Full run push

trainer_state.json

@@ -11,1412 +11,1412 @@
   "log_history": [
     {
       "epoch": 0.005,
-      "grad_norm": 1.0231467485427856,
+      "grad_norm": 0.9899646043777466,
       "learning_rate": 3.326666666666667e-06,
-      "loss": 0.7736,
+      "loss": 0.7756,
       "step": 500
     },
     {
       "epoch": 0.01,
-      "grad_norm": 1.4338911771774292,
+      "grad_norm": 1.3953757286071777,
       "learning_rate": 6.660000000000001e-06,
-      "loss": 0.5028,
+      "loss": 0.5041,
       "step": 1000
     },
     {
       "epoch": 0.015,
-      "grad_norm": 2.6965255737304688,
+      "grad_norm": 2.672257900238037,
       "learning_rate": 9.993333333333333e-06,
-      "loss": 0.4404,
+      "loss": 0.4406,
       "step": 1500
     },
     {
       "epoch": 0.02,
-      "grad_norm": 1.0795048475265503,
+      "grad_norm": 1.0422719717025757,
       "learning_rate": 1.3326666666666667e-05,
-      "loss": 0.3904,
+      "loss": 0.3907,
       "step": 2000
     },
     {
       "epoch": 0.025,
-      "grad_norm": 2.8491880893707275,
+      "grad_norm": 2.749046564102173,
       "learning_rate": 1.666e-05,
-      "loss": 0.3792,
+      "loss": 0.3794,
       "step": 2500
     },
     {
       "epoch": 0.03,
-      "grad_norm": 1.920973539352417,
+      "grad_norm": 1.8957346677780151,
       "learning_rate": 1.9993333333333337e-05,
-      "loss": 0.3676,
+      "loss": 0.3677,
       "step": 3000
     },
     {
       "epoch": 0.035,
-      "grad_norm": 3.041874408721924,
+      "grad_norm": 2.9847750663757324,
       "learning_rate": 1.9998694075951004e-05,
-      "loss": 0.3706,
+      "loss": 0.3705,
       "step": 3500
     },
     {
       "epoch": 0.04,
-      "grad_norm": 3.2993004322052,
+      "grad_norm": 3.22480845451355,
       "learning_rate": 1.9994766172915367e-05,
-      "loss": 0.3403,
+      "loss": 0.3402,
       "step": 4000
     },
     {
       "epoch": 0.045,
-      "grad_norm": 1.6200640201568604,
+      "grad_norm": 1.5718226432800293,
       "learning_rate": 1.9988217315672814e-05,
-      "loss": 0.3421,
+      "loss": 0.3419,
       "step": 4500
     },
     {
       "epoch": 0.05,
-      "grad_norm": 2.1550967693328857,
+      "grad_norm": 2.2217137813568115,
       "learning_rate": 1.997904922154766e-05,
-      "loss": 0.3255,
+      "loss": 0.325,
       "step": 5000
     },
     {
       "epoch": 0.055,
-      "grad_norm": 3.408877372741699,
+      "grad_norm": 3.4321208000183105,
       "learning_rate": 1.99672642947137e-05,
-      "loss": 0.3298,
+      "loss": 0.3295,
       "step": 5500
     },
     {
       "epoch": 0.06,
-      "grad_norm": 3.2565906047821045,
+      "grad_norm": 3.138777017593384,
       "learning_rate": 1.995286562556374e-05,
-      "loss": 0.33,
+      "loss": 0.3298,
       "step": 6000
     },
     {
       "epoch": 0.065,
-      "grad_norm": 4.7875447273254395,
+      "grad_norm": 4.528077602386475,
       "learning_rate": 1.993585698989922e-05,
-      "loss": 0.3327,
+      "loss": 0.3324,
       "step": 6500
     },
     {
       "epoch": 0.07,
-      "grad_norm": 2.0756988525390625,
+      "grad_norm": 1.9870072603225708,
       "learning_rate": 1.9916242847940064e-05,
-      "loss": 0.3282,
+      "loss": 0.3279,
       "step": 7000
     },
     {
       "epoch": 0.075,
-      "grad_norm": 2.247107982635498,
+      "grad_norm": 2.230520248413086,
       "learning_rate": 1.9894028343155063e-05,
-      "loss": 0.3312,
+      "loss": 0.3309,
       "step": 7500
     },
     {
       "epoch": 0.08,
-      "grad_norm": 1.810956597328186,
+      "grad_norm": 1.8087975978851318,
       "learning_rate": 1.9869219300913088e-05,
-      "loss": 0.3286,
+      "loss": 0.3281,
       "step": 8000
     },
     {
       "epoch": 0.085,
-      "grad_norm": 3.0170199871063232,
+      "grad_norm": 2.8722152709960938,
       "learning_rate": 1.9841822226955496e-05,
-      "loss": 0.3125,
+      "loss": 0.312,
       "step": 8500
     },
     {
       "epoch": 0.09,
-      "grad_norm": 2.1969668865203857,
+      "grad_norm": 1.97301185131073,
       "learning_rate": 1.9811844305690107e-05,
-      "loss": 0.3144,
+      "loss": 0.314,
       "step": 9000
     },
     {
       "epoch": 0.095,
-      "grad_norm": 2.5006604194641113,
+      "grad_norm": 2.5529353618621826,
       "learning_rate": 1.977929339830722e-05,
-      "loss": 0.316,
+      "loss": 0.3149,
       "step": 9500
     },
     {
       "epoch": 0.1,
-      "grad_norm": 2.108166217803955,
+      "grad_norm": 2.081092596054077,
       "learning_rate": 1.9744178040718144e-05,
-      "loss": 0.3078,
+      "loss": 0.3075,
       "step": 10000
     },
     {
       "epoch": 0.105,
-      "grad_norm": 3.1543705463409424,
+      "grad_norm": 3.256613254547119,
       "learning_rate": 1.9706507441316814e-05,
-      "loss": 0.3181,
+      "loss": 0.3174,
       "step": 10500
     },
     {
       "epoch": 0.11,
-      "grad_norm": 1.9502928256988525,
+      "grad_norm": 1.9820867776870728,
       "learning_rate": 1.9666291478565047e-05,
-      "loss": 0.3244,
+      "loss": 0.3239,
       "step": 11000
     },
     {
       "epoch": 0.115,
-      "grad_norm": 2.6310606002807617,
+      "grad_norm": 2.5789074897766113,
       "learning_rate": 1.962354069840208e-05,
-      "loss": 0.3124,
+      "loss": 0.3118,
       "step": 11500
     },
     {
       "epoch": 0.12,
-      "grad_norm": 3.0800294876098633,
+      "grad_norm": 3.117645740509033,
       "learning_rate": 1.9578266311479082e-05,
-      "loss": 0.314,
+      "loss": 0.3139,
       "step": 12000
     },
     {
       "epoch": 0.125,
-      "grad_norm": 2.6295111179351807,
+      "grad_norm": 2.703904151916504,
       "learning_rate": 1.9530480190219365e-05,
-      "loss": 0.319,
+      "loss": 0.3191,
       "step": 12500
     },
     {
       "epoch": 0.13,
-      "grad_norm": 3.222616195678711,
+      "grad_norm": 3.2976551055908203,
       "learning_rate": 1.948019486570503e-05,
-      "loss": 0.3047,
+      "loss": 0.3041,
       "step": 13000
     },
     {
       "epoch": 0.135,
-      "grad_norm": 3.3163318634033203,
+      "grad_norm": 3.4138541221618652,
       "learning_rate": 1.942742352439093e-05,
-      "loss": 0.3018,
+      "loss": 0.3012,
       "step": 13500
     },
     {
       "epoch": 0.14,
-      "grad_norm": 3.113029956817627,
+      "grad_norm": 3.0907506942749023,
       "learning_rate": 1.937218000464673e-05,
-      "loss": 0.3107,
+      "loss": 0.3099,
       "step": 14000
     },
     {
       "epoch": 0.145,
-      "grad_norm": 3.6275253295898438,
+      "grad_norm": 3.5500905513763428,
       "learning_rate": 1.931447879312804e-05,
-      "loss": 0.311,
+      "loss": 0.3103,
       "step": 14500
     },
     {
       "epoch": 0.15,
-      "grad_norm": 2.6240341663360596,
+      "grad_norm": 2.548569917678833,
       "learning_rate": 1.9254335020977542e-05,
-      "loss": 0.3078,
+      "loss": 0.3076,
       "step": 15000
     },
     {
       "epoch": 0.155,
-      "grad_norm": 3.1110498905181885,
+      "grad_norm": 3.012566566467285,
       "learning_rate": 1.9191764459857124e-05,
-      "loss": 0.3025,
+      "loss": 0.3022,
       "step": 15500
     },
     {
       "epoch": 0.16,
-      "grad_norm": 2.5400900840759277,
+      "grad_norm": 2.3697686195373535,
       "learning_rate": 1.9126783517812005e-05,
-      "loss": 0.2879,
+      "loss": 0.2877,
       "step": 16000
     },
     {
       "epoch": 0.165,
-      "grad_norm": 2.7605581283569336,
+      "grad_norm": 2.7005627155303955,
       "learning_rate": 1.9059409234968034e-05,
-      "loss": 0.2957,
+      "loss": 0.2952,
       "step": 16500
     },
     {
       "epoch": 0.17,
-      "grad_norm": 1.2014743089675903,
+      "grad_norm": 1.2764240503311157,
       "learning_rate": 1.8989659279063214e-05,
-      "loss": 0.2886,
+      "loss": 0.2881,
       "step": 17000
     },
     {
       "epoch": 0.175,
-      "grad_norm": 3.9782371520996094,
+      "grad_norm": 4.40521240234375,
       "learning_rate": 1.8917551940814615e-05,
-      "loss": 0.2947,
+      "loss": 0.2942,
       "step": 17500
     },
     {
       "epoch": 0.18,
-      "grad_norm": 2.198760509490967,
+      "grad_norm": 2.161640167236328,
       "learning_rate": 1.8843106129121993e-05,
-      "loss": 0.2978,
+      "loss": 0.2972,
       "step": 18000
     },
     {
       "epoch": 0.185,
-      "grad_norm": 28.132869720458984,
+      "grad_norm": 23.8270320892334,
       "learning_rate": 1.8766341366109214e-05,
-      "loss": 0.2902,
+      "loss": 0.2898,
       "step": 18500
     },
     {
       "epoch": 0.19,
-      "grad_norm": 2.148406505584717,
+      "grad_norm": 2.1488304138183594,
       "learning_rate": 1.8687277782004954e-05,
-      "loss": 0.2783,
+      "loss": 0.2778,
       "step": 19000
     },
     {
       "epoch": 0.195,
-      "grad_norm": 3.125153064727783,
+      "grad_norm": 2.925344705581665,
       "learning_rate": 1.8605936109863872e-05,
-      "loss": 0.2987,
+      "loss": 0.2982,
       "step": 19500
     },
     {
       "epoch": 0.2,
-      "grad_norm": 2.2595837116241455,
+      "grad_norm": 2.227436065673828,
       "learning_rate": 1.8522337680129728e-05,
-      "loss": 0.2903,
+      "loss": 0.2899,
       "step": 20000
     },
     {
       "epoch": 0.205,
-      "grad_norm": 2.338547945022583,
+      "grad_norm": 2.2893733978271484,
       "learning_rate": 1.843650441504186e-05,
-      "loss": 0.2926,
+      "loss": 0.2927,
       "step": 20500
     },
     {
       "epoch": 0.21,
-      "grad_norm": 1.230920672416687,
+      "grad_norm": 1.1803125143051147,
       "learning_rate": 1.8348458822886437e-05,
-      "loss": 0.2888,
+      "loss": 0.2885,
       "step": 21000
     },
     {
       "epoch": 0.215,
-      "grad_norm": 2.5802927017211914,
+      "grad_norm": 2.5023751258850098,
       "learning_rate": 1.8258223992094055e-05,
-      "loss": 0.299,
+      "loss": 0.2983,
       "step": 21500
     },
     {
       "epoch": 0.22,
-      "grad_norm": 1.0901334285736084,
+      "grad_norm": 1.0207619667053223,
       "learning_rate": 1.816582358518521e-05,
-      "loss": 0.2888,
+      "loss": 0.2885,
       "step": 22000
     },
     {
       "epoch": 0.225,
-      "grad_norm": 2.762578248977661,
+      "grad_norm": 2.7328686714172363,
       "learning_rate": 1.80712818325652e-05,
-      "loss": 0.2904,
+      "loss": 0.2902,
       "step": 22500
     },
     {
       "epoch": 0.23,
-      "grad_norm": 2.782043695449829,
+      "grad_norm": 2.740802049636841,
       "learning_rate": 1.7974623526170113e-05,
-      "loss": 0.2952,
+      "loss": 0.2953,
       "step": 23000
     },
     {
       "epoch": 0.235,
-      "grad_norm": 1.5426374673843384,
+      "grad_norm": 1.5233033895492554,
       "learning_rate": 1.78758740129656e-05,
-      "loss": 0.2847,
+      "loss": 0.2845,
       "step": 23500
     },
     {
       "epoch": 0.24,
-      "grad_norm": 4.683619499206543,
+      "grad_norm": 4.8131866455078125,
       "learning_rate": 1.777505918830004e-05,
-      "loss": 0.2854,
+      "loss": 0.2851,
       "step": 24000
     },
     {
       "epoch": 0.245,
-      "grad_norm": 2.5666069984436035,
+      "grad_norm": 2.3986711502075195,
       "learning_rate": 1.767220548911395e-05,
-      "loss": 0.27,
+      "loss": 0.2697,
       "step": 24500
     },
     {
       "epoch": 0.25,
-      "grad_norm": 2.759404420852661,
+      "grad_norm": 2.7533788681030273,
       "learning_rate": 1.756733988700737e-05,
-      "loss": 0.2744,
+      "loss": 0.274,
       "step": 25000
     },
     {
       "epoch": 0.255,
-      "grad_norm": 2.2151215076446533,
+      "grad_norm": 2.176361560821533,
       "learning_rate": 1.7460489881166986e-05,
-      "loss": 0.285,
+      "loss": 0.2848,
       "step": 25500
     },
     {
       "epoch": 0.26,
-      "grad_norm": 1.9585872888565063,
+      "grad_norm": 2.001997470855713,
       "learning_rate": 1.7351683491154984e-05,
-      "loss": 0.2964,
+      "loss": 0.2959,
       "step": 26000
     },
     {
       "epoch": 0.265,
-      "grad_norm": 2.9606637954711914,
+      "grad_norm": 2.8803274631500244,
       "learning_rate": 1.7240949249561402e-05,
-      "loss": 0.2821,
+      "loss": 0.2816,
       "step": 26500
     },
     {
       "epoch": 0.27,
-      "grad_norm": 2.198655366897583,
+      "grad_norm": 2.2319493293762207,
       "learning_rate": 1.7128316194521923e-05,
-      "loss": 0.2694,
+      "loss": 0.2692,
       "step": 27000
     },
     {
       "epoch": 0.275,
-      "grad_norm": 3.7448890209198,
+      "grad_norm": 3.515261173248291,
       "learning_rate": 1.7013813862103163e-05,
-      "loss": 0.2748,
+      "loss": 0.2744,
       "step": 27500
     },
     {
       "epoch": 0.28,
-      "grad_norm": 1.034110426902771,
+      "grad_norm": 1.112574815750122,
       "learning_rate": 1.689747227855733e-05,
-      "loss": 0.2795,
+      "loss": 0.279,
       "step": 28000
     },
     {
       "epoch": 0.285,
-      "grad_norm": 2.9149484634399414,
+      "grad_norm": 2.715858221054077,
       "learning_rate": 1.677932195244836e-05,
-      "loss": 0.2797,
+      "loss": 0.2795,
       "step": 28500
     },
     {
       "epoch": 0.29,
-      "grad_norm": 3.301887273788452,
+      "grad_norm": 3.356738805770874,
       "learning_rate": 1.6659393866651585e-05,
-      "loss": 0.27,
+      "loss": 0.2698,
       "step": 29000
     },
     {
       "epoch": 0.295,
-      "grad_norm": 3.6001880168914795,
+      "grad_norm": 3.5027854442596436,
       "learning_rate": 1.6537719470228993e-05,
-      "loss": 0.2762,
+      "loss": 0.2751,
       "step": 29500
     },
     {
       "epoch": 0.3,
-      "grad_norm": 2.743347406387329,
+      "grad_norm": 2.723360776901245,
       "learning_rate": 1.6414330670182278e-05,
-      "loss": 0.2766,
+      "loss": 0.2761,
       "step": 30000
     },
     {
       "epoch": 0.305,
-      "grad_norm": 2.654660224914551,
+      "grad_norm": 2.674299955368042,
       "learning_rate": 1.6289259823085765e-05,
       "loss": 0.2834,
       "step": 30500
     },
     {
       "epoch": 0.31,
-      "grad_norm": 1.923891544342041,
+      "grad_norm": 1.7354482412338257,
       "learning_rate": 1.6162539726601477e-05,
-      "loss": 0.2806,
+      "loss": 0.2802,
       "step": 31000
     },
     {
       "epoch": 0.315,
-      "grad_norm": 3.3875720500946045,
+      "grad_norm": 3.5115904808044434,
       "learning_rate": 1.603420361087849e-05,
       "loss": 0.2791,
       "step": 31500
     },
     {
       "epoch": 0.32,
-      "grad_norm": 1.5244992971420288,
+      "grad_norm": 1.5374090671539307,
       "learning_rate": 1.5904285129838924e-05,
-      "loss": 0.2808,
+      "loss": 0.2803,
       "step": 32000
     },
     {
       "epoch": 0.325,
-      "grad_norm": 2.7460532188415527,
+      "grad_norm": 2.6573033332824707,
       "learning_rate": 1.577281835235279e-05,
       "loss": 0.2719,
       "step": 32500
     },
     {
       "epoch": 0.33,
-      "grad_norm": 2.344585418701172,
+      "grad_norm": 2.28947377204895,
       "learning_rate": 1.563983775330399e-05,
-      "loss": 0.2703,
+      "loss": 0.2699,
       "step": 33000
     },
     {
       "epoch": 0.335,
-      "grad_norm": 2.091862678527832,
+      "grad_norm": 2.1672635078430176,
       "learning_rate": 1.5505378204549907e-05,
-      "loss": 0.2782,
+      "loss": 0.2779,
       "step": 33500
     },
     {
       "epoch": 0.34,
-      "grad_norm": 5.391460418701172,
+      "grad_norm": 5.222517490386963,
       "learning_rate": 1.536947496577684e-05,
-      "loss": 0.2789,
+      "loss": 0.2792,
       "step": 34000
     },
     {
       "epoch": 0.345,
-      "grad_norm": 2.717209577560425,
+      "grad_norm": 2.769939661026001,
       "learning_rate": 1.5232163675253793e-05,
-      "loss": 0.2838,
+      "loss": 0.2829,
       "step": 34500
     },
     {
       "epoch": 0.35,
-      "grad_norm": 1.9279614686965942,
+      "grad_norm": 1.86508309841156,
       "learning_rate": 1.5093480340486922e-05,
-      "loss": 0.2695,
+      "loss": 0.2696,
       "step": 35000
     },
     {
       "epoch": 0.355,
-      "grad_norm": 2.6572341918945312,
+      "grad_norm": 2.6956353187561035,
       "learning_rate": 1.495346132877721e-05,
-      "loss": 0.2696,
+      "loss": 0.2694,
       "step": 35500
     },
     {
       "epoch": 0.36,
-      "grad_norm": 3.5666520595550537,
+      "grad_norm": 3.8267648220062256,
       "learning_rate": 1.4812143357683773e-05,
-      "loss": 0.2559,
+      "loss": 0.2557,
       "step": 36000
     },
     {
       "epoch": 0.365,
-      "grad_norm": 2.403407096862793,
+      "grad_norm": 2.288548231124878,
       "learning_rate": 1.4669563485395315e-05,
-      "loss": 0.2737,
+      "loss": 0.2736,
       "step": 36500
     },
     {
       "epoch": 0.37,
-      "grad_norm": 2.280306577682495,
+      "grad_norm": 2.274164915084839,
       "learning_rate": 1.4525759101012249e-05,
-      "loss": 0.2776,
+      "loss": 0.2778,
       "step": 37000
     },
     {
       "epoch": 0.375,
-      "grad_norm": 2.5620460510253906,
+      "grad_norm": 2.6275923252105713,
       "learning_rate": 1.4380767914742074e-05,
-      "loss": 0.271,
+      "loss": 0.2706,
       "step": 37500
     },
     {
       "epoch": 0.38,
-      "grad_norm": 1.8679001331329346,
+      "grad_norm": 1.9517382383346558,
       "learning_rate": 1.4234627948010497e-05,
-      "loss": 0.2605,
+      "loss": 0.2604,
       "step": 38000
     },
     {
       "epoch": 0.385,
-      "grad_norm": 2.4315834045410156,
+      "grad_norm": 2.431234121322632,
       "learning_rate": 1.4087377523491016e-05,
-      "loss": 0.2751,
+      "loss": 0.2752,
       "step": 38500
     },
     {
       "epoch": 0.39,
-      "grad_norm": 2.193459987640381,
+      "grad_norm": 2.1692144870758057,
       "learning_rate": 1.3939055255055419e-05,
-      "loss": 0.2604,
+      "loss": 0.2602,
       "step": 39000
     },
     {
       "epoch": 0.395,
-      "grad_norm": 3.0785605907440186,
+      "grad_norm": 3.2636561393737793,
       "learning_rate": 1.3789700037647988e-05,
-      "loss": 0.2678,
+      "loss": 0.2676,
       "step": 39500
     },
     {
       "epoch": 0.4,
-      "grad_norm": 4.347239971160889,
+      "grad_norm": 4.37136173248291,
       "learning_rate": 1.363935103708598e-05,
-      "loss": 0.2587,
+      "loss": 0.2586,
       "step": 40000
     },
     {
       "epoch": 0.405,
-      "grad_norm": 4.212979793548584,
+      "grad_norm": 4.321747779846191,
       "learning_rate": 1.3488047679789052e-05,
-      "loss": 0.2742,
+      "loss": 0.2738,
       "step": 40500
     },
     {
       "epoch": 0.41,
-      "grad_norm": 2.030700206756592,
+      "grad_norm": 2.1585159301757812,
       "learning_rate": 1.3335829642440372e-05,
-      "loss": 0.2632,
+      "loss": 0.2633,
       "step": 41000
     },
     {
       "epoch": 0.415,
-      "grad_norm": 2.8373637199401855,
+      "grad_norm": 2.787271738052368,
       "learning_rate": 1.3182736841582109e-05,
-      "loss": 0.2726,
+      "loss": 0.2722,
       "step": 41500
     },
     {
       "epoch": 0.42,
-      "grad_norm": 2.2016520500183105,
+      "grad_norm": 2.1286282539367676,
       "learning_rate": 1.3028809423147997e-05,
-      "loss": 0.2795,
+      "loss": 0.2794,
       "step": 42000
     },
     {
       "epoch": 0.425,
-      "grad_norm": 2.687300682067871,
+      "grad_norm": 2.528885841369629,
       "learning_rate": 1.2874087751935756e-05,
-      "loss": 0.2708,
+      "loss": 0.2704,
       "step": 42500
     },
     {
       "epoch": 0.43,
-      "grad_norm": 2.179485559463501,
+      "grad_norm": 2.1489899158477783,
       "learning_rate": 1.2718612401022148e-05,
-      "loss": 0.2687,
+      "loss": 0.2683,
       "step": 43000
     },
     {
       "epoch": 0.435,
-      "grad_norm": 2.3243203163146973,
+      "grad_norm": 2.373365640640259,
       "learning_rate": 1.2562424141123364e-05,
-      "loss": 0.2704,
+      "loss": 0.2703,
       "step": 43500
     },
     {
       "epoch": 0.44,
-      "grad_norm": 3.0649986267089844,
+      "grad_norm": 2.9887542724609375,
       "learning_rate": 1.2405563929903614e-05,
-      "loss": 0.2694,
+      "loss": 0.2692,
       "step": 44000
     },
     {
       "epoch": 0.445,
-      "grad_norm": 3.376861333847046,
+      "grad_norm": 3.2291131019592285,
       "learning_rate": 1.2248072901234715e-05,
-      "loss": 0.2642,
+      "loss": 0.2639,
       "step": 44500
     },
     {
       "epoch": 0.45,
-      "grad_norm": 2.2706363201141357,
+      "grad_norm": 2.0922982692718506,
       "learning_rate": 1.2089992354409415e-05,
-      "loss": 0.268,
+      "loss": 0.2679,
       "step": 45000
     },
     {
       "epoch": 0.455,
-      "grad_norm": 2.4422154426574707,
+      "grad_norm": 2.3987605571746826,
       "learning_rate": 1.1931363743311377e-05,
-      "loss": 0.2642,
+      "loss": 0.2644,
       "step": 45500
     },
     {
       "epoch": 0.46,
-      "grad_norm": 2.9449546337127686,
+      "grad_norm": 2.9259192943573,
       "learning_rate": 1.1772228665544634e-05,
-      "loss": 0.2631,
+      "loss": 0.2627,
       "step": 46000
     },
     {
       "epoch": 0.465,
-      "grad_norm": 2.0101988315582275,
+      "grad_norm": 1.9953309297561646,
       "learning_rate": 1.1612628851525313e-05,
-      "loss": 0.2691,
+      "loss": 0.2694,
       "step": 46500
     },
     {
       "epoch": 0.47,
-      "grad_norm": 2.6115427017211914,
+      "grad_norm": 2.523066759109497,
       "learning_rate": 1.1452606153538577e-05,
       "loss": 0.2634,
       "step": 47000
     },
     {
       "epoch": 0.475,
-      "grad_norm": 1.2970906496047974,
+      "grad_norm": 1.3646570444107056,
       "learning_rate": 1.129220253476358e-05,
-      "loss": 0.2656,
+      "loss": 0.2648,
       "step": 47500
     },
     {
       "epoch": 0.48,
-      "grad_norm": 3.1013381481170654,
+      "grad_norm": 2.9045891761779785,
       "learning_rate": 1.1131460058269342e-05,
-      "loss": 0.2546,
+      "loss": 0.2545,
       "step": 48000
     },
     {
       "epoch": 0.485,
-      "grad_norm": 2.3590879440307617,
+      "grad_norm": 2.251849889755249,
       "learning_rate": 1.0970420875984453e-05,
-      "loss": 0.2637,
+      "loss": 0.2634,
       "step": 48500
     },
     {
       "epoch": 0.49,
-      "grad_norm": 3.6027493476867676,
+      "grad_norm": 3.584717035293579,
       "learning_rate": 1.0809127217643462e-05,
-      "loss": 0.2637,
+      "loss": 0.2632,
       "step": 49000
     },
     {
       "epoch": 0.495,
-      "grad_norm": 2.193842649459839,
+      "grad_norm": 2.139634132385254,
       "learning_rate": 1.0647621379712843e-05,
-      "loss": 0.2659,
+      "loss": 0.2655,
       "step": 49500
     },
     {
       "epoch": 0.5,
-      "grad_norm": 1.900044560432434,
+      "grad_norm": 1.8866384029388428,
       "learning_rate": 1.0485945714299487e-05,
-      "loss": 0.2546,
+      "loss": 0.2543,
       "step": 50000
     },
     {
       "epoch": 0.505,
-      "grad_norm": 3.0313541889190674,
+      "grad_norm": 2.74698543548584,
       "learning_rate": 1.0324142618044612e-05,
-      "loss": 0.2692,
+      "loss": 0.2686,
       "step": 50500
     },
     {
       "epoch": 0.51,
-      "grad_norm": 1.9150141477584839,
+      "grad_norm": 1.884582281112671,
       "learning_rate": 1.0162254521005956e-05,
-      "loss": 0.2637,
+      "loss": 0.2631,
       "step": 51000
     },
     {
       "epoch": 0.515,
-      "grad_norm": 2.062260389328003,
+      "grad_norm": 2.098027467727661,
       "learning_rate": 1.0000323875531241e-05,
-      "loss": 0.2605,
+      "loss": 0.2598,
       "step": 51500
     },
     {
       "epoch": 0.52,
-      "grad_norm": 2.8927578926086426,
+      "grad_norm": 2.741763114929199,
       "learning_rate": 9.838393145125787e-06,
-      "loss": 0.2612,
+      "loss": 0.2611,
       "step": 52000
     },
     {
       "epoch": 0.525,
-      "grad_norm": 3.342867136001587,
+      "grad_norm": 3.3235342502593994,
       "learning_rate": 9.676504793317163e-06,
-      "loss": 0.2588,
+      "loss": 0.2585,
       "step": 52500
     },
     {
       "epoch": 0.53,
-      "grad_norm": 2.1746459007263184,
+      "grad_norm": 2.0972297191619873,
       "learning_rate": 9.514701272519918e-06,
-      "loss": 0.2547,
+      "loss": 0.2542,
       "step": 53000
     },
     {
       "epoch": 0.535,
-      "grad_norm": 1.5649690628051758,
+      "grad_norm": 1.5382198095321655,
       "learning_rate": 9.353025012903112e-06,
       "loss": 0.2642,
       "step": 53500
     },
     {
       "epoch": 0.54,
-      "grad_norm": 2.0328540802001953,
+      "grad_norm": 1.974549412727356,
       "learning_rate": 9.191518411263777e-06,
-      "loss": 0.2426,
+      "loss": 0.2419,
       "step": 54000
     },
     {
       "epoch": 0.545,
-      "grad_norm": 2.822202682495117,
+      "grad_norm": 2.693479061126709,
       "learning_rate": 9.03022381990908e-06,
-      "loss": 0.2694,
+      "loss": 0.2691,
       "step": 54500
     },
     {
       "epoch": 0.55,
-      "grad_norm": 4.873462200164795,
+      "grad_norm": 5.070830821990967,
       "learning_rate": 8.869183535550167e-06,
-      "loss": 0.2723,
+      "loss": 0.2734,
       "step": 55000
     },
     {
       "epoch": 0.555,
-      "grad_norm": 3.1478092670440674,
+      "grad_norm": 3.0315475463867188,
       "learning_rate": 8.708439788210573e-06,
-      "loss": 0.2691,
+      "loss": 0.2692,
       "step": 55500
     },
     {
       "epoch": 0.56,
-      "grad_norm": 3.296886920928955,
+      "grad_norm": 3.1777706146240234,
       "learning_rate": 8.548034730152153e-06,
-      "loss": 0.2628,
+      "loss": 0.2625,
       "step": 56000
     },
     {
       "epoch": 0.565,
-      "grad_norm": 2.208437442779541,
+      "grad_norm": 2.1810688972473145,
       "learning_rate": 8.388010424821354e-06,
-      "loss": 0.2655,
+      "loss": 0.2651,
       "step": 56500
     },
     {
       "epoch": 0.57,
-      "grad_norm": 2.3353774547576904,
+      "grad_norm": 2.2163448333740234,
       "learning_rate": 8.22840883581883e-06,
-      "loss": 0.2527,
+      "loss": 0.2526,
       "step": 57000
     },
     {
       "epoch": 0.575,
-      "grad_norm": 2.4625635147094727,
+      "grad_norm": 2.2531285285949707,
       "learning_rate": 8.069271815895196e-06,
-      "loss": 0.281,
+      "loss": 0.2809,
       "step": 57500
     },
     {
       "epoch": 0.58,
-      "grad_norm": 4.710414886474609,
+      "grad_norm": 4.996068954467773,
       "learning_rate": 7.910641095975886e-06,
-      "loss": 0.2583,
+      "loss": 0.2585,
       "step": 58000
     },
     {
       "epoch": 0.585,
-      "grad_norm": 1.024167776107788,
+      "grad_norm": 1.0134843587875366,
       "learning_rate": 7.752558274217919e-06,
-      "loss": 0.2538,
+      "loss": 0.2537,
       "step": 58500
     },
     {
       "epoch": 0.59,
-      "grad_norm": 3.2519004344940186,
+      "grad_norm": 3.20263671875,
       "learning_rate": 7.595064805101538e-06,
-      "loss": 0.2509,
+      "loss": 0.2506,
       "step": 59000
     },
     {
       "epoch": 0.595,
-      "grad_norm": 1.9836984872817993,
+      "grad_norm": 2.070533275604248,
       "learning_rate": 7.438201988559449e-06,
       "loss": 0.2648,
       "step": 59500
     },
     {
       "epoch": 0.6,
-      "grad_norm": 5.451993942260742,
+      "grad_norm": 5.259614944458008,
       "learning_rate": 7.282010959146662e-06,
-      "loss": 0.2653,
+      "loss": 0.2647,
       "step": 60000
     },
     {
       "epoch": 0.605,
-      "grad_norm": 4.014955043792725,
+      "grad_norm": 4.084048748016357,
       "learning_rate": 7.1265326752536545e-06,
       "loss": 0.2632,
       "step": 60500
     },
     {
       "epoch": 0.61,
-      "grad_norm": 2.711050510406494,
+      "grad_norm": 2.7259840965270996,
       "learning_rate": 6.971807908365737e-06,
-      "loss": 0.2587,
+      "loss": 0.2584,
       "step": 61000
     },
     {
       "epoch": 0.615,
-      "grad_norm": 2.784024238586426,
+      "grad_norm": 2.913233995437622,
       "learning_rate": 6.817877232371453e-06,
-      "loss": 0.2656,
+      "loss": 0.2655,
       "step": 61500
     },
     {
       "epoch": 0.62,
-      "grad_norm": 2.6961591243743896,
+      "grad_norm": 2.728580951690674,
       "learning_rate": 6.66478101292277e-06,
       "loss": 0.2502,
       "step": 62000
     },
     {
       "epoch": 0.625,
-      "grad_norm": 3.2568044662475586,
+      "grad_norm": 3.386467933654785,
       "learning_rate": 6.5125593968498855e-06,
-      "loss": 0.2457,
+      "loss": 0.2458,
       "step": 62500
     },
     {
       "epoch": 0.63,
-      "grad_norm": 3.6847946643829346,
+      "grad_norm": 3.570934534072876,
       "learning_rate": 6.361252301633422e-06,
-      "loss": 0.2639,
+      "loss": 0.2637,
       "step": 63000
     },
     {
       "epoch": 0.635,
-      "grad_norm": 2.6187539100646973,
+      "grad_norm": 2.603757619857788,
       "learning_rate": 6.210899404936761e-06,
-      "loss": 0.2536,
+      "loss": 0.2535,
       "step": 63500
     },
     {
       "epoch": 0.64,
-      "grad_norm": 3.7911410331726074,
+      "grad_norm": 3.9069035053253174,
       "learning_rate": 6.061540134201248e-06,
-      "loss": 0.2573,
+      "loss": 0.2564,
       "step": 64000
     },
     {
       "epoch": 0.645,
-      "grad_norm": 4.211312770843506,
+      "grad_norm": 4.281370162963867,
       "learning_rate": 5.913213656307049e-06,
-      "loss": 0.2607,
+      "loss": 0.2602,
       "step": 64500
     },
     {
       "epoch": 0.65,
-      "grad_norm": 2.4206204414367676,
+      "grad_norm": 2.389120101928711,
       "learning_rate": 5.765958867302311e-06,
-      "loss": 0.2524,
+      "loss": 0.252,
       "step": 65000
     },
     {
       "epoch": 0.655,
-      "grad_norm": 2.050504207611084,
+      "grad_norm": 2.0732436180114746,
       "learning_rate": 5.619814382203341e-06,
-      "loss": 0.2444,
+      "loss": 0.2439,
       "step": 65500
     },
     {
       "epoch": 0.66,
-      "grad_norm": 3.716695547103882,
+      "grad_norm": 3.80635142326355,
       "learning_rate": 5.474818524868482e-06,
-      "loss": 0.2741,
+      "loss": 0.274,
       "step": 66000
     },
     {
       "epoch": 0.665,
-      "grad_norm": 2.9382705688476562,
+      "grad_norm": 2.9537928104400635,
       "learning_rate": 5.331009317948359e-06,
-      "loss": 0.2671,
+      "loss": 0.2676,
       "step": 66500
     },
     {
       "epoch": 0.67,
-      "grad_norm": 2.7678606510162354,
+      "grad_norm": 2.7541537284851074,
       "learning_rate": 5.188424472915091e-06,
       "loss": 0.2708,
       "step": 67000
     },
     {
       "epoch": 0.675,
-      "grad_norm": 2.4799718856811523,
+      "grad_norm": 2.415677070617676,
       "learning_rate": 5.047101380173111e-06,
-      "loss": 0.2451,
+      "loss": 0.2449,
       "step": 67500
     },
     {
       "epoch": 0.68,
-      "grad_norm": 3.983339309692383,
+      "grad_norm": 3.56915020942688,
       "learning_rate": 4.907077099254194e-06,
-      "loss": 0.2611,
+      "loss": 0.2608,
       "step": 68000
     },
     {
       "epoch": 0.685,
-      "grad_norm": 2.8727800846099854,
+      "grad_norm": 2.8096132278442383,
       "learning_rate": 4.7683883490991985e-06,
-      "loss": 0.2623,
+      "loss": 0.263,
       "step": 68500
     },
     {
       "epoch": 0.69,
-      "grad_norm": 1.0295288562774658,
+      "grad_norm": 1.026166558265686,
       "learning_rate": 4.6310714984292096e-06,
-      "loss": 0.2624,
+      "loss": 0.2621,
       "step": 69000
     },
     {
       "epoch": 0.695,
-      "grad_norm": 1.1261619329452515,
+      "grad_norm": 0.8848626613616943,
       "learning_rate": 4.495162556208438e-06,
-      "loss": 0.2506,
+      "loss": 0.2505,
       "step": 69500
     },
     {
       "epoch": 0.7,
-      "grad_norm": 6.0587077140808105,
+      "grad_norm": 6.4718217849731445,
       "learning_rate": 4.360697162201508e-06,
-      "loss": 0.2545,
+      "loss": 0.2543,
       "step": 70000
     },
     {
       "epoch": 0.705,
-      "grad_norm": 3.328782081604004,
+      "grad_norm": 3.0504908561706543,
       "learning_rate": 4.2277105776275395e-06,
-      "loss": 0.2499,
+      "loss": 0.2496,
       "step": 70500
     },
     {
       "epoch": 0.71,
-      "grad_norm": 2.418349027633667,
+      "grad_norm": 2.385812997817993,
       "learning_rate": 4.096237675913521e-06,
-      "loss": 0.262,
+      "loss": 0.2618,
       "step": 71000
     },
     {
       "epoch": 0.715,
-      "grad_norm": 2.3388543128967285,
+      "grad_norm": 2.370842695236206,
       "learning_rate": 3.966312933549325e-06,
-      "loss": 0.2411,
+      "loss": 0.2407,
       "step": 71500
     },
     {
       "epoch": 0.72,
-      "grad_norm": 4.206648349761963,
+      "grad_norm": 4.309947967529297,
       "learning_rate": 3.837970421046881e-06,
-      "loss": 0.2502,
+      "loss": 0.25,
       "step": 72000
     },
     {
       "epoch": 0.725,
-      "grad_norm": 3.9209518432617188,
+      "grad_norm": 3.7516446113586426,
       "learning_rate": 3.7112437940057543e-06,
-      "loss": 0.2503,
+      "loss": 0.2499,
       "step": 72500
     },
     {
       "epoch": 0.73,
-      "grad_norm": 2.7175557613372803,
+      "grad_norm": 2.502753973007202,
       "learning_rate": 3.5861662842875534e-06,
-      "loss": 0.2422,
+      "loss": 0.2424,
       "step": 73000
     },
     {
       "epoch": 0.735,
-      "grad_norm": 4.289679527282715,
+      "grad_norm": 4.212218761444092,
       "learning_rate": 3.4627706913014503e-06,
-      "loss": 0.2553,
+      "loss": 0.2549,
       "step": 73500
     },
     {
       "epoch": 0.74,
-      "grad_norm": 2.622404098510742,
+      "grad_norm": 2.6515533924102783,
       "learning_rate": 3.3410893734031046e-06,
-      "loss": 0.2632,
+      "loss": 0.2633,
       "step": 74000
     },
     {
       "epoch": 0.745,
-      "grad_norm": 2.628495216369629,
+      "grad_norm": 2.6154916286468506,
       "learning_rate": 3.2211542394092443e-06,
-      "loss": 0.2526,
+      "loss": 0.2523,
       "step": 74500
     },
     {
       "epoch": 0.75,
-      "grad_norm": 2.4453587532043457,
+      "grad_norm": 2.3544111251831055,
       "learning_rate": 3.102996740230131e-06,
       "loss": 0.2445,
       "step": 75000
     },
     {
       "epoch": 0.755,
-      "grad_norm": 1.8122838735580444,
+      "grad_norm": 1.8452802896499634,
       "learning_rate": 2.986647860622105e-06,
-      "loss": 0.2611,
+      "loss": 0.2612,
       "step": 75500
     },
     {
       "epoch": 0.76,
-      "grad_norm": 1.76727294921875,
+      "grad_norm": 1.7021652460098267,
       "learning_rate": 2.8721381110623724e-06,
       "loss": 0.2494,
       "step": 76000
     },
     {
       "epoch": 0.765,
-      "grad_norm": 2.5952329635620117,
+      "grad_norm": 2.5925469398498535,
       "learning_rate": 2.7594975197481546e-06,
-      "loss": 0.2503,
+      "loss": 0.25,
       "step": 76500
     },
     {
       "epoch": 0.77,
-      "grad_norm": 3.9623279571533203,
+      "grad_norm": 3.866541862487793,
       "learning_rate": 2.64875562472232e-06,
-      "loss": 0.2481,
+      "loss": 0.2483,
       "step": 77000
     },
     {
       "epoch": 0.775,
-      "grad_norm": 4.493262767791748,
+      "grad_norm": 4.389611721038818,
       "learning_rate": 2.5399414661275444e-06,
-      "loss": 0.2593,
+      "loss": 0.2588,
       "step": 77500
     },
     {
       "epoch": 0.78,
-      "grad_norm": 1.9906212091445923,
+      "grad_norm": 1.7422980070114136,
       "learning_rate": 2.433083578591029e-06,
-      "loss": 0.2482,
+      "loss": 0.248,
       "step": 78000
     },
     {
       "epoch": 0.785,
-      "grad_norm": 3.287825107574463,
+      "grad_norm": 3.3398666381835938,
       "learning_rate": 2.3282099837418004e-06,
-      "loss": 0.2578,
+      "loss": 0.2582,
       "step": 78500
     },
     {
       "epoch": 0.79,
-      "grad_norm": 3.720820665359497,
+      "grad_norm": 3.718583822250366,
       "learning_rate": 2.2253481828625144e-06,
-      "loss": 0.2658,
+      "loss": 0.2661,
       "step": 79000
     },
     {
       "epoch": 0.795,
-      "grad_norm": 4.187972545623779,
+      "grad_norm": 3.974256753921509,
       "learning_rate": 2.1245251496777274e-06,
-      "loss": 0.2555,
+      "loss": 0.2553,
       "step": 79500
     },
     {
       "epoch": 0.8,
-      "grad_norm": 2.308875560760498,
+      "grad_norm": 2.317969799041748,
       "learning_rate": 2.025767323280502e-06,
       "loss": 0.2556,
       "step": 80000
     },
     {
       "epoch": 0.805,
-      "grad_norm": 5.726957321166992,
+      "grad_norm": 5.751351356506348,
       "learning_rate": 1.929100601199211e-06,
-      "loss": 0.239,
+      "loss": 0.2388,
       "step": 80500
     },
     {
       "epoch": 0.81,
-      "grad_norm": 3.462347984313965,
+      "grad_norm": 3.3387367725372314,
       "learning_rate": 1.834550332606365e-06,
-      "loss": 0.2357,
+      "loss": 0.2353,
       "step": 81000
     },
     {
       "epoch": 0.815,
-      "grad_norm": 2.8061912059783936,
+      "grad_norm": 2.7964117527008057,
       "learning_rate": 1.742141311671227e-06,
-      "loss": 0.2496,
+      "loss": 0.2499,
       "step": 81500
     },
     {
       "epoch": 0.82,
-      "grad_norm": 1.9805946350097656,
+      "grad_norm": 2.3127243518829346,
       "learning_rate": 1.6518977710579708e-06,
-      "loss": 0.2692,
+      "loss": 0.2693,
       "step": 82000
     },
     {
       "epoch": 0.825,
-      "grad_norm": 2.1562674045562744,
+      "grad_norm": 2.1335015296936035,
       "learning_rate": 1.5638433755710892e-06,
-      "loss": 0.252,
+      "loss": 0.2516,
       "step": 82500
     },
     {
       "epoch": 0.83,
-      "grad_norm": 2.197392225265503,
+      "grad_norm": 2.2353951930999756,
       "learning_rate": 1.4780012159497037e-06,
-      "loss": 0.2645,
+      "loss": 0.2643,
       "step": 83000
     },
     {
       "epoch": 0.835,
-      "grad_norm": 1.9730626344680786,
+      "grad_norm": 2.0724055767059326,
       "learning_rate": 1.3943938028124216e-06,
-      "loss": 0.2579,
+      "loss": 0.2568,
       "step": 83500
     },
     {
       "epoch": 0.84,
-      "grad_norm": 6.6025214195251465,
+      "grad_norm": 6.218677520751953,
       "learning_rate": 1.3130430607543166e-06,
       "loss": 0.2567,
       "step": 84000
     },
     {
       "epoch": 0.845,
-      "grad_norm": 5.605392932891846,
+      "grad_norm": 5.340097904205322,
       "learning_rate": 1.2339703225975818e-06,
-      "loss": 0.2514,
+      "loss": 0.2517,
       "step": 84500
     },
     {
       "epoch": 0.85,
-      "grad_norm": 2.08113694190979,
+      "grad_norm": 2.1726861000061035,
       "learning_rate": 1.1571963237973693e-06,
-      "loss": 0.2465,
+      "loss": 0.2463,
       "step": 85000
     },
     {
       "epoch": 0.855,
-      "grad_norm": 3.3233418464660645,
+      "grad_norm": 3.2792880535125732,
       "learning_rate": 1.0827411970042723e-06,
-      "loss": 0.2537,
+      "loss": 0.2534,
       "step": 85500
     },
     {
       "epoch": 0.86,
-      "grad_norm": 3.9959723949432373,
+      "grad_norm": 3.8043406009674072,
       "learning_rate": 1.010624466784892e-06,
-      "loss": 0.2488,
+      "loss": 0.249,
       "step": 86000
     },
     {
       "epoch": 0.865,
-      "grad_norm": 4.4443678855896,
+      "grad_norm": 4.719531059265137,
       "learning_rate": 9.40865044501853e-07,
-      "loss": 0.2502,
+      "loss": 0.2503,
       "step": 86500
     },
     {
       "epoch": 0.87,
-      "grad_norm": 2.794731378555298,
+      "grad_norm": 2.73796010017395,
       "learning_rate": 8.734812233546319e-07,
-      "loss": 0.2533,
+      "loss": 0.2535,
       "step": 87000
     },
     {
       "epoch": 0.875,
-      "grad_norm": 3.05753755569458,
+      "grad_norm": 2.953639030456543,
       "learning_rate": 8.084906735824771e-07,
-      "loss": 0.2541,
+      "loss": 0.2544,
       "step": 87500
     },
     {
       "epoch": 0.88,
-      "grad_norm": 5.455068588256836,
+      "grad_norm": 5.714886665344238,
       "learning_rate": 7.45910437830697e-07,
       "loss": 0.2505,
       "step": 88000
     },
     {
       "epoch": 0.885,
-      "grad_norm": 3.3121325969696045,
+      "grad_norm": 3.277892827987671,
       "learning_rate": 6.857569266815233e-07,
-      "loss": 0.2472,
+      "loss": 0.2474,
       "step": 88500
     },
     {
       "epoch": 0.89,
-      "grad_norm": 3.913578510284424,
+      "grad_norm": 3.6936960220336914,
       "learning_rate": 6.2804591435072e-07,
-      "loss": 0.2621,
+      "loss": 0.2611,
       "step": 89000
     },
     {
       "epoch": 0.895,
-      "grad_norm": 1.292697787284851,
+      "grad_norm": 1.2861735820770264,
       "learning_rate": 5.727925345510644e-07,
-      "loss": 0.2434,
+      "loss": 0.2432,
       "step": 89500
     },
     {
       "epoch": 0.9,
-      "grad_norm": 1.8864761590957642,
+      "grad_norm": 1.9155391454696655,
       "learning_rate": 5.200112765237985e-07,
-      "loss": 0.2511,
+      "loss": 0.2508,
       "step": 90000
     },
     {
       "epoch": 0.905,
-      "grad_norm": 1.9154138565063477,
+      "grad_norm": 1.7853316068649292,
       "learning_rate": 4.697159812390739e-07,
       "loss": 0.2452,
       "step": 90500
     },
     {
       "epoch": 0.91,
-      "grad_norm": 3.7760703563690186,
+      "grad_norm": 4.1483235359191895,
       "learning_rate": 4.21919837766398e-07,
-      "loss": 0.2488,
+      "loss": 0.2483,
       "step": 91000
     },
     {
       "epoch": 0.915,
-      "grad_norm": 2.573406219482422,
+      "grad_norm": 2.7010982036590576,
       "learning_rate": 3.766353798160283e-07,
-      "loss": 0.2441,
+      "loss": 0.2435,
       "step": 91500
     },
     {
       "epoch": 0.92,
-      "grad_norm": 2.5799012184143066,
+      "grad_norm": 2.720015048980713,
       "learning_rate": 3.338744824522266e-07,
-      "loss": 0.249,
+      "loss": 0.2488,
       "step": 92000
     },
     {
       "epoch": 0.925,
-      "grad_norm": 2.0907909870147705,
+      "grad_norm": 2.154423952102661,
       "learning_rate": 2.9364835897922606e-07,
-      "loss": 0.2522,
+      "loss": 0.2523,
       "step": 92500
     },
     {
       "epoch": 0.93,
-      "grad_norm": 5.767177104949951,
+      "grad_norm": 5.67596960067749,
       "learning_rate": 2.5596755800073593e-07,
-      "loss": 0.2487,
+      "loss": 0.2483,
       "step": 93000
     },
     {
       "epoch": 0.935,
-      "grad_norm": 2.8661417961120605,
+      "grad_norm": 2.8991081714630127,
       "learning_rate": 2.208419606537593e-07,
-      "loss": 0.2533,
+      "loss": 0.2536,
       "step": 93500
     },
     {
       "epoch": 0.94,
-      "grad_norm": 3.0605692863464355,
+      "grad_norm": 2.9144272804260254,
       "learning_rate": 1.8828077801743006e-07,
-      "loss": 0.2566,
+      "loss": 0.2557,
       "step": 94000
     },
     {
       "epoch": 0.945,
-      "grad_norm": 2.973830461502075,
+      "grad_norm": 3.0168917179107666,
       "learning_rate": 1.5829254869756395e-07,
-      "loss": 0.2569,
+      "loss": 0.2567,
       "step": 94500
     },
     {
       "epoch": 0.95,
-      "grad_norm": 1.0513603687286377,
+      "grad_norm": 0.992645800113678,
       "learning_rate": 1.3088513658756962e-07,
-      "loss": 0.2495,
+      "loss": 0.2498,
       "step": 95000
     },
     {
       "epoch": 0.955,
-      "grad_norm": 2.0021743774414062,
+      "grad_norm": 2.234450101852417,
       "learning_rate": 1.060657288062672e-07,
-      "loss": 0.2504,
+      "loss": 0.2503,
       "step": 95500
     },
     {
       "epoch": 0.96,
-      "grad_norm": 2.6165597438812256,
+      "grad_norm": 2.546241521835327,
       "learning_rate": 8.384083381319464e-08,
-      "loss": 0.2442,
+      "loss": 0.2446,
       "step": 96000
     },
     {
       "epoch": 0.965,
-      "grad_norm": 3.900705099105835,
+      "grad_norm": 4.158427715301514,
       "learning_rate": 6.421627970187971e-08,
-      "loss": 0.2469,
+      "loss": 0.2467,
       "step": 96500
     },
     {
       "epoch": 0.97,
-      "grad_norm": 2.7001099586486816,
+      "grad_norm": 2.6824352741241455,
       "learning_rate": 4.719721267152033e-08,
-      "loss": 0.2593,
+      "loss": 0.2586,
       "step": 97000
     },
     {
       "epoch": 0.975,
-      "grad_norm": 4.71160888671875,
+      "grad_norm": 4.74870491027832,
       "learning_rate": 3.2788095677480644e-08,
-      "loss": 0.2523,
+      "loss": 0.252,
       "step": 97500
     },
     {
       "epoch": 0.98,
-      "grad_norm": 2.4307923316955566,
+      "grad_norm": 2.3958590030670166,
       "learning_rate": 2.099270726096503e-08,
-      "loss": 0.2532,
+      "loss": 0.2527,
       "step": 98000
     },
     {
       "epoch": 0.985,
-      "grad_norm": 4.588560581207275,
+      "grad_norm": 4.80270528793335,
       "learning_rate": 1.1814140558158438e-08,
-      "loss": 0.2466,
+      "loss": 0.2459,
       "step": 98500
     },
     {
       "epoch": 0.99,
-      "grad_norm": 3.476001024246216,
+      "grad_norm": 3.4724161624908447,
       "learning_rate": 5.254802489100819e-09,
-      "loss": 0.2541,
+      "loss": 0.254,
       "step": 99000
     },
     {
       "epoch": 0.995,
-      "grad_norm": 1.7518401145935059,
+      "grad_norm": 1.695827603340149,
       "learning_rate": 1.3164131265197555e-09,
-      "loss": 0.2542,
+      "loss": 0.2544,
       "step": 99500
     },
     {
       "epoch": 1.0,
-      "grad_norm": 2.3076059818267822,
+      "grad_norm": 2.3921620845794678,
       "learning_rate": 5.2447679532718895e-15,
-      "loss": 0.2467,
+      "loss": 0.2468,
       "step": 100000
     },
     {
       "epoch": 1.0,
       "step": 100000,
       "total_flos": 1.1222027981664584e+18,
-      "train_loss": 0.27731778343200686,
-      "train_runtime": 19372.6268,
-      "train_samples_per_second": 5.162,
-      "train_steps_per_second": 5.162
+      "train_loss": 0.2771104446411133,
+      "train_runtime": 18624.8013,
+      "train_samples_per_second": 5.369,
+      "train_steps_per_second": 5.369
     }
   ],
   "logging_steps": 500,