diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,7103 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 2.995163584637269,
+  "eval_steps": 50,
+  "global_step": 987,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0030346135609293505,
+      "grad_norm": 5.274968147277832,
+      "learning_rate": 2.0202020202020205e-07,
+      "loss": 0.8522,
+      "step": 1
+    },
+    {
+      "epoch": 0.006069227121858701,
+      "grad_norm": 5.754838466644287,
+      "learning_rate": 4.040404040404041e-07,
+      "loss": 0.854,
+      "step": 2
+    },
+    {
+      "epoch": 0.009103840682788052,
+      "grad_norm": 5.632802963256836,
+      "learning_rate": 6.060606060606061e-07,
+      "loss": 0.873,
+      "step": 3
+    },
+    {
+      "epoch": 0.012138454243717402,
+      "grad_norm": 5.394160747528076,
+      "learning_rate": 8.080808080808082e-07,
+      "loss": 0.8563,
+      "step": 4
+    },
+    {
+      "epoch": 0.015173067804646752,
+      "grad_norm": 4.5878424644470215,
+      "learning_rate": 1.01010101010101e-06,
+      "loss": 0.8425,
+      "step": 5
+    },
+    {
+      "epoch": 0.018207681365576104,
+      "grad_norm": 4.280163764953613,
+      "learning_rate": 1.2121212121212122e-06,
+      "loss": 0.8569,
+      "step": 6
+    },
+    {
+      "epoch": 0.021242294926505454,
+      "grad_norm": 3.5766515731811523,
+      "learning_rate": 1.4141414141414143e-06,
+      "loss": 0.8309,
+      "step": 7
+    },
+    {
+      "epoch": 0.024276908487434804,
+      "grad_norm": 4.070709228515625,
+      "learning_rate": 1.6161616161616164e-06,
+      "loss": 0.825,
+      "step": 8
+    },
+    {
+      "epoch": 0.027311522048364154,
+      "grad_norm": 4.395944595336914,
+      "learning_rate": 1.8181818181818183e-06,
+      "loss": 0.8419,
+      "step": 9
+    },
+    {
+      "epoch": 0.030346135609293504,
+      "grad_norm": 3.765310764312744,
+      "learning_rate": 2.02020202020202e-06,
+      "loss": 0.8426,
+      "step": 10
+    },
+    {
+      "epoch": 0.03338074917022286,
+      "grad_norm": 2.906996488571167,
+      "learning_rate": 2.222222222222222e-06,
+      "loss": 0.8263,
+      "step": 11
+    },
+    {
+      "epoch": 0.03641536273115221,
+      "grad_norm": 2.8421216011047363,
+      "learning_rate": 2.4242424242424244e-06,
+      "loss": 0.8242,
+      "step": 12
+    },
+    {
+      "epoch": 0.03944997629208156,
+      "grad_norm": 2.6818017959594727,
+      "learning_rate": 2.6262626262626267e-06,
+      "loss": 0.8206,
+      "step": 13
+    },
+    {
+      "epoch": 0.04248458985301091,
+      "grad_norm": 2.8597841262817383,
+      "learning_rate": 2.8282828282828286e-06,
+      "loss": 0.8201,
+      "step": 14
+    },
+    {
+      "epoch": 0.04551920341394026,
+      "grad_norm": 2.673666000366211,
+      "learning_rate": 3.0303030303030305e-06,
+      "loss": 0.8215,
+      "step": 15
+    },
+    {
+      "epoch": 0.04855381697486961,
+      "grad_norm": 2.636281728744507,
+      "learning_rate": 3.232323232323233e-06,
+      "loss": 0.8037,
+      "step": 16
+    },
+    {
+      "epoch": 0.05158843053579896,
+      "grad_norm": 2.502042770385742,
+      "learning_rate": 3.4343434343434347e-06,
+      "loss": 0.8162,
+      "step": 17
+    },
+    {
+      "epoch": 0.05462304409672831,
+      "grad_norm": 2.562307834625244,
+      "learning_rate": 3.6363636363636366e-06,
+      "loss": 0.8066,
+      "step": 18
+    },
+    {
+      "epoch": 0.05765765765765766,
+      "grad_norm": 2.6667816638946533,
+      "learning_rate": 3.8383838383838385e-06,
+      "loss": 0.8197,
+      "step": 19
+    },
+    {
+      "epoch": 0.06069227121858701,
+      "grad_norm": 2.293628215789795,
+      "learning_rate": 4.04040404040404e-06,
+      "loss": 0.8092,
+      "step": 20
+    },
+    {
+      "epoch": 0.06372688477951635,
+      "grad_norm": 2.4843485355377197,
+      "learning_rate": 4.242424242424243e-06,
+      "loss": 0.8082,
+      "step": 21
+    },
+    {
+      "epoch": 0.06676149834044572,
+      "grad_norm": 2.5178024768829346,
+      "learning_rate": 4.444444444444444e-06,
+      "loss": 0.8165,
+      "step": 22
+    },
+    {
+      "epoch": 0.06979611190137507,
+      "grad_norm": 2.6686830520629883,
+      "learning_rate": 4.646464646464647e-06,
+      "loss": 0.773,
+      "step": 23
+    },
+    {
+      "epoch": 0.07283072546230442,
+      "grad_norm": 2.2187390327453613,
+      "learning_rate": 4.848484848484849e-06,
+      "loss": 0.8026,
+      "step": 24
+    },
+    {
+      "epoch": 0.07586533902323377,
+      "grad_norm": 2.4837453365325928,
+      "learning_rate": 5.0505050505050515e-06,
+      "loss": 0.7962,
+      "step": 25
+    },
+    {
+      "epoch": 0.07889995258416312,
+      "grad_norm": 2.3953895568847656,
+      "learning_rate": 5.252525252525253e-06,
+      "loss": 0.7902,
+      "step": 26
+    },
+    {
+      "epoch": 0.08193456614509247,
+      "grad_norm": 2.384993553161621,
+      "learning_rate": 5.4545454545454545e-06,
+      "loss": 0.8133,
+      "step": 27
+    },
+    {
+      "epoch": 0.08496917970602182,
+      "grad_norm": 2.5588831901550293,
+      "learning_rate": 5.656565656565657e-06,
+      "loss": 0.8043,
+      "step": 28
+    },
+    {
+      "epoch": 0.08800379326695117,
+      "grad_norm": 2.254196882247925,
+      "learning_rate": 5.858585858585859e-06,
+      "loss": 0.7967,
+      "step": 29
+    },
+    {
+      "epoch": 0.09103840682788052,
+      "grad_norm": 2.2686617374420166,
+      "learning_rate": 6.060606060606061e-06,
+      "loss": 0.8036,
+      "step": 30
+    },
+    {
+      "epoch": 0.09407302038880987,
+      "grad_norm": 2.3785650730133057,
+      "learning_rate": 6.262626262626264e-06,
+      "loss": 0.799,
+      "step": 31
+    },
+    {
+      "epoch": 0.09710763394973922,
+      "grad_norm": 2.4035139083862305,
+      "learning_rate": 6.464646464646466e-06,
+      "loss": 0.8086,
+      "step": 32
+    },
+    {
+      "epoch": 0.10014224751066857,
+      "grad_norm": 2.2935521602630615,
+      "learning_rate": 6.666666666666667e-06,
+      "loss": 0.7842,
+      "step": 33
+    },
+    {
+      "epoch": 0.10317686107159792,
+      "grad_norm": 2.3163795471191406,
+      "learning_rate": 6.868686868686869e-06,
+      "loss": 0.8041,
+      "step": 34
+    },
+    {
+      "epoch": 0.10621147463252727,
+      "grad_norm": 2.3953793048858643,
+      "learning_rate": 7.070707070707071e-06,
+      "loss": 0.7929,
+      "step": 35
+    },
+    {
+      "epoch": 0.10924608819345662,
+      "grad_norm": 2.4116768836975098,
+      "learning_rate": 7.272727272727273e-06,
+      "loss": 0.7943,
+      "step": 36
+    },
+    {
+      "epoch": 0.11228070175438597,
+      "grad_norm": 2.5221264362335205,
+      "learning_rate": 7.474747474747476e-06,
+      "loss": 0.8036,
+      "step": 37
+    },
+    {
+      "epoch": 0.11531531531531532,
+      "grad_norm": 2.4730491638183594,
+      "learning_rate": 7.676767676767677e-06,
+      "loss": 0.8063,
+      "step": 38
+    },
+    {
+      "epoch": 0.11834992887624467,
+      "grad_norm": 2.1653268337249756,
+      "learning_rate": 7.87878787878788e-06,
+      "loss": 0.8009,
+      "step": 39
+    },
+    {
+      "epoch": 0.12138454243717402,
+      "grad_norm": 2.266632080078125,
+      "learning_rate": 8.08080808080808e-06,
+      "loss": 0.7825,
+      "step": 40
+    },
+    {
+      "epoch": 0.12441915599810337,
+      "grad_norm": 2.2630982398986816,
+      "learning_rate": 8.282828282828283e-06,
+      "loss": 0.8004,
+      "step": 41
+    },
+    {
+      "epoch": 0.1274537695590327,
+      "grad_norm": 2.403125762939453,
+      "learning_rate": 8.484848484848486e-06,
+      "loss": 0.7911,
+      "step": 42
+    },
+    {
+      "epoch": 0.13048838311996205,
+      "grad_norm": 2.4483842849731445,
+      "learning_rate": 8.686868686868687e-06,
+      "loss": 0.7861,
+      "step": 43
+    },
+    {
+      "epoch": 0.13352299668089143,
+      "grad_norm": 2.346780776977539,
+      "learning_rate": 8.888888888888888e-06,
+      "loss": 0.7969,
+      "step": 44
+    },
+    {
+      "epoch": 0.13655761024182078,
+      "grad_norm": 2.3152859210968018,
+      "learning_rate": 9.090909090909091e-06,
+      "loss": 0.7842,
+      "step": 45
+    },
+    {
+      "epoch": 0.13959222380275013,
+      "grad_norm": 2.2333388328552246,
+      "learning_rate": 9.292929292929294e-06,
+      "loss": 0.7886,
+      "step": 46
+    },
+    {
+      "epoch": 0.14262683736367948,
+      "grad_norm": 2.360898017883301,
+      "learning_rate": 9.494949494949497e-06,
+      "loss": 0.8146,
+      "step": 47
+    },
+    {
+      "epoch": 0.14566145092460883,
+      "grad_norm": 2.5461983680725098,
+      "learning_rate": 9.696969696969698e-06,
+      "loss": 0.8005,
+      "step": 48
+    },
+    {
+      "epoch": 0.14869606448553818,
+      "grad_norm": 2.4684877395629883,
+      "learning_rate": 9.8989898989899e-06,
+      "loss": 0.8031,
+      "step": 49
+    },
+    {
+      "epoch": 0.15173067804646753,
+      "grad_norm": 2.3025074005126953,
+      "learning_rate": 1.0101010101010103e-05,
+      "loss": 0.7828,
+      "step": 50
+    },
+    {
+      "epoch": 0.15173067804646753,
+      "eval_loss": NaN,
+      "eval_runtime": 233.599,
+      "eval_samples_per_second": 17.8,
+      "eval_steps_per_second": 4.452,
+      "step": 50
+    },
+    {
+      "epoch": 0.15476529160739688,
+      "grad_norm": 2.4276065826416016,
+      "learning_rate": 1.0303030303030304e-05,
+      "loss": 0.7801,
+      "step": 51
+    },
+    {
+      "epoch": 0.15779990516832623,
+      "grad_norm": 2.279287815093994,
+      "learning_rate": 1.0505050505050507e-05,
+      "loss": 0.8055,
+      "step": 52
+    },
+    {
+      "epoch": 0.16083451872925558,
+      "grad_norm": 2.7459404468536377,
+      "learning_rate": 1.0707070707070708e-05,
+      "loss": 0.7864,
+      "step": 53
+    },
+    {
+      "epoch": 0.16386913229018493,
+      "grad_norm": 2.3488147258758545,
+      "learning_rate": 1.0909090909090909e-05,
+      "loss": 0.7869,
+      "step": 54
+    },
+    {
+      "epoch": 0.16690374585111428,
+      "grad_norm": 2.547166109085083,
+      "learning_rate": 1.1111111111111113e-05,
+      "loss": 0.7869,
+      "step": 55
+    },
+    {
+      "epoch": 0.16993835941204363,
+      "grad_norm": 2.4360105991363525,
+      "learning_rate": 1.1313131313131314e-05,
+      "loss": 0.7883,
+      "step": 56
+    },
+    {
+      "epoch": 0.17297297297297298,
+      "grad_norm": 2.8864669799804688,
+      "learning_rate": 1.1515151515151517e-05,
+      "loss": 0.7927,
+      "step": 57
+    },
+    {
+      "epoch": 0.17600758653390233,
+      "grad_norm": 2.3167998790740967,
+      "learning_rate": 1.1717171717171718e-05,
+      "loss": 0.7999,
+      "step": 58
+    },
+    {
+      "epoch": 0.17904220009483168,
+      "grad_norm": 3.075058698654175,
+      "learning_rate": 1.191919191919192e-05,
+      "loss": 0.7888,
+      "step": 59
+    },
+    {
+      "epoch": 0.18207681365576103,
+      "grad_norm": 2.38899827003479,
+      "learning_rate": 1.2121212121212122e-05,
+      "loss": 0.787,
+      "step": 60
+    },
+    {
+      "epoch": 0.18511142721669038,
+      "grad_norm": 3.121044874191284,
+      "learning_rate": 1.2323232323232323e-05,
+      "loss": 0.8053,
+      "step": 61
+    },
+    {
+      "epoch": 0.18814604077761973,
+      "grad_norm": 2.580725908279419,
+      "learning_rate": 1.2525252525252527e-05,
+      "loss": 0.7819,
+      "step": 62
+    },
+    {
+      "epoch": 0.19118065433854908,
+      "grad_norm": 3.1028575897216797,
+      "learning_rate": 1.2727272727272728e-05,
+      "loss": 0.7886,
+      "step": 63
+    },
+    {
+      "epoch": 0.19421526789947843,
+      "grad_norm": 2.6575424671173096,
+      "learning_rate": 1.2929292929292931e-05,
+      "loss": 0.7904,
+      "step": 64
+    },
+    {
+      "epoch": 0.19724988146040778,
+      "grad_norm": 2.8595755100250244,
+      "learning_rate": 1.3131313131313132e-05,
+      "loss": 0.7955,
+      "step": 65
+    },
+    {
+      "epoch": 0.20028449502133713,
+      "grad_norm": 2.235410451889038,
+      "learning_rate": 1.3333333333333333e-05,
+      "loss": 0.796,
+      "step": 66
+    },
+    {
+      "epoch": 0.20331910858226648,
+      "grad_norm": 2.7857375144958496,
+      "learning_rate": 1.3535353535353538e-05,
+      "loss": 0.8008,
+      "step": 67
+    },
+    {
+      "epoch": 0.20635372214319583,
+      "grad_norm": 2.310331106185913,
+      "learning_rate": 1.3737373737373739e-05,
+      "loss": 0.7802,
+      "step": 68
+    },
+    {
+      "epoch": 0.20938833570412518,
+      "grad_norm": 2.8039300441741943,
+      "learning_rate": 1.3939393939393942e-05,
+      "loss": 0.7946,
+      "step": 69
+    },
+    {
+      "epoch": 0.21242294926505453,
+      "grad_norm": 2.345369815826416,
+      "learning_rate": 1.4141414141414143e-05,
+      "loss": 0.7835,
+      "step": 70
+    },
+    {
+      "epoch": 0.21545756282598388,
+      "grad_norm": 2.8384764194488525,
+      "learning_rate": 1.4343434343434344e-05,
+      "loss": 0.7855,
+      "step": 71
+    },
+    {
+      "epoch": 0.21849217638691323,
+      "grad_norm": 2.371159076690674,
+      "learning_rate": 1.4545454545454546e-05,
+      "loss": 0.7689,
+      "step": 72
+    },
+    {
+      "epoch": 0.22152678994784258,
+      "grad_norm": 2.9578475952148438,
+      "learning_rate": 1.4747474747474747e-05,
+      "loss": 0.7918,
+      "step": 73
+    },
+    {
+      "epoch": 0.22456140350877193,
+      "grad_norm": 2.8900325298309326,
+      "learning_rate": 1.4949494949494952e-05,
+      "loss": 0.7746,
+      "step": 74
+    },
+    {
+      "epoch": 0.22759601706970128,
+      "grad_norm": 2.4469008445739746,
+      "learning_rate": 1.5151515151515153e-05,
+      "loss": 0.7919,
+      "step": 75
+    },
+    {
+      "epoch": 0.23063063063063063,
+      "grad_norm": 2.410154104232788,
+      "learning_rate": 1.5353535353535354e-05,
+      "loss": 0.7987,
+      "step": 76
+    },
+    {
+      "epoch": 0.23366524419155998,
+      "grad_norm": 2.599086284637451,
+      "learning_rate": 1.555555555555556e-05,
+      "loss": 0.7714,
+      "step": 77
+    },
+    {
+      "epoch": 0.23669985775248933,
+      "grad_norm": 2.4329092502593994,
+      "learning_rate": 1.575757575757576e-05,
+      "loss": 0.8041,
+      "step": 78
+    },
+    {
+      "epoch": 0.23973447131341868,
+      "grad_norm": 2.7239413261413574,
+      "learning_rate": 1.595959595959596e-05,
+      "loss": 0.7941,
+      "step": 79
+    },
+    {
+      "epoch": 0.24276908487434803,
+      "grad_norm": 2.8684301376342773,
+      "learning_rate": 1.616161616161616e-05,
+      "loss": 0.8076,
+      "step": 80
+    },
+    {
+      "epoch": 0.24580369843527738,
+      "grad_norm": 3.004608631134033,
+      "learning_rate": 1.6363636363636366e-05,
+      "loss": 0.7893,
+      "step": 81
+    },
+    {
+      "epoch": 0.24883831199620673,
+      "grad_norm": 2.6391353607177734,
+      "learning_rate": 1.6565656565656567e-05,
+      "loss": 0.8116,
+      "step": 82
+    },
+    {
+      "epoch": 0.2518729255571361,
+      "grad_norm": 3.3293023109436035,
+      "learning_rate": 1.6767676767676768e-05,
+      "loss": 0.7952,
+      "step": 83
+    },
+    {
+      "epoch": 0.2549075391180654,
+      "grad_norm": 2.5788087844848633,
+      "learning_rate": 1.6969696969696972e-05,
+      "loss": 0.7905,
+      "step": 84
+    },
+    {
+      "epoch": 0.2579421526789948,
+      "grad_norm": 3.4907588958740234,
+      "learning_rate": 1.7171717171717173e-05,
+      "loss": 0.7903,
+      "step": 85
+    },
+    {
+      "epoch": 0.2609767662399241,
+      "grad_norm": 2.8050403594970703,
+      "learning_rate": 1.7373737373737375e-05,
+      "loss": 0.7813,
+      "step": 86
+    },
+    {
+      "epoch": 0.2640113798008535,
+      "grad_norm": 3.180819034576416,
+      "learning_rate": 1.7575757575757576e-05,
+      "loss": 0.7797,
+      "step": 87
+    },
+    {
+      "epoch": 0.26704599336178286,
+      "grad_norm": 2.614614486694336,
+      "learning_rate": 1.7777777777777777e-05,
+      "loss": 0.7903,
+      "step": 88
+    },
+    {
+      "epoch": 0.2700806069227122,
+      "grad_norm": 3.212218761444092,
+      "learning_rate": 1.797979797979798e-05,
+      "loss": 0.7927,
+      "step": 89
+    },
+    {
+      "epoch": 0.27311522048364156,
+      "grad_norm": 2.6879336833953857,
+      "learning_rate": 1.8181818181818182e-05,
+      "loss": 0.8099,
+      "step": 90
+    },
+    {
+      "epoch": 0.2761498340445709,
+      "grad_norm": 3.2700448036193848,
+      "learning_rate": 1.8383838383838387e-05,
+      "loss": 0.794,
+      "step": 91
+    },
+    {
+      "epoch": 0.27918444760550026,
+      "grad_norm": 3.1249783039093018,
+      "learning_rate": 1.8585858585858588e-05,
+      "loss": 0.7807,
+      "step": 92
+    },
+    {
+      "epoch": 0.2822190611664296,
+      "grad_norm": 2.4789459705352783,
+      "learning_rate": 1.8787878787878792e-05,
+      "loss": 0.7829,
+      "step": 93
+    },
+    {
+      "epoch": 0.28525367472735896,
+      "grad_norm": 2.624887466430664,
+      "learning_rate": 1.8989898989898993e-05,
+      "loss": 0.8043,
+      "step": 94
+    },
+    {
+      "epoch": 0.2882882882882883,
+      "grad_norm": 2.3336539268493652,
+      "learning_rate": 1.9191919191919194e-05,
+      "loss": 0.7827,
+      "step": 95
+    },
+    {
+      "epoch": 0.29132290184921766,
+      "grad_norm": 2.5490119457244873,
+      "learning_rate": 1.9393939393939395e-05,
+      "loss": 0.8072,
+      "step": 96
+    },
+    {
+      "epoch": 0.294357515410147,
+      "grad_norm": 2.5659990310668945,
+      "learning_rate": 1.9595959595959596e-05,
+      "loss": 0.7986,
+      "step": 97
+    },
+    {
+      "epoch": 0.29739212897107636,
+      "grad_norm": 2.3554656505584717,
+      "learning_rate": 1.97979797979798e-05,
+      "loss": 0.8059,
+      "step": 98
+    },
+    {
+      "epoch": 0.3004267425320057,
+      "grad_norm": 2.4364328384399414,
+      "learning_rate": 2e-05,
+      "loss": 0.7874,
+      "step": 99
+    },
+    {
+      "epoch": 0.30346135609293506,
+      "grad_norm": 2.3030965328216553,
+      "learning_rate": 1.9999937418875125e-05,
+      "loss": 0.7854,
+      "step": 100
+    },
+    {
+      "epoch": 0.30346135609293506,
+      "eval_loss": NaN,
+      "eval_runtime": 204.7053,
+      "eval_samples_per_second": 20.312,
+      "eval_steps_per_second": 5.08,
+      "step": 100
+    },
+    {
+      "epoch": 0.3064959696538644,
+      "grad_norm": 2.318124294281006,
+      "learning_rate": 1.9999749676283775e-05,
+      "loss": 0.7964,
+      "step": 101
+    },
+    {
+      "epoch": 0.30953058321479376,
+      "grad_norm": 2.2906177043914795,
+      "learning_rate": 1.9999436774575783e-05,
+      "loss": 0.8049,
+      "step": 102
+    },
+    {
+      "epoch": 0.3125651967757231,
+      "grad_norm": 2.2947778701782227,
+      "learning_rate": 1.999899871766749e-05,
+      "loss": 0.7881,
+      "step": 103
+    },
+    {
+      "epoch": 0.31559981033665246,
+      "grad_norm": 2.548788547515869,
+      "learning_rate": 1.999843551104172e-05,
+      "loss": 0.803,
+      "step": 104
+    },
+    {
+      "epoch": 0.3186344238975818,
+      "grad_norm": 2.3735954761505127,
+      "learning_rate": 1.9997747161747696e-05,
+      "loss": 0.8044,
+      "step": 105
+    },
+    {
+      "epoch": 0.32166903745851116,
+      "grad_norm": 2.620364189147949,
+      "learning_rate": 1.9996933678400948e-05,
+      "loss": 0.7806,
+      "step": 106
+    },
+    {
+      "epoch": 0.3247036510194405,
+      "grad_norm": 2.3897509574890137,
+      "learning_rate": 1.999599507118322e-05,
+      "loss": 0.7862,
+      "step": 107
+    },
+    {
+      "epoch": 0.32773826458036986,
+      "grad_norm": 2.6068966388702393,
+      "learning_rate": 1.9994931351842327e-05,
+      "loss": 0.8051,
+      "step": 108
+    },
+    {
+      "epoch": 0.3307728781412992,
+      "grad_norm": 2.560683488845825,
+      "learning_rate": 1.999374253369202e-05,
+      "loss": 0.7956,
+      "step": 109
+    },
+    {
+      "epoch": 0.33380749170222856,
+      "grad_norm": 2.1629045009613037,
+      "learning_rate": 1.999242863161182e-05,
+      "loss": 0.8074,
+      "step": 110
+    },
+    {
+      "epoch": 0.3368421052631579,
+      "grad_norm": 2.653731346130371,
+      "learning_rate": 1.999098966204682e-05,
+      "loss": 0.8071,
+      "step": 111
+    },
+    {
+      "epoch": 0.33987671882408726,
+      "grad_norm": 2.484330177307129,
+      "learning_rate": 1.9989425643007475e-05,
+      "loss": 0.8135,
+      "step": 112
+    },
+    {
+      "epoch": 0.3429113323850166,
+      "grad_norm": 2.3897571563720703,
+      "learning_rate": 1.9987736594069417e-05,
+      "loss": 0.8066,
+      "step": 113
+    },
+    {
+      "epoch": 0.34594594594594597,
+      "grad_norm": 2.653904676437378,
+      "learning_rate": 1.998592253637315e-05,
+      "loss": 0.7908,
+      "step": 114
+    },
+    {
+      "epoch": 0.3489805595068753,
+      "grad_norm": 2.2755186557769775,
+      "learning_rate": 1.9983983492623832e-05,
+      "loss": 0.7824,
+      "step": 115
+    },
+    {
+      "epoch": 0.35201517306780467,
+      "grad_norm": 2.260126829147339,
+      "learning_rate": 1.9981919487090973e-05,
+      "loss": 0.7916,
+      "step": 116
+    },
+    {
+      "epoch": 0.355049786628734,
+      "grad_norm": 2.6538352966308594,
+      "learning_rate": 1.9979730545608128e-05,
+      "loss": 0.7927,
+      "step": 117
+    },
+    {
+      "epoch": 0.35808440018966337,
+      "grad_norm": 2.2571558952331543,
+      "learning_rate": 1.9977416695572577e-05,
+      "loss": 0.7826,
+      "step": 118
+    },
+    {
+      "epoch": 0.3611190137505927,
+      "grad_norm": 2.5154271125793457,
+      "learning_rate": 1.9974977965945e-05,
+      "loss": 0.807,
+      "step": 119
+    },
+    {
+      "epoch": 0.36415362731152207,
+      "grad_norm": 2.4908971786499023,
+      "learning_rate": 1.9972414387249074e-05,
+      "loss": 0.7831,
+      "step": 120
+    },
+    {
+      "epoch": 0.3671882408724514,
+      "grad_norm": 2.3925859928131104,
+      "learning_rate": 1.996972599157113e-05,
+      "loss": 0.7844,
+      "step": 121
+    },
+    {
+      "epoch": 0.37022285443338077,
+      "grad_norm": 2.3527793884277344,
+      "learning_rate": 1.9966912812559733e-05,
+      "loss": 0.7921,
+      "step": 122
+    },
+    {
+      "epoch": 0.3732574679943101,
+      "grad_norm": 2.2694365978240967,
+      "learning_rate": 1.9963974885425267e-05,
+      "loss": 0.7816,
+      "step": 123
+    },
+    {
+      "epoch": 0.37629208155523947,
+      "grad_norm": 2.436676025390625,
+      "learning_rate": 1.9960912246939486e-05,
+      "loss": 0.7782,
+      "step": 124
+    },
+    {
+      "epoch": 0.3793266951161688,
+      "grad_norm": 2.3587653636932373,
+      "learning_rate": 1.9957724935435065e-05,
+      "loss": 0.8024,
+      "step": 125
+    },
+    {
+      "epoch": 0.38236130867709817,
+      "grad_norm": 2.3145172595977783,
+      "learning_rate": 1.9954412990805107e-05,
+      "loss": 0.8115,
+      "step": 126
+    },
+    {
+      "epoch": 0.3853959222380275,
+      "grad_norm": 2.271946430206299,
+      "learning_rate": 1.995097645450266e-05,
+      "loss": 0.7975,
+      "step": 127
+    },
+    {
+      "epoch": 0.38843053579895687,
+      "grad_norm": 2.2533860206604004,
+      "learning_rate": 1.994741536954019e-05,
+      "loss": 0.8187,
+      "step": 128
+    },
+    {
+      "epoch": 0.3914651493598862,
+      "grad_norm": 2.5504581928253174,
+      "learning_rate": 1.994372978048903e-05,
+      "loss": 0.7913,
+      "step": 129
+    },
+    {
+      "epoch": 0.39449976292081557,
+      "grad_norm": 2.3467888832092285,
+      "learning_rate": 1.993991973347884e-05,
+      "loss": 0.7955,
+      "step": 130
+    },
+    {
+      "epoch": 0.3975343764817449,
+      "grad_norm": 2.356933832168579,
+      "learning_rate": 1.9935985276197033e-05,
+      "loss": 0.808,
+      "step": 131
+    },
+    {
+      "epoch": 0.40056899004267427,
+      "grad_norm": 2.608546257019043,
+      "learning_rate": 1.9931926457888155e-05,
+      "loss": 0.785,
+      "step": 132
+    },
+    {
+      "epoch": 0.4036036036036036,
+      "grad_norm": 2.2730495929718018,
+      "learning_rate": 1.9927743329353295e-05,
+      "loss": 0.79,
+      "step": 133
+    },
+    {
+      "epoch": 0.40663821716453297,
+      "grad_norm": 2.2720224857330322,
+      "learning_rate": 1.992343594294943e-05,
+      "loss": 0.8084,
+      "step": 134
+    },
+    {
+      "epoch": 0.4096728307254623,
+      "grad_norm": 2.2325122356414795,
+      "learning_rate": 1.9919004352588768e-05,
+      "loss": 0.8008,
+      "step": 135
+    },
+    {
+      "epoch": 0.41270744428639167,
+      "grad_norm": 2.4259414672851562,
+      "learning_rate": 1.9914448613738107e-05,
+      "loss": 0.7827,
+      "step": 136
+    },
+    {
+      "epoch": 0.415742057847321,
+      "grad_norm": 2.2663819789886475,
+      "learning_rate": 1.9909768783418086e-05,
+      "loss": 0.8059,
+      "step": 137
+    },
+    {
+      "epoch": 0.41877667140825037,
+      "grad_norm": 2.910830020904541,
+      "learning_rate": 1.990496492020252e-05,
+      "loss": 0.8159,
+      "step": 138
+    },
+    {
+      "epoch": 0.4218112849691797,
+      "grad_norm": 2.4485902786254883,
+      "learning_rate": 1.9900037084217637e-05,
+      "loss": 0.7921,
+      "step": 139
+    },
+    {
+      "epoch": 0.42484589853010907,
+      "grad_norm": 2.3199424743652344,
+      "learning_rate": 1.989498533714135e-05,
+      "loss": 0.8006,
+      "step": 140
+    },
+    {
+      "epoch": 0.4278805120910384,
+      "grad_norm": 2.529900550842285,
+      "learning_rate": 1.9889809742202454e-05,
+      "loss": 0.7805,
+      "step": 141
+    },
+    {
+      "epoch": 0.43091512565196777,
+      "grad_norm": 2.5135438442230225,
+      "learning_rate": 1.988451036417986e-05,
+      "loss": 0.8088,
+      "step": 142
+    },
+    {
+      "epoch": 0.4339497392128971,
+      "grad_norm": 2.359200954437256,
+      "learning_rate": 1.9879087269401782e-05,
+      "loss": 0.7963,
+      "step": 143
+    },
+    {
+      "epoch": 0.43698435277382647,
+      "grad_norm": 2.307164192199707,
+      "learning_rate": 1.9873540525744888e-05,
+      "loss": 0.79,
+      "step": 144
+    },
+    {
+      "epoch": 0.4400189663347558,
+      "grad_norm": 2.341834545135498,
+      "learning_rate": 1.986787020263347e-05,
+      "loss": 0.7955,
+      "step": 145
+    },
+    {
+      "epoch": 0.44305357989568517,
+      "grad_norm": 2.409449338912964,
+      "learning_rate": 1.986207637103857e-05,
+      "loss": 0.7761,
+      "step": 146
+    },
+    {
+      "epoch": 0.4460881934566145,
+      "grad_norm": 2.3119232654571533,
+      "learning_rate": 1.9856159103477085e-05,
+      "loss": 0.7992,
+      "step": 147
+    },
+    {
+      "epoch": 0.44912280701754387,
+      "grad_norm": 2.3376715183258057,
+      "learning_rate": 1.9850118474010873e-05,
+      "loss": 0.784,
+      "step": 148
+    },
+    {
+      "epoch": 0.4521574205784732,
+      "grad_norm": 2.387392997741699,
+      "learning_rate": 1.98439545582458e-05,
+      "loss": 0.7806,
+      "step": 149
+    },
+    {
+      "epoch": 0.45519203413940257,
+      "grad_norm": 2.2223150730133057,
+      "learning_rate": 1.983766743333084e-05,
+      "loss": 0.7914,
+      "step": 150
+    },
+    {
+      "epoch": 0.45519203413940257,
+      "eval_loss": NaN,
+      "eval_runtime": 205.0592,
+      "eval_samples_per_second": 20.277,
+      "eval_steps_per_second": 5.072,
+      "step": 150
+    },
+    {
+      "epoch": 0.4582266477003319,
+      "grad_norm": 2.727046012878418,
+      "learning_rate": 1.9831257177957045e-05,
+      "loss": 0.793,
+      "step": 151
+    },
+    {
+      "epoch": 0.46126126126126127,
+      "grad_norm": 2.266788959503174,
+      "learning_rate": 1.9824723872356623e-05,
+      "loss": 0.8127,
+      "step": 152
+    },
+    {
+      "epoch": 0.4642958748221906,
+      "grad_norm": 2.530904531478882,
+      "learning_rate": 1.9818067598301894e-05,
+      "loss": 0.7973,
+      "step": 153
+    },
+    {
+      "epoch": 0.46733048838311997,
+      "grad_norm": 2.3305609226226807,
+      "learning_rate": 1.981128843910428e-05,
+      "loss": 0.7961,
+      "step": 154
+    },
+    {
+      "epoch": 0.4703651019440493,
+      "grad_norm": 2.1240079402923584,
+      "learning_rate": 1.9804386479613268e-05,
+      "loss": 0.8093,
+      "step": 155
+    },
+    {
+      "epoch": 0.47339971550497867,
+      "grad_norm": 2.625185489654541,
+      "learning_rate": 1.9797361806215335e-05,
+      "loss": 0.7988,
+      "step": 156
+    },
+    {
+      "epoch": 0.476434329065908,
+      "grad_norm": 2.4407734870910645,
+      "learning_rate": 1.9790214506832868e-05,
+      "loss": 0.8166,
+      "step": 157
+    },
+    {
+      "epoch": 0.47946894262683737,
+      "grad_norm": 2.4013476371765137,
+      "learning_rate": 1.9782944670923075e-05,
+      "loss": 0.7935,
+      "step": 158
+    },
+    {
+      "epoch": 0.4825035561877667,
+      "grad_norm": 2.357010841369629,
+      "learning_rate": 1.9775552389476865e-05,
+      "loss": 0.7818,
+      "step": 159
+    },
+    {
+      "epoch": 0.48553816974869607,
+      "grad_norm": 2.1862401962280273,
+      "learning_rate": 1.9768037755017687e-05,
+      "loss": 0.7879,
+      "step": 160
+    },
+    {
+      "epoch": 0.4885727833096254,
+      "grad_norm": 2.3130927085876465,
+      "learning_rate": 1.97604008616004e-05,
+      "loss": 0.7914,
+      "step": 161
+    },
+    {
+      "epoch": 0.49160739687055477,
+      "grad_norm": 2.0661509037017822,
+      "learning_rate": 1.9752641804810083e-05,
+      "loss": 0.8023,
+      "step": 162
+    },
+    {
+      "epoch": 0.4946420104314841,
+      "grad_norm": 2.2117955684661865,
+      "learning_rate": 1.9744760681760832e-05,
+      "loss": 0.7972,
+      "step": 163
+    },
+    {
+      "epoch": 0.49767662399241347,
+      "grad_norm": 2.603163242340088,
+      "learning_rate": 1.973675759109456e-05,
+      "loss": 0.7913,
+      "step": 164
+    },
+    {
+      "epoch": 0.5007112375533428,
+      "grad_norm": 2.2511062622070312,
+      "learning_rate": 1.9728632632979746e-05,
+      "loss": 0.7914,
+      "step": 165
+    },
+    {
+      "epoch": 0.5037458511142722,
+      "grad_norm": 2.375213861465454,
+      "learning_rate": 1.9720385909110197e-05,
+      "loss": 0.7928,
+      "step": 166
+    },
+    {
+      "epoch": 0.5067804646752015,
+      "grad_norm": 2.5348660945892334,
+      "learning_rate": 1.9712017522703764e-05,
+      "loss": 0.7894,
+      "step": 167
+    },
+    {
+      "epoch": 0.5098150782361308,
+      "grad_norm": 2.2094035148620605,
+      "learning_rate": 1.9703527578501052e-05,
+      "loss": 0.7813,
+      "step": 168
+    },
+    {
+      "epoch": 0.5128496917970602,
+      "grad_norm": 2.2283380031585693,
+      "learning_rate": 1.9694916182764113e-05,
+      "loss": 0.7877,
+      "step": 169
+    },
+    {
+      "epoch": 0.5158843053579896,
+      "grad_norm": 2.189119338989258,
+      "learning_rate": 1.9686183443275118e-05,
+      "loss": 0.7989,
+      "step": 170
+    },
+    {
+      "epoch": 0.518918918918919,
+      "grad_norm": 2.2636640071868896,
+      "learning_rate": 1.967732946933499e-05,
+      "loss": 0.8059,
+      "step": 171
+    },
+    {
+      "epoch": 0.5219535324798482,
+      "grad_norm": 2.243251085281372,
+      "learning_rate": 1.9668354371762066e-05,
+      "loss": 0.7904,
+      "step": 172
+    },
+    {
+      "epoch": 0.5249881460407776,
+      "grad_norm": 2.215536117553711,
+      "learning_rate": 1.9659258262890683e-05,
+      "loss": 0.7912,
+      "step": 173
+    },
+    {
+      "epoch": 0.528022759601707,
+      "grad_norm": 2.2998199462890625,
+      "learning_rate": 1.9650041256569792e-05,
+      "loss": 0.797,
+      "step": 174
+    },
+    {
+      "epoch": 0.5310573731626363,
+      "grad_norm": 2.3586950302124023,
+      "learning_rate": 1.9640703468161508e-05,
+      "loss": 0.7907,
+      "step": 175
+    },
+    {
+      "epoch": 0.5340919867235657,
+      "grad_norm": 2.257404088973999,
+      "learning_rate": 1.96312450145397e-05,
+      "loss": 0.7977,
+      "step": 176
+    },
+    {
+      "epoch": 0.537126600284495,
+      "grad_norm": 2.4808526039123535,
+      "learning_rate": 1.9621666014088495e-05,
+      "loss": 0.7929,
+      "step": 177
+    },
+    {
+      "epoch": 0.5401612138454244,
+      "grad_norm": 2.208704710006714,
+      "learning_rate": 1.9611966586700825e-05,
+      "loss": 0.7975,
+      "step": 178
+    },
+    {
+      "epoch": 0.5431958274063537,
+      "grad_norm": 2.4615161418914795,
+      "learning_rate": 1.9602146853776894e-05,
+      "loss": 0.7991,
+      "step": 179
+    },
+    {
+      "epoch": 0.5462304409672831,
+      "grad_norm": 2.4766757488250732,
+      "learning_rate": 1.9592206938222703e-05,
+      "loss": 0.7911,
+      "step": 180
+    },
+    {
+      "epoch": 0.5492650545282124,
+      "grad_norm": 2.2009432315826416,
+      "learning_rate": 1.9582146964448457e-05,
+      "loss": 0.788,
+      "step": 181
+    },
+    {
+      "epoch": 0.5522996680891418,
+      "grad_norm": 2.258129358291626,
+      "learning_rate": 1.9571967058367067e-05,
+      "loss": 0.7893,
+      "step": 182
+    },
+    {
+      "epoch": 0.5553342816500711,
+      "grad_norm": 2.1914985179901123,
+      "learning_rate": 1.956166734739251e-05,
+      "loss": 0.8057,
+      "step": 183
+    },
+    {
+      "epoch": 0.5583688952110005,
+      "grad_norm": 2.2815279960632324,
+      "learning_rate": 1.9551247960438298e-05,
+      "loss": 0.7823,
+      "step": 184
+    },
+    {
+      "epoch": 0.5614035087719298,
+      "grad_norm": 2.2393579483032227,
+      "learning_rate": 1.954070902791582e-05,
+      "loss": 0.7899,
+      "step": 185
+    },
+    {
+      "epoch": 0.5644381223328592,
+      "grad_norm": 2.56640625,
+      "learning_rate": 1.953005068173272e-05,
+      "loss": 0.7731,
+      "step": 186
+    },
+    {
+      "epoch": 0.5674727358937885,
+      "grad_norm": 2.3896234035491943,
+      "learning_rate": 1.9519273055291266e-05,
+      "loss": 0.7936,
+      "step": 187
+    },
+    {
+      "epoch": 0.5705073494547179,
+      "grad_norm": 2.21494722366333,
+      "learning_rate": 1.9508376283486653e-05,
+      "loss": 0.8121,
+      "step": 188
+    },
+    {
+      "epoch": 0.5735419630156472,
+      "grad_norm": 2.400538444519043,
+      "learning_rate": 1.949736050270532e-05,
+      "loss": 0.7938,
+      "step": 189
+    },
+    {
+      "epoch": 0.5765765765765766,
+      "grad_norm": 2.1337621212005615,
+      "learning_rate": 1.9486225850823265e-05,
+      "loss": 0.8049,
+      "step": 190
+    },
+    {
+      "epoch": 0.579611190137506,
+      "grad_norm": 2.314168930053711,
+      "learning_rate": 1.9474972467204298e-05,
+      "loss": 0.8109,
+      "step": 191
+    },
+    {
+      "epoch": 0.5826458036984353,
+      "grad_norm": 2.2364132404327393,
+      "learning_rate": 1.9463600492698297e-05,
+      "loss": 0.78,
+      "step": 192
+    },
+    {
+      "epoch": 0.5856804172593646,
+      "grad_norm": 2.204206705093384,
+      "learning_rate": 1.945211006963945e-05,
+      "loss": 0.8187,
+      "step": 193
+    },
+    {
+      "epoch": 0.588715030820294,
+      "grad_norm": 2.1865625381469727,
+      "learning_rate": 1.9440501341844484e-05,
+      "loss": 0.7844,
+      "step": 194
+    },
+    {
+      "epoch": 0.5917496443812233,
+      "grad_norm": 2.143092155456543,
+      "learning_rate": 1.9428774454610845e-05,
+      "loss": 0.7967,
+      "step": 195
+    },
+    {
+      "epoch": 0.5947842579421527,
+      "grad_norm": 2.3440985679626465,
+      "learning_rate": 1.9416929554714887e-05,
+      "loss": 0.8096,
+      "step": 196
+    },
+    {
+      "epoch": 0.597818871503082,
+      "grad_norm": 2.43977689743042,
+      "learning_rate": 1.9404966790410047e-05,
+      "loss": 0.795,
+      "step": 197
+    },
+    {
+      "epoch": 0.6008534850640114,
+      "grad_norm": 2.2740390300750732,
+      "learning_rate": 1.9392886311424975e-05,
+      "loss": 0.7937,
+      "step": 198
+    },
+    {
+      "epoch": 0.6038880986249408,
+      "grad_norm": 2.1775126457214355,
+      "learning_rate": 1.938068826896166e-05,
+      "loss": 0.8019,
+      "step": 199
+    },
+    {
+      "epoch": 0.6069227121858701,
+      "grad_norm": 2.1858487129211426,
+      "learning_rate": 1.9368372815693547e-05,
+      "loss": 0.7962,
+      "step": 200
+    },
+    {
+      "epoch": 0.6069227121858701,
+      "eval_loss": NaN,
+      "eval_runtime": 204.6421,
+      "eval_samples_per_second": 20.318,
+      "eval_steps_per_second": 5.082,
+      "step": 200
+    },
+    {
+      "epoch": 0.6099573257467994,
+      "grad_norm": 2.2958755493164062,
+      "learning_rate": 1.9355940105763622e-05,
+      "loss": 0.8003,
+      "step": 201
+    },
+    {
+      "epoch": 0.6129919393077288,
+      "grad_norm": 2.1330931186676025,
+      "learning_rate": 1.934339029478248e-05,
+      "loss": 0.7803,
+      "step": 202
+    },
+    {
+      "epoch": 0.6160265528686582,
+      "grad_norm": 2.3881402015686035,
+      "learning_rate": 1.9330723539826373e-05,
+      "loss": 0.8044,
+      "step": 203
+    },
+    {
+      "epoch": 0.6190611664295875,
+      "grad_norm": 2.2404513359069824,
+      "learning_rate": 1.9317939999435262e-05,
+      "loss": 0.8097,
+      "step": 204
+    },
+    {
+      "epoch": 0.6220957799905168,
+      "grad_norm": 2.194645404815674,
+      "learning_rate": 1.930503983361081e-05,
+      "loss": 0.8178,
+      "step": 205
+    },
+    {
+      "epoch": 0.6251303935514462,
+      "grad_norm": 2.509723424911499,
+      "learning_rate": 1.92920232038144e-05,
+      "loss": 0.7936,
+      "step": 206
+    },
+    {
+      "epoch": 0.6281650071123756,
+      "grad_norm": 2.2457869052886963,
+      "learning_rate": 1.9278890272965097e-05,
+      "loss": 0.8162,
+      "step": 207
+    },
+    {
+      "epoch": 0.6311996206733049,
+      "grad_norm": 2.2989683151245117,
+      "learning_rate": 1.9265641205437612e-05,
+      "loss": 0.8012,
+      "step": 208
+    },
+    {
+      "epoch": 0.6342342342342342,
+      "grad_norm": 2.3188092708587646,
+      "learning_rate": 1.925227616706026e-05,
+      "loss": 0.8094,
+      "step": 209
+    },
+    {
+      "epoch": 0.6372688477951636,
+      "grad_norm": 2.2978076934814453,
+      "learning_rate": 1.9238795325112867e-05,
+      "loss": 0.7811,
+      "step": 210
+    },
+    {
+      "epoch": 0.640303461356093,
+      "grad_norm": 2.331542730331421,
+      "learning_rate": 1.9225198848324687e-05,
+      "loss": 0.7946,
+      "step": 211
+    },
+    {
+      "epoch": 0.6433380749170223,
+      "grad_norm": 2.1989738941192627,
+      "learning_rate": 1.921148690687228e-05,
+      "loss": 0.7983,
+      "step": 212
+    },
+    {
+      "epoch": 0.6463726884779516,
+      "grad_norm": 2.4261419773101807,
+      "learning_rate": 1.9197659672377388e-05,
+      "loss": 0.8134,
+      "step": 213
+    },
+    {
+      "epoch": 0.649407302038881,
+      "grad_norm": 2.3790082931518555,
+      "learning_rate": 1.918371731790479e-05,
+      "loss": 0.79,
+      "step": 214
+    },
+    {
+      "epoch": 0.6524419155998104,
+      "grad_norm": 2.2966949939727783,
+      "learning_rate": 1.9169660017960135e-05,
+      "loss": 0.8029,
+      "step": 215
+    },
+    {
+      "epoch": 0.6554765291607397,
+      "grad_norm": 2.5911426544189453,
+      "learning_rate": 1.915548794848775e-05,
+      "loss": 0.8118,
+      "step": 216
+    },
+    {
+      "epoch": 0.658511142721669,
+      "grad_norm": 2.3282856941223145,
+      "learning_rate": 1.9141201286868435e-05,
+      "loss": 0.8092,
+      "step": 217
+    },
+    {
+      "epoch": 0.6615457562825984,
+      "grad_norm": 2.415398359298706,
+      "learning_rate": 1.9126800211917277e-05,
+      "loss": 0.8156,
+      "step": 218
+    },
+    {
+      "epoch": 0.6645803698435278,
+      "grad_norm": 2.2823410034179688,
+      "learning_rate": 1.911228490388136e-05,
+      "loss": 0.8004,
+      "step": 219
+    },
+    {
+      "epoch": 0.6676149834044571,
+      "grad_norm": 2.2104527950286865,
+      "learning_rate": 1.9097655544437544e-05,
+      "loss": 0.8023,
+      "step": 220
+    },
+    {
+      "epoch": 0.6706495969653864,
+      "grad_norm": 2.3354063034057617,
+      "learning_rate": 1.908291231669019e-05,
+      "loss": 0.8117,
+      "step": 221
+    },
+    {
+      "epoch": 0.6736842105263158,
+      "grad_norm": 2.230656147003174,
+      "learning_rate": 1.906805540516885e-05,
+      "loss": 0.7797,
+      "step": 222
+    },
+    {
+      "epoch": 0.6767188240872452,
+      "grad_norm": 2.464111328125,
+      "learning_rate": 1.905308499582597e-05,
+      "loss": 0.7929,
+      "step": 223
+    },
+    {
+      "epoch": 0.6797534376481745,
+      "grad_norm": 2.188788652420044,
+      "learning_rate": 1.903800127603456e-05,
+      "loss": 0.7971,
+      "step": 224
+    },
+    {
+      "epoch": 0.6827880512091038,
+      "grad_norm": 2.202427387237549,
+      "learning_rate": 1.9022804434585854e-05,
+      "loss": 0.8026,
+      "step": 225
+    },
+    {
+      "epoch": 0.6858226647700332,
+      "grad_norm": 2.2621190547943115,
+      "learning_rate": 1.9007494661686937e-05,
+      "loss": 0.8112,
+      "step": 226
+    },
+    {
+      "epoch": 0.6888572783309626,
+      "grad_norm": 2.333603620529175,
+      "learning_rate": 1.8992072148958368e-05,
+      "loss": 0.7937,
+      "step": 227
+    },
+    {
+      "epoch": 0.6918918918918919,
+      "grad_norm": 2.626451253890991,
+      "learning_rate": 1.8976537089431793e-05,
+      "loss": 0.8005,
+      "step": 228
+    },
+    {
+      "epoch": 0.6949265054528212,
+      "grad_norm": 2.4107227325439453,
+      "learning_rate": 1.8960889677547506e-05,
+      "loss": 0.7813,
+      "step": 229
+    },
+    {
+      "epoch": 0.6979611190137506,
+      "grad_norm": 2.7147607803344727,
+      "learning_rate": 1.8945130109152035e-05,
+      "loss": 0.8036,
+      "step": 230
+    },
+    {
+      "epoch": 0.70099573257468,
+      "grad_norm": 2.281703233718872,
+      "learning_rate": 1.8929258581495688e-05,
+      "loss": 0.7946,
+      "step": 231
+    },
+    {
+      "epoch": 0.7040303461356093,
+      "grad_norm": 3.016942262649536,
+      "learning_rate": 1.891327529323007e-05,
+      "loss": 0.7786,
+      "step": 232
+    },
+    {
+      "epoch": 0.7070649596965386,
+      "grad_norm": 2.5729317665100098,
+      "learning_rate": 1.8897180444405615e-05,
+      "loss": 0.8141,
+      "step": 233
+    },
+    {
+      "epoch": 0.710099573257468,
+      "grad_norm": 2.75722336769104,
+      "learning_rate": 1.888097423646907e-05,
+      "loss": 0.8079,
+      "step": 234
+    },
+    {
+      "epoch": 0.7131341868183974,
+      "grad_norm": 2.843980073928833,
+      "learning_rate": 1.8864656872260985e-05,
+      "loss": 0.795,
+      "step": 235
+    },
+    {
+      "epoch": 0.7161688003793267,
+      "grad_norm": 3.3999879360198975,
+      "learning_rate": 1.884822855601316e-05,
+      "loss": 0.8086,
+      "step": 236
+    },
+    {
+      "epoch": 0.719203413940256,
+      "grad_norm": 3.1997487545013428,
+      "learning_rate": 1.8831689493346095e-05,
+      "loss": 0.7919,
+      "step": 237
+    },
+    {
+      "epoch": 0.7222380275011854,
+      "grad_norm": 2.4851367473602295,
+      "learning_rate": 1.881503989126642e-05,
+      "loss": 0.7914,
+      "step": 238
+    },
+    {
+      "epoch": 0.7252726410621148,
+      "grad_norm": 2.738428831100464,
+      "learning_rate": 1.8798279958164295e-05,
+      "loss": 0.7929,
+      "step": 239
+    },
+    {
+      "epoch": 0.7283072546230441,
+      "grad_norm": 2.1467783451080322,
+      "learning_rate": 1.8781409903810823e-05,
+      "loss": 0.815,
+      "step": 240
+    },
+    {
+      "epoch": 0.7313418681839734,
+      "grad_norm": 16.773521423339844,
+      "learning_rate": 1.8764429939355394e-05,
+      "loss": 0.8014,
+      "step": 241
+    },
+    {
+      "epoch": 0.7343764817449028,
+      "grad_norm": 3.171633720397949,
+      "learning_rate": 1.874734027732306e-05,
+      "loss": 0.805,
+      "step": 242
+    },
+    {
+      "epoch": 0.7374110953058322,
+      "grad_norm": 2.249514102935791,
+      "learning_rate": 1.8730141131611882e-05,
+      "loss": 0.8109,
+      "step": 243
+    },
+    {
+      "epoch": 0.7404457088667615,
+      "grad_norm": 2.816488742828369,
+      "learning_rate": 1.8712832717490238e-05,
+      "loss": 0.8044,
+      "step": 244
+    },
+    {
+      "epoch": 0.7434803224276908,
+      "grad_norm": 2.558295488357544,
+      "learning_rate": 1.8695415251594123e-05,
+      "loss": 0.8269,
+      "step": 245
+    },
+    {
+      "epoch": 0.7465149359886202,
+      "grad_norm": 2.5777034759521484,
+      "learning_rate": 1.8677888951924473e-05,
+      "loss": 0.7971,
+      "step": 246
+    },
+    {
+      "epoch": 0.7495495495495496,
+      "grad_norm": 2.394287586212158,
+      "learning_rate": 1.866025403784439e-05,
+      "loss": 0.8092,
+      "step": 247
+    },
+    {
+      "epoch": 0.7525841631104789,
+      "grad_norm": 2.2798614501953125,
+      "learning_rate": 1.864251073007642e-05,
+      "loss": 0.7964,
+      "step": 248
+    },
+    {
+      "epoch": 0.7556187766714082,
+      "grad_norm": 2.3587262630462646,
+      "learning_rate": 1.8624659250699807e-05,
+      "loss": 0.7928,
+      "step": 249
+    },
+    {
+      "epoch": 0.7586533902323376,
+      "grad_norm": 2.189763307571411,
+      "learning_rate": 1.8606699823147675e-05,
+      "loss": 0.7812,
+      "step": 250
+    },
+    {
+      "epoch": 0.7586533902323376,
+      "eval_loss": NaN,
+      "eval_runtime": 204.0596,
+      "eval_samples_per_second": 20.376,
+      "eval_steps_per_second": 5.097,
+      "step": 250
+    },
+    {
+      "epoch": 0.761688003793267,
+      "grad_norm": 2.433803081512451,
+      "learning_rate": 1.8588632672204264e-05,
+      "loss": 0.8111,
+      "step": 251
+    },
+    {
+      "epoch": 0.7647226173541963,
+      "grad_norm": 2.28267765045166,
+      "learning_rate": 1.8570458024002094e-05,
+      "loss": 0.8001,
+      "step": 252
+    },
+    {
+      "epoch": 0.7677572309151256,
+      "grad_norm": 2.3339545726776123,
+      "learning_rate": 1.8552176106019156e-05,
+      "loss": 0.8158,
+      "step": 253
+    },
+    {
+      "epoch": 0.770791844476055,
+      "grad_norm": 2.284759998321533,
+      "learning_rate": 1.8533787147076046e-05,
+      "loss": 0.7852,
+      "step": 254
+    },
+    {
+      "epoch": 0.7738264580369844,
+      "grad_norm": 2.35969614982605,
+      "learning_rate": 1.8515291377333114e-05,
+      "loss": 0.7909,
+      "step": 255
+    },
+    {
+      "epoch": 0.7768610715979137,
+      "grad_norm": 2.4525341987609863,
+      "learning_rate": 1.8496689028287572e-05,
+      "loss": 0.817,
+      "step": 256
+    },
+    {
+      "epoch": 0.779895685158843,
+      "grad_norm": 2.7126755714416504,
+      "learning_rate": 1.847798033277061e-05,
+      "loss": 0.797,
+      "step": 257
+    },
+    {
+      "epoch": 0.7829302987197724,
+      "grad_norm": 2.2292561531066895,
+      "learning_rate": 1.8459165524944463e-05,
+      "loss": 0.8044,
+      "step": 258
+    },
+    {
+      "epoch": 0.7859649122807018,
+      "grad_norm": 2.447347640991211,
+      "learning_rate": 1.8440244840299507e-05,
+      "loss": 0.7979,
+      "step": 259
+    },
+    {
+      "epoch": 0.7889995258416311,
+      "grad_norm": 2.2336087226867676,
+      "learning_rate": 1.842121851565128e-05,
+      "loss": 0.8036,
+      "step": 260
+    },
+    {
+      "epoch": 0.7920341394025604,
+      "grad_norm": 2.3720502853393555,
+      "learning_rate": 1.8402086789137547e-05,
+      "loss": 0.7979,
+      "step": 261
+    },
+    {
+      "epoch": 0.7950687529634898,
+      "grad_norm": 2.1780807971954346,
+      "learning_rate": 1.8382849900215297e-05,
+      "loss": 0.7876,
+      "step": 262
+    },
+    {
+      "epoch": 0.7981033665244192,
+      "grad_norm": 2.3325858116149902,
+      "learning_rate": 1.8363508089657763e-05,
+      "loss": 0.7997,
+      "step": 263
+    },
+    {
+      "epoch": 0.8011379800853485,
+      "grad_norm": 2.3341164588928223,
+      "learning_rate": 1.8344061599551397e-05,
+      "loss": 0.7844,
+      "step": 264
+    },
+    {
+      "epoch": 0.8041725936462779,
+      "grad_norm": 2.4678280353546143,
+      "learning_rate": 1.8324510673292844e-05,
+      "loss": 0.7946,
+      "step": 265
+    },
+    {
+      "epoch": 0.8072072072072072,
+      "grad_norm": 2.424893856048584,
+      "learning_rate": 1.8304855555585893e-05,
+      "loss": 0.7916,
+      "step": 266
+    },
+    {
+      "epoch": 0.8102418207681366,
+      "grad_norm": 2.5433976650238037,
+      "learning_rate": 1.8285096492438424e-05,
+      "loss": 0.7983,
+      "step": 267
+    },
+    {
+      "epoch": 0.8132764343290659,
+      "grad_norm": 2.499178647994995,
+      "learning_rate": 1.826523373115931e-05,
+      "loss": 0.7944,
+      "step": 268
+    },
+    {
+      "epoch": 0.8163110478899953,
+      "grad_norm": 2.2453994750976562,
+      "learning_rate": 1.8245267520355348e-05,
+      "loss": 0.8148,
+      "step": 269
+    },
+    {
+      "epoch": 0.8193456614509246,
+      "grad_norm": 2.350146770477295,
+      "learning_rate": 1.8225198109928116e-05,
+      "loss": 0.7986,
+      "step": 270
+    },
+    {
+      "epoch": 0.822380275011854,
+      "grad_norm": 2.2048559188842773,
+      "learning_rate": 1.8205025751070878e-05,
+      "loss": 0.8093,
+      "step": 271
+    },
+    {
+      "epoch": 0.8254148885727833,
+      "grad_norm": 2.2730185985565186,
+      "learning_rate": 1.8184750696265408e-05,
+      "loss": 0.7787,
+      "step": 272
+    },
+    {
+      "epoch": 0.8284495021337127,
+      "grad_norm": 2.423301935195923,
+      "learning_rate": 1.8164373199278858e-05,
+      "loss": 0.823,
+      "step": 273
+    },
+    {
+      "epoch": 0.831484115694642,
+      "grad_norm": 2.309649705886841,
+      "learning_rate": 1.8143893515160565e-05,
+      "loss": 0.7901,
+      "step": 274
+    },
+    {
+      "epoch": 0.8345187292555714,
+      "grad_norm": 2.249284267425537,
+      "learning_rate": 1.812331190023886e-05,
+      "loss": 0.8095,
+      "step": 275
+    },
+    {
+      "epoch": 0.8375533428165007,
+      "grad_norm": 2.2063703536987305,
+      "learning_rate": 1.8102628612117868e-05,
+      "loss": 0.8008,
+      "step": 276
+    },
+    {
+      "epoch": 0.8405879563774301,
+      "grad_norm": 2.2518839836120605,
+      "learning_rate": 1.8081843909674277e-05,
+      "loss": 0.8051,
+      "step": 277
+    },
+    {
+      "epoch": 0.8436225699383594,
+      "grad_norm": 2.226356267929077,
+      "learning_rate": 1.8060958053054095e-05,
+      "loss": 0.8036,
+      "step": 278
+    },
+    {
+      "epoch": 0.8466571834992888,
+      "grad_norm": 2.186485767364502,
+      "learning_rate": 1.8039971303669407e-05,
+      "loss": 0.8025,
+      "step": 279
+    },
+    {
+      "epoch": 0.8496917970602181,
+      "grad_norm": 2.4235646724700928,
+      "learning_rate": 1.8018883924195085e-05,
+      "loss": 0.7799,
+      "step": 280
+    },
+    {
+      "epoch": 0.8527264106211475,
+      "grad_norm": 2.2470943927764893,
+      "learning_rate": 1.799769617856552e-05,
+      "loss": 0.8025,
+      "step": 281
+    },
+    {
+      "epoch": 0.8557610241820768,
+      "grad_norm": 2.2514889240264893,
+      "learning_rate": 1.79764083319713e-05,
+      "loss": 0.7978,
+      "step": 282
+    },
+    {
+      "epoch": 0.8587956377430062,
+      "grad_norm": 2.220952033996582,
+      "learning_rate": 1.79550206508559e-05,
+      "loss": 0.7978,
+      "step": 283
+    },
+    {
+      "epoch": 0.8618302513039355,
+      "grad_norm": 2.4699270725250244,
+      "learning_rate": 1.7933533402912354e-05,
+      "loss": 0.7835,
+      "step": 284
+    },
+    {
+      "epoch": 0.8648648648648649,
+      "grad_norm": 2.3011207580566406,
+      "learning_rate": 1.7911946857079886e-05,
+      "loss": 0.8008,
+      "step": 285
+    },
+    {
+      "epoch": 0.8678994784257942,
+      "grad_norm": 2.2239327430725098,
+      "learning_rate": 1.7890261283540563e-05,
+      "loss": 0.7904,
+      "step": 286
+    },
+    {
+      "epoch": 0.8709340919867236,
+      "grad_norm": 2.077845573425293,
+      "learning_rate": 1.78684769537159e-05,
+      "loss": 0.7962,
+      "step": 287
+    },
+    {
+      "epoch": 0.8739687055476529,
+      "grad_norm": 2.2492687702178955,
+      "learning_rate": 1.7846594140263475e-05,
+      "loss": 0.8076,
+      "step": 288
+    },
+    {
+      "epoch": 0.8770033191085823,
+      "grad_norm": 2.1000773906707764,
+      "learning_rate": 1.78246131170735e-05,
+      "loss": 0.8107,
+      "step": 289
+    },
+    {
+      "epoch": 0.8800379326695116,
+      "grad_norm": 2.2440242767333984,
+      "learning_rate": 1.7802534159265407e-05,
+      "loss": 0.784,
+      "step": 290
+    },
+    {
+      "epoch": 0.883072546230441,
+      "grad_norm": 2.1370065212249756,
+      "learning_rate": 1.7780357543184396e-05,
+      "loss": 0.7926,
+      "step": 291
+    },
+    {
+      "epoch": 0.8861071597913703,
+      "grad_norm": 2.2182703018188477,
+      "learning_rate": 1.775808354639799e-05,
+      "loss": 0.8079,
+      "step": 292
+    },
+    {
+      "epoch": 0.8891417733522997,
+      "grad_norm": 2.236370325088501,
+      "learning_rate": 1.773571244769254e-05,
+      "loss": 0.7851,
+      "step": 293
+    },
+    {
+      "epoch": 0.892176386913229,
+      "grad_norm": 2.216042995452881,
+      "learning_rate": 1.771324452706975e-05,
+      "loss": 0.798,
+      "step": 294
+    },
+    {
+      "epoch": 0.8952110004741584,
+      "grad_norm": 2.2804715633392334,
+      "learning_rate": 1.769068006574317e-05,
+      "loss": 0.7916,
+      "step": 295
+    },
+    {
+      "epoch": 0.8982456140350877,
+      "grad_norm": 2.188271999359131,
+      "learning_rate": 1.7668019346134674e-05,
+      "loss": 0.7993,
+      "step": 296
+    },
+    {
+      "epoch": 0.9012802275960171,
+      "grad_norm": 2.372596025466919,
+      "learning_rate": 1.7645262651870926e-05,
+      "loss": 0.816,
+      "step": 297
+    },
+    {
+      "epoch": 0.9043148411569464,
+      "grad_norm": 2.174302339553833,
+      "learning_rate": 1.7622410267779834e-05,
+      "loss": 0.8247,
+      "step": 298
+    },
+    {
+      "epoch": 0.9073494547178758,
+      "grad_norm": 2.0944302082061768,
+      "learning_rate": 1.7599462479886976e-05,
+      "loss": 0.7979,
+      "step": 299
+    },
+    {
+      "epoch": 0.9103840682788051,
+      "grad_norm": 2.118502140045166,
+      "learning_rate": 1.7576419575412028e-05,
+      "loss": 0.8007,
+      "step": 300
+    },
+    {
+      "epoch": 0.9103840682788051,
+      "eval_loss": NaN,
+      "eval_runtime": 203.6703,
+      "eval_samples_per_second": 20.415,
+      "eval_steps_per_second": 5.106,
+      "step": 300
+    },
+    {
+      "epoch": 0.9134186818397345,
+      "grad_norm": 2.2053334712982178,
+      "learning_rate": 1.755328184276517e-05,
+      "loss": 0.7913,
+      "step": 301
+    },
+    {
+      "epoch": 0.9164532954006638,
+      "grad_norm": 2.243788957595825,
+      "learning_rate": 1.7530049571543464e-05,
+      "loss": 0.801,
+      "step": 302
+    },
+    {
+      "epoch": 0.9194879089615932,
+      "grad_norm": 2.363306760787964,
+      "learning_rate": 1.7506723052527243e-05,
+      "loss": 0.8278,
+      "step": 303
+    },
+    {
+      "epoch": 0.9225225225225225,
+      "grad_norm": 2.2566967010498047,
+      "learning_rate": 1.7483302577676475e-05,
+      "loss": 0.7929,
+      "step": 304
+    },
+    {
+      "epoch": 0.9255571360834519,
+      "grad_norm": 2.1566100120544434,
+      "learning_rate": 1.7459788440127083e-05,
+      "loss": 0.7953,
+      "step": 305
+    },
+    {
+      "epoch": 0.9285917496443812,
+      "grad_norm": 2.279130697250366,
+      "learning_rate": 1.7436180934187307e-05,
+      "loss": 0.8125,
+      "step": 306
+    },
+    {
+      "epoch": 0.9316263632053106,
+      "grad_norm": 2.172891616821289,
+      "learning_rate": 1.7412480355334006e-05,
+      "loss": 0.8007,
+      "step": 307
+    },
+    {
+      "epoch": 0.9346609767662399,
+      "grad_norm": 2.2873098850250244,
+      "learning_rate": 1.738868700020895e-05,
+      "loss": 0.779,
+      "step": 308
+    },
+    {
+      "epoch": 0.9376955903271693,
+      "grad_norm": 2.3155357837677,
+      "learning_rate": 1.7364801166615124e-05,
+      "loss": 0.8025,
+      "step": 309
+    },
+    {
+      "epoch": 0.9407302038880986,
+      "grad_norm": 2.20151686668396,
+      "learning_rate": 1.7340823153513003e-05,
+      "loss": 0.8173,
+      "step": 310
+    },
+    {
+      "epoch": 0.943764817449028,
+      "grad_norm": 2.2033958435058594,
+      "learning_rate": 1.7316753261016782e-05,
+      "loss": 0.8095,
+      "step": 311
+    },
+    {
+      "epoch": 0.9467994310099573,
+      "grad_norm": 2.2805070877075195,
+      "learning_rate": 1.7292591790390668e-05,
+      "loss": 0.8139,
+      "step": 312
+    },
+    {
+      "epoch": 0.9498340445708867,
+      "grad_norm": 2.335238456726074,
+      "learning_rate": 1.7268339044045044e-05,
+      "loss": 0.7898,
+      "step": 313
+    },
+    {
+      "epoch": 0.952868658131816,
+      "grad_norm": 2.1889333724975586,
+      "learning_rate": 1.7243995325532755e-05,
+      "loss": 0.8002,
+      "step": 314
+    },
+    {
+      "epoch": 0.9559032716927454,
+      "grad_norm": 2.5007987022399902,
+      "learning_rate": 1.7219560939545246e-05,
+      "loss": 0.7892,
+      "step": 315
+    },
+    {
+      "epoch": 0.9589378852536747,
+      "grad_norm": 2.1342813968658447,
+      "learning_rate": 1.7195036191908798e-05,
+      "loss": 0.8028,
+      "step": 316
+    },
+    {
+      "epoch": 0.9619724988146041,
+      "grad_norm": 2.199307918548584,
+      "learning_rate": 1.7170421389580666e-05,
+      "loss": 0.793,
+      "step": 317
+    },
+    {
+      "epoch": 0.9650071123755334,
+      "grad_norm": 2.1955904960632324,
+      "learning_rate": 1.7145716840645253e-05,
+      "loss": 0.8085,
+      "step": 318
+    },
+    {
+      "epoch": 0.9680417259364628,
+      "grad_norm": 2.41741943359375,
+      "learning_rate": 1.712092285431026e-05,
+      "loss": 0.7964,
+      "step": 319
+    },
+    {
+      "epoch": 0.9710763394973921,
+      "grad_norm": 2.3595402240753174,
+      "learning_rate": 1.7096039740902782e-05,
+      "loss": 0.7999,
+      "step": 320
+    },
+    {
+      "epoch": 0.9741109530583215,
+      "grad_norm": 2.15049147605896,
+      "learning_rate": 1.7071067811865477e-05,
+      "loss": 0.7813,
+      "step": 321
+    },
+    {
+      "epoch": 0.9771455666192508,
+      "grad_norm": 2.240618944168091,
+      "learning_rate": 1.7046007379752624e-05,
+      "loss": 0.8038,
+      "step": 322
+    },
+    {
+      "epoch": 0.9801801801801802,
+      "grad_norm": 2.418973922729492,
+      "learning_rate": 1.702085875822623e-05,
+      "loss": 0.7672,
+      "step": 323
+    },
+    {
+      "epoch": 0.9832147937411095,
+      "grad_norm": 2.2924294471740723,
+      "learning_rate": 1.6995622262052093e-05,
+      "loss": 0.8013,
+      "step": 324
+    },
+    {
+      "epoch": 0.9862494073020389,
+      "grad_norm": 2.2135136127471924,
+      "learning_rate": 1.6970298207095887e-05,
+      "loss": 0.8112,
+      "step": 325
+    },
+    {
+      "epoch": 0.9892840208629682,
+      "grad_norm": 2.2720751762390137,
+      "learning_rate": 1.6944886910319173e-05,
+      "loss": 0.7896,
+      "step": 326
+    },
+    {
+      "epoch": 0.9923186344238976,
+      "grad_norm": 2.172165632247925,
+      "learning_rate": 1.6919388689775463e-05,
+      "loss": 0.8084,
+      "step": 327
+    },
+    {
+      "epoch": 0.9953532479848269,
+      "grad_norm": 2.1608591079711914,
+      "learning_rate": 1.6893803864606224e-05,
+      "loss": 0.7914,
+      "step": 328
+    },
+    {
+      "epoch": 0.9983878615457563,
+      "grad_norm": 2.153231143951416,
+      "learning_rate": 1.6868132755036875e-05,
+      "loss": 0.803,
+      "step": 329
+    },
+    {
+      "epoch": 1.0014224751066856,
+      "grad_norm": 2.242035150527954,
+      "learning_rate": 1.6842375682372803e-05,
+      "loss": 0.7964,
+      "step": 330
+    },
+    {
+      "epoch": 1.004457088667615,
+      "grad_norm": 2.490118980407715,
+      "learning_rate": 1.681653296899533e-05,
+      "loss": 0.7645,
+      "step": 331
+    },
+    {
+      "epoch": 1.0074917022285443,
+      "grad_norm": 2.258561372756958,
+      "learning_rate": 1.6790604938357664e-05,
+      "loss": 0.7579,
+      "step": 332
+    },
+    {
+      "epoch": 1.0105263157894737,
+      "grad_norm": 2.3750267028808594,
+      "learning_rate": 1.676459191498087e-05,
+      "loss": 0.7849,
+      "step": 333
+    },
+    {
+      "epoch": 1.013560929350403,
+      "grad_norm": 2.5088164806365967,
+      "learning_rate": 1.6738494224449802e-05,
+      "loss": 0.7814,
+      "step": 334
+    },
+    {
+      "epoch": 1.0165955429113325,
+      "grad_norm": 2.4220824241638184,
+      "learning_rate": 1.6712312193409032e-05,
+      "loss": 0.7774,
+      "step": 335
+    },
+    {
+      "epoch": 1.0196301564722616,
+      "grad_norm": 2.2030773162841797,
+      "learning_rate": 1.6686046149558736e-05,
+      "loss": 0.7495,
+      "step": 336
+    },
+    {
+      "epoch": 1.022664770033191,
+      "grad_norm": 2.336583375930786,
+      "learning_rate": 1.6659696421650645e-05,
+      "loss": 0.7508,
+      "step": 337
+    },
+    {
+      "epoch": 1.0256993835941204,
+      "grad_norm": 2.3112359046936035,
+      "learning_rate": 1.6633263339483867e-05,
+      "loss": 0.7525,
+      "step": 338
+    },
+    {
+      "epoch": 1.0287339971550498,
+      "grad_norm": 2.165022611618042,
+      "learning_rate": 1.6606747233900816e-05,
+      "loss": 0.787,
+      "step": 339
+    },
+    {
+      "epoch": 1.0317686107159791,
+      "grad_norm": 2.3545596599578857,
+      "learning_rate": 1.658014843678303e-05,
+      "loss": 0.7665,
+      "step": 340
+    },
+    {
+      "epoch": 1.0348032242769085,
+      "grad_norm": 2.231351137161255,
+      "learning_rate": 1.655346728104704e-05,
+      "loss": 0.7726,
+      "step": 341
+    },
+    {
+      "epoch": 1.037837837837838,
+      "grad_norm": 2.292428970336914,
+      "learning_rate": 1.652670410064019e-05,
+      "loss": 0.7722,
+      "step": 342
+    },
+    {
+      "epoch": 1.0408724513987673,
+      "grad_norm": 2.2555713653564453,
+      "learning_rate": 1.6499859230536468e-05,
+      "loss": 0.755,
+      "step": 343
+    },
+    {
+      "epoch": 1.0439070649596967,
+      "grad_norm": 2.5129449367523193,
+      "learning_rate": 1.647293300673231e-05,
+      "loss": 0.7736,
+      "step": 344
+    },
+    {
+      "epoch": 1.0469416785206258,
+      "grad_norm": 2.269122362136841,
+      "learning_rate": 1.6445925766242392e-05,
+      "loss": 0.7898,
+      "step": 345
+    },
+    {
+      "epoch": 1.0499762920815552,
+      "grad_norm": 2.21991229057312,
+      "learning_rate": 1.641883784709541e-05,
+      "loss": 0.7767,
+      "step": 346
+    },
+    {
+      "epoch": 1.0530109056424846,
+      "grad_norm": 2.324253797531128,
+      "learning_rate": 1.639166958832985e-05,
+      "loss": 0.7728,
+      "step": 347
+    },
+    {
+      "epoch": 1.056045519203414,
+      "grad_norm": 2.3205628395080566,
+      "learning_rate": 1.6364421329989758e-05,
+      "loss": 0.7845,
+      "step": 348
+    },
+    {
+      "epoch": 1.0590801327643433,
+      "grad_norm": 2.361678123474121,
+      "learning_rate": 1.6337093413120463e-05,
+      "loss": 0.7455,
+      "step": 349
+    },
+    {
+      "epoch": 1.0621147463252727,
+      "grad_norm": 2.3375606536865234,
+      "learning_rate": 1.6309686179764317e-05,
+      "loss": 0.7754,
+      "step": 350
+    },
+    {
+      "epoch": 1.0621147463252727,
+      "eval_loss": NaN,
+      "eval_runtime": 203.6306,
+      "eval_samples_per_second": 20.419,
+      "eval_steps_per_second": 5.107,
+      "step": 350
+    },
+    {
+      "epoch": 1.065149359886202,
+      "grad_norm": 2.3522422313690186,
+      "learning_rate": 1.6282199972956425e-05,
+      "loss": 0.7759,
+      "step": 351
+    },
+    {
+      "epoch": 1.0681839734471312,
+      "grad_norm": 2.4227213859558105,
+      "learning_rate": 1.6254635136720328e-05,
+      "loss": 0.7772,
+      "step": 352
+    },
+    {
+      "epoch": 1.0712185870080606,
+      "grad_norm": 2.235722303390503,
+      "learning_rate": 1.6226992016063726e-05,
+      "loss": 0.7694,
+      "step": 353
+    },
+    {
+      "epoch": 1.07425320056899,
+      "grad_norm": 2.2417314052581787,
+      "learning_rate": 1.6199270956974128e-05,
+      "loss": 0.7628,
+      "step": 354
+    },
+    {
+      "epoch": 1.0772878141299194,
+      "grad_norm": 2.348954200744629,
+      "learning_rate": 1.6171472306414554e-05,
+      "loss": 0.7656,
+      "step": 355
+    },
+    {
+      "epoch": 1.0803224276908487,
+      "grad_norm": 2.346963882446289,
+      "learning_rate": 1.614359641231916e-05,
+      "loss": 0.7839,
+      "step": 356
+    },
+    {
+      "epoch": 1.0833570412517781,
+      "grad_norm": 2.2969138622283936,
+      "learning_rate": 1.6115643623588915e-05,
+      "loss": 0.7728,
+      "step": 357
+    },
+    {
+      "epoch": 1.0863916548127075,
+      "grad_norm": 2.3338327407836914,
+      "learning_rate": 1.608761429008721e-05,
+      "loss": 0.7902,
+      "step": 358
+    },
+    {
+      "epoch": 1.0894262683736369,
+      "grad_norm": 2.2462401390075684,
+      "learning_rate": 1.6059508762635482e-05,
+      "loss": 0.7906,
+      "step": 359
+    },
+    {
+      "epoch": 1.0924608819345663,
+      "grad_norm": 2.2091758251190186,
+      "learning_rate": 1.6031327393008848e-05,
+      "loss": 0.7587,
+      "step": 360
+    },
+    {
+      "epoch": 1.0954954954954954,
+      "grad_norm": 2.2392489910125732,
+      "learning_rate": 1.6003070533931657e-05,
+      "loss": 0.7598,
+      "step": 361
+    },
+    {
+      "epoch": 1.0985301090564248,
+      "grad_norm": 2.193833112716675,
+      "learning_rate": 1.5974738539073125e-05,
+      "loss": 0.7622,
+      "step": 362
+    },
+    {
+      "epoch": 1.1015647226173542,
+      "grad_norm": 2.187610149383545,
+      "learning_rate": 1.594633176304287e-05,
+      "loss": 0.7796,
+      "step": 363
+    },
+    {
+      "epoch": 1.1045993361782835,
+      "grad_norm": 2.2753069400787354,
+      "learning_rate": 1.5917850561386487e-05,
+      "loss": 0.7783,
+      "step": 364
+    },
+    {
+      "epoch": 1.107633949739213,
+      "grad_norm": 2.2835614681243896,
+      "learning_rate": 1.588929529058111e-05,
+      "loss": 0.7801,
+      "step": 365
+    },
+    {
+      "epoch": 1.1106685633001423,
+      "grad_norm": 2.46498441696167,
+      "learning_rate": 1.5860666308030933e-05,
+      "loss": 0.7683,
+      "step": 366
+    },
+    {
+      "epoch": 1.1137031768610717,
+      "grad_norm": 2.361351490020752,
+      "learning_rate": 1.5831963972062734e-05,
+      "loss": 0.783,
+      "step": 367
+    },
+    {
+      "epoch": 1.1167377904220008,
+      "grad_norm": 2.2396347522735596,
+      "learning_rate": 1.5803188641921417e-05,
+      "loss": 0.7563,
+      "step": 368
+    },
+    {
+      "epoch": 1.1197724039829302,
+      "grad_norm": 2.2810609340667725,
+      "learning_rate": 1.5774340677765483e-05,
+      "loss": 0.7865,
+      "step": 369
+    },
+    {
+      "epoch": 1.1228070175438596,
+      "grad_norm": 2.147937297821045,
+      "learning_rate": 1.5745420440662543e-05,
+      "loss": 0.7684,
+      "step": 370
+    },
+    {
+      "epoch": 1.125841631104789,
+      "grad_norm": 2.355337142944336,
+      "learning_rate": 1.5716428292584788e-05,
+      "loss": 0.7693,
+      "step": 371
+    },
+    {
+      "epoch": 1.1288762446657183,
+      "grad_norm": 2.4234957695007324,
+      "learning_rate": 1.568736459640447e-05,
+      "loss": 0.7567,
+      "step": 372
+    },
+    {
+      "epoch": 1.1319108582266477,
+      "grad_norm": 2.2321126461029053,
+      "learning_rate": 1.5658229715889345e-05,
+      "loss": 0.7984,
+      "step": 373
+    },
+    {
+      "epoch": 1.134945471787577,
+      "grad_norm": 2.2270772457122803,
+      "learning_rate": 1.5629024015698137e-05,
+      "loss": 0.7868,
+      "step": 374
+    },
+    {
+      "epoch": 1.1379800853485065,
+      "grad_norm": 2.4906022548675537,
+      "learning_rate": 1.5599747861375957e-05,
+      "loss": 0.7761,
+      "step": 375
+    },
+    {
+      "epoch": 1.1410146989094359,
+      "grad_norm": 2.4099533557891846,
+      "learning_rate": 1.5570401619349737e-05,
+      "loss": 0.7727,
+      "step": 376
+    },
+    {
+      "epoch": 1.144049312470365,
+      "grad_norm": 2.167451858520508,
+      "learning_rate": 1.5540985656923648e-05,
+      "loss": 0.7449,
+      "step": 377
+    },
+    {
+      "epoch": 1.1470839260312944,
+      "grad_norm": 2.4137990474700928,
+      "learning_rate": 1.551150034227449e-05,
+      "loss": 0.7836,
+      "step": 378
+    },
+    {
+      "epoch": 1.1501185395922238,
+      "grad_norm": 2.0170676708221436,
+      "learning_rate": 1.54819460444471e-05,
+      "loss": 0.771,
+      "step": 379
+    },
+    {
+      "epoch": 1.1531531531531531,
+      "grad_norm": 2.3556909561157227,
+      "learning_rate": 1.5452323133349712e-05,
+      "loss": 0.769,
+      "step": 380
+    },
+    {
+      "epoch": 1.1561877667140825,
+      "grad_norm": 2.1490001678466797,
+      "learning_rate": 1.5422631979749354e-05,
+      "loss": 0.7467,
+      "step": 381
+    },
+    {
+      "epoch": 1.159222380275012,
+      "grad_norm": 2.240185260772705,
+      "learning_rate": 1.5392872955267176e-05,
+      "loss": 0.7677,
+      "step": 382
+    },
+    {
+      "epoch": 1.1622569938359413,
+      "grad_norm": 2.1402430534362793,
+      "learning_rate": 1.5363046432373824e-05,
+      "loss": 0.7706,
+      "step": 383
+    },
+    {
+      "epoch": 1.1652916073968704,
+      "grad_norm": 2.043536424636841,
+      "learning_rate": 1.5333152784384777e-05,
+      "loss": 0.7679,
+      "step": 384
+    },
+    {
+      "epoch": 1.1683262209577998,
+      "grad_norm": 2.146432399749756,
+      "learning_rate": 1.5303192385455652e-05,
+      "loss": 0.7746,
+      "step": 385
+    },
+    {
+      "epoch": 1.1713608345187292,
+      "grad_norm": 2.216525077819824,
+      "learning_rate": 1.5273165610577543e-05,
+      "loss": 0.7735,
+      "step": 386
+    },
+    {
+      "epoch": 1.1743954480796586,
+      "grad_norm": 2.134411334991455,
+      "learning_rate": 1.5243072835572319e-05,
+      "loss": 0.771,
+      "step": 387
+    },
+    {
+      "epoch": 1.177430061640588,
+      "grad_norm": 2.210275173187256,
+      "learning_rate": 1.5212914437087921e-05,
+      "loss": 0.7665,
+      "step": 388
+    },
+    {
+      "epoch": 1.1804646752015173,
+      "grad_norm": 2.2126176357269287,
+      "learning_rate": 1.5182690792593659e-05,
+      "loss": 0.7658,
+      "step": 389
+    },
+    {
+      "epoch": 1.1834992887624467,
+      "grad_norm": 1.9885146617889404,
+      "learning_rate": 1.5152402280375454e-05,
+      "loss": 0.7509,
+      "step": 390
+    },
+    {
+      "epoch": 1.186533902323376,
+      "grad_norm": 2.2027952671051025,
+      "learning_rate": 1.5122049279531143e-05,
+      "loss": 0.7811,
+      "step": 391
+    },
+    {
+      "epoch": 1.1895685158843055,
+      "grad_norm": 2.1936960220336914,
+      "learning_rate": 1.509163216996572e-05,
+      "loss": 0.7785,
+      "step": 392
+    },
+    {
+      "epoch": 1.1926031294452346,
+      "grad_norm": 2.166518449783325,
+      "learning_rate": 1.5061151332386565e-05,
+      "loss": 0.7775,
+      "step": 393
+    },
+    {
+      "epoch": 1.195637743006164,
+      "grad_norm": 2.0909955501556396,
+      "learning_rate": 1.5030607148298697e-05,
+      "loss": 0.7783,
+      "step": 394
+    },
+    {
+      "epoch": 1.1986723565670934,
+      "grad_norm": 2.287322521209717,
+      "learning_rate": 1.5000000000000002e-05,
+      "loss": 0.7678,
+      "step": 395
+    },
+    {
+      "epoch": 1.2017069701280227,
+      "grad_norm": 2.265106678009033,
+      "learning_rate": 1.4969330270576428e-05,
+      "loss": 0.7772,
+      "step": 396
+    },
+    {
+      "epoch": 1.2047415836889521,
+      "grad_norm": 2.2475264072418213,
+      "learning_rate": 1.4938598343897215e-05,
+      "loss": 0.7509,
+      "step": 397
+    },
+    {
+      "epoch": 1.2077761972498815,
+      "grad_norm": 2.3240981101989746,
+      "learning_rate": 1.4907804604610064e-05,
+      "loss": 0.7849,
+      "step": 398
+    },
+    {
+      "epoch": 1.2108108108108109,
+      "grad_norm": 2.148869276046753,
+      "learning_rate": 1.4876949438136348e-05,
+      "loss": 0.7781,
+      "step": 399
+    },
+    {
+      "epoch": 1.2138454243717403,
+      "grad_norm": 2.318875789642334,
+      "learning_rate": 1.484603323066627e-05,
+      "loss": 0.7987,
+      "step": 400
+    },
+    {
+      "epoch": 1.2138454243717403,
+      "eval_loss": NaN,
+      "eval_runtime": 204.8025,
+      "eval_samples_per_second": 20.302,
+      "eval_steps_per_second": 5.078,
+      "step": 400
+    },
+    {
+      "epoch": 1.2168800379326696,
+      "grad_norm": 2.1469035148620605,
+      "learning_rate": 1.4815056369154039e-05,
+      "loss": 0.7924,
+      "step": 401
+    },
+    {
+      "epoch": 1.2199146514935988,
+      "grad_norm": 2.3983654975891113,
+      "learning_rate": 1.4784019241313025e-05,
+      "loss": 0.7431,
+      "step": 402
+    },
+    {
+      "epoch": 1.2229492650545282,
+      "grad_norm": 2.1171765327453613,
+      "learning_rate": 1.47529222356109e-05,
+      "loss": 0.7583,
+      "step": 403
+    },
+    {
+      "epoch": 1.2259838786154575,
+      "grad_norm": 2.3186557292938232,
+      "learning_rate": 1.4721765741264786e-05,
+      "loss": 0.7545,
+      "step": 404
+    },
+    {
+      "epoch": 1.229018492176387,
+      "grad_norm": 2.308945417404175,
+      "learning_rate": 1.4690550148236371e-05,
+      "loss": 0.7752,
+      "step": 405
+    },
+    {
+      "epoch": 1.2320531057373163,
+      "grad_norm": 2.141418933868408,
+      "learning_rate": 1.4659275847227044e-05,
+      "loss": 0.7501,
+      "step": 406
+    },
+    {
+      "epoch": 1.2350877192982457,
+      "grad_norm": 2.1447696685791016,
+      "learning_rate": 1.4627943229672992e-05,
+      "loss": 0.7446,
+      "step": 407
+    },
+    {
+      "epoch": 1.238122332859175,
+      "grad_norm": 2.062683582305908,
+      "learning_rate": 1.4596552687740304e-05,
+      "loss": 0.7729,
+      "step": 408
+    },
+    {
+      "epoch": 1.2411569464201042,
+      "grad_norm": 2.283247232437134,
+      "learning_rate": 1.4565104614320065e-05,
+      "loss": 0.7752,
+      "step": 409
+    },
+    {
+      "epoch": 1.2441915599810336,
+      "grad_norm": 2.299151659011841,
+      "learning_rate": 1.453359940302344e-05,
+      "loss": 0.7794,
+      "step": 410
+    },
+    {
+      "epoch": 1.247226173541963,
+      "grad_norm": 2.2340760231018066,
+      "learning_rate": 1.4502037448176734e-05,
+      "loss": 0.7811,
+      "step": 411
+    },
+    {
+      "epoch": 1.2502607871028923,
+      "grad_norm": 2.305233955383301,
+      "learning_rate": 1.4470419144816483e-05,
+      "loss": 0.7663,
+      "step": 412
+    },
+    {
+      "epoch": 1.2532954006638217,
+      "grad_norm": 2.1460888385772705,
+      "learning_rate": 1.4438744888684481e-05,
+      "loss": 0.7584,
+      "step": 413
+    },
+    {
+      "epoch": 1.256330014224751,
+      "grad_norm": 2.3122851848602295,
+      "learning_rate": 1.4407015076222845e-05,
+      "loss": 0.7817,
+      "step": 414
+    },
+    {
+      "epoch": 1.2593646277856805,
+      "grad_norm": 2.0583643913269043,
+      "learning_rate": 1.4375230104569044e-05,
+      "loss": 0.7695,
+      "step": 415
+    },
+    {
+      "epoch": 1.2623992413466096,
+      "grad_norm": 2.262274980545044,
+      "learning_rate": 1.4343390371550936e-05,
+      "loss": 0.7739,
+      "step": 416
+    },
+    {
+      "epoch": 1.2654338549075392,
+      "grad_norm": 2.151893377304077,
+      "learning_rate": 1.4311496275681785e-05,
+      "loss": 0.7789,
+      "step": 417
+    },
+    {
+      "epoch": 1.2684684684684684,
+      "grad_norm": 2.28367280960083,
+      "learning_rate": 1.4279548216155265e-05,
+      "loss": 0.775,
+      "step": 418
+    },
+    {
+      "epoch": 1.2715030820293978,
+      "grad_norm": 2.3059751987457275,
+      "learning_rate": 1.424754659284048e-05,
+      "loss": 0.7613,
+      "step": 419
+    },
+    {
+      "epoch": 1.2745376955903271,
+      "grad_norm": 2.436896800994873,
+      "learning_rate": 1.4215491806276944e-05,
+      "loss": 0.7835,
+      "step": 420
+    },
+    {
+      "epoch": 1.2775723091512565,
+      "grad_norm": 2.102220058441162,
+      "learning_rate": 1.418338425766958e-05,
+      "loss": 0.7932,
+      "step": 421
+    },
+    {
+      "epoch": 1.280606922712186,
+      "grad_norm": 2.1711723804473877,
+      "learning_rate": 1.4151224348883692e-05,
+      "loss": 0.7668,
+      "step": 422
+    },
+    {
+      "epoch": 1.2836415362731153,
+      "grad_norm": 2.213289737701416,
+      "learning_rate": 1.4119012482439929e-05,
+      "loss": 0.7745,
+      "step": 423
+    },
+    {
+      "epoch": 1.2866761498340447,
+      "grad_norm": 2.0960137844085693,
+      "learning_rate": 1.408674906150926e-05,
+      "loss": 0.7742,
+      "step": 424
+    },
+    {
+      "epoch": 1.2897107633949738,
+      "grad_norm": 2.081200122833252,
+      "learning_rate": 1.4054434489907916e-05,
+      "loss": 0.7652,
+      "step": 425
+    },
+    {
+      "epoch": 1.2927453769559032,
+      "grad_norm": 2.284423351287842,
+      "learning_rate": 1.4022069172092354e-05,
+      "loss": 0.7762,
+      "step": 426
+    },
+    {
+      "epoch": 1.2957799905168326,
+      "grad_norm": 2.292639970779419,
+      "learning_rate": 1.3989653513154165e-05,
+      "loss": 0.7644,
+      "step": 427
+    },
+    {
+      "epoch": 1.298814604077762,
+      "grad_norm": 2.282759428024292,
+      "learning_rate": 1.3957187918815032e-05,
+      "loss": 0.7658,
+      "step": 428
+    },
+    {
+      "epoch": 1.3018492176386913,
+      "grad_norm": 2.2170190811157227,
+      "learning_rate": 1.3924672795421638e-05,
+      "loss": 0.7661,
+      "step": 429
+    },
+    {
+      "epoch": 1.3048838311996207,
+      "grad_norm": 2.202991247177124,
+      "learning_rate": 1.3892108549940583e-05,
+      "loss": 0.7881,
+      "step": 430
+    },
+    {
+      "epoch": 1.30791844476055,
+      "grad_norm": 2.148986339569092,
+      "learning_rate": 1.3859495589953289e-05,
+      "loss": 0.7865,
+      "step": 431
+    },
+    {
+      "epoch": 1.3109530583214792,
+      "grad_norm": 2.1265058517456055,
+      "learning_rate": 1.3826834323650899e-05,
+      "loss": 0.789,
+      "step": 432
+    },
+    {
+      "epoch": 1.3139876718824088,
+      "grad_norm": 2.1798737049102783,
+      "learning_rate": 1.3794125159829173e-05,
+      "loss": 0.7707,
+      "step": 433
+    },
+    {
+      "epoch": 1.317022285443338,
+      "grad_norm": 2.2072978019714355,
+      "learning_rate": 1.376136850788336e-05,
+      "loss": 0.763,
+      "step": 434
+    },
+    {
+      "epoch": 1.3200568990042674,
+      "grad_norm": 2.12349534034729,
+      "learning_rate": 1.3728564777803089e-05,
+      "loss": 0.7505,
+      "step": 435
+    },
+    {
+      "epoch": 1.3230915125651967,
+      "grad_norm": 2.104276180267334,
+      "learning_rate": 1.3695714380167221e-05,
+      "loss": 0.7891,
+      "step": 436
+    },
+    {
+      "epoch": 1.3261261261261261,
+      "grad_norm": 2.038515090942383,
+      "learning_rate": 1.3662817726138729e-05,
+      "loss": 0.7668,
+      "step": 437
+    },
+    {
+      "epoch": 1.3291607396870555,
+      "grad_norm": 2.0557668209075928,
+      "learning_rate": 1.3629875227459532e-05,
+      "loss": 0.7685,
+      "step": 438
+    },
+    {
+      "epoch": 1.3321953532479849,
+      "grad_norm": 2.221299886703491,
+      "learning_rate": 1.359688729644536e-05,
+      "loss": 0.7765,
+      "step": 439
+    },
+    {
+      "epoch": 1.3352299668089143,
+      "grad_norm": 2.383873462677002,
+      "learning_rate": 1.356385434598057e-05,
+      "loss": 0.7863,
+      "step": 440
+    },
+    {
+      "epoch": 1.3382645803698434,
+      "grad_norm": 2.144969940185547,
+      "learning_rate": 1.3530776789513009e-05,
+      "loss": 0.7854,
+      "step": 441
+    },
+    {
+      "epoch": 1.341299193930773,
+      "grad_norm": 2.3431177139282227,
+      "learning_rate": 1.3497655041048812e-05,
+      "loss": 0.7491,
+      "step": 442
+    },
+    {
+      "epoch": 1.3443338074917022,
+      "grad_norm": 2.1558756828308105,
+      "learning_rate": 1.3464489515147239e-05,
+      "loss": 0.7935,
+      "step": 443
+    },
+    {
+      "epoch": 1.3473684210526315,
+      "grad_norm": 2.4032328128814697,
+      "learning_rate": 1.3431280626915466e-05,
+      "loss": 0.765,
+      "step": 444
+    },
+    {
+      "epoch": 1.350403034613561,
+      "grad_norm": 2.373549461364746,
+      "learning_rate": 1.3398028792003413e-05,
+      "loss": 0.766,
+      "step": 445
+    },
+    {
+      "epoch": 1.3534376481744903,
+      "grad_norm": 2.38440203666687,
+      "learning_rate": 1.3364734426598527e-05,
+      "loss": 0.7849,
+      "step": 446
+    },
+    {
+      "epoch": 1.3564722617354197,
+      "grad_norm": 2.319101333618164,
+      "learning_rate": 1.3331397947420578e-05,
+      "loss": 0.7738,
+      "step": 447
+    },
+    {
+      "epoch": 1.359506875296349,
+      "grad_norm": 2.1911866664886475,
+      "learning_rate": 1.3298019771716435e-05,
+      "loss": 0.7779,
+      "step": 448
+    },
+    {
+      "epoch": 1.3625414888572784,
+      "grad_norm": 2.273451328277588,
+      "learning_rate": 1.3264600317254854e-05,
+      "loss": 0.76,
+      "step": 449
+    },
+    {
+      "epoch": 1.3655761024182076,
+      "grad_norm": 2.2576355934143066,
+      "learning_rate": 1.3231140002321252e-05,
+      "loss": 0.7687,
+      "step": 450
+    },
+    {
+      "epoch": 1.3655761024182076,
+      "eval_loss": NaN,
+      "eval_runtime": 203.9814,
+      "eval_samples_per_second": 20.384,
+      "eval_steps_per_second": 5.099,
+      "step": 450
+    },
+    {
+      "epoch": 1.368610715979137,
+      "grad_norm": 2.2846243381500244,
+      "learning_rate": 1.3197639245712454e-05,
+      "loss": 0.77,
+      "step": 451
+    },
+    {
+      "epoch": 1.3716453295400663,
+      "grad_norm": 2.1583635807037354,
+      "learning_rate": 1.3164098466731467e-05,
+      "loss": 0.7681,
+      "step": 452
+    },
+    {
+      "epoch": 1.3746799431009957,
+      "grad_norm": 2.126995325088501,
+      "learning_rate": 1.3130518085182224e-05,
+      "loss": 0.7755,
+      "step": 453
+    },
+    {
+      "epoch": 1.377714556661925,
+      "grad_norm": 2.2357747554779053,
+      "learning_rate": 1.3096898521364338e-05,
+      "loss": 0.7509,
+      "step": 454
+    },
+    {
+      "epoch": 1.3807491702228545,
+      "grad_norm": 2.0657386779785156,
+      "learning_rate": 1.3063240196067837e-05,
+      "loss": 0.8043,
+      "step": 455
+    },
+    {
+      "epoch": 1.3837837837837839,
+      "grad_norm": 2.1653802394866943,
+      "learning_rate": 1.3029543530567884e-05,
+      "loss": 0.7676,
+      "step": 456
+    },
+    {
+      "epoch": 1.386818397344713,
+      "grad_norm": 2.161508798599243,
+      "learning_rate": 1.2995808946619533e-05,
+      "loss": 0.7735,
+      "step": 457
+    },
+    {
+      "epoch": 1.3898530109056426,
+      "grad_norm": 2.185350179672241,
+      "learning_rate": 1.2962036866452423e-05,
+      "loss": 0.7891,
+      "step": 458
+    },
+    {
+      "epoch": 1.3928876244665718,
+      "grad_norm": 2.198807954788208,
+      "learning_rate": 1.2928227712765504e-05,
+      "loss": 0.7657,
+      "step": 459
+    },
+    {
+      "epoch": 1.3959222380275011,
+      "grad_norm": 2.2574117183685303,
+      "learning_rate": 1.2894381908721757e-05,
+      "loss": 0.7893,
+      "step": 460
+    },
+    {
+      "epoch": 1.3989568515884305,
+      "grad_norm": 2.2577221393585205,
+      "learning_rate": 1.2860499877942876e-05,
+      "loss": 0.7787,
+      "step": 461
+    },
+    {
+      "epoch": 1.40199146514936,
+      "grad_norm": 2.265421152114868,
+      "learning_rate": 1.282658204450398e-05,
+      "loss": 0.7714,
+      "step": 462
+    },
+    {
+      "epoch": 1.4050260787102893,
+      "grad_norm": 2.0731489658355713,
+      "learning_rate": 1.2792628832928302e-05,
+      "loss": 0.782,
+      "step": 463
+    },
+    {
+      "epoch": 1.4080606922712187,
+      "grad_norm": 2.130533218383789,
+      "learning_rate": 1.275864066818188e-05,
+      "loss": 0.7873,
+      "step": 464
+    },
+    {
+      "epoch": 1.411095305832148,
+      "grad_norm": 2.0858917236328125,
+      "learning_rate": 1.2724617975668229e-05,
+      "loss": 0.8005,
+      "step": 465
+    },
+    {
+      "epoch": 1.4141299193930772,
+      "grad_norm": 2.0890703201293945,
+      "learning_rate": 1.2690561181223024e-05,
+      "loss": 0.775,
+      "step": 466
+    },
+    {
+      "epoch": 1.4171645329540066,
+      "grad_norm": 2.3934412002563477,
+      "learning_rate": 1.2656470711108763e-05,
+      "loss": 0.7931,
+      "step": 467
+    },
+    {
+      "epoch": 1.420199146514936,
+      "grad_norm": 2.2045512199401855,
+      "learning_rate": 1.2622346992009447e-05,
+      "loss": 0.7747,
+      "step": 468
+    },
+    {
+      "epoch": 1.4232337600758653,
+      "grad_norm": 2.371346950531006,
+      "learning_rate": 1.2588190451025209e-05,
+      "loss": 0.7581,
+      "step": 469
+    },
+    {
+      "epoch": 1.4262683736367947,
+      "grad_norm": 2.1523914337158203,
+      "learning_rate": 1.2554001515667009e-05,
+      "loss": 0.7741,
+      "step": 470
+    },
+    {
+      "epoch": 1.429302987197724,
+      "grad_norm": 2.2948853969573975,
+      "learning_rate": 1.2519780613851254e-05,
+      "loss": 0.7925,
+      "step": 471
+    },
+    {
+      "epoch": 1.4323376007586535,
+      "grad_norm": 2.2751922607421875,
+      "learning_rate": 1.2485528173894447e-05,
+      "loss": 0.7784,
+      "step": 472
+    },
+    {
+      "epoch": 1.4353722143195826,
+      "grad_norm": 2.1596484184265137,
+      "learning_rate": 1.2451244624507831e-05,
+      "loss": 0.7895,
+      "step": 473
+    },
+    {
+      "epoch": 1.4384068278805122,
+      "grad_norm": 2.2292542457580566,
+      "learning_rate": 1.2416930394792026e-05,
+      "loss": 0.7698,
+      "step": 474
+    },
+    {
+      "epoch": 1.4414414414414414,
+      "grad_norm": 2.1559245586395264,
+      "learning_rate": 1.238258591423165e-05,
+      "loss": 0.776,
+      "step": 475
+    },
+    {
+      "epoch": 1.4444760550023708,
+      "grad_norm": 2.072768211364746,
+      "learning_rate": 1.234821161268995e-05,
+      "loss": 0.7665,
+      "step": 476
+    },
+    {
+      "epoch": 1.4475106685633001,
+      "grad_norm": 2.3749032020568848,
+      "learning_rate": 1.2313807920403419e-05,
+      "loss": 0.7765,
+      "step": 477
+    },
+    {
+      "epoch": 1.4505452821242295,
+      "grad_norm": 2.1614534854888916,
+      "learning_rate": 1.22793752679764e-05,
+      "loss": 0.7908,
+      "step": 478
+    },
+    {
+      "epoch": 1.4535798956851589,
+      "grad_norm": 2.2183918952941895,
+      "learning_rate": 1.2244914086375726e-05,
+      "loss": 0.7662,
+      "step": 479
+    },
+    {
+      "epoch": 1.4566145092460883,
+      "grad_norm": 2.230243444442749,
+      "learning_rate": 1.22104248069253e-05,
+      "loss": 0.7758,
+      "step": 480
+    },
+    {
+      "epoch": 1.4596491228070176,
+      "grad_norm": 2.503591775894165,
+      "learning_rate": 1.2175907861300698e-05,
+      "loss": 0.7739,
+      "step": 481
+    },
+    {
+      "epoch": 1.4626837363679468,
+      "grad_norm": 2.4481165409088135,
+      "learning_rate": 1.2141363681523777e-05,
+      "loss": 0.788,
+      "step": 482
+    },
+    {
+      "epoch": 1.4657183499288762,
+      "grad_norm": 2.30068302154541,
+      "learning_rate": 1.2106792699957264e-05,
+      "loss": 0.7905,
+      "step": 483
+    },
+    {
+      "epoch": 1.4687529634898056,
+      "grad_norm": 2.2382349967956543,
+      "learning_rate": 1.2072195349299344e-05,
+      "loss": 0.7617,
+      "step": 484
+    },
+    {
+      "epoch": 1.471787577050735,
+      "grad_norm": 2.2054882049560547,
+      "learning_rate": 1.2037572062578238e-05,
+      "loss": 0.7802,
+      "step": 485
+    },
+    {
+      "epoch": 1.4748221906116643,
+      "grad_norm": 2.282318115234375,
+      "learning_rate": 1.2002923273146793e-05,
+      "loss": 0.7711,
+      "step": 486
+    },
+    {
+      "epoch": 1.4778568041725937,
+      "grad_norm": 2.1077611446380615,
+      "learning_rate": 1.1968249414677055e-05,
+      "loss": 0.761,
+      "step": 487
+    },
+    {
+      "epoch": 1.480891417733523,
+      "grad_norm": 2.1558871269226074,
+      "learning_rate": 1.1933550921154836e-05,
+      "loss": 0.7872,
+      "step": 488
+    },
+    {
+      "epoch": 1.4839260312944522,
+      "grad_norm": 2.332897901535034,
+      "learning_rate": 1.1898828226874284e-05,
+      "loss": 0.7791,
+      "step": 489
+    },
+    {
+      "epoch": 1.4869606448553818,
+      "grad_norm": 2.0427420139312744,
+      "learning_rate": 1.1864081766432457e-05,
+      "loss": 0.782,
+      "step": 490
+    },
+    {
+      "epoch": 1.489995258416311,
+      "grad_norm": 2.2689926624298096,
+      "learning_rate": 1.1829311974723868e-05,
+      "loss": 0.7659,
+      "step": 491
+    },
+    {
+      "epoch": 1.4930298719772404,
+      "grad_norm": 2.2432594299316406,
+      "learning_rate": 1.1794519286935056e-05,
+      "loss": 0.7543,
+      "step": 492
+    },
+    {
+      "epoch": 1.4960644855381697,
+      "grad_norm": 2.0946452617645264,
+      "learning_rate": 1.1759704138539121e-05,
+      "loss": 0.7712,
+      "step": 493
+    },
+    {
+      "epoch": 1.499099099099099,
+      "grad_norm": 2.1154541969299316,
+      "learning_rate": 1.1724866965290302e-05,
+      "loss": 0.7732,
+      "step": 494
+    },
+    {
+      "epoch": 1.5021337126600285,
+      "grad_norm": 2.195223569869995,
+      "learning_rate": 1.1690008203218493e-05,
+      "loss": 0.7596,
+      "step": 495
+    },
+    {
+      "epoch": 1.5051683262209576,
+      "grad_norm": 2.105151653289795,
+      "learning_rate": 1.1655128288623803e-05,
+      "loss": 0.775,
+      "step": 496
+    },
+    {
+      "epoch": 1.5082029397818872,
+      "grad_norm": 2.196159601211548,
+      "learning_rate": 1.1620227658071088e-05,
+      "loss": 0.7893,
+      "step": 497
+    },
+    {
+      "epoch": 1.5112375533428164,
+      "grad_norm": 2.3409769535064697,
+      "learning_rate": 1.158530674838449e-05,
+      "loss": 0.7494,
+      "step": 498
+    },
+    {
+      "epoch": 1.514272166903746,
+      "grad_norm": 2.177128553390503,
+      "learning_rate": 1.155036599664198e-05,
+      "loss": 0.7595,
+      "step": 499
+    },
+    {
+      "epoch": 1.5173067804646752,
+      "grad_norm": 2.3704683780670166,
+      "learning_rate": 1.1515405840169861e-05,
+      "loss": 0.7607,
+      "step": 500
+    },
+    {
+      "epoch": 1.5173067804646752,
+      "eval_loss": NaN,
+      "eval_runtime": 204.5697,
+      "eval_samples_per_second": 20.326,
+      "eval_steps_per_second": 5.084,
+      "step": 500
+    },
+    {
+      "epoch": 1.5203413940256045,
+      "grad_norm": 2.175325393676758,
+      "learning_rate": 1.1480426716537316e-05,
+      "loss": 0.7626,
+      "step": 501
+    },
+    {
+      "epoch": 1.523376007586534,
+      "grad_norm": 2.093395471572876,
+      "learning_rate": 1.1445429063550925e-05,
+      "loss": 0.7787,
+      "step": 502
+    },
+    {
+      "epoch": 1.5264106211474633,
+      "grad_norm": 2.2766175270080566,
+      "learning_rate": 1.1410413319249193e-05,
+      "loss": 0.7592,
+      "step": 503
+    },
+    {
+      "epoch": 1.5294452347083927,
+      "grad_norm": 2.00833797454834,
+      "learning_rate": 1.1375379921897052e-05,
+      "loss": 0.7685,
+      "step": 504
+    },
+    {
+      "epoch": 1.5324798482693218,
+      "grad_norm": 2.2018191814422607,
+      "learning_rate": 1.1340329309980379e-05,
+      "loss": 0.753,
+      "step": 505
+    },
+    {
+      "epoch": 1.5355144618302514,
+      "grad_norm": 2.228724479675293,
+      "learning_rate": 1.130526192220052e-05,
+      "loss": 0.7687,
+      "step": 506
+    },
+    {
+      "epoch": 1.5385490753911806,
+      "grad_norm": 2.241725444793701,
+      "learning_rate": 1.1270178197468788e-05,
+      "loss": 0.7674,
+      "step": 507
+    },
+    {
+      "epoch": 1.54158368895211,
+      "grad_norm": 2.1697564125061035,
+      "learning_rate": 1.1235078574900984e-05,
+      "loss": 0.7726,
+      "step": 508
+    },
+    {
+      "epoch": 1.5446183025130393,
+      "grad_norm": 2.197449207305908,
+      "learning_rate": 1.119996349381187e-05,
+      "loss": 0.7672,
+      "step": 509
+    },
+    {
+      "epoch": 1.5476529160739687,
+      "grad_norm": 2.0595834255218506,
+      "learning_rate": 1.1164833393709707e-05,
+      "loss": 0.7706,
+      "step": 510
+    },
+    {
+      "epoch": 1.550687529634898,
+      "grad_norm": 2.3002591133117676,
+      "learning_rate": 1.112968871429073e-05,
+      "loss": 0.7875,
+      "step": 511
+    },
+    {
+      "epoch": 1.5537221431958272,
+      "grad_norm": 2.175219774246216,
+      "learning_rate": 1.1094529895433653e-05,
+      "loss": 0.7809,
+      "step": 512
+    },
+    {
+      "epoch": 1.5567567567567568,
+      "grad_norm": 2.1368846893310547,
+      "learning_rate": 1.1059357377194161e-05,
+      "loss": 0.7878,
+      "step": 513
+    },
+    {
+      "epoch": 1.559791370317686,
+      "grad_norm": 2.210344076156616,
+      "learning_rate": 1.102417159979941e-05,
+      "loss": 0.7543,
+      "step": 514
+    },
+    {
+      "epoch": 1.5628259838786156,
+      "grad_norm": 2.1888577938079834,
+      "learning_rate": 1.09889730036425e-05,
+      "loss": 0.7731,
+      "step": 515
+    },
+    {
+      "epoch": 1.5658605974395448,
+      "grad_norm": 2.3701112270355225,
+      "learning_rate": 1.0953762029276982e-05,
+      "loss": 0.7689,
+      "step": 516
+    },
+    {
+      "epoch": 1.5688952110004741,
+      "grad_norm": 2.069556713104248,
+      "learning_rate": 1.0918539117411334e-05,
+      "loss": 0.767,
+      "step": 517
+    },
+    {
+      "epoch": 1.5719298245614035,
+      "grad_norm": 2.209773302078247,
+      "learning_rate": 1.0883304708903441e-05,
+      "loss": 0.7696,
+      "step": 518
+    },
+    {
+      "epoch": 1.5749644381223329,
+      "grad_norm": 2.1156795024871826,
+      "learning_rate": 1.0848059244755093e-05,
+      "loss": 0.7827,
+      "step": 519
+    },
+    {
+      "epoch": 1.5779990516832623,
+      "grad_norm": 2.359513521194458,
+      "learning_rate": 1.0812803166106445e-05,
+      "loss": 0.7612,
+      "step": 520
+    },
+    {
+      "epoch": 1.5810336652441914,
+      "grad_norm": 2.2499759197235107,
+      "learning_rate": 1.0777536914230509e-05,
+      "loss": 0.7671,
+      "step": 521
+    },
+    {
+      "epoch": 1.584068278805121,
+      "grad_norm": 2.219525098800659,
+      "learning_rate": 1.0742260930527625e-05,
+      "loss": 0.776,
+      "step": 522
+    },
+    {
+      "epoch": 1.5871028923660502,
+      "grad_norm": 2.339210033416748,
+      "learning_rate": 1.0706975656519946e-05,
+      "loss": 0.7669,
+      "step": 523
+    },
+    {
+      "epoch": 1.5901375059269798,
+      "grad_norm": 2.3282480239868164,
+      "learning_rate": 1.06716815338459e-05,
+      "loss": 0.7843,
+      "step": 524
+    },
+    {
+      "epoch": 1.593172119487909,
+      "grad_norm": 2.09635329246521,
+      "learning_rate": 1.0636379004254665e-05,
+      "loss": 0.7598,
+      "step": 525
+    },
+    {
+      "epoch": 1.5962067330488383,
+      "grad_norm": 2.3128199577331543,
+      "learning_rate": 1.0601068509600642e-05,
+      "loss": 0.7673,
+      "step": 526
+    },
+    {
+      "epoch": 1.5992413466097677,
+      "grad_norm": 2.3669700622558594,
+      "learning_rate": 1.0565750491837925e-05,
+      "loss": 0.7697,
+      "step": 527
+    },
+    {
+      "epoch": 1.602275960170697,
+      "grad_norm": 2.2540953159332275,
+      "learning_rate": 1.0530425393014773e-05,
+      "loss": 0.7641,
+      "step": 528
+    },
+    {
+      "epoch": 1.6053105737316264,
+      "grad_norm": 2.3089439868927,
+      "learning_rate": 1.049509365526807e-05,
+      "loss": 0.768,
+      "step": 529
+    },
+    {
+      "epoch": 1.6083451872925556,
+      "grad_norm": 2.0084259510040283,
+      "learning_rate": 1.0459755720817797e-05,
+      "loss": 0.7504,
+      "step": 530
+    },
+    {
+      "epoch": 1.6113798008534852,
+      "grad_norm": 2.261798143386841,
+      "learning_rate": 1.0424412031961485e-05,
+      "loss": 0.7629,
+      "step": 531
+    },
+    {
+      "epoch": 1.6144144144144144,
+      "grad_norm": 2.2108161449432373,
+      "learning_rate": 1.0389063031068698e-05,
+      "loss": 0.738,
+      "step": 532
+    },
+    {
+      "epoch": 1.6174490279753437,
+      "grad_norm": 2.2087652683258057,
+      "learning_rate": 1.0353709160575488e-05,
+      "loss": 0.7758,
+      "step": 533
+    },
+    {
+      "epoch": 1.6204836415362731,
+      "grad_norm": 2.218838691711426,
+      "learning_rate": 1.0318350862978848e-05,
+      "loss": 0.7641,
+      "step": 534
+    },
+    {
+      "epoch": 1.6235182550972025,
+      "grad_norm": 2.156991720199585,
+      "learning_rate": 1.0282988580831183e-05,
+      "loss": 0.7577,
+      "step": 535
+    },
+    {
+      "epoch": 1.6265528686581319,
+      "grad_norm": 2.1950886249542236,
+      "learning_rate": 1.0247622756734775e-05,
+      "loss": 0.7888,
+      "step": 536
+    },
+    {
+      "epoch": 1.629587482219061,
+      "grad_norm": 2.114649534225464,
+      "learning_rate": 1.0212253833336237e-05,
+      "loss": 0.7766,
+      "step": 537
+    },
+    {
+      "epoch": 1.6326220957799906,
+      "grad_norm": 2.217928409576416,
+      "learning_rate": 1.0176882253320968e-05,
+      "loss": 0.7529,
+      "step": 538
+    },
+    {
+      "epoch": 1.6356567093409198,
+      "grad_norm": 2.2367630004882812,
+      "learning_rate": 1.0141508459407622e-05,
+      "loss": 0.7699,
+      "step": 539
+    },
+    {
+      "epoch": 1.6386913229018494,
+      "grad_norm": 2.1163787841796875,
+      "learning_rate": 1.0106132894342564e-05,
+      "loss": 0.7637,
+      "step": 540
+    },
+    {
+      "epoch": 1.6417259364627785,
+      "grad_norm": 2.083292245864868,
+      "learning_rate": 1.0070756000894321e-05,
+      "loss": 0.783,
+      "step": 541
+    },
+    {
+      "epoch": 1.644760550023708,
+      "grad_norm": 2.1091578006744385,
+      "learning_rate": 1.0035378221848053e-05,
+      "loss": 0.7848,
+      "step": 542
+    },
+    {
+      "epoch": 1.6477951635846373,
+      "grad_norm": 2.0994367599487305,
+      "learning_rate": 1e-05,
+      "loss": 0.7698,
+      "step": 543
+    },
+    {
+      "epoch": 1.6508297771455667,
+      "grad_norm": 2.2539420127868652,
+      "learning_rate": 9.964621778151947e-06,
+      "loss": 0.769,
+      "step": 544
+    },
+    {
+      "epoch": 1.653864390706496,
+      "grad_norm": 2.154353380203247,
+      "learning_rate": 9.929243999105682e-06,
+      "loss": 0.7496,
+      "step": 545
+    },
+    {
+      "epoch": 1.6568990042674252,
+      "grad_norm": 2.265385150909424,
+      "learning_rate": 9.89386710565744e-06,
+      "loss": 0.768,
+      "step": 546
+    },
+    {
+      "epoch": 1.6599336178283548,
+      "grad_norm": 2.073289632797241,
+      "learning_rate": 9.858491540592383e-06,
+      "loss": 0.7807,
+      "step": 547
+    },
+    {
+      "epoch": 1.662968231389284,
+      "grad_norm": 2.135737657546997,
+      "learning_rate": 9.823117746679034e-06,
+      "loss": 0.766,
+      "step": 548
+    },
+    {
+      "epoch": 1.6660028449502133,
+      "grad_norm": 2.1423635482788086,
+      "learning_rate": 9.787746166663765e-06,
+      "loss": 0.7588,
+      "step": 549
+    },
+    {
+      "epoch": 1.6690374585111427,
+      "grad_norm": 2.0772769451141357,
+      "learning_rate": 9.752377243265229e-06,
+      "loss": 0.7647,
+      "step": 550
+    },
+    {
+      "epoch": 1.6690374585111427,
+      "eval_loss": NaN,
+      "eval_runtime": 204.4783,
+      "eval_samples_per_second": 20.335,
+      "eval_steps_per_second": 5.086,
+      "step": 550
+    },
+    {
+      "epoch": 1.672072072072072,
+      "grad_norm": 2.118854284286499,
+      "learning_rate": 9.71701141916882e-06,
+      "loss": 0.7728,
+      "step": 551
+    },
+    {
+      "epoch": 1.6751066856330015,
+      "grad_norm": 2.1113169193267822,
+      "learning_rate": 9.681649137021158e-06,
+      "loss": 0.7796,
+      "step": 552
+    },
+    {
+      "epoch": 1.6781412991939306,
+      "grad_norm": 2.064993143081665,
+      "learning_rate": 9.646290839424515e-06,
+      "loss": 0.762,
+      "step": 553
+    },
+    {
+      "epoch": 1.6811759127548602,
+      "grad_norm": 2.151028633117676,
+      "learning_rate": 9.610936968931302e-06,
+      "loss": 0.7709,
+      "step": 554
+    },
+    {
+      "epoch": 1.6842105263157894,
+      "grad_norm": 2.201429843902588,
+      "learning_rate": 9.57558796803852e-06,
+      "loss": 0.7619,
+      "step": 555
+    },
+    {
+      "epoch": 1.687245139876719,
+      "grad_norm": 2.065720796585083,
+      "learning_rate": 9.540244279182206e-06,
+      "loss": 0.798,
+      "step": 556
+    },
+    {
+      "epoch": 1.6902797534376481,
+      "grad_norm": 2.1895217895507812,
+      "learning_rate": 9.504906344731933e-06,
+      "loss": 0.7802,
+      "step": 557
+    },
+    {
+      "epoch": 1.6933143669985775,
+      "grad_norm": 2.2065324783325195,
+      "learning_rate": 9.46957460698523e-06,
+      "loss": 0.7767,
+      "step": 558
+    },
+    {
+      "epoch": 1.696348980559507,
+      "grad_norm": 2.17842698097229,
+      "learning_rate": 9.434249508162076e-06,
+      "loss": 0.7604,
+      "step": 559
+    },
+    {
+      "epoch": 1.6993835941204363,
+      "grad_norm": 2.164586305618286,
+      "learning_rate": 9.398931490399363e-06,
+      "loss": 0.756,
+      "step": 560
+    },
+    {
+      "epoch": 1.7024182076813656,
+      "grad_norm": 2.235797643661499,
+      "learning_rate": 9.363620995745337e-06,
+      "loss": 0.777,
+      "step": 561
+    },
+    {
+      "epoch": 1.7054528212422948,
+      "grad_norm": 2.084228038787842,
+      "learning_rate": 9.328318466154102e-06,
+      "loss": 0.765,
+      "step": 562
+    },
+    {
+      "epoch": 1.7084874348032244,
+      "grad_norm": 2.1382224559783936,
+      "learning_rate": 9.293024343480056e-06,
+      "loss": 0.7712,
+      "step": 563
+    },
+    {
+      "epoch": 1.7115220483641536,
+      "grad_norm": 2.304201126098633,
+      "learning_rate": 9.257739069472375e-06,
+      "loss": 0.7748,
+      "step": 564
+    },
+    {
+      "epoch": 1.714556661925083,
+      "grad_norm": 2.1342921257019043,
+      "learning_rate": 9.222463085769495e-06,
+      "loss": 0.762,
+      "step": 565
+    },
+    {
+      "epoch": 1.7175912754860123,
+      "grad_norm": 2.189845323562622,
+      "learning_rate": 9.187196833893559e-06,
+      "loss": 0.7605,
+      "step": 566
+    },
+    {
+      "epoch": 1.7206258890469417,
+      "grad_norm": 2.3078930377960205,
+      "learning_rate": 9.151940755244912e-06,
+      "loss": 0.7677,
+      "step": 567
+    },
+    {
+      "epoch": 1.723660502607871,
+      "grad_norm": 2.303213357925415,
+      "learning_rate": 9.11669529109656e-06,
+      "loss": 0.7735,
+      "step": 568
+    },
+    {
+      "epoch": 1.7266951161688002,
+      "grad_norm": 2.3082435131073,
+      "learning_rate": 9.081460882588668e-06,
+      "loss": 0.7683,
+      "step": 569
+    },
+    {
+      "epoch": 1.7297297297297298,
+      "grad_norm": 2.2995338439941406,
+      "learning_rate": 9.046237970723022e-06,
+      "loss": 0.752,
+      "step": 570
+    },
+    {
+      "epoch": 1.732764343290659,
+      "grad_norm": 2.0647215843200684,
+      "learning_rate": 9.011026996357504e-06,
+      "loss": 0.7731,
+      "step": 571
+    },
+    {
+      "epoch": 1.7357989568515886,
+      "grad_norm": 2.160496473312378,
+      "learning_rate": 8.975828400200592e-06,
+      "loss": 0.7543,
+      "step": 572
+    },
+    {
+      "epoch": 1.7388335704125177,
+      "grad_norm": 2.3624351024627686,
+      "learning_rate": 8.94064262280584e-06,
+      "loss": 0.774,
+      "step": 573
+    },
+    {
+      "epoch": 1.7418681839734471,
+      "grad_norm": 2.165922164916992,
+      "learning_rate": 8.90547010456635e-06,
+      "loss": 0.7815,
+      "step": 574
+    },
+    {
+      "epoch": 1.7449027975343765,
+      "grad_norm": 2.138921022415161,
+      "learning_rate": 8.870311285709274e-06,
+      "loss": 0.7593,
+      "step": 575
+    },
+    {
+      "epoch": 1.7479374110953059,
+      "grad_norm": 2.122746706008911,
+      "learning_rate": 8.835166606290295e-06,
+      "loss": 0.783,
+      "step": 576
+    },
+    {
+      "epoch": 1.7509720246562352,
+      "grad_norm": 2.119318723678589,
+      "learning_rate": 8.80003650618813e-06,
+      "loss": 0.7778,
+      "step": 577
+    },
+    {
+      "epoch": 1.7540066382171644,
+      "grad_norm": 2.092092990875244,
+      "learning_rate": 8.76492142509902e-06,
+      "loss": 0.765,
+      "step": 578
+    },
+    {
+      "epoch": 1.757041251778094,
+      "grad_norm": 2.156517267227173,
+      "learning_rate": 8.729821802531213e-06,
+      "loss": 0.7583,
+      "step": 579
+    },
+    {
+      "epoch": 1.7600758653390232,
+      "grad_norm": 2.2096939086914062,
+      "learning_rate": 8.694738077799487e-06,
+      "loss": 0.7698,
+      "step": 580
+    },
+    {
+      "epoch": 1.7631104788999525,
+      "grad_norm": 1.981307029724121,
+      "learning_rate": 8.659670690019626e-06,
+      "loss": 0.7771,
+      "step": 581
+    },
+    {
+      "epoch": 1.766145092460882,
+      "grad_norm": 2.1369874477386475,
+      "learning_rate": 8.624620078102952e-06,
+      "loss": 0.777,
+      "step": 582
+    },
+    {
+      "epoch": 1.7691797060218113,
+      "grad_norm": 2.1329455375671387,
+      "learning_rate": 8.58958668075081e-06,
+      "loss": 0.7748,
+      "step": 583
+    },
+    {
+      "epoch": 1.7722143195827407,
+      "grad_norm": 2.1630642414093018,
+      "learning_rate": 8.554570936449073e-06,
+      "loss": 0.7675,
+      "step": 584
+    },
+    {
+      "epoch": 1.7752489331436698,
+      "grad_norm": 2.1383938789367676,
+      "learning_rate": 8.519573283462688e-06,
+      "loss": 0.7702,
+      "step": 585
+    },
+    {
+      "epoch": 1.7782835467045994,
+      "grad_norm": 2.001569986343384,
+      "learning_rate": 8.484594159830142e-06,
+      "loss": 0.7553,
+      "step": 586
+    },
+    {
+      "epoch": 1.7813181602655286,
+      "grad_norm": 2.1266000270843506,
+      "learning_rate": 8.449634003358022e-06,
+      "loss": 0.7757,
+      "step": 587
+    },
+    {
+      "epoch": 1.7843527738264582,
+      "grad_norm": 2.172466278076172,
+      "learning_rate": 8.414693251615513e-06,
+      "loss": 0.7625,
+      "step": 588
+    },
+    {
+      "epoch": 1.7873873873873873,
+      "grad_norm": 2.2140755653381348,
+      "learning_rate": 8.379772341928916e-06,
+      "loss": 0.7653,
+      "step": 589
+    },
+    {
+      "epoch": 1.7904220009483167,
+      "grad_norm": 2.1517114639282227,
+      "learning_rate": 8.344871711376202e-06,
+      "loss": 0.7744,
+      "step": 590
+    },
+    {
+      "epoch": 1.793456614509246,
+      "grad_norm": 2.1749329566955566,
+      "learning_rate": 8.309991796781512e-06,
+      "loss": 0.7663,
+      "step": 591
+    },
+    {
+      "epoch": 1.7964912280701755,
+      "grad_norm": 2.2055773735046387,
+      "learning_rate": 8.2751330347097e-06,
+      "loss": 0.7768,
+      "step": 592
+    },
+    {
+      "epoch": 1.7995258416311048,
+      "grad_norm": 2.117532253265381,
+      "learning_rate": 8.24029586146088e-06,
+      "loss": 0.7774,
+      "step": 593
+    },
+    {
+      "epoch": 1.802560455192034,
+      "grad_norm": 2.2061290740966797,
+      "learning_rate": 8.205480713064947e-06,
+      "loss": 0.7819,
+      "step": 594
+    },
+    {
+      "epoch": 1.8055950687529636,
+      "grad_norm": 2.2168619632720947,
+      "learning_rate": 8.170688025276134e-06,
+      "loss": 0.7646,
+      "step": 595
+    },
+    {
+      "epoch": 1.8086296823138928,
+      "grad_norm": 2.238391637802124,
+      "learning_rate": 8.135918233567545e-06,
+      "loss": 0.7782,
+      "step": 596
+    },
+    {
+      "epoch": 1.8116642958748221,
+      "grad_norm": 2.105041265487671,
+      "learning_rate": 8.101171773125716e-06,
+      "loss": 0.7525,
+      "step": 597
+    },
+    {
+      "epoch": 1.8146989094357515,
+      "grad_norm": 2.0234577655792236,
+      "learning_rate": 8.066449078845168e-06,
+      "loss": 0.7658,
+      "step": 598
+    },
+    {
+      "epoch": 1.817733522996681,
+      "grad_norm": 2.188751220703125,
+      "learning_rate": 8.031750585322948e-06,
+      "loss": 0.7616,
+      "step": 599
+    },
+    {
+      "epoch": 1.8207681365576103,
+      "grad_norm": 2.1460800170898438,
+      "learning_rate": 7.99707672685321e-06,
+      "loss": 0.7617,
+      "step": 600
+    },
+    {
+      "epoch": 1.8207681365576103,
+      "eval_loss": NaN,
+      "eval_runtime": 204.6188,
+      "eval_samples_per_second": 20.321,
+      "eval_steps_per_second": 5.083,
+      "step": 600
+    },
+    {
+      "epoch": 1.8238027501185394,
+      "grad_norm": 1.964609146118164,
+      "learning_rate": 7.962427937421763e-06,
+      "loss": 0.7876,
+      "step": 601
+    },
+    {
+      "epoch": 1.826837363679469,
+      "grad_norm": 2.0918426513671875,
+      "learning_rate": 7.92780465070066e-06,
+      "loss": 0.7744,
+      "step": 602
+    },
+    {
+      "epoch": 1.8298719772403982,
+      "grad_norm": 2.059992790222168,
+      "learning_rate": 7.89320730004274e-06,
+      "loss": 0.7665,
+      "step": 603
+    },
+    {
+      "epoch": 1.8329065908013278,
+      "grad_norm": 2.0450618267059326,
+      "learning_rate": 7.858636318476226e-06,
+      "loss": 0.7636,
+      "step": 604
+    },
+    {
+      "epoch": 1.835941204362257,
+      "grad_norm": 2.1291024684906006,
+      "learning_rate": 7.824092138699307e-06,
+      "loss": 0.7652,
+      "step": 605
+    },
+    {
+      "epoch": 1.8389758179231863,
+      "grad_norm": 2.0631322860717773,
+      "learning_rate": 7.789575193074703e-06,
+      "loss": 0.7819,
+      "step": 606
+    },
+    {
+      "epoch": 1.8420104314841157,
+      "grad_norm": 2.1132242679595947,
+      "learning_rate": 7.755085913624274e-06,
+      "loss": 0.7507,
+      "step": 607
+    },
+    {
+      "epoch": 1.845045045045045,
+      "grad_norm": 2.1836376190185547,
+      "learning_rate": 7.720624732023604e-06,
+      "loss": 0.7798,
+      "step": 608
+    },
+    {
+      "epoch": 1.8480796586059745,
+      "grad_norm": 2.087453842163086,
+      "learning_rate": 7.686192079596586e-06,
+      "loss": 0.7817,
+      "step": 609
+    },
+    {
+      "epoch": 1.8511142721669036,
+      "grad_norm": 2.0749568939208984,
+      "learning_rate": 7.651788387310053e-06,
+      "loss": 0.7893,
+      "step": 610
+    },
+    {
+      "epoch": 1.8541488857278332,
+      "grad_norm": 2.085495710372925,
+      "learning_rate": 7.617414085768352e-06,
+      "loss": 0.7765,
+      "step": 611
+    },
+    {
+      "epoch": 1.8571834992887624,
+      "grad_norm": 2.1710901260375977,
+      "learning_rate": 7.5830696052079754e-06,
+      "loss": 0.7745,
+      "step": 612
+    },
+    {
+      "epoch": 1.860218112849692,
+      "grad_norm": 2.12780499458313,
+      "learning_rate": 7.548755375492173e-06,
+      "loss": 0.7845,
+      "step": 613
+    },
+    {
+      "epoch": 1.8632527264106211,
+      "grad_norm": 2.240459680557251,
+      "learning_rate": 7.514471826105556e-06,
+      "loss": 0.7731,
+      "step": 614
+    },
+    {
+      "epoch": 1.8662873399715505,
+      "grad_norm": 2.130253553390503,
+      "learning_rate": 7.480219386148751e-06,
+      "loss": 0.7904,
+      "step": 615
+    },
+    {
+      "epoch": 1.8693219535324799,
+      "grad_norm": 2.216806650161743,
+      "learning_rate": 7.445998484332994e-06,
+      "loss": 0.7529,
+      "step": 616
+    },
+    {
+      "epoch": 1.8723565670934093,
+      "grad_norm": 2.250821352005005,
+      "learning_rate": 7.411809548974792e-06,
+      "loss": 0.7566,
+      "step": 617
+    },
+    {
+      "epoch": 1.8753911806543386,
+      "grad_norm": 2.088843822479248,
+      "learning_rate": 7.377653007990559e-06,
+      "loss": 0.7568,
+      "step": 618
+    },
+    {
+      "epoch": 1.8784257942152678,
+      "grad_norm": 2.108356237411499,
+      "learning_rate": 7.343529288891239e-06,
+      "loss": 0.7769,
+      "step": 619
+    },
+    {
+      "epoch": 1.8814604077761974,
+      "grad_norm": 2.1201224327087402,
+      "learning_rate": 7.3094388187769815e-06,
+      "loss": 0.7651,
+      "step": 620
+    },
+    {
+      "epoch": 1.8844950213371265,
+      "grad_norm": 2.066652536392212,
+      "learning_rate": 7.275382024331773e-06,
+      "loss": 0.7773,
+      "step": 621
+    },
+    {
+      "epoch": 1.887529634898056,
+      "grad_norm": 2.1200613975524902,
+      "learning_rate": 7.241359331818121e-06,
+      "loss": 0.7835,
+      "step": 622
+    },
+    {
+      "epoch": 1.8905642484589853,
+      "grad_norm": 2.160304546356201,
+      "learning_rate": 7.2073711670717e-06,
+      "loss": 0.7489,
+      "step": 623
+    },
+    {
+      "epoch": 1.8935988620199147,
+      "grad_norm": 2.2403597831726074,
+      "learning_rate": 7.173417955496025e-06,
+      "loss": 0.7579,
+      "step": 624
+    },
+    {
+      "epoch": 1.896633475580844,
+      "grad_norm": 2.083029270172119,
+      "learning_rate": 7.13950012205713e-06,
+      "loss": 0.7715,
+      "step": 625
+    },
+    {
+      "epoch": 1.8996680891417732,
+      "grad_norm": 2.031341075897217,
+      "learning_rate": 7.105618091278246e-06,
+      "loss": 0.7494,
+      "step": 626
+    },
+    {
+      "epoch": 1.9027027027027028,
+      "grad_norm": 2.155595541000366,
+      "learning_rate": 7.071772287234497e-06,
+      "loss": 0.7797,
+      "step": 627
+    },
+    {
+      "epoch": 1.905737316263632,
+      "grad_norm": 2.2580630779266357,
+      "learning_rate": 7.037963133547583e-06,
+      "loss": 0.7801,
+      "step": 628
+    },
+    {
+      "epoch": 1.9087719298245616,
+      "grad_norm": 2.2416763305664062,
+      "learning_rate": 7.004191053380469e-06,
+      "loss": 0.7753,
+      "step": 629
+    },
+    {
+      "epoch": 1.9118065433854907,
+      "grad_norm": 2.0690417289733887,
+      "learning_rate": 6.970456469432116e-06,
+      "loss": 0.7693,
+      "step": 630
+    },
+    {
+      "epoch": 1.91484115694642,
+      "grad_norm": 2.0331270694732666,
+      "learning_rate": 6.936759803932167e-06,
+      "loss": 0.7578,
+      "step": 631
+    },
+    {
+      "epoch": 1.9178757705073495,
+      "grad_norm": 2.1094417572021484,
+      "learning_rate": 6.903101478635663e-06,
+      "loss": 0.7589,
+      "step": 632
+    },
+    {
+      "epoch": 1.9209103840682789,
+      "grad_norm": 2.042919635772705,
+      "learning_rate": 6.869481914817779e-06,
+      "loss": 0.7659,
+      "step": 633
+    },
+    {
+      "epoch": 1.9239449976292082,
+      "grad_norm": 2.0610294342041016,
+      "learning_rate": 6.835901533268536e-06,
+      "loss": 0.7648,
+      "step": 634
+    },
+    {
+      "epoch": 1.9269796111901374,
+      "grad_norm": 2.0473015308380127,
+      "learning_rate": 6.802360754287548e-06,
+      "loss": 0.7763,
+      "step": 635
+    },
+    {
+      "epoch": 1.930014224751067,
+      "grad_norm": 2.1137821674346924,
+      "learning_rate": 6.768859997678751e-06,
+      "loss": 0.7849,
+      "step": 636
+    },
+    {
+      "epoch": 1.9330488383119961,
+      "grad_norm": 2.12430477142334,
+      "learning_rate": 6.735399682745145e-06,
+      "loss": 0.7644,
+      "step": 637
+    },
+    {
+      "epoch": 1.9360834518729255,
+      "grad_norm": 2.0620853900909424,
+      "learning_rate": 6.701980228283569e-06,
+      "loss": 0.7733,
+      "step": 638
+    },
+    {
+      "epoch": 1.939118065433855,
+      "grad_norm": 2.124861478805542,
+      "learning_rate": 6.668602052579425e-06,
+      "loss": 0.7538,
+      "step": 639
+    },
+    {
+      "epoch": 1.9421526789947843,
+      "grad_norm": 2.003215789794922,
+      "learning_rate": 6.635265573401475e-06,
+      "loss": 0.7574,
+      "step": 640
+    },
+    {
+      "epoch": 1.9451872925557137,
+      "grad_norm": 2.1591827869415283,
+      "learning_rate": 6.601971207996592e-06,
+      "loss": 0.7591,
+      "step": 641
+    },
+    {
+      "epoch": 1.9482219061166428,
+      "grad_norm": 2.1029608249664307,
+      "learning_rate": 6.5687193730845375e-06,
+      "loss": 0.7927,
+      "step": 642
+    },
+    {
+      "epoch": 1.9512565196775724,
+      "grad_norm": 2.069796085357666,
+      "learning_rate": 6.535510484852767e-06,
+      "loss": 0.7491,
+      "step": 643
+    },
+    {
+      "epoch": 1.9542911332385016,
+      "grad_norm": 2.0876927375793457,
+      "learning_rate": 6.50234495895119e-06,
+      "loss": 0.7615,
+      "step": 644
+    },
+    {
+      "epoch": 1.9573257467994312,
+      "grad_norm": 2.0911192893981934,
+      "learning_rate": 6.469223210486992e-06,
+      "loss": 0.7579,
+      "step": 645
+    },
+    {
+      "epoch": 1.9603603603603603,
+      "grad_norm": 2.2714855670928955,
+      "learning_rate": 6.4361456540194325e-06,
+      "loss": 0.7543,
+      "step": 646
+    },
+    {
+      "epoch": 1.9633949739212897,
+      "grad_norm": 2.075011968612671,
+      "learning_rate": 6.403112703554643e-06,
+      "loss": 0.752,
+      "step": 647
+    },
+    {
+      "epoch": 1.966429587482219,
+      "grad_norm": 2.097029447555542,
+      "learning_rate": 6.370124772540469e-06,
+      "loss": 0.7338,
+      "step": 648
+    },
+    {
+      "epoch": 1.9694642010431485,
+      "grad_norm": 2.147951602935791,
+      "learning_rate": 6.337182273861273e-06,
+      "loss": 0.7735,
+      "step": 649
+    },
+    {
+      "epoch": 1.9724988146040778,
+      "grad_norm": 2.0643298625946045,
+      "learning_rate": 6.3042856198327795e-06,
+      "loss": 0.7775,
+      "step": 650
+    },
+    {
+      "epoch": 1.9724988146040778,
+      "eval_loss": NaN,
+      "eval_runtime": 204.7127,
+      "eval_samples_per_second": 20.311,
+      "eval_steps_per_second": 5.08,
+      "step": 650
+    },
+    {
+      "epoch": 1.975533428165007,
+      "grad_norm": 2.043440580368042,
+      "learning_rate": 6.2714352221969155e-06,
+      "loss": 0.7593,
+      "step": 651
+    },
+    {
+      "epoch": 1.9785680417259366,
+      "grad_norm": 2.250242233276367,
+      "learning_rate": 6.238631492116643e-06,
+      "loss": 0.7657,
+      "step": 652
+    },
+    {
+      "epoch": 1.9816026552868657,
+      "grad_norm": 2.068997383117676,
+      "learning_rate": 6.205874840170833e-06,
+      "loss": 0.7774,
+      "step": 653
+    },
+    {
+      "epoch": 1.9846372688477951,
+      "grad_norm": 2.1658010482788086,
+      "learning_rate": 6.173165676349103e-06,
+      "loss": 0.7792,
+      "step": 654
+    },
+    {
+      "epoch": 1.9876718824087245,
+      "grad_norm": 2.1947412490844727,
+      "learning_rate": 6.140504410046712e-06,
+      "loss": 0.7631,
+      "step": 655
+    },
+    {
+      "epoch": 1.9907064959696539,
+      "grad_norm": 2.082087278366089,
+      "learning_rate": 6.107891450059419e-06,
+      "loss": 0.7564,
+      "step": 656
+    },
+    {
+      "epoch": 1.9937411095305833,
+      "grad_norm": 2.043905019760132,
+      "learning_rate": 6.075327204578363e-06,
+      "loss": 0.7834,
+      "step": 657
+    },
+    {
+      "epoch": 1.9967757230915124,
+      "grad_norm": 2.1089298725128174,
+      "learning_rate": 6.042812081184973e-06,
+      "loss": 0.7595,
+      "step": 658
+    },
+    {
+      "epoch": 1.999810336652442,
+      "grad_norm": 2.0419466495513916,
+      "learning_rate": 6.010346486845837e-06,
+      "loss": 0.7481,
+      "step": 659
+    },
+    {
+      "epoch": 2.002844950213371,
+      "grad_norm": 2.1154048442840576,
+      "learning_rate": 5.97793082790765e-06,
+      "loss": 0.7479,
+      "step": 660
+    },
+    {
+      "epoch": 2.0058795637743008,
+      "grad_norm": 1.9878915548324585,
+      "learning_rate": 5.945565510092086e-06,
+      "loss": 0.7547,
+      "step": 661
+    },
+    {
+      "epoch": 2.00891417733523,
+      "grad_norm": 2.1136927604675293,
+      "learning_rate": 5.913250938490744e-06,
+      "loss": 0.7294,
+      "step": 662
+    },
+    {
+      "epoch": 2.0119487908961595,
+      "grad_norm": 2.1599817276000977,
+      "learning_rate": 5.880987517560075e-06,
+      "loss": 0.744,
+      "step": 663
+    },
+    {
+      "epoch": 2.0149834044570887,
+      "grad_norm": 2.200779438018799,
+      "learning_rate": 5.84877565111631e-06,
+      "loss": 0.7353,
+      "step": 664
+    },
+    {
+      "epoch": 2.018018018018018,
+      "grad_norm": 2.05635404586792,
+      "learning_rate": 5.81661574233042e-06,
+      "loss": 0.7534,
+      "step": 665
+    },
+    {
+      "epoch": 2.0210526315789474,
+      "grad_norm": 2.193406820297241,
+      "learning_rate": 5.784508193723058e-06,
+      "loss": 0.7443,
+      "step": 666
+    },
+    {
+      "epoch": 2.0240872451398766,
+      "grad_norm": 2.120541572570801,
+      "learning_rate": 5.752453407159521e-06,
+      "loss": 0.7414,
+      "step": 667
+    },
+    {
+      "epoch": 2.027121858700806,
+      "grad_norm": 2.075017213821411,
+      "learning_rate": 5.720451783844741e-06,
+      "loss": 0.7439,
+      "step": 668
+    },
+    {
+      "epoch": 2.0301564722617353,
+      "grad_norm": 2.1643104553222656,
+      "learning_rate": 5.688503724318217e-06,
+      "loss": 0.7565,
+      "step": 669
+    },
+    {
+      "epoch": 2.033191085822665,
+      "grad_norm": 2.148364782333374,
+      "learning_rate": 5.656609628449064e-06,
+      "loss": 0.7353,
+      "step": 670
+    },
+    {
+      "epoch": 2.036225699383594,
+      "grad_norm": 2.205618143081665,
+      "learning_rate": 5.6247698954309616e-06,
+      "loss": 0.7529,
+      "step": 671
+    },
+    {
+      "epoch": 2.0392603129445233,
+      "grad_norm": 2.1727452278137207,
+      "learning_rate": 5.592984923777156e-06,
+      "loss": 0.7439,
+      "step": 672
+    },
+    {
+      "epoch": 2.042294926505453,
+      "grad_norm": 2.1897435188293457,
+      "learning_rate": 5.561255111315525e-06,
+      "loss": 0.7578,
+      "step": 673
+    },
+    {
+      "epoch": 2.045329540066382,
+      "grad_norm": 2.155768871307373,
+      "learning_rate": 5.529580855183518e-06,
+      "loss": 0.7687,
+      "step": 674
+    },
+    {
+      "epoch": 2.0483641536273116,
+      "grad_norm": 2.100489616394043,
+      "learning_rate": 5.497962551823266e-06,
+      "loss": 0.7333,
+      "step": 675
+    },
+    {
+      "epoch": 2.0513987671882408,
+      "grad_norm": 2.0687026977539062,
+      "learning_rate": 5.466400596976568e-06,
+      "loss": 0.745,
+      "step": 676
+    },
+    {
+      "epoch": 2.0544333807491704,
+      "grad_norm": 2.144482135772705,
+      "learning_rate": 5.434895385679937e-06,
+      "loss": 0.7369,
+      "step": 677
+    },
+    {
+      "epoch": 2.0574679943100995,
+      "grad_norm": 2.1690595149993896,
+      "learning_rate": 5.403447312259702e-06,
+      "loss": 0.7443,
+      "step": 678
+    },
+    {
+      "epoch": 2.060502607871029,
+      "grad_norm": 2.0905041694641113,
+      "learning_rate": 5.3720567703270135e-06,
+      "loss": 0.7333,
+      "step": 679
+    },
+    {
+      "epoch": 2.0635372214319583,
+      "grad_norm": 2.205411911010742,
+      "learning_rate": 5.340724152772956e-06,
+      "loss": 0.7399,
+      "step": 680
+    },
+    {
+      "epoch": 2.0665718349928874,
+      "grad_norm": 2.3007540702819824,
+      "learning_rate": 5.3094498517636324e-06,
+      "loss": 0.7452,
+      "step": 681
+    },
+    {
+      "epoch": 2.069606448553817,
+      "grad_norm": 2.2882180213928223,
+      "learning_rate": 5.278234258735215e-06,
+      "loss": 0.7285,
+      "step": 682
+    },
+    {
+      "epoch": 2.072641062114746,
+      "grad_norm": 2.2169337272644043,
+      "learning_rate": 5.247077764389099e-06,
+      "loss": 0.7437,
+      "step": 683
+    },
+    {
+      "epoch": 2.075675675675676,
+      "grad_norm": 2.1462368965148926,
+      "learning_rate": 5.215980758686978e-06,
+      "loss": 0.7366,
+      "step": 684
+    },
+    {
+      "epoch": 2.078710289236605,
+      "grad_norm": 2.200030565261841,
+      "learning_rate": 5.18494363084596e-06,
+      "loss": 0.7546,
+      "step": 685
+    },
+    {
+      "epoch": 2.0817449027975345,
+      "grad_norm": 2.1369688510894775,
+      "learning_rate": 5.153966769333734e-06,
+      "loss": 0.7388,
+      "step": 686
+    },
+    {
+      "epoch": 2.0847795163584637,
+      "grad_norm": 2.191958427429199,
+      "learning_rate": 5.1230505618636575e-06,
+      "loss": 0.7297,
+      "step": 687
+    },
+    {
+      "epoch": 2.0878141299193933,
+      "grad_norm": 2.1999082565307617,
+      "learning_rate": 5.092195395389937e-06,
+      "loss": 0.7428,
+      "step": 688
+    },
+    {
+      "epoch": 2.0908487434803225,
+      "grad_norm": 2.373140335083008,
+      "learning_rate": 5.061401656102791e-06,
+      "loss": 0.7585,
+      "step": 689
+    },
+    {
+      "epoch": 2.0938833570412516,
+      "grad_norm": 2.2395219802856445,
+      "learning_rate": 5.030669729423572e-06,
+      "loss": 0.7382,
+      "step": 690
+    },
+    {
+      "epoch": 2.096917970602181,
+      "grad_norm": 2.1638119220733643,
+      "learning_rate": 5.000000000000003e-06,
+      "loss": 0.7538,
+      "step": 691
+    },
+    {
+      "epoch": 2.0999525841631104,
+      "grad_norm": 2.173945188522339,
+      "learning_rate": 4.969392851701306e-06,
+      "loss": 0.7403,
+      "step": 692
+    },
+    {
+      "epoch": 2.10298719772404,
+      "grad_norm": 2.2739264965057373,
+      "learning_rate": 4.938848667613436e-06,
+      "loss": 0.7459,
+      "step": 693
+    },
+    {
+      "epoch": 2.106021811284969,
+      "grad_norm": 2.2303757667541504,
+      "learning_rate": 4.908367830034284e-06,
+      "loss": 0.717,
+      "step": 694
+    },
+    {
+      "epoch": 2.1090564248458987,
+      "grad_norm": 2.083385944366455,
+      "learning_rate": 4.8779507204688595e-06,
+      "loss": 0.747,
+      "step": 695
+    },
+    {
+      "epoch": 2.112091038406828,
+      "grad_norm": 2.252095937728882,
+      "learning_rate": 4.84759771962455e-06,
+      "loss": 0.7414,
+      "step": 696
+    },
+    {
+      "epoch": 2.115125651967757,
+      "grad_norm": 2.192476272583008,
+      "learning_rate": 4.817309207406347e-06,
+      "loss": 0.7078,
+      "step": 697
+    },
+    {
+      "epoch": 2.1181602655286866,
+      "grad_norm": 2.170762062072754,
+      "learning_rate": 4.787085562912076e-06,
+      "loss": 0.7482,
+      "step": 698
+    },
+    {
+      "epoch": 2.121194879089616,
+      "grad_norm": 2.1275007724761963,
+      "learning_rate": 4.756927164427685e-06,
+      "loss": 0.7415,
+      "step": 699
+    },
+    {
+      "epoch": 2.1242294926505454,
+      "grad_norm": 2.1509687900543213,
+      "learning_rate": 4.726834389422461e-06,
+      "loss": 0.7326,
+      "step": 700
+    },
+    {
+      "epoch": 2.1242294926505454,
+      "eval_loss": NaN,
+      "eval_runtime": 204.0742,
+      "eval_samples_per_second": 20.375,
+      "eval_steps_per_second": 5.096,
+      "step": 700
+    },
+    {
+      "epoch": 2.1272641062114745,
+      "grad_norm": 2.0226032733917236,
+      "learning_rate": 4.696807614544352e-06,
+      "loss": 0.7411,
+      "step": 701
+    },
+    {
+      "epoch": 2.130298719772404,
+      "grad_norm": 2.1866512298583984,
+      "learning_rate": 4.666847215615225e-06,
+      "loss": 0.7394,
+      "step": 702
+    },
+    {
+      "epoch": 2.1333333333333333,
+      "grad_norm": 2.2013630867004395,
+      "learning_rate": 4.636953567626176e-06,
+      "loss": 0.7356,
+      "step": 703
+    },
+    {
+      "epoch": 2.1363679468942625,
+      "grad_norm": 2.4076430797576904,
+      "learning_rate": 4.607127044732827e-06,
+      "loss": 0.7501,
+      "step": 704
+    },
+    {
+      "epoch": 2.139402560455192,
+      "grad_norm": 2.1296753883361816,
+      "learning_rate": 4.57736802025065e-06,
+      "loss": 0.7399,
+      "step": 705
+    },
+    {
+      "epoch": 2.142437174016121,
+      "grad_norm": 2.22512149810791,
+      "learning_rate": 4.5476768666502895e-06,
+      "loss": 0.7553,
+      "step": 706
+    },
+    {
+      "epoch": 2.145471787577051,
+      "grad_norm": 2.178199052810669,
+      "learning_rate": 4.518053955552903e-06,
+      "loss": 0.7458,
+      "step": 707
+    },
+    {
+      "epoch": 2.14850640113798,
+      "grad_norm": 2.0975606441497803,
+      "learning_rate": 4.488499657725511e-06,
+      "loss": 0.7234,
+      "step": 708
+    },
+    {
+      "epoch": 2.1515410146989096,
+      "grad_norm": 2.1103413105010986,
+      "learning_rate": 4.459014343076356e-06,
+      "loss": 0.7431,
+      "step": 709
+    },
+    {
+      "epoch": 2.1545756282598387,
+      "grad_norm": 2.2594456672668457,
+      "learning_rate": 4.429598380650266e-06,
+      "loss": 0.7578,
+      "step": 710
+    },
+    {
+      "epoch": 2.1576102418207683,
+      "grad_norm": 2.2270963191986084,
+      "learning_rate": 4.400252138624047e-06,
+      "loss": 0.7461,
+      "step": 711
+    },
+    {
+      "epoch": 2.1606448553816975,
+      "grad_norm": 2.191631555557251,
+      "learning_rate": 4.370975984301866e-06,
+      "loss": 0.7425,
+      "step": 712
+    },
+    {
+      "epoch": 2.1636794689426266,
+      "grad_norm": 2.124616861343384,
+      "learning_rate": 4.341770284110655e-06,
+      "loss": 0.7495,
+      "step": 713
+    },
+    {
+      "epoch": 2.1667140825035562,
+      "grad_norm": 2.18497371673584,
+      "learning_rate": 4.312635403595532e-06,
+      "loss": 0.7239,
+      "step": 714
+    },
+    {
+      "epoch": 2.1697486960644854,
+      "grad_norm": 2.149658441543579,
+      "learning_rate": 4.283571707415214e-06,
+      "loss": 0.7325,
+      "step": 715
+    },
+    {
+      "epoch": 2.172783309625415,
+      "grad_norm": 2.133171558380127,
+      "learning_rate": 4.25457955933746e-06,
+      "loss": 0.7324,
+      "step": 716
+    },
+    {
+      "epoch": 2.175817923186344,
+      "grad_norm": 2.15606689453125,
+      "learning_rate": 4.2256593222345185e-06,
+      "loss": 0.7273,
+      "step": 717
+    },
+    {
+      "epoch": 2.1788525367472737,
+      "grad_norm": 2.237285614013672,
+      "learning_rate": 4.196811358078585e-06,
+      "loss": 0.7463,
+      "step": 718
+    },
+    {
+      "epoch": 2.181887150308203,
+      "grad_norm": 2.267974853515625,
+      "learning_rate": 4.168036027937267e-06,
+      "loss": 0.7405,
+      "step": 719
+    },
+    {
+      "epoch": 2.1849217638691325,
+      "grad_norm": 2.0772793292999268,
+      "learning_rate": 4.139333691969071e-06,
+      "loss": 0.7418,
+      "step": 720
+    },
+    {
+      "epoch": 2.1879563774300617,
+      "grad_norm": 2.1171600818634033,
+      "learning_rate": 4.1107047094188946e-06,
+      "loss": 0.7222,
+      "step": 721
+    },
+    {
+      "epoch": 2.190990990990991,
+      "grad_norm": 2.3039426803588867,
+      "learning_rate": 4.082149438613514e-06,
+      "loss": 0.731,
+      "step": 722
+    },
+    {
+      "epoch": 2.1940256045519204,
+      "grad_norm": 2.207404375076294,
+      "learning_rate": 4.053668236957135e-06,
+      "loss": 0.7499,
+      "step": 723
+    },
+    {
+      "epoch": 2.1970602181128496,
+      "grad_norm": 2.1732053756713867,
+      "learning_rate": 4.025261460926877e-06,
+      "loss": 0.7617,
+      "step": 724
+    },
+    {
+      "epoch": 2.200094831673779,
+      "grad_norm": 2.537623167037964,
+      "learning_rate": 3.996929466068344e-06,
+      "loss": 0.72,
+      "step": 725
+    },
+    {
+      "epoch": 2.2031294452347083,
+      "grad_norm": 2.2418384552001953,
+      "learning_rate": 3.96867260699116e-06,
+      "loss": 0.727,
+      "step": 726
+    },
+    {
+      "epoch": 2.206164058795638,
+      "grad_norm": 2.189655065536499,
+      "learning_rate": 3.940491237364519e-06,
+      "loss": 0.7321,
+      "step": 727
+    },
+    {
+      "epoch": 2.209198672356567,
+      "grad_norm": 2.1620960235595703,
+      "learning_rate": 3.912385709912794e-06,
+      "loss": 0.7421,
+      "step": 728
+    },
+    {
+      "epoch": 2.2122332859174962,
+      "grad_norm": 2.216374397277832,
+      "learning_rate": 3.884356376411089e-06,
+      "loss": 0.7632,
+      "step": 729
+    },
+    {
+      "epoch": 2.215267899478426,
+      "grad_norm": 2.152038335800171,
+      "learning_rate": 3.8564035876808405e-06,
+      "loss": 0.7349,
+      "step": 730
+    },
+    {
+      "epoch": 2.218302513039355,
+      "grad_norm": 2.1349170207977295,
+      "learning_rate": 3.828527693585451e-06,
+      "loss": 0.7373,
+      "step": 731
+    },
+    {
+      "epoch": 2.2213371266002846,
+      "grad_norm": 2.1102511882781982,
+      "learning_rate": 3.8007290430258712e-06,
+      "loss": 0.7278,
+      "step": 732
+    },
+    {
+      "epoch": 2.2243717401612138,
+      "grad_norm": 2.124293088912964,
+      "learning_rate": 3.7730079839362755e-06,
+      "loss": 0.7315,
+      "step": 733
+    },
+    {
+      "epoch": 2.2274063537221434,
+      "grad_norm": 2.0747625827789307,
+      "learning_rate": 3.7453648632796745e-06,
+      "loss": 0.7391,
+      "step": 734
+    },
+    {
+      "epoch": 2.2304409672830725,
+      "grad_norm": 2.1584184169769287,
+      "learning_rate": 3.7178000270435765e-06,
+      "loss": 0.7526,
+      "step": 735
+    },
+    {
+      "epoch": 2.2334755808440017,
+      "grad_norm": 2.0907371044158936,
+      "learning_rate": 3.690313820235686e-06,
+      "loss": 0.7176,
+      "step": 736
+    },
+    {
+      "epoch": 2.2365101944049313,
+      "grad_norm": 2.100431442260742,
+      "learning_rate": 3.662906586879542e-06,
+      "loss": 0.7619,
+      "step": 737
+    },
+    {
+      "epoch": 2.2395448079658604,
+      "grad_norm": 2.1690993309020996,
+      "learning_rate": 3.6355786700102426e-06,
+      "loss": 0.7385,
+      "step": 738
+    },
+    {
+      "epoch": 2.24257942152679,
+      "grad_norm": 2.2165980339050293,
+      "learning_rate": 3.6083304116701535e-06,
+      "loss": 0.7577,
+      "step": 739
+    },
+    {
+      "epoch": 2.245614035087719,
+      "grad_norm": 2.1825928688049316,
+      "learning_rate": 3.5811621529045927e-06,
+      "loss": 0.7353,
+      "step": 740
+    },
+    {
+      "epoch": 2.2486486486486488,
+      "grad_norm": 2.1485071182250977,
+      "learning_rate": 3.5540742337576083e-06,
+      "loss": 0.7534,
+      "step": 741
+    },
+    {
+      "epoch": 2.251683262209578,
+      "grad_norm": 2.193574905395508,
+      "learning_rate": 3.5270669932676926e-06,
+      "loss": 0.7395,
+      "step": 742
+    },
+    {
+      "epoch": 2.2547178757705075,
+      "grad_norm": 2.152099847793579,
+      "learning_rate": 3.5001407694635326e-06,
+      "loss": 0.7265,
+      "step": 743
+    },
+    {
+      "epoch": 2.2577524893314367,
+      "grad_norm": 2.2395544052124023,
+      "learning_rate": 3.4732958993598153e-06,
+      "loss": 0.7482,
+      "step": 744
+    },
+    {
+      "epoch": 2.2607871028923663,
+      "grad_norm": 2.209214448928833,
+      "learning_rate": 3.446532718952966e-06,
+      "loss": 0.7366,
+      "step": 745
+    },
+    {
+      "epoch": 2.2638217164532954,
+      "grad_norm": 2.282007932662964,
+      "learning_rate": 3.4198515632169703e-06,
+      "loss": 0.7641,
+      "step": 746
+    },
+    {
+      "epoch": 2.2668563300142246,
+      "grad_norm": 2.1614651679992676,
+      "learning_rate": 3.3932527660991877e-06,
+      "loss": 0.7502,
+      "step": 747
+    },
+    {
+      "epoch": 2.269890943575154,
+      "grad_norm": 2.306121826171875,
+      "learning_rate": 3.3667366605161322e-06,
+      "loss": 0.7464,
+      "step": 748
+    },
+    {
+      "epoch": 2.2729255571360834,
+      "grad_norm": 2.2220146656036377,
+      "learning_rate": 3.340303578349361e-06,
+      "loss": 0.7497,
+      "step": 749
+    },
+    {
+      "epoch": 2.275960170697013,
+      "grad_norm": 2.1184372901916504,
+      "learning_rate": 3.313953850441266e-06,
+      "loss": 0.7434,
+      "step": 750
+    },
+    {
+      "epoch": 2.275960170697013,
+      "eval_loss": NaN,
+      "eval_runtime": 203.9554,
+      "eval_samples_per_second": 20.387,
+      "eval_steps_per_second": 5.099,
+      "step": 750
+    },
+    {
+      "epoch": 2.278994784257942,
+      "grad_norm": 2.2895419597625732,
+      "learning_rate": 3.2876878065909714e-06,
+      "loss": 0.7362,
+      "step": 751
+    },
+    {
+      "epoch": 2.2820293978188717,
+      "grad_norm": 2.148529291152954,
+      "learning_rate": 3.2615057755502e-06,
+      "loss": 0.7558,
+      "step": 752
+    },
+    {
+      "epoch": 2.285064011379801,
+      "grad_norm": 2.1223018169403076,
+      "learning_rate": 3.2354080850191328e-06,
+      "loss": 0.7591,
+      "step": 753
+    },
+    {
+      "epoch": 2.28809862494073,
+      "grad_norm": 2.2450056076049805,
+      "learning_rate": 3.2093950616423397e-06,
+      "loss": 0.7335,
+      "step": 754
+    },
+    {
+      "epoch": 2.2911332385016596,
+      "grad_norm": 2.173351526260376,
+      "learning_rate": 3.1834670310046735e-06,
+      "loss": 0.7546,
+      "step": 755
+    },
+    {
+      "epoch": 2.2941678520625888,
+      "grad_norm": 2.0341904163360596,
+      "learning_rate": 3.157624317627195e-06,
+      "loss": 0.7447,
+      "step": 756
+    },
+    {
+      "epoch": 2.2972024656235184,
+      "grad_norm": 2.0900380611419678,
+      "learning_rate": 3.1318672449631283e-06,
+      "loss": 0.7544,
+      "step": 757
+    },
+    {
+      "epoch": 2.3002370791844475,
+      "grad_norm": 2.2536418437957764,
+      "learning_rate": 3.106196135393782e-06,
+      "loss": 0.7326,
+      "step": 758
+    },
+    {
+      "epoch": 2.303271692745377,
+      "grad_norm": 2.1906511783599854,
+      "learning_rate": 3.0806113102245395e-06,
+      "loss": 0.7522,
+      "step": 759
+    },
+    {
+      "epoch": 2.3063063063063063,
+      "grad_norm": 2.1476638317108154,
+      "learning_rate": 3.055113089680829e-06,
+      "loss": 0.7423,
+      "step": 760
+    },
+    {
+      "epoch": 2.3093409198672354,
+      "grad_norm": 2.1808698177337646,
+      "learning_rate": 3.029701792904117e-06,
+      "loss": 0.7612,
+      "step": 761
+    },
+    {
+      "epoch": 2.312375533428165,
+      "grad_norm": 2.085947036743164,
+      "learning_rate": 3.00437773794791e-06,
+      "loss": 0.7439,
+      "step": 762
+    },
+    {
+      "epoch": 2.315410146989094,
+      "grad_norm": 2.1122984886169434,
+      "learning_rate": 2.979141241773775e-06,
+      "loss": 0.7383,
+      "step": 763
+    },
+    {
+      "epoch": 2.318444760550024,
+      "grad_norm": 2.1743266582489014,
+      "learning_rate": 2.953992620247379e-06,
+      "loss": 0.7458,
+      "step": 764
+    },
+    {
+      "epoch": 2.321479374110953,
+      "grad_norm": 2.1469690799713135,
+      "learning_rate": 2.9289321881345257e-06,
+      "loss": 0.756,
+      "step": 765
+    },
+    {
+      "epoch": 2.3245139876718826,
+      "grad_norm": 2.077439069747925,
+      "learning_rate": 2.9039602590972203e-06,
+      "loss": 0.7316,
+      "step": 766
+    },
+    {
+      "epoch": 2.3275486012328117,
+      "grad_norm": 2.4407780170440674,
+      "learning_rate": 2.879077145689746e-06,
+      "loss": 0.7343,
+      "step": 767
+    },
+    {
+      "epoch": 2.330583214793741,
+      "grad_norm": 2.1479241847991943,
+      "learning_rate": 2.8542831593547483e-06,
+      "loss": 0.7407,
+      "step": 768
+    },
+    {
+      "epoch": 2.3336178283546705,
+      "grad_norm": 2.2073941230773926,
+      "learning_rate": 2.829578610419337e-06,
+      "loss": 0.7471,
+      "step": 769
+    },
+    {
+      "epoch": 2.3366524419155996,
+      "grad_norm": 2.0292317867279053,
+      "learning_rate": 2.8049638080912045e-06,
+      "loss": 0.7332,
+      "step": 770
+    },
+    {
+      "epoch": 2.339687055476529,
+      "grad_norm": 2.2681589126586914,
+      "learning_rate": 2.780439060454756e-06,
+      "loss": 0.7436,
+      "step": 771
+    },
+    {
+      "epoch": 2.3427216690374584,
+      "grad_norm": 2.057676076889038,
+      "learning_rate": 2.75600467446725e-06,
+      "loss": 0.7352,
+      "step": 772
+    },
+    {
+      "epoch": 2.345756282598388,
+      "grad_norm": 2.2097206115722656,
+      "learning_rate": 2.7316609559549568e-06,
+      "loss": 0.7275,
+      "step": 773
+    },
+    {
+      "epoch": 2.348790896159317,
+      "grad_norm": 2.3039231300354004,
+      "learning_rate": 2.707408209609339e-06,
+      "loss": 0.7556,
+      "step": 774
+    },
+    {
+      "epoch": 2.3518255097202467,
+      "grad_norm": 2.097167730331421,
+      "learning_rate": 2.683246738983217e-06,
+      "loss": 0.7404,
+      "step": 775
+    },
+    {
+      "epoch": 2.354860123281176,
+      "grad_norm": 2.083677053451538,
+      "learning_rate": 2.6591768464870016e-06,
+      "loss": 0.7359,
+      "step": 776
+    },
+    {
+      "epoch": 2.3578947368421055,
+      "grad_norm": 2.0094263553619385,
+      "learning_rate": 2.6351988333848787e-06,
+      "loss": 0.7297,
+      "step": 777
+    },
+    {
+      "epoch": 2.3609293504030346,
+      "grad_norm": 2.13173770904541,
+      "learning_rate": 2.611312999791055e-06,
+      "loss": 0.7315,
+      "step": 778
+    },
+    {
+      "epoch": 2.363963963963964,
+      "grad_norm": 2.1343095302581787,
+      "learning_rate": 2.587519644666001e-06,
+      "loss": 0.7309,
+      "step": 779
+    },
+    {
+      "epoch": 2.3669985775248934,
+      "grad_norm": 2.122774600982666,
+      "learning_rate": 2.5638190658126937e-06,
+      "loss": 0.7428,
+      "step": 780
+    },
+    {
+      "epoch": 2.3700331910858226,
+      "grad_norm": 2.1483750343322754,
+      "learning_rate": 2.5402115598729182e-06,
+      "loss": 0.7386,
+      "step": 781
+    },
+    {
+      "epoch": 2.373067804646752,
+      "grad_norm": 2.189150810241699,
+      "learning_rate": 2.5166974223235295e-06,
+      "loss": 0.7237,
+      "step": 782
+    },
+    {
+      "epoch": 2.3761024182076813,
+      "grad_norm": 2.156003475189209,
+      "learning_rate": 2.493276947472756e-06,
+      "loss": 0.7614,
+      "step": 783
+    },
+    {
+      "epoch": 2.379137031768611,
+      "grad_norm": 2.100486993789673,
+      "learning_rate": 2.4699504284565413e-06,
+      "loss": 0.7543,
+      "step": 784
+    },
+    {
+      "epoch": 2.38217164532954,
+      "grad_norm": 2.1420738697052,
+      "learning_rate": 2.446718157234832e-06,
+      "loss": 0.7475,
+      "step": 785
+    },
+    {
+      "epoch": 2.385206258890469,
+      "grad_norm": 2.205798625946045,
+      "learning_rate": 2.4235804245879723e-06,
+      "loss": 0.7362,
+      "step": 786
+    },
+    {
+      "epoch": 2.388240872451399,
+      "grad_norm": 2.193894147872925,
+      "learning_rate": 2.4005375201130275e-06,
+      "loss": 0.7456,
+      "step": 787
+    },
+    {
+      "epoch": 2.391275486012328,
+      "grad_norm": 2.1630735397338867,
+      "learning_rate": 2.3775897322201667e-06,
+      "loss": 0.7553,
+      "step": 788
+    },
+    {
+      "epoch": 2.3943100995732576,
+      "grad_norm": 2.247066020965576,
+      "learning_rate": 2.354737348129077e-06,
+      "loss": 0.7324,
+      "step": 789
+    },
+    {
+      "epoch": 2.3973447131341867,
+      "grad_norm": 2.1332616806030273,
+      "learning_rate": 2.33198065386533e-06,
+      "loss": 0.7413,
+      "step": 790
+    },
+    {
+      "epoch": 2.4003793266951163,
+      "grad_norm": 2.1438822746276855,
+      "learning_rate": 2.3093199342568316e-06,
+      "loss": 0.7215,
+      "step": 791
+    },
+    {
+      "epoch": 2.4034139402560455,
+      "grad_norm": 2.1765286922454834,
+      "learning_rate": 2.2867554729302545e-06,
+      "loss": 0.7362,
+      "step": 792
+    },
+    {
+      "epoch": 2.4064485538169746,
+      "grad_norm": 2.059136390686035,
+      "learning_rate": 2.2642875523074613e-06,
+      "loss": 0.7373,
+      "step": 793
+    },
+    {
+      "epoch": 2.4094831673779042,
+      "grad_norm": 2.195065975189209,
+      "learning_rate": 2.2419164536020112e-06,
+      "loss": 0.7363,
+      "step": 794
+    },
+    {
+      "epoch": 2.4125177809388334,
+      "grad_norm": 2.2990760803222656,
+      "learning_rate": 2.2196424568156073e-06,
+      "loss": 0.7348,
+      "step": 795
+    },
+    {
+      "epoch": 2.415552394499763,
+      "grad_norm": 2.251936435699463,
+      "learning_rate": 2.197465840734596e-06,
+      "loss": 0.7257,
+      "step": 796
+    },
+    {
+      "epoch": 2.418587008060692,
+      "grad_norm": 2.142132520675659,
+      "learning_rate": 2.1753868829265046e-06,
+      "loss": 0.7526,
+      "step": 797
+    },
+    {
+      "epoch": 2.4216216216216218,
+      "grad_norm": 2.235971450805664,
+      "learning_rate": 2.1534058597365284e-06,
+      "loss": 0.7402,
+      "step": 798
+    },
+    {
+      "epoch": 2.424656235182551,
+      "grad_norm": 2.1040003299713135,
+      "learning_rate": 2.1315230462840985e-06,
+      "loss": 0.7271,
+      "step": 799
+    },
+    {
+      "epoch": 2.4276908487434805,
+      "grad_norm": 2.153122663497925,
+      "learning_rate": 2.1097387164594406e-06,
+      "loss": 0.7274,
+      "step": 800
+    },
+    {
+      "epoch": 2.4276908487434805,
+      "eval_loss": NaN,
+      "eval_runtime": 204.5921,
+      "eval_samples_per_second": 20.323,
+      "eval_steps_per_second": 5.083,
+      "step": 800
+    },
+    {
+      "epoch": 2.4307254623044097,
+      "grad_norm": 2.2538816928863525,
+      "learning_rate": 2.0880531429201146e-06,
+      "loss": 0.749,
+      "step": 801
+    },
+    {
+      "epoch": 2.4337600758653393,
+      "grad_norm": 2.194967031478882,
+      "learning_rate": 2.0664665970876496e-06,
+      "loss": 0.7395,
+      "step": 802
+    },
+    {
+      "epoch": 2.4367946894262684,
+      "grad_norm": 2.1779520511627197,
+      "learning_rate": 2.0449793491441026e-06,
+      "loss": 0.7476,
+      "step": 803
+    },
+    {
+      "epoch": 2.4398293029871976,
+      "grad_norm": 2.243781805038452,
+      "learning_rate": 2.0235916680287015e-06,
+      "loss": 0.7268,
+      "step": 804
+    },
+    {
+      "epoch": 2.442863916548127,
+      "grad_norm": 2.1497793197631836,
+      "learning_rate": 2.0023038214344827e-06,
+      "loss": 0.76,
+      "step": 805
+    },
+    {
+      "epoch": 2.4458985301090563,
+      "grad_norm": 2.209360122680664,
+      "learning_rate": 1.9811160758049163e-06,
+      "loss": 0.7133,
+      "step": 806
+    },
+    {
+      "epoch": 2.448933143669986,
+      "grad_norm": 2.1285481452941895,
+      "learning_rate": 1.960028696330596e-06,
+      "loss": 0.7349,
+      "step": 807
+    },
+    {
+      "epoch": 2.451967757230915,
+      "grad_norm": 2.16249680519104,
+      "learning_rate": 1.9390419469459066e-06,
+      "loss": 0.7392,
+      "step": 808
+    },
+    {
+      "epoch": 2.4550023707918447,
+      "grad_norm": 2.1789779663085938,
+      "learning_rate": 1.9181560903257234e-06,
+      "loss": 0.7299,
+      "step": 809
+    },
+    {
+      "epoch": 2.458036984352774,
+      "grad_norm": 2.196904420852661,
+      "learning_rate": 1.8973713878821343e-06,
+      "loss": 0.74,
+      "step": 810
+    },
+    {
+      "epoch": 2.461071597913703,
+      "grad_norm": 2.087092399597168,
+      "learning_rate": 1.8766880997611424e-06,
+      "loss": 0.7503,
+      "step": 811
+    },
+    {
+      "epoch": 2.4641062114746326,
+      "grad_norm": 2.023298978805542,
+      "learning_rate": 1.8561064848394384e-06,
+      "loss": 0.7409,
+      "step": 812
+    },
+    {
+      "epoch": 2.4671408250355618,
+      "grad_norm": 2.2470693588256836,
+      "learning_rate": 1.8356268007211442e-06,
+      "loss": 0.7553,
+      "step": 813
+    },
+    {
+      "epoch": 2.4701754385964914,
+      "grad_norm": 2.0922911167144775,
+      "learning_rate": 1.8152493037345942e-06,
+      "loss": 0.7257,
+      "step": 814
+    },
+    {
+      "epoch": 2.4732100521574205,
+      "grad_norm": 2.100867986679077,
+      "learning_rate": 1.7949742489291256e-06,
+      "loss": 0.7275,
+      "step": 815
+    },
+    {
+      "epoch": 2.47624466571835,
+      "grad_norm": 2.309476375579834,
+      "learning_rate": 1.7748018900718856e-06,
+      "loss": 0.7349,
+      "step": 816
+    },
+    {
+      "epoch": 2.4792792792792793,
+      "grad_norm": 2.1686556339263916,
+      "learning_rate": 1.7547324796446553e-06,
+      "loss": 0.746,
+      "step": 817
+    },
+    {
+      "epoch": 2.4823138928402084,
+      "grad_norm": 2.161126136779785,
+      "learning_rate": 1.7347662688406907e-06,
+      "loss": 0.7637,
+      "step": 818
+    },
+    {
+      "epoch": 2.485348506401138,
+      "grad_norm": 2.0931811332702637,
+      "learning_rate": 1.7149035075615795e-06,
+      "loss": 0.7417,
+      "step": 819
+    },
+    {
+      "epoch": 2.488383119962067,
+      "grad_norm": 2.220621347427368,
+      "learning_rate": 1.6951444444141084e-06,
+      "loss": 0.7426,
+      "step": 820
+    },
+    {
+      "epoch": 2.491417733522997,
+      "grad_norm": 2.1453609466552734,
+      "learning_rate": 1.6754893267071593e-06,
+      "loss": 0.7277,
+      "step": 821
+    },
+    {
+      "epoch": 2.494452347083926,
+      "grad_norm": 2.204213857650757,
+      "learning_rate": 1.6559384004486057e-06,
+      "loss": 0.7456,
+      "step": 822
+    },
+    {
+      "epoch": 2.4974869606448555,
+      "grad_norm": 2.1252691745758057,
+      "learning_rate": 1.6364919103422394e-06,
+      "loss": 0.7423,
+      "step": 823
+    },
+    {
+      "epoch": 2.5005215742057847,
+      "grad_norm": 2.285757064819336,
+      "learning_rate": 1.6171500997847056e-06,
+      "loss": 0.7377,
+      "step": 824
+    },
+    {
+      "epoch": 2.503556187766714,
+      "grad_norm": 2.1519298553466797,
+      "learning_rate": 1.5979132108624572e-06,
+      "loss": 0.7397,
+      "step": 825
+    },
+    {
+      "epoch": 2.5065908013276434,
+      "grad_norm": 2.1684153079986572,
+      "learning_rate": 1.5787814843487226e-06,
+      "loss": 0.7355,
+      "step": 826
+    },
+    {
+      "epoch": 2.509625414888573,
+      "grad_norm": 2.1787893772125244,
+      "learning_rate": 1.5597551597004968e-06,
+      "loss": 0.7356,
+      "step": 827
+    },
+    {
+      "epoch": 2.512660028449502,
+      "grad_norm": 2.22249436378479,
+      "learning_rate": 1.5408344750555382e-06,
+      "loss": 0.7439,
+      "step": 828
+    },
+    {
+      "epoch": 2.5156946420104314,
+      "grad_norm": 2.2051734924316406,
+      "learning_rate": 1.522019667229393e-06,
+      "loss": 0.7271,
+      "step": 829
+    },
+    {
+      "epoch": 2.518729255571361,
+      "grad_norm": 2.0986456871032715,
+      "learning_rate": 1.5033109717124284e-06,
+      "loss": 0.7385,
+      "step": 830
+    },
+    {
+      "epoch": 2.52176386913229,
+      "grad_norm": 2.115734100341797,
+      "learning_rate": 1.4847086226668871e-06,
+      "loss": 0.7317,
+      "step": 831
+    },
+    {
+      "epoch": 2.5247984826932193,
+      "grad_norm": 2.02058482170105,
+      "learning_rate": 1.4662128529239572e-06,
+      "loss": 0.7524,
+      "step": 832
+    },
+    {
+      "epoch": 2.527833096254149,
+      "grad_norm": 2.1982414722442627,
+      "learning_rate": 1.4478238939808454e-06,
+      "loss": 0.7409,
+      "step": 833
+    },
+    {
+      "epoch": 2.5308677098150785,
+      "grad_norm": 2.1642115116119385,
+      "learning_rate": 1.4295419759979079e-06,
+      "loss": 0.7365,
+      "step": 834
+    },
+    {
+      "epoch": 2.5339023233760076,
+      "grad_norm": 2.118788003921509,
+      "learning_rate": 1.4113673277957395e-06,
+      "loss": 0.7276,
+      "step": 835
+    },
+    {
+      "epoch": 2.536936936936937,
+      "grad_norm": 2.0929243564605713,
+      "learning_rate": 1.393300176852327e-06,
+      "loss": 0.7449,
+      "step": 836
+    },
+    {
+      "epoch": 2.5399715504978664,
+      "grad_norm": 2.1339359283447266,
+      "learning_rate": 1.3753407493001968e-06,
+      "loss": 0.7241,
+      "step": 837
+    },
+    {
+      "epoch": 2.5430061640587955,
+      "grad_norm": 2.15981125831604,
+      "learning_rate": 1.3574892699235798e-06,
+      "loss": 0.7373,
+      "step": 838
+    },
+    {
+      "epoch": 2.546040777619725,
+      "grad_norm": 2.184105396270752,
+      "learning_rate": 1.339745962155613e-06,
+      "loss": 0.7283,
+      "step": 839
+    },
+    {
+      "epoch": 2.5490753911806543,
+      "grad_norm": 2.0236942768096924,
+      "learning_rate": 1.3221110480755306e-06,
+      "loss": 0.7279,
+      "step": 840
+    },
+    {
+      "epoch": 2.552110004741584,
+      "grad_norm": 2.0451905727386475,
+      "learning_rate": 1.3045847484058748e-06,
+      "loss": 0.7417,
+      "step": 841
+    },
+    {
+      "epoch": 2.555144618302513,
+      "grad_norm": 2.113330602645874,
+      "learning_rate": 1.287167282509767e-06,
+      "loss": 0.7604,
+      "step": 842
+    },
+    {
+      "epoch": 2.558179231863442,
+      "grad_norm": 2.0830495357513428,
+      "learning_rate": 1.2698588683881185e-06,
+      "loss": 0.7579,
+      "step": 843
+    },
+    {
+      "epoch": 2.561213845424372,
+      "grad_norm": 2.065068483352661,
+      "learning_rate": 1.252659722676941e-06,
+      "loss": 0.7268,
+      "step": 844
+    },
+    {
+      "epoch": 2.564248458985301,
+      "grad_norm": 2.157485008239746,
+      "learning_rate": 1.2355700606446119e-06,
+      "loss": 0.7305,
+      "step": 845
+    },
+    {
+      "epoch": 2.5672830725462306,
+      "grad_norm": 2.135056257247925,
+      "learning_rate": 1.2185900961891794e-06,
+      "loss": 0.7332,
+      "step": 846
+    },
+    {
+      "epoch": 2.5703176861071597,
+      "grad_norm": 2.1909451484680176,
+      "learning_rate": 1.2017200418357077e-06,
+      "loss": 0.7402,
+      "step": 847
+    },
+    {
+      "epoch": 2.5733522996680893,
+      "grad_norm": 2.2838058471679688,
+      "learning_rate": 1.184960108733586e-06,
+      "loss": 0.7336,
+      "step": 848
+    },
+    {
+      "epoch": 2.5763869132290185,
+      "grad_norm": 2.0348660945892334,
+      "learning_rate": 1.1683105066539068e-06,
+      "loss": 0.7367,
+      "step": 849
+    },
+    {
+      "epoch": 2.5794215267899476,
+      "grad_norm": 2.088353157043457,
+      "learning_rate": 1.151771443986842e-06,
+      "loss": 0.749,
+      "step": 850
+    },
+    {
+      "epoch": 2.5794215267899476,
+      "eval_loss": NaN,
+      "eval_runtime": 205.2076,
+      "eval_samples_per_second": 20.262,
+      "eval_steps_per_second": 5.068,
+      "step": 850
+    },
+    {
+      "epoch": 2.5824561403508772,
+      "grad_norm": 2.1730659008026123,
+      "learning_rate": 1.1353431277390125e-06,
+      "loss": 0.7359,
+      "step": 851
+    },
+    {
+      "epoch": 2.5854907539118064,
+      "grad_norm": 2.10697078704834,
+      "learning_rate": 1.1190257635309276e-06,
+      "loss": 0.7223,
+      "step": 852
+    },
+    {
+      "epoch": 2.588525367472736,
+      "grad_norm": 2.1502668857574463,
+      "learning_rate": 1.1028195555943877e-06,
+      "loss": 0.7265,
+      "step": 853
+    },
+    {
+      "epoch": 2.591559981033665,
+      "grad_norm": 2.0735018253326416,
+      "learning_rate": 1.0867247067699315e-06,
+      "loss": 0.7345,
+      "step": 854
+    },
+    {
+      "epoch": 2.5945945945945947,
+      "grad_norm": 2.1325292587280273,
+      "learning_rate": 1.0707414185043163e-06,
+      "loss": 0.746,
+      "step": 855
+    },
+    {
+      "epoch": 2.597629208155524,
+      "grad_norm": 2.1418328285217285,
+      "learning_rate": 1.0548698908479671e-06,
+      "loss": 0.7401,
+      "step": 856
+    },
+    {
+      "epoch": 2.600663821716453,
+      "grad_norm": 2.061847448348999,
+      "learning_rate": 1.0391103224524957e-06,
+      "loss": 0.7319,
+      "step": 857
+    },
+    {
+      "epoch": 2.6036984352773827,
+      "grad_norm": 2.1931376457214355,
+      "learning_rate": 1.0234629105682104e-06,
+      "loss": 0.7429,
+      "step": 858
+    },
+    {
+      "epoch": 2.6067330488383122,
+      "grad_norm": 2.1396734714508057,
+      "learning_rate": 1.0079278510416313e-06,
+      "loss": 0.7369,
+      "step": 859
+    },
+    {
+      "epoch": 2.6097676623992414,
+      "grad_norm": 2.2338297367095947,
+      "learning_rate": 9.925053383130667e-07,
+      "loss": 0.745,
+      "step": 860
+    },
+    {
+      "epoch": 2.6128022759601706,
+      "grad_norm": 2.120897054672241,
+      "learning_rate": 9.771955654141496e-07,
+      "loss": 0.7444,
+      "step": 861
+    },
+    {
+      "epoch": 2.6158368895211,
+      "grad_norm": 2.204875946044922,
+      "learning_rate": 9.619987239654405e-07,
+      "loss": 0.7293,
+      "step": 862
+    },
+    {
+      "epoch": 2.6188715030820293,
+      "grad_norm": 2.228181838989258,
+      "learning_rate": 9.469150041740338e-07,
+      "loss": 0.7251,
+      "step": 863
+    },
+    {
+      "epoch": 2.6219061166429585,
+      "grad_norm": 2.170053720474243,
+      "learning_rate": 9.319445948311534e-07,
+      "loss": 0.7444,
+      "step": 864
+    },
+    {
+      "epoch": 2.624940730203888,
+      "grad_norm": 2.163527011871338,
+      "learning_rate": 9.170876833098119e-07,
+      "loss": 0.7568,
+      "step": 865
+    },
+    {
+      "epoch": 2.6279753437648177,
+      "grad_norm": 2.1692657470703125,
+      "learning_rate": 9.023444555624572e-07,
+      "loss": 0.736,
+      "step": 866
+    },
+    {
+      "epoch": 2.631009957325747,
+      "grad_norm": 2.0737569332122803,
+      "learning_rate": 8.87715096118642e-07,
+      "loss": 0.7368,
+      "step": 867
+    },
+    {
+      "epoch": 2.634044570886676,
+      "grad_norm": 2.05617094039917,
+      "learning_rate": 8.731997880827258e-07,
+      "loss": 0.7334,
+      "step": 868
+    },
+    {
+      "epoch": 2.6370791844476056,
+      "grad_norm": 1.9645661115646362,
+      "learning_rate": 8.587987131315656e-07,
+      "loss": 0.7422,
+      "step": 869
+    },
+    {
+      "epoch": 2.6401137980085347,
+      "grad_norm": 2.0881567001342773,
+      "learning_rate": 8.445120515122552e-07,
+      "loss": 0.7296,
+      "step": 870
+    },
+    {
+      "epoch": 2.6431484115694643,
+      "grad_norm": 2.2319769859313965,
+      "learning_rate": 8.303399820398672e-07,
+      "loss": 0.7502,
+      "step": 871
+    },
+    {
+      "epoch": 2.6461830251303935,
+      "grad_norm": 2.1117403507232666,
+      "learning_rate": 8.162826820952097e-07,
+      "loss": 0.7319,
+      "step": 872
+    },
+    {
+      "epoch": 2.649217638691323,
+      "grad_norm": 2.056861400604248,
+      "learning_rate": 8.023403276226127e-07,
+      "loss": 0.7429,
+      "step": 873
+    },
+    {
+      "epoch": 2.6522522522522523,
+      "grad_norm": 2.140435218811035,
+      "learning_rate": 7.885130931277219e-07,
+      "loss": 0.7418,
+      "step": 874
+    },
+    {
+      "epoch": 2.6552868658131814,
+      "grad_norm": 2.11114239692688,
+      "learning_rate": 7.74801151675314e-07,
+      "loss": 0.7521,
+      "step": 875
+    },
+    {
+      "epoch": 2.658321479374111,
+      "grad_norm": 2.0694892406463623,
+      "learning_rate": 7.612046748871327e-07,
+      "loss": 0.7385,
+      "step": 876
+    },
+    {
+      "epoch": 2.66135609293504,
+      "grad_norm": 2.155579090118408,
+      "learning_rate": 7.477238329397419e-07,
+      "loss": 0.7427,
+      "step": 877
+    },
+    {
+      "epoch": 2.6643907064959698,
+      "grad_norm": 2.223646402359009,
+      "learning_rate": 7.343587945623908e-07,
+      "loss": 0.731,
+      "step": 878
+    },
+    {
+      "epoch": 2.667425320056899,
+      "grad_norm": 2.047801971435547,
+      "learning_rate": 7.211097270349065e-07,
+      "loss": 0.7426,
+      "step": 879
+    },
+    {
+      "epoch": 2.6704599336178285,
+      "grad_norm": 2.119241714477539,
+      "learning_rate": 7.07976796185601e-07,
+      "loss": 0.7407,
+      "step": 880
+    },
+    {
+      "epoch": 2.6734945471787577,
+      "grad_norm": 2.2282469272613525,
+      "learning_rate": 6.949601663891891e-07,
+      "loss": 0.7541,
+      "step": 881
+    },
+    {
+      "epoch": 2.676529160739687,
+      "grad_norm": 2.175909996032715,
+      "learning_rate": 6.820600005647382e-07,
+      "loss": 0.7461,
+      "step": 882
+    },
+    {
+      "epoch": 2.6795637743006164,
+      "grad_norm": 2.047321081161499,
+      "learning_rate": 6.692764601736268e-07,
+      "loss": 0.7197,
+      "step": 883
+    },
+    {
+      "epoch": 2.682598387861546,
+      "grad_norm": 2.0702617168426514,
+      "learning_rate": 6.566097052175213e-07,
+      "loss": 0.7656,
+      "step": 884
+    },
+    {
+      "epoch": 2.685633001422475,
+      "grad_norm": 2.1683459281921387,
+      "learning_rate": 6.440598942363796e-07,
+      "loss": 0.7521,
+      "step": 885
+    },
+    {
+      "epoch": 2.6886676149834043,
+      "grad_norm": 2.067746162414551,
+      "learning_rate": 6.316271843064536e-07,
+      "loss": 0.7531,
+      "step": 886
+    },
+    {
+      "epoch": 2.691702228544334,
+      "grad_norm": 2.1310875415802,
+      "learning_rate": 6.193117310383412e-07,
+      "loss": 0.7538,
+      "step": 887
+    },
+    {
+      "epoch": 2.694736842105263,
+      "grad_norm": 2.1403868198394775,
+      "learning_rate": 6.071136885750272e-07,
+      "loss": 0.7373,
+      "step": 888
+    },
+    {
+      "epoch": 2.6977714556661923,
+      "grad_norm": 2.049807071685791,
+      "learning_rate": 5.950332095899547e-07,
+      "loss": 0.738,
+      "step": 889
+    },
+    {
+      "epoch": 2.700806069227122,
+      "grad_norm": 2.080238103866577,
+      "learning_rate": 5.830704452851166e-07,
+      "loss": 0.7212,
+      "step": 890
+    },
+    {
+      "epoch": 2.7038406827880515,
+      "grad_norm": 2.1021158695220947,
+      "learning_rate": 5.71225545389158e-07,
+      "loss": 0.7411,
+      "step": 891
+    },
+    {
+      "epoch": 2.7068752963489806,
+      "grad_norm": 2.0948758125305176,
+      "learning_rate": 5.594986581555173e-07,
+      "loss": 0.7369,
+      "step": 892
+    },
+    {
+      "epoch": 2.7099099099099098,
+      "grad_norm": 2.1047518253326416,
+      "learning_rate": 5.478899303605512e-07,
+      "loss": 0.7289,
+      "step": 893
+    },
+    {
+      "epoch": 2.7129445234708394,
+      "grad_norm": 2.1207687854766846,
+      "learning_rate": 5.363995073017047e-07,
+      "loss": 0.7385,
+      "step": 894
+    },
+    {
+      "epoch": 2.7159791370317685,
+      "grad_norm": 2.051896572113037,
+      "learning_rate": 5.250275327957033e-07,
+      "loss": 0.7244,
+      "step": 895
+    },
+    {
+      "epoch": 2.719013750592698,
+      "grad_norm": 2.220273017883301,
+      "learning_rate": 5.137741491767345e-07,
+      "loss": 0.7522,
+      "step": 896
+    },
+    {
+      "epoch": 2.7220483641536273,
+      "grad_norm": 2.1100893020629883,
+      "learning_rate": 5.026394972946813e-07,
+      "loss": 0.7276,
+      "step": 897
+    },
+    {
+      "epoch": 2.725082977714557,
+      "grad_norm": 2.300666332244873,
+      "learning_rate": 4.91623716513352e-07,
+      "loss": 0.7404,
+      "step": 898
+    },
+    {
+      "epoch": 2.728117591275486,
+      "grad_norm": 2.2343952655792236,
+      "learning_rate": 4.807269447087348e-07,
+      "loss": 0.7474,
+      "step": 899
+    },
+    {
+      "epoch": 2.731152204836415,
+      "grad_norm": 2.1582469940185547,
+      "learning_rate": 4.6994931826728094e-07,
+      "loss": 0.7275,
+      "step": 900
+    },
+    {
+      "epoch": 2.731152204836415,
+      "eval_loss": NaN,
+      "eval_runtime": 204.8855,
+      "eval_samples_per_second": 20.294,
+      "eval_steps_per_second": 5.076,
+      "step": 900
+    },
+    {
+      "epoch": 2.734186818397345,
+      "grad_norm": 2.057573080062866,
+      "learning_rate": 4.592909720841843e-07,
+      "loss": 0.7351,
+      "step": 901
+    },
+    {
+      "epoch": 2.737221431958274,
+      "grad_norm": 2.320993423461914,
+      "learning_rate": 4.487520395617029e-07,
+      "loss": 0.7086,
+      "step": 902
+    },
+    {
+      "epoch": 2.7402560455192035,
+      "grad_norm": 2.2511098384857178,
+      "learning_rate": 4.3833265260749157e-07,
+      "loss": 0.723,
+      "step": 903
+    },
+    {
+      "epoch": 2.7432906590801327,
+      "grad_norm": 2.182513952255249,
+      "learning_rate": 4.280329416329365e-07,
+      "loss": 0.7196,
+      "step": 904
+    },
+    {
+      "epoch": 2.7463252726410623,
+      "grad_norm": 2.1579484939575195,
+      "learning_rate": 4.178530355515409e-07,
+      "loss": 0.7334,
+      "step": 905
+    },
+    {
+      "epoch": 2.7493598862019915,
+      "grad_norm": 2.091196060180664,
+      "learning_rate": 4.077930617773007e-07,
+      "loss": 0.7387,
+      "step": 906
+    },
+    {
+      "epoch": 2.7523944997629206,
+      "grad_norm": 2.1611766815185547,
+      "learning_rate": 3.97853146223105e-07,
+      "loss": 0.7515,
+      "step": 907
+    },
+    {
+      "epoch": 2.75542911332385,
+      "grad_norm": 2.240387201309204,
+      "learning_rate": 3.880334132991792e-07,
+      "loss": 0.7377,
+      "step": 908
+    },
+    {
+      "epoch": 2.7584637268847794,
+      "grad_norm": 2.1546630859375,
+      "learning_rate": 3.783339859115065e-07,
+      "loss": 0.7388,
+      "step": 909
+    },
+    {
+      "epoch": 2.761498340445709,
+      "grad_norm": 2.0892395973205566,
+      "learning_rate": 3.687549854603023e-07,
+      "loss": 0.7488,
+      "step": 910
+    },
+    {
+      "epoch": 2.764532954006638,
+      "grad_norm": 2.182509422302246,
+      "learning_rate": 3.5929653183849444e-07,
+      "loss": 0.736,
+      "step": 911
+    },
+    {
+      "epoch": 2.7675675675675677,
+      "grad_norm": 2.198559522628784,
+      "learning_rate": 3.49958743430211e-07,
+      "loss": 0.736,
+      "step": 912
+    },
+    {
+      "epoch": 2.770602181128497,
+      "grad_norm": 2.188133955001831,
+      "learning_rate": 3.4074173710931804e-07,
+      "loss": 0.7345,
+      "step": 913
+    },
+    {
+      "epoch": 2.773636794689426,
+      "grad_norm": 2.133892774581909,
+      "learning_rate": 3.3164562823793654e-07,
+      "loss": 0.7275,
+      "step": 914
+    },
+    {
+      "epoch": 2.7766714082503556,
+      "grad_norm": 2.159824848175049,
+      "learning_rate": 3.226705306650113e-07,
+      "loss": 0.7198,
+      "step": 915
+    },
+    {
+      "epoch": 2.7797060218112852,
+      "grad_norm": 2.1087584495544434,
+      "learning_rate": 3.1381655672488634e-07,
+      "loss": 0.7381,
+      "step": 916
+    },
+    {
+      "epoch": 2.7827406353722144,
+      "grad_norm": 2.0807528495788574,
+      "learning_rate": 3.050838172358883e-07,
+      "loss": 0.7386,
+      "step": 917
+    },
+    {
+      "epoch": 2.7857752489331435,
+      "grad_norm": 2.1473801136016846,
+      "learning_rate": 2.9647242149895005e-07,
+      "loss": 0.7336,
+      "step": 918
+    },
+    {
+      "epoch": 2.788809862494073,
+      "grad_norm": 2.219571352005005,
+      "learning_rate": 2.879824772962381e-07,
+      "loss": 0.7438,
+      "step": 919
+    },
+    {
+      "epoch": 2.7918444760550023,
+      "grad_norm": 2.144059419631958,
+      "learning_rate": 2.796140908898026e-07,
+      "loss": 0.7338,
+      "step": 920
+    },
+    {
+      "epoch": 2.7948790896159315,
+      "grad_norm": 2.1552734375,
+      "learning_rate": 2.7136736702025436e-07,
+      "loss": 0.7345,
+      "step": 921
+    },
+    {
+      "epoch": 2.797913703176861,
+      "grad_norm": 2.18730092048645,
+      "learning_rate": 2.632424089054419e-07,
+      "loss": 0.7222,
+      "step": 922
+    },
+    {
+      "epoch": 2.8009483167377907,
+      "grad_norm": 2.233304500579834,
+      "learning_rate": 2.552393182391677e-07,
+      "loss": 0.7474,
+      "step": 923
+    },
+    {
+      "epoch": 2.80398293029872,
+      "grad_norm": 2.041757583618164,
+      "learning_rate": 2.473581951899184e-07,
+      "loss": 0.7329,
+      "step": 924
+    },
+    {
+      "epoch": 2.807017543859649,
+      "grad_norm": 2.1639747619628906,
+      "learning_rate": 2.395991383995999e-07,
+      "loss": 0.748,
+      "step": 925
+    },
+    {
+      "epoch": 2.8100521574205786,
+      "grad_norm": 2.0612905025482178,
+      "learning_rate": 2.3196224498231447e-07,
+      "loss": 0.7311,
+      "step": 926
+    },
+    {
+      "epoch": 2.8130867709815077,
+      "grad_norm": 2.260965347290039,
+      "learning_rate": 2.2444761052313857e-07,
+      "loss": 0.7353,
+      "step": 927
+    },
+    {
+      "epoch": 2.8161213845424373,
+      "grad_norm": 2.1008665561676025,
+      "learning_rate": 2.1705532907692617e-07,
+      "loss": 0.7677,
+      "step": 928
+    },
+    {
+      "epoch": 2.8191559981033665,
+      "grad_norm": 2.20302414894104,
+      "learning_rate": 2.0978549316713615e-07,
+      "loss": 0.7447,
+      "step": 929
+    },
+    {
+      "epoch": 2.822190611664296,
+      "grad_norm": 2.100362539291382,
+      "learning_rate": 2.0263819378466888e-07,
+      "loss": 0.7393,
+      "step": 930
+    },
+    {
+      "epoch": 2.8252252252252252,
+      "grad_norm": 2.2228076457977295,
+      "learning_rate": 1.9561352038673264e-07,
+      "loss": 0.7501,
+      "step": 931
+    },
+    {
+      "epoch": 2.8282598387861544,
+      "grad_norm": 2.2403063774108887,
+      "learning_rate": 1.8871156089572018e-07,
+      "loss": 0.7508,
+      "step": 932
+    },
+    {
+      "epoch": 2.831294452347084,
+      "grad_norm": 2.1292290687561035,
+      "learning_rate": 1.8193240169810943e-07,
+      "loss": 0.7362,
+      "step": 933
+    },
+    {
+      "epoch": 2.834329065908013,
+      "grad_norm": 2.029907703399658,
+      "learning_rate": 1.752761276433801e-07,
+      "loss": 0.7504,
+      "step": 934
+    },
+    {
+      "epoch": 2.8373636794689427,
+      "grad_norm": 2.175046682357788,
+      "learning_rate": 1.6874282204295765e-07,
+      "loss": 0.7437,
+      "step": 935
+    },
+    {
+      "epoch": 2.840398293029872,
+      "grad_norm": 2.150054693222046,
+      "learning_rate": 1.623325666691644e-07,
+      "loss": 0.7388,
+      "step": 936
+    },
+    {
+      "epoch": 2.8434329065908015,
+      "grad_norm": 2.2034387588500977,
+      "learning_rate": 1.5604544175419901e-07,
+      "loss": 0.7451,
+      "step": 937
+    },
+    {
+      "epoch": 2.8464675201517307,
+      "grad_norm": 2.1042325496673584,
+      "learning_rate": 1.4988152598913063e-07,
+      "loss": 0.7535,
+      "step": 938
+    },
+    {
+      "epoch": 2.84950213371266,
+      "grad_norm": 2.1048696041107178,
+      "learning_rate": 1.4384089652291544e-07,
+      "loss": 0.745,
+      "step": 939
+    },
+    {
+      "epoch": 2.8525367472735894,
+      "grad_norm": 2.3105156421661377,
+      "learning_rate": 1.3792362896143164e-07,
+      "loss": 0.7233,
+      "step": 940
+    },
+    {
+      "epoch": 2.8555713608345186,
+      "grad_norm": 2.1417715549468994,
+      "learning_rate": 1.3212979736653142e-07,
+      "loss": 0.7467,
+      "step": 941
+    },
+    {
+      "epoch": 2.858605974395448,
+      "grad_norm": 2.1809537410736084,
+      "learning_rate": 1.2645947425511397e-07,
+      "loss": 0.7432,
+      "step": 942
+    },
+    {
+      "epoch": 2.8616405879563773,
+      "grad_norm": 2.2112855911254883,
+      "learning_rate": 1.209127305982205e-07,
+      "loss": 0.7487,
+      "step": 943
+    },
+    {
+      "epoch": 2.864675201517307,
+      "grad_norm": 2.059002161026001,
+      "learning_rate": 1.1548963582013961e-07,
+      "loss": 0.734,
+      "step": 944
+    },
+    {
+      "epoch": 2.867709815078236,
+      "grad_norm": 2.04377818107605,
+      "learning_rate": 1.1019025779754666e-07,
+      "loss": 0.7213,
+      "step": 945
+    },
+    {
+      "epoch": 2.8707444286391652,
+      "grad_norm": 2.0912530422210693,
+      "learning_rate": 1.0501466285865124e-07,
+      "loss": 0.7548,
+      "step": 946
+    },
+    {
+      "epoch": 2.873779042200095,
+      "grad_norm": 2.2096190452575684,
+      "learning_rate": 9.996291578236228e-08,
+      "loss": 0.7246,
+      "step": 947
+    },
+    {
+      "epoch": 2.8768136557610244,
+      "grad_norm": 2.284653663635254,
+      "learning_rate": 9.503507979748305e-08,
+      "loss": 0.7339,
+      "step": 948
+    },
+    {
+      "epoch": 2.8798482693219536,
+      "grad_norm": 2.097752332687378,
+      "learning_rate": 9.023121658191636e-08,
+      "loss": 0.7303,
+      "step": 949
+    },
+    {
+      "epoch": 2.8828828828828827,
+      "grad_norm": 2.171391010284424,
+      "learning_rate": 8.555138626189619e-08,
+      "loss": 0.7503,
+      "step": 950
+    },
+    {
+      "epoch": 2.8828828828828827,
+      "eval_loss": NaN,
+      "eval_runtime": 204.3654,
+      "eval_samples_per_second": 20.346,
+      "eval_steps_per_second": 5.089,
+      "step": 950
+    },
+    {
+      "epoch": 2.8859174964438123,
+      "grad_norm": 2.1366286277770996,
+      "learning_rate": 8.099564741123167e-08,
+      "loss": 0.7241,
+      "step": 951
+    },
+    {
+      "epoch": 2.8889521100047415,
+      "grad_norm": 2.0811874866485596,
+      "learning_rate": 7.656405705057434e-08,
+      "loss": 0.7504,
+      "step": 952
+    },
+    {
+      "epoch": 2.8919867235656707,
+      "grad_norm": 2.080226421356201,
+      "learning_rate": 7.225667064670761e-08,
+      "loss": 0.7355,
+      "step": 953
+    },
+    {
+      "epoch": 2.8950213371266003,
+      "grad_norm": 2.200331211090088,
+      "learning_rate": 6.807354211184613e-08,
+      "loss": 0.7427,
+      "step": 954
+    },
+    {
+      "epoch": 2.89805595068753,
+      "grad_norm": 2.038591146469116,
+      "learning_rate": 6.401472380297091e-08,
+      "loss": 0.7379,
+      "step": 955
+    },
+    {
+      "epoch": 2.901090564248459,
+      "grad_norm": 2.1141886711120605,
+      "learning_rate": 6.008026652116306e-08,
+      "loss": 0.7376,
+      "step": 956
+    },
+    {
+      "epoch": 2.904125177809388,
+      "grad_norm": 2.0803427696228027,
+      "learning_rate": 5.6270219510975445e-08,
+      "loss": 0.7424,
+      "step": 957
+    },
+    {
+      "epoch": 2.9071597913703178,
+      "grad_norm": 2.0900638103485107,
+      "learning_rate": 5.258463045981432e-08,
+      "loss": 0.7243,
+      "step": 958
+    },
+    {
+      "epoch": 2.910194404931247,
+      "grad_norm": 2.133312463760376,
+      "learning_rate": 4.902354549733979e-08,
+      "loss": 0.7257,
+      "step": 959
+    },
+    {
+      "epoch": 2.9132290184921765,
+      "grad_norm": 2.2082343101501465,
+      "learning_rate": 4.5587009194894005e-08,
+      "loss": 0.7346,
+      "step": 960
+    },
+    {
+      "epoch": 2.9162636320531057,
+      "grad_norm": 2.0872416496276855,
+      "learning_rate": 4.227506456493835e-08,
+      "loss": 0.7437,
+      "step": 961
+    },
+    {
+      "epoch": 2.9192982456140353,
+      "grad_norm": 2.136866807937622,
+      "learning_rate": 3.908775306051604e-08,
+      "loss": 0.7362,
+      "step": 962
+    },
+    {
+      "epoch": 2.9223328591749644,
+      "grad_norm": 2.125523805618286,
+      "learning_rate": 3.602511457473479e-08,
+      "loss": 0.7298,
+      "step": 963
+    },
+    {
+      "epoch": 2.9253674727358936,
+      "grad_norm": 2.055690050125122,
+      "learning_rate": 3.3087187440268284e-08,
+      "loss": 0.7285,
+      "step": 964
+    },
+    {
+      "epoch": 2.928402086296823,
+      "grad_norm": 2.26811146736145,
+      "learning_rate": 3.027400842887218e-08,
+      "loss": 0.732,
+      "step": 965
+    },
+    {
+      "epoch": 2.9314366998577523,
+      "grad_norm": 2.0984137058258057,
+      "learning_rate": 2.758561275092886e-08,
+      "loss": 0.7238,
+      "step": 966
+    },
+    {
+      "epoch": 2.934471313418682,
+      "grad_norm": 2.117258310317993,
+      "learning_rate": 2.5022034055003363e-08,
+      "loss": 0.7604,
+      "step": 967
+    },
+    {
+      "epoch": 2.937505926979611,
+      "grad_norm": 2.135441541671753,
+      "learning_rate": 2.2583304427421516e-08,
+      "loss": 0.7358,
+      "step": 968
+    },
+    {
+      "epoch": 2.9405405405405407,
+      "grad_norm": 2.1416280269622803,
+      "learning_rate": 2.0269454391874665e-08,
+      "loss": 0.7436,
+      "step": 969
+    },
+    {
+      "epoch": 2.94357515410147,
+      "grad_norm": 2.1733927726745605,
+      "learning_rate": 1.80805129090289e-08,
+      "loss": 0.7308,
+      "step": 970
+    },
+    {
+      "epoch": 2.946609767662399,
+      "grad_norm": 2.1229727268218994,
+      "learning_rate": 1.6016507376169776e-08,
+      "loss": 0.7577,
+      "step": 971
+    },
+    {
+      "epoch": 2.9496443812233286,
+      "grad_norm": 2.283834218978882,
+      "learning_rate": 1.4077463626852584e-08,
+      "loss": 0.748,
+      "step": 972
+    },
+    {
+      "epoch": 2.952678994784258,
+      "grad_norm": 2.205937623977661,
+      "learning_rate": 1.2263405930585947e-08,
+      "loss": 0.7335,
+      "step": 973
+    },
+    {
+      "epoch": 2.9557136083451874,
+      "grad_norm": 2.111551284790039,
+      "learning_rate": 1.0574356992525403e-08,
+      "loss": 0.7295,
+      "step": 974
+    },
+    {
+      "epoch": 2.9587482219061165,
+      "grad_norm": 2.058469295501709,
+      "learning_rate": 9.010337953185843e-09,
+      "loss": 0.7547,
+      "step": 975
+    },
+    {
+      "epoch": 2.961782835467046,
+      "grad_norm": 2.1459245681762695,
+      "learning_rate": 7.571368388181732e-09,
+      "loss": 0.7368,
+      "step": 976
+    },
+    {
+      "epoch": 2.9648174490279753,
+      "grad_norm": 2.1419780254364014,
+      "learning_rate": 6.257466307980631e-09,
+      "loss": 0.7364,
+      "step": 977
+    },
+    {
+      "epoch": 2.9678520625889044,
+      "grad_norm": 2.1368930339813232,
+      "learning_rate": 5.068648157675604e-09,
+      "loss": 0.7243,
+      "step": 978
+    },
+    {
+      "epoch": 2.970886676149834,
+      "grad_norm": 2.072345733642578,
+      "learning_rate": 4.00492881678427e-09,
+      "loss": 0.7229,
+      "step": 979
+    },
+    {
+      "epoch": 2.9739212897107636,
+      "grad_norm": 2.173828363418579,
+      "learning_rate": 3.0663215990534013e-09,
+      "loss": 0.711,
+      "step": 980
+    },
+    {
+      "epoch": 2.976955903271693,
+      "grad_norm": 2.117293357849121,
+      "learning_rate": 2.2528382523057115e-09,
+      "loss": 0.7648,
+      "step": 981
+    },
+    {
+      "epoch": 2.979990516832622,
+      "grad_norm": 2.126967191696167,
+      "learning_rate": 1.564488958279986e-09,
+      "loss": 0.7359,
+      "step": 982
+    },
+    {
+      "epoch": 2.9830251303935515,
+      "grad_norm": 2.086338520050049,
+      "learning_rate": 1.0012823325111776e-09,
+      "loss": 0.7575,
+      "step": 983
+    },
+    {
+      "epoch": 2.9860597439544807,
+      "grad_norm": 2.0174200534820557,
+      "learning_rate": 5.632254242204926e-10,
+      "loss": 0.7324,
+      "step": 984
+    },
+    {
+      "epoch": 2.9890943575154103,
+      "grad_norm": 2.0716025829315186,
+      "learning_rate": 2.503237162254646e-10,
+      "loss": 0.7482,
+      "step": 985
+    },
+    {
+      "epoch": 2.9921289710763395,
+      "grad_norm": 2.0842347145080566,
+      "learning_rate": 6.258112487667056e-11,
+      "loss": 0.7334,
+      "step": 986
+    },
+    {
+      "epoch": 2.995163584637269,
+      "grad_norm": 2.1061208248138428,
+      "learning_rate": 0.0,
+      "loss": 0.7299,
+      "step": 987
+    },
+    {
+      "epoch": 2.995163584637269,
+      "step": 987,
+      "total_flos": 4.415483185831556e+19,
+      "train_loss": 0.7698779804849093,
+      "train_runtime": 131039.1076,
+      "train_samples_per_second": 3.863,
+      "train_steps_per_second": 0.008
+    }
+  ],
+  "logging_steps": 1.0,
+  "max_steps": 987,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 3,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 4.415483185831556e+19,
+  "train_batch_size": 2,
+  "trial_name": null,
+  "trial_params": null
+}