 9.324876177990959, + "learning_rate": 1.9878960159362572e-05, + "loss": 0.3496, "step": 3320 }, { "epoch": 0.39, - "grad_norm": 3.5743234754551723, - "learning_rate": 1.9649769966528606e-05, - "loss": 0.2607, + "grad_norm": 7.1979554134343635, + "learning_rate": 1.987594055258496e-05, + "loss": 0.3167, "step": 3340 }, { "epoch": 0.39, - "grad_norm": 6.011608021133435, - "learning_rate": 1.9642970695583317e-05, - "loss": 0.2493, + "grad_norm": 29.076306154493622, + "learning_rate": 1.987288397809993e-05, + "loss": 0.34, "step": 3360 }, { "epoch": 0.4, - "grad_norm": 4.773080465946433, - "learning_rate": 1.9636107257396144e-05, - "loss": 0.231, + "grad_norm": 7.861933206527291, + "learning_rate": 1.986979044735002e-05, + "loss": 0.3424, "step": 3380 }, { "epoch": 0.4, - "grad_norm": 5.449977370748874, - "learning_rate": 1.9629179697643214e-05, - "loss": 0.2466, + "grad_norm": 5.7878723156599445, + "learning_rate": 1.9866659971916146e-05, + "loss": 0.3157, "step": 3400 }, { "epoch": 0.4, - "grad_norm": 5.36397841910189, - "learning_rate": 1.9622188062427393e-05, - "loss": 0.2531, + "grad_norm": 8.470531768859948, + "learning_rate": 1.9863492563517513e-05, + "loss": 0.3201, "step": 3420 }, { "epoch": 0.4, - "grad_norm": 5.451392655529927, - "learning_rate": 1.9615132398277943e-05, - "loss": 0.2354, + "grad_norm": 8.104846079245535, + "learning_rate": 1.98602882340116e-05, + "loss": 0.322, "step": 3440 }, { "epoch": 0.4, - "grad_norm": 7.143537532553735, - "learning_rate": 1.960801275215027e-05, - "loss": 0.2632, + "grad_norm": 6.180289951670451, + "learning_rate": 1.9857046995394084e-05, + "loss": 0.2991, "step": 3460 }, { "epoch": 0.41, - "grad_norm": 3.4010860952344717, - "learning_rate": 1.960082917142556e-05, - "loss": 0.2362, + "grad_norm": 4.962160167060266, + "learning_rate": 1.985376885979884e-05, + "loss": 0.3097, "step": 3480 }, { "epoch": 0.41, - "grad_norm": 5.840079557321102, - "learning_rate": 1.959358170391049e-05, - "loss": 0.2402, + "grad_norm": 6.813966925636774, + "learning_rate": 1.9850453839497844e-05, + "loss": 0.3294, "step": 3500 }, { "epoch": 0.41, - "grad_norm": 5.938052824284585, - "learning_rate": 1.9586270397836898e-05, - "loss": 0.2407, + "grad_norm": 5.902763508299491, + "learning_rate": 1.9847101946901172e-05, + "loss": 0.2967, "step": 3520 }, { "epoch": 0.41, - "grad_norm": 5.132153426638341, - "learning_rate": 1.9578895301861475e-05, - "loss": 0.2401, + "grad_norm": 7.359798475906595, + "learning_rate": 1.984371319455693e-05, + "loss": 0.3268, "step": 3540 }, { "epoch": 0.42, - "grad_norm": 5.87746718526918, - "learning_rate": 1.9571456465065425e-05, - "loss": 0.2749, + "grad_norm": 7.377881796357637, + "learning_rate": 1.9840287595151205e-05, + "loss": 0.3251, "step": 3560 }, { "epoch": 0.42, - "grad_norm": 11.830889239002573, - "learning_rate": 1.956395393695415e-05, - "loss": 0.2302, + "grad_norm": 7.261601866242624, + "learning_rate": 1.9836825161508032e-05, + "loss": 0.284, "step": 3580 }, { "epoch": 0.42, - "grad_norm": 4.661847737277734, - "learning_rate": 1.955638776745692e-05, - "loss": 0.2359, + "grad_norm": 6.123609423918951, + "learning_rate": 1.983332590658934e-05, + "loss": 0.3137, "step": 3600 }, { "epoch": 0.42, - "grad_norm": 4.245057766614552, - "learning_rate": 1.9548758006926537e-05, - "loss": 0.2534, + "grad_norm": 7.205290164862431, + "learning_rate": 1.9829789843494885e-05, + "loss": 0.3, "step": 3620 }, { "epoch": 0.43, - "grad_norm": 6.400102149625017, - "learning_rate": 1.9541064706138998e-05, - "loss": 0.2513, + "grad_norm": 6.0471563544848035, + "learning_rate": 1.9826216985462245e-05, + "loss": 0.2991, "step": 3640 }, { "epoch": 0.43, - "grad_norm": 7.7926185012927105, - "learning_rate": 1.953330791629316e-05, - "loss": 0.2508, + "grad_norm": 7.436589545934819, + "learning_rate": 1.9822607345866715e-05, + "loss": 0.3299, "step": 3660 }, { "epoch": 0.43, - "grad_norm": 5.502373096713324, - "learning_rate": 1.9525487689010402e-05, - "loss": 0.2316, + "grad_norm": 8.240853065109379, + "learning_rate": 1.9818960938221308e-05, + "loss": 0.2886, "step": 3680 }, { "epoch": 0.43, - "grad_norm": 4.600917699743126, - "learning_rate": 1.951760407633428e-05, - "loss": 0.2207, + "grad_norm": 5.451034188795461, + "learning_rate": 1.9815277776176663e-05, + "loss": 0.3138, "step": 3700 }, { "epoch": 0.44, - "grad_norm": 5.748567153866237, - "learning_rate": 1.950965713073017e-05, - "loss": 0.2383, + "grad_norm": 7.874196329434049, + "learning_rate": 1.9811557873521027e-05, + "loss": 0.2946, "step": 3720 }, { "epoch": 0.44, - "grad_norm": 7.113580486910231, - "learning_rate": 1.950164690508494e-05, - "loss": 0.2362, + "grad_norm": 6.547528436867746, + "learning_rate": 1.9807801244180174e-05, + "loss": 0.2978, "step": 3740 }, { "epoch": 0.44, - "grad_norm": 4.385449600398098, - "learning_rate": 1.9493573452706576e-05, - "loss": 0.2646, + "grad_norm": 5.816214082411462, + "learning_rate": 1.980400790221739e-05, + "loss": 0.3067, "step": 3760 }, { "epoch": 0.44, - "grad_norm": 6.270456330213243, - "learning_rate": 1.948543682732385e-05, - "loss": 0.2558, + "grad_norm": 6.245697982302123, + "learning_rate": 1.9800177861833372e-05, + "loss": 0.3191, "step": 3780 }, { "epoch": 0.44, - "grad_norm": 6.787086330896131, - "learning_rate": 1.9477237083085934e-05, - "loss": 0.2446, + "grad_norm": 7.932045331416635, + "learning_rate": 1.9796311137366212e-05, + "loss": 0.3439, "step": 3800 }, { "epoch": 0.45, - "grad_norm": 4.424978723852011, - "learning_rate": 1.946897427456207e-05, - "loss": 0.2658, + "grad_norm": 6.088006858952151, + "learning_rate": 1.9792407743291343e-05, + "loss": 0.33, "step": 3820 }, { "epoch": 0.45, - "grad_norm": 5.168068951870145, - "learning_rate": 1.946064845674118e-05, - "loss": 0.2553, + "grad_norm": 5.770262468473867, + "learning_rate": 1.9788467694221453e-05, + "loss": 0.2897, "step": 3840 }, { "epoch": 0.45, - "grad_norm": 5.442116191965272, - "learning_rate": 1.9452259685031528e-05, - "loss": 0.27, + "grad_norm": 7.978864884873927, + "learning_rate": 1.9784491004906468e-05, + "loss": 0.3118, "step": 3860 }, { "epoch": 0.45, - "grad_norm": 6.214077368189247, - "learning_rate": 1.944380801526032e-05, - "loss": 0.2472, + "grad_norm": 6.631368778757732, + "learning_rate": 1.9780477690233476e-05, + "loss": 0.3165, "step": 3880 }, { "epoch": 0.46, - "grad_norm": 5.0773790016952125, - "learning_rate": 1.943529350367335e-05, - "loss": 0.2416, + "grad_norm": 6.541290304773085, + "learning_rate": 1.9776427765226664e-05, + "loss": 0.3379, "step": 3900 }, { "epoch": 0.46, - "grad_norm": 5.691248067465491, - "learning_rate": 1.9426716206934634e-05, - "loss": 0.2666, + "grad_norm": 6.42886929503141, + "learning_rate": 1.9772341245047285e-05, + "loss": 0.3136, "step": 3920 }, { "epoch": 0.46, - "grad_norm": 8.072722349391876, - "learning_rate": 1.9418076182126016e-05, - "loss": 0.2642, + "grad_norm": 9.004651033459043, + "learning_rate": 1.9768218144993593e-05, + "loss": 0.311, "step": 3940 }, { "epoch": 0.46, - "grad_norm": 6.147111800674382, - "learning_rate": 1.9409373486746796e-05, - "loss": 0.2736, + "grad_norm": 6.241942743461084, + "learning_rate": 1.976405848050077e-05, + "loss": 0.3224, "step": 3960 }, { "epoch": 0.47, - "grad_norm": 6.10278933178705, - "learning_rate": 1.9400608178713344e-05, - "loss": 0.2241, + "grad_norm": 7.030346892956473, + "learning_rate": 1.975986226714089e-05, + "loss": 0.3513, "step": 3980 }, { "epoch": 0.47, - "grad_norm": 11.62252289006471, - "learning_rate": 1.939178031635872e-05, - "loss": 0.2305, + "grad_norm": 6.984184655106913, + "learning_rate": 1.9755629520622835e-05, + "loss": 0.3031, "step": 4000 }, { "epoch": 0.47, - "grad_norm": 5.344565702945431, - "learning_rate": 1.9382889958432278e-05, - "loss": 0.2534, + "grad_norm": 3.6594620639728253, + "learning_rate": 1.9751360256792274e-05, + "loss": 0.2819, "step": 4020 }, { "epoch": 0.47, - "grad_norm": 5.186472227935148, - "learning_rate": 1.9373937164099286e-05, - "loss": 0.2196, + "grad_norm": 3.9453364305431418, + "learning_rate": 1.9747054491631567e-05, + "loss": 0.2806, "step": 4040 }, { "epoch": 0.48, - "grad_norm": 5.878287511563509, - "learning_rate": 1.9364921992940528e-05, - "loss": 0.2304, + "grad_norm": 3.8004852838340475, + "learning_rate": 1.9742712241259726e-05, + "loss": 0.2712, "step": 4060 }, { "epoch": 0.48, - "grad_norm": 5.994204901496365, - "learning_rate": 1.935584450495188e-05, - "loss": 0.2237, + "grad_norm": 4.671864793440983, + "learning_rate": 1.9738333521932348e-05, + "loss": 0.264, "step": 4080 }, { "epoch": 0.48, - "grad_norm": 6.796206246284836, - "learning_rate": 1.9346704760543978e-05, - "loss": 0.2437, + "grad_norm": 3.9158215253486452, + "learning_rate": 1.9733918350041546e-05, + "loss": 0.2466, "step": 4100 }, { "epoch": 0.48, - "grad_norm": 7.852752747158656, - "learning_rate": 1.9337502820541742e-05, - "loss": 0.2438, + "grad_norm": 3.232730170815344, + "learning_rate": 1.972946674211591e-05, + "loss": 0.28, "step": 4120 }, { "epoch": 0.48, - "grad_norm": 4.969180490330714, - "learning_rate": 1.9328238746184015e-05, - "loss": 0.2216, + "grad_norm": 5.26102706338969, + "learning_rate": 1.972497871482043e-05, + "loss": 0.2793, "step": 4140 }, { "epoch": 0.49, - "grad_norm": 6.896232416719303, - "learning_rate": 1.9318912599123136e-05, - "loss": 0.2268, + "grad_norm": 6.036896745386087, + "learning_rate": 1.972045428495642e-05, + "loss": 0.2551, "step": 4160 }, { "epoch": 0.49, - "grad_norm": 7.925370929112004, - "learning_rate": 1.930952444142455e-05, - "loss": 0.2182, + "grad_norm": 3.4899809349872823, + "learning_rate": 1.971589346946149e-05, + "loss": 0.2746, "step": 4180 }, { "epoch": 0.49, - "grad_norm": 5.198662571209722, - "learning_rate": 1.930007433556637e-05, - "loss": 0.2506, + "grad_norm": 4.003065603442425, + "learning_rate": 1.9711296285409443e-05, + "loss": 0.2639, "step": 4200 }, { "epoch": 0.49, - "grad_norm": 5.826437370857378, - "learning_rate": 1.9290562344438983e-05, - "loss": 0.2437, + "grad_norm": 2.8477299266017626, + "learning_rate": 1.9706662750010255e-05, + "loss": 0.2689, "step": 4220 }, { "epoch": 0.5, - "grad_norm": 6.687619217139761, - "learning_rate": 1.9280988531344616e-05, - "loss": 0.224, + "grad_norm": 3.5491058459283837, + "learning_rate": 1.970199288060996e-05, + "loss": 0.2642, "step": 4240 }, { "epoch": 0.5, - "grad_norm": 5.17828230403087, - "learning_rate": 1.9271352959996922e-05, - "loss": 0.2304, + "grad_norm": 3.658049963633355, + "learning_rate": 1.969728669469063e-05, + "loss": 0.2699, "step": 4260 }, { "epoch": 0.5, - "grad_norm": 5.633269271636816, - "learning_rate": 1.9261655694520554e-05, - "loss": 0.2501, + "grad_norm": 3.7093939532444127, + "learning_rate": 1.9692544209870287e-05, + "loss": 0.2697, "step": 4280 }, { "epoch": 0.5, - "grad_norm": 8.027110417077195, - "learning_rate": 1.925189679945075e-05, - "loss": 0.24, + "grad_norm": 3.0342042036143444, + "learning_rate": 1.968776544390283e-05, + "loss": 0.2654, "step": 4300 }, { "epoch": 0.51, - "grad_norm": 3.9526632288043206, - "learning_rate": 1.9242076339732868e-05, - "loss": 0.2464, + "grad_norm": 3.8174199112021174, + "learning_rate": 1.9682950414678004e-05, + "loss": 0.2525, "step": 4320 }, { "epoch": 0.51, - "grad_norm": 24.011155396941934, - "learning_rate": 1.9232194380722e-05, - "loss": 0.242, + "grad_norm": 5.35961771156863, + "learning_rate": 1.9678099140221284e-05, + "loss": 0.2464, "step": 4340 }, { "epoch": 0.51, - "grad_norm": 6.901238880338521, - "learning_rate": 1.922225098818251e-05, - "loss": 0.2233, + "grad_norm": 5.063501894675061, + "learning_rate": 1.9673211638693844e-05, + "loss": 0.2788, "step": 4360 }, { "epoch": 0.51, - "grad_norm": 4.6063725407318925, - "learning_rate": 1.9212246228287598e-05, - "loss": 0.2436, + "grad_norm": 3.8581072620927723, + "learning_rate": 1.9668287928392478e-05, + "loss": 0.2743, "step": 4380 }, { "epoch": 0.51, - "grad_norm": 6.595648758844639, - "learning_rate": 1.9202180167618868e-05, - "loss": 0.2274, + "grad_norm": 3.386524586463125, + "learning_rate": 1.9663328027749523e-05, + "loss": 0.2547, "step": 4400 }, { "epoch": 0.52, - "grad_norm": 4.8087240491206, - "learning_rate": 1.9192052873165873e-05, - "loss": 0.2259, + "grad_norm": 4.049063661443696, + "learning_rate": 1.9658331955332807e-05, + "loss": 0.2787, "step": 4420 }, { "epoch": 0.52, - "grad_norm": 5.967897364247756, - "learning_rate": 1.9181864412325683e-05, - "loss": 0.2087, + "grad_norm": 5.12761026294579, + "learning_rate": 1.9653299729845564e-05, + "loss": 0.2577, "step": 4440 }, { "epoch": 0.52, - "grad_norm": 5.531641886993624, - "learning_rate": 1.9171614852902424e-05, - "loss": 0.2467, + "grad_norm": 3.808933278135121, + "learning_rate": 1.9648231370126375e-05, + "loss": 0.2654, "step": 4460 }, { "epoch": 0.52, - "grad_norm": 4.809099854052731, - "learning_rate": 1.9161304263106844e-05, - "loss": 0.253, + "grad_norm": 4.402659150192465, + "learning_rate": 1.9643126895149087e-05, + "loss": 0.2603, "step": 4480 }, { "epoch": 0.53, - "grad_norm": 5.503146225359231, - "learning_rate": 1.9150932711555832e-05, - "loss": 0.2634, + "grad_norm": 8.492295307503369, + "learning_rate": 1.963798632402275e-05, + "loss": 0.2691, "step": 4500 }, { "epoch": 0.53, - "grad_norm": 6.074743433428479, - "learning_rate": 1.914050026727199e-05, - "loss": 0.241, + "grad_norm": 3.628014593555578, + "learning_rate": 1.9632809675991544e-05, + "loss": 0.2711, "step": 4520 }, { "epoch": 0.53, - "grad_norm": 9.683286476569032, - "learning_rate": 1.9130006999683156e-05, - "loss": 0.2378, + "grad_norm": 5.207256154885955, + "learning_rate": 1.9627596970434708e-05, + "loss": 0.2604, "step": 4540 }, { "epoch": 0.53, - "grad_norm": 4.549479050209584, - "learning_rate": 1.9119452978621942e-05, - "loss": 0.234, + "grad_norm": 3.892539047603965, + "learning_rate": 1.962234822686646e-05, + "loss": 0.2693, "step": 4560 }, { "epoch": 0.54, - "grad_norm": 5.11133278951115, - "learning_rate": 1.9108838274325284e-05, - "loss": 0.1995, + "grad_norm": 4.005334872796194, + "learning_rate": 1.961706346493594e-05, + "loss": 0.2774, "step": 4580 }, { "epoch": 0.54, - "grad_norm": 3.9316257072882044, - "learning_rate": 1.9098162957433953e-05, - "loss": 0.2238, + "grad_norm": 4.456881895914548, + "learning_rate": 1.9611742704427106e-05, + "loss": 0.2531, "step": 4600 }, { "epoch": 0.54, - "grad_norm": 5.645831888158784, - "learning_rate": 1.90874270989921e-05, - "loss": 0.2437, + "grad_norm": 3.925512807597538, + "learning_rate": 1.9606385965258697e-05, + "loss": 0.2507, "step": 4620 }, { "epoch": 0.54, - "grad_norm": 6.604181075055343, - "learning_rate": 1.9076630770446795e-05, - "loss": 0.2181, + "grad_norm": 4.401211069070843, + "learning_rate": 1.9600993267484147e-05, + "loss": 0.262, "step": 4640 }, { "epoch": 0.55, - "grad_norm": 5.8750461008413986, - "learning_rate": 1.9065774043647508e-05, - "loss": 0.2271, + "grad_norm": 5.3437799820979075, + "learning_rate": 1.9595564631291478e-05, + "loss": 0.2516, "step": 4660 }, { "epoch": 0.55, - "grad_norm": 4.745625740492316, - "learning_rate": 1.9054856990845676e-05, - "loss": 0.2568, + "grad_norm": 4.553539308066039, + "learning_rate": 1.9590100077003278e-05, + "loss": 0.2603, "step": 4680 }, { "epoch": 0.55, - "grad_norm": 6.812269574389316, - "learning_rate": 1.904387968469421e-05, - "loss": 0.2371, + "grad_norm": 3.169670353545694, + "learning_rate": 1.958459962507658e-05, + "loss": 0.2875, "step": 4700 }, { "epoch": 0.55, - "grad_norm": 5.799364897057997, - "learning_rate": 1.903284219824699e-05, - "loss": 0.2314, + "grad_norm": 3.572973282405317, + "learning_rate": 1.9579063296102812e-05, + "loss": 0.2671, "step": 4720 }, { "epoch": 0.55, - "grad_norm": 6.48625343415292, - "learning_rate": 1.9021744604958413e-05, - "loss": 0.219, + "grad_norm": 3.59782258517998, + "learning_rate": 1.957349111080771e-05, + "loss": 0.2837, "step": 4740 }, { "epoch": 0.56, - "grad_norm": 6.278707264795213, - "learning_rate": 1.9010586978682876e-05, - "loss": 0.2276, + "grad_norm": 5.199297439358037, + "learning_rate": 1.956788309005123e-05, + "loss": 0.2753, "step": 4760 }, { "epoch": 0.56, - "grad_norm": 5.371555555431128, - "learning_rate": 1.899936939367429e-05, - "loss": 0.2363, + "grad_norm": 4.993257936146954, + "learning_rate": 1.95622392548275e-05, + "loss": 0.2714, "step": 4780 }, { "epoch": 0.56, - "grad_norm": 5.40795130778857, - "learning_rate": 1.898809192458561e-05, - "loss": 0.2366, + "grad_norm": 3.7288744650364642, + "learning_rate": 1.955655962626471e-05, + "loss": 0.2609, "step": 4800 }, { "epoch": 0.56, - "grad_norm": 7.243141086845891, - "learning_rate": 1.8976754646468297e-05, - "loss": 0.2236, + "grad_norm": 5.377784540016259, + "learning_rate": 1.9550844225625043e-05, + "loss": 0.2605, "step": 4820 }, { "epoch": 0.57, - "grad_norm": 5.568031389685097, - "learning_rate": 1.8965357634771863e-05, - "loss": 0.2233, + "grad_norm": 3.1237811954279353, + "learning_rate": 1.954509307430461e-05, + "loss": 0.2569, "step": 4840 }, { "epoch": 0.57, - "grad_norm": 4.403544142992904, - "learning_rate": 1.895390096534332e-05, - "loss": 0.231, + "grad_norm": 4.101705029145583, + "learning_rate": 1.953930619383335e-05, + "loss": 0.2719, "step": 4860 }, { "epoch": 0.57, - "grad_norm": 5.810384057592795, - "learning_rate": 1.8942384714426724e-05, - "loss": 0.227, + "grad_norm": 2.734092828632754, + "learning_rate": 1.953348360587496e-05, + "loss": 0.2503, "step": 4880 }, { "epoch": 0.57, - "grad_norm": 7.307934240436817, - "learning_rate": 1.8930808958662634e-05, - "loss": 0.2374, + "grad_norm": 3.4516150854343604, + "learning_rate": 1.9527625332226804e-05, + "loss": 0.2588, "step": 4900 }, { "epoch": 0.58, - "grad_norm": 6.000264161970516, - "learning_rate": 1.891917377508762e-05, - "loss": 0.2357, + "grad_norm": 3.0561153265178085, + "learning_rate": 1.952173139481985e-05, + "loss": 0.2779, "step": 4920 }, { "epoch": 0.58, - "grad_norm": 6.840584636933369, - "learning_rate": 1.8907479241133743e-05, - "loss": 0.2209, + "grad_norm": 3.5096519082422986, + "learning_rate": 1.9515801815718566e-05, + "loss": 0.2567, "step": 4940 }, { "epoch": 0.58, - "grad_norm": 5.060412317899446, - "learning_rate": 1.8895725434628033e-05, - "loss": 0.2165, + "grad_norm": 3.1690240083159993, + "learning_rate": 1.9509836617120856e-05, + "loss": 0.2466, "step": 4960 }, { "epoch": 0.58, - "grad_norm": 5.819876963045717, - "learning_rate": 1.888391243379199e-05, - "loss": 0.2222, + "grad_norm": 3.2970371296868484, + "learning_rate": 1.950383582135796e-05, + "loss": 0.2762, "step": 4980 }, { "epoch": 0.59, - "grad_norm": 5.534379838696889, - "learning_rate": 1.8872040317241033e-05, - "loss": 0.2319, + "grad_norm": 4.326207366177782, + "learning_rate": 1.9497799450894387e-05, + "loss": 0.2681, "step": 5000 }, { "epoch": 0.59, - "grad_norm": 5.311173217247157, - "learning_rate": 1.8860109163984017e-05, - "loss": 0.2184, + "grad_norm": 3.977255410491104, + "learning_rate": 1.949172752832782e-05, + "loss": 0.235, "step": 5020 }, { "epoch": 0.59, - "grad_norm": 4.196681646351116, - "learning_rate": 1.884811905342267e-05, - "loss": 0.2136, + "grad_norm": 3.5634772315616665, + "learning_rate": 1.9485620076389036e-05, + "loss": 0.254, "step": 5040 }, { "epoch": 0.59, - "grad_norm": 4.147633735105589, - "learning_rate": 1.8836070065351094e-05, - "loss": 0.2228, + "grad_norm": 2.9667663371741377, + "learning_rate": 1.9479477117941815e-05, + "loss": 0.2582, "step": 5060 }, { "epoch": 0.59, - "grad_norm": 5.977670444689344, - "learning_rate": 1.882396227995521e-05, - "loss": 0.2196, + "grad_norm": 3.628580495546253, + "learning_rate": 1.9473298675982863e-05, + "loss": 0.2526, "step": 5080 }, { "epoch": 0.6, - "grad_norm": 5.9509298928281344, - "learning_rate": 1.881179577781224e-05, - "loss": 0.2128, + "grad_norm": 7.393689026504584, + "learning_rate": 1.946708477364172e-05, + "loss": 0.2587, "step": 5100 }, { "epoch": 0.6, - "grad_norm": 4.308487061660745, - "learning_rate": 1.8799570639890162e-05, - "loss": 0.2034, + "grad_norm": 3.1975448011133407, + "learning_rate": 1.9460835434180677e-05, + "loss": 0.2679, "step": 5120 }, { "epoch": 0.6, - "grad_norm": 5.579788771975886, - "learning_rate": 1.8787286947547173e-05, - "loss": 0.2163, + "grad_norm": 3.231683615655966, + "learning_rate": 1.9454550680994684e-05, + "loss": 0.2589, "step": 5140 }, { "epoch": 0.6, - "grad_norm": 5.331086082666531, - "learning_rate": 1.8774944782531157e-05, - "loss": 0.2387, + "grad_norm": 2.956736460495959, + "learning_rate": 1.9448230537611266e-05, + "loss": 0.2594, "step": 5160 }, { "epoch": 0.61, - "grad_norm": 4.859846184262665, - "learning_rate": 1.8762544226979123e-05, - "loss": 0.228, + "grad_norm": 3.326543297808526, + "learning_rate": 1.9441875027690435e-05, + "loss": 0.2556, "step": 5180 }, { "epoch": 0.61, - "grad_norm": 8.956508243561895, - "learning_rate": 1.8750085363416685e-05, - "loss": 0.205, + "grad_norm": 3.635188025395725, + "learning_rate": 1.94354841750246e-05, + "loss": 0.2391, "step": 5200 }, { "epoch": 0.61, - "grad_norm": 6.011375134855825, - "learning_rate": 1.8737568274757472e-05, - "loss": 0.2554, + "grad_norm": 4.934642109517512, + "learning_rate": 1.942905800353848e-05, + "loss": 0.2576, "step": 5220 }, { "epoch": 0.61, - "grad_norm": 4.884307583607841, - "learning_rate": 1.8724993044302627e-05, - "loss": 0.2426, + "grad_norm": 3.8685357103793767, + "learning_rate": 1.9422596537289016e-05, + "loss": 0.2735, "step": 5240 }, { "epoch": 0.62, - "grad_norm": 5.502332192912132, - "learning_rate": 1.87123597557402e-05, - "loss": 0.2113, + "grad_norm": 4.387694986472, + "learning_rate": 1.941609980046527e-05, + "loss": 0.2584, "step": 5260 }, { "epoch": 0.62, - "grad_norm": 4.10529243533445, - "learning_rate": 1.8699668493144644e-05, - "loss": 0.212, + "grad_norm": 2.8046325848231977, + "learning_rate": 1.9409567817388347e-05, + "loss": 0.2527, "step": 5280 }, { "epoch": 0.62, - "grad_norm": 5.764936204410069, - "learning_rate": 1.8686919340976204e-05, - "loss": 0.2411, + "grad_norm": 3.4012361074638466, + "learning_rate": 1.94030006125113e-05, + "loss": 0.2566, "step": 5300 }, { "epoch": 0.62, - "grad_norm": 5.522034635540896, - "learning_rate": 1.86741123840804e-05, - "loss": 0.2216, + "grad_norm": 5.450089536158873, + "learning_rate": 1.9396398210419038e-05, + "loss": 0.2667, "step": 5320 }, { "epoch": 0.62, - "grad_norm": 7.991532135733726, - "learning_rate": 1.8661247707687427e-05, - "loss": 0.2194, + "grad_norm": 3.7967573233998677, + "learning_rate": 1.938976063582824e-05, + "loss": 0.2536, "step": 5340 }, { "epoch": 0.63, - "grad_norm": 9.518656030242417, - "learning_rate": 1.864832539741161e-05, - "loss": 0.2587, + "grad_norm": 3.0487888683110866, + "learning_rate": 1.938308791358723e-05, + "loss": 0.2673, "step": 5360 }, { "epoch": 0.63, - "grad_norm": 3.577063390505517, - "learning_rate": 1.863534553925083e-05, - "loss": 0.2102, + "grad_norm": 3.2736066169116285, + "learning_rate": 1.9376380068675956e-05, + "loss": 0.2592, "step": 5380 }, { "epoch": 0.63, - "grad_norm": 6.534884231114006, - "learning_rate": 1.862230821958594e-05, - "loss": 0.2557, + "grad_norm": 9.727952335441667, + "learning_rate": 1.93696371262058e-05, + "loss": 0.273, "step": 5400 }, { "epoch": 0.63, - "grad_norm": 11.231289372192048, - "learning_rate": 1.8609213525180207e-05, - "loss": 0.2181, + "grad_norm": 3.3824664284674686, + "learning_rate": 1.936285911141958e-05, + "loss": 0.2787, "step": 5420 }, { "epoch": 0.64, - "grad_norm": 6.42773971626943, - "learning_rate": 1.8596061543178722e-05, - "loss": 0.2048, + "grad_norm": 6.879080376487243, + "learning_rate": 1.935604604969138e-05, + "loss": 0.2698, "step": 5440 }, { "epoch": 0.64, - "grad_norm": 5.645681381526835, - "learning_rate": 1.858285236110782e-05, - "loss": 0.2239, + "grad_norm": 4.579168728014292, + "learning_rate": 1.934919796652649e-05, + "loss": 0.2589, "step": 5460 }, { "epoch": 0.64, - "grad_norm": 5.521382987053305, - "learning_rate": 1.8569586066874516e-05, - "loss": 0.2536, + "grad_norm": 4.6776324362587784, + "learning_rate": 1.9342314887561323e-05, + "loss": 0.2544, "step": 5480 }, { "epoch": 0.64, - "grad_norm": 6.458490316472477, - "learning_rate": 1.8556262748765894e-05, - "loss": 0.2274, + "grad_norm": 3.693997622998789, + "learning_rate": 1.9335396838563282e-05, + "loss": 0.2446, "step": 5500 }, { "epoch": 0.65, - "grad_norm": 5.396911317032735, - "learning_rate": 1.854288249544853e-05, - "loss": 0.2208, + "grad_norm": 5.06873813162095, + "learning_rate": 1.932844384543069e-05, + "loss": 0.2618, "step": 5520 }, { "epoch": 0.65, - "grad_norm": 4.837238093735751, - "learning_rate": 1.852944539596792e-05, - "loss": 0.2088, + "grad_norm": 6.816411566527297, + "learning_rate": 1.9321455934192693e-05, + "loss": 0.2614, "step": 5540 }, { "epoch": 0.65, - "grad_norm": 4.897482454931544, - "learning_rate": 1.851595153974785e-05, - "loss": 0.2387, + "grad_norm": 3.5475232040366853, + "learning_rate": 1.9314433131009146e-05, + "loss": 0.2942, "step": 5560 }, { "epoch": 0.65, - "grad_norm": 4.932207578796922, - "learning_rate": 1.850240101658983e-05, - "loss": 0.204, + "grad_norm": 5.287550198652632, + "learning_rate": 1.9307375462170527e-05, + "loss": 0.2482, "step": 5580 }, { "epoch": 0.66, - "grad_norm": 5.536196980329085, - "learning_rate": 1.8488793916672495e-05, - "loss": 0.212, + "grad_norm": 11.389854038766874, + "learning_rate": 1.9300282954097847e-05, + "loss": 0.245, "step": 5600 }, { "epoch": 0.66, - "grad_norm": 7.074855563033988, - "learning_rate": 1.8475130330550987e-05, - "loss": 0.2258, + "grad_norm": 4.178435806663215, + "learning_rate": 1.9293155633342534e-05, + "loss": 0.2534, "step": 5620 }, { "epoch": 0.66, - "grad_norm": 4.714164678662987, - "learning_rate": 1.8461410349156373e-05, - "loss": 0.2142, + "grad_norm": 3.1710922703595035, + "learning_rate": 1.928599352658633e-05, + "loss": 0.2622, "step": 5640 }, { "epoch": 0.66, - "grad_norm": 5.118777335330323, - "learning_rate": 1.8447634063795017e-05, - "loss": 0.2258, + "grad_norm": 2.6946627953356996, + "learning_rate": 1.927879666064122e-05, + "loss": 0.2626, "step": 5660 }, { "epoch": 0.66, - "grad_norm": 5.367581593178051, - "learning_rate": 1.8433801566148005e-05, - "loss": 0.1945, + "grad_norm": 4.8581812333670955, + "learning_rate": 1.92715650624493e-05, + "loss": 0.2346, "step": 5680 }, { "epoch": 0.67, - "grad_norm": 1.9698062666888698, - "learning_rate": 1.8419912948270494e-05, - "loss": 0.201, + "grad_norm": 3.092927765677092, + "learning_rate": 1.926429875908269e-05, + "loss": 0.2585, "step": 5700 }, { "epoch": 0.67, - "grad_norm": 4.6618833323265, - "learning_rate": 1.8405968302591135e-05, - "loss": 0.2409, + "grad_norm": 4.029661826813332, + "learning_rate": 1.9256997777743437e-05, + "loss": 0.2467, "step": 5720 }, { "epoch": 0.67, - "grad_norm": 5.546962683615939, - "learning_rate": 1.839196772191144e-05, - "loss": 0.1949, + "grad_norm": 4.642691104141709, + "learning_rate": 1.924966214576341e-05, + "loss": 0.2681, "step": 5740 }, { "epoch": 0.67, - "grad_norm": 5.457193425471358, - "learning_rate": 1.8377911299405163e-05, - "loss": 0.2171, + "grad_norm": 7.315047122076067, + "learning_rate": 1.9242291890604186e-05, + "loss": 0.2508, "step": 5760 }, { "epoch": 0.68, - "grad_norm": 15.65983957649386, - "learning_rate": 1.8363799128617697e-05, - "loss": 0.2076, + "grad_norm": 3.540549397482083, + "learning_rate": 1.923488703985696e-05, + "loss": 0.2592, "step": 5780 }, { "epoch": 0.68, - "grad_norm": 5.987492877103834, - "learning_rate": 1.8349631303465426e-05, - "loss": 0.2199, + "grad_norm": 3.475018859373003, + "learning_rate": 1.9227447621242458e-05, + "loss": 0.2422, "step": 5800 }, { "epoch": 0.68, - "grad_norm": 6.583898334965102, - "learning_rate": 1.8335407918235116e-05, - "loss": 0.1861, + "grad_norm": 3.8676644079356355, + "learning_rate": 1.921997366261078e-05, + "loss": 0.2603, "step": 5820 }, { "epoch": 0.68, - "grad_norm": 5.51874555542228, - "learning_rate": 1.832112906758329e-05, - "loss": 0.2392, + "grad_norm": 3.5885966628483, + "learning_rate": 1.9212465191941346e-05, + "loss": 0.2542, "step": 5840 }, { "epoch": 0.69, - "grad_norm": 3.579186427839362, - "learning_rate": 1.83067948465356e-05, - "loss": 0.2204, + "grad_norm": 3.6495890137770886, + "learning_rate": 1.920492223734279e-05, + "loss": 0.2374, "step": 5860 }, { "epoch": 0.69, - "grad_norm": 3.519966665890388, - "learning_rate": 1.8292405350486165e-05, - "loss": 0.2151, + "grad_norm": 4.323488251305847, + "learning_rate": 1.9197344827052813e-05, + "loss": 0.2687, "step": 5880 }, { "epoch": 0.69, - "grad_norm": 4.244150062547031, - "learning_rate": 1.8277960675196983e-05, - "loss": 0.2297, + "grad_norm": 5.051497760408599, + "learning_rate": 1.9189732989438123e-05, + "loss": 0.2678, "step": 5900 }, { "epoch": 0.69, - "grad_norm": 5.344839063171692, - "learning_rate": 1.826346091679725e-05, - "loss": 0.2245, + "grad_norm": 3.5559342765441495, + "learning_rate": 1.9182086752994295e-05, + "loss": 0.2726, "step": 5920 }, { "epoch": 0.7, - "grad_norm": 5.0436570623407775, - "learning_rate": 1.8248906171782755e-05, - "loss": 0.2378, + "grad_norm": 3.029184640641301, + "learning_rate": 1.9174406146345694e-05, + "loss": 0.246, "step": 5940 }, { "epoch": 0.7, - "grad_norm": 7.185397163940519, - "learning_rate": 1.8234296537015207e-05, - "loss": 0.2294, + "grad_norm": 3.4792605785250994, + "learning_rate": 1.9166691198245343e-05, + "loss": 0.2843, "step": 5960 }, { "epoch": 0.7, - "grad_norm": 4.591681189724662, - "learning_rate": 1.8219632109721624e-05, - "loss": 0.2256, + "grad_norm": 3.0247490770922, + "learning_rate": 1.915894193757483e-05, + "loss": 0.2528, "step": 5980 }, { "epoch": 0.7, - "grad_norm": 5.248378321548811, - "learning_rate": 1.8204912987493658e-05, - "loss": 0.2351, + "grad_norm": 3.7727567007247855, + "learning_rate": 1.915115839334419e-05, + "loss": 0.2442, "step": 6000 }, { "epoch": 0.7, - "grad_norm": 3.251787867657218, - "learning_rate": 1.819013926828695e-05, - "loss": 0.1976, + "grad_norm": 2.31881357193246, + "learning_rate": 1.9143340594691806e-05, + "loss": 0.2561, "step": 6020 }, { "epoch": 0.71, - "grad_norm": 3.355709045119318, - "learning_rate": 1.817531105042049e-05, - "loss": 0.1958, + "grad_norm": 3.138315746753083, + "learning_rate": 1.913548857088429e-05, + "loss": 0.2636, "step": 6040 }, { "epoch": 0.71, - "grad_norm": 2.8975140933466705, - "learning_rate": 1.8160428432575958e-05, - "loss": 0.1763, + "grad_norm": 8.010145906221986, + "learning_rate": 1.9127602351316393e-05, + "loss": 0.2876, "step": 6060 }, { "epoch": 0.71, - "grad_norm": 7.07650583370445, - "learning_rate": 1.814549151379706e-05, - "loss": 0.1839, + "grad_norm": 3.867596475572881, + "learning_rate": 1.9119681965510858e-05, + "loss": 0.2659, "step": 6080 }, { "epoch": 0.71, - "grad_norm": 2.3292924654795906, - "learning_rate": 1.8130500393488883e-05, - "loss": 0.1737, + "grad_norm": 3.3587166074477777, + "learning_rate": 1.9111727443118357e-05, + "loss": 0.2634, "step": 6100 }, { "epoch": 0.72, - "grad_norm": 3.6110322128545116, - "learning_rate": 1.811545517141722e-05, - "loss": 0.2161, + "grad_norm": 2.89895288044486, + "learning_rate": 1.9103738813917336e-05, + "loss": 0.2533, "step": 6120 }, { "epoch": 0.72, - "grad_norm": 3.2080012723487368, - "learning_rate": 1.8100355947707896e-05, - "loss": 0.1741, + "grad_norm": 4.037752097559083, + "learning_rate": 1.909571610781393e-05, + "loss": 0.2485, "step": 6140 }, { "epoch": 0.72, - "grad_norm": 3.4674409337915906, - "learning_rate": 1.8085202822846145e-05, - "loss": 0.1983, + "grad_norm": 3.763608653829743, + "learning_rate": 1.908765935484185e-05, + "loss": 0.259, "step": 6160 }, { "epoch": 0.72, - "grad_norm": 3.24481638760689, - "learning_rate": 1.8069995897675893e-05, - "loss": 0.1974, + "grad_norm": 3.161229351032073, + "learning_rate": 1.9079568585162248e-05, + "loss": 0.2508, "step": 6180 }, { "epoch": 0.73, - "grad_norm": 2.1527007372890807, - "learning_rate": 1.8054735273399105e-05, - "loss": 0.1787, + "grad_norm": 2.725004696623228, + "learning_rate": 1.9071443829063644e-05, + "loss": 0.2524, "step": 6200 }, { "epoch": 0.73, - "grad_norm": 3.8433408441880994, - "learning_rate": 1.8039421051575134e-05, - "loss": 0.2045, + "grad_norm": 3.7116897623627985, + "learning_rate": 1.906328511696177e-05, + "loss": 0.2598, "step": 6220 }, { "epoch": 0.73, - "grad_norm": 3.1482360623951675, - "learning_rate": 1.802405333412e-05, - "loss": 0.1823, + "grad_norm": 2.9004110110815544, + "learning_rate": 1.9055092479399474e-05, + "loss": 0.2553, "step": 6240 }, { "epoch": 0.73, - "grad_norm": 3.638738970462984, - "learning_rate": 1.800863222330575e-05, - "loss": 0.163, + "grad_norm": 4.311030925472568, + "learning_rate": 1.9046865947046627e-05, + "loss": 0.2678, "step": 6260 }, { "epoch": 0.73, - "grad_norm": 2.6939386819489473, - "learning_rate": 1.7993157821759756e-05, - "loss": 0.1506, + "grad_norm": 3.5377476420175977, + "learning_rate": 1.903860555069997e-05, + "loss": 0.2379, "step": 6280 }, { "epoch": 0.74, - "grad_norm": 2.460255555625638, - "learning_rate": 1.7977630232464046e-05, - "loss": 0.2013, + "grad_norm": 3.58934533910836, + "learning_rate": 1.9030311321283022e-05, + "loss": 0.2608, "step": 6300 }, { "epoch": 0.74, - "grad_norm": 3.419539678056426, - "learning_rate": 1.7962049558754607e-05, - "loss": 0.1746, + "grad_norm": 3.4612535457241624, + "learning_rate": 1.9021983289845956e-05, + "loss": 0.246, "step": 6320 }, { "epoch": 0.74, - "grad_norm": 4.6827563888655295, - "learning_rate": 1.794641590432071e-05, - "loss": 0.1907, + "grad_norm": 3.9763337015158258, + "learning_rate": 1.9013621487565498e-05, + "loss": 0.238, "step": 6340 }, { "epoch": 0.74, - "grad_norm": 3.156495792902755, - "learning_rate": 1.7930729373204197e-05, - "loss": 0.1932, + "grad_norm": 3.05488683337997, + "learning_rate": 1.9005225945744782e-05, + "loss": 0.2472, "step": 6360 }, { "epoch": 0.75, - "grad_norm": 1.8329842047169478, - "learning_rate": 1.7914990069798825e-05, - "loss": 0.1839, + "grad_norm": 3.2350295137123313, + "learning_rate": 1.8996796695813256e-05, + "loss": 0.261, "step": 6380 }, { "epoch": 0.75, - "grad_norm": 3.1302938253922066, - "learning_rate": 1.7899198098849532e-05, - "loss": 0.1958, + "grad_norm": 3.2500906711548123, + "learning_rate": 1.8988333769326564e-05, + "loss": 0.2641, "step": 6400 }, { "epoch": 0.75, - "grad_norm": 3.389701612050505, - "learning_rate": 1.788335356545176e-05, - "loss": 0.1915, + "grad_norm": 2.815087548041305, + "learning_rate": 1.8979837197966415e-05, + "loss": 0.2583, "step": 6420 }, { "epoch": 0.75, - "grad_norm": 2.4493770274440805, - "learning_rate": 1.786745657505077e-05, - "loss": 0.1864, + "grad_norm": 3.7397965536014257, + "learning_rate": 1.8971307013540464e-05, + "loss": 0.2485, "step": 6440 }, { "epoch": 0.76, - "grad_norm": 3.7081248335498316, - "learning_rate": 1.7851507233440907e-05, - "loss": 0.1737, + "grad_norm": 2.7592373806083423, + "learning_rate": 1.8962743247982205e-05, + "loss": 0.2641, "step": 6460 }, { "epoch": 0.76, - "grad_norm": 2.9880342269224034, - "learning_rate": 1.7835505646764915e-05, - "loss": 0.1737, + "grad_norm": 2.818080452105616, + "learning_rate": 1.895414593335085e-05, + "loss": 0.2554, "step": 6480 }, { "epoch": 0.76, - "grad_norm": 3.243289769328672, - "learning_rate": 1.781945192151324e-05, - "loss": 0.1719, + "grad_norm": 3.627385505536174, + "learning_rate": 1.8945515101831203e-05, + "loss": 0.2693, "step": 6500 }, { "epoch": 0.76, - "grad_norm": 3.4374399353328338, - "learning_rate": 1.7803346164523294e-05, - "loss": 0.1758, + "grad_norm": 3.255509767944512, + "learning_rate": 1.893685078573354e-05, + "loss": 0.2459, "step": 6520 }, { "epoch": 0.77, - "grad_norm": 2.8036052347770863, - "learning_rate": 1.778718848297877e-05, - "loss": 0.1782, + "grad_norm": 4.135182845698409, + "learning_rate": 1.8928153017493488e-05, + "loss": 0.2499, "step": 6540 }, { "epoch": 0.77, - "grad_norm": 2.2879959069195395, - "learning_rate": 1.777097898440892e-05, - "loss": 0.191, + "grad_norm": 5.892816677202276, + "learning_rate": 1.8919421829671904e-05, + "loss": 0.2786, "step": 6560 }, { "epoch": 0.77, - "grad_norm": 3.040554749166379, - "learning_rate": 1.7754717776687835e-05, - "loss": 0.1831, + "grad_norm": 3.479892875732629, + "learning_rate": 1.891065725495476e-05, + "loss": 0.265, "step": 6580 }, { "epoch": 0.77, - "grad_norm": 3.071599714764158, - "learning_rate": 1.773840496803372e-05, - "loss": 0.185, + "grad_norm": 3.8343809741781216, + "learning_rate": 1.890185932615301e-05, + "loss": 0.2527, "step": 6600 }, { "epoch": 0.77, - "grad_norm": 3.501269791372487, - "learning_rate": 1.7722040667008204e-05, - "loss": 0.1719, + "grad_norm": 3.1643033159708107, + "learning_rate": 1.8893028076202472e-05, + "loss": 0.2741, "step": 6620 }, { "epoch": 0.78, - "grad_norm": 4.16368161404422, - "learning_rate": 1.7705624982515573e-05, - "loss": 0.184, + "grad_norm": 4.477352585232187, + "learning_rate": 1.8884163538163708e-05, + "loss": 0.2525, "step": 6640 }, { "epoch": 0.78, - "grad_norm": 2.7540351566960117, - "learning_rate": 1.7689158023802095e-05, - "loss": 0.1946, + "grad_norm": 3.529760400999529, + "learning_rate": 1.8875265745221894e-05, + "loss": 0.2643, "step": 6660 }, { "epoch": 0.78, - "grad_norm": 3.2355776407579153, - "learning_rate": 1.7672639900455252e-05, - "loss": 0.1712, + "grad_norm": 2.3110189076619525, + "learning_rate": 1.8866334730686703e-05, + "loss": 0.2721, "step": 6680 }, { "epoch": 0.78, - "grad_norm": 2.6997084000974327, - "learning_rate": 1.765607072240303e-05, - "loss": 0.1784, + "grad_norm": 3.685226774238388, + "learning_rate": 1.8857370527992165e-05, + "loss": 0.2549, "step": 6700 }, { "epoch": 0.79, - "grad_norm": 2.841998053177577, - "learning_rate": 1.763945059991318e-05, - "loss": 0.1742, + "grad_norm": 3.8074891978039864, + "learning_rate": 1.8848373170696567e-05, + "loss": 0.2418, "step": 6720 }, { "epoch": 0.79, - "grad_norm": 2.7453638357487713, - "learning_rate": 1.7622779643592492e-05, - "loss": 0.1944, + "grad_norm": 7.507877790185021, + "learning_rate": 1.8839342692482302e-05, + "loss": 0.2451, "step": 6740 }, { "epoch": 0.79, - "grad_norm": 3.3176797639024938, - "learning_rate": 1.760605796438606e-05, - "loss": 0.1899, + "grad_norm": 2.7359733051983186, + "learning_rate": 1.8830279127155764e-05, + "loss": 0.2484, "step": 6760 }, { "epoch": 0.79, - "grad_norm": 2.366476424276501, - "learning_rate": 1.758928567357652e-05, - "loss": 0.1759, + "grad_norm": 2.4635290638854657, + "learning_rate": 1.88211825086472e-05, + "loss": 0.2407, "step": 6780 }, { "epoch": 0.8, - "grad_norm": 3.7220494355163054, - "learning_rate": 1.7572462882783353e-05, - "loss": 0.1837, + "grad_norm": 3.040130192327906, + "learning_rate": 1.8812052871010605e-05, + "loss": 0.2544, "step": 6800 }, { "epoch": 0.8, - "grad_norm": 3.038915792420121, - "learning_rate": 1.7555589703962098e-05, - "loss": 0.1646, + "grad_norm": 5.610778632978326, + "learning_rate": 1.8802890248423574e-05, + "loss": 0.2397, "step": 6820 }, { "epoch": 0.8, - "grad_norm": 3.4617122704323244, - "learning_rate": 1.7538666249403633e-05, - "loss": 0.1758, + "grad_norm": 4.156547822535594, + "learning_rate": 1.8793694675187197e-05, + "loss": 0.2635, "step": 6840 }, { "epoch": 0.8, - "grad_norm": 2.362854772501653, - "learning_rate": 1.752169263173342e-05, - "loss": 0.1634, + "grad_norm": 4.057314806159278, + "learning_rate": 1.8784466185725902e-05, + "loss": 0.2598, "step": 6860 }, { "epoch": 0.81, - "grad_norm": 3.00174582863279, - "learning_rate": 1.7504668963910752e-05, - "loss": 0.1701, + "grad_norm": 8.314874066232434, + "learning_rate": 1.877520481458736e-05, + "loss": 0.2706, "step": 6880 }, { "epoch": 0.81, - "grad_norm": 3.2309255588354, - "learning_rate": 1.748759535922802e-05, - "loss": 0.1995, + "grad_norm": 3.7767818313058195, + "learning_rate": 1.8765910596442316e-05, + "loss": 0.2483, "step": 6900 }, { "epoch": 0.81, - "grad_norm": 2.662821998288258, - "learning_rate": 1.7470471931309922e-05, - "loss": 0.1672, + "grad_norm": 3.53905010431742, + "learning_rate": 1.8756583566084502e-05, + "loss": 0.2683, "step": 6920 }, { "epoch": 0.81, - "grad_norm": 2.8772004041302623, - "learning_rate": 1.7453298794112747e-05, - "loss": 0.1898, + "grad_norm": 3.2053331160944625, + "learning_rate": 1.8747223758430467e-05, + "loss": 0.2567, "step": 6940 }, { "epoch": 0.81, - "grad_norm": 2.9237695763280365, - "learning_rate": 1.7436076061923592e-05, - "loss": 0.1762, + "grad_norm": 3.655977418308093, + "learning_rate": 1.8737831208519478e-05, + "loss": 0.2433, "step": 6960 }, { "epoch": 0.82, - "grad_norm": 2.5504855909062036, - "learning_rate": 1.741880384935961e-05, - "loss": 0.1992, + "grad_norm": 3.334646935711793, + "learning_rate": 1.8728405951513376e-05, + "loss": 0.2442, "step": 6980 }, { "epoch": 0.82, - "grad_norm": 7.919165394281912, - "learning_rate": 1.7401482271367242e-05, - "loss": 0.1692, + "grad_norm": 3.3678493336120505, + "learning_rate": 1.8718948022696427e-05, + "loss": 0.2634, "step": 7000 }, { "epoch": 0.82, - "grad_norm": 3.05371171825022, - "learning_rate": 1.7384111443221457e-05, - "loss": 0.1908, + "grad_norm": 2.9941614875638956, + "learning_rate": 1.8709457457475224e-05, + "loss": 0.2719, "step": 7020 }, { "epoch": 0.82, - "grad_norm": 2.621096842688656, - "learning_rate": 1.7366691480524992e-05, - "loss": 0.1685, + "grad_norm": 3.3212764283242495, + "learning_rate": 1.8699934291378534e-05, + "loss": 0.2616, "step": 7040 }, { "epoch": 0.83, - "grad_norm": 3.1230713042352862, - "learning_rate": 1.734922249920756e-05, - "loss": 0.1792, + "grad_norm": 3.7449113015848963, + "learning_rate": 1.8690378560057165e-05, + "loss": 0.251, "step": 7060 }, { "epoch": 0.83, - "grad_norm": 3.104105869867001, - "learning_rate": 1.73317046155251e-05, - "loss": 0.1729, + "grad_norm": 2.588981565454773, + "learning_rate": 1.868079029928383e-05, + "loss": 0.2345, "step": 7080 }, { "epoch": 0.83, - "grad_norm": 3.1591947811656436, - "learning_rate": 1.7314137946059e-05, - "loss": 0.1718, + "grad_norm": 4.188807496871825, + "learning_rate": 1.8671169544953033e-05, + "loss": 0.2537, "step": 7100 }, { "epoch": 0.83, - "grad_norm": 3.519623783770741, - "learning_rate": 1.7296522607715304e-05, - "loss": 0.1764, + "grad_norm": 4.66518430781191, + "learning_rate": 1.866151633308091e-05, + "loss": 0.2397, "step": 7120 }, { "epoch": 0.84, - "grad_norm": 2.915870412492841, - "learning_rate": 1.7278858717723962e-05, - "loss": 0.1558, + "grad_norm": 4.197676860816234, + "learning_rate": 1.8651830699805108e-05, + "loss": 0.2446, "step": 7140 }, { "epoch": 0.84, - "grad_norm": 2.3279234523707295, - "learning_rate": 1.726114639363803e-05, - "loss": 0.1913, + "grad_norm": 3.6434672643458543, + "learning_rate": 1.8642112681384645e-05, + "loss": 0.2402, "step": 7160 }, { "epoch": 0.84, - "grad_norm": 2.6716434311486, - "learning_rate": 1.724338575333288e-05, - "loss": 0.1745, + "grad_norm": 3.950387723218738, + "learning_rate": 1.8632362314199778e-05, + "loss": 0.2536, "step": 7180 }, { "epoch": 0.84, - "grad_norm": 2.878117736740971, - "learning_rate": 1.7225576915005445e-05, - "loss": 0.1727, + "grad_norm": 9.658211616844094, + "learning_rate": 1.8622579634751864e-05, + "loss": 0.2541, "step": 7200 }, { "epoch": 0.84, - "grad_norm": 2.7046376665043277, - "learning_rate": 1.720771999717341e-05, - "loss": 0.1772, + "grad_norm": 3.799337836345952, + "learning_rate": 1.8612764679663222e-05, + "loss": 0.2586, "step": 7220 }, { "epoch": 0.85, - "grad_norm": 3.194537005719014, - "learning_rate": 1.7189815118674426e-05, - "loss": 0.1796, + "grad_norm": 3.737906856782671, + "learning_rate": 1.8602917485677e-05, + "loss": 0.2539, "step": 7240 }, { "epoch": 0.85, - "grad_norm": 3.463727321763896, - "learning_rate": 1.7171862398665327e-05, - "loss": 0.1919, + "grad_norm": 3.2570157254536016, + "learning_rate": 1.859303808965703e-05, + "loss": 0.2533, "step": 7260 }, { "epoch": 0.85, - "grad_norm": 3.13725471905633, - "learning_rate": 1.7153861956621326e-05, - "loss": 0.1811, + "grad_norm": 2.8736525985585226, + "learning_rate": 1.85831265285877e-05, + "loss": 0.2532, "step": 7280 }, { "epoch": 0.85, - "grad_norm": 4.544353611874395, - "learning_rate": 1.7135813912335225e-05, - "loss": 0.1692, + "grad_norm": 3.5447389329516503, + "learning_rate": 1.857318283957381e-05, + "loss": 0.2387, "step": 7300 }, { "epoch": 0.86, - "grad_norm": 2.1203551297270153, - "learning_rate": 1.7117718385916627e-05, - "loss": 0.1794, + "grad_norm": 2.7468496831393403, + "learning_rate": 1.8563207059840436e-05, + "loss": 0.2455, "step": 7320 }, { "epoch": 0.86, - "grad_norm": 2.8114219294750917, - "learning_rate": 1.709957549779112e-05, - "loss": 0.1568, + "grad_norm": 3.483830293921577, + "learning_rate": 1.8553199226732784e-05, + "loss": 0.2456, "step": 7340 }, { "epoch": 0.86, - "grad_norm": 2.772982105960882, - "learning_rate": 1.7081385368699487e-05, - "loss": 0.1731, + "grad_norm": 3.3132796899404635, + "learning_rate": 1.8543159377716054e-05, + "loss": 0.2564, "step": 7360 }, { "epoch": 0.86, - "grad_norm": 2.169539808531856, - "learning_rate": 1.7063148119696903e-05, - "loss": 0.196, + "grad_norm": 3.3795508326649575, + "learning_rate": 1.8533087550375302e-05, + "loss": 0.2527, "step": 7380 }, { "epoch": 0.87, - "grad_norm": 3.269632171568159, - "learning_rate": 1.704486387215212e-05, - "loss": 0.1733, + "grad_norm": 4.001941448174156, + "learning_rate": 1.8522983782415306e-05, + "loss": 0.2433, "step": 7400 }, { "epoch": 0.87, - "grad_norm": 2.304673017036061, - "learning_rate": 1.7026532747746664e-05, - "loss": 0.1752, + "grad_norm": 3.4205382361927117, + "learning_rate": 1.85128481116604e-05, + "loss": 0.2538, "step": 7420 }, { "epoch": 0.87, - "grad_norm": 3.871844945364525, - "learning_rate": 1.700815486847403e-05, - "loss": 0.1614, + "grad_norm": 3.896330422632185, + "learning_rate": 1.8502680576054365e-05, + "loss": 0.2478, "step": 7440 }, { "epoch": 0.87, - "grad_norm": 3.4589296462620522, - "learning_rate": 1.698973035663887e-05, - "loss": 0.1795, + "grad_norm": 5.020228094868969, + "learning_rate": 1.849248121366026e-05, + "loss": 0.2427, "step": 7460 }, { "epoch": 0.88, - "grad_norm": 2.449963810711825, - "learning_rate": 1.6971259334856162e-05, - "loss": 0.1607, + "grad_norm": 3.574276782956127, + "learning_rate": 1.8482250062660293e-05, + "loss": 0.2389, "step": 7480 }, { "epoch": 0.88, - "grad_norm": 3.769043316663817, - "learning_rate": 1.695274192605042e-05, - "loss": 0.1649, + "grad_norm": 3.5309579103911184, + "learning_rate": 1.847198716135568e-05, + "loss": 0.2567, "step": 7500 }, { "epoch": 0.88, - "grad_norm": 3.1777138173025103, - "learning_rate": 1.6934178253454868e-05, - "loss": 0.1579, + "grad_norm": 3.1760414458004, + "learning_rate": 1.846169254816649e-05, + "loss": 0.2291, "step": 7520 }, { "epoch": 0.88, - "grad_norm": 2.55585511382277, - "learning_rate": 1.691556844061059e-05, - "loss": 0.1483, + "grad_norm": 4.214334621561301, + "learning_rate": 1.8451366261631514e-05, + "loss": 0.2624, "step": 7540 }, { "epoch": 0.88, - "grad_norm": 2.9610046540146118, - "learning_rate": 1.6896912611365762e-05, - "loss": 0.1671, + "grad_norm": 2.977221959069712, + "learning_rate": 1.8441008340408115e-05, + "loss": 0.2425, "step": 7560 }, { "epoch": 0.89, - "grad_norm": 2.916410998083297, - "learning_rate": 1.687821088987478e-05, - "loss": 0.1715, + "grad_norm": 3.78383188925913, + "learning_rate": 1.8430618823272076e-05, + "loss": 0.2437, "step": 7580 }, { "epoch": 0.89, - "grad_norm": 2.5312840023818173, - "learning_rate": 1.6859463400597465e-05, - "loss": 0.1715, + "grad_norm": 3.2432934448936224, + "learning_rate": 1.842019774911747e-05, + "loss": 0.2628, "step": 7600 }, { "epoch": 0.89, - "grad_norm": 3.5116131080293282, - "learning_rate": 1.684067026829821e-05, - "loss": 0.1579, + "grad_norm": 3.05382414419798, + "learning_rate": 1.8409745156956503e-05, + "loss": 0.2332, "step": 7620 }, { "epoch": 0.89, - "grad_norm": 2.5798326789951567, - "learning_rate": 1.6821831618045166e-05, - "loss": 0.1713, + "grad_norm": 3.098744070302099, + "learning_rate": 1.839926108591937e-05, + "loss": 0.2359, "step": 7640 }, { "epoch": 0.9, - "grad_norm": 2.7929950419435565, - "learning_rate": 1.6802947575209415e-05, - "loss": 0.173, + "grad_norm": 3.427098715766002, + "learning_rate": 1.8388745575254117e-05, + "loss": 0.2429, "step": 7660 }, { "epoch": 0.9, - "grad_norm": 3.2236373979916646, - "learning_rate": 1.6784018265464113e-05, - "loss": 0.1808, + "grad_norm": 5.630711008711337, + "learning_rate": 1.8378198664326474e-05, + "loss": 0.2416, "step": 7680 }, { "epoch": 0.9, - "grad_norm": 3.1749239394321243, - "learning_rate": 1.6765043814783676e-05, - "loss": 0.1759, + "grad_norm": 3.1144723181358622, + "learning_rate": 1.836762039261974e-05, + "loss": 0.2406, "step": 7700 }, { "epoch": 0.9, - "grad_norm": 2.032005169681671, - "learning_rate": 1.674602434944293e-05, - "loss": 0.1814, + "grad_norm": 4.070301722184627, + "learning_rate": 1.8357010799734588e-05, + "loss": 0.234, "step": 7720 }, { "epoch": 0.91, - "grad_norm": 3.2976168841859366, - "learning_rate": 1.6726959996016273e-05, - "loss": 0.1487, + "grad_norm": 3.255428182680105, + "learning_rate": 1.834636992538898e-05, + "loss": 0.2365, "step": 7740 }, { "epoch": 0.91, - "grad_norm": 2.3455040026914995, - "learning_rate": 1.6707850881376823e-05, - "loss": 0.1686, + "grad_norm": 3.4823775104086576, + "learning_rate": 1.8335697809417952e-05, + "loss": 0.2376, "step": 7760 }, { "epoch": 0.91, - "grad_norm": 3.271968976468327, - "learning_rate": 1.668869713269561e-05, - "loss": 0.2027, + "grad_norm": 5.528269163918348, + "learning_rate": 1.8324994491773507e-05, + "loss": 0.2679, "step": 7780 }, { "epoch": 0.91, - "grad_norm": 3.220964133175747, - "learning_rate": 1.666949887744067e-05, - "loss": 0.1694, + "grad_norm": 2.7221100312640467, + "learning_rate": 1.8314260012524462e-05, + "loss": 0.2544, "step": 7800 }, { "epoch": 0.92, - "grad_norm": 3.5721944777889676, - "learning_rate": 1.6650256243376257e-05, - "loss": 0.1908, + "grad_norm": 3.224515621271091, + "learning_rate": 1.8303494411856278e-05, + "loss": 0.2586, "step": 7820 }, { "epoch": 0.92, - "grad_norm": 2.3813766405767627, - "learning_rate": 1.6630969358561948e-05, - "loss": 0.1824, + "grad_norm": 2.798714485709818, + "learning_rate": 1.829269773007092e-05, + "loss": 0.2277, "step": 7840 }, { "epoch": 0.92, - "grad_norm": 3.129649556197066, - "learning_rate": 1.661163835135182e-05, - "loss": 0.1787, + "grad_norm": 3.890849401904209, + "learning_rate": 1.8281870007586735e-05, + "loss": 0.2565, "step": 7860 }, { "epoch": 0.92, - "grad_norm": 3.415549227335631, - "learning_rate": 1.659226335039358e-05, - "loss": 0.1825, + "grad_norm": 3.701194549732654, + "learning_rate": 1.8271011284938233e-05, + "loss": 0.2307, "step": 7880 }, { "epoch": 0.92, - "grad_norm": 2.3764217538181778, - "learning_rate": 1.6572844484627707e-05, - "loss": 0.1462, + "grad_norm": 3.9057568048025875, + "learning_rate": 1.8260121602776003e-05, + "loss": 0.2464, "step": 7900 }, { "epoch": 0.93, - "grad_norm": 2.9671229672357287, - "learning_rate": 1.6553381883286612e-05, - "loss": 0.1914, + "grad_norm": 3.397196786583742, + "learning_rate": 1.8249201001866533e-05, + "loss": 0.25, "step": 7920 }, { "epoch": 0.93, - "grad_norm": 3.060295245568844, - "learning_rate": 1.6533875675893755e-05, - "loss": 0.181, + "grad_norm": 3.6136942859367496, + "learning_rate": 1.8238249523092046e-05, + "loss": 0.2398, "step": 7940 }, { "epoch": 0.93, - "grad_norm": 3.7842972511980086, - "learning_rate": 1.6514325992262798e-05, - "loss": 0.1507, + "grad_norm": 3.6009293204428765, + "learning_rate": 1.8227267207450363e-05, + "loss": 0.2356, "step": 7960 }, { "epoch": 0.93, - "grad_norm": 2.8230206703837326, - "learning_rate": 1.6494732962496735e-05, - "loss": 0.1736, + "grad_norm": 3.1462270393488585, + "learning_rate": 1.8216254096054747e-05, + "loss": 0.2292, "step": 7980 }, { "epoch": 0.94, - "grad_norm": 3.453554704952226, - "learning_rate": 1.6475096716987044e-05, - "loss": 0.1765, + "grad_norm": 3.453766085807792, + "learning_rate": 1.8205210230133756e-05, + "loss": 0.2532, "step": 8000 }, { "epoch": 0.94, - "grad_norm": 2.9060450655966794, - "learning_rate": 1.6455417386412776e-05, - "loss": 0.1755, + "grad_norm": 4.718419953986993, + "learning_rate": 1.8194135651031063e-05, + "loss": 0.2379, "step": 8020 }, { "epoch": 0.94, - "grad_norm": 2.54357724739597, - "learning_rate": 1.6435695101739728e-05, - "loss": 0.178, + "grad_norm": 4.074122309800583, + "learning_rate": 1.8183030400205335e-05, + "loss": 0.247, "step": 8040 }, { "epoch": 0.94, - "grad_norm": 2.2841639263364417, - "learning_rate": 1.641592999421956e-05, - "loss": 0.1671, + "grad_norm": 3.597629476607082, + "learning_rate": 1.817189451923005e-05, + "loss": 0.2563, "step": 8060 }, { "epoch": 0.95, - "grad_norm": 2.922479250844846, - "learning_rate": 1.639612219538891e-05, - "loss": 0.1722, + "grad_norm": 3.2550443969136436, + "learning_rate": 1.8160728049793357e-05, + "loss": 0.2425, "step": 8080 }, { "epoch": 0.95, - "grad_norm": 3.364819041086794, - "learning_rate": 1.6376271837068535e-05, - "loss": 0.1743, + "grad_norm": 2.6027515890006505, + "learning_rate": 1.8149531033697913e-05, + "loss": 0.2329, "step": 8100 }, { "epoch": 0.95, - "grad_norm": 3.2343077721726976, - "learning_rate": 1.6356379051362408e-05, - "loss": 0.1754, + "grad_norm": 3.705349797095471, + "learning_rate": 1.8138303512860737e-05, + "loss": 0.2521, "step": 8120 }, { "epoch": 0.95, - "grad_norm": 3.988702555642947, - "learning_rate": 1.6336443970656877e-05, - "loss": 0.1719, + "grad_norm": 3.510055266811937, + "learning_rate": 1.812704552931303e-05, + "loss": 0.2413, "step": 8140 }, { "epoch": 0.95, - "grad_norm": 3.586344112040257, - "learning_rate": 1.6316466727619746e-05, - "loss": 0.1727, + "grad_norm": 3.2863076604207997, + "learning_rate": 1.8115757125200054e-05, + "loss": 0.2471, "step": 8160 }, { "epoch": 0.96, - "grad_norm": 3.49168839611008, - "learning_rate": 1.629644745519942e-05, - "loss": 0.1812, + "grad_norm": 3.174018473859266, + "learning_rate": 1.810443834278094e-05, + "loss": 0.2429, "step": 8180 }, { "epoch": 0.96, - "grad_norm": 3.103292218300282, - "learning_rate": 1.6276386286624007e-05, - "loss": 0.1703, + "grad_norm": 4.78684161965386, + "learning_rate": 1.8093089224428533e-05, + "loss": 0.21, "step": 8200 }, { "epoch": 0.96, - "grad_norm": 3.5623729750088957, - "learning_rate": 1.6256283355400428e-05, - "loss": 0.1873, + "grad_norm": 3.778736256263061, + "learning_rate": 1.8081709812629265e-05, + "loss": 0.2438, "step": 8220 }, { "epoch": 0.96, - "grad_norm": 2.295795628214393, - "learning_rate": 1.6236138795313535e-05, - "loss": 0.1807, + "grad_norm": 2.011964890319697, + "learning_rate": 1.807030014998296e-05, + "loss": 0.2235, "step": 8240 }, { "epoch": 0.97, - "grad_norm": 3.365176421212747, - "learning_rate": 1.6215952740425232e-05, - "loss": 0.1433, + "grad_norm": 2.469231474076661, + "learning_rate": 1.8058860279202698e-05, + "loss": 0.2361, "step": 8260 }, { "epoch": 0.97, - "grad_norm": 3.3396715621732143, - "learning_rate": 1.6195725325073552e-05, - "loss": 0.1686, + "grad_norm": 4.194473608952758, + "learning_rate": 1.804739024311463e-05, + "loss": 0.2485, "step": 8280 }, { "epoch": 0.97, - "grad_norm": 4.290597554096145, - "learning_rate": 1.6175456683871797e-05, - "loss": 0.1687, + "grad_norm": 3.648802255045255, + "learning_rate": 1.8035890084657854e-05, + "loss": 0.2448, "step": 8300 }, { "epoch": 0.97, - "grad_norm": 2.5200221330185415, - "learning_rate": 1.6155146951707615e-05, - "loss": 0.1562, + "grad_norm": 2.8256525011924443, + "learning_rate": 1.8024359846884213e-05, + "loss": 0.2688, "step": 8320 }, { "epoch": 0.98, - "grad_norm": 4.318015952324211, - "learning_rate": 1.6134796263742125e-05, - "loss": 0.1731, + "grad_norm": 4.409589867929299, + "learning_rate": 1.8012799572958173e-05, + "loss": 0.2405, "step": 8340 }, { "epoch": 0.98, - "grad_norm": 2.503160957906815, - "learning_rate": 1.6114404755409e-05, - "loss": 0.1592, + "grad_norm": 5.024619525530542, + "learning_rate": 1.8001209306156634e-05, + "loss": 0.2284, "step": 8360 }, { "epoch": 0.98, - "grad_norm": 2.9078381208780253, - "learning_rate": 1.6093972562413572e-05, - "loss": 0.1664, + "grad_norm": 3.5087106485202026, + "learning_rate": 1.7989589089868774e-05, + "loss": 0.2298, "step": 8380 }, { "epoch": 0.98, - "grad_norm": 2.480511330066373, - "learning_rate": 1.607349982073193e-05, - "loss": 0.181, + "grad_norm": 4.40165070000877, + "learning_rate": 1.7977938967595897e-05, + "loss": 0.2507, "step": 8400 }, { "epoch": 0.99, - "grad_norm": 2.548494912067414, - "learning_rate": 1.6052986666610015e-05, - "loss": 0.1612, + "grad_norm": 3.9660729217673705, + "learning_rate": 1.7966258982951255e-05, + "loss": 0.2459, "step": 8420 }, { "epoch": 0.99, - "grad_norm": 2.9336394244218176, - "learning_rate": 1.6032433236562715e-05, - "loss": 0.1597, + "grad_norm": 4.665706172372416, + "learning_rate": 1.7954549179659903e-05, + "loss": 0.2458, "step": 8440 }, { "epoch": 0.99, - "grad_norm": 3.7724275678856776, - "learning_rate": 1.6011839667372948e-05, - "loss": 0.1782, + "grad_norm": 3.2908069759526333, + "learning_rate": 1.794280960155851e-05, + "loss": 0.2251, "step": 8460 }, { "epoch": 0.99, - "grad_norm": 4.105531004225903, - "learning_rate": 1.599120609609076e-05, - "loss": 0.1746, + "grad_norm": 6.7504752662905725, + "learning_rate": 1.7931040292595227e-05, + "loss": 0.254, "step": 8480 }, { "epoch": 0.99, - "grad_norm": 2.6233575314573505, - "learning_rate": 1.5970532660032413e-05, - "loss": 0.1654, + "grad_norm": 2.2047073417435787, + "learning_rate": 1.791924129682949e-05, + "loss": 0.2456, "step": 8500 }, { "epoch": 1.0, - "grad_norm": 2.71752665389956, - "learning_rate": 1.5949819496779467e-05, - "loss": 0.1866, + "grad_norm": 3.9048911210632236, + "learning_rate": 1.7907412658431877e-05, + "loss": 0.2307, "step": 8520 }, { "epoch": 1.0, - "grad_norm": 3.052677830070527, - "learning_rate": 1.592906674417787e-05, - "loss": 0.1705, + "grad_norm": 3.33290909610624, + "learning_rate": 1.7895554421683938e-05, + "loss": 0.2514, "step": 8540 } ], "logging_steps": 20, - "max_steps": 25638, + "max_steps": 34184, "num_input_tokens_seen": 0, - "num_train_epochs": 3, + "num_train_epochs": 4, "save_steps": 500, "total_flos": 0.0, "train_batch_size": 1,