diff --git "a/trainer_state.json" "b/trainer_state.json"
--- "a/trainer_state.json"
+++ "b/trainer_state.json"
@@ -19,7 +19,7 @@
       "epoch": 0.0,
       "grad_norm": 0.546875,
       "learning_rate": 1.6920473773265652e-06,
-      "loss": 2.5283,
+      "loss": 2.5282,
       "step": 5
     },
     {
@@ -31,21 +31,21 @@
     },
     {
       "epoch": 0.01,
-      "grad_norm": 0.859375,
+      "grad_norm": 0.8671875,
       "learning_rate": 5.076142131979695e-06,
       "loss": 2.5184,
       "step": 15
     },
     {
       "epoch": 0.01,
-      "grad_norm": 0.89453125,
+      "grad_norm": 0.88671875,
       "learning_rate": 6.768189509306261e-06,
-      "loss": 2.5027,
+      "loss": 2.5026,
       "step": 20
     },
     {
       "epoch": 0.01,
-      "grad_norm": 0.76953125,
+      "grad_norm": 0.7734375,
       "learning_rate": 8.460236886632826e-06,
       "loss": 2.5038,
       "step": 25
@@ -61,12 +61,12 @@
       "epoch": 0.01,
       "grad_norm": 0.5546875,
       "learning_rate": 1.1844331641285957e-05,
-      "loss": 2.4494,
+      "loss": 2.4495,
       "step": 35
     },
     {
       "epoch": 0.01,
-      "grad_norm": 0.474609375,
+      "grad_norm": 0.47265625,
       "learning_rate": 1.3536379018612521e-05,
       "loss": 2.4364,
       "step": 40
@@ -75,7 +75,7 @@
       "epoch": 0.02,
       "grad_norm": 0.34765625,
       "learning_rate": 1.5228426395939088e-05,
-      "loss": 2.4704,
+      "loss": 2.4702,
       "step": 45
     },
     {
@@ -89,26 +89,26 @@
       "epoch": 0.02,
       "grad_norm": 0.2265625,
       "learning_rate": 1.8612521150592218e-05,
-      "loss": 2.4019,
+      "loss": 2.402,
       "step": 55
     },
     {
       "epoch": 0.02,
       "grad_norm": 0.19140625,
       "learning_rate": 2.030456852791878e-05,
-      "loss": 2.3979,
+      "loss": 2.3978,
       "step": 60
     },
     {
       "epoch": 0.02,
       "grad_norm": 0.1708984375,
       "learning_rate": 2.199661590524535e-05,
-      "loss": 2.3894,
+      "loss": 2.3895,
       "step": 65
     },
     {
       "epoch": 0.02,
-      "grad_norm": 0.1748046875,
+      "grad_norm": 0.1767578125,
       "learning_rate": 2.3688663282571914e-05,
       "loss": 2.4104,
       "step": 70
@@ -129,30 +129,30 @@
     },
     {
       "epoch": 0.03,
-      "grad_norm": 0.1640625,
+      "grad_norm": 0.1650390625,
       "learning_rate": 2.876480541455161e-05,
-      "loss": 2.3435,
+      "loss": 2.3434,
       "step": 85
     },
     {
       "epoch": 0.03,
-      "grad_norm": 0.1806640625,
+      "grad_norm": 0.1796875,
       "learning_rate": 3.0456852791878175e-05,
-      "loss": 2.3394,
+      "loss": 2.3395,
       "step": 90
     },
     {
       "epoch": 0.03,
-      "grad_norm": 0.1611328125,
+      "grad_norm": 0.16015625,
       "learning_rate": 3.214890016920474e-05,
-      "loss": 2.3178,
+      "loss": 2.3176,
       "step": 95
     },
     {
       "epoch": 0.03,
       "grad_norm": 0.1513671875,
       "learning_rate": 3.3840947546531304e-05,
-      "loss": 2.3124,
+      "loss": 2.3123,
       "step": 100
     },
     {
@@ -173,14 +173,14 @@
       "epoch": 0.04,
       "grad_norm": 0.1416015625,
       "learning_rate": 3.8917089678510996e-05,
-      "loss": 2.293,
+      "loss": 2.2931,
       "step": 115
     },
     {
       "epoch": 0.04,
       "grad_norm": 0.1435546875,
       "learning_rate": 4.060913705583756e-05,
-      "loss": 2.3071,
+      "loss": 2.307,
       "step": 120
     },
     {
@@ -194,14 +194,14 @@
       "epoch": 0.04,
       "grad_norm": 0.134765625,
       "learning_rate": 4.39932318104907e-05,
-      "loss": 2.2847,
+      "loss": 2.2846,
       "step": 130
     },
     {
       "epoch": 0.05,
       "grad_norm": 0.134765625,
       "learning_rate": 4.568527918781726e-05,
-      "loss": 2.2561,
+      "loss": 2.256,
       "step": 135
     },
     {
@@ -215,7 +215,7 @@
       "epoch": 0.05,
       "grad_norm": 0.1298828125,
       "learning_rate": 4.906937394247039e-05,
-      "loss": 2.2716,
+      "loss": 2.2717,
       "step": 145
     },
     {
@@ -234,9 +234,9 @@
     },
     {
       "epoch": 0.05,
-      "grad_norm": 0.1298828125,
+      "grad_norm": 0.130859375,
       "learning_rate": 5.4145516074450085e-05,
-      "loss": 2.2446,
+      "loss": 2.2448,
       "step": 160
     },
     {
@@ -255,9 +255,9 @@
     },
     {
       "epoch": 0.06,
-      "grad_norm": 0.13671875,
+      "grad_norm": 0.1376953125,
       "learning_rate": 5.9221658206429784e-05,
-      "loss": 2.2511,
+      "loss": 2.2512,
       "step": 175
     },
     {
@@ -278,7 +278,7 @@
       "epoch": 0.06,
       "grad_norm": 0.140625,
       "learning_rate": 6.429780033840948e-05,
-      "loss": 2.2085,
+      "loss": 2.2084,
       "step": 190
     },
     {
@@ -292,28 +292,28 @@
       "epoch": 0.07,
       "grad_norm": 0.1416015625,
       "learning_rate": 6.768189509306261e-05,
-      "loss": 2.2032,
+      "loss": 2.2031,
       "step": 200
     },
     {
       "epoch": 0.07,
-      "grad_norm": 0.14453125,
+      "grad_norm": 0.1435546875,
       "learning_rate": 6.937394247038918e-05,
-      "loss": 2.212,
+      "loss": 2.2121,
       "step": 205
     },
     {
       "epoch": 0.07,
       "grad_norm": 0.1376953125,
       "learning_rate": 7.106598984771574e-05,
-      "loss": 2.2063,
+      "loss": 2.2062,
       "step": 210
     },
     {
       "epoch": 0.07,
       "grad_norm": 0.138671875,
       "learning_rate": 7.275803722504231e-05,
-      "loss": 2.2317,
+      "loss": 2.2315,
       "step": 215
     },
     {
@@ -327,21 +327,21 @@
       "epoch": 0.08,
       "grad_norm": 0.1435546875,
       "learning_rate": 7.614213197969543e-05,
-      "loss": 2.2178,
+      "loss": 2.2177,
       "step": 225
     },
     {
       "epoch": 0.08,
       "grad_norm": 0.1455078125,
       "learning_rate": 7.783417935702199e-05,
-      "loss": 2.2295,
+      "loss": 2.2296,
       "step": 230
     },
     {
       "epoch": 0.08,
-      "grad_norm": 0.150390625,
+      "grad_norm": 0.15234375,
       "learning_rate": 7.952622673434857e-05,
-      "loss": 2.2275,
+      "loss": 2.2274,
       "step": 235
     },
     {
@@ -383,19 +383,19 @@
       "epoch": 0.09,
       "grad_norm": 0.1513671875,
       "learning_rate": 8.967851099830795e-05,
-      "loss": 2.2361,
+      "loss": 2.2362,
       "step": 265
     },
     {
       "epoch": 0.09,
       "grad_norm": 0.1494140625,
       "learning_rate": 9.137055837563452e-05,
-      "loss": 2.2203,
+      "loss": 2.2204,
       "step": 270
     },
     {
       "epoch": 0.09,
-      "grad_norm": 0.15234375,
+      "grad_norm": 0.1513671875,
       "learning_rate": 9.306260575296108e-05,
       "loss": 2.1898,
       "step": 275
@@ -404,21 +404,21 @@
       "epoch": 0.09,
       "grad_norm": 0.1474609375,
       "learning_rate": 9.475465313028765e-05,
-      "loss": 2.1787,
+      "loss": 2.1788,
       "step": 280
     },
     {
       "epoch": 0.1,
-      "grad_norm": 0.1552734375,
+      "grad_norm": 0.15625,
       "learning_rate": 9.644670050761421e-05,
-      "loss": 2.1783,
+      "loss": 2.1784,
       "step": 285
     },
     {
       "epoch": 0.1,
       "grad_norm": 0.150390625,
       "learning_rate": 9.813874788494079e-05,
-      "loss": 2.1904,
+      "loss": 2.1903,
       "step": 290
     },
     {
@@ -432,35 +432,35 @@
       "epoch": 0.1,
       "grad_norm": 0.1513671875,
       "learning_rate": 0.0001015228426395939,
-      "loss": 2.1938,
+      "loss": 2.1936,
       "step": 300
     },
     {
       "epoch": 0.1,
-      "grad_norm": 0.1513671875,
+      "grad_norm": 0.150390625,
       "learning_rate": 0.00010321489001692048,
-      "loss": 2.1987,
+      "loss": 2.1986,
       "step": 305
     },
     {
       "epoch": 0.1,
       "grad_norm": 0.1591796875,
       "learning_rate": 0.00010490693739424704,
-      "loss": 2.1945,
+      "loss": 2.1946,
       "step": 310
     },
     {
       "epoch": 0.11,
-      "grad_norm": 0.1591796875,
+      "grad_norm": 0.16015625,
       "learning_rate": 0.00010659898477157362,
-      "loss": 2.2216,
+      "loss": 2.2217,
       "step": 315
     },
     {
       "epoch": 0.11,
-      "grad_norm": 0.158203125,
+      "grad_norm": 0.1572265625,
       "learning_rate": 0.00010829103214890017,
-      "loss": 2.208,
+      "loss": 2.2082,
       "step": 320
     },
     {
@@ -474,28 +474,28 @@
       "epoch": 0.11,
       "grad_norm": 0.1513671875,
       "learning_rate": 0.0001116751269035533,
-      "loss": 2.1854,
+      "loss": 2.1853,
       "step": 330
     },
     {
       "epoch": 0.11,
       "grad_norm": 0.158203125,
       "learning_rate": 0.00011336717428087986,
-      "loss": 2.1777,
+      "loss": 2.1776,
       "step": 335
     },
     {
       "epoch": 0.12,
       "grad_norm": 0.158203125,
       "learning_rate": 0.00011505922165820644,
-      "loss": 2.1846,
+      "loss": 2.1847,
       "step": 340
     },
     {
       "epoch": 0.12,
       "grad_norm": 0.162109375,
       "learning_rate": 0.000116751269035533,
-      "loss": 2.2163,
+      "loss": 2.2162,
       "step": 345
     },
     {
@@ -516,7 +516,7 @@
       "epoch": 0.12,
       "grad_norm": 0.1552734375,
       "learning_rate": 0.0001218274111675127,
-      "loss": 2.1977,
+      "loss": 2.1975,
       "step": 360
     },
     {
@@ -528,9 +528,9 @@
     },
     {
       "epoch": 0.13,
-      "grad_norm": 0.1640625,
+      "grad_norm": 0.1650390625,
       "learning_rate": 0.0001252115059221658,
-      "loss": 2.1746,
+      "loss": 2.1745,
       "step": 370
     },
     {
@@ -544,12 +544,12 @@
       "epoch": 0.13,
       "grad_norm": 0.158203125,
       "learning_rate": 0.00012859560067681895,
-      "loss": 2.1899,
+      "loss": 2.19,
       "step": 380
     },
     {
       "epoch": 0.13,
-      "grad_norm": 0.1591796875,
+      "grad_norm": 0.158203125,
       "learning_rate": 0.00013028764805414554,
       "loss": 2.2186,
       "step": 385
@@ -558,21 +558,21 @@
       "epoch": 0.13,
       "grad_norm": 0.162109375,
       "learning_rate": 0.00013197969543147207,
-      "loss": 2.1991,
+      "loss": 2.1992,
       "step": 390
     },
     {
       "epoch": 0.13,
       "grad_norm": 0.16015625,
       "learning_rate": 0.00013367174280879866,
-      "loss": 2.1906,
+      "loss": 2.1905,
       "step": 395
     },
     {
       "epoch": 0.14,
-      "grad_norm": 0.1572265625,
+      "grad_norm": 0.158203125,
       "learning_rate": 0.00013536379018612522,
-      "loss": 2.1837,
+      "loss": 2.1836,
       "step": 400
     },
     {
@@ -593,14 +593,14 @@
       "epoch": 0.14,
       "grad_norm": 0.150390625,
       "learning_rate": 0.0001404399323181049,
-      "loss": 2.1866,
+      "loss": 2.1865,
       "step": 415
     },
     {
       "epoch": 0.14,
       "grad_norm": 0.1552734375,
       "learning_rate": 0.00014213197969543148,
-      "loss": 2.1919,
+      "loss": 2.1918,
       "step": 420
     },
     {
@@ -614,14 +614,14 @@
       "epoch": 0.15,
       "grad_norm": 0.15234375,
       "learning_rate": 0.00014551607445008463,
-      "loss": 2.182,
+      "loss": 2.1819,
       "step": 430
     },
     {
       "epoch": 0.15,
       "grad_norm": 0.1611328125,
       "learning_rate": 0.00014720812182741116,
-      "loss": 2.1641,
+      "loss": 2.1639,
       "step": 435
     },
     {
@@ -640,9 +640,9 @@
     },
     {
       "epoch": 0.15,
-      "grad_norm": 0.166015625,
+      "grad_norm": 0.1669921875,
       "learning_rate": 0.00015228426395939087,
-      "loss": 2.1962,
+      "loss": 2.1961,
       "step": 450
     },
     {
@@ -656,14 +656,14 @@
       "epoch": 0.16,
       "grad_norm": 0.1474609375,
       "learning_rate": 0.00015566835871404398,
-      "loss": 2.1702,
+      "loss": 2.1705,
       "step": 460
     },
     {
       "epoch": 0.16,
       "grad_norm": 0.15234375,
       "learning_rate": 0.00015736040609137057,
-      "loss": 2.2,
+      "loss": 2.1999,
       "step": 465
     },
     {
@@ -677,14 +677,14 @@
       "epoch": 0.16,
       "grad_norm": 0.146484375,
       "learning_rate": 0.00016074450084602372,
-      "loss": 2.1905,
+      "loss": 2.1903,
       "step": 475
     },
     {
       "epoch": 0.16,
       "grad_norm": 0.1494140625,
       "learning_rate": 0.00016243654822335025,
-      "loss": 2.1742,
+      "loss": 2.174,
       "step": 480
     },
     {
@@ -705,28 +705,28 @@
       "epoch": 0.17,
       "grad_norm": 0.1552734375,
       "learning_rate": 0.00016751269035532995,
-      "loss": 2.1828,
+      "loss": 2.1827,
       "step": 495
     },
     {
       "epoch": 0.17,
       "grad_norm": 0.15234375,
       "learning_rate": 0.00016920473773265651,
-      "loss": 2.1915,
+      "loss": 2.1913,
       "step": 500
     },
     {
       "epoch": 0.17,
       "grad_norm": 0.16015625,
       "learning_rate": 0.00017089678510998307,
-      "loss": 2.1874,
+      "loss": 2.1873,
       "step": 505
     },
     {
       "epoch": 0.17,
       "grad_norm": 0.154296875,
       "learning_rate": 0.00017258883248730966,
-      "loss": 2.1935,
+      "loss": 2.1936,
       "step": 510
     },
     {
@@ -740,7 +740,7 @@
       "epoch": 0.18,
       "grad_norm": 0.1474609375,
       "learning_rate": 0.0001759729272419628,
-      "loss": 2.2071,
+      "loss": 2.2072,
       "step": 520
     },
     {
@@ -754,35 +754,35 @@
       "epoch": 0.18,
       "grad_norm": 0.146484375,
       "learning_rate": 0.0001793570219966159,
-      "loss": 2.193,
+      "loss": 2.1929,
       "step": 530
     },
     {
       "epoch": 0.18,
       "grad_norm": 0.154296875,
       "learning_rate": 0.00018104906937394248,
-      "loss": 2.1653,
+      "loss": 2.1655,
       "step": 535
     },
     {
       "epoch": 0.18,
       "grad_norm": 0.1513671875,
       "learning_rate": 0.00018274111675126904,
-      "loss": 2.1909,
+      "loss": 2.191,
       "step": 540
     },
     {
       "epoch": 0.18,
       "grad_norm": 0.154296875,
       "learning_rate": 0.0001844331641285956,
-      "loss": 2.2016,
+      "loss": 2.2017,
       "step": 545
     },
     {
       "epoch": 0.19,
       "grad_norm": 0.1533203125,
       "learning_rate": 0.00018612521150592216,
-      "loss": 2.1922,
+      "loss": 2.1921,
       "step": 550
     },
     {
@@ -803,7 +803,7 @@
       "epoch": 0.19,
       "grad_norm": 0.1494140625,
       "learning_rate": 0.00019120135363790187,
-      "loss": 2.1756,
+      "loss": 2.1757,
       "step": 565
     },
     {
@@ -822,9 +822,9 @@
     },
     {
       "epoch": 0.2,
-      "grad_norm": 0.1474609375,
+      "grad_norm": 0.1484375,
       "learning_rate": 0.00019627749576988157,
-      "loss": 2.1808,
+      "loss": 2.1807,
       "step": 580
     },
     {
@@ -838,21 +838,21 @@
       "epoch": 0.2,
       "grad_norm": 0.14453125,
       "learning_rate": 0.0001996615905245347,
-      "loss": 2.1607,
+      "loss": 2.1605,
       "step": 590
     },
     {
       "epoch": 0.2,
       "grad_norm": 0.15234375,
       "learning_rate": 0.00019999972028877317,
-      "loss": 2.1788,
+      "loss": 2.1787,
       "step": 595
     },
     {
       "epoch": 0.2,
       "grad_norm": 0.1494140625,
       "learning_rate": 0.00019999858396459598,
-      "loss": 2.1947,
+      "loss": 2.1948,
       "step": 600
     },
     {
@@ -866,7 +866,7 @@
       "epoch": 0.21,
       "grad_norm": 0.142578125,
       "learning_rate": 0.00019999368907888313,
-      "loss": 2.1582,
+      "loss": 2.1583,
       "step": 610
     },
     {
@@ -878,7 +878,7 @@
     },
     {
       "epoch": 0.21,
-      "grad_norm": 0.1474609375,
+      "grad_norm": 0.1484375,
       "learning_rate": 0.0001999852980320461,
       "loss": 2.1962,
       "step": 620
@@ -887,7 +887,7 @@
       "epoch": 0.21,
       "grad_norm": 0.142578125,
       "learning_rate": 0.0001999797915351135,
-      "loss": 2.1669,
+      "loss": 2.167,
       "step": 625
     },
     {
@@ -901,35 +901,35 @@
       "epoch": 0.22,
       "grad_norm": 0.146484375,
       "learning_rate": 0.00019996615683488039,
-      "loss": 2.1759,
+      "loss": 2.1757,
       "step": 635
     },
     {
       "epoch": 0.22,
       "grad_norm": 0.1484375,
       "learning_rate": 0.00019995802875076042,
-      "loss": 2.1692,
+      "loss": 2.169,
       "step": 640
     },
     {
       "epoch": 0.22,
       "grad_norm": 0.14453125,
       "learning_rate": 0.0001999490269361554,
-      "loss": 2.171,
+      "loss": 2.1709,
       "step": 645
     },
     {
       "epoch": 0.22,
       "grad_norm": 0.1455078125,
       "learning_rate": 0.00019993915146974996,
-      "loss": 2.1641,
+      "loss": 2.1642,
       "step": 650
     },
     {
       "epoch": 0.22,
       "grad_norm": 0.142578125,
       "learning_rate": 0.00019992840243786525,
-      "loss": 2.1699,
+      "loss": 2.17,
       "step": 655
     },
     {
@@ -950,7 +950,7 @@
       "epoch": 0.23,
       "grad_norm": 0.142578125,
       "learning_rate": 0.00019989091492707975,
-      "loss": 2.1987,
+      "loss": 2.1986,
       "step": 670
     },
     {
@@ -964,21 +964,21 @@
       "epoch": 0.23,
       "grad_norm": 0.142578125,
       "learning_rate": 0.00019986155735195372,
-      "loss": 2.1756,
+      "loss": 2.1754,
       "step": 680
     },
     {
       "epoch": 0.23,
-      "grad_norm": 0.1474609375,
+      "grad_norm": 0.1484375,
       "learning_rate": 0.0001998455691674828,
       "loss": 2.1832,
       "step": 685
     },
     {
       "epoch": 0.23,
-      "grad_norm": 0.1396484375,
+      "grad_norm": 0.138671875,
       "learning_rate": 0.00019982870823553308,
-      "loss": 2.1832,
+      "loss": 2.1831,
       "step": 690
     },
     {
@@ -990,9 +990,9 @@
     },
     {
       "epoch": 0.24,
-      "grad_norm": 0.140625,
+      "grad_norm": 0.1416015625,
       "learning_rate": 0.00019979236872634838,
-      "loss": 2.1821,
+      "loss": 2.1822,
       "step": 700
     },
     {
@@ -1020,7 +1020,7 @@
       "epoch": 0.24,
       "grad_norm": 0.1416015625,
       "learning_rate": 0.00019970922373395176,
-      "loss": 2.1797,
+      "loss": 2.1798,
       "step": 720
     },
     {
@@ -1039,7 +1039,7 @@
     },
     {
       "epoch": 0.25,
-      "grad_norm": 0.1474609375,
+      "grad_norm": 0.1484375,
       "learning_rate": 0.00019963771304567387,
       "loss": 2.1733,
       "step": 735
@@ -1048,7 +1048,7 @@
       "epoch": 0.25,
       "grad_norm": 0.146484375,
       "learning_rate": 0.00019961213400292762,
-      "loss": 2.1611,
+      "loss": 2.161,
       "step": 740
     },
     {
@@ -1069,42 +1069,42 @@
       "epoch": 0.26,
       "grad_norm": 0.1533203125,
       "learning_rate": 0.00019953017356491457,
-      "loss": 2.1972,
+      "loss": 2.1973,
       "step": 755
     },
     {
       "epoch": 0.26,
       "grad_norm": 0.150390625,
       "learning_rate": 0.0001995011131116679,
-      "loss": 2.1809,
+      "loss": 2.1808,
       "step": 760
     },
     {
       "epoch": 0.26,
       "grad_norm": 0.1513671875,
       "learning_rate": 0.00019947118292182377,
-      "loss": 2.1648,
+      "loss": 2.1649,
       "step": 765
     },
     {
       "epoch": 0.26,
       "grad_norm": 0.146484375,
       "learning_rate": 0.00019944038325700103,
-      "loss": 2.1709,
+      "loss": 2.171,
       "step": 770
     },
     {
       "epoch": 0.26,
       "grad_norm": 0.1474609375,
       "learning_rate": 0.00019940871438641882,
-      "loss": 2.1709,
+      "loss": 2.171,
       "step": 775
     },
     {
       "epoch": 0.26,
       "grad_norm": 0.1416015625,
       "learning_rate": 0.00019937617658689384,
-      "loss": 2.1944,
+      "loss": 2.1943,
       "step": 780
     },
     {
@@ -1118,7 +1118,7 @@
       "epoch": 0.27,
       "grad_norm": 0.1513671875,
       "learning_rate": 0.0001993084953462567,
-      "loss": 2.1785,
+      "loss": 2.1786,
       "step": 790
     },
     {
@@ -1174,7 +1174,7 @@
       "epoch": 0.28,
       "grad_norm": 0.1474609375,
       "learning_rate": 0.00019900307382679638,
-      "loss": 2.138,
+      "loss": 2.1378,
       "step": 830
     },
     {
@@ -1188,12 +1188,12 @@
       "epoch": 0.28,
       "grad_norm": 0.1474609375,
       "learning_rate": 0.00019891805765612794,
-      "loss": 2.1588,
+      "loss": 2.1589,
       "step": 840
     },
     {
       "epoch": 0.29,
-      "grad_norm": 0.1435546875,
+      "grad_norm": 0.14453125,
       "learning_rate": 0.0001988742524229346,
       "loss": 2.1581,
       "step": 845
@@ -1202,26 +1202,26 @@
       "epoch": 0.29,
       "grad_norm": 0.142578125,
       "learning_rate": 0.00019882958293251636,
-      "loss": 2.1803,
+      "loss": 2.1802,
       "step": 850
     },
     {
       "epoch": 0.29,
       "grad_norm": 0.138671875,
       "learning_rate": 0.00019878404957532814,
-      "loss": 2.1448,
+      "loss": 2.1449,
       "step": 855
     },
     {
       "epoch": 0.29,
-      "grad_norm": 0.1416015625,
+      "grad_norm": 0.142578125,
       "learning_rate": 0.00019873765274937578,
       "loss": 2.1585,
       "step": 860
     },
     {
       "epoch": 0.29,
-      "grad_norm": 0.1435546875,
+      "grad_norm": 0.14453125,
       "learning_rate": 0.00019869039286021271,
       "loss": 2.1749,
       "step": 865
@@ -1230,7 +1230,7 @@
       "epoch": 0.29,
       "grad_norm": 0.1474609375,
       "learning_rate": 0.00019864227032093633,
-      "loss": 2.1606,
+      "loss": 2.1607,
       "step": 870
     },
     {
@@ -1244,21 +1244,21 @@
       "epoch": 0.3,
       "grad_norm": 0.1435546875,
       "learning_rate": 0.0001985434389821319,
-      "loss": 2.1608,
+      "loss": 2.1609,
       "step": 880
     },
     {
       "epoch": 0.3,
       "grad_norm": 0.142578125,
       "learning_rate": 0.00019849273104648592,
-      "loss": 2.1602,
+      "loss": 2.1601,
       "step": 885
     },
     {
       "epoch": 0.3,
       "grad_norm": 0.1455078125,
       "learning_rate": 0.00019844116218848334,
-      "loss": 2.15,
+      "loss": 2.1498,
       "step": 890
     },
     {
@@ -1286,7 +1286,7 @@
       "epoch": 0.31,
       "grad_norm": 0.1474609375,
       "learning_rate": 0.0001982262866609439,
-      "loss": 2.1337,
+      "loss": 2.1336,
       "step": 910
     },
     {
@@ -1300,21 +1300,21 @@
       "epoch": 0.31,
       "grad_norm": 0.1435546875,
       "learning_rate": 0.00019811369543992183,
-      "loss": 2.1663,
+      "loss": 2.1662,
       "step": 920
     },
     {
       "epoch": 0.31,
-      "grad_norm": 0.14453125,
+      "grad_norm": 0.1435546875,
       "learning_rate": 0.0001980561131676718,
-      "loss": 2.1468,
+      "loss": 2.1467,
       "step": 925
     },
     {
       "epoch": 0.31,
       "grad_norm": 0.1474609375,
       "learning_rate": 0.0001979976737895305,
-      "loss": 2.1823,
+      "loss": 2.1824,
       "step": 930
     },
     {
@@ -1328,7 +1328,7 @@
       "epoch": 0.32,
       "grad_norm": 0.142578125,
       "learning_rate": 0.0001978782257663299,
-      "loss": 2.152,
+      "loss": 2.1521,
       "step": 940
     },
     {
@@ -1349,7 +1349,7 @@
       "epoch": 0.32,
       "grad_norm": 0.1455078125,
       "learning_rate": 0.00019769263845101828,
-      "loss": 2.1894,
+      "loss": 2.1895,
       "step": 955
     },
     {
@@ -1363,7 +1363,7 @@
       "epoch": 0.33,
       "grad_norm": 0.1484375,
       "learning_rate": 0.00019756464302906465,
-      "loss": 2.1603,
+      "loss": 2.1602,
       "step": 965
     },
     {
@@ -1384,7 +1384,7 @@
       "epoch": 0.33,
       "grad_norm": 0.140625,
       "learning_rate": 0.00019736625526650269,
-      "loss": 2.1671,
+      "loss": 2.1672,
       "step": 980
     },
     {
@@ -1398,12 +1398,12 @@
       "epoch": 0.34,
       "grad_norm": 0.1435546875,
       "learning_rate": 0.00019722974041542203,
-      "loss": 2.1563,
+      "loss": 2.1562,
       "step": 990
     },
     {
       "epoch": 0.34,
-      "grad_norm": 0.1396484375,
+      "grad_norm": 0.138671875,
       "learning_rate": 0.00019716020786582169,
       "loss": 2.1351,
       "step": 995
@@ -1412,14 +1412,14 @@
       "epoch": 0.34,
       "grad_norm": 0.1435546875,
       "learning_rate": 0.0001970898260414146,
-      "loss": 2.1661,
+      "loss": 2.1662,
       "step": 1000
     },
     {
       "epoch": 0.34,
       "grad_norm": 0.1572265625,
       "learning_rate": 0.00019701859555740648,
-      "loss": 2.16,
+      "loss": 2.1601,
       "step": 1005
     },
     {
@@ -1431,9 +1431,9 @@
     },
     {
       "epoch": 0.34,
-      "grad_norm": 0.1484375,
+      "grad_norm": 0.1474609375,
       "learning_rate": 0.0001968735911084948,
-      "loss": 2.1696,
+      "loss": 2.1697,
       "step": 1015
     },
     {
@@ -1447,7 +1447,7 @@
       "epoch": 0.35,
       "grad_norm": 0.1484375,
       "learning_rate": 0.00019672519958899583,
-      "loss": 2.1445,
+      "loss": 2.1444,
       "step": 1025
     },
     {
@@ -1461,14 +1461,14 @@
       "epoch": 0.35,
       "grad_norm": 0.146484375,
       "learning_rate": 0.00019657342618724358,
-      "loss": 2.1496,
+      "loss": 2.1495,
       "step": 1035
     },
     {
       "epoch": 0.35,
       "grad_norm": 0.146484375,
       "learning_rate": 0.00019649627293421413,
-      "loss": 2.1179,
+      "loss": 2.1181,
       "step": 1040
     },
     {
@@ -1489,21 +1489,21 @@
       "epoch": 0.36,
       "grad_norm": 0.1416015625,
       "learning_rate": 0.00019625975508134755,
-      "loss": 2.1596,
+      "loss": 2.1598,
       "step": 1055
     },
     {
       "epoch": 0.36,
       "grad_norm": 0.142578125,
       "learning_rate": 0.00019617923206290692,
-      "loss": 2.1651,
+      "loss": 2.165,
       "step": 1060
     },
     {
       "epoch": 0.36,
       "grad_norm": 0.150390625,
       "learning_rate": 0.00019609786834434313,
-      "loss": 2.1671,
+      "loss": 2.1672,
       "step": 1065
     },
     {
@@ -1517,7 +1517,7 @@
       "epoch": 0.36,
       "grad_norm": 0.14453125,
       "learning_rate": 0.00019593262165898076,
-      "loss": 2.1581,
+      "loss": 2.1579,
       "step": 1075
     },
     {
@@ -1531,7 +1531,7 @@
       "epoch": 0.37,
       "grad_norm": 0.1416015625,
       "learning_rate": 0.00019576402080291545,
-      "loss": 2.1963,
+      "loss": 2.1964,
       "step": 1085
     },
     {
@@ -1557,9 +1557,9 @@
     },
     {
       "epoch": 0.37,
-      "grad_norm": 0.1435546875,
+      "grad_norm": 0.142578125,
       "learning_rate": 0.00019541678027546296,
-      "loss": 2.1654,
+      "loss": 2.1653,
       "step": 1105
     },
     {
@@ -1578,7 +1578,7 @@
     },
     {
       "epoch": 0.38,
-      "grad_norm": 0.15234375,
+      "grad_norm": 0.1533203125,
       "learning_rate": 0.00019514758987623784,
       "loss": 2.1723,
       "step": 1120
@@ -1601,7 +1601,7 @@
       "epoch": 0.38,
       "grad_norm": 0.1455078125,
       "learning_rate": 0.00019487091437753594,
-      "loss": 2.1579,
+      "loss": 2.1581,
       "step": 1135
     },
     {
@@ -1620,7 +1620,7 @@
     },
     {
       "epoch": 0.39,
-      "grad_norm": 0.14453125,
+      "grad_norm": 0.1455078125,
       "learning_rate": 0.00019458677554494932,
       "loss": 2.1645,
       "step": 1150
@@ -1636,21 +1636,21 @@
       "epoch": 0.39,
       "grad_norm": 0.14453125,
       "learning_rate": 0.00019439321437422695,
-      "loss": 2.1561,
+      "loss": 2.1563,
       "step": 1160
     },
     {
       "epoch": 0.39,
       "grad_norm": 0.146484375,
       "learning_rate": 0.00019429519573119794,
-      "loss": 2.1558,
+      "loss": 2.1559,
       "step": 1165
     },
     {
       "epoch": 0.4,
-      "grad_norm": 0.150390625,
+      "grad_norm": 0.1494140625,
       "learning_rate": 0.00019419635285635746,
-      "loss": 2.1707,
+      "loss": 2.1708,
       "step": 1170
     },
     {
@@ -1664,7 +1664,7 @@
       "epoch": 0.4,
       "grad_norm": 0.1494140625,
       "learning_rate": 0.00019399619787437104,
-      "loss": 2.1422,
+      "loss": 2.1421,
       "step": 1180
     },
     {
@@ -1678,63 +1678,63 @@
       "epoch": 0.4,
       "grad_norm": 0.1455078125,
       "learning_rate": 0.00019379275642645002,
-      "loss": 2.1618,
+      "loss": 2.1617,
       "step": 1190
     },
     {
       "epoch": 0.4,
       "grad_norm": 0.1435546875,
       "learning_rate": 0.0001936898054961228,
-      "loss": 2.1826,
+      "loss": 2.1828,
       "step": 1195
     },
     {
       "epoch": 0.41,
       "grad_norm": 0.1484375,
       "learning_rate": 0.00019358603562568416,
-      "loss": 2.1238,
+      "loss": 2.1236,
       "step": 1200
     },
     {
       "epoch": 0.41,
       "grad_norm": 0.1435546875,
       "learning_rate": 0.0001934814477221838,
-      "loss": 2.1468,
+      "loss": 2.1469,
       "step": 1205
     },
     {
       "epoch": 0.41,
-      "grad_norm": 0.140625,
+      "grad_norm": 0.1416015625,
       "learning_rate": 0.0001933760426998218,
-      "loss": 2.1826,
+      "loss": 2.1827,
       "step": 1210
     },
     {
       "epoch": 0.41,
       "grad_norm": 0.14453125,
       "learning_rate": 0.0001932698214799407,
-      "loss": 2.1484,
+      "loss": 2.1483,
       "step": 1215
     },
     {
       "epoch": 0.41,
       "grad_norm": 0.14453125,
       "learning_rate": 0.0001931627849910174,
-      "loss": 2.1427,
+      "loss": 2.1429,
       "step": 1220
     },
     {
       "epoch": 0.41,
       "grad_norm": 0.1435546875,
       "learning_rate": 0.00019305493416865493,
-      "loss": 2.1281,
+      "loss": 2.128,
       "step": 1225
     },
     {
       "epoch": 0.42,
       "grad_norm": 0.14453125,
       "learning_rate": 0.00019294626995557457,
-      "loss": 2.1611,
+      "loss": 2.1609,
       "step": 1230
     },
     {
@@ -1755,56 +1755,56 @@
       "epoch": 0.42,
       "grad_norm": 0.14453125,
       "learning_rate": 0.00019261540650583522,
-      "loss": 2.1855,
+      "loss": 2.1856,
       "step": 1245
     },
     {
       "epoch": 0.42,
       "grad_norm": 0.142578125,
       "learning_rate": 0.00019250349829916661,
-      "loss": 2.1404,
+      "loss": 2.1403,
       "step": 1250
     },
     {
       "epoch": 0.43,
       "grad_norm": 0.1474609375,
       "learning_rate": 0.0001923907815218664,
-      "loss": 2.1797,
+      "loss": 2.1798,
       "step": 1255
     },
     {
       "epoch": 0.43,
       "grad_norm": 0.1416015625,
       "learning_rate": 0.00019227725715918897,
-      "loss": 2.1369,
+      "loss": 2.137,
       "step": 1260
     },
     {
       "epoch": 0.43,
       "grad_norm": 0.14453125,
       "learning_rate": 0.00019216292620344777,
-      "loss": 2.1737,
+      "loss": 2.1738,
       "step": 1265
     },
     {
       "epoch": 0.43,
       "grad_norm": 0.1533203125,
       "learning_rate": 0.00019204778965400667,
-      "loss": 2.1469,
+      "loss": 2.1471,
       "step": 1270
     },
     {
       "epoch": 0.43,
       "grad_norm": 0.1455078125,
       "learning_rate": 0.0001919318485172712,
-      "loss": 2.1758,
+      "loss": 2.1756,
       "step": 1275
     },
     {
       "epoch": 0.43,
       "grad_norm": 0.15234375,
       "learning_rate": 0.00019181510380667977,
-      "loss": 2.1649,
+      "loss": 2.165,
       "step": 1280
     },
     {
@@ -1818,19 +1818,19 @@
       "epoch": 0.44,
       "grad_norm": 0.1455078125,
       "learning_rate": 0.00019157920775279383,
-      "loss": 2.1468,
+      "loss": 2.1469,
       "step": 1290
     },
     {
       "epoch": 0.44,
       "grad_norm": 0.1513671875,
       "learning_rate": 0.0001914600584714605,
-      "loss": 2.1454,
+      "loss": 2.1457,
       "step": 1295
     },
     {
       "epoch": 0.44,
-      "grad_norm": 0.1552734375,
+      "grad_norm": 0.154296875,
       "learning_rate": 0.0001913401097401755,
       "loss": 2.1362,
       "step": 1300
@@ -1839,49 +1839,49 @@
       "epoch": 0.44,
       "grad_norm": 0.1474609375,
       "learning_rate": 0.00019121936260740752,
-      "loss": 2.1761,
+      "loss": 2.1762,
       "step": 1305
     },
     {
       "epoch": 0.44,
       "grad_norm": 0.1435546875,
       "learning_rate": 0.0001910978181286041,
-      "loss": 2.1609,
+      "loss": 2.161,
       "step": 1310
     },
     {
       "epoch": 0.45,
-      "grad_norm": 0.1474609375,
+      "grad_norm": 0.146484375,
       "learning_rate": 0.00019097547736618228,
       "loss": 2.1565,
       "step": 1315
     },
     {
       "epoch": 0.45,
-      "grad_norm": 0.1474609375,
+      "grad_norm": 0.146484375,
       "learning_rate": 0.0001908523413895194,
-      "loss": 2.1391,
+      "loss": 2.139,
       "step": 1320
     },
     {
       "epoch": 0.45,
       "grad_norm": 0.14453125,
       "learning_rate": 0.0001907284112749438,
-      "loss": 2.1393,
+      "loss": 2.1392,
       "step": 1325
     },
     {
       "epoch": 0.45,
-      "grad_norm": 0.1474609375,
+      "grad_norm": 0.146484375,
       "learning_rate": 0.00019060368810572539,
-      "loss": 2.1562,
+      "loss": 2.1563,
       "step": 1330
     },
     {
       "epoch": 0.45,
-      "grad_norm": 0.1494140625,
+      "grad_norm": 0.150390625,
       "learning_rate": 0.00019047817297206598,
-      "loss": 2.1202,
+      "loss": 2.1203,
       "step": 1335
     },
     {
@@ -1895,28 +1895,28 @@
       "epoch": 0.46,
       "grad_norm": 0.14453125,
       "learning_rate": 0.0001902247712068352,
-      "loss": 2.1503,
+      "loss": 2.1504,
       "step": 1345
     },
     {
       "epoch": 0.46,
-      "grad_norm": 0.142578125,
+      "grad_norm": 0.1435546875,
       "learning_rate": 0.0001900968867902419,
-      "loss": 2.1799,
+      "loss": 2.1798,
       "step": 1350
     },
     {
       "epoch": 0.46,
       "grad_norm": 0.150390625,
       "learning_rate": 0.0001899682148391446,
-      "loss": 2.154,
+      "loss": 2.1541,
       "step": 1355
     },
     {
       "epoch": 0.46,
       "grad_norm": 0.1513671875,
       "learning_rate": 0.00018983875647826136,
-      "loss": 2.1285,
+      "loss": 2.1286,
       "step": 1360
     },
     {
@@ -1930,7 +1930,7 @@
       "epoch": 0.46,
       "grad_norm": 0.1435546875,
       "learning_rate": 0.00018957748506036957,
-      "loss": 2.1369,
+      "loss": 2.137,
       "step": 1370
     },
     {
@@ -1942,7 +1942,7 @@
     },
     {
       "epoch": 0.47,
-      "grad_norm": 0.1474609375,
+      "grad_norm": 0.1484375,
       "learning_rate": 0.0001893130816716129,
       "loss": 2.1668,
       "step": 1380
@@ -1956,7 +1956,7 @@
     },
     {
       "epoch": 0.47,
-      "grad_norm": 0.1484375,
+      "grad_norm": 0.1474609375,
       "learning_rate": 0.00018904555555654317,
       "loss": 2.1846,
       "step": 1390
@@ -1972,21 +1972,21 @@
       "epoch": 0.47,
       "grad_norm": 0.1484375,
       "learning_rate": 0.00018877491606889476,
-      "loss": 2.1312,
+      "loss": 2.1311,
       "step": 1400
     },
     {
       "epoch": 0.48,
-      "grad_norm": 0.146484375,
+      "grad_norm": 0.1474609375,
       "learning_rate": 0.0001886384317631697,
-      "loss": 2.1694,
+      "loss": 2.1693,
       "step": 1405
     },
     {
       "epoch": 0.48,
       "grad_norm": 0.146484375,
       "learning_rate": 0.00018850117267125738,
-      "loss": 2.1772,
+      "loss": 2.1773,
       "step": 1410
     },
     {
@@ -2028,42 +2028,42 @@
       "epoch": 0.49,
       "grad_norm": 0.1474609375,
       "learning_rate": 0.00018766141527016533,
-      "loss": 2.1336,
+      "loss": 2.1334,
       "step": 1440
     },
     {
       "epoch": 0.49,
       "grad_norm": 0.14453125,
       "learning_rate": 0.00018751876664665367,
-      "loss": 2.1352,
+      "loss": 2.1351,
       "step": 1445
     },
     {
       "epoch": 0.49,
-      "grad_norm": 0.150390625,
+      "grad_norm": 0.1513671875,
       "learning_rate": 0.00018737535302391795,
       "loss": 2.1777,
       "step": 1450
     },
     {
       "epoch": 0.49,
-      "grad_norm": 0.1484375,
+      "grad_norm": 0.1474609375,
       "learning_rate": 0.00018723117565553284,
-      "loss": 2.1516,
+      "loss": 2.1515,
       "step": 1455
     },
     {
       "epoch": 0.49,
       "grad_norm": 0.150390625,
       "learning_rate": 0.00018708623580174889,
-      "loss": 2.1431,
+      "loss": 2.1432,
       "step": 1460
     },
     {
       "epoch": 0.5,
       "grad_norm": 0.1484375,
       "learning_rate": 0.00018694053472948156,
-      "loss": 2.1385,
+      "loss": 2.1384,
       "step": 1465
     },
     {
@@ -2077,42 +2077,42 @@
       "epoch": 0.5,
       "grad_norm": 0.1484375,
       "learning_rate": 0.00018664685403041619,
-      "loss": 2.1623,
+      "loss": 2.1624,
       "step": 1475
     },
     {
       "epoch": 0.5,
       "grad_norm": 0.142578125,
       "learning_rate": 0.0001864988769706734,
-      "loss": 2.1619,
+      "loss": 2.1621,
       "step": 1480
     },
     {
       "epoch": 0.5,
-      "grad_norm": 0.1455078125,
+      "grad_norm": 0.146484375,
       "learning_rate": 0.0001863501438265352,
-      "loss": 2.1514,
+      "loss": 2.1513,
       "step": 1485
     },
     {
       "epoch": 0.5,
       "grad_norm": 0.1494140625,
       "learning_rate": 0.00018620065589807413,
-      "loss": 2.1649,
+      "loss": 2.1648,
       "step": 1490
     },
     {
       "epoch": 0.51,
       "grad_norm": 0.1484375,
       "learning_rate": 0.00018605041449196012,
-      "loss": 2.1802,
+      "loss": 2.18,
       "step": 1495
     },
     {
       "epoch": 0.51,
       "grad_norm": 0.14453125,
       "learning_rate": 0.00018589942092144942,
-      "loss": 2.1805,
+      "loss": 2.1806,
       "step": 1500
     },
     {
@@ -2126,7 +2126,7 @@
       "epoch": 0.51,
       "grad_norm": 0.1484375,
       "learning_rate": 0.0001855951825731241,
-      "loss": 2.1333,
+      "loss": 2.1332,
       "step": 1510
     },
     {
@@ -2138,23 +2138,23 @@
     },
     {
       "epoch": 0.51,
-      "grad_norm": 0.1474609375,
+      "grad_norm": 0.146484375,
       "learning_rate": 0.00018528795149043236,
-      "loss": 2.158,
+      "loss": 2.1581,
       "step": 1520
     },
     {
       "epoch": 0.52,
-      "grad_norm": 0.1484375,
+      "grad_norm": 0.1494140625,
       "learning_rate": 0.00018513321702648807,
-      "loss": 2.1432,
+      "loss": 2.1433,
       "step": 1525
     },
     {
       "epoch": 0.52,
-      "grad_norm": 0.1474609375,
+      "grad_norm": 0.1484375,
       "learning_rate": 0.0001849777384153458,
-      "loss": 2.1542,
+      "loss": 2.1543,
       "step": 1530
     },
     {
@@ -2175,14 +2175,14 @@
       "epoch": 0.52,
       "grad_norm": 0.1455078125,
       "learning_rate": 0.0001845068513215271,
-      "loss": 2.1429,
+      "loss": 2.143,
       "step": 1545
     },
     {
       "epoch": 0.52,
       "grad_norm": 0.146484375,
       "learning_rate": 0.00018434840977680453,
-      "loss": 2.1376,
+      "loss": 2.1375,
       "step": 1550
     },
     {
@@ -2203,28 +2203,28 @@
       "epoch": 0.53,
       "grad_norm": 0.1494140625,
       "learning_rate": 0.00018386866699126973,
-      "loss": 2.1788,
+      "loss": 2.1789,
       "step": 1565
     },
     {
       "epoch": 0.53,
       "grad_norm": 0.1474609375,
       "learning_rate": 0.00018370728467165828,
-      "loss": 2.142,
+      "loss": 2.1419,
       "step": 1570
     },
     {
       "epoch": 0.53,
       "grad_norm": 0.146484375,
       "learning_rate": 0.0001835451706688859,
-      "loss": 2.175,
+      "loss": 2.1748,
       "step": 1575
     },
     {
       "epoch": 0.54,
       "grad_norm": 0.1455078125,
       "learning_rate": 0.0001833823263999867,
-      "loss": 2.1268,
+      "loss": 2.1267,
       "step": 1580
     },
     {
@@ -2238,14 +2238,14 @@
       "epoch": 0.54,
       "grad_norm": 0.1494140625,
       "learning_rate": 0.00018305445276384875,
-      "loss": 2.1436,
+      "loss": 2.1438,
       "step": 1590
     },
     {
       "epoch": 0.54,
       "grad_norm": 0.1474609375,
       "learning_rate": 0.00018288942626254473,
-      "loss": 2.1573,
+      "loss": 2.1572,
       "step": 1595
     },
     {
@@ -2259,7 +2259,7 @@
       "epoch": 0.54,
       "grad_norm": 0.1455078125,
       "learning_rate": 0.00018255720110591533,
-      "loss": 2.157,
+      "loss": 2.1568,
       "step": 1605
     },
     {
@@ -2287,7 +2287,7 @@
       "epoch": 0.55,
       "grad_norm": 0.1533203125,
       "learning_rate": 0.0001818841029646321,
-      "loss": 2.1532,
+      "loss": 2.153,
       "step": 1625
     },
     {
@@ -2301,14 +2301,14 @@
       "epoch": 0.55,
       "grad_norm": 0.1474609375,
       "learning_rate": 0.00018154325351405897,
-      "loss": 2.1849,
+      "loss": 2.185,
       "step": 1635
     },
     {
       "epoch": 0.56,
       "grad_norm": 0.1513671875,
       "learning_rate": 0.00018137175889123978,
-      "loss": 2.1444,
+      "loss": 2.1443,
       "step": 1640
     },
     {
@@ -2322,21 +2322,21 @@
       "epoch": 0.56,
       "grad_norm": 0.1494140625,
       "learning_rate": 0.00018102663734567283,
-      "loss": 2.1413,
+      "loss": 2.1414,
       "step": 1650
     },
     {
       "epoch": 0.56,
       "grad_norm": 0.1494140625,
       "learning_rate": 0.0001808530134396234,
-      "loss": 2.1554,
+      "loss": 2.1552,
       "step": 1655
     },
     {
       "epoch": 0.56,
       "grad_norm": 0.1474609375,
       "learning_rate": 0.00018067868279952236,
-      "loss": 2.1375,
+      "loss": 2.1374,
       "step": 1660
     },
     {
@@ -2355,44 +2355,44 @@
     },
     {
       "epoch": 0.57,
-      "grad_norm": 0.1513671875,
+      "grad_norm": 0.150390625,
       "learning_rate": 0.00018015146574391233,
-      "loss": 2.1284,
+      "loss": 2.1283,
       "step": 1675
     },
     {
       "epoch": 0.57,
       "grad_norm": 0.1484375,
       "learning_rate": 0.00017997432346737524,
-      "loss": 2.1505,
+      "loss": 2.1504,
       "step": 1680
     },
     {
       "epoch": 0.57,
       "grad_norm": 0.1474609375,
       "learning_rate": 0.00017979648213739232,
-      "loss": 2.1339,
+      "loss": 2.1338,
       "step": 1685
     },
     {
       "epoch": 0.57,
       "grad_norm": 0.1474609375,
       "learning_rate": 0.00017961794330846994,
-      "loss": 2.1554,
+      "loss": 2.1555,
       "step": 1690
     },
     {
       "epoch": 0.57,
-      "grad_norm": 0.150390625,
+      "grad_norm": 0.1513671875,
       "learning_rate": 0.00017943870854121124,
       "loss": 2.1543,
       "step": 1695
     },
     {
       "epoch": 0.58,
-      "grad_norm": 0.1533203125,
+      "grad_norm": 0.15234375,
       "learning_rate": 0.0001792587794023027,
-      "loss": 2.1429,
+      "loss": 2.1428,
       "step": 1700
     },
     {
@@ -2418,9 +2418,9 @@
     },
     {
       "epoch": 0.58,
-      "grad_norm": 0.150390625,
+      "grad_norm": 0.1484375,
       "learning_rate": 0.00017853215067603926,
-      "loss": 2.1919,
+      "loss": 2.1918,
       "step": 1720
     },
     {
@@ -2439,7 +2439,7 @@
     },
     {
       "epoch": 0.59,
-      "grad_norm": 0.1474609375,
+      "grad_norm": 0.146484375,
       "learning_rate": 0.0001779799658965153,
       "loss": 2.137,
       "step": 1735
@@ -2448,49 +2448,49 @@
       "epoch": 0.59,
       "grad_norm": 0.146484375,
       "learning_rate": 0.00017779453891055412,
-      "loss": 2.1714,
+      "loss": 2.1713,
       "step": 1740
     },
     {
       "epoch": 0.59,
-      "grad_norm": 0.1533203125,
+      "grad_norm": 0.154296875,
       "learning_rate": 0.00017760843192458626,
-      "loss": 2.1456,
+      "loss": 2.1457,
       "step": 1745
     },
     {
       "epoch": 0.59,
       "grad_norm": 0.150390625,
       "learning_rate": 0.00017742164656536798,
-      "loss": 2.1575,
+      "loss": 2.1576,
       "step": 1750
     },
     {
       "epoch": 0.59,
-      "grad_norm": 0.1552734375,
+      "grad_norm": 0.154296875,
       "learning_rate": 0.00017723418446558516,
       "loss": 2.1461,
       "step": 1755
     },
     {
       "epoch": 0.6,
-      "grad_norm": 0.1484375,
+      "grad_norm": 0.1474609375,
       "learning_rate": 0.00017704604726383904,
       "loss": 2.1383,
       "step": 1760
     },
     {
       "epoch": 0.6,
-      "grad_norm": 0.1474609375,
+      "grad_norm": 0.1484375,
       "learning_rate": 0.00017685723660463193,
-      "loss": 2.13,
+      "loss": 2.1299,
       "step": 1765
     },
     {
       "epoch": 0.6,
-      "grad_norm": 0.1484375,
+      "grad_norm": 0.1474609375,
       "learning_rate": 0.00017666775413835282,
-      "loss": 2.1463,
+      "loss": 2.1461,
       "step": 1770
     },
     {
@@ -2504,19 +2504,19 @@
       "epoch": 0.6,
       "grad_norm": 0.1484375,
       "learning_rate": 0.0001762867804154812,
-      "loss": 2.144,
+      "loss": 2.1439,
       "step": 1780
     },
     {
       "epoch": 0.6,
-      "grad_norm": 0.1484375,
+      "grad_norm": 0.1494140625,
       "learning_rate": 0.00017609529248896997,
       "loss": 2.1802,
       "step": 1785
     },
     {
       "epoch": 0.61,
-      "grad_norm": 0.15234375,
+      "grad_norm": 0.1513671875,
       "learning_rate": 0.00017590313941552002,
       "loss": 2.1746,
       "step": 1790
@@ -2525,28 +2525,28 @@
       "epoch": 0.61,
       "grad_norm": 0.1513671875,
       "learning_rate": 0.00017571032287473642,
-      "loss": 2.1199,
+      "loss": 2.1198,
       "step": 1795
     },
     {
       "epoch": 0.61,
       "grad_norm": 0.15234375,
       "learning_rate": 0.00017551684455202336,
-      "loss": 2.1279,
+      "loss": 2.128,
       "step": 1800
     },
     {
       "epoch": 0.61,
       "grad_norm": 0.1494140625,
       "learning_rate": 0.00017532270613856976,
-      "loss": 2.1532,
+      "loss": 2.153,
       "step": 1805
     },
     {
       "epoch": 0.61,
       "grad_norm": 0.1494140625,
       "learning_rate": 0.00017512790933133437,
-      "loss": 2.1274,
+      "loss": 2.1273,
       "step": 1810
     },
     {
@@ -2560,28 +2560,28 @@
       "epoch": 0.62,
       "grad_norm": 0.146484375,
       "learning_rate": 0.0001747363473521132,
-      "loss": 2.1483,
+      "loss": 2.1482,
       "step": 1820
     },
     {
       "epoch": 0.62,
       "grad_norm": 0.1533203125,
       "learning_rate": 0.00017453958560276038,
-      "loss": 2.1474,
+      "loss": 2.1475,
       "step": 1825
     },
     {
       "epoch": 0.62,
       "grad_norm": 0.1494140625,
       "learning_rate": 0.00017434217230486164,
-      "loss": 2.1605,
+      "loss": 2.1606,
       "step": 1830
     },
     {
       "epoch": 0.62,
-      "grad_norm": 0.15234375,
+      "grad_norm": 0.1513671875,
       "learning_rate": 0.0001741441091840014,
-      "loss": 2.1658,
+      "loss": 2.1657,
       "step": 1835
     },
     {
@@ -2595,56 +2595,56 @@
       "epoch": 0.62,
       "grad_norm": 0.150390625,
       "learning_rate": 0.00017374604040411935,
-      "loss": 2.1318,
+      "loss": 2.1317,
       "step": 1845
     },
     {
       "epoch": 0.63,
       "grad_norm": 0.150390625,
       "learning_rate": 0.00017354603822460621,
-      "loss": 2.1321,
+      "loss": 2.1323,
       "step": 1850
     },
     {
       "epoch": 0.63,
       "grad_norm": 0.154296875,
       "learning_rate": 0.00017334539318111856,
-      "loss": 2.1417,
+      "loss": 2.1418,
       "step": 1855
     },
     {
       "epoch": 0.63,
       "grad_norm": 0.150390625,
       "learning_rate": 0.00017314410702748932,
-      "loss": 2.1343,
+      "loss": 2.1342,
       "step": 1860
     },
     {
       "epoch": 0.63,
-      "grad_norm": 0.15234375,
+      "grad_norm": 0.1533203125,
       "learning_rate": 0.00017294218152315546,
       "loss": 2.1471,
       "step": 1865
     },
     {
       "epoch": 0.63,
-      "grad_norm": 0.150390625,
+      "grad_norm": 0.1494140625,
       "learning_rate": 0.00017273961843314252,
-      "loss": 2.147,
+      "loss": 2.1469,
       "step": 1870
     },
     {
       "epoch": 0.64,
       "grad_norm": 0.1552734375,
       "learning_rate": 0.0001725364195280491,
-      "loss": 2.1528,
+      "loss": 2.1527,
       "step": 1875
     },
     {
       "epoch": 0.64,
       "grad_norm": 0.1533203125,
       "learning_rate": 0.00017233258658403138,
-      "loss": 2.1462,
+      "loss": 2.1461,
       "step": 1880
     },
     {
@@ -2658,7 +2658,7 @@
       "epoch": 0.64,
       "grad_norm": 0.1494140625,
       "learning_rate": 0.00017192302571154331,
-      "loss": 2.1613,
+      "loss": 2.1612,
       "step": 1890
     },
     {
@@ -2672,7 +2672,7 @@
       "epoch": 0.64,
       "grad_norm": 0.1474609375,
       "learning_rate": 0.00017151095013548994,
-      "loss": 2.1443,
+      "loss": 2.1444,
       "step": 1900
     },
     {
@@ -2684,37 +2684,37 @@
     },
     {
       "epoch": 0.65,
-      "grad_norm": 0.1455078125,
+      "grad_norm": 0.146484375,
       "learning_rate": 0.0001710963742636067,
-      "loss": 2.1351,
+      "loss": 2.1352,
       "step": 1910
     },
     {
       "epoch": 0.65,
-      "grad_norm": 0.1494140625,
+      "grad_norm": 0.150390625,
       "learning_rate": 0.00017088815324306392,
       "loss": 2.1285,
       "step": 1915
     },
     {
       "epoch": 0.65,
-      "grad_norm": 0.154296875,
+      "grad_norm": 0.1552734375,
       "learning_rate": 0.00017067931259104885,
       "loss": 2.1599,
       "step": 1920
     },
     {
       "epoch": 0.65,
-      "grad_norm": 0.1484375,
+      "grad_norm": 0.1494140625,
       "learning_rate": 0.00017046985413303215,
-      "loss": 2.1219,
+      "loss": 2.1218,
       "step": 1925
     },
     {
       "epoch": 0.65,
-      "grad_norm": 0.1484375,
+      "grad_norm": 0.1494140625,
       "learning_rate": 0.00017025977969988465,
-      "loss": 2.1373,
+      "loss": 2.1374,
       "step": 1930
     },
     {
@@ -2735,14 +2735,14 @@
       "epoch": 0.66,
       "grad_norm": 0.1494140625,
       "learning_rate": 0.00016962587893903276,
-      "loss": 2.1502,
+      "loss": 2.1501,
       "step": 1945
     },
     {
       "epoch": 0.66,
       "grad_norm": 0.150390625,
       "learning_rate": 0.0001694133590215139,
-      "loss": 2.1611,
+      "loss": 2.1612,
       "step": 1950
     },
     {
@@ -2756,14 +2756,14 @@
       "epoch": 0.66,
       "grad_norm": 0.146484375,
       "learning_rate": 0.00016898650082840572,
-      "loss": 2.1338,
+      "loss": 2.1339,
       "step": 1960
     },
     {
       "epoch": 0.67,
       "grad_norm": 0.1474609375,
       "learning_rate": 0.00016877216628397257,
-      "loss": 2.1257,
+      "loss": 2.1258,
       "step": 1965
     },
     {
@@ -2784,70 +2784,70 @@
       "epoch": 0.67,
       "grad_norm": 0.1513671875,
       "learning_rate": 0.0001681255633567823,
-      "loss": 2.1571,
+      "loss": 2.1573,
       "step": 1980
     },
     {
       "epoch": 0.67,
       "grad_norm": 0.150390625,
       "learning_rate": 0.00016790883556302272,
-      "loss": 2.152,
+      "loss": 2.1519,
       "step": 1985
     },
     {
       "epoch": 0.67,
-      "grad_norm": 0.158203125,
+      "grad_norm": 0.1572265625,
       "learning_rate": 0.00016769151417992791,
-      "loss": 2.1425,
+      "loss": 2.1426,
       "step": 1990
     },
     {
       "epoch": 0.68,
       "grad_norm": 0.1513671875,
       "learning_rate": 0.00016747360110709838,
-      "loss": 2.1405,
+      "loss": 2.1406,
       "step": 1995
     },
     {
       "epoch": 0.68,
       "grad_norm": 0.1552734375,
       "learning_rate": 0.00016725509824930645,
-      "loss": 2.1496,
+      "loss": 2.1495,
       "step": 2000
     },
     {
       "epoch": 0.68,
       "grad_norm": 0.154296875,
       "learning_rate": 0.0001670360075164799,
-      "loss": 2.1588,
+      "loss": 2.1589,
       "step": 2005
     },
     {
       "epoch": 0.68,
       "grad_norm": 0.150390625,
       "learning_rate": 0.00016681633082368498,
-      "loss": 2.1571,
+      "loss": 2.1569,
       "step": 2010
     },
     {
       "epoch": 0.68,
       "grad_norm": 0.150390625,
       "learning_rate": 0.00016659607009110984,
-      "loss": 2.1428,
+      "loss": 2.1429,
       "step": 2015
     },
     {
       "epoch": 0.68,
       "grad_norm": 0.15234375,
       "learning_rate": 0.00016637522724404774,
-      "loss": 2.1548,
+      "loss": 2.1547,
       "step": 2020
     },
     {
       "epoch": 0.69,
-      "grad_norm": 0.1494140625,
+      "grad_norm": 0.150390625,
       "learning_rate": 0.00016615380421288018,
-      "loss": 2.1186,
+      "loss": 2.1187,
       "step": 2025
     },
     {
@@ -2866,9 +2866,9 @@
     },
     {
       "epoch": 0.69,
-      "grad_norm": 0.1474609375,
+      "grad_norm": 0.1484375,
       "learning_rate": 0.00016548607339452853,
-      "loss": 2.1105,
+      "loss": 2.1106,
       "step": 2040
     },
     {
@@ -2882,7 +2882,7 @@
       "epoch": 0.69,
       "grad_norm": 0.1484375,
       "learning_rate": 0.00016503805421285968,
-      "loss": 2.1515,
+      "loss": 2.1514,
       "step": 2050
     },
     {
@@ -2896,28 +2896,28 @@
       "epoch": 0.7,
       "grad_norm": 0.1494140625,
       "learning_rate": 0.00016458776105251447,
-      "loss": 2.1587,
+      "loss": 2.1588,
       "step": 2060
     },
     {
       "epoch": 0.7,
       "grad_norm": 0.1494140625,
       "learning_rate": 0.0001643617666472376,
-      "loss": 2.1292,
+      "loss": 2.1291,
       "step": 2065
     },
     {
       "epoch": 0.7,
       "grad_norm": 0.1513671875,
       "learning_rate": 0.00016413520965746097,
-      "loss": 2.1662,
+      "loss": 2.1661,
       "step": 2070
     },
     {
       "epoch": 0.7,
       "grad_norm": 0.1484375,
       "learning_rate": 0.0001639080920635134,
-      "loss": 2.1416,
+      "loss": 2.1415,
       "step": 2075
     },
     {
@@ -2929,28 +2929,28 @@
     },
     {
       "epoch": 0.71,
-      "grad_norm": 0.1533203125,
+      "grad_norm": 0.15234375,
       "learning_rate": 0.00016345218300890357,
-      "loss": 2.1246,
+      "loss": 2.1245,
       "step": 2085
     },
     {
       "epoch": 0.71,
       "grad_norm": 0.1533203125,
       "learning_rate": 0.00016322339553333034,
-      "loss": 2.1425,
+      "loss": 2.1427,
       "step": 2090
     },
     {
       "epoch": 0.71,
       "grad_norm": 0.1484375,
       "learning_rate": 0.00016299405542372924,
-      "loss": 2.1527,
+      "loss": 2.1526,
       "step": 2095
     },
     {
       "epoch": 0.71,
-      "grad_norm": 0.15625,
+      "grad_norm": 0.1572265625,
       "learning_rate": 0.00016276416468475607,
       "loss": 2.1265,
       "step": 2100
@@ -2959,28 +2959,28 @@
       "epoch": 0.71,
       "grad_norm": 0.1484375,
       "learning_rate": 0.00016253372532587976,
-      "loss": 2.1505,
+      "loss": 2.1504,
       "step": 2105
     },
     {
       "epoch": 0.71,
       "grad_norm": 0.1533203125,
       "learning_rate": 0.0001623027393613646,
-      "loss": 2.1566,
+      "loss": 2.1564,
       "step": 2110
     },
     {
       "epoch": 0.72,
       "grad_norm": 0.1552734375,
       "learning_rate": 0.00016207120881025282,
-      "loss": 2.1498,
+      "loss": 2.1497,
       "step": 2115
     },
     {
       "epoch": 0.72,
       "grad_norm": 0.1474609375,
       "learning_rate": 0.0001618391356963468,
-      "loss": 2.1094,
+      "loss": 2.1096,
       "step": 2120
     },
     {
@@ -3001,14 +3001,14 @@
       "epoch": 0.72,
       "grad_norm": 0.150390625,
       "learning_rate": 0.00016113968128691933,
-      "loss": 2.1403,
+      "loss": 2.1404,
       "step": 2135
     },
     {
       "epoch": 0.72,
       "grad_norm": 0.15234375,
       "learning_rate": 0.00016090545825444506,
-      "loss": 2.1621,
+      "loss": 2.1619,
       "step": 2140
     },
     {
@@ -3027,7 +3027,7 @@
     },
     {
       "epoch": 0.73,
-      "grad_norm": 0.150390625,
+      "grad_norm": 0.1513671875,
       "learning_rate": 0.00016019960313162434,
       "loss": 2.1193,
       "step": 2155
@@ -3048,7 +3048,7 @@
     },
     {
       "epoch": 0.73,
-      "grad_norm": 0.154296875,
+      "grad_norm": 0.1552734375,
       "learning_rate": 0.00015948901220815445,
       "loss": 2.1639,
       "step": 2170
@@ -3064,7 +3064,7 @@
       "epoch": 0.74,
       "grad_norm": 0.1533203125,
       "learning_rate": 0.00015901268151570491,
-      "loss": 2.1863,
+      "loss": 2.1862,
       "step": 2180
     },
     {
@@ -3092,21 +3092,21 @@
       "epoch": 0.75,
       "grad_norm": 0.15234375,
       "learning_rate": 0.0001580538469313566,
-      "loss": 2.1352,
+      "loss": 2.1354,
       "step": 2200
     },
     {
       "epoch": 0.75,
       "grad_norm": 0.1484375,
       "learning_rate": 0.00015781286441802534,
-      "loss": 2.135,
+      "loss": 2.1349,
       "step": 2205
     },
     {
       "epoch": 0.75,
       "grad_norm": 0.1552734375,
       "learning_rate": 0.00015757137656397557,
-      "loss": 2.1041,
+      "loss": 2.1039,
       "step": 2210
     },
     {
@@ -3120,47 +3120,47 @@
       "epoch": 0.75,
       "grad_norm": 0.1533203125,
       "learning_rate": 0.00015708689328147493,
-      "loss": 2.1311,
+      "loss": 2.131,
       "step": 2220
     },
     {
       "epoch": 0.75,
       "grad_norm": 0.1513671875,
       "learning_rate": 0.00015684390208787962,
-      "loss": 2.143,
+      "loss": 2.1432,
       "step": 2225
     },
     {
       "epoch": 0.76,
-      "grad_norm": 0.15234375,
+      "grad_norm": 0.1533203125,
       "learning_rate": 0.0001566004140232399,
-      "loss": 2.1528,
+      "loss": 2.1529,
       "step": 2230
     },
     {
       "epoch": 0.76,
-      "grad_norm": 0.146484375,
+      "grad_norm": 0.1474609375,
       "learning_rate": 0.00015635643121587848,
-      "loss": 2.164,
+      "loss": 2.1641,
       "step": 2235
     },
     {
       "epoch": 0.76,
       "grad_norm": 0.1552734375,
       "learning_rate": 0.00015611195579844265,
-      "loss": 2.1369,
+      "loss": 2.1368,
       "step": 2240
     },
     {
       "epoch": 0.76,
-      "grad_norm": 0.1572265625,
+      "grad_norm": 0.15625,
       "learning_rate": 0.00015586698990788554,
-      "loss": 2.1538,
+      "loss": 2.1537,
       "step": 2245
     },
     {
       "epoch": 0.76,
-      "grad_norm": 0.1484375,
+      "grad_norm": 0.1474609375,
       "learning_rate": 0.00015562153568544752,
       "loss": 2.1305,
       "step": 2250
@@ -3169,21 +3169,21 @@
       "epoch": 0.76,
       "grad_norm": 0.15234375,
       "learning_rate": 0.00015537559527663744,
-      "loss": 2.1832,
+      "loss": 2.1831,
       "step": 2255
     },
     {
       "epoch": 0.77,
       "grad_norm": 0.15625,
       "learning_rate": 0.00015512917083121397,
-      "loss": 2.1522,
+      "loss": 2.1523,
       "step": 2260
     },
     {
       "epoch": 0.77,
       "grad_norm": 0.15234375,
       "learning_rate": 0.00015488226450316664,
-      "loss": 2.1116,
+      "loss": 2.1115,
       "step": 2265
     },
     {
@@ -3197,7 +3197,7 @@
       "epoch": 0.77,
       "grad_norm": 0.1572265625,
       "learning_rate": 0.00015438701483620027,
-      "loss": 2.1485,
+      "loss": 2.1486,
       "step": 2275
     },
     {
@@ -3211,12 +3211,12 @@
       "epoch": 0.77,
       "grad_norm": 0.15234375,
       "learning_rate": 0.00015388986359155758,
-      "loss": 2.1407,
+      "loss": 2.1408,
       "step": 2285
     },
     {
       "epoch": 0.78,
-      "grad_norm": 0.1572265625,
+      "grad_norm": 0.1552734375,
       "learning_rate": 0.0001536405803069975,
       "loss": 2.1033,
       "step": 2290
@@ -3225,7 +3225,7 @@
       "epoch": 0.78,
       "grad_norm": 0.1513671875,
       "learning_rate": 0.00015339082815154394,
-      "loss": 2.1388,
+      "loss": 2.1389,
       "step": 2295
     },
     {
@@ -3246,7 +3246,7 @@
       "epoch": 0.78,
       "grad_norm": 0.1513671875,
       "learning_rate": 0.00015263878031096975,
-      "loss": 2.1443,
+      "loss": 2.1442,
       "step": 2310
     },
     {
@@ -3281,21 +3281,21 @@
       "epoch": 0.79,
       "grad_norm": 0.1552734375,
       "learning_rate": 0.00015137619437681767,
-      "loss": 2.1289,
+      "loss": 2.1287,
       "step": 2335
     },
     {
       "epoch": 0.79,
       "grad_norm": 0.162109375,
       "learning_rate": 0.00015112232110728015,
-      "loss": 2.1517,
+      "loss": 2.1516,
       "step": 2340
     },
     {
       "epoch": 0.79,
       "grad_norm": 0.15234375,
       "learning_rate": 0.00015086800097888624,
-      "loss": 2.1423,
+      "loss": 2.1424,
       "step": 2345
     },
     {
@@ -3314,7 +3314,7 @@
     },
     {
       "epoch": 0.8,
-      "grad_norm": 0.1513671875,
+      "grad_norm": 0.15234375,
       "learning_rate": 0.0001501023816900342,
       "loss": 2.168,
       "step": 2360
@@ -3330,7 +3330,7 @@
       "epoch": 0.8,
       "grad_norm": 0.150390625,
       "learning_rate": 0.0001495897753948833,
-      "loss": 2.1486,
+      "loss": 2.1485,
       "step": 2370
     },
     {
@@ -3342,7 +3342,7 @@
     },
     {
       "epoch": 0.81,
-      "grad_norm": 0.1552734375,
+      "grad_norm": 0.15625,
       "learning_rate": 0.00014907543525186166,
       "loss": 2.1244,
       "step": 2380
@@ -3351,7 +3351,7 @@
       "epoch": 0.81,
       "grad_norm": 0.1533203125,
       "learning_rate": 0.00014881762060482814,
-      "loss": 2.123,
+      "loss": 2.1229,
       "step": 2385
     },
     {
@@ -3365,26 +3365,26 @@
       "epoch": 0.81,
       "grad_norm": 0.1513671875,
       "learning_rate": 0.00014830071342745112,
-      "loss": 2.1471,
+      "loss": 2.1472,
       "step": 2395
     },
     {
       "epoch": 0.81,
-      "grad_norm": 0.1533203125,
+      "grad_norm": 0.15234375,
       "learning_rate": 0.00014804162541537955,
-      "loss": 2.1544,
+      "loss": 2.1543,
       "step": 2400
     },
     {
       "epoch": 0.81,
-      "grad_norm": 0.15234375,
+      "grad_norm": 0.1513671875,
       "learning_rate": 0.00014778211747273114,
       "loss": 2.1675,
       "step": 2405
     },
     {
       "epoch": 0.82,
-      "grad_norm": 0.146484375,
+      "grad_norm": 0.1455078125,
       "learning_rate": 0.00014752219186785784,
       "loss": 2.1267,
       "step": 2410
@@ -3393,7 +3393,7 @@
       "epoch": 0.82,
       "grad_norm": 0.154296875,
       "learning_rate": 0.0001472618508727626,
-      "loss": 2.1661,
+      "loss": 2.1662,
       "step": 2415
     },
     {
@@ -3428,7 +3428,7 @@
       "epoch": 0.83,
       "grad_norm": 0.1552734375,
       "learning_rate": 0.00014595399481710515,
-      "loss": 2.1296,
+      "loss": 2.1295,
       "step": 2440
     },
     {
@@ -3442,7 +3442,7 @@
       "epoch": 0.83,
       "grad_norm": 0.1552734375,
       "learning_rate": 0.00014542802458659152,
-      "loss": 2.1482,
+      "loss": 2.1483,
       "step": 2450
     },
     {
@@ -3454,16 +3454,16 @@
     },
     {
       "epoch": 0.83,
-      "grad_norm": 0.1552734375,
+      "grad_norm": 0.15625,
       "learning_rate": 0.00014490046601890405,
-      "loss": 2.1445,
+      "loss": 2.1444,
       "step": 2460
     },
     {
       "epoch": 0.83,
       "grad_norm": 0.1513671875,
       "learning_rate": 0.00014463609687067526,
-      "loss": 2.1528,
+      "loss": 2.1529,
       "step": 2465
     },
     {
@@ -3484,21 +3484,21 @@
       "epoch": 0.84,
       "grad_norm": 0.1552734375,
       "learning_rate": 0.00014384065770874373,
-      "loss": 2.1379,
+      "loss": 2.1381,
       "step": 2480
     },
     {
       "epoch": 0.84,
       "grad_norm": 0.15625,
       "learning_rate": 0.00014357474180781232,
-      "loss": 2.1293,
+      "loss": 2.1294,
       "step": 2485
     },
     {
       "epoch": 0.84,
       "grad_norm": 0.15625,
       "learning_rate": 0.00014330844502121547,
-      "loss": 2.1311,
+      "loss": 2.1312,
       "step": 2490
     },
     {
@@ -3524,7 +3524,7 @@
     },
     {
       "epoch": 0.85,
-      "grad_norm": 0.1552734375,
+      "grad_norm": 0.154296875,
       "learning_rate": 0.00014223949562154929,
       "loss": 2.1459,
       "step": 2510
@@ -3538,7 +3538,7 @@
     },
     {
       "epoch": 0.85,
-      "grad_norm": 0.1591796875,
+      "grad_norm": 0.158203125,
       "learning_rate": 0.0001417027962839634,
       "loss": 2.15,
       "step": 2520
@@ -3547,40 +3547,40 @@
       "epoch": 0.86,
       "grad_norm": 0.1513671875,
       "learning_rate": 0.00014143389865699132,
-      "loss": 2.1567,
+      "loss": 2.1569,
       "step": 2525
     },
     {
       "epoch": 0.86,
-      "grad_norm": 0.154296875,
+      "grad_norm": 0.1552734375,
       "learning_rate": 0.00014116463885740723,
-      "loss": 2.1289,
+      "loss": 2.129,
       "step": 2530
     },
     {
       "epoch": 0.86,
       "grad_norm": 0.154296875,
       "learning_rate": 0.00014089501923880384,
-      "loss": 2.1547,
+      "loss": 2.1546,
       "step": 2535
     },
     {
       "epoch": 0.86,
       "grad_norm": 0.1533203125,
       "learning_rate": 0.00014062504215791905,
-      "loss": 2.1306,
+      "loss": 2.1305,
       "step": 2540
     },
     {
       "epoch": 0.86,
       "grad_norm": 0.1552734375,
       "learning_rate": 0.00014035470997461548,
-      "loss": 2.1208,
+      "loss": 2.1207,
       "step": 2545
     },
     {
       "epoch": 0.86,
-      "grad_norm": 0.15625,
+      "grad_norm": 0.1572265625,
       "learning_rate": 0.00014008402505185952,
       "loss": 2.1112,
       "step": 2550
@@ -3589,28 +3589,28 @@
       "epoch": 0.87,
       "grad_norm": 0.154296875,
       "learning_rate": 0.000139812989755701,
-      "loss": 2.1231,
+      "loss": 2.1232,
       "step": 2555
     },
     {
       "epoch": 0.87,
       "grad_norm": 0.1552734375,
       "learning_rate": 0.00013954160645525217,
-      "loss": 2.1254,
+      "loss": 2.1253,
       "step": 2560
     },
     {
       "epoch": 0.87,
       "grad_norm": 0.1591796875,
       "learning_rate": 0.00013926987752266735,
-      "loss": 2.1362,
+      "loss": 2.1361,
       "step": 2565
     },
     {
       "epoch": 0.87,
       "grad_norm": 0.1552734375,
       "learning_rate": 0.0001389978053331219,
-      "loss": 2.1495,
+      "loss": 2.1494,
       "step": 2570
     },
     {
@@ -3624,14 +3624,14 @@
       "epoch": 0.87,
       "grad_norm": 0.16015625,
       "learning_rate": 0.00013845264069883216,
-      "loss": 2.1305,
+      "loss": 2.1307,
       "step": 2580
     },
     {
       "epoch": 0.88,
       "grad_norm": 0.15625,
       "learning_rate": 0.00013817955301935743,
-      "loss": 2.1316,
+      "loss": 2.1318,
       "step": 2585
     },
     {
@@ -3643,7 +3643,7 @@
     },
     {
       "epoch": 0.88,
-      "grad_norm": 0.162109375,
+      "grad_norm": 0.1611328125,
       "learning_rate": 0.00013763237887098843,
       "loss": 2.1299,
       "step": 2595
@@ -3666,7 +3666,7 @@
       "epoch": 0.88,
       "grad_norm": 0.15625,
       "learning_rate": 0.00013680915656774265,
-      "loss": 2.1452,
+      "loss": 2.1451,
       "step": 2610
     },
     {
@@ -3694,28 +3694,28 @@
       "epoch": 0.89,
       "grad_norm": 0.1552734375,
       "learning_rate": 0.00013570703361249188,
-      "loss": 2.154,
+      "loss": 2.1541,
       "step": 2630
     },
     {
       "epoch": 0.89,
       "grad_norm": 0.158203125,
       "learning_rate": 0.00013543071656015084,
-      "loss": 2.1221,
+      "loss": 2.122,
       "step": 2635
     },
     {
       "epoch": 0.89,
       "grad_norm": 0.158203125,
       "learning_rate": 0.0001351540898088536,
-      "loss": 2.1542,
+      "loss": 2.1543,
       "step": 2640
     },
     {
       "epoch": 0.9,
       "grad_norm": 0.1552734375,
       "learning_rate": 0.00013487715577658726,
-      "loss": 2.1586,
+      "loss": 2.1587,
       "step": 2645
     },
     {
@@ -3729,19 +3729,19 @@
       "epoch": 0.9,
       "grad_norm": 0.154296875,
       "learning_rate": 0.00013432237555450444,
-      "loss": 2.1498,
+      "loss": 2.1499,
       "step": 2655
     },
     {
       "epoch": 0.9,
       "grad_norm": 0.154296875,
       "learning_rate": 0.00013404453421400714,
-      "loss": 2.1519,
+      "loss": 2.152,
       "step": 2660
     },
     {
       "epoch": 0.9,
-      "grad_norm": 0.1572265625,
+      "grad_norm": 0.15625,
       "learning_rate": 0.00013376639529113688,
       "loss": 2.1233,
       "step": 2665
@@ -3771,28 +3771,28 @@
       "epoch": 0.91,
       "grad_norm": 0.1591796875,
       "learning_rate": 0.00013265091243859652,
-      "loss": 2.1236,
+      "loss": 2.1235,
       "step": 2685
     },
     {
       "epoch": 0.91,
       "grad_norm": 0.15625,
       "learning_rate": 0.00013237132212321487,
-      "loss": 2.1443,
+      "loss": 2.1442,
       "step": 2690
     },
     {
       "epoch": 0.91,
-      "grad_norm": 0.1591796875,
+      "grad_norm": 0.158203125,
       "learning_rate": 0.0001320914488509629,
-      "loss": 2.1559,
+      "loss": 2.1558,
       "step": 2695
     },
     {
       "epoch": 0.91,
       "grad_norm": 0.1630859375,
       "learning_rate": 0.00013181129506820545,
-      "loss": 2.1195,
+      "loss": 2.1194,
       "step": 2700
     },
     {
@@ -3813,14 +3813,14 @@
       "epoch": 0.92,
       "grad_norm": 0.15625,
       "learning_rate": 0.00013096917515719952,
-      "loss": 2.1442,
+      "loss": 2.1441,
       "step": 2715
     },
     {
       "epoch": 0.92,
       "grad_norm": 0.16015625,
       "learning_rate": 0.00013068792384478636,
-      "loss": 2.1488,
+      "loss": 2.1487,
       "step": 2720
     },
     {
@@ -3834,19 +3834,19 @@
       "epoch": 0.92,
       "grad_norm": 0.1513671875,
       "learning_rate": 0.00013012461895372344,
-      "loss": 2.1394,
+      "loss": 2.1395,
       "step": 2730
     },
     {
       "epoch": 0.93,
       "grad_norm": 0.15625,
       "learning_rate": 0.00012984257029890683,
-      "loss": 2.1418,
+      "loss": 2.1419,
       "step": 2735
     },
     {
       "epoch": 0.93,
-      "grad_norm": 0.15625,
+      "grad_norm": 0.1552734375,
       "learning_rate": 0.00012956026079097272,
       "loss": 2.1476,
       "step": 2740
@@ -3862,7 +3862,7 @@
       "epoch": 0.93,
       "grad_norm": 0.158203125,
       "learning_rate": 0.00012899486908865012,
-      "loss": 2.1379,
+      "loss": 2.138,
       "step": 2750
     },
     {
@@ -3874,7 +3874,7 @@
     },
     {
       "epoch": 0.93,
-      "grad_norm": 0.1572265625,
+      "grad_norm": 0.15625,
       "learning_rate": 0.00012842846361500816,
       "loss": 2.1108,
       "step": 2760
@@ -3890,14 +3890,14 @@
       "epoch": 0.94,
       "grad_norm": 0.1572265625,
       "learning_rate": 0.00012786106417374455,
-      "loss": 2.1762,
+      "loss": 2.1761,
       "step": 2770
     },
     {
       "epoch": 0.94,
       "grad_norm": 0.158203125,
       "learning_rate": 0.00012757699791343186,
-      "loss": 2.1445,
+      "loss": 2.1446,
       "step": 2775
     },
     {
@@ -3925,7 +3925,7 @@
       "epoch": 0.95,
       "grad_norm": 0.16015625,
       "learning_rate": 0.0001264383472357202,
-      "loss": 2.1394,
+      "loss": 2.1395,
       "step": 2795
     },
     {
@@ -3937,9 +3937,9 @@
     },
     {
       "epoch": 0.95,
-      "grad_norm": 0.1572265625,
+      "grad_norm": 0.158203125,
       "learning_rate": 0.00012586762535738374,
-      "loss": 2.1344,
+      "loss": 2.1345,
       "step": 2805
     },
     {
@@ -3953,7 +3953,7 @@
       "epoch": 0.95,
       "grad_norm": 0.158203125,
       "learning_rate": 0.00012529599904810784,
-      "loss": 2.1555,
+      "loss": 2.1554,
       "step": 2815
     },
     {
@@ -3967,12 +3967,12 @@
       "epoch": 0.96,
       "grad_norm": 0.1591796875,
       "learning_rate": 0.00012472348829413064,
-      "loss": 2.1479,
+      "loss": 2.1478,
       "step": 2825
     },
     {
       "epoch": 0.96,
-      "grad_norm": 0.16015625,
+      "grad_norm": 0.1591796875,
       "learning_rate": 0.00012443690750455326,
       "loss": 2.1159,
       "step": 2830
@@ -3988,35 +3988,35 @@
       "epoch": 0.96,
       "grad_norm": 0.1572265625,
       "learning_rate": 0.00012386310762517452,
-      "loss": 2.1339,
+      "loss": 2.134,
       "step": 2840
     },
     {
       "epoch": 0.96,
       "grad_norm": 0.158203125,
       "learning_rate": 0.00012357589355094275,
-      "loss": 2.109,
+      "loss": 2.1089,
       "step": 2845
     },
     {
       "epoch": 0.97,
       "grad_norm": 0.15625,
       "learning_rate": 0.0001232884734004491,
-      "loss": 2.1317,
+      "loss": 2.1316,
       "step": 2850
     },
     {
       "epoch": 0.97,
       "grad_norm": 0.154296875,
       "learning_rate": 0.00012300084968602549,
-      "loss": 2.1117,
+      "loss": 2.1116,
       "step": 2855
     },
     {
       "epoch": 0.97,
       "grad_norm": 0.154296875,
       "learning_rate": 0.00012271302492178327,
-      "loss": 2.1378,
+      "loss": 2.138,
       "step": 2860
     },
     {
@@ -4044,7 +4044,7 @@
       "epoch": 0.98,
       "grad_norm": 0.1611328125,
       "learning_rate": 0.00012155976570989949,
-      "loss": 2.1242,
+      "loss": 2.124,
       "step": 2880
     },
     {
@@ -4058,14 +4058,14 @@
       "epoch": 0.98,
       "grad_norm": 0.158203125,
       "learning_rate": 0.00012098199529902648,
-      "loss": 2.1341,
+      "loss": 2.134,
       "step": 2890
     },
     {
       "epoch": 0.98,
       "grad_norm": 0.154296875,
       "learning_rate": 0.0001206928337260159,
-      "loss": 2.1265,
+      "loss": 2.1264,
       "step": 2895
     },
     {
@@ -4077,16 +4077,16 @@
     },
     {
       "epoch": 0.98,
-      "grad_norm": 0.1552734375,
+      "grad_norm": 0.15625,
       "learning_rate": 0.00012011397048259285,
-      "loss": 2.1379,
+      "loss": 2.1381,
       "step": 2905
     },
     {
       "epoch": 0.99,
       "grad_norm": 0.154296875,
       "learning_rate": 0.00011982427387200867,
-      "loss": 2.1186,
+      "loss": 2.1187,
       "step": 2910
     },
     {
@@ -4100,7 +4100,7 @@
       "epoch": 0.99,
       "grad_norm": 0.1552734375,
       "learning_rate": 0.00011924436333422489,
-      "loss": 2.1316,
+      "loss": 2.1315,
       "step": 2920
     },
     {
@@ -4121,7 +4121,7 @@
       "epoch": 0.99,
       "grad_norm": 0.1552734375,
       "learning_rate": 0.00011837324226444169,
-      "loss": 2.1388,
+      "loss": 2.1389,
       "step": 2935
     },
     {
@@ -4135,7 +4135,7 @@
       "epoch": 1.0,
       "grad_norm": 0.1552734375,
       "learning_rate": 0.00011779168765418079,
-      "loss": 2.1314,
+      "loss": 2.1315,
       "step": 2945
     },
     {
@@ -4147,29 +4147,29 @@
     },
     {
       "epoch": 1.0,
-      "eval_loss": 2.13199520111084,
-      "eval_runtime": 156.8723,
-      "eval_samples_per_second": 8.465,
-      "eval_steps_per_second": 1.058,
+      "eval_loss": 2.133820056915283,
+      "eval_runtime": 154.3826,
+      "eval_samples_per_second": 8.453,
+      "eval_steps_per_second": 1.062,
       "step": 2952
     },
     {
       "epoch": 1.0,
       "grad_norm": 0.1572265625,
       "learning_rate": 0.00011720951097859476,
-      "loss": 2.1113,
+      "loss": 2.1114,
       "step": 2955
     },
     {
       "epoch": 1.0,
-      "grad_norm": 0.1533203125,
+      "grad_norm": 0.154296875,
       "learning_rate": 0.00011691819572644939,
-      "loss": 2.0923,
+      "loss": 2.0924,
       "step": 2960
     },
     {
       "epoch": 1.0,
-      "grad_norm": 0.1513671875,
+      "grad_norm": 0.15234375,
       "learning_rate": 0.00011662673259280276,
       "loss": 2.1001,
       "step": 2965
@@ -4178,14 +4178,14 @@
       "epoch": 1.01,
       "grad_norm": 0.1572265625,
       "learning_rate": 0.00011633512412532637,
-      "loss": 2.1235,
+      "loss": 2.1237,
       "step": 2970
     },
     {
       "epoch": 1.01,
       "grad_norm": 0.158203125,
       "learning_rate": 0.0001160433728729621,
-      "loss": 2.1127,
+      "loss": 2.1129,
       "step": 2975
     },
     {
@@ -4199,21 +4199,21 @@
       "epoch": 1.01,
       "grad_norm": 0.158203125,
       "learning_rate": 0.0001154594522155557,
-      "loss": 2.1346,
+      "loss": 2.1345,
       "step": 2985
     },
     {
       "epoch": 1.01,
       "grad_norm": 0.1630859375,
       "learning_rate": 0.00011516728791454861,
-      "loss": 2.1046,
+      "loss": 2.1045,
       "step": 2990
     },
     {
       "epoch": 1.01,
-      "grad_norm": 0.1611328125,
+      "grad_norm": 0.162109375,
       "learning_rate": 0.00011487499103667904,
-      "loss": 2.0974,
+      "loss": 2.0972,
       "step": 2995
     },
     {
@@ -4225,7 +4225,7 @@
     },
     {
       "epoch": 1.02,
-      "grad_norm": 0.1630859375,
+      "grad_norm": 0.162109375,
       "learning_rate": 0.00011429000977132629,
       "loss": 2.1148,
       "step": 3005
@@ -4234,7 +4234,7 @@
       "epoch": 1.02,
       "grad_norm": 0.162109375,
       "learning_rate": 0.00011399733049714884,
-      "loss": 2.1048,
+      "loss": 2.1049,
       "step": 3010
     },
     {
@@ -4248,28 +4248,28 @@
       "epoch": 1.02,
       "grad_norm": 0.1640625,
       "learning_rate": 0.00011341160745727844,
-      "loss": 2.1225,
+      "loss": 2.1226,
       "step": 3020
     },
     {
       "epoch": 1.02,
       "grad_norm": 0.16015625,
       "learning_rate": 0.00011311856881137506,
-      "loss": 2.1125,
+      "loss": 2.1123,
       "step": 3025
     },
     {
       "epoch": 1.03,
       "grad_norm": 0.162109375,
       "learning_rate": 0.00011282541549640873,
-      "loss": 2.0913,
+      "loss": 2.0914,
       "step": 3030
     },
     {
       "epoch": 1.03,
       "grad_norm": 0.1630859375,
       "learning_rate": 0.0001125321500748248,
-      "loss": 2.0832,
+      "loss": 2.0834,
       "step": 3035
     },
     {
@@ -4281,7 +4281,7 @@
     },
     {
       "epoch": 1.03,
-      "grad_norm": 0.1591796875,
+      "grad_norm": 0.16015625,
       "learning_rate": 0.00011194529316646293,
       "loss": 2.1353,
       "step": 3045
@@ -4290,42 +4290,42 @@
       "epoch": 1.03,
       "grad_norm": 0.1591796875,
       "learning_rate": 0.00011165170680938572,
-      "loss": 2.1065,
+      "loss": 2.1066,
       "step": 3050
     },
     {
       "epoch": 1.03,
       "grad_norm": 0.1572265625,
       "learning_rate": 0.00011135801860504749,
-      "loss": 2.1287,
+      "loss": 2.1288,
       "step": 3055
     },
     {
       "epoch": 1.04,
-      "grad_norm": 0.162109375,
+      "grad_norm": 0.1611328125,
       "learning_rate": 0.00011106423112056911,
-      "loss": 2.1505,
+      "loss": 2.1506,
       "step": 3060
     },
     {
       "epoch": 1.04,
       "grad_norm": 0.16015625,
       "learning_rate": 0.00011077034692393917,
-      "loss": 2.1052,
+      "loss": 2.1051,
       "step": 3065
     },
     {
       "epoch": 1.04,
       "grad_norm": 0.1591796875,
       "learning_rate": 0.00011047636858399169,
-      "loss": 2.1197,
+      "loss": 2.1196,
       "step": 3070
     },
     {
       "epoch": 1.04,
       "grad_norm": 0.1611328125,
       "learning_rate": 0.00011018229867038356,
-      "loss": 2.086,
+      "loss": 2.0859,
       "step": 3075
     },
     {
@@ -4346,7 +4346,7 @@
       "epoch": 1.05,
       "grad_norm": 0.1611328125,
       "learning_rate": 0.00010929956519603594,
-      "loss": 2.1101,
+      "loss": 2.1099,
       "step": 3090
     },
     {
@@ -4358,49 +4358,49 @@
     },
     {
       "epoch": 1.05,
-      "grad_norm": 0.1630859375,
+      "grad_norm": 0.1640625,
       "learning_rate": 0.00010871066549019688,
-      "loss": 2.1027,
+      "loss": 2.1026,
       "step": 3100
     },
     {
       "epoch": 1.05,
       "grad_norm": 0.162109375,
       "learning_rate": 0.0001084161001406712,
-      "loss": 2.1149,
+      "loss": 2.1147,
       "step": 3105
     },
     {
       "epoch": 1.05,
       "grad_norm": 0.16015625,
       "learning_rate": 0.00010812146122623683,
-      "loss": 2.0942,
+      "loss": 2.0943,
       "step": 3110
     },
     {
       "epoch": 1.06,
       "grad_norm": 0.162109375,
       "learning_rate": 0.00010782675132232474,
-      "loss": 2.1331,
+      "loss": 2.133,
       "step": 3115
     },
     {
       "epoch": 1.06,
-      "grad_norm": 0.1611328125,
+      "grad_norm": 0.16015625,
       "learning_rate": 0.00010753197300498638,
-      "loss": 2.1118,
+      "loss": 2.1119,
       "step": 3120
     },
     {
       "epoch": 1.06,
-      "grad_norm": 0.1650390625,
+      "grad_norm": 0.166015625,
       "learning_rate": 0.00010723712885087123,
       "loss": 2.0959,
       "step": 3125
     },
     {
       "epoch": 1.06,
-      "grad_norm": 0.162109375,
+      "grad_norm": 0.1611328125,
       "learning_rate": 0.00010694222143720423,
       "loss": 2.0654,
       "step": 3130
@@ -4414,42 +4414,42 @@
     },
     {
       "epoch": 1.06,
-      "grad_norm": 0.166015625,
+      "grad_norm": 0.1650390625,
       "learning_rate": 0.00010635222714285676,
-      "loss": 2.1053,
+      "loss": 2.1052,
       "step": 3140
     },
     {
       "epoch": 1.07,
-      "grad_norm": 0.1669921875,
+      "grad_norm": 0.16796875,
       "learning_rate": 0.0001060571454193008,
-      "loss": 2.1174,
+      "loss": 2.1175,
       "step": 3145
     },
     {
       "epoch": 1.07,
       "grad_norm": 0.1640625,
       "learning_rate": 0.00010576201075039696,
-      "loss": 2.0996,
+      "loss": 2.0998,
       "step": 3150
     },
     {
       "epoch": 1.07,
       "grad_norm": 0.1650390625,
       "learning_rate": 0.00010546682571590958,
-      "loss": 2.12,
+      "loss": 2.1202,
       "step": 3155
     },
     {
       "epoch": 1.07,
       "grad_norm": 0.1640625,
       "learning_rate": 0.00010517159289604324,
-      "loss": 2.1057,
+      "loss": 2.1058,
       "step": 3160
     },
     {
       "epoch": 1.07,
-      "grad_norm": 0.1630859375,
+      "grad_norm": 0.162109375,
       "learning_rate": 0.00010487631487142017,
       "loss": 2.135,
       "step": 3165
@@ -4465,28 +4465,28 @@
       "epoch": 1.08,
       "grad_norm": 0.1669921875,
       "learning_rate": 0.0001042856335323462,
-      "loss": 2.1096,
+      "loss": 2.1095,
       "step": 3175
     },
     {
       "epoch": 1.08,
       "grad_norm": 0.16015625,
       "learning_rate": 0.00010399023538102522,
-      "loss": 2.1008,
+      "loss": 2.1009,
       "step": 3180
     },
     {
       "epoch": 1.08,
       "grad_norm": 0.16015625,
       "learning_rate": 0.00010369480235116229,
-      "loss": 2.1124,
+      "loss": 2.1125,
       "step": 3185
     },
     {
       "epoch": 1.08,
       "grad_norm": 0.1650390625,
       "learning_rate": 0.00010339933702512979,
-      "loss": 2.1561,
+      "loss": 2.156,
       "step": 3190
     },
     {
@@ -4519,16 +4519,16 @@
     },
     {
       "epoch": 1.09,
-      "grad_norm": 0.1630859375,
+      "grad_norm": 0.162109375,
       "learning_rate": 0.00010192161635395026,
-      "loss": 2.0713,
+      "loss": 2.0715,
       "step": 3215
     },
     {
       "epoch": 1.09,
       "grad_norm": 0.1591796875,
       "learning_rate": 0.00010162601149494676,
-      "loss": 2.1249,
+      "loss": 2.1251,
       "step": 3220
     },
     {
@@ -4542,7 +4542,7 @@
       "epoch": 1.09,
       "grad_norm": 0.1611328125,
       "learning_rate": 0.00010103476172216792,
-      "loss": 2.1181,
+      "loss": 2.1182,
       "step": 3230
     },
     {
@@ -4556,14 +4556,14 @@
       "epoch": 1.1,
       "grad_norm": 0.162109375,
       "learning_rate": 0.0001004434757701688,
-      "loss": 2.1403,
+      "loss": 2.1404,
       "step": 3240
     },
     {
       "epoch": 1.1,
       "grad_norm": 0.1630859375,
       "learning_rate": 0.00010014782568743641,
-      "loss": 2.1352,
+      "loss": 2.1353,
       "step": 3245
     },
     {
@@ -4575,9 +4575,9 @@
     },
     {
       "epoch": 1.1,
-      "grad_norm": 0.1640625,
+      "grad_norm": 0.1630859375,
       "learning_rate": 9.955652422983122e-05,
-      "loss": 2.1037,
+      "loss": 2.1039,
       "step": 3255
     },
     {
@@ -4605,21 +4605,21 @@
       "epoch": 1.11,
       "grad_norm": 0.162109375,
       "learning_rate": 9.837398850505324e-05,
-      "loss": 2.1151,
+      "loss": 2.1152,
       "step": 3275
     },
     {
       "epoch": 1.11,
       "grad_norm": 0.162109375,
       "learning_rate": 9.807838364604978e-05,
-      "loss": 2.1192,
+      "loss": 2.119,
       "step": 3280
     },
     {
       "epoch": 1.11,
       "grad_norm": 0.1611328125,
       "learning_rate": 9.7782795583844e-05,
-      "loss": 2.1127,
+      "loss": 2.1126,
       "step": 3285
     },
     {
@@ -4633,7 +4633,7 @@
       "epoch": 1.12,
       "grad_norm": 0.162109375,
       "learning_rate": 9.719168018456598e-05,
-      "loss": 2.1043,
+      "loss": 2.1042,
       "step": 3295
     },
     {
@@ -4647,14 +4647,14 @@
       "epoch": 1.12,
       "grad_norm": 0.1640625,
       "learning_rate": 9.660066297487022e-05,
-      "loss": 2.1261,
+      "loss": 2.126,
       "step": 3305
     },
     {
       "epoch": 1.12,
       "grad_norm": 0.1611328125,
       "learning_rate": 9.630519764883772e-05,
-      "loss": 2.1013,
+      "loss": 2.1014,
       "step": 3310
     },
     {
@@ -4668,14 +4668,14 @@
       "epoch": 1.12,
       "grad_norm": 0.1650390625,
       "learning_rate": 9.571436646765382e-05,
-      "loss": 2.113,
+      "loss": 2.1129,
       "step": 3320
     },
     {
       "epoch": 1.13,
       "grad_norm": 0.162109375,
       "learning_rate": 9.541900577694217e-05,
-      "loss": 2.1211,
+      "loss": 2.1212,
       "step": 3325
     },
     {
@@ -4696,14 +4696,14 @@
       "epoch": 1.13,
       "grad_norm": 0.1650390625,
       "learning_rate": 9.453317428409044e-05,
-      "loss": 2.1096,
+      "loss": 2.1095,
       "step": 3340
     },
     {
       "epoch": 1.13,
       "grad_norm": 0.166015625,
       "learning_rate": 9.423798924960306e-05,
-      "loss": 2.1065,
+      "loss": 2.1064,
       "step": 3345
     },
     {
@@ -4715,7 +4715,7 @@
     },
     {
       "epoch": 1.14,
-      "grad_norm": 0.166015625,
+      "grad_norm": 0.1669921875,
       "learning_rate": 9.364777285714324e-05,
       "loss": 2.1006,
       "step": 3355
@@ -4724,28 +4724,28 @@
       "epoch": 1.14,
       "grad_norm": 0.1650390625,
       "learning_rate": 9.33527466582367e-05,
-      "loss": 2.1188,
+      "loss": 2.1189,
       "step": 3360
     },
     {
       "epoch": 1.14,
       "grad_norm": 0.166015625,
       "learning_rate": 9.30577785627958e-05,
-      "loss": 2.1265,
+      "loss": 2.1267,
       "step": 3365
     },
     {
       "epoch": 1.14,
       "grad_norm": 0.1630859375,
       "learning_rate": 9.276287114912878e-05,
-      "loss": 2.1252,
+      "loss": 2.1253,
       "step": 3370
     },
     {
       "epoch": 1.14,
       "grad_norm": 0.16796875,
       "learning_rate": 9.246802699501363e-05,
-      "loss": 2.1333,
+      "loss": 2.1334,
       "step": 3375
     },
     {
@@ -4766,14 +4766,14 @@
       "epoch": 1.15,
       "grad_norm": 0.1669921875,
       "learning_rate": 9.158389985932881e-05,
-      "loss": 2.1238,
+      "loss": 2.1237,
       "step": 3390
     },
     {
       "epoch": 1.15,
       "grad_norm": 0.1640625,
       "learning_rate": 9.128933450980314e-05,
-      "loss": 2.1229,
+      "loss": 2.1228,
       "step": 3395
     },
     {
@@ -4787,12 +4787,12 @@
       "epoch": 1.15,
       "grad_norm": 0.166015625,
       "learning_rate": 9.070043480396404e-05,
-      "loss": 2.0925,
+      "loss": 2.0926,
       "step": 3405
     },
     {
       "epoch": 1.15,
-      "grad_norm": 0.1630859375,
+      "grad_norm": 0.1640625,
       "learning_rate": 9.04061055952074e-05,
       "loss": 2.1346,
       "step": 3410
@@ -4801,12 +4801,12 @@
       "epoch": 1.16,
       "grad_norm": 0.162109375,
       "learning_rate": 9.011186024642793e-05,
-      "loss": 2.096,
+      "loss": 2.0959,
       "step": 3415
     },
     {
       "epoch": 1.16,
-      "grad_norm": 0.162109375,
+      "grad_norm": 0.1611328125,
       "learning_rate": 8.981770132961649e-05,
       "loss": 2.1153,
       "step": 3420
@@ -4843,14 +4843,14 @@
       "epoch": 1.17,
       "grad_norm": 0.166015625,
       "learning_rate": 8.834829319061431e-05,
-      "loss": 2.114,
+      "loss": 2.1139,
       "step": 3445
     },
     {
       "epoch": 1.17,
       "grad_norm": 0.16796875,
       "learning_rate": 8.805470683353708e-05,
-      "loss": 2.1197,
+      "loss": 2.1199,
       "step": 3450
     },
     {
@@ -4864,7 +4864,7 @@
       "epoch": 1.17,
       "grad_norm": 0.1630859375,
       "learning_rate": 8.746784992517518e-05,
-      "loss": 2.0911,
+      "loss": 2.0909,
       "step": 3460
     },
     {
@@ -4878,14 +4878,14 @@
       "epoch": 1.18,
       "grad_norm": 0.1640625,
       "learning_rate": 8.688143118862499e-05,
-      "loss": 2.1213,
+      "loss": 2.1214,
       "step": 3470
     },
     {
       "epoch": 1.18,
       "grad_norm": 0.16796875,
       "learning_rate": 8.658839254272157e-05,
-      "loss": 2.1115,
+      "loss": 2.1114,
       "step": 3475
     },
     {
@@ -4906,21 +4906,21 @@
       "epoch": 1.18,
       "grad_norm": 0.1689453125,
       "learning_rate": 8.570999022867373e-05,
-      "loss": 2.1145,
+      "loss": 2.1147,
       "step": 3490
     },
     {
       "epoch": 1.18,
       "grad_norm": 0.16796875,
       "learning_rate": 8.541743586309365e-05,
-      "loss": 2.0887,
+      "loss": 2.0888,
       "step": 3495
     },
     {
       "epoch": 1.19,
-      "grad_norm": 0.1650390625,
+      "grad_norm": 0.166015625,
       "learning_rate": 8.512500896332097e-05,
-      "loss": 2.1073,
+      "loss": 2.1074,
       "step": 3500
     },
     {
@@ -4934,14 +4934,14 @@
       "epoch": 1.19,
       "grad_norm": 0.1689453125,
       "learning_rate": 8.454054778444431e-05,
-      "loss": 2.1306,
+      "loss": 2.1305,
       "step": 3510
     },
     {
       "epoch": 1.19,
       "grad_norm": 0.166015625,
       "learning_rate": 8.424851861410007e-05,
-      "loss": 2.0804,
+      "loss": 2.0802,
       "step": 3515
     },
     {
@@ -4955,28 +4955,28 @@
       "epoch": 1.19,
       "grad_norm": 0.16796875,
       "learning_rate": 8.366487587467368e-05,
-      "loss": 2.1138,
+      "loss": 2.1136,
       "step": 3525
     },
     {
       "epoch": 1.2,
       "grad_norm": 0.16796875,
       "learning_rate": 8.337326740719726e-05,
-      "loss": 2.0983,
+      "loss": 2.0982,
       "step": 3530
     },
     {
       "epoch": 1.2,
       "grad_norm": 0.1650390625,
       "learning_rate": 8.308180427355062e-05,
-      "loss": 2.1253,
+      "loss": 2.1252,
       "step": 3535
     },
     {
       "epoch": 1.2,
       "grad_norm": 0.1640625,
       "learning_rate": 8.279048902140528e-05,
-      "loss": 2.1103,
+      "loss": 2.1102,
       "step": 3540
     },
     {
@@ -5004,7 +5004,7 @@
       "epoch": 1.21,
       "grad_norm": 0.16015625,
       "learning_rate": 8.162675773555836e-05,
-      "loss": 2.0978,
+      "loss": 2.0979,
       "step": 3560
     },
     {
@@ -5046,7 +5046,7 @@
       "epoch": 1.22,
       "grad_norm": 0.162109375,
       "learning_rate": 7.988602951740717e-05,
-      "loss": 2.1214,
+      "loss": 2.1215,
       "step": 3590
     },
     {
@@ -5060,7 +5060,7 @@
       "epoch": 1.22,
       "grad_norm": 0.1650390625,
       "learning_rate": 7.930716627398412e-05,
-      "loss": 2.1155,
+      "loss": 2.1156,
       "step": 3600
     },
     {
@@ -5081,7 +5081,7 @@
       "epoch": 1.22,
       "grad_norm": 0.169921875,
       "learning_rate": 7.84402342901005e-05,
-      "loss": 2.1381,
+      "loss": 2.1382,
       "step": 3615
     },
     {
@@ -5095,7 +5095,7 @@
       "epoch": 1.23,
       "grad_norm": 0.1630859375,
       "learning_rate": 7.786321769094717e-05,
-      "loss": 2.1158,
+      "loss": 2.1157,
       "step": 3625
     },
     {
@@ -5123,14 +5123,14 @@
       "epoch": 1.23,
       "grad_norm": 0.1689453125,
       "learning_rate": 7.671152659955096e-05,
-      "loss": 2.1199,
+      "loss": 2.12,
       "step": 3645
     },
     {
       "epoch": 1.24,
       "grad_norm": 0.1630859375,
       "learning_rate": 7.642410644905726e-05,
-      "loss": 2.1144,
+      "loss": 2.1143,
       "step": 3650
     },
     {
@@ -5151,7 +5151,7 @@
       "epoch": 1.24,
       "grad_norm": 0.1650390625,
       "learning_rate": 7.556309249544678e-05,
-      "loss": 2.1457,
+      "loss": 2.1458,
       "step": 3665
     },
     {
@@ -5165,28 +5165,28 @@
       "epoch": 1.24,
       "grad_norm": 0.1640625,
       "learning_rate": 7.499014702364913e-05,
-      "loss": 2.1119,
+      "loss": 2.112,
       "step": 3675
     },
     {
       "epoch": 1.25,
       "grad_norm": 0.166015625,
       "learning_rate": 7.470400095189219e-05,
-      "loss": 2.1395,
+      "loss": 2.1394,
       "step": 3680
     },
     {
       "epoch": 1.25,
       "grad_norm": 0.1640625,
       "learning_rate": 7.44180759917937e-05,
-      "loss": 2.0965,
+      "loss": 2.0966,
       "step": 3685
     },
     {
       "epoch": 1.25,
       "grad_norm": 0.169921875,
       "learning_rate": 7.413237464261627e-05,
-      "loss": 2.1036,
+      "loss": 2.1037,
       "step": 3690
     },
     {
@@ -5200,14 +5200,14 @@
       "epoch": 1.25,
       "grad_norm": 0.1669921875,
       "learning_rate": 7.356165276427983e-05,
-      "loss": 2.1214,
+      "loss": 2.1213,
       "step": 3700
     },
     {
       "epoch": 1.25,
       "grad_norm": 0.1611328125,
       "learning_rate": 7.327663722378561e-05,
-      "loss": 2.0925,
+      "loss": 2.0926,
       "step": 3705
     },
     {
@@ -5221,14 +5221,14 @@
       "epoch": 1.26,
       "grad_norm": 0.166015625,
       "learning_rate": 7.270730939669006e-05,
-      "loss": 2.0904,
+      "loss": 2.0903,
       "step": 3715
     },
     {
       "epoch": 1.26,
       "grad_norm": 0.166015625,
       "learning_rate": 7.242300208656814e-05,
-      "loss": 2.1251,
+      "loss": 2.1252,
       "step": 3720
     },
     {
@@ -5242,7 +5242,7 @@
       "epoch": 1.26,
       "grad_norm": 0.169921875,
       "learning_rate": 7.185511309876775e-05,
-      "loss": 2.0908,
+      "loss": 2.0909,
       "step": 3730
     },
     {
@@ -5256,7 +5256,7 @@
       "epoch": 1.27,
       "grad_norm": 0.1640625,
       "learning_rate": 7.128820816366442e-05,
-      "loss": 2.1062,
+      "loss": 2.1063,
       "step": 3740
     },
     {
@@ -5333,7 +5333,7 @@
       "epoch": 1.29,
       "grad_norm": 0.16796875,
       "learning_rate": 6.818870493179458e-05,
-      "loss": 2.0933,
+      "loss": 2.0934,
       "step": 3795
     },
     {
@@ -5354,7 +5354,7 @@
       "epoch": 1.29,
       "grad_norm": 0.1650390625,
       "learning_rate": 6.73490875614035e-05,
-      "loss": 2.1112,
+      "loss": 2.1111,
       "step": 3810
     },
     {
@@ -5368,28 +5368,28 @@
       "epoch": 1.29,
       "grad_norm": 0.1708984375,
       "learning_rate": 6.679076557432278e-05,
-      "loss": 2.122,
+      "loss": 2.1221,
       "step": 3820
     },
     {
       "epoch": 1.3,
       "grad_norm": 0.16796875,
       "learning_rate": 6.651203878290139e-05,
-      "loss": 2.124,
+      "loss": 2.1238,
       "step": 3825
     },
     {
       "epoch": 1.3,
       "grad_norm": 0.1650390625,
       "learning_rate": 6.623360470886314e-05,
-      "loss": 2.123,
+      "loss": 2.1229,
       "step": 3830
     },
     {
       "epoch": 1.3,
-      "grad_norm": 0.1669921875,
+      "grad_norm": 0.166015625,
       "learning_rate": 6.59554657859929e-05,
-      "loss": 2.0984,
+      "loss": 2.0983,
       "step": 3835
     },
     {
@@ -5403,7 +5403,7 @@
       "epoch": 1.3,
       "grad_norm": 0.1630859375,
       "learning_rate": 6.540008311597507e-05,
-      "loss": 2.1201,
+      "loss": 2.1202,
       "step": 3845
     },
     {
@@ -5417,7 +5417,7 @@
       "epoch": 1.31,
       "grad_norm": 0.166015625,
       "learning_rate": 6.484591019114646e-05,
-      "loss": 2.1121,
+      "loss": 2.1122,
       "step": 3855
     },
     {
@@ -5431,7 +5431,7 @@
       "epoch": 1.31,
       "grad_norm": 0.16796875,
       "learning_rate": 6.429296638750814e-05,
-      "loss": 2.1089,
+      "loss": 2.1091,
       "step": 3865
     },
     {
@@ -5445,21 +5445,21 @@
       "epoch": 1.31,
       "grad_norm": 0.1669921875,
       "learning_rate": 6.374127103808654e-05,
-      "loss": 2.1108,
+      "loss": 2.1106,
       "step": 3875
     },
     {
       "epoch": 1.31,
       "grad_norm": 0.1669921875,
       "learning_rate": 6.34658975633605e-05,
-      "loss": 2.1064,
+      "loss": 2.1063,
       "step": 3880
     },
     {
       "epoch": 1.32,
-      "grad_norm": 0.1689453125,
+      "grad_norm": 0.16796875,
       "learning_rate": 6.319084343225738e-05,
-      "loss": 2.131,
+      "loss": 2.1309,
       "step": 3885
     },
     {
@@ -5471,9 +5471,9 @@
     },
     {
       "epoch": 1.32,
-      "grad_norm": 0.166015625,
+      "grad_norm": 0.1650390625,
       "learning_rate": 6.264170281507111e-05,
-      "loss": 2.1196,
+      "loss": 2.1197,
       "step": 3895
     },
     {
@@ -5485,7 +5485,7 @@
     },
     {
       "epoch": 1.32,
-      "grad_norm": 0.1689453125,
+      "grad_norm": 0.169921875,
       "learning_rate": 6.209386838658024e-05,
       "loss": 2.1263,
       "step": 3905
@@ -5494,14 +5494,14 @@
       "epoch": 1.32,
       "grad_norm": 0.169921875,
       "learning_rate": 6.182044698064256e-05,
-      "loss": 2.1083,
+      "loss": 2.1084,
       "step": 3910
     },
     {
       "epoch": 1.33,
       "grad_norm": 0.166015625,
       "learning_rate": 6.154735930116786e-05,
-      "loss": 2.1206,
+      "loss": 2.1207,
       "step": 3915
     },
     {
@@ -5515,63 +5515,63 @@
       "epoch": 1.33,
       "grad_norm": 0.1650390625,
       "learning_rate": 6.1002194666878106e-05,
-      "loss": 2.1249,
+      "loss": 2.1247,
       "step": 3925
     },
     {
       "epoch": 1.33,
       "grad_norm": 0.1669921875,
       "learning_rate": 6.0730122477332675e-05,
-      "loss": 2.1023,
+      "loss": 2.1024,
       "step": 3930
     },
     {
       "epoch": 1.33,
       "grad_norm": 0.169921875,
       "learning_rate": 6.045839354474786e-05,
-      "loss": 2.0902,
+      "loss": 2.0901,
       "step": 3935
     },
     {
       "epoch": 1.33,
       "grad_norm": 0.1630859375,
       "learning_rate": 6.0187010244299046e-05,
-      "loss": 2.0995,
+      "loss": 2.0994,
       "step": 3940
     },
     {
       "epoch": 1.34,
       "grad_norm": 0.169921875,
       "learning_rate": 5.9915974948140474e-05,
-      "loss": 2.101,
+      "loss": 2.1009,
       "step": 3945
     },
     {
       "epoch": 1.34,
       "grad_norm": 0.1669921875,
       "learning_rate": 5.964529002538455e-05,
-      "loss": 2.0826,
+      "loss": 2.0828,
       "step": 3950
     },
     {
       "epoch": 1.34,
-      "grad_norm": 0.1669921875,
+      "grad_norm": 0.166015625,
       "learning_rate": 5.937495784208096e-05,
-      "loss": 2.1324,
+      "loss": 2.1322,
       "step": 3955
     },
     {
       "epoch": 1.34,
       "grad_norm": 0.171875,
       "learning_rate": 5.910498076119622e-05,
-      "loss": 2.1116,
+      "loss": 2.1117,
       "step": 3960
     },
     {
       "epoch": 1.34,
       "grad_norm": 0.1669921875,
       "learning_rate": 5.883536114259277e-05,
-      "loss": 2.0813,
+      "loss": 2.0814,
       "step": 3965
     },
     {
@@ -5585,28 +5585,28 @@
       "epoch": 1.35,
       "grad_norm": 0.171875,
       "learning_rate": 5.829720371603664e-05,
-      "loss": 2.1116,
+      "loss": 2.1117,
       "step": 3975
     },
     {
       "epoch": 1.35,
-      "grad_norm": 0.16796875,
+      "grad_norm": 0.1689453125,
       "learning_rate": 5.802867061210375e-05,
-      "loss": 2.1095,
+      "loss": 2.1094,
       "step": 3980
     },
     {
       "epoch": 1.35,
       "grad_norm": 0.169921875,
       "learning_rate": 5.776050437845075e-05,
-      "loss": 2.1008,
+      "loss": 2.1009,
       "step": 3985
     },
     {
       "epoch": 1.35,
       "grad_norm": 0.1708984375,
       "learning_rate": 5.749270735911158e-05,
-      "loss": 2.1075,
+      "loss": 2.1074,
       "step": 3990
     },
     {
@@ -5620,28 +5620,28 @@
       "epoch": 1.35,
       "grad_norm": 0.166015625,
       "learning_rate": 5.695823032335366e-05,
-      "loss": 2.1118,
+      "loss": 2.1119,
       "step": 4000
     },
     {
       "epoch": 1.36,
       "grad_norm": 0.1689453125,
       "learning_rate": 5.669155497878454e-05,
-      "loss": 2.1185,
+      "loss": 2.1184,
       "step": 4005
     },
     {
       "epoch": 1.36,
       "grad_norm": 0.1650390625,
       "learning_rate": 5.642525819218769e-05,
-      "loss": 2.0828,
+      "loss": 2.0827,
       "step": 4010
     },
     {
       "epoch": 1.36,
       "grad_norm": 0.166015625,
       "learning_rate": 5.6159342291256254e-05,
-      "loss": 2.0953,
+      "loss": 2.0952,
       "step": 4015
     },
     {
@@ -5655,14 +5655,14 @@
       "epoch": 1.36,
       "grad_norm": 0.1640625,
       "learning_rate": 5.562866244049557e-05,
-      "loss": 2.1283,
+      "loss": 2.1282,
       "step": 4025
     },
     {
       "epoch": 1.36,
       "grad_norm": 0.16796875,
       "learning_rate": 5.53639031293248e-05,
-      "loss": 2.1114,
+      "loss": 2.1115,
       "step": 4030
     },
     {
@@ -5676,14 +5676,14 @@
       "epoch": 1.37,
       "grad_norm": 0.166015625,
       "learning_rate": 5.483555730665282e-05,
-      "loss": 2.0859,
+      "loss": 2.0858,
       "step": 4040
     },
     {
       "epoch": 1.37,
       "grad_norm": 0.16796875,
       "learning_rate": 5.457197541340853e-05,
-      "loss": 2.1137,
+      "loss": 2.1138,
       "step": 4045
     },
     {
@@ -5704,7 +5704,7 @@
       "epoch": 1.38,
       "grad_norm": 0.166015625,
       "learning_rate": 5.3783621443117414e-05,
-      "loss": 2.1037,
+      "loss": 2.1038,
       "step": 4060
     },
     {
@@ -5716,7 +5716,7 @@
     },
     {
       "epoch": 1.38,
-      "grad_norm": 0.1689453125,
+      "grad_norm": 0.16796875,
       "learning_rate": 5.326006818194782e-05,
       "loss": 2.1037,
       "step": 4070
@@ -5725,21 +5725,21 @@
       "epoch": 1.38,
       "grad_norm": 0.1669921875,
       "learning_rate": 5.2998903236920895e-05,
-      "loss": 2.1188,
+      "loss": 2.1187,
       "step": 4075
     },
     {
       "epoch": 1.38,
       "grad_norm": 0.171875,
       "learning_rate": 5.273814912723742e-05,
-      "loss": 2.1134,
+      "loss": 2.1133,
       "step": 4080
     },
     {
       "epoch": 1.38,
       "grad_norm": 0.169921875,
       "learning_rate": 5.247780813214214e-05,
-      "loss": 2.126,
+      "loss": 2.1261,
       "step": 4085
     },
     {
@@ -5753,14 +5753,14 @@
       "epoch": 1.39,
       "grad_norm": 0.16796875,
       "learning_rate": 5.195837458462045e-05,
-      "loss": 2.1119,
+      "loss": 2.112,
       "step": 4095
     },
     {
       "epoch": 1.39,
       "grad_norm": 0.16796875,
       "learning_rate": 5.16992865725489e-05,
-      "loss": 2.1249,
+      "loss": 2.125,
       "step": 4100
     },
     {
@@ -5781,35 +5781,35 @@
       "epoch": 1.39,
       "grad_norm": 0.1748046875,
       "learning_rate": 5.092456474813841e-05,
-      "loss": 2.1067,
+      "loss": 2.1066,
       "step": 4115
     },
     {
       "epoch": 1.4,
       "grad_norm": 0.1640625,
       "learning_rate": 5.066717906818618e-05,
-      "loss": 2.1206,
+      "loss": 2.1207,
       "step": 4120
     },
     {
       "epoch": 1.4,
       "grad_norm": 0.1611328125,
       "learning_rate": 5.041022460511673e-05,
-      "loss": 2.1134,
+      "loss": 2.1132,
       "step": 4125
     },
     {
       "epoch": 1.4,
       "grad_norm": 0.169921875,
       "learning_rate": 5.015370360496219e-05,
-      "loss": 2.1385,
+      "loss": 2.1384,
       "step": 4130
     },
     {
       "epoch": 1.4,
       "grad_norm": 0.166015625,
       "learning_rate": 4.989761830996581e-05,
-      "loss": 2.1043,
+      "loss": 2.1042,
       "step": 4135
     },
     {
@@ -5828,37 +5828,37 @@
     },
     {
       "epoch": 1.41,
-      "grad_norm": 0.177734375,
+      "grad_norm": 0.1787109375,
       "learning_rate": 4.913199902111385e-05,
-      "loss": 2.1188,
+      "loss": 2.1187,
       "step": 4150
     },
     {
       "epoch": 1.41,
       "grad_norm": 0.166015625,
       "learning_rate": 4.8877678892719866e-05,
-      "loss": 2.0727,
+      "loss": 2.0729,
       "step": 4155
     },
     {
       "epoch": 1.41,
       "grad_norm": 0.171875,
       "learning_rate": 4.862380562318236e-05,
-      "loss": 2.1085,
+      "loss": 2.1086,
       "step": 4160
     },
     {
       "epoch": 1.41,
       "grad_norm": 0.16796875,
       "learning_rate": 4.837038143160082e-05,
-      "loss": 2.1181,
+      "loss": 2.118,
       "step": 4165
     },
     {
       "epoch": 1.41,
       "grad_norm": 0.166015625,
       "learning_rate": 4.811740853314939e-05,
-      "loss": 2.118,
+      "loss": 2.1179,
       "step": 4170
     },
     {
@@ -5879,12 +5879,12 @@
       "epoch": 1.42,
       "grad_norm": 0.16796875,
       "learning_rate": 4.736121968903027e-05,
-      "loss": 2.1318,
+      "loss": 2.1319,
       "step": 4185
     },
     {
       "epoch": 1.42,
-      "grad_norm": 0.16796875,
+      "grad_norm": 0.1669921875,
       "learning_rate": 4.7110074035656316e-05,
       "loss": 2.0898,
       "step": 4190
@@ -5900,7 +5900,7 @@
       "epoch": 1.42,
       "grad_norm": 0.1669921875,
       "learning_rate": 4.6609171848456066e-05,
-      "loss": 2.0966,
+      "loss": 2.0967,
       "step": 4200
     },
     {
@@ -5914,33 +5914,33 @@
       "epoch": 1.43,
       "grad_norm": 0.1728515625,
       "learning_rate": 4.611013640844245e-05,
-      "loss": 2.118,
+      "loss": 2.1181,
       "step": 4210
     },
     {
       "epoch": 1.43,
-      "grad_norm": 0.169921875,
+      "grad_norm": 0.1689453125,
       "learning_rate": 4.5861324173754484e-05,
-      "loss": 2.1136,
+      "loss": 2.1135,
       "step": 4215
     },
     {
       "epoch": 1.43,
-      "grad_norm": 0.16796875,
+      "grad_norm": 0.1689453125,
       "learning_rate": 4.561298516379974e-05,
-      "loss": 2.1026,
+      "loss": 2.1027,
       "step": 4220
     },
     {
       "epoch": 1.43,
       "grad_norm": 0.166015625,
       "learning_rate": 4.5365121549302914e-05,
-      "loss": 2.0938,
+      "loss": 2.094,
       "step": 4225
     },
     {
       "epoch": 1.43,
-      "grad_norm": 0.1689453125,
+      "grad_norm": 0.169921875,
       "learning_rate": 4.5117735496833415e-05,
       "loss": 2.1279,
       "step": 4230
@@ -5949,35 +5949,35 @@
       "epoch": 1.43,
       "grad_norm": 0.166015625,
       "learning_rate": 4.487082916878606e-05,
-      "loss": 2.1078,
+      "loss": 2.1079,
       "step": 4235
     },
     {
       "epoch": 1.44,
       "grad_norm": 0.1650390625,
       "learning_rate": 4.4624404723362576e-05,
-      "loss": 2.1142,
+      "loss": 2.1143,
       "step": 4240
     },
     {
       "epoch": 1.44,
-      "grad_norm": 0.1640625,
+      "grad_norm": 0.1650390625,
       "learning_rate": 4.437846431455249e-05,
-      "loss": 2.0946,
+      "loss": 2.0944,
       "step": 4245
     },
     {
       "epoch": 1.44,
       "grad_norm": 0.171875,
       "learning_rate": 4.4133010092114494e-05,
-      "loss": 2.1203,
+      "loss": 2.1205,
       "step": 4250
     },
     {
       "epoch": 1.44,
       "grad_norm": 0.166015625,
       "learning_rate": 4.3888044201557376e-05,
-      "loss": 2.1113,
+      "loss": 2.1112,
       "step": 4255
     },
     {
@@ -5989,23 +5989,23 @@
     },
     {
       "epoch": 1.44,
-      "grad_norm": 0.173828125,
+      "grad_norm": 0.1728515625,
       "learning_rate": 4.3399585976760105e-05,
-      "loss": 2.1055,
+      "loss": 2.1054,
       "step": 4265
     },
     {
       "epoch": 1.45,
       "grad_norm": 0.169921875,
       "learning_rate": 4.3156097912120385e-05,
-      "loss": 2.1009,
+      "loss": 2.1011,
       "step": 4270
     },
     {
       "epoch": 1.45,
-      "grad_norm": 0.1728515625,
+      "grad_norm": 0.173828125,
       "learning_rate": 4.29131067185251e-05,
-      "loss": 2.1301,
+      "loss": 2.1303,
       "step": 4275
     },
     {
@@ -6019,7 +6019,7 @@
       "epoch": 1.45,
       "grad_norm": 0.169921875,
       "learning_rate": 4.242862343602447e-05,
-      "loss": 2.0793,
+      "loss": 2.0792,
       "step": 4285
     },
     {
@@ -6033,7 +6033,7 @@
       "epoch": 1.45,
       "grad_norm": 0.1708984375,
       "learning_rate": 4.19461530686434e-05,
-      "loss": 2.1081,
+      "loss": 2.1082,
       "step": 4295
     },
     {
@@ -6047,7 +6047,7 @@
       "epoch": 1.46,
       "grad_norm": 0.169921875,
       "learning_rate": 4.1465712485387966e-05,
-      "loss": 2.1102,
+      "loss": 2.1103,
       "step": 4305
     },
     {
@@ -6061,33 +6061,33 @@
       "epoch": 1.46,
       "grad_norm": 0.1708984375,
       "learning_rate": 4.0987318484295135e-05,
-      "loss": 2.103,
+      "loss": 2.1029,
       "step": 4315
     },
     {
       "epoch": 1.46,
       "grad_norm": 0.1748046875,
       "learning_rate": 4.074889418189608e-05,
-      "loss": 2.1171,
+      "loss": 2.1172,
       "step": 4320
     },
     {
       "epoch": 1.46,
       "grad_norm": 0.1640625,
       "learning_rate": 4.051098779184559e-05,
-      "loss": 2.0826,
+      "loss": 2.0825,
       "step": 4325
     },
     {
       "epoch": 1.47,
-      "grad_norm": 0.1630859375,
+      "grad_norm": 0.1640625,
       "learning_rate": 4.0273601393677064e-05,
-      "loss": 2.1348,
+      "loss": 2.1349,
       "step": 4330
     },
     {
       "epoch": 1.47,
-      "grad_norm": 0.1708984375,
+      "grad_norm": 0.171875,
       "learning_rate": 4.0036737062378823e-05,
       "loss": 2.1057,
       "step": 4335
@@ -6096,33 +6096,33 @@
       "epoch": 1.47,
       "grad_norm": 0.16796875,
       "learning_rate": 3.980039686837568e-05,
-      "loss": 2.1364,
+      "loss": 2.1363,
       "step": 4340
     },
     {
       "epoch": 1.47,
       "grad_norm": 0.171875,
       "learning_rate": 3.956458287751097e-05,
-      "loss": 2.1176,
+      "loss": 2.1177,
       "step": 4345
     },
     {
       "epoch": 1.47,
       "grad_norm": 0.16796875,
       "learning_rate": 3.932929715102863e-05,
-      "loss": 2.0995,
+      "loss": 2.0994,
       "step": 4350
     },
     {
       "epoch": 1.48,
       "grad_norm": 0.1728515625,
       "learning_rate": 3.9094541745554946e-05,
-      "loss": 2.0963,
+      "loss": 2.0962,
       "step": 4355
     },
     {
       "epoch": 1.48,
-      "grad_norm": 0.1708984375,
+      "grad_norm": 0.171875,
       "learning_rate": 3.8860318713080725e-05,
       "loss": 2.1128,
       "step": 4360
@@ -6138,54 +6138,54 @@
       "epoch": 1.48,
       "grad_norm": 0.169921875,
       "learning_rate": 3.8393477951808444e-05,
-      "loss": 2.1023,
+      "loss": 2.1022,
       "step": 4370
     },
     {
       "epoch": 1.48,
       "grad_norm": 0.16796875,
       "learning_rate": 3.816086430365321e-05,
-      "loss": 2.1185,
+      "loss": 2.1184,
       "step": 4375
     },
     {
       "epoch": 1.48,
       "grad_norm": 0.1669921875,
       "learning_rate": 3.7928791189747195e-05,
-      "loss": 2.1078,
+      "loss": 2.1077,
       "step": 4380
     },
     {
       "epoch": 1.49,
       "grad_norm": 0.16796875,
       "learning_rate": 3.769726063863541e-05,
-      "loss": 2.1135,
+      "loss": 2.1133,
       "step": 4385
     },
     {
       "epoch": 1.49,
       "grad_norm": 0.1650390625,
       "learning_rate": 3.746627467412026e-05,
-      "loss": 2.1013,
+      "loss": 2.1012,
       "step": 4390
     },
     {
       "epoch": 1.49,
-      "grad_norm": 0.1669921875,
+      "grad_norm": 0.16796875,
       "learning_rate": 3.723583531524394e-05,
-      "loss": 2.115,
+      "loss": 2.1151,
       "step": 4395
     },
     {
       "epoch": 1.49,
-      "grad_norm": 0.1689453125,
+      "grad_norm": 0.16796875,
       "learning_rate": 3.700594457627079e-05,
       "loss": 2.1374,
       "step": 4400
     },
     {
       "epoch": 1.49,
-      "grad_norm": 0.1669921875,
+      "grad_norm": 0.16796875,
       "learning_rate": 3.6776604466669686e-05,
       "loss": 2.1082,
       "step": 4405
@@ -6194,21 +6194,21 @@
       "epoch": 1.49,
       "grad_norm": 0.169921875,
       "learning_rate": 3.654781699109645e-05,
-      "loss": 2.1172,
+      "loss": 2.1171,
       "step": 4410
     },
     {
       "epoch": 1.5,
       "grad_norm": 0.16796875,
       "learning_rate": 3.631958414937633e-05,
-      "loss": 2.1055,
+      "loss": 2.1057,
       "step": 4415
     },
     {
       "epoch": 1.5,
       "grad_norm": 0.166015625,
       "learning_rate": 3.609190793648661e-05,
-      "loss": 2.1365,
+      "loss": 2.1366,
       "step": 4420
     },
     {
@@ -6222,14 +6222,14 @@
       "epoch": 1.5,
       "grad_norm": 0.1689453125,
       "learning_rate": 3.563823335276244e-05,
-      "loss": 2.1005,
+      "loss": 2.1004,
       "step": 4430
     },
     {
       "epoch": 1.5,
       "grad_norm": 0.1689453125,
       "learning_rate": 3.541223894748553e-05,
-      "loss": 2.096,
+      "loss": 2.0959,
       "step": 4435
     },
     {
@@ -6257,28 +6257,28 @@
       "epoch": 1.51,
       "grad_norm": 0.166015625,
       "learning_rate": 3.45139266054715e-05,
-      "loss": 2.1069,
+      "loss": 2.1068,
       "step": 4455
     },
     {
       "epoch": 1.51,
       "grad_norm": 0.16796875,
       "learning_rate": 3.429077465490551e-05,
-      "loss": 2.1214,
+      "loss": 2.1213,
       "step": 4460
     },
     {
       "epoch": 1.51,
       "grad_norm": 0.1728515625,
       "learning_rate": 3.406819706694003e-05,
-      "loss": 2.0987,
+      "loss": 2.0988,
       "step": 4465
     },
     {
       "epoch": 1.51,
       "grad_norm": 0.1708984375,
       "learning_rate": 3.3846195787119814e-05,
-      "loss": 2.0997,
+      "loss": 2.0996,
       "step": 4470
     },
     {
@@ -6292,7 +6292,7 @@
       "epoch": 1.52,
       "grad_norm": 0.171875,
       "learning_rate": 3.340392990889018e-05,
-      "loss": 2.0976,
+      "loss": 2.0977,
       "step": 4480
     },
     {
@@ -6311,30 +6311,30 @@
     },
     {
       "epoch": 1.52,
-      "grad_norm": 0.1689453125,
+      "grad_norm": 0.16796875,
       "learning_rate": 3.2744901750693556e-05,
-      "loss": 2.0953,
+      "loss": 2.0952,
       "step": 4495
     },
     {
       "epoch": 1.52,
       "grad_norm": 0.169921875,
       "learning_rate": 3.2526398892901654e-05,
-      "loss": 2.0882,
+      "loss": 2.0881,
       "step": 4500
     },
     {
       "epoch": 1.53,
       "grad_norm": 0.169921875,
       "learning_rate": 3.2308485820072075e-05,
-      "loss": 2.1045,
+      "loss": 2.1046,
       "step": 4505
     },
     {
       "epoch": 1.53,
       "grad_norm": 0.1669921875,
       "learning_rate": 3.2091164436977294e-05,
-      "loss": 2.109,
+      "loss": 2.1092,
       "step": 4510
     },
     {
@@ -6355,68 +6355,68 @@
       "epoch": 1.53,
       "grad_norm": 0.162109375,
       "learning_rate": 3.144276939614683e-05,
-      "loss": 2.1295,
+      "loss": 2.1297,
       "step": 4525
     },
     {
       "epoch": 1.53,
       "grad_norm": 0.16796875,
       "learning_rate": 3.122783371602747e-05,
-      "loss": 2.1322,
+      "loss": 2.1324,
       "step": 4530
     },
     {
       "epoch": 1.54,
       "grad_norm": 0.166015625,
       "learning_rate": 3.101349917159433e-05,
-      "loss": 2.1366,
+      "loss": 2.1365,
       "step": 4535
     },
     {
       "epoch": 1.54,
-      "grad_norm": 0.1669921875,
+      "grad_norm": 0.16796875,
       "learning_rate": 3.079976763633996e-05,
-      "loss": 2.1136,
+      "loss": 2.1137,
       "step": 4540
     },
     {
       "epoch": 1.54,
       "grad_norm": 0.16796875,
       "learning_rate": 3.058664097848612e-05,
-      "loss": 2.1107,
+      "loss": 2.1106,
       "step": 4545
     },
     {
       "epoch": 1.54,
       "grad_norm": 0.171875,
       "learning_rate": 3.0374121060967255e-05,
-      "loss": 2.1025,
+      "loss": 2.1024,
       "step": 4550
     },
     {
       "epoch": 1.54,
       "grad_norm": 0.1669921875,
       "learning_rate": 3.0162209741414304e-05,
-      "loss": 2.1045,
+      "loss": 2.1043,
       "step": 4555
     },
     {
       "epoch": 1.54,
       "grad_norm": 0.1708984375,
       "learning_rate": 2.9950908872138584e-05,
-      "loss": 2.0756,
+      "loss": 2.0755,
       "step": 4560
     },
     {
       "epoch": 1.55,
       "grad_norm": 0.1669921875,
       "learning_rate": 2.9740220300115386e-05,
-      "loss": 2.119,
+      "loss": 2.1192,
       "step": 4565
     },
     {
       "epoch": 1.55,
-      "grad_norm": 0.171875,
+      "grad_norm": 0.1708984375,
       "learning_rate": 2.9530145866967895e-05,
       "loss": 2.1415,
       "step": 4570
@@ -6430,44 +6430,44 @@
     },
     {
       "epoch": 1.55,
-      "grad_norm": 0.1669921875,
+      "grad_norm": 0.166015625,
       "learning_rate": 2.9111846756936113e-05,
-      "loss": 2.1181,
+      "loss": 2.1182,
       "step": 4580
     },
     {
       "epoch": 1.55,
       "grad_norm": 0.1669921875,
       "learning_rate": 2.8903625736393304e-05,
-      "loss": 2.088,
+      "loss": 2.0882,
       "step": 4585
     },
     {
       "epoch": 1.55,
       "grad_norm": 0.1689453125,
       "learning_rate": 2.8696026167377155e-05,
-      "loss": 2.1498,
+      "loss": 2.1499,
       "step": 4590
     },
     {
       "epoch": 1.56,
-      "grad_norm": 0.1748046875,
+      "grad_norm": 0.17578125,
       "learning_rate": 2.8489049864510054e-05,
-      "loss": 2.1044,
+      "loss": 2.1045,
       "step": 4595
     },
     {
       "epoch": 1.56,
       "grad_norm": 0.1708984375,
       "learning_rate": 2.8282698636966375e-05,
-      "loss": 2.139,
+      "loss": 2.1391,
       "step": 4600
     },
     {
       "epoch": 1.56,
       "grad_norm": 0.169921875,
       "learning_rate": 2.8076974288456726e-05,
-      "loss": 2.0717,
+      "loss": 2.0716,
       "step": 4605
     },
     {
@@ -6481,21 +6481,21 @@
       "epoch": 1.56,
       "grad_norm": 0.169921875,
       "learning_rate": 2.7667413415968635e-05,
-      "loss": 2.0918,
+      "loss": 2.0917,
       "step": 4615
     },
     {
       "epoch": 1.56,
       "grad_norm": 0.1689453125,
       "learning_rate": 2.7463580471950943e-05,
-      "loss": 2.1356,
+      "loss": 2.1355,
       "step": 4620
     },
     {
       "epoch": 1.57,
       "grad_norm": 0.1650390625,
       "learning_rate": 2.7260381566857473e-05,
-      "loss": 2.0996,
+      "loss": 2.0997,
       "step": 4625
     },
     {
@@ -6509,7 +6509,7 @@
       "epoch": 1.57,
       "grad_norm": 0.16796875,
       "learning_rate": 2.68558929725107e-05,
-      "loss": 2.0948,
+      "loss": 2.095,
       "step": 4635
     },
     {
@@ -6523,7 +6523,7 @@
       "epoch": 1.57,
       "grad_norm": 0.1640625,
       "learning_rate": 2.645396177539379e-05,
-      "loss": 2.1148,
+      "loss": 2.1147,
       "step": 4645
     },
     {
@@ -6537,14 +6537,14 @@
       "epoch": 1.58,
       "grad_norm": 0.16796875,
       "learning_rate": 2.6054602028555885e-05,
-      "loss": 2.113,
+      "loss": 2.1129,
       "step": 4655
     },
     {
       "epoch": 1.58,
       "grad_norm": 0.16796875,
       "learning_rate": 2.585589081599862e-05,
-      "loss": 2.1115,
+      "loss": 2.1113,
       "step": 4660
     },
     {
@@ -6558,19 +6558,19 @@
       "epoch": 1.58,
       "grad_norm": 0.177734375,
       "learning_rate": 2.546041439723963e-05,
-      "loss": 2.1269,
+      "loss": 2.1268,
       "step": 4670
     },
     {
       "epoch": 1.58,
-      "grad_norm": 0.1669921875,
+      "grad_norm": 0.16796875,
       "learning_rate": 2.5263652647886803e-05,
-      "loss": 2.1149,
+      "loss": 2.115,
       "step": 4675
     },
     {
       "epoch": 1.59,
-      "grad_norm": 0.16796875,
+      "grad_norm": 0.1669921875,
       "learning_rate": 2.5067544166969114e-05,
       "loss": 2.1177,
       "step": 4680
@@ -6579,7 +6579,7 @@
       "epoch": 1.59,
       "grad_norm": 0.169921875,
       "learning_rate": 2.487209066866565e-05,
-      "loss": 2.0814,
+      "loss": 2.0815,
       "step": 4685
     },
     {
@@ -6600,63 +6600,63 @@
       "epoch": 1.59,
       "grad_norm": 0.1708984375,
       "learning_rate": 2.42896771252636e-05,
-      "loss": 2.0944,
+      "loss": 2.0945,
       "step": 4700
     },
     {
       "epoch": 1.59,
       "grad_norm": 0.1669921875,
       "learning_rate": 2.4096860584479974e-05,
-      "loss": 2.1351,
+      "loss": 2.1349,
       "step": 4705
     },
     {
       "epoch": 1.6,
       "grad_norm": 0.16796875,
       "learning_rate": 2.390470751103008e-05,
-      "loss": 2.09,
+      "loss": 2.0902,
       "step": 4710
     },
     {
       "epoch": 1.6,
-      "grad_norm": 0.16796875,
+      "grad_norm": 0.1669921875,
       "learning_rate": 2.37132195845188e-05,
-      "loss": 2.1335,
+      "loss": 2.1333,
       "step": 4715
     },
     {
       "epoch": 1.6,
       "grad_norm": 0.16796875,
       "learning_rate": 2.3522398478737108e-05,
-      "loss": 2.097,
+      "loss": 2.0972,
       "step": 4720
     },
     {
       "epoch": 1.6,
       "grad_norm": 0.1689453125,
       "learning_rate": 2.3332245861647206e-05,
-      "loss": 2.0869,
+      "loss": 2.087,
       "step": 4725
     },
     {
       "epoch": 1.6,
       "grad_norm": 0.1728515625,
       "learning_rate": 2.3142763395368095e-05,
-      "loss": 2.0909,
+      "loss": 2.0908,
       "step": 4730
     },
     {
       "epoch": 1.6,
       "grad_norm": 0.1669921875,
       "learning_rate": 2.295395273616099e-05,
-      "loss": 2.1025,
+      "loss": 2.1023,
       "step": 4735
     },
     {
       "epoch": 1.61,
-      "grad_norm": 0.1708984375,
+      "grad_norm": 0.171875,
       "learning_rate": 2.2765815534414868e-05,
-      "loss": 2.0932,
+      "loss": 2.0931,
       "step": 4740
     },
     {
@@ -6670,7 +6670,7 @@
       "epoch": 1.61,
       "grad_norm": 0.1728515625,
       "learning_rate": 2.239156807541375e-05,
-      "loss": 2.1045,
+      "loss": 2.1044,
       "step": 4750
     },
     {
@@ -6684,7 +6684,7 @@
       "epoch": 1.61,
       "grad_norm": 0.1806640625,
       "learning_rate": 2.202003410348473e-05,
-      "loss": 2.0936,
+      "loss": 2.0934,
       "step": 4760
     },
     {
@@ -6712,21 +6712,21 @@
       "epoch": 1.62,
       "grad_norm": 0.1748046875,
       "learning_rate": 2.128515848649929e-05,
-      "loss": 2.1241,
+      "loss": 2.1242,
       "step": 4780
     },
     {
       "epoch": 1.62,
       "grad_norm": 0.16796875,
       "learning_rate": 2.1103155693385136e-05,
-      "loss": 2.1371,
+      "loss": 2.137,
       "step": 4785
     },
     {
       "epoch": 1.62,
       "grad_norm": 0.1748046875,
       "learning_rate": 2.092184253549998e-05,
-      "loss": 2.1204,
+      "loss": 2.1203,
       "step": 4790
     },
     {
@@ -6740,14 +6740,14 @@
       "epoch": 1.63,
       "grad_norm": 0.16796875,
       "learning_rate": 2.0561291458788733e-05,
-      "loss": 2.1013,
+      "loss": 2.1012,
       "step": 4800
     },
     {
       "epoch": 1.63,
       "grad_norm": 0.171875,
       "learning_rate": 2.0382056691530084e-05,
-      "loss": 2.1382,
+      "loss": 2.138,
       "step": 4805
     },
     {
@@ -6761,21 +6761,21 @@
       "epoch": 1.63,
       "grad_norm": 0.166015625,
       "learning_rate": 2.002567653262479e-05,
-      "loss": 2.0965,
+      "loss": 2.0964,
       "step": 4815
     },
     {
       "epoch": 1.63,
       "grad_norm": 0.1640625,
       "learning_rate": 1.984853425608769e-05,
-      "loss": 2.1354,
+      "loss": 2.1355,
       "step": 4820
     },
     {
       "epoch": 1.63,
       "grad_norm": 0.169921875,
       "learning_rate": 1.9672092581392375e-05,
-      "loss": 2.1271,
+      "loss": 2.1272,
       "step": 4825
     },
     {
@@ -6789,40 +6789,40 @@
       "epoch": 1.64,
       "grad_norm": 0.1630859375,
       "learning_rate": 1.9321317200477653e-05,
-      "loss": 2.1449,
+      "loss": 2.1447,
       "step": 4835
     },
     {
       "epoch": 1.64,
       "grad_norm": 0.16796875,
       "learning_rate": 1.91469865603766e-05,
-      "loss": 2.0944,
+      "loss": 2.0945,
       "step": 4840
     },
     {
       "epoch": 1.64,
       "grad_norm": 0.16796875,
       "learning_rate": 1.8973362654327175e-05,
-      "loss": 2.1171,
+      "loss": 2.117,
       "step": 4845
     },
     {
       "epoch": 1.64,
       "grad_norm": 0.166015625,
       "learning_rate": 1.8800446999971346e-05,
-      "loss": 2.1076,
+      "loss": 2.1077,
       "step": 4850
     },
     {
       "epoch": 1.64,
       "grad_norm": 0.1767578125,
       "learning_rate": 1.8628241108760268e-05,
-      "loss": 2.0964,
+      "loss": 2.0966,
       "step": 4855
     },
     {
       "epoch": 1.65,
-      "grad_norm": 0.169921875,
+      "grad_norm": 0.1708984375,
       "learning_rate": 1.845674648594108e-05,
       "loss": 2.1312,
       "step": 4860
@@ -6831,7 +6831,7 @@
       "epoch": 1.65,
       "grad_norm": 0.1669921875,
       "learning_rate": 1.828596463054375e-05,
-      "loss": 2.0881,
+      "loss": 2.0882,
       "step": 4865
     },
     {
@@ -6859,7 +6859,7 @@
       "epoch": 1.65,
       "grad_norm": 0.169921875,
       "learning_rate": 1.760999464543869e-05,
-      "loss": 2.0972,
+      "loss": 2.0973,
       "step": 4885
     },
     {
@@ -6873,19 +6873,19 @@
       "epoch": 1.66,
       "grad_norm": 0.1689453125,
       "learning_rate": 1.7276324773041565e-05,
-      "loss": 2.1326,
+      "loss": 2.1325,
       "step": 4895
     },
     {
       "epoch": 1.66,
       "grad_norm": 0.171875,
       "learning_rate": 1.7110573737455295e-05,
-      "loss": 2.1088,
+      "loss": 2.1089,
       "step": 4900
     },
     {
       "epoch": 1.66,
-      "grad_norm": 0.166015625,
+      "grad_norm": 0.1669921875,
       "learning_rate": 1.694554723615126e-05,
       "loss": 2.1011,
       "step": 4905
@@ -6894,14 +6894,14 @@
       "epoch": 1.66,
       "grad_norm": 0.16796875,
       "learning_rate": 1.6781246711621744e-05,
-      "loss": 2.1268,
+      "loss": 2.1267,
       "step": 4910
     },
     {
       "epoch": 1.66,
       "grad_norm": 0.1650390625,
       "learning_rate": 1.6617673600013296e-05,
-      "loss": 2.1262,
+      "loss": 2.1263,
       "step": 4915
     },
     {
@@ -6915,35 +6915,35 @@
       "epoch": 1.67,
       "grad_norm": 0.16796875,
       "learning_rate": 1.6292715328341712e-05,
-      "loss": 2.0922,
+      "loss": 2.0923,
       "step": 4925
     },
     {
       "epoch": 1.67,
       "grad_norm": 0.1728515625,
       "learning_rate": 1.6131333008730277e-05,
-      "loss": 2.1354,
+      "loss": 2.1355,
       "step": 4930
     },
     {
       "epoch": 1.67,
       "grad_norm": 0.1728515625,
       "learning_rate": 1.5970683782918374e-05,
-      "loss": 2.1079,
+      "loss": 2.108,
       "step": 4935
     },
     {
       "epoch": 1.67,
       "grad_norm": 0.16796875,
       "learning_rate": 1.5810769055136644e-05,
-      "loss": 2.1116,
+      "loss": 2.1115,
       "step": 4940
     },
     {
       "epoch": 1.67,
       "grad_norm": 0.171875,
       "learning_rate": 1.56515902231955e-05,
-      "loss": 2.1022,
+      "loss": 2.1023,
       "step": 4945
     },
     {
@@ -6992,14 +6992,14 @@
       "epoch": 1.69,
       "grad_norm": 0.16796875,
       "learning_rate": 1.4558059545351143e-05,
-      "loss": 2.1329,
+      "loss": 2.1328,
       "step": 4980
     },
     {
       "epoch": 1.69,
       "grad_norm": 0.16796875,
       "learning_rate": 1.4404817426875938e-05,
-      "loss": 2.0845,
+      "loss": 2.0847,
       "step": 4985
     },
     {
@@ -7013,28 +7013,28 @@
       "epoch": 1.69,
       "grad_norm": 0.169921875,
       "learning_rate": 1.4100579078550613e-05,
-      "loss": 2.0901,
+      "loss": 2.09,
       "step": 4995
     },
     {
       "epoch": 1.69,
       "grad_norm": 0.1748046875,
       "learning_rate": 1.3949585508039886e-05,
-      "loss": 2.123,
+      "loss": 2.1229,
       "step": 5000
     },
     {
       "epoch": 1.7,
       "grad_norm": 0.1669921875,
       "learning_rate": 1.3799344101925904e-05,
-      "loss": 2.1344,
+      "loss": 2.1345,
       "step": 5005
     },
     {
       "epoch": 1.7,
       "grad_norm": 0.177734375,
       "learning_rate": 1.3649856173464781e-05,
-      "loss": 2.1068,
+      "loss": 2.1066,
       "step": 5010
     },
     {
@@ -7055,7 +7055,7 @@
       "epoch": 1.7,
       "grad_norm": 0.1669921875,
       "learning_rate": 1.3205926287699988e-05,
-      "loss": 2.114,
+      "loss": 2.1139,
       "step": 5025
     },
     {
@@ -7074,14 +7074,14 @@
     },
     {
       "epoch": 1.71,
-      "grad_norm": 0.1708984375,
+      "grad_norm": 0.169921875,
       "learning_rate": 1.276882434446719e-05,
-      "loss": 2.1104,
+      "loss": 2.1105,
       "step": 5040
     },
     {
       "epoch": 1.71,
-      "grad_norm": 0.1689453125,
+      "grad_norm": 0.169921875,
       "learning_rate": 1.2624646976082066e-05,
       "loss": 2.1108,
       "step": 5045
@@ -7097,14 +7097,14 @@
       "epoch": 1.71,
       "grad_norm": 0.1689453125,
       "learning_rate": 1.2338584729834701e-05,
-      "loss": 2.0853,
+      "loss": 2.0852,
       "step": 5055
     },
     {
       "epoch": 1.71,
       "grad_norm": 0.16796875,
       "learning_rate": 1.2196702352434953e-05,
-      "loss": 2.1188,
+      "loss": 2.1189,
       "step": 5060
     },
     {
@@ -7118,7 +7118,7 @@
       "epoch": 1.72,
       "grad_norm": 0.16796875,
       "learning_rate": 1.1915241290023115e-05,
-      "loss": 2.0969,
+      "loss": 2.097,
       "step": 5070
     },
     {
@@ -7146,7 +7146,7 @@
       "epoch": 1.72,
       "grad_norm": 0.1689453125,
       "learning_rate": 1.1361568236830323e-05,
-      "loss": 2.129,
+      "loss": 2.1292,
       "step": 5090
     },
     {
@@ -7174,7 +7174,7 @@
       "epoch": 1.73,
       "grad_norm": 0.169921875,
       "learning_rate": 1.0820291627186107e-05,
-      "loss": 2.1059,
+      "loss": 2.106,
       "step": 5110
     },
     {
@@ -7195,7 +7195,7 @@
       "epoch": 1.74,
       "grad_norm": 0.173828125,
       "learning_rate": 1.0422514939630424e-05,
-      "loss": 2.1088,
+      "loss": 2.1087,
       "step": 5125
     },
     {
@@ -7209,7 +7209,7 @@
       "epoch": 1.74,
       "grad_norm": 0.1669921875,
       "learning_rate": 1.0161243521738661e-05,
-      "loss": 2.1255,
+      "loss": 2.1257,
       "step": 5135
     },
     {
@@ -7242,7 +7242,7 @@
     },
     {
       "epoch": 1.75,
-      "grad_norm": 0.16796875,
+      "grad_norm": 0.1669921875,
       "learning_rate": 9.521827027934038e-06,
       "loss": 2.1189,
       "step": 5160
@@ -7251,7 +7251,7 @@
       "epoch": 1.75,
       "grad_norm": 0.1689453125,
       "learning_rate": 9.396311894274645e-06,
-      "loss": 2.1266,
+      "loss": 2.1268,
       "step": 5165
     },
     {
@@ -7272,7 +7272,7 @@
       "epoch": 1.75,
       "grad_norm": 0.1689453125,
       "learning_rate": 9.024522633817756e-06,
-      "loss": 2.1242,
+      "loss": 2.124,
       "step": 5180
     },
     {
@@ -7286,19 +7286,19 @@
       "epoch": 1.76,
       "grad_norm": 0.1689453125,
       "learning_rate": 8.780637392592495e-06,
-      "loss": 2.1045,
+      "loss": 2.1046,
       "step": 5190
     },
     {
       "epoch": 1.76,
       "grad_norm": 0.16796875,
       "learning_rate": 8.659890259824532e-06,
-      "loss": 2.1174,
+      "loss": 2.1173,
       "step": 5195
     },
     {
       "epoch": 1.76,
-      "grad_norm": 0.1640625,
+      "grad_norm": 0.1630859375,
       "learning_rate": 8.53994152853952e-06,
       "loss": 2.1193,
       "step": 5200
@@ -7307,14 +7307,14 @@
       "epoch": 1.76,
       "grad_norm": 0.1689453125,
       "learning_rate": 8.420792247206177e-06,
-      "loss": 2.1263,
+      "loss": 2.1264,
       "step": 5205
     },
     {
       "epoch": 1.76,
       "grad_norm": 0.16796875,
       "learning_rate": 8.302443457305209e-06,
-      "loss": 2.0897,
+      "loss": 2.0896,
       "step": 5210
     },
     {
@@ -7328,28 +7328,28 @@
       "epoch": 1.77,
       "grad_norm": 0.1689453125,
       "learning_rate": 8.068151482728802e-06,
-      "loss": 2.0994,
+      "loss": 2.0992,
       "step": 5220
     },
     {
       "epoch": 1.77,
       "grad_norm": 0.16796875,
       "learning_rate": 7.952210345993339e-06,
-      "loss": 2.1262,
+      "loss": 2.1263,
       "step": 5225
     },
     {
       "epoch": 1.77,
       "grad_norm": 0.16796875,
       "learning_rate": 7.837073796552241e-06,
-      "loss": 2.096,
+      "loss": 2.0959,
       "step": 5230
     },
     {
       "epoch": 1.77,
       "grad_norm": 0.169921875,
       "learning_rate": 7.72274284081106e-06,
-      "loss": 2.107,
+      "loss": 2.1071,
       "step": 5235
     },
     {
@@ -7363,12 +7363,12 @@
       "epoch": 1.78,
       "grad_norm": 0.1669921875,
       "learning_rate": 7.4965017008334245e-06,
-      "loss": 2.0934,
+      "loss": 2.0935,
       "step": 5245
     },
     {
       "epoch": 1.78,
-      "grad_norm": 0.1708984375,
+      "grad_norm": 0.169921875,
       "learning_rate": 7.3845934941648046e-06,
       "loss": 2.1126,
       "step": 5250
@@ -7377,21 +7377,21 @@
       "epoch": 1.78,
       "grad_norm": 0.16796875,
       "learning_rate": 7.2734948363144206e-06,
-      "loss": 2.0974,
+      "loss": 2.0975,
       "step": 5255
     },
     {
       "epoch": 1.78,
       "grad_norm": 0.16796875,
       "learning_rate": 7.163206698392744e-06,
-      "loss": 2.1165,
+      "loss": 2.1164,
       "step": 5260
     },
     {
       "epoch": 1.78,
       "grad_norm": 0.1650390625,
       "learning_rate": 7.0537300444254435e-06,
-      "loss": 2.1232,
+      "loss": 2.1231,
       "step": 5265
     },
     {
@@ -7412,28 +7412,28 @@
       "epoch": 1.79,
       "grad_norm": 0.169921875,
       "learning_rate": 6.7301785200593046e-06,
-      "loss": 2.1474,
+      "loss": 2.1473,
       "step": 5280
     },
     {
       "epoch": 1.79,
       "grad_norm": 0.1748046875,
       "learning_rate": 6.623957300178207e-06,
-      "loss": 2.0954,
+      "loss": 2.0955,
       "step": 5285
     },
     {
       "epoch": 1.79,
       "grad_norm": 0.16796875,
       "learning_rate": 6.518552277816215e-06,
-      "loss": 2.1279,
+      "loss": 2.1281,
       "step": 5290
     },
     {
       "epoch": 1.79,
       "grad_norm": 0.169921875,
       "learning_rate": 6.413964374315851e-06,
-      "loss": 2.129,
+      "loss": 2.1291,
       "step": 5295
     },
     {
@@ -7447,98 +7447,98 @@
       "epoch": 1.8,
       "grad_norm": 0.17578125,
       "learning_rate": 6.207243573549959e-06,
-      "loss": 2.1047,
+      "loss": 2.1046,
       "step": 5305
     },
     {
       "epoch": 1.8,
       "grad_norm": 0.1689453125,
       "learning_rate": 6.1051124832254944e-06,
-      "loss": 2.0973,
+      "loss": 2.0975,
       "step": 5310
     },
     {
       "epoch": 1.8,
       "grad_norm": 0.1689453125,
       "learning_rate": 6.003802125628966e-06,
-      "loss": 2.133,
+      "loss": 2.1329,
       "step": 5315
     },
     {
       "epoch": 1.8,
       "grad_norm": 0.171875,
       "learning_rate": 5.903313386311527e-06,
-      "loss": 2.094,
+      "loss": 2.0939,
       "step": 5320
     },
     {
       "epoch": 1.8,
       "grad_norm": 0.1689453125,
       "learning_rate": 5.803647143642554e-06,
-      "loss": 2.1193,
+      "loss": 2.1194,
       "step": 5325
     },
     {
       "epoch": 1.81,
-      "grad_norm": 0.1728515625,
+      "grad_norm": 0.169921875,
       "learning_rate": 5.704804268802077e-06,
       "loss": 2.1236,
       "step": 5330
     },
     {
       "epoch": 1.81,
-      "grad_norm": 0.17578125,
+      "grad_norm": 0.1748046875,
       "learning_rate": 5.606785625773048e-06,
-      "loss": 2.1485,
+      "loss": 2.1484,
       "step": 5335
     },
     {
       "epoch": 1.81,
       "grad_norm": 0.1748046875,
       "learning_rate": 5.5095920713338134e-06,
-      "loss": 2.1401,
+      "loss": 2.14,
       "step": 5340
     },
     {
       "epoch": 1.81,
-      "grad_norm": 0.1748046875,
+      "grad_norm": 0.17578125,
       "learning_rate": 5.413224455050692e-06,
-      "loss": 2.1279,
+      "loss": 2.128,
       "step": 5345
     },
     {
       "epoch": 1.81,
       "grad_norm": 0.1689453125,
       "learning_rate": 5.3176836192704414e-06,
-      "loss": 2.1355,
+      "loss": 2.1357,
       "step": 5350
     },
     {
       "epoch": 1.81,
       "grad_norm": 0.16796875,
       "learning_rate": 5.222970399112981e-06,
-      "loss": 2.119,
+      "loss": 2.1191,
       "step": 5355
     },
     {
       "epoch": 1.82,
       "grad_norm": 0.16796875,
       "learning_rate": 5.12908562246408e-06,
-      "loss": 2.1322,
+      "loss": 2.1321,
       "step": 5360
     },
     {
       "epoch": 1.82,
       "grad_norm": 0.171875,
       "learning_rate": 5.036030109968082e-06,
-      "loss": 2.1283,
+      "loss": 2.1284,
       "step": 5365
     },
     {
       "epoch": 1.82,
       "grad_norm": 0.1650390625,
       "learning_rate": 4.9438046750207465e-06,
-      "loss": 2.1171,
+      "loss": 2.117,
       "step": 5370
     },
     {
@@ -7566,7 +7566,7 @@
       "epoch": 1.83,
       "grad_norm": 0.1728515625,
       "learning_rate": 4.583219724537046e-06,
-      "loss": 2.1144,
+      "loss": 2.1145,
       "step": 5390
     },
     {
@@ -7580,7 +7580,7 @@
       "epoch": 1.83,
       "grad_norm": 0.169921875,
       "learning_rate": 4.407928328923194e-06,
-      "loss": 2.1348,
+      "loss": 2.1347,
       "step": 5400
     },
     {
@@ -7608,14 +7608,14 @@
       "epoch": 1.84,
       "grad_norm": 0.1689453125,
       "learning_rate": 4.067378341019257e-06,
-      "loss": 2.1269,
+      "loss": 2.1268,
       "step": 5420
     },
     {
       "epoch": 1.84,
       "grad_norm": 0.173828125,
       "learning_rate": 3.984335363145752e-06,
-      "loss": 2.1027,
+      "loss": 2.1028,
       "step": 5425
     },
     {
@@ -7636,7 +7636,7 @@
       "epoch": 1.84,
       "grad_norm": 0.166015625,
       "learning_rate": 3.740244918652469e-06,
-      "loss": 2.1148,
+      "loss": 2.1147,
       "step": 5440
     },
     {
@@ -7648,7 +7648,7 @@
     },
     {
       "epoch": 1.85,
-      "grad_norm": 0.16796875,
+      "grad_norm": 0.1689453125,
       "learning_rate": 3.581723790184377e-06,
       "loss": 2.1154,
       "step": 5450
@@ -7657,7 +7657,7 @@
       "epoch": 1.85,
       "grad_norm": 0.1650390625,
       "learning_rate": 3.503727065785878e-06,
-      "loss": 2.1344,
+      "loss": 2.1346,
       "step": 5455
     },
     {
@@ -7678,42 +7678,42 @@
       "epoch": 1.85,
       "grad_norm": 0.1708984375,
       "learning_rate": 3.2748004110041863e-06,
-      "loss": 2.1073,
+      "loss": 2.1074,
       "step": 5470
     },
     {
       "epoch": 1.85,
       "grad_norm": 0.1669921875,
       "learning_rate": 3.2001815889286856e-06,
-      "loss": 2.0838,
+      "loss": 2.0837,
       "step": 5475
     },
     {
       "epoch": 1.86,
       "grad_norm": 0.1689453125,
       "learning_rate": 3.126408891505217e-06,
-      "loss": 2.0874,
+      "loss": 2.0873,
       "step": 5480
     },
     {
       "epoch": 1.86,
       "grad_norm": 0.16796875,
       "learning_rate": 3.0534829635789684e-06,
-      "loss": 2.1232,
+      "loss": 2.1231,
       "step": 5485
     },
     {
       "epoch": 1.86,
       "grad_norm": 0.1669921875,
       "learning_rate": 2.9814044425935606e-06,
-      "loss": 2.1074,
+      "loss": 2.1073,
       "step": 5490
     },
     {
       "epoch": 1.86,
       "grad_norm": 0.1689453125,
       "learning_rate": 2.910173958585416e-06,
-      "loss": 2.0819,
+      "loss": 2.0818,
       "step": 5495
     },
     {
@@ -7734,7 +7734,7 @@
       "epoch": 1.87,
       "grad_norm": 0.1650390625,
       "learning_rate": 2.7015769175665063e-06,
-      "loss": 2.0805,
+      "loss": 2.0804,
       "step": 5510
     },
     {
@@ -7748,14 +7748,14 @@
       "epoch": 1.87,
       "grad_norm": 0.1767578125,
       "learning_rate": 2.5667636252897143e-06,
-      "loss": 2.109,
+      "loss": 2.1088,
       "step": 5520
     },
     {
       "epoch": 1.87,
       "grad_norm": 0.1630859375,
       "learning_rate": 2.5006341784238107e-06,
-      "loss": 2.096,
+      "loss": 2.0958,
       "step": 5525
     },
     {
@@ -7767,9 +7767,9 @@
     },
     {
       "epoch": 1.87,
-      "grad_norm": 0.1728515625,
+      "grad_norm": 0.171875,
       "learning_rate": 2.370932573410667e-06,
-      "loss": 2.1081,
+      "loss": 2.1082,
       "step": 5535
     },
     {
@@ -7783,7 +7783,7 @@
       "epoch": 1.88,
       "grad_norm": 0.16796875,
       "learning_rate": 2.244644453321165e-06,
-      "loss": 2.0947,
+      "loss": 2.0948,
       "step": 5545
     },
     {
@@ -7811,21 +7811,21 @@
       "epoch": 1.88,
       "grad_norm": 0.1708984375,
       "learning_rate": 2.0023262104694852e-06,
-      "loss": 2.1125,
+      "loss": 2.1126,
       "step": 5565
     },
     {
       "epoch": 1.89,
       "grad_norm": 0.1708984375,
       "learning_rate": 1.9438868323282124e-06,
-      "loss": 2.1181,
+      "loss": 2.118,
       "step": 5570
     },
     {
       "epoch": 1.89,
       "grad_norm": 0.1689453125,
       "learning_rate": 1.8863045600782003e-06,
-      "loss": 2.1209,
+      "loss": 2.1207,
       "step": 5575
     },
     {
@@ -7837,7 +7837,7 @@
     },
     {
       "epoch": 1.89,
-      "grad_norm": 0.171875,
+      "grad_norm": 0.1728515625,
       "learning_rate": 1.7737133390561046e-06,
       "loss": 2.112,
       "step": 5585
@@ -7851,16 +7851,16 @@
     },
     {
       "epoch": 1.9,
-      "grad_norm": 0.1708984375,
+      "grad_norm": 0.169921875,
       "learning_rate": 1.6645564840221396e-06,
-      "loss": 2.1044,
+      "loss": 2.1042,
       "step": 5595
     },
     {
       "epoch": 1.9,
       "grad_norm": 0.1669921875,
       "learning_rate": 1.6112671411138036e-06,
-      "loss": 2.0861,
+      "loss": 2.086,
       "step": 5600
     },
     {
@@ -7881,7 +7881,7 @@
       "epoch": 1.9,
       "grad_norm": 0.16796875,
       "learning_rate": 1.4565610178681299e-06,
-      "loss": 2.1317,
+      "loss": 2.1318,
       "step": 5615
     },
     {
@@ -7895,14 +7895,14 @@
       "epoch": 1.91,
       "grad_norm": 0.1689453125,
       "learning_rate": 1.3577296790636684e-06,
-      "loss": 2.1301,
+      "loss": 2.1302,
       "step": 5625
     },
     {
       "epoch": 1.91,
       "grad_norm": 0.1708984375,
       "learning_rate": 1.3096071397873056e-06,
-      "loss": 2.0979,
+      "loss": 2.0978,
       "step": 5630
     },
     {
@@ -7916,14 +7916,14 @@
       "epoch": 1.91,
       "grad_norm": 0.171875,
       "learning_rate": 1.2159504246718522e-06,
-      "loss": 2.1103,
+      "loss": 2.1102,
       "step": 5640
     },
     {
       "epoch": 1.91,
       "grad_norm": 0.16796875,
       "learning_rate": 1.1704170674836313e-06,
-      "loss": 2.1058,
+      "loss": 2.1059,
       "step": 5645
     },
     {
@@ -7951,7 +7951,7 @@
       "epoch": 1.92,
       "grad_norm": 0.17578125,
       "learning_rate": 9.969261732036605e-07,
-      "loss": 2.1311,
+      "loss": 2.1312,
       "step": 5665
     },
     {
@@ -7984,23 +7984,23 @@
     },
     {
       "epoch": 1.93,
-      "grad_norm": 0.169921875,
+      "grad_norm": 0.1689453125,
       "learning_rate": 7.99536124756517e-07,
-      "loss": 2.0929,
+      "loss": 2.0928,
       "step": 5690
     },
     {
       "epoch": 1.93,
       "grad_norm": 0.1708984375,
       "learning_rate": 7.62658098515856e-07,
-      "loss": 2.0956,
+      "loss": 2.0955,
       "step": 5695
     },
     {
       "epoch": 1.93,
       "grad_norm": 0.1689453125,
       "learning_rate": 7.266475032555619e-07,
-      "loss": 2.12,
+      "loss": 2.1201,
       "step": 5700
     },
     {
@@ -8028,28 +8028,28 @@
       "epoch": 1.94,
       "grad_norm": 0.1689453125,
       "learning_rate": 5.912856135812051e-07,
-      "loss": 2.1018,
+      "loss": 2.1019,
       "step": 5720
     },
     {
       "epoch": 1.94,
       "grad_norm": 0.16796875,
       "learning_rate": 5.596167429989807e-07,
-      "loss": 2.1054,
+      "loss": 2.1053,
       "step": 5725
     },
     {
       "epoch": 1.94,
       "grad_norm": 0.169921875,
       "learning_rate": 5.288170781762469e-07,
-      "loss": 2.1114,
+      "loss": 2.1113,
       "step": 5730
     },
     {
       "epoch": 1.94,
       "grad_norm": 0.1689453125,
       "learning_rate": 4.98886888332073e-07,
-      "loss": 2.1091,
+      "loss": 2.109,
       "step": 5735
     },
     {
@@ -8063,14 +8063,14 @@
       "epoch": 1.95,
       "grad_norm": 0.1767578125,
       "learning_rate": 4.416359724530139e-07,
-      "loss": 2.1143,
+      "loss": 2.1142,
       "step": 5745
     },
     {
       "epoch": 1.95,
       "grad_norm": 0.1708984375,
       "learning_rate": 4.143157468468717e-07,
-      "loss": 2.1154,
+      "loss": 2.1152,
       "step": 5750
     },
     {
@@ -8089,7 +8089,7 @@
     },
     {
       "epoch": 1.95,
-      "grad_norm": 0.1748046875,
+      "grad_norm": 0.17578125,
       "learning_rate": 3.3757884219383085e-07,
       "loss": 2.1078,
       "step": 5765
@@ -8098,14 +8098,14 @@
       "epoch": 1.95,
       "grad_norm": 0.1728515625,
       "learning_rate": 3.1374187664844346e-07,
-      "loss": 2.115,
+      "loss": 2.1151,
       "step": 5770
     },
     {
       "epoch": 1.96,
       "grad_norm": 0.16796875,
       "learning_rate": 2.907762660482538e-07,
-      "loss": 2.1022,
+      "loss": 2.1021,
       "step": 5775
     },
     {
@@ -8119,14 +8119,14 @@
       "epoch": 1.96,
       "grad_norm": 0.166015625,
       "learning_rate": 2.474599050323989e-07,
-      "loss": 2.1228,
+      "loss": 2.123,
       "step": 5785
     },
     {
       "epoch": 1.96,
       "grad_norm": 0.169921875,
       "learning_rate": 2.271095332438966e-07,
-      "loss": 2.0893,
+      "loss": 2.0891,
       "step": 5790
     },
     {
@@ -8138,9 +8138,9 @@
     },
     {
       "epoch": 1.96,
-      "grad_norm": 0.16796875,
+      "grad_norm": 0.1689453125,
       "learning_rate": 1.890252965145112e-07,
-      "loss": 2.0929,
+      "loss": 2.0931,
       "step": 5800
     },
     {
@@ -8161,7 +8161,7 @@
       "epoch": 1.97,
       "grad_norm": 0.1689453125,
       "learning_rate": 1.384426480462997e-07,
-      "loss": 2.0969,
+      "loss": 2.0971,
       "step": 5815
     },
     {
@@ -8173,131 +8173,131 @@
     },
     {
       "epoch": 1.97,
-      "grad_norm": 0.1708984375,
+      "grad_norm": 0.171875,
       "learning_rate": 1.0908507292026748e-07,
-      "loss": 2.133,
+      "loss": 2.1332,
       "step": 5825
     },
     {
       "epoch": 1.97,
       "grad_norm": 0.1650390625,
       "learning_rate": 9.571593887891528e-08,
-      "loss": 2.1076,
+      "loss": 2.1077,
       "step": 5830
     },
     {
       "epoch": 1.98,
-      "grad_norm": 0.1708984375,
+      "grad_norm": 0.171875,
       "learning_rate": 8.322006554171146e-08,
-      "loss": 2.142,
+      "loss": 2.1421,
       "step": 5835
     },
     {
       "epoch": 1.98,
       "grad_norm": 0.1669921875,
       "learning_rate": 7.159756213476199e-08,
-      "loss": 2.0921,
+      "loss": 2.0922,
       "step": 5840
     },
     {
       "epoch": 1.98,
       "grad_norm": 0.1640625,
       "learning_rate": 6.084853025005721e-08,
-      "loss": 2.1118,
+      "loss": 2.1117,
       "step": 5845
     },
     {
       "epoch": 1.98,
       "grad_norm": 0.169921875,
       "learning_rate": 5.0973063844605986e-08,
-      "loss": 2.0982,
+      "loss": 2.0981,
       "step": 5850
     },
     {
       "epoch": 1.98,
       "grad_norm": 0.169921875,
       "learning_rate": 4.1971249239591834e-08,
-      "loss": 2.0941,
+      "loss": 2.094,
       "step": 5855
     },
     {
       "epoch": 1.98,
       "grad_norm": 0.1640625,
       "learning_rate": 3.384316511964025e-08,
-      "loss": 2.1074,
+      "loss": 2.1075,
       "step": 5860
     },
     {
       "epoch": 1.99,
       "grad_norm": 0.1689453125,
       "learning_rate": 2.658888253211922e-08,
-      "loss": 2.107,
+      "loss": 2.1071,
       "step": 5865
     },
     {
       "epoch": 1.99,
       "grad_norm": 0.171875,
       "learning_rate": 2.0208464886517508e-08,
-      "loss": 2.1051,
+      "loss": 2.105,
       "step": 5870
     },
     {
       "epoch": 1.99,
       "grad_norm": 0.16796875,
       "learning_rate": 1.4701967953911766e-08,
-      "loss": 2.1026,
+      "loss": 2.1025,
       "step": 5875
     },
     {
       "epoch": 1.99,
       "grad_norm": 0.1689453125,
       "learning_rate": 1.0069439866422503e-08,
-      "loss": 2.1107,
+      "loss": 2.1108,
       "step": 5880
     },
     {
       "epoch": 1.99,
       "grad_norm": 0.171875,
       "learning_rate": 6.3109211168699275e-09,
-      "loss": 2.1133,
+      "loss": 2.1134,
       "step": 5885
     },
     {
       "epoch": 1.99,
       "grad_norm": 0.1630859375,
       "learning_rate": 3.4264445583631622e-09,
-      "loss": 2.1141,
+      "loss": 2.114,
       "step": 5890
     },
     {
       "epoch": 2.0,
       "grad_norm": 0.169921875,
       "learning_rate": 1.4160354040448908e-09,
-      "loss": 2.0934,
+      "loss": 2.0933,
       "step": 5895
     },
     {
       "epoch": 2.0,
       "grad_norm": 0.169921875,
       "learning_rate": 2.7971122683601023e-10,
-      "loss": 2.1251,
+      "loss": 2.125,
       "step": 5900
     },
     {
       "epoch": 2.0,
-      "eval_loss": 2.1270718574523926,
-      "eval_runtime": 156.9365,
-      "eval_samples_per_second": 8.462,
-      "eval_steps_per_second": 1.058,
+      "eval_loss": 2.1289539337158203,
+      "eval_runtime": 154.4105,
+      "eval_samples_per_second": 8.451,
+      "eval_steps_per_second": 1.062,
       "step": 5904
     },
     {
       "epoch": 2.0,
       "step": 5904,
       "total_flos": 6.077264558911652e+17,
-      "train_loss": 2.1395560028107186,
-      "train_runtime": 20794.4522,
-      "train_samples_per_second": 2.272,
+      "train_loss": 2.139555886670503,
+      "train_runtime": 20822.1805,
+      "train_samples_per_second": 2.269,
       "train_steps_per_second": 0.284
     }
   ],