Training in progress, step 2000, checkpoint

checkpoint-2000/model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d1740abdbfc9cc1a0d4c4ed56571a326e8c702d30710738c5674fe029f1a2225
+oid sha256:13385550b666fe989b17ff4c0dafcbbbc5320c4b8cd07bcf293425e0d4fff27d
 size 2471661992

checkpoint-2000/optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:75a191cf28769110f495786a93173ee4b2dcb9fb5a745be4cccd562e39bb636a
+oid sha256:ecea9ac226af40f82b354a2e5cde85c9289537bc10964f816c1c2737eb9c7c6d
 size 3725003

checkpoint-2000/trainer_state.json

@@ -11,1402 +11,1402 @@
   "log_history": [
     {
       "epoch": 0.005,
-      "grad_norm": 256.0,
+      "grad_norm": 195.0,
       "learning_rate": 9.000000000000001e-07,
-      "loss": 1.0984,
+      "loss": 1.1039,
       "step": 10
     },
     {
       "epoch": 0.01,
-      "grad_norm": 141.0,
+      "grad_norm": 161.0,
       "learning_rate": 1.9000000000000002e-06,
-      "loss": 1.1754,
+      "loss": 1.1855,
       "step": 20
     },
     {
       "epoch": 0.015,
-      "grad_norm": 209.0,
+      "grad_norm": 226.0,
       "learning_rate": 2.9e-06,
-      "loss": 0.9668,
+      "loss": 0.9566,
       "step": 30
     },
     {
       "epoch": 0.02,
-      "grad_norm": 224.0,
+      "grad_norm": 252.0,
       "learning_rate": 3.900000000000001e-06,
-      "loss": 1.0859,
+      "loss": 1.0664,
       "step": 40
     },
     {
       "epoch": 0.025,
-      "grad_norm": 98.0,
+      "grad_norm": 111.5,
       "learning_rate": 4.9000000000000005e-06,
-      "loss": 0.8516,
+      "loss": 0.8496,
       "step": 50
     },
     {
       "epoch": 0.03,
-      "grad_norm": 12.25,
+      "grad_norm": 14.5,
       "learning_rate": 5.9e-06,
-      "loss": 0.6496,
+      "loss": 0.6879,
       "step": 60
     },
     {
       "epoch": 0.035,
-      "grad_norm": 6.03125,
+      "grad_norm": 7.59375,
       "learning_rate": 6.9e-06,
-      "loss": 0.4813,
+      "loss": 0.4643,
       "step": 70
     },
     {
       "epoch": 0.04,
-      "grad_norm": 16.25,
+      "grad_norm": 14.375,
       "learning_rate": 7.9e-06,
-      "loss": 0.3936,
+      "loss": 0.4078,
       "step": 80
     },
     {
       "epoch": 0.045,
-      "grad_norm": 21.625,
+      "grad_norm": 30.625,
       "learning_rate": 8.900000000000001e-06,
-      "loss": 0.3643,
+      "loss": 0.3529,
       "step": 90
     },
     {
       "epoch": 0.05,
-      "grad_norm": 252.0,
+      "grad_norm": 165.0,
       "learning_rate": 9.9e-06,
-      "loss": 0.3336,
+      "loss": 0.3338,
       "step": 100
     },
     {
       "epoch": 0.055,
-      "grad_norm": 3.078125,
+      "grad_norm": 3.03125,
       "learning_rate": 1.0900000000000002e-05,
-      "loss": 0.3113,
+      "loss": 0.3053,
       "step": 110
     },
     {
       "epoch": 0.06,
-      "grad_norm": 3.328125,
+      "grad_norm": 3.078125,
       "learning_rate": 1.1900000000000001e-05,
-      "loss": 0.299,
+      "loss": 0.2962,
       "step": 120
     },
     {
       "epoch": 0.065,
-      "grad_norm": 28.5,
+      "grad_norm": 2.78125,
       "learning_rate": 1.2900000000000002e-05,
-      "loss": 0.2716,
+      "loss": 0.2529,
       "step": 130
     },
     {
       "epoch": 0.07,
-      "grad_norm": 2.953125,
+      "grad_norm": 3.015625,
       "learning_rate": 1.39e-05,
-      "loss": 0.2648,
+      "loss": 0.2544,
       "step": 140
     },
     {
       "epoch": 0.075,
-      "grad_norm": 35.25,
+      "grad_norm": 2.28125,
       "learning_rate": 1.4900000000000001e-05,
-      "loss": 0.2335,
+      "loss": 0.2417,
       "step": 150
     },
     {
       "epoch": 0.08,
-      "grad_norm": 2.34375,
+      "grad_norm": 2.765625,
       "learning_rate": 1.5900000000000004e-05,
-      "loss": 0.2365,
+      "loss": 0.2387,
       "step": 160
     },
     {
       "epoch": 0.085,
-      "grad_norm": 1.921875,
+      "grad_norm": 1.8125,
       "learning_rate": 1.69e-05,
-      "loss": 0.2247,
+      "loss": 0.2291,
       "step": 170
     },
     {
       "epoch": 0.09,
-      "grad_norm": 2.890625,
+      "grad_norm": 2.71875,
       "learning_rate": 1.79e-05,
-      "loss": 0.2183,
+      "loss": 0.2181,
       "step": 180
     },
     {
       "epoch": 0.095,
-      "grad_norm": 1.7265625,
+      "grad_norm": 1.984375,
       "learning_rate": 1.8900000000000002e-05,
-      "loss": 0.2058,
+      "loss": 0.217,
       "step": 190
     },
     {
       "epoch": 0.1,
-      "grad_norm": 1.5625,
+      "grad_norm": 2.046875,
       "learning_rate": 1.9900000000000003e-05,
-      "loss": 0.2061,
+      "loss": 0.2021,
       "step": 200
     },
     {
       "epoch": 0.105,
-      "grad_norm": 1.9375,
+      "grad_norm": 1.5390625,
       "learning_rate": 1.9998766324816606e-05,
-      "loss": 0.2005,
+      "loss": 0.1965,
       "step": 210
     },
     {
       "epoch": 0.11,
-      "grad_norm": 1.921875,
+      "grad_norm": 1.34375,
       "learning_rate": 1.9994502159417576e-05,
-      "loss": 0.1814,
+      "loss": 0.1746,
       "step": 220
     },
     {
       "epoch": 0.115,
-      "grad_norm": 2.125,
+      "grad_norm": 1.5078125,
       "learning_rate": 1.9987193571841865e-05,
-      "loss": 0.1793,
+      "loss": 0.1643,
       "step": 230
     },
     {
       "epoch": 0.12,
-      "grad_norm": 1.7109375,
+      "grad_norm": 1.25,
       "learning_rate": 1.9976842788356054e-05,
-      "loss": 0.1606,
+      "loss": 0.1653,
       "step": 240
     },
     {
       "epoch": 0.125,
-      "grad_norm": 2.75,
+      "grad_norm": 2.578125,
       "learning_rate": 1.9963452961909065e-05,
-      "loss": 0.1579,
+      "loss": 0.1516,
       "step": 250
     },
     {
       "epoch": 0.13,
-      "grad_norm": 1.546875,
+      "grad_norm": 1.703125,
       "learning_rate": 1.9947028171171742e-05,
-      "loss": 0.1607,
+      "loss": 0.1457,
       "step": 260
     },
     {
       "epoch": 0.135,
-      "grad_norm": 1.5078125,
+      "grad_norm": 1.921875,
       "learning_rate": 1.9927573419294456e-05,
-      "loss": 0.1466,
+      "loss": 0.1579,
       "step": 270
     },
     {
       "epoch": 0.14,
       "grad_norm": 1.3671875,
       "learning_rate": 1.990509463238309e-05,
-      "loss": 0.1426,
+      "loss": 0.1339,
       "step": 280
     },
     {
       "epoch": 0.145,
-      "grad_norm": 2.171875,
+      "grad_norm": 1.34375,
       "learning_rate": 1.9879598657693894e-05,
-      "loss": 0.1438,
+      "loss": 0.1425,
       "step": 290
     },
     {
       "epoch": 0.15,
-      "grad_norm": 0.984375,
+      "grad_norm": 1.1796875,
       "learning_rate": 1.985109326154774e-05,
-      "loss": 0.1368,
+      "loss": 0.1331,
       "step": 300
     },
     {
       "epoch": 0.155,
       "grad_norm": 1.1640625,
       "learning_rate": 1.981958712696444e-05,
-      "loss": 0.1335,
+      "loss": 0.1409,
       "step": 310
     },
     {
       "epoch": 0.16,
-      "grad_norm": 1.6015625,
+      "grad_norm": 1.5703125,
       "learning_rate": 1.9785089851017788e-05,
-      "loss": 0.1437,
+      "loss": 0.1392,
       "step": 320
     },
     {
       "epoch": 0.165,
-      "grad_norm": 1.3203125,
+      "grad_norm": 1.484375,
       "learning_rate": 1.974761194191222e-05,
-      "loss": 0.1373,
+      "loss": 0.1407,
       "step": 330
     },
     {
       "epoch": 0.17,
-      "grad_norm": 1.1875,
+      "grad_norm": 1.203125,
       "learning_rate": 1.970716481578191e-05,
-      "loss": 0.1328,
+      "loss": 0.1327,
       "step": 340
     },
     {
       "epoch": 0.175,
-      "grad_norm": 1.53125,
+      "grad_norm": 1.28125,
       "learning_rate": 1.9663760793213297e-05,
-      "loss": 0.1287,
+      "loss": 0.1285,
       "step": 350
     },
     {
       "epoch": 0.18,
-      "grad_norm": 1.3203125,
+      "grad_norm": 1.4453125,
       "learning_rate": 1.9617413095492114e-05,
-      "loss": 0.1263,
+      "loss": 0.1251,
       "step": 360
     },
     {
       "epoch": 0.185,
-      "grad_norm": 1.078125,
+      "grad_norm": 1.3046875,
       "learning_rate": 1.956813584057608e-05,
-      "loss": 0.1211,
+      "loss": 0.128,
       "step": 370
     },
     {
       "epoch": 0.19,
-      "grad_norm": 1.71875,
+      "grad_norm": 1.4453125,
       "learning_rate": 1.9515944038794384e-05,
-      "loss": 0.1295,
+      "loss": 0.1269,
       "step": 380
     },
     {
       "epoch": 0.195,
-      "grad_norm": 1.421875,
+      "grad_norm": 1.1953125,
       "learning_rate": 1.9460853588275454e-05,
-      "loss": 0.1217,
+      "loss": 0.1201,
       "step": 390
     },
     {
       "epoch": 0.2,
-      "grad_norm": 1.125,
+      "grad_norm": 1.3046875,
       "learning_rate": 1.940288127010419e-05,
-      "loss": 0.1166,
+      "loss": 0.1278,
       "step": 400
     },
     {
       "epoch": 0.205,
-      "grad_norm": 1.328125,
+      "grad_norm": 1.1953125,
       "learning_rate": 1.9342044743210295e-05,
-      "loss": 0.1285,
+      "loss": 0.1284,
       "step": 410
     },
     {
       "epoch": 0.21,
-      "grad_norm": 1.109375,
+      "grad_norm": 1.1328125,
       "learning_rate": 1.92783625389892e-05,
-      "loss": 0.1231,
+      "loss": 0.1135,
       "step": 420
     },
     {
       "epoch": 0.215,
-      "grad_norm": 1.1875,
+      "grad_norm": 1.7265625,
       "learning_rate": 1.9211854055657216e-05,
-      "loss": 0.1211,
+      "loss": 0.1268,
       "step": 430
     },
     {
       "epoch": 0.22,
-      "grad_norm": 1.078125,
+      "grad_norm": 0.96875,
       "learning_rate": 1.9142539552342638e-05,
-      "loss": 0.1229,
+      "loss": 0.1134,
       "step": 440
     },
     {
       "epoch": 0.225,
-      "grad_norm": 1.453125,
+      "grad_norm": 1.3984375,
       "learning_rate": 1.907044014291465e-05,
-      "loss": 0.1222,
+      "loss": 0.1245,
       "step": 450
     },
     {
       "epoch": 0.23,
-      "grad_norm": 1.3359375,
+      "grad_norm": 1.5546875,
       "learning_rate": 1.8995577789551806e-05,
-      "loss": 0.1191,
+      "loss": 0.1188,
       "step": 460
     },
     {
       "epoch": 0.235,
-      "grad_norm": 1.28125,
+      "grad_norm": 1.3125,
       "learning_rate": 1.8917975296052143e-05,
-      "loss": 0.1159,
+      "loss": 0.1102,
       "step": 470
     },
     {
       "epoch": 0.24,
-      "grad_norm": 1.203125,
+      "grad_norm": 1.296875,
       "learning_rate": 1.8837656300886937e-05,
-      "loss": 0.112,
+      "loss": 0.1127,
       "step": 480
     },
     {
       "epoch": 0.245,
-      "grad_norm": 1.3828125,
+      "grad_norm": 1.0703125,
       "learning_rate": 1.875464527000018e-05,
-      "loss": 0.123,
+      "loss": 0.1104,
       "step": 490
     },
     {
       "epoch": 0.25,
-      "grad_norm": 1.1953125,
+      "grad_norm": 0.98046875,
       "learning_rate": 1.866896748935603e-05,
-      "loss": 0.1073,
+      "loss": 0.1035,
       "step": 500
     },
     {
       "epoch": 0.255,
-      "grad_norm": 1.0,
+      "grad_norm": 1.0234375,
       "learning_rate": 1.858064905723645e-05,
-      "loss": 0.1082,
+      "loss": 0.1125,
       "step": 510
     },
     {
       "epoch": 0.26,
       "grad_norm": 1.4765625,
       "learning_rate": 1.8489716876291417e-05,
-      "loss": 0.1128,
+      "loss": 0.1105,
       "step": 520
     },
     {
       "epoch": 0.265,
-      "grad_norm": 1.53125,
+      "grad_norm": 1.140625,
       "learning_rate": 1.8396198645344133e-05,
-      "loss": 0.1101,
+      "loss": 0.1078,
       "step": 530
     },
     {
       "epoch": 0.27,
-      "grad_norm": 1.296875,
+      "grad_norm": 1.015625,
       "learning_rate": 1.8300122850953678e-05,
-      "loss": 0.1068,
+      "loss": 0.1128,
       "step": 540
     },
     {
       "epoch": 0.275,
-      "grad_norm": 1.1640625,
+      "grad_norm": 1.1953125,
       "learning_rate": 1.8201518758737726e-05,
-      "loss": 0.1196,
+      "loss": 0.1115,
       "step": 550
     },
     {
       "epoch": 0.28,
-      "grad_norm": 0.9921875,
+      "grad_norm": 0.953125,
       "learning_rate": 1.8100416404457962e-05,
-      "loss": 0.1068,
+      "loss": 0.112,
       "step": 560
     },
     {
       "epoch": 0.285,
-      "grad_norm": 1.1875,
+      "grad_norm": 0.83984375,
       "learning_rate": 1.799684658487091e-05,
-      "loss": 0.1133,
+      "loss": 0.1088,
       "step": 570
     },
     {
       "epoch": 0.29,
-      "grad_norm": 1.34375,
+      "grad_norm": 1.3671875,
       "learning_rate": 1.789084084834691e-05,
-      "loss": 0.1176,
+      "loss": 0.111,
       "step": 580
     },
     {
       "epoch": 0.295,
-      "grad_norm": 1.0234375,
+      "grad_norm": 0.921875,
       "learning_rate": 1.778243148526021e-05,
-      "loss": 0.114,
+      "loss": 0.1086,
       "step": 590
     },
     {
       "epoch": 0.3,
-      "grad_norm": 1.03125,
+      "grad_norm": 1.3203125,
       "learning_rate": 1.7671651518153e-05,
       "loss": 0.1051,
       "step": 600
     },
     {
       "epoch": 0.305,
-      "grad_norm": 1.453125,
+      "grad_norm": 1.25,
       "learning_rate": 1.7558534691676396e-05,
-      "loss": 0.1061,
+      "loss": 0.1136,
       "step": 610
     },
     {
       "epoch": 0.31,
-      "grad_norm": 1.171875,
+      "grad_norm": 1.1484375,
       "learning_rate": 1.744311546231154e-05,
-      "loss": 0.1084,
+      "loss": 0.1052,
       "step": 620
     },
     {
       "epoch": 0.315,
-      "grad_norm": 1.3828125,
+      "grad_norm": 1.0,
       "learning_rate": 1.732542898787379e-05,
-      "loss": 0.1064,
+      "loss": 0.1065,
       "step": 630
     },
     {
       "epoch": 0.32,
-      "grad_norm": 0.7578125,
+      "grad_norm": 0.8671875,
       "learning_rate": 1.7205511116803306e-05,
-      "loss": 0.1049,
+      "loss": 0.1129,
       "step": 640
     },
     {
       "epoch": 0.325,
-      "grad_norm": 0.97265625,
+      "grad_norm": 1.15625,
       "learning_rate": 1.708339837724529e-05,
-      "loss": 0.1157,
+      "loss": 0.1055,
       "step": 650
     },
     {
       "epoch": 0.33,
-      "grad_norm": 1.4296875,
+      "grad_norm": 0.734375,
       "learning_rate": 1.6959127965923144e-05,
-      "loss": 0.1121,
+      "loss": 0.1066,
       "step": 660
     },
     {
       "epoch": 0.335,
-      "grad_norm": 1.0703125,
+      "grad_norm": 1.09375,
       "learning_rate": 1.6832737736807994e-05,
-      "loss": 0.1073,
+      "loss": 0.1074,
       "step": 670
     },
     {
       "epoch": 0.34,
-      "grad_norm": 1.1640625,
+      "grad_norm": 1.2578125,
       "learning_rate": 1.6704266189587992e-05,
-      "loss": 0.104,
+      "loss": 0.1021,
       "step": 680
     },
     {
       "epoch": 0.345,
-      "grad_norm": 0.91796875,
+      "grad_norm": 0.8125,
       "learning_rate": 1.657375245794096e-05,
-      "loss": 0.0994,
+      "loss": 0.0991,
       "step": 690
     },
     {
       "epoch": 0.35,
-      "grad_norm": 1.46875,
+      "grad_norm": 1.09375,
       "learning_rate": 1.644123629761387e-05,
-      "loss": 0.1047,
+      "loss": 0.1066,
       "step": 700
     },
     {
       "epoch": 0.355,
-      "grad_norm": 0.79296875,
+      "grad_norm": 1.1015625,
       "learning_rate": 1.6306758074312866e-05,
-      "loss": 0.0963,
+      "loss": 0.103,
       "step": 710
     },
     {
       "epoch": 0.36,
-      "grad_norm": 1.1015625,
+      "grad_norm": 1.0234375,
       "learning_rate": 1.617035875140749e-05,
-      "loss": 0.0971,
+      "loss": 0.1026,
       "step": 720
     },
     {
       "epoch": 0.365,
-      "grad_norm": 0.9765625,
+      "grad_norm": 0.859375,
       "learning_rate": 1.6032079877452825e-05,
-      "loss": 0.1009,
+      "loss": 0.1048,
       "step": 730
     },
     {
       "epoch": 0.37,
-      "grad_norm": 1.234375,
+      "grad_norm": 1.0390625,
       "learning_rate": 1.5891963573533424e-05,
-      "loss": 0.1057,
+      "loss": 0.0967,
       "step": 740
     },
     {
       "epoch": 0.375,
-      "grad_norm": 1.09375,
+      "grad_norm": 1.265625,
       "learning_rate": 1.575005252043279e-05,
-      "loss": 0.0979,
+      "loss": 0.1021,
       "step": 750
     },
     {
       "epoch": 0.38,
-      "grad_norm": 1.09375,
+      "grad_norm": 0.90234375,
       "learning_rate": 1.560638994563242e-05,
-      "loss": 0.0974,
+      "loss": 0.0941,
       "step": 760
     },
     {
       "epoch": 0.385,
-      "grad_norm": 0.75,
+      "grad_norm": 0.984375,
       "learning_rate": 1.5461019610144292e-05,
-      "loss": 0.0943,
+      "loss": 0.0974,
       "step": 770
     },
     {
       "epoch": 0.39,
-      "grad_norm": 0.84375,
+      "grad_norm": 0.91015625,
       "learning_rate": 1.531398579518083e-05,
-      "loss": 0.0967,
+      "loss": 0.0934,
       "step": 780
     },
     {
       "epoch": 0.395,
-      "grad_norm": 1.5,
+      "grad_norm": 1.03125,
       "learning_rate": 1.516533328866642e-05,
-      "loss": 0.0953,
+      "loss": 0.0988,
       "step": 790
     },
     {
       "epoch": 0.4,
-      "grad_norm": 0.671875,
+      "grad_norm": 1.5078125,
       "learning_rate": 1.5015107371594576e-05,
-      "loss": 0.1023,
+      "loss": 0.0979,
       "step": 800
     },
     {
       "epoch": 0.405,
-      "grad_norm": 1.03125,
+      "grad_norm": 1.109375,
       "learning_rate": 1.4863353804234906e-05,
-      "loss": 0.0945,
+      "loss": 0.0988,
       "step": 810
     },
     {
       "epoch": 0.41,
-      "grad_norm": 0.96484375,
+      "grad_norm": 1.2265625,
       "learning_rate": 1.47101188121941e-05,
-      "loss": 0.095,
+      "loss": 0.0964,
       "step": 820
     },
     {
       "epoch": 0.415,
-      "grad_norm": 1.2890625,
+      "grad_norm": 0.91796875,
       "learning_rate": 1.4555449072335157e-05,
-      "loss": 0.1022,
+      "loss": 0.0987,
       "step": 830
     },
     {
       "epoch": 0.42,
-      "grad_norm": 0.83984375,
+      "grad_norm": 0.953125,
       "learning_rate": 1.4399391698559153e-05,
-      "loss": 0.1005,
+      "loss": 0.0994,
       "step": 840
     },
     {
       "epoch": 0.425,
-      "grad_norm": 1.5234375,
+      "grad_norm": 1.0390625,
       "learning_rate": 1.4241994227453902e-05,
-      "loss": 0.0983,
+      "loss": 0.0884,
       "step": 850
     },
     {
       "epoch": 0.43,
-      "grad_norm": 1.0234375,
+      "grad_norm": 0.890625,
       "learning_rate": 1.408330460381385e-05,
-      "loss": 0.0861,
+      "loss": 0.0989,
       "step": 860
     },
     {
       "epoch": 0.435,
-      "grad_norm": 1.2109375,
+      "grad_norm": 0.90625,
       "learning_rate": 1.3923371166035615e-05,
-      "loss": 0.0945,
+      "loss": 0.0968,
       "step": 870
     },
     {
       "epoch": 0.44,
       "grad_norm": 0.9296875,
       "learning_rate": 1.3762242631393656e-05,
-      "loss": 0.0935,
+      "loss": 0.0947,
       "step": 880
     },
     {
       "epoch": 0.445,
-      "grad_norm": 0.98046875,
+      "grad_norm": 1.0234375,
       "learning_rate": 1.3599968081200515e-05,
-      "loss": 0.0996,
+      "loss": 0.0912,
       "step": 890
     },
     {
       "epoch": 0.45,
-      "grad_norm": 0.9609375,
+      "grad_norm": 0.93359375,
       "learning_rate": 1.3436596945856164e-05,
-      "loss": 0.094,
+      "loss": 0.1021,
       "step": 900
     },
     {
       "epoch": 0.455,
-      "grad_norm": 1.15625,
+      "grad_norm": 0.8125,
       "learning_rate": 1.327217898979104e-05,
-      "loss": 0.0959,
+      "loss": 0.0984,
       "step": 910
     },
     {
       "epoch": 0.46,
-      "grad_norm": 0.8515625,
+      "grad_norm": 1.296875,
       "learning_rate": 1.310676429630732e-05,
-      "loss": 0.0821,
+      "loss": 0.0914,
       "step": 920
     },
     {
       "epoch": 0.465,
-      "grad_norm": 0.796875,
+      "grad_norm": 0.86328125,
       "learning_rate": 1.294040325232304e-05,
-      "loss": 0.0921,
+      "loss": 0.0986,
       "step": 930
     },
     {
       "epoch": 0.47,
-      "grad_norm": 0.796875,
+      "grad_norm": 0.95703125,
       "learning_rate": 1.2773146533023782e-05,
-      "loss": 0.0894,
+      "loss": 0.0937,
       "step": 940
     },
     {
       "epoch": 0.475,
-      "grad_norm": 0.87890625,
+      "grad_norm": 0.91796875,
       "learning_rate": 1.2605045086426487e-05,
-      "loss": 0.0901,
+      "loss": 0.0896,
       "step": 950
     },
     {
       "epoch": 0.48,
-      "grad_norm": 0.7421875,
+      "grad_norm": 0.99609375,
       "learning_rate": 1.2436150117860226e-05,
-      "loss": 0.0834,
+      "loss": 0.0925,
       "step": 960
     },
     {
       "epoch": 0.485,
-      "grad_norm": 1.046875,
+      "grad_norm": 0.89453125,
       "learning_rate": 1.2266513074368552e-05,
-      "loss": 0.0961,
+      "loss": 0.0865,
       "step": 970
     },
     {
       "epoch": 0.49,
-      "grad_norm": 0.890625,
+      "grad_norm": 0.9765625,
       "learning_rate": 1.2096185629038219e-05,
-      "loss": 0.0926,
+      "loss": 0.0916,
       "step": 980
     },
     {
       "epoch": 0.495,
-      "grad_norm": 0.890625,
+      "grad_norm": 0.8125,
       "learning_rate": 1.1925219665259076e-05,
-      "loss": 0.0979,
+      "loss": 0.0823,
       "step": 990
     },
     {
       "epoch": 0.5,
-      "grad_norm": 0.8828125,
+      "grad_norm": 0.79296875,
       "learning_rate": 1.1753667260919872e-05,
-      "loss": 0.0922,
+      "loss": 0.093,
       "step": 1000
     },
     {
       "epoch": 0.505,
-      "grad_norm": 0.828125,
+      "grad_norm": 1.015625,
       "learning_rate": 1.1581580672544839e-05,
-      "loss": 0.087,
+      "loss": 0.0928,
       "step": 1010
     },
     {
       "epoch": 0.51,
-      "grad_norm": 0.75390625,
+      "grad_norm": 0.89453125,
       "learning_rate": 1.1409012319375828e-05,
-      "loss": 0.0856,
+      "loss": 0.0884,
       "step": 1020
     },
     {
       "epoch": 0.515,
-      "grad_norm": 1.4921875,
+      "grad_norm": 0.765625,
       "learning_rate": 1.1236014767404929e-05,
-      "loss": 0.0936,
+      "loss": 0.0902,
       "step": 1030
     },
     {
       "epoch": 0.52,
-      "grad_norm": 0.93359375,
+      "grad_norm": 0.78125,
       "learning_rate": 1.1062640713362333e-05,
-      "loss": 0.0931,
+      "loss": 0.0921,
       "step": 1040
     },
     {
       "epoch": 0.525,
-      "grad_norm": 0.91015625,
+      "grad_norm": 0.796875,
       "learning_rate": 1.0888942968664417e-05,
-      "loss": 0.0907,
+      "loss": 0.0901,
       "step": 1050
     },
     {
       "epoch": 0.53,
-      "grad_norm": 0.8125,
+      "grad_norm": 0.921875,
       "learning_rate": 1.071497444332686e-05,
-      "loss": 0.0901,
+      "loss": 0.0861,
       "step": 1060
     },
     {
       "epoch": 0.535,
-      "grad_norm": 0.828125,
+      "grad_norm": 0.73046875,
       "learning_rate": 1.0540788129847757e-05,
-      "loss": 0.0874,
+      "loss": 0.0926,
       "step": 1070
     },
     {
       "epoch": 0.54,
-      "grad_norm": 0.8828125,
+      "grad_norm": 1.21875,
       "learning_rate": 1.0366437087065564e-05,
-      "loss": 0.0861,
+      "loss": 0.0911,
       "step": 1080
     },
     {
       "epoch": 0.545,
-      "grad_norm": 1.234375,
+      "grad_norm": 1.4765625,
       "learning_rate": 1.01919744239969e-05,
-      "loss": 0.0804,
+      "loss": 0.0883,
       "step": 1090
     },
     {
       "epoch": 0.55,
-      "grad_norm": 1.578125,
+      "grad_norm": 0.82421875,
       "learning_rate": 1.0017453283658984e-05,
-      "loss": 0.0905,
+      "loss": 0.0888,
       "step": 1100
     },
     {
       "epoch": 0.555,
-      "grad_norm": 0.6875,
+      "grad_norm": 1.0390625,
       "learning_rate": 9.842926826881796e-06,
-      "loss": 0.0868,
+      "loss": 0.087,
       "step": 1110
     },
     {
       "epoch": 0.56,
-      "grad_norm": 0.859375,
+      "grad_norm": 0.828125,
       "learning_rate": 9.668448216114739e-06,
-      "loss": 0.0837,
+      "loss": 0.0967,
       "step": 1120
     },
     {
       "epoch": 0.565,
-      "grad_norm": 0.90625,
+      "grad_norm": 0.875,
       "learning_rate": 9.494070599232868e-06,
-      "loss": 0.0917,
+      "loss": 0.0838,
       "step": 1130
     },
     {
       "epoch": 0.57,
-      "grad_norm": 0.74609375,
+      "grad_norm": 1.1328125,
       "learning_rate": 9.319847093347522e-06,
-      "loss": 0.0864,
+      "loss": 0.088,
       "step": 1140
     },
     {
       "epoch": 0.575,
-      "grad_norm": 1.15625,
+      "grad_norm": 0.93359375,
       "learning_rate": 9.145830768626326e-06,
-      "loss": 0.0914,
+      "loss": 0.0879,
       "step": 1150
     },
     {
       "epoch": 0.58,
-      "grad_norm": 0.9140625,
+      "grad_norm": 0.88671875,
       "learning_rate": 8.972074632127533e-06,
-      "loss": 0.0839,
+      "loss": 0.0846,
       "step": 1160
     },
     {
       "epoch": 0.585,
-      "grad_norm": 0.80859375,
+      "grad_norm": 1.015625,
       "learning_rate": 8.79863161165353e-06,
-      "loss": 0.0909,
+      "loss": 0.0858,
       "step": 1170
     },
     {
       "epoch": 0.59,
-      "grad_norm": 0.88671875,
+      "grad_norm": 0.71875,
       "learning_rate": 8.625554539628536e-06,
-      "loss": 0.0818,
+      "loss": 0.0847,
       "step": 1180
     },
     {
       "epoch": 0.595,
-      "grad_norm": 0.62109375,
+      "grad_norm": 0.90234375,
       "learning_rate": 8.452896137005322e-06,
-      "loss": 0.083,
+      "loss": 0.0855,
       "step": 1190
     },
     {
       "epoch": 0.6,
-      "grad_norm": 0.6953125,
+      "grad_norm": 0.61328125,
       "learning_rate": 8.280708997205904e-06,
-      "loss": 0.0906,
+      "loss": 0.0809,
       "step": 1200
     },
     {
       "epoch": 0.605,
-      "grad_norm": 0.69921875,
+      "grad_norm": 0.77734375,
       "learning_rate": 8.109045570101086e-06,
-      "loss": 0.0792,
+      "loss": 0.0866,
       "step": 1210
     },
     {
       "epoch": 0.61,
-      "grad_norm": 0.83984375,
+      "grad_norm": 0.86328125,
       "learning_rate": 7.937958146033706e-06,
-      "loss": 0.081,
+      "loss": 0.086,
       "step": 1220
     },
     {
       "epoch": 0.615,
-      "grad_norm": 0.75,
+      "grad_norm": 0.6328125,
       "learning_rate": 7.767498839890489e-06,
-      "loss": 0.0848,
+      "loss": 0.0816,
       "step": 1230
     },
     {
       "epoch": 0.62,
-      "grad_norm": 0.8203125,
+      "grad_norm": 0.7421875,
       "learning_rate": 7.597719575227364e-06,
-      "loss": 0.089,
+      "loss": 0.0874,
       "step": 1240
     },
     {
       "epoch": 0.625,
-      "grad_norm": 0.5234375,
+      "grad_norm": 0.72265625,
       "learning_rate": 7.428672068453041e-06,
-      "loss": 0.0814,
+      "loss": 0.0882,
       "step": 1250
     },
     {
       "epoch": 0.63,
-      "grad_norm": 1.234375,
+      "grad_norm": 0.828125,
       "learning_rate": 7.260407813075676e-06,
-      "loss": 0.0836,
+      "loss": 0.0856,
       "step": 1260
     },
     {
       "epoch": 0.635,
-      "grad_norm": 0.84765625,
+      "grad_norm": 0.84375,
       "learning_rate": 7.092978064017475e-06,
-      "loss": 0.087,
+      "loss": 0.0885,
       "step": 1270
     },
     {
       "epoch": 0.64,
-      "grad_norm": 0.6953125,
+      "grad_norm": 0.95703125,
       "learning_rate": 6.92643382200193e-06,
-      "loss": 0.0833,
+      "loss": 0.0877,
       "step": 1280
     },
     {
       "epoch": 0.645,
-      "grad_norm": 0.90234375,
+      "grad_norm": 0.921875,
       "learning_rate": 6.7608258180185085e-06,
-      "loss": 0.0958,
+      "loss": 0.0828,
       "step": 1290
     },
     {
       "epoch": 0.65,
-      "grad_norm": 0.91015625,
+      "grad_norm": 0.80078125,
       "learning_rate": 6.596204497869501e-06,
-      "loss": 0.0887,
+      "loss": 0.084,
       "step": 1300
     },
     {
       "epoch": 0.655,
-      "grad_norm": 0.796875,
+      "grad_norm": 0.99609375,
       "learning_rate": 6.432620006803747e-06,
-      "loss": 0.0842,
+      "loss": 0.0825,
       "step": 1310
     },
     {
       "epoch": 0.66,
-      "grad_norm": 0.84765625,
+      "grad_norm": 0.8359375,
       "learning_rate": 6.2701221742419106e-06,
-      "loss": 0.0854,
+      "loss": 0.0789,
       "step": 1320
     },
     {
       "epoch": 0.665,
-      "grad_norm": 1.0390625,
+      "grad_norm": 0.6953125,
       "learning_rate": 6.108760498597939e-06,
-      "loss": 0.0785,
+      "loss": 0.0885,
       "step": 1330
     },
     {
       "epoch": 0.67,
-      "grad_norm": 0.76953125,
+      "grad_norm": 0.87890625,
       "learning_rate": 5.948584132201376e-06,
-      "loss": 0.0813,
+      "loss": 0.085,
       "step": 1340
     },
     {
       "epoch": 0.675,
-      "grad_norm": 0.796875,
+      "grad_norm": 0.80078125,
       "learning_rate": 5.789641866325091e-06,
-      "loss": 0.0827,
+      "loss": 0.0875,
       "step": 1350
     },
     {
       "epoch": 0.68,
-      "grad_norm": 0.64453125,
+      "grad_norm": 0.84765625,
       "learning_rate": 5.631982116322981e-06,
-      "loss": 0.0846,
+      "loss": 0.0813,
       "step": 1360
     },
     {
       "epoch": 0.685,
-      "grad_norm": 0.78515625,
+      "grad_norm": 0.89453125,
       "learning_rate": 5.475652906882173e-06,
-      "loss": 0.0844,
+      "loss": 0.0872,
       "step": 1370
     },
     {
       "epoch": 0.69,
-      "grad_norm": 1.1484375,
+      "grad_norm": 0.7890625,
       "learning_rate": 5.3207018573942684e-06,
-      "loss": 0.0866,
+      "loss": 0.0841,
       "step": 1380
     },
     {
       "epoch": 0.695,
-      "grad_norm": 0.703125,
+      "grad_norm": 0.71875,
       "learning_rate": 5.167176167449977e-06,
-      "loss": 0.0834,
+      "loss": 0.0805,
       "step": 1390
     },
     {
       "epoch": 0.7,
-      "grad_norm": 0.81640625,
+      "grad_norm": 0.6796875,
       "learning_rate": 5.015122602461698e-06,
-      "loss": 0.0793,
+      "loss": 0.0771,
       "step": 1400
     },
     {
       "epoch": 0.705,
-      "grad_norm": 0.73046875,
+      "grad_norm": 0.953125,
       "learning_rate": 4.864587479418302e-06,
-      "loss": 0.0825,
+      "loss": 0.086,
       "step": 1410
     },
     {
       "epoch": 0.71,
-      "grad_norm": 0.609375,
+      "grad_norm": 0.72265625,
       "learning_rate": 4.71561665277653e-06,
-      "loss": 0.0786,
+      "loss": 0.0803,
       "step": 1420
     },
     {
       "epoch": 0.715,
-      "grad_norm": 0.80078125,
+      "grad_norm": 0.8984375,
       "learning_rate": 4.568255500493292e-06,
-      "loss": 0.0836,
+      "loss": 0.0816,
       "step": 1430
     },
     {
       "epoch": 0.72,
-      "grad_norm": 0.76953125,
+      "grad_norm": 0.73046875,
       "learning_rate": 4.422548910203099e-06,
-      "loss": 0.078,
+      "loss": 0.0826,
       "step": 1440
     },
     {
       "epoch": 0.725,
-      "grad_norm": 0.8515625,
+      "grad_norm": 0.85546875,
       "learning_rate": 4.27854126554484e-06,
-      "loss": 0.0864,
+      "loss": 0.08,
       "step": 1450
     },
     {
       "epoch": 0.73,
-      "grad_norm": 0.6484375,
+      "grad_norm": 0.87890625,
       "learning_rate": 4.136276432642107e-06,
-      "loss": 0.0848,
+      "loss": 0.0813,
       "step": 1460
     },
     {
       "epoch": 0.735,
-      "grad_norm": 1.046875,
+      "grad_norm": 1.0625,
       "learning_rate": 3.9957977467411615e-06,
-      "loss": 0.0816,
+      "loss": 0.0772,
       "step": 1470
     },
     {
       "epoch": 0.74,
-      "grad_norm": 0.63671875,
+      "grad_norm": 0.66796875,
       "learning_rate": 3.857147999010568e-06,
-      "loss": 0.0807,
+      "loss": 0.0791,
       "step": 1480
     },
     {
       "epoch": 0.745,
-      "grad_norm": 0.640625,
+      "grad_norm": 0.61328125,
       "learning_rate": 3.7203694235066224e-06,
-      "loss": 0.0751,
+      "loss": 0.0801,
       "step": 1490
     },
     {
       "epoch": 0.75,
-      "grad_norm": 0.74609375,
+      "grad_norm": 0.73828125,
       "learning_rate": 3.5855036843084213e-06,
-      "loss": 0.081,
+      "loss": 0.0749,
       "step": 1500
     },
     {
       "epoch": 0.755,
-      "grad_norm": 0.70703125,
+      "grad_norm": 0.828125,
       "learning_rate": 3.452591862826603e-06,
-      "loss": 0.0862,
+      "loss": 0.0852,
       "step": 1510
     },
     {
       "epoch": 0.76,
-      "grad_norm": 0.81640625,
+      "grad_norm": 0.76953125,
       "learning_rate": 3.3216744452895356e-06,
-      "loss": 0.077,
+      "loss": 0.0817,
       "step": 1520
     },
     {
       "epoch": 0.765,
-      "grad_norm": 0.73828125,
+      "grad_norm": 0.9375,
       "learning_rate": 3.192791310410822e-06,
-      "loss": 0.0778,
+      "loss": 0.0845,
       "step": 1530
     },
     {
       "epoch": 0.77,
-      "grad_norm": 0.69140625,
+      "grad_norm": 0.83984375,
       "learning_rate": 3.0659817172418694e-06,
-      "loss": 0.0891,
+      "loss": 0.0872,
       "step": 1540
     },
     {
       "epoch": 0.775,
-      "grad_norm": 0.859375,
+      "grad_norm": 0.9609375,
       "learning_rate": 2.9412842932131904e-06,
-      "loss": 0.0873,
+      "loss": 0.0797,
       "step": 1550
     },
     {
       "epoch": 0.78,
-      "grad_norm": 0.94921875,
+      "grad_norm": 0.7734375,
       "learning_rate": 2.8187370223681134e-06,
-      "loss": 0.0827,
+      "loss": 0.0774,
       "step": 1560
     },
     {
       "epoch": 0.785,
-      "grad_norm": 0.71875,
+      "grad_norm": 0.796875,
       "learning_rate": 2.698377233792476e-06,
-      "loss": 0.0799,
+      "loss": 0.0793,
       "step": 1570
     },
     {
       "epoch": 0.79,
-      "grad_norm": 0.6484375,
+      "grad_norm": 0.7109375,
       "learning_rate": 2.5802415902438373e-06,
-      "loss": 0.0735,
+      "loss": 0.0822,
       "step": 1580
     },
     {
       "epoch": 0.795,
-      "grad_norm": 0.953125,
+      "grad_norm": 0.6796875,
       "learning_rate": 2.464366076983623e-06,
-      "loss": 0.0867,
+      "loss": 0.0808,
       "step": 1590
     },
     {
       "epoch": 0.8,
-      "grad_norm": 0.9375,
+      "grad_norm": 0.9296875,
       "learning_rate": 2.3507859908156828e-06,
-      "loss": 0.0824,
+      "loss": 0.0811,
       "step": 1600
     },
     {
       "epoch": 0.805,
-      "grad_norm": 0.828125,
+      "grad_norm": 0.80859375,
       "learning_rate": 2.2395359293345396e-06,
-      "loss": 0.0771,
+      "loss": 0.083,
       "step": 1610
     },
     {
       "epoch": 0.81,
-      "grad_norm": 0.57421875,
+      "grad_norm": 0.82421875,
       "learning_rate": 2.130649780386628e-06,
-      "loss": 0.0829,
+      "loss": 0.0882,
       "step": 1620
     },
     {
       "epoch": 0.815,
-      "grad_norm": 0.70703125,
+      "grad_norm": 0.76171875,
       "learning_rate": 2.024160711747717e-06,
-      "loss": 0.0802,
+      "loss": 0.0775,
       "step": 1630
     },
     {
       "epoch": 0.82,
-      "grad_norm": 0.73828125,
+      "grad_norm": 0.75,
       "learning_rate": 1.9201011610196972e-06,
-      "loss": 0.0737,
+      "loss": 0.0841,
       "step": 1640
     },
     {
       "epoch": 0.825,
-      "grad_norm": 0.94140625,
+      "grad_norm": 0.58203125,
       "learning_rate": 1.818502825749764e-06,
-      "loss": 0.0796,
+      "loss": 0.0762,
       "step": 1650
     },
     {
       "epoch": 0.83,
-      "grad_norm": 0.88671875,
+      "grad_norm": 0.77734375,
       "learning_rate": 1.7193966537750561e-06,
-      "loss": 0.0784,
+      "loss": 0.0761,
       "step": 1660
     },
     {
       "epoch": 0.835,
-      "grad_norm": 0.78515625,
+      "grad_norm": 0.7734375,
       "learning_rate": 1.6228128337956128e-06,
-      "loss": 0.0755,
+      "loss": 0.0845,
       "step": 1670
     },
     {
       "epoch": 0.84,
-      "grad_norm": 0.7421875,
+      "grad_norm": 0.6015625,
       "learning_rate": 1.5287807861786308e-06,
-      "loss": 0.0857,
+      "loss": 0.0749,
       "step": 1680
     },
     {
       "epoch": 0.845,
-      "grad_norm": 0.90625,
+      "grad_norm": 0.62890625,
       "learning_rate": 1.4373291539967182e-06,
-      "loss": 0.086,
+      "loss": 0.0813,
       "step": 1690
     },
     {
       "epoch": 0.85,
-      "grad_norm": 0.60546875,
+      "grad_norm": 0.85546875,
       "learning_rate": 1.3484857943029572e-06,
-      "loss": 0.0815,
+      "loss": 0.0802,
       "step": 1700
     },
     {
       "epoch": 0.855,
-      "grad_norm": 0.77734375,
+      "grad_norm": 0.94140625,
       "learning_rate": 1.2622777696453482e-06,
-      "loss": 0.079,
+      "loss": 0.0778,
       "step": 1710
     },
     {
       "epoch": 0.86,
-      "grad_norm": 0.6640625,
+      "grad_norm": 0.76953125,
       "learning_rate": 1.1787313398233235e-06,
-      "loss": 0.0822,
+      "loss": 0.0864,
       "step": 1720
     },
     {
       "epoch": 0.865,
-      "grad_norm": 0.7734375,
+      "grad_norm": 0.81640625,
       "learning_rate": 1.097871953888735e-06,
-      "loss": 0.0813,
+      "loss": 0.0793,
       "step": 1730
     },
     {
       "epoch": 0.87,
-      "grad_norm": 0.67578125,
+      "grad_norm": 0.77734375,
       "learning_rate": 1.0197242423938447e-06,
-      "loss": 0.0783,
+      "loss": 0.0801,
       "step": 1740
     },
     {
       "epoch": 0.875,
-      "grad_norm": 0.75,
+      "grad_norm": 2.15625,
       "learning_rate": 9.44312009888606e-07,
-      "loss": 0.0802,
+      "loss": 0.0848,
       "step": 1750
     },
     {
       "epoch": 0.88,
-      "grad_norm": 0.6953125,
+      "grad_norm": 0.9140625,
       "learning_rate": 8.716582276695729e-07,
-      "loss": 0.0813,
+      "loss": 0.0849,
       "step": 1760
     },
     {
       "epoch": 0.885,
-      "grad_norm": 0.78515625,
+      "grad_norm": 0.84375,
       "learning_rate": 8.017850267826233e-07,
-      "loss": 0.0838,
+      "loss": 0.0766,
       "step": 1770
     },
     {
       "epoch": 0.89,
-      "grad_norm": 0.88671875,
+      "grad_norm": 0.734375,
       "learning_rate": 7.347136912816277e-07,
-      "loss": 0.0807,
+      "loss": 0.0792,
       "step": 1780
     },
     {
       "epoch": 0.895,
-      "grad_norm": 0.61328125,
+      "grad_norm": 0.9140625,
       "learning_rate": 6.704646517451108e-07,
-      "loss": 0.0747,
+      "loss": 0.083,
       "step": 1790
     },
     {
       "epoch": 0.9,
-      "grad_norm": 0.921875,
+      "grad_norm": 0.890625,
       "learning_rate": 6.090574790529091e-07,
-      "loss": 0.0794,
+      "loss": 0.0776,
       "step": 1800
     },
     {
       "epoch": 0.905,
-      "grad_norm": 0.97265625,
+      "grad_norm": 0.796875,
       "learning_rate": 5.505108784246926e-07,
-      "loss": 0.0807,
+      "loss": 0.0833,
       "step": 1810
     },
     {
       "epoch": 0.91,
-      "grad_norm": 0.76171875,
+      "grad_norm": 0.76953125,
       "learning_rate": 4.948426837221632e-07,
-      "loss": 0.0794,
+      "loss": 0.0782,
       "step": 1820
     },
     {
       "epoch": 0.915,
-      "grad_norm": 0.75,
+      "grad_norm": 0.90234375,
       "learning_rate": 4.420698520166988e-07,
-      "loss": 0.078,
+      "loss": 0.0879,
       "step": 1830
     },
     {
       "epoch": 0.92,
-      "grad_norm": 0.7265625,
+      "grad_norm": 0.69140625,
       "learning_rate": 3.922084584240582e-07,
-      "loss": 0.085,
+      "loss": 0.0805,
       "step": 1840
     },
     {
       "epoch": 0.925,
-      "grad_norm": 0.88671875,
+      "grad_norm": 0.67578125,
       "learning_rate": 3.4527369120775036e-07,
-      "loss": 0.0824,
+      "loss": 0.0764,
       "step": 1850
     },
     {
       "epoch": 0.93,
-      "grad_norm": 3.421875,
+      "grad_norm": 0.625,
       "learning_rate": 3.0127984715253246e-07,
-      "loss": 0.0845,
+      "loss": 0.0854,
       "step": 1860
     },
     {
       "epoch": 0.935,
-      "grad_norm": 0.8515625,
+      "grad_norm": 0.6640625,
       "learning_rate": 2.6024032720948446e-07,
-      "loss": 0.0776,
+      "loss": 0.078,
       "step": 1870
     },
     {
       "epoch": 0.94,
-      "grad_norm": 0.62109375,
+      "grad_norm": 0.64453125,
       "learning_rate": 2.221676324139377e-07,
-      "loss": 0.0834,
+      "loss": 0.0839,
       "step": 1880
     },
     {
       "epoch": 0.945,
-      "grad_norm": 0.80859375,
+      "grad_norm": 1.0,
       "learning_rate": 1.8707336007754873e-07,
-      "loss": 0.0818,
+      "loss": 0.0792,
       "step": 1890
     },
     {
       "epoch": 0.95,
-      "grad_norm": 0.91015625,
+      "grad_norm": 0.71484375,
       "learning_rate": 1.549682002556341e-07,
-      "loss": 0.0805,
+      "loss": 0.0806,
       "step": 1900
     },
     {
       "epoch": 0.955,
-      "grad_norm": 0.63671875,
+      "grad_norm": 1.0,
       "learning_rate": 1.2586193249088607e-07,
-      "loss": 0.0786,
+      "loss": 0.0788,
       "step": 1910
     },
     {
       "epoch": 0.96,
-      "grad_norm": 0.6328125,
+      "grad_norm": 0.53515625,
       "learning_rate": 9.976342283442464e-08,
-      "loss": 0.0834,
+      "loss": 0.0796,
       "step": 1920
     },
     {
       "epoch": 0.965,
-      "grad_norm": 0.828125,
+      "grad_norm": 0.87109375,
       "learning_rate": 7.66806211451132e-08,
-      "loss": 0.0781,
+      "loss": 0.0778,
       "step": 1930
     },
     {
       "epoch": 0.97,
-      "grad_norm": 1.1171875,
+      "grad_norm": 0.86328125,
       "learning_rate": 5.662055866795357e-08,
-      "loss": 0.0791,
+      "loss": 0.0794,
       "step": 1940
     },
     {
       "epoch": 0.975,
-      "grad_norm": 0.64453125,
+      "grad_norm": 0.9140625,
       "learning_rate": 3.9589345892304673e-08,
-      "loss": 0.0809,
+      "loss": 0.0806,
       "step": 1950
     },
     {
       "epoch": 0.98,
-      "grad_norm": 0.6640625,
+      "grad_norm": 0.74609375,
       "learning_rate": 2.5592170690560415e-08,
-      "loss": 0.0829,
+      "loss": 0.0813,
       "step": 1960
     },
     {
       "epoch": 0.985,
-      "grad_norm": 1.09375,
+      "grad_norm": 0.796875,
       "learning_rate": 1.4633296737882607e-08,
-      "loss": 0.0795,
+      "loss": 0.0911,
       "step": 1970
     },
     {
       "epoch": 0.99,
-      "grad_norm": 0.82421875,
+      "grad_norm": 0.83984375,
       "learning_rate": 6.716062213437679e-09,
-      "loss": 0.084,
+      "loss": 0.0816,
       "step": 1980
     },
     {
       "epoch": 0.995,
-      "grad_norm": 1.265625,
+      "grad_norm": 0.76171875,
       "learning_rate": 1.8428787835578222e-09,
-      "loss": 0.0812,
+      "loss": 0.0854,
       "step": 1990
     },
     {
       "epoch": 1.0,
-      "grad_norm": 0.88671875,
+      "grad_norm": 0.62890625,
       "learning_rate": 1.5230867123072757e-11,
-      "loss": 0.0848,
+      "loss": 0.0819,
       "step": 2000
     }
   ],

checkpoint-2000/training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:71e2eadc0cbabb87821fe5a0534c3c9870b7ebcb958c2a67d3ea28671647481e
+oid sha256:f675fbf3c5fe21c225b007d45d281c9507dea4441d272bdc86973c61472d11ec
 size 5841