diff --git "a/finetuned_codet5base_512/trainer_state.json" "b/finetuned_codet5base_512/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/finetuned_codet5base_512/trainer_state.json"
@@ -0,0 +1,27016 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.7911735506579634,
+  "global_step": 675000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "learning_rate": 3.999545168410774e-05,
+      "loss": 2.8381,
+      "step": 200
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 3.9986073713195866e-05,
+      "loss": 1.6065,
+      "step": 400
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 3.997669574228399e-05,
+      "loss": 1.5159,
+      "step": 600
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 3.996731777137211e-05,
+      "loss": 1.4606,
+      "step": 800
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 3.995793980046023e-05,
+      "loss": 1.4182,
+      "step": 1000
+    },
+    {
+      "epoch": 0.0,
+      "eval_em_first_line_ratio": 0.3847,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.2805,
+      "eval_loss": 1.2461129426956177,
+      "eval_runtime": 289.7766,
+      "eval_samples_per_second": 34.509,
+      "eval_steps_per_second": 0.542,
+      "step": 1000
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 3.994856182954835e-05,
+      "loss": 1.3826,
+      "step": 1200
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 3.993918385863647e-05,
+      "loss": 1.3565,
+      "step": 1400
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 3.992980588772459e-05,
+      "loss": 1.344,
+      "step": 1600
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 3.992042791681271e-05,
+      "loss": 1.3241,
+      "step": 1800
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 3.9911049945900834e-05,
+      "loss": 1.3084,
+      "step": 2000
+    },
+    {
+      "epoch": 0.0,
+      "eval_em_first_line_ratio": 0.3952,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.2899,
+      "eval_loss": 1.1736323833465576,
+      "eval_runtime": 281.1898,
+      "eval_samples_per_second": 35.563,
+      "eval_steps_per_second": 0.558,
+      "step": 2000
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 3.9901671974988955e-05,
+      "loss": 1.2913,
+      "step": 2200
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 3.9892294004077076e-05,
+      "loss": 1.286,
+      "step": 2400
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 3.98829160331652e-05,
+      "loss": 1.2663,
+      "step": 2600
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 3.987353806225332e-05,
+      "loss": 1.2613,
+      "step": 2800
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 3.986416009134144e-05,
+      "loss": 1.2602,
+      "step": 3000
+    },
+    {
+      "epoch": 0.0,
+      "eval_em_first_line_ratio": 0.4105,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.301,
+      "eval_loss": 1.1301541328430176,
+      "eval_runtime": 277.3208,
+      "eval_samples_per_second": 36.059,
+      "eval_steps_per_second": 0.566,
+      "step": 3000
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 3.985478212042956e-05,
+      "loss": 1.2502,
+      "step": 3200
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 3.984540414951768e-05,
+      "loss": 1.2511,
+      "step": 3400
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 3.98360261786058e-05,
+      "loss": 1.2305,
+      "step": 3600
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 3.982664820769392e-05,
+      "loss": 1.2305,
+      "step": 3800
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 3.981727023678205e-05,
+      "loss": 1.2121,
+      "step": 4000
+    },
+    {
+      "epoch": 0.0,
+      "eval_em_first_line_ratio": 0.4125,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3031,
+      "eval_loss": 1.1024450063705444,
+      "eval_runtime": 261.9327,
+      "eval_samples_per_second": 38.178,
+      "eval_steps_per_second": 0.599,
+      "step": 4000
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 3.9807892265870164e-05,
+      "loss": 1.2163,
+      "step": 4200
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.979851429495829e-05,
+      "loss": 1.2053,
+      "step": 4400
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.9789136324046406e-05,
+      "loss": 1.196,
+      "step": 4600
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.9779758353134534e-05,
+      "loss": 1.1985,
+      "step": 4800
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.977038038222265e-05,
+      "loss": 1.1953,
+      "step": 5000
+    },
+    {
+      "epoch": 0.01,
+      "eval_em_first_line_ratio": 0.4188,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3093,
+      "eval_loss": 1.081965446472168,
+      "eval_runtime": 266.2694,
+      "eval_samples_per_second": 37.556,
+      "eval_steps_per_second": 0.59,
+      "step": 5000
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.9761002411310776e-05,
+      "loss": 1.1883,
+      "step": 5200
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.97516244403989e-05,
+      "loss": 1.1849,
+      "step": 5400
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.974224646948702e-05,
+      "loss": 1.1817,
+      "step": 5600
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.973286849857514e-05,
+      "loss": 1.1709,
+      "step": 5800
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.972349052766326e-05,
+      "loss": 1.1774,
+      "step": 6000
+    },
+    {
+      "epoch": 0.01,
+      "eval_em_first_line_ratio": 0.4202,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3103,
+      "eval_loss": 1.0661252737045288,
+      "eval_runtime": 277.8519,
+      "eval_samples_per_second": 35.99,
+      "eval_steps_per_second": 0.565,
+      "step": 6000
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.971411255675138e-05,
+      "loss": 1.1636,
+      "step": 6200
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.97047345858395e-05,
+      "loss": 1.1654,
+      "step": 6400
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.969535661492762e-05,
+      "loss": 1.1618,
+      "step": 6600
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.9685978644015744e-05,
+      "loss": 1.156,
+      "step": 6800
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.9676600673103865e-05,
+      "loss": 1.1563,
+      "step": 7000
+    },
+    {
+      "epoch": 0.01,
+      "eval_em_first_line_ratio": 0.4232,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3133,
+      "eval_loss": 1.052996039390564,
+      "eval_runtime": 276.6284,
+      "eval_samples_per_second": 36.15,
+      "eval_steps_per_second": 0.568,
+      "step": 7000
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.9667222702191986e-05,
+      "loss": 1.1475,
+      "step": 7200
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.965784473128011e-05,
+      "loss": 1.1462,
+      "step": 7400
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.964846676036823e-05,
+      "loss": 1.1454,
+      "step": 7600
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.9639088789456355e-05,
+      "loss": 1.1439,
+      "step": 7800
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.962971081854447e-05,
+      "loss": 1.1392,
+      "step": 8000
+    },
+    {
+      "epoch": 0.01,
+      "eval_em_first_line_ratio": 0.4245,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3148,
+      "eval_loss": 1.0396713018417358,
+      "eval_runtime": 273.2538,
+      "eval_samples_per_second": 36.596,
+      "eval_steps_per_second": 0.575,
+      "step": 8000
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.96203328476326e-05,
+      "loss": 1.1345,
+      "step": 8200
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.961095487672071e-05,
+      "loss": 1.1337,
+      "step": 8400
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.9601623795663394e-05,
+      "loss": 1.1328,
+      "step": 8600
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.9592245824751515e-05,
+      "loss": 1.121,
+      "step": 8800
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.9582867853839636e-05,
+      "loss": 1.1242,
+      "step": 9000
+    },
+    {
+      "epoch": 0.01,
+      "eval_em_first_line_ratio": 0.4298,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3193,
+      "eval_loss": 1.0310505628585815,
+      "eval_runtime": 263.1681,
+      "eval_samples_per_second": 37.999,
+      "eval_steps_per_second": 0.597,
+      "step": 9000
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.957348988292776e-05,
+      "loss": 1.1289,
+      "step": 9200
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.956411191201588e-05,
+      "loss": 1.1208,
+      "step": 9400
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.9554733941104e-05,
+      "loss": 1.1187,
+      "step": 9600
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.954535597019212e-05,
+      "loss": 1.1162,
+      "step": 9800
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.953597799928024e-05,
+      "loss": 1.1136,
+      "step": 10000
+    },
+    {
+      "epoch": 0.01,
+      "eval_em_first_line_ratio": 0.429,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3192,
+      "eval_loss": 1.0203908681869507,
+      "eval_runtime": 253.3716,
+      "eval_samples_per_second": 39.468,
+      "eval_steps_per_second": 0.62,
+      "step": 10000
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.952660002836837e-05,
+      "loss": 1.1127,
+      "step": 10200
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.951722205745648e-05,
+      "loss": 1.1052,
+      "step": 10400
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.950784408654461e-05,
+      "loss": 1.1096,
+      "step": 10600
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.9498466115632725e-05,
+      "loss": 1.1015,
+      "step": 10800
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.948908814472085e-05,
+      "loss": 1.1056,
+      "step": 11000
+    },
+    {
+      "epoch": 0.01,
+      "eval_em_first_line_ratio": 0.4335,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3235,
+      "eval_loss": 1.012048602104187,
+      "eval_runtime": 230.2445,
+      "eval_samples_per_second": 43.432,
+      "eval_steps_per_second": 0.682,
+      "step": 11000
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.9479757063663535e-05,
+      "loss": 1.0968,
+      "step": 11200
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.947037909275165e-05,
+      "loss": 1.0949,
+      "step": 11400
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.946100112183978e-05,
+      "loss": 1.0961,
+      "step": 11600
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.945162315092789e-05,
+      "loss": 1.1008,
+      "step": 11800
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.944224518001602e-05,
+      "loss": 1.0912,
+      "step": 12000
+    },
+    {
+      "epoch": 0.01,
+      "eval_em_first_line_ratio": 0.4337,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3232,
+      "eval_loss": 1.0061216354370117,
+      "eval_runtime": 231.8151,
+      "eval_samples_per_second": 43.138,
+      "eval_steps_per_second": 0.677,
+      "step": 12000
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.943286720910413e-05,
+      "loss": 1.0787,
+      "step": 12200
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.942348923819226e-05,
+      "loss": 1.0919,
+      "step": 12400
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.941411126728038e-05,
+      "loss": 1.0913,
+      "step": 12600
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.94047332963685e-05,
+      "loss": 1.089,
+      "step": 12800
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.9395355325456624e-05,
+      "loss": 1.0846,
+      "step": 13000
+    },
+    {
+      "epoch": 0.02,
+      "eval_em_first_line_ratio": 0.4338,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3237,
+      "eval_loss": 0.9972391724586487,
+      "eval_runtime": 233.6003,
+      "eval_samples_per_second": 42.808,
+      "eval_steps_per_second": 0.672,
+      "step": 13000
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.9385977354544745e-05,
+      "loss": 1.0834,
+      "step": 13200
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.9376599383632866e-05,
+      "loss": 1.0765,
+      "step": 13400
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.936722141272099e-05,
+      "loss": 1.0757,
+      "step": 13600
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.935784344180911e-05,
+      "loss": 1.072,
+      "step": 13800
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.934846547089723e-05,
+      "loss": 1.0743,
+      "step": 14000
+    },
+    {
+      "epoch": 0.02,
+      "eval_em_first_line_ratio": 0.4346,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3245,
+      "eval_loss": 0.9912696480751038,
+      "eval_runtime": 235.4529,
+      "eval_samples_per_second": 42.471,
+      "eval_steps_per_second": 0.667,
+      "step": 14000
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.933908749998535e-05,
+      "loss": 1.0703,
+      "step": 14200
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.932970952907347e-05,
+      "loss": 1.0764,
+      "step": 14400
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.932033155816159e-05,
+      "loss": 1.07,
+      "step": 14600
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.931095358724971e-05,
+      "loss": 1.0732,
+      "step": 14800
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.930157561633784e-05,
+      "loss": 1.066,
+      "step": 15000
+    },
+    {
+      "epoch": 0.02,
+      "eval_em_first_line_ratio": 0.4366,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3267,
+      "eval_loss": 0.9854965806007385,
+      "eval_runtime": 246.7833,
+      "eval_samples_per_second": 40.521,
+      "eval_steps_per_second": 0.636,
+      "step": 15000
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.9292197645425955e-05,
+      "loss": 1.0726,
+      "step": 15200
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.928286656436864e-05,
+      "loss": 1.0647,
+      "step": 15400
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.927353548331131e-05,
+      "loss": 1.0649,
+      "step": 15600
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.926415751239944e-05,
+      "loss": 1.0615,
+      "step": 15800
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.925477954148756e-05,
+      "loss": 1.065,
+      "step": 16000
+    },
+    {
+      "epoch": 0.02,
+      "eval_em_first_line_ratio": 0.4369,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3267,
+      "eval_loss": 0.9802919030189514,
+      "eval_runtime": 246.4072,
+      "eval_samples_per_second": 40.583,
+      "eval_steps_per_second": 0.637,
+      "step": 16000
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.924540157057568e-05,
+      "loss": 1.0631,
+      "step": 16200
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.9236023599663803e-05,
+      "loss": 1.0592,
+      "step": 16400
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.9226645628751924e-05,
+      "loss": 1.0613,
+      "step": 16600
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.9217267657840045e-05,
+      "loss": 1.06,
+      "step": 16800
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.9207889686928166e-05,
+      "loss": 1.055,
+      "step": 17000
+    },
+    {
+      "epoch": 0.02,
+      "eval_em_first_line_ratio": 0.4393,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3291,
+      "eval_loss": 0.9742295145988464,
+      "eval_runtime": 245.5441,
+      "eval_samples_per_second": 40.726,
+      "eval_steps_per_second": 0.639,
+      "step": 17000
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.919851171601629e-05,
+      "loss": 1.0558,
+      "step": 17200
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.918913374510441e-05,
+      "loss": 1.0532,
+      "step": 17400
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.917975577419253e-05,
+      "loss": 1.056,
+      "step": 17600
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.917037780328065e-05,
+      "loss": 1.0521,
+      "step": 17800
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.916099983236877e-05,
+      "loss": 1.0539,
+      "step": 18000
+    },
+    {
+      "epoch": 0.02,
+      "eval_em_first_line_ratio": 0.4392,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3287,
+      "eval_loss": 0.9696520566940308,
+      "eval_runtime": 249.1142,
+      "eval_samples_per_second": 40.142,
+      "eval_steps_per_second": 0.63,
+      "step": 18000
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.915162186145689e-05,
+      "loss": 1.047,
+      "step": 18200
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.914224389054502e-05,
+      "loss": 1.0446,
+      "step": 18400
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.9132865919633134e-05,
+      "loss": 1.0409,
+      "step": 18600
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.912348794872126e-05,
+      "loss": 1.0536,
+      "step": 18800
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.9114109977809376e-05,
+      "loss": 1.0439,
+      "step": 19000
+    },
+    {
+      "epoch": 0.02,
+      "eval_em_first_line_ratio": 0.4387,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3286,
+      "eval_loss": 0.9655564427375793,
+      "eval_runtime": 248.103,
+      "eval_samples_per_second": 40.306,
+      "eval_steps_per_second": 0.633,
+      "step": 19000
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.9104732006897504e-05,
+      "loss": 1.0412,
+      "step": 19200
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.9095354035985625e-05,
+      "loss": 1.0435,
+      "step": 19400
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.9085976065073746e-05,
+      "loss": 1.0403,
+      "step": 19600
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.907659809416187e-05,
+      "loss": 1.0341,
+      "step": 19800
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.906722012324999e-05,
+      "loss": 1.0333,
+      "step": 20000
+    },
+    {
+      "epoch": 0.02,
+      "eval_em_first_line_ratio": 0.4428,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3319,
+      "eval_loss": 0.9613921642303467,
+      "eval_runtime": 241.1533,
+      "eval_samples_per_second": 41.467,
+      "eval_steps_per_second": 0.651,
+      "step": 20000
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.905788904219267e-05,
+      "loss": 1.0363,
+      "step": 20200
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.9048511071280784e-05,
+      "loss": 1.0426,
+      "step": 20400
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.903913310036891e-05,
+      "loss": 1.0389,
+      "step": 20600
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.902975512945703e-05,
+      "loss": 1.0386,
+      "step": 20800
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.902037715854515e-05,
+      "loss": 1.0317,
+      "step": 21000
+    },
+    {
+      "epoch": 0.02,
+      "eval_em_first_line_ratio": 0.4427,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3319,
+      "eval_loss": 0.956505537033081,
+      "eval_runtime": 240.6061,
+      "eval_samples_per_second": 41.562,
+      "eval_steps_per_second": 0.653,
+      "step": 21000
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.9010999187633275e-05,
+      "loss": 1.0345,
+      "step": 21200
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.900162121672139e-05,
+      "loss": 1.0349,
+      "step": 21400
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.899224324580952e-05,
+      "loss": 1.0295,
+      "step": 21600
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.89829121647522e-05,
+      "loss": 1.0316,
+      "step": 21800
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.8973534193840314e-05,
+      "loss": 1.0312,
+      "step": 22000
+    },
+    {
+      "epoch": 0.03,
+      "eval_em_first_line_ratio": 0.4407,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3307,
+      "eval_loss": 0.9542904496192932,
+      "eval_runtime": 247.6476,
+      "eval_samples_per_second": 40.38,
+      "eval_steps_per_second": 0.634,
+      "step": 22000
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.896415622292844e-05,
+      "loss": 1.0266,
+      "step": 22200
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.8954778252016556e-05,
+      "loss": 1.0344,
+      "step": 22400
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.8945400281104683e-05,
+      "loss": 1.0264,
+      "step": 22600
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.8936022310192804e-05,
+      "loss": 1.026,
+      "step": 22800
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.8926644339280925e-05,
+      "loss": 1.0232,
+      "step": 23000
+    },
+    {
+      "epoch": 0.03,
+      "eval_em_first_line_ratio": 0.4423,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3319,
+      "eval_loss": 0.9504188299179077,
+      "eval_runtime": 246.5153,
+      "eval_samples_per_second": 40.565,
+      "eval_steps_per_second": 0.637,
+      "step": 23000
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.8917266368369046e-05,
+      "loss": 1.0222,
+      "step": 23200
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.890788839745717e-05,
+      "loss": 1.0246,
+      "step": 23400
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.889851042654529e-05,
+      "loss": 1.0228,
+      "step": 23600
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.888913245563341e-05,
+      "loss": 1.0256,
+      "step": 23800
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.887975448472153e-05,
+      "loss": 1.0254,
+      "step": 24000
+    },
+    {
+      "epoch": 0.03,
+      "eval_em_first_line_ratio": 0.4404,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3303,
+      "eval_loss": 0.9470880627632141,
+      "eval_runtime": 243.8658,
+      "eval_samples_per_second": 41.006,
+      "eval_steps_per_second": 0.644,
+      "step": 24000
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.887042340366421e-05,
+      "loss": 1.0214,
+      "step": 24200
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.8861045432752334e-05,
+      "loss": 1.0183,
+      "step": 24400
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.8851667461840455e-05,
+      "loss": 1.0186,
+      "step": 24600
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.8842289490928576e-05,
+      "loss": 1.0183,
+      "step": 24800
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.88329115200167e-05,
+      "loss": 1.0129,
+      "step": 25000
+    },
+    {
+      "epoch": 0.03,
+      "eval_em_first_line_ratio": 0.4459,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3349,
+      "eval_loss": 0.9437031149864197,
+      "eval_runtime": 242.6988,
+      "eval_samples_per_second": 41.203,
+      "eval_steps_per_second": 0.647,
+      "step": 25000
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.882353354910482e-05,
+      "loss": 1.0146,
+      "step": 25200
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.881415557819294e-05,
+      "loss": 1.0148,
+      "step": 25400
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.880477760728106e-05,
+      "loss": 1.0105,
+      "step": 25600
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.879539963636918e-05,
+      "loss": 1.016,
+      "step": 25800
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.87860216654573e-05,
+      "loss": 1.0114,
+      "step": 26000
+    },
+    {
+      "epoch": 0.03,
+      "eval_em_first_line_ratio": 0.4472,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3362,
+      "eval_loss": 0.9406141638755798,
+      "eval_runtime": 232.8085,
+      "eval_samples_per_second": 42.954,
+      "eval_steps_per_second": 0.674,
+      "step": 26000
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.877664369454542e-05,
+      "loss": 1.0118,
+      "step": 26200
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.8767265723633543e-05,
+      "loss": 1.0123,
+      "step": 26400
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.875788775272167e-05,
+      "loss": 1.0048,
+      "step": 26600
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.8748509781809785e-05,
+      "loss": 1.0097,
+      "step": 26800
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.8739131810897906e-05,
+      "loss": 1.0091,
+      "step": 27000
+    },
+    {
+      "epoch": 0.03,
+      "eval_em_first_line_ratio": 0.4433,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.333,
+      "eval_loss": 0.9355378150939941,
+      "eval_runtime": 243.7522,
+      "eval_samples_per_second": 41.025,
+      "eval_steps_per_second": 0.644,
+      "step": 27000
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.872975383998603e-05,
+      "loss": 1.0025,
+      "step": 27200
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.872037586907415e-05,
+      "loss": 1.0037,
+      "step": 27400
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.8710997898162276e-05,
+      "loss": 1.0072,
+      "step": 27600
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.870161992725039e-05,
+      "loss": 1.0114,
+      "step": 27800
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.869224195633852e-05,
+      "loss": 1.0042,
+      "step": 28000
+    },
+    {
+      "epoch": 0.03,
+      "eval_em_first_line_ratio": 0.448,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3371,
+      "eval_loss": 0.9330340623855591,
+      "eval_runtime": 247.5786,
+      "eval_samples_per_second": 40.391,
+      "eval_steps_per_second": 0.634,
+      "step": 28000
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.868286398542663e-05,
+      "loss": 1.0122,
+      "step": 28200
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.867348601451476e-05,
+      "loss": 1.0051,
+      "step": 28400
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.8664108043602874e-05,
+      "loss": 1.0043,
+      "step": 28600
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.8654730072691e-05,
+      "loss": 1.0044,
+      "step": 28800
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.864535210177912e-05,
+      "loss": 0.9959,
+      "step": 29000
+    },
+    {
+      "epoch": 0.03,
+      "eval_em_first_line_ratio": 0.4476,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3366,
+      "eval_loss": 0.9307966828346252,
+      "eval_runtime": 243.2993,
+      "eval_samples_per_second": 41.102,
+      "eval_steps_per_second": 0.645,
+      "step": 29000
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.8635974130867244e-05,
+      "loss": 0.9998,
+      "step": 29200
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.8626596159955365e-05,
+      "loss": 1.001,
+      "step": 29400
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.8617218189043486e-05,
+      "loss": 1.0008,
+      "step": 29600
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.860788710798617e-05,
+      "loss": 1.0007,
+      "step": 29800
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 3.859850913707429e-05,
+      "loss": 1.0012,
+      "step": 30000
+    },
+    {
+      "epoch": 0.04,
+      "eval_em_first_line_ratio": 0.4485,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3376,
+      "eval_loss": 0.9280819296836853,
+      "eval_runtime": 239.9875,
+      "eval_samples_per_second": 41.669,
+      "eval_steps_per_second": 0.654,
+      "step": 30000
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 3.858913116616241e-05,
+      "loss": 0.9968,
+      "step": 30200
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 3.857975319525053e-05,
+      "loss": 0.9967,
+      "step": 30400
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 3.857037522433865e-05,
+      "loss": 0.9947,
+      "step": 30600
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 3.856099725342677e-05,
+      "loss": 0.9977,
+      "step": 30800
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 3.8551619282514894e-05,
+      "loss": 0.999,
+      "step": 31000
+    },
+    {
+      "epoch": 0.04,
+      "eval_em_first_line_ratio": 0.4477,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3372,
+      "eval_loss": 0.9261435270309448,
+      "eval_runtime": 241.6078,
+      "eval_samples_per_second": 41.389,
+      "eval_steps_per_second": 0.65,
+      "step": 31000
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 3.8542241311603015e-05,
+      "loss": 0.9989,
+      "step": 31200
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 3.8532863340691136e-05,
+      "loss": 0.9975,
+      "step": 31400
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 3.852348536977926e-05,
+      "loss": 0.9961,
+      "step": 31600
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 3.851410739886738e-05,
+      "loss": 0.9987,
+      "step": 31800
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 3.85047294279555e-05,
+      "loss": 0.9914,
+      "step": 32000
+    },
+    {
+      "epoch": 0.04,
+      "eval_em_first_line_ratio": 0.4489,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3381,
+      "eval_loss": 0.9217458963394165,
+      "eval_runtime": 271.1469,
+      "eval_samples_per_second": 36.88,
+      "eval_steps_per_second": 0.579,
+      "step": 32000
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 3.849535145704362e-05,
+      "loss": 0.9951,
+      "step": 32200
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 3.848597348613175e-05,
+      "loss": 0.9921,
+      "step": 32400
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 3.847659551521986e-05,
+      "loss": 0.9952,
+      "step": 32600
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 3.8467264434162544e-05,
+      "loss": 0.9934,
+      "step": 32800
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 3.8457886463250665e-05,
+      "loss": 0.9875,
+      "step": 33000
+    },
+    {
+      "epoch": 0.04,
+      "eval_em_first_line_ratio": 0.4494,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3387,
+      "eval_loss": 0.9223683476448059,
+      "eval_runtime": 274.817,
+      "eval_samples_per_second": 36.388,
+      "eval_steps_per_second": 0.571,
+      "step": 33000
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 3.8448508492338786e-05,
+      "loss": 0.9854,
+      "step": 33200
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 3.843913052142691e-05,
+      "loss": 0.9887,
+      "step": 33400
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 3.842975255051503e-05,
+      "loss": 0.9893,
+      "step": 33600
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 3.842042146945771e-05,
+      "loss": 0.9881,
+      "step": 33800
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 3.8411090388400387e-05,
+      "loss": 0.9842,
+      "step": 34000
+    },
+    {
+      "epoch": 0.04,
+      "eval_em_first_line_ratio": 0.449,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3374,
+      "eval_loss": 0.9189767837524414,
+      "eval_runtime": 273.6261,
+      "eval_samples_per_second": 36.546,
+      "eval_steps_per_second": 0.574,
+      "step": 34000
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 3.8401712417488514e-05,
+      "loss": 0.985,
+      "step": 34200
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 3.8392334446576635e-05,
+      "loss": 0.9878,
+      "step": 34400
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 3.8382956475664756e-05,
+      "loss": 0.985,
+      "step": 34600
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 3.837357850475288e-05,
+      "loss": 0.9809,
+      "step": 34800
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 3.8364200533841e-05,
+      "loss": 0.9794,
+      "step": 35000
+    },
+    {
+      "epoch": 0.04,
+      "eval_em_first_line_ratio": 0.4506,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3392,
+      "eval_loss": 0.9168775677680969,
+      "eval_runtime": 264.5542,
+      "eval_samples_per_second": 37.799,
+      "eval_steps_per_second": 0.593,
+      "step": 35000
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 3.835482256292912e-05,
+      "loss": 0.9808,
+      "step": 35200
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 3.834544459201724e-05,
+      "loss": 0.9812,
+      "step": 35400
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 3.833606662110536e-05,
+      "loss": 0.9842,
+      "step": 35600
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 3.832668865019348e-05,
+      "loss": 0.9818,
+      "step": 35800
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 3.83173106792816e-05,
+      "loss": 0.9833,
+      "step": 36000
+    },
+    {
+      "epoch": 0.04,
+      "eval_em_first_line_ratio": 0.4482,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3366,
+      "eval_loss": 0.9132855534553528,
+      "eval_runtime": 265.1195,
+      "eval_samples_per_second": 37.719,
+      "eval_steps_per_second": 0.592,
+      "step": 36000
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 3.8307932708369724e-05,
+      "loss": 0.9848,
+      "step": 36200
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 3.8298554737457845e-05,
+      "loss": 0.9824,
+      "step": 36400
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 3.8289176766545966e-05,
+      "loss": 0.9812,
+      "step": 36600
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 3.827979879563409e-05,
+      "loss": 0.9749,
+      "step": 36800
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 3.827042082472221e-05,
+      "loss": 0.9829,
+      "step": 37000
+    },
+    {
+      "epoch": 0.04,
+      "eval_em_first_line_ratio": 0.4497,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3385,
+      "eval_loss": 0.9126341938972473,
+      "eval_runtime": 272.4296,
+      "eval_samples_per_second": 36.707,
+      "eval_steps_per_second": 0.576,
+      "step": 37000
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 3.8261042853810336e-05,
+      "loss": 0.9794,
+      "step": 37200
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 3.825166488289845e-05,
+      "loss": 0.9776,
+      "step": 37400
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 3.824228691198658e-05,
+      "loss": 0.9787,
+      "step": 37600
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 3.823290894107469e-05,
+      "loss": 0.9764,
+      "step": 37800
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 3.822353097016282e-05,
+      "loss": 0.9727,
+      "step": 38000
+    },
+    {
+      "epoch": 0.04,
+      "eval_em_first_line_ratio": 0.4498,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3393,
+      "eval_loss": 0.9100714921951294,
+      "eval_runtime": 270.0662,
+      "eval_samples_per_second": 37.028,
+      "eval_steps_per_second": 0.581,
+      "step": 38000
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 3.821415299925094e-05,
+      "loss": 0.9752,
+      "step": 38200
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 3.820477502833906e-05,
+      "loss": 0.9802,
+      "step": 38400
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 3.819539705742718e-05,
+      "loss": 0.977,
+      "step": 38600
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 3.818606597636986e-05,
+      "loss": 0.9782,
+      "step": 38800
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 3.817668800545798e-05,
+      "loss": 0.9764,
+      "step": 39000
+    },
+    {
+      "epoch": 0.05,
+      "eval_em_first_line_ratio": 0.4521,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3408,
+      "eval_loss": 0.9077931642532349,
+      "eval_runtime": 276.0084,
+      "eval_samples_per_second": 36.231,
+      "eval_steps_per_second": 0.569,
+      "step": 39000
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 3.816731003454611e-05,
+      "loss": 0.9707,
+      "step": 39200
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 3.815793206363422e-05,
+      "loss": 0.9704,
+      "step": 39400
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 3.814855409272235e-05,
+      "loss": 0.9679,
+      "step": 39600
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 3.813917612181046e-05,
+      "loss": 0.9691,
+      "step": 39800
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 3.812979815089859e-05,
+      "loss": 0.9735,
+      "step": 40000
+    },
+    {
+      "epoch": 0.05,
+      "eval_em_first_line_ratio": 0.4503,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3391,
+      "eval_loss": 0.9063853621482849,
+      "eval_runtime": 276.5324,
+      "eval_samples_per_second": 36.162,
+      "eval_steps_per_second": 0.568,
+      "step": 40000
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 3.8120420179986705e-05,
+      "loss": 0.9711,
+      "step": 40200
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 3.811104220907483e-05,
+      "loss": 0.9717,
+      "step": 40400
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 3.8101664238162954e-05,
+      "loss": 0.9648,
+      "step": 40600
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 3.8092286267251075e-05,
+      "loss": 0.97,
+      "step": 40800
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 3.8082908296339196e-05,
+      "loss": 0.9707,
+      "step": 41000
+    },
+    {
+      "epoch": 0.05,
+      "eval_em_first_line_ratio": 0.4537,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.342,
+      "eval_loss": 0.9037744402885437,
+      "eval_runtime": 259.697,
+      "eval_samples_per_second": 38.506,
+      "eval_steps_per_second": 0.605,
+      "step": 41000
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 3.807353032542732e-05,
+      "loss": 0.9685,
+      "step": 41200
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 3.806415235451544e-05,
+      "loss": 0.9679,
+      "step": 41400
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 3.805477438360356e-05,
+      "loss": 0.9697,
+      "step": 41600
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 3.804539641269168e-05,
+      "loss": 0.9719,
+      "step": 41800
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 3.80360184417798e-05,
+      "loss": 0.9644,
+      "step": 42000
+    },
+    {
+      "epoch": 0.05,
+      "eval_em_first_line_ratio": 0.4522,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3416,
+      "eval_loss": 0.9013987183570862,
+      "eval_runtime": 228.9242,
+      "eval_samples_per_second": 43.683,
+      "eval_steps_per_second": 0.686,
+      "step": 42000
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 3.802664047086792e-05,
+      "loss": 0.9726,
+      "step": 42200
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 3.801726249995604e-05,
+      "loss": 0.9675,
+      "step": 42400
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 3.8007884529044163e-05,
+      "loss": 0.9729,
+      "step": 42600
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 3.7998506558132284e-05,
+      "loss": 0.967,
+      "step": 42800
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 3.798917547707497e-05,
+      "loss": 0.9609,
+      "step": 43000
+    },
+    {
+      "epoch": 0.05,
+      "eval_em_first_line_ratio": 0.4522,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3404,
+      "eval_loss": 0.8999049067497253,
+      "eval_runtime": 222.6871,
+      "eval_samples_per_second": 44.906,
+      "eval_steps_per_second": 0.705,
+      "step": 43000
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 3.797979750616309e-05,
+      "loss": 0.9646,
+      "step": 43200
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 3.797041953525121e-05,
+      "loss": 0.9647,
+      "step": 43400
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 3.796104156433933e-05,
+      "loss": 0.9617,
+      "step": 43600
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 3.795166359342745e-05,
+      "loss": 0.9637,
+      "step": 43800
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 3.794228562251558e-05,
+      "loss": 0.9611,
+      "step": 44000
+    },
+    {
+      "epoch": 0.05,
+      "eval_em_first_line_ratio": 0.453,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3423,
+      "eval_loss": 0.8978204131126404,
+      "eval_runtime": 260.5652,
+      "eval_samples_per_second": 38.378,
+      "eval_steps_per_second": 0.603,
+      "step": 44000
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 3.793290765160369e-05,
+      "loss": 0.963,
+      "step": 44200
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 3.792352968069182e-05,
+      "loss": 0.9626,
+      "step": 44400
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 3.7914151709779935e-05,
+      "loss": 0.9542,
+      "step": 44600
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 3.790477373886806e-05,
+      "loss": 0.9518,
+      "step": 44800
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 3.789539576795618e-05,
+      "loss": 0.964,
+      "step": 45000
+    },
+    {
+      "epoch": 0.05,
+      "eval_em_first_line_ratio": 0.4537,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.343,
+      "eval_loss": 0.8964654207229614,
+      "eval_runtime": 266.9444,
+      "eval_samples_per_second": 37.461,
+      "eval_steps_per_second": 0.588,
+      "step": 45000
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 3.78860177970443e-05,
+      "loss": 0.9624,
+      "step": 45200
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 3.7876639826132425e-05,
+      "loss": 0.9622,
+      "step": 45400
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 3.786726185522054e-05,
+      "loss": 0.9533,
+      "step": 45600
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 3.785788388430867e-05,
+      "loss": 0.9541,
+      "step": 45800
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 3.784850591339678e-05,
+      "loss": 0.9578,
+      "step": 46000
+    },
+    {
+      "epoch": 0.05,
+      "eval_em_first_line_ratio": 0.4541,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3432,
+      "eval_loss": 0.8948951363563538,
+      "eval_runtime": 267.8691,
+      "eval_samples_per_second": 37.332,
+      "eval_steps_per_second": 0.586,
+      "step": 46000
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 3.783912794248491e-05,
+      "loss": 0.9595,
+      "step": 46200
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 3.782974997157303e-05,
+      "loss": 0.9607,
+      "step": 46400
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 3.782037200066115e-05,
+      "loss": 0.959,
+      "step": 46600
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 3.781099402974927e-05,
+      "loss": 0.9529,
+      "step": 46800
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 3.780161605883739e-05,
+      "loss": 0.9656,
+      "step": 47000
+    },
+    {
+      "epoch": 0.06,
+      "eval_em_first_line_ratio": 0.4557,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3442,
+      "eval_loss": 0.8929873108863831,
+      "eval_runtime": 253.7854,
+      "eval_samples_per_second": 39.403,
+      "eval_steps_per_second": 0.619,
+      "step": 47000
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 3.7792238087925514e-05,
+      "loss": 0.9538,
+      "step": 47200
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 3.7782860117013635e-05,
+      "loss": 0.9537,
+      "step": 47400
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 3.7773482146101756e-05,
+      "loss": 0.9528,
+      "step": 47600
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 3.776410417518988e-05,
+      "loss": 0.9575,
+      "step": 47800
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 3.775477309413256e-05,
+      "loss": 0.951,
+      "step": 48000
+    },
+    {
+      "epoch": 0.06,
+      "eval_em_first_line_ratio": 0.4536,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3424,
+      "eval_loss": 0.890616774559021,
+      "eval_runtime": 273.3585,
+      "eval_samples_per_second": 36.582,
+      "eval_steps_per_second": 0.574,
+      "step": 48000
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 3.774539512322068e-05,
+      "loss": 0.9519,
+      "step": 48200
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 3.77360171523088e-05,
+      "loss": 0.9543,
+      "step": 48400
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 3.772663918139692e-05,
+      "loss": 0.9554,
+      "step": 48600
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 3.7717261210485043e-05,
+      "loss": 0.9549,
+      "step": 48800
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 3.7707930129427726e-05,
+      "loss": 0.9517,
+      "step": 49000
+    },
+    {
+      "epoch": 0.06,
+      "eval_em_first_line_ratio": 0.4552,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3442,
+      "eval_loss": 0.8898947238922119,
+      "eval_runtime": 252.1225,
+      "eval_samples_per_second": 39.663,
+      "eval_steps_per_second": 0.623,
+      "step": 49000
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 3.769855215851585e-05,
+      "loss": 0.9511,
+      "step": 49200
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 3.768917418760397e-05,
+      "loss": 0.9525,
+      "step": 49400
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 3.767984310654665e-05,
+      "loss": 0.9501,
+      "step": 49600
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 3.767046513563477e-05,
+      "loss": 0.952,
+      "step": 49800
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 3.766108716472289e-05,
+      "loss": 0.9506,
+      "step": 50000
+    },
+    {
+      "epoch": 0.06,
+      "eval_em_first_line_ratio": 0.4555,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3448,
+      "eval_loss": 0.8868198990821838,
+      "eval_runtime": 259.4953,
+      "eval_samples_per_second": 38.536,
+      "eval_steps_per_second": 0.605,
+      "step": 50000
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 3.765170919381101e-05,
+      "loss": 0.9542,
+      "step": 50200
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 3.7642331222899134e-05,
+      "loss": 0.9505,
+      "step": 50400
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 3.7632953251987255e-05,
+      "loss": 0.9451,
+      "step": 50600
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 3.7623575281075376e-05,
+      "loss": 0.9422,
+      "step": 50800
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 3.76141973101635e-05,
+      "loss": 0.9521,
+      "step": 51000
+    },
+    {
+      "epoch": 0.06,
+      "eval_em_first_line_ratio": 0.4567,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3457,
+      "eval_loss": 0.8867935538291931,
+      "eval_runtime": 283.1249,
+      "eval_samples_per_second": 35.32,
+      "eval_steps_per_second": 0.555,
+      "step": 51000
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 3.760481933925162e-05,
+      "loss": 0.9461,
+      "step": 51200
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 3.759544136833974e-05,
+      "loss": 0.9473,
+      "step": 51400
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 3.758606339742786e-05,
+      "loss": 0.9468,
+      "step": 51600
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 3.757668542651598e-05,
+      "loss": 0.9419,
+      "step": 51800
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 3.75673074556041e-05,
+      "loss": 0.9464,
+      "step": 52000
+    },
+    {
+      "epoch": 0.06,
+      "eval_em_first_line_ratio": 0.4558,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3443,
+      "eval_loss": 0.8852089047431946,
+      "eval_runtime": 276.8891,
+      "eval_samples_per_second": 36.116,
+      "eval_steps_per_second": 0.567,
+      "step": 52000
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 3.755792948469222e-05,
+      "loss": 0.9489,
+      "step": 52200
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 3.7548551513780344e-05,
+      "loss": 0.9452,
+      "step": 52400
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 3.7539173542868465e-05,
+      "loss": 0.9446,
+      "step": 52600
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 3.7529795571956586e-05,
+      "loss": 0.9408,
+      "step": 52800
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 3.752041760104471e-05,
+      "loss": 0.9404,
+      "step": 53000
+    },
+    {
+      "epoch": 0.06,
+      "eval_em_first_line_ratio": 0.4586,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3466,
+      "eval_loss": 0.8833824396133423,
+      "eval_runtime": 226.7684,
+      "eval_samples_per_second": 44.098,
+      "eval_steps_per_second": 0.692,
+      "step": 53000
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 3.751103963013283e-05,
+      "loss": 0.9405,
+      "step": 53200
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 3.750166165922095e-05,
+      "loss": 0.9428,
+      "step": 53400
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 3.749228368830908e-05,
+      "loss": 0.9466,
+      "step": 53600
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 3.748290571739719e-05,
+      "loss": 0.94,
+      "step": 53800
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 3.747352774648532e-05,
+      "loss": 0.9454,
+      "step": 54000
+    },
+    {
+      "epoch": 0.06,
+      "eval_em_first_line_ratio": 0.4554,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3439,
+      "eval_loss": 0.8826947212219238,
+      "eval_runtime": 234.8237,
+      "eval_samples_per_second": 42.585,
+      "eval_steps_per_second": 0.669,
+      "step": 54000
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 3.746414977557343e-05,
+      "loss": 0.9422,
+      "step": 54200
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 3.745477180466156e-05,
+      "loss": 0.9467,
+      "step": 54400
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 3.744539383374968e-05,
+      "loss": 0.937,
+      "step": 54600
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 3.74360158628378e-05,
+      "loss": 0.9461,
+      "step": 54800
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 3.7426637891925923e-05,
+      "loss": 0.9394,
+      "step": 55000
+    },
+    {
+      "epoch": 0.06,
+      "eval_em_first_line_ratio": 0.458,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3468,
+      "eval_loss": 0.8802450299263,
+      "eval_runtime": 251.4679,
+      "eval_samples_per_second": 39.767,
+      "eval_steps_per_second": 0.624,
+      "step": 55000
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 3.7417259921014044e-05,
+      "loss": 0.9417,
+      "step": 55200
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 3.7407881950102165e-05,
+      "loss": 0.935,
+      "step": 55400
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 3.7398503979190286e-05,
+      "loss": 0.9336,
+      "step": 55600
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 3.738912600827841e-05,
+      "loss": 0.9414,
+      "step": 55800
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 3.737979492722109e-05,
+      "loss": 0.9391,
+      "step": 56000
+    },
+    {
+      "epoch": 0.07,
+      "eval_em_first_line_ratio": 0.4569,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3451,
+      "eval_loss": 0.8788434863090515,
+      "eval_runtime": 255.1162,
+      "eval_samples_per_second": 39.198,
+      "eval_steps_per_second": 0.615,
+      "step": 56000
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 3.737041695630921e-05,
+      "loss": 0.9404,
+      "step": 56200
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 3.736103898539733e-05,
+      "loss": 0.9399,
+      "step": 56400
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 3.735166101448545e-05,
+      "loss": 0.9365,
+      "step": 56600
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 3.7342329933428135e-05,
+      "loss": 0.9426,
+      "step": 56800
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 3.7332951962516256e-05,
+      "loss": 0.9434,
+      "step": 57000
+    },
+    {
+      "epoch": 0.07,
+      "eval_em_first_line_ratio": 0.4556,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3443,
+      "eval_loss": 0.8779515624046326,
+      "eval_runtime": 267.3725,
+      "eval_samples_per_second": 37.401,
+      "eval_steps_per_second": 0.587,
+      "step": 57000
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 3.732357399160437e-05,
+      "loss": 0.9364,
+      "step": 57200
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 3.73141960206925e-05,
+      "loss": 0.9376,
+      "step": 57400
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 3.730481804978061e-05,
+      "loss": 0.9393,
+      "step": 57600
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 3.729544007886874e-05,
+      "loss": 0.9346,
+      "step": 57800
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 3.728606210795686e-05,
+      "loss": 0.9373,
+      "step": 58000
+    },
+    {
+      "epoch": 0.07,
+      "eval_em_first_line_ratio": 0.4558,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3442,
+      "eval_loss": 0.8759181499481201,
+      "eval_runtime": 272.7371,
+      "eval_samples_per_second": 36.665,
+      "eval_steps_per_second": 0.576,
+      "step": 58000
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 3.727668413704498e-05,
+      "loss": 0.9371,
+      "step": 58200
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 3.72673061661331e-05,
+      "loss": 0.9383,
+      "step": 58400
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 3.7257928195221224e-05,
+      "loss": 0.9314,
+      "step": 58600
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 3.7248550224309345e-05,
+      "loss": 0.9336,
+      "step": 58800
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 3.7239172253397466e-05,
+      "loss": 0.9347,
+      "step": 59000
+    },
+    {
+      "epoch": 0.07,
+      "eval_em_first_line_ratio": 0.4566,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3454,
+      "eval_loss": 0.875309944152832,
+      "eval_runtime": 255.3506,
+      "eval_samples_per_second": 39.162,
+      "eval_steps_per_second": 0.615,
+      "step": 59000
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 3.722979428248559e-05,
+      "loss": 0.9356,
+      "step": 59200
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 3.722041631157371e-05,
+      "loss": 0.9329,
+      "step": 59400
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 3.721103834066183e-05,
+      "loss": 0.9337,
+      "step": 59600
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 3.720166036974995e-05,
+      "loss": 0.9291,
+      "step": 59800
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 3.719228239883807e-05,
+      "loss": 0.9367,
+      "step": 60000
+    },
+    {
+      "epoch": 0.07,
+      "eval_em_first_line_ratio": 0.4575,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3452,
+      "eval_loss": 0.8750749826431274,
+      "eval_runtime": 264.3661,
+      "eval_samples_per_second": 37.826,
+      "eval_steps_per_second": 0.594,
+      "step": 60000
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 3.718295131778075e-05,
+      "loss": 0.9352,
+      "step": 60200
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 3.7173573346868874e-05,
+      "loss": 0.9314,
+      "step": 60400
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 3.7164195375956995e-05,
+      "loss": 0.9304,
+      "step": 60600
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 3.7154817405045116e-05,
+      "loss": 0.9326,
+      "step": 60800
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 3.714543943413324e-05,
+      "loss": 0.9348,
+      "step": 61000
+    },
+    {
+      "epoch": 0.07,
+      "eval_em_first_line_ratio": 0.4577,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3461,
+      "eval_loss": 0.8724747896194458,
+      "eval_runtime": 259.6707,
+      "eval_samples_per_second": 38.51,
+      "eval_steps_per_second": 0.605,
+      "step": 61000
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 3.713606146322136e-05,
+      "loss": 0.9299,
+      "step": 61200
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 3.712668349230948e-05,
+      "loss": 0.9357,
+      "step": 61400
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 3.71173055213976e-05,
+      "loss": 0.9305,
+      "step": 61600
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 3.710792755048573e-05,
+      "loss": 0.9293,
+      "step": 61800
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 3.709854957957384e-05,
+      "loss": 0.9257,
+      "step": 62000
+    },
+    {
+      "epoch": 0.07,
+      "eval_em_first_line_ratio": 0.4571,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3453,
+      "eval_loss": 0.870647668838501,
+      "eval_runtime": 262.5202,
+      "eval_samples_per_second": 38.092,
+      "eval_steps_per_second": 0.598,
+      "step": 62000
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 3.708917160866197e-05,
+      "loss": 0.9246,
+      "step": 62200
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 3.7079793637750084e-05,
+      "loss": 0.9288,
+      "step": 62400
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 3.707041566683821e-05,
+      "loss": 0.9331,
+      "step": 62600
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 3.706103769592633e-05,
+      "loss": 0.9275,
+      "step": 62800
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 3.7051659725014454e-05,
+      "loss": 0.9325,
+      "step": 63000
+    },
+    {
+      "epoch": 0.07,
+      "eval_em_first_line_ratio": 0.4579,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3465,
+      "eval_loss": 0.8716476559638977,
+      "eval_runtime": 254.2003,
+      "eval_samples_per_second": 39.339,
+      "eval_steps_per_second": 0.618,
+      "step": 63000
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 3.704232864395713e-05,
+      "loss": 0.9291,
+      "step": 63200
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 3.703299756289981e-05,
+      "loss": 0.9234,
+      "step": 63400
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 3.702361959198793e-05,
+      "loss": 0.9278,
+      "step": 63600
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 3.7014241621076054e-05,
+      "loss": 0.9238,
+      "step": 63800
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 3.7004863650164175e-05,
+      "loss": 0.928,
+      "step": 64000
+    },
+    {
+      "epoch": 0.08,
+      "eval_em_first_line_ratio": 0.4612,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3492,
+      "eval_loss": 0.868577241897583,
+      "eval_runtime": 232.2077,
+      "eval_samples_per_second": 43.065,
+      "eval_steps_per_second": 0.676,
+      "step": 64000
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 3.6995485679252296e-05,
+      "loss": 0.9308,
+      "step": 64200
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 3.698610770834042e-05,
+      "loss": 0.9229,
+      "step": 64400
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 3.697672973742854e-05,
+      "loss": 0.9295,
+      "step": 64600
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 3.696739865637122e-05,
+      "loss": 0.9297,
+      "step": 64800
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 3.695802068545934e-05,
+      "loss": 0.9214,
+      "step": 65000
+    },
+    {
+      "epoch": 0.08,
+      "eval_em_first_line_ratio": 0.4591,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3471,
+      "eval_loss": 0.8682117462158203,
+      "eval_runtime": 223.1071,
+      "eval_samples_per_second": 44.822,
+      "eval_steps_per_second": 0.704,
+      "step": 65000
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 3.694864271454746e-05,
+      "loss": 0.9227,
+      "step": 65200
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 3.693926474363558e-05,
+      "loss": 0.9243,
+      "step": 65400
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 3.6929886772723704e-05,
+      "loss": 0.9248,
+      "step": 65600
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 3.6920508801811825e-05,
+      "loss": 0.9273,
+      "step": 65800
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 3.6911130830899946e-05,
+      "loss": 0.9247,
+      "step": 66000
+    },
+    {
+      "epoch": 0.08,
+      "eval_em_first_line_ratio": 0.4582,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3462,
+      "eval_loss": 0.8676705360412598,
+      "eval_runtime": 250.6495,
+      "eval_samples_per_second": 39.896,
+      "eval_steps_per_second": 0.626,
+      "step": 66000
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 3.6901752859988074e-05,
+      "loss": 0.9284,
+      "step": 66200
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 3.689237488907619e-05,
+      "loss": 0.9319,
+      "step": 66400
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 3.6882996918164316e-05,
+      "loss": 0.926,
+      "step": 66600
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 3.687361894725243e-05,
+      "loss": 0.9244,
+      "step": 66800
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 3.686424097634056e-05,
+      "loss": 0.9251,
+      "step": 67000
+    },
+    {
+      "epoch": 0.08,
+      "eval_em_first_line_ratio": 0.4611,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3495,
+      "eval_loss": 0.8662257790565491,
+      "eval_runtime": 264.7353,
+      "eval_samples_per_second": 37.774,
+      "eval_steps_per_second": 0.593,
+      "step": 67000
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 3.685486300542867e-05,
+      "loss": 0.925,
+      "step": 67200
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 3.68454850345168e-05,
+      "loss": 0.924,
+      "step": 67400
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 3.683610706360492e-05,
+      "loss": 0.9247,
+      "step": 67600
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 3.682672909269304e-05,
+      "loss": 0.9224,
+      "step": 67800
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 3.681735112178116e-05,
+      "loss": 0.9223,
+      "step": 68000
+    },
+    {
+      "epoch": 0.08,
+      "eval_em_first_line_ratio": 0.4588,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3474,
+      "eval_loss": 0.8646082878112793,
+      "eval_runtime": 251.495,
+      "eval_samples_per_second": 39.762,
+      "eval_steps_per_second": 0.624,
+      "step": 68000
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 3.6807973150869284e-05,
+      "loss": 0.9163,
+      "step": 68200
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 3.6798595179957405e-05,
+      "loss": 0.9237,
+      "step": 68400
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 3.6789217209045526e-05,
+      "loss": 0.917,
+      "step": 68600
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 3.6779839238133647e-05,
+      "loss": 0.9229,
+      "step": 68800
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 3.677046126722177e-05,
+      "loss": 0.9181,
+      "step": 69000
+    },
+    {
+      "epoch": 0.08,
+      "eval_em_first_line_ratio": 0.46,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3482,
+      "eval_loss": 0.8637815713882446,
+      "eval_runtime": 247.6716,
+      "eval_samples_per_second": 40.376,
+      "eval_steps_per_second": 0.634,
+      "step": 69000
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 3.676108329630989e-05,
+      "loss": 0.9229,
+      "step": 69200
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 3.675170532539801e-05,
+      "loss": 0.9204,
+      "step": 69400
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 3.674232735448613e-05,
+      "loss": 0.9138,
+      "step": 69600
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 3.673294938357425e-05,
+      "loss": 0.9213,
+      "step": 69800
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 3.672357141266237e-05,
+      "loss": 0.9186,
+      "step": 70000
+    },
+    {
+      "epoch": 0.08,
+      "eval_em_first_line_ratio": 0.4619,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3501,
+      "eval_loss": 0.8636025190353394,
+      "eval_runtime": 256.4164,
+      "eval_samples_per_second": 38.999,
+      "eval_steps_per_second": 0.612,
+      "step": 70000
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 3.6714193441750493e-05,
+      "loss": 0.9178,
+      "step": 70200
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 3.6704815470838614e-05,
+      "loss": 0.9145,
+      "step": 70400
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 3.6695437499926735e-05,
+      "loss": 0.9217,
+      "step": 70600
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 3.6686059529014856e-05,
+      "loss": 0.9174,
+      "step": 70800
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 3.667672844795754e-05,
+      "loss": 0.923,
+      "step": 71000
+    },
+    {
+      "epoch": 0.08,
+      "eval_em_first_line_ratio": 0.4623,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.35,
+      "eval_loss": 0.8620550036430359,
+      "eval_runtime": 248.2682,
+      "eval_samples_per_second": 40.279,
+      "eval_steps_per_second": 0.632,
+      "step": 71000
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 3.666735047704566e-05,
+      "loss": 0.9123,
+      "step": 71200
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 3.665797250613378e-05,
+      "loss": 0.9198,
+      "step": 71400
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 3.66485945352219e-05,
+      "loss": 0.9132,
+      "step": 71600
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 3.663921656431002e-05,
+      "loss": 0.9244,
+      "step": 71800
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 3.6629838593398144e-05,
+      "loss": 0.9166,
+      "step": 72000
+    },
+    {
+      "epoch": 0.08,
+      "eval_em_first_line_ratio": 0.4582,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3466,
+      "eval_loss": 0.8603691458702087,
+      "eval_runtime": 253.768,
+      "eval_samples_per_second": 39.406,
+      "eval_steps_per_second": 0.619,
+      "step": 72000
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 3.6620460622486265e-05,
+      "loss": 0.9208,
+      "step": 72200
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 3.661108265157439e-05,
+      "loss": 0.9174,
+      "step": 72400
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 3.6601704680662507e-05,
+      "loss": 0.9183,
+      "step": 72600
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 3.6592326709750634e-05,
+      "loss": 0.9102,
+      "step": 72800
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 3.658294873883875e-05,
+      "loss": 0.9204,
+      "step": 73000
+    },
+    {
+      "epoch": 0.09,
+      "eval_em_first_line_ratio": 0.4626,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.35,
+      "eval_loss": 0.8592997789382935,
+      "eval_runtime": 248.2607,
+      "eval_samples_per_second": 40.28,
+      "eval_steps_per_second": 0.632,
+      "step": 73000
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 3.6573570767926876e-05,
+      "loss": 0.919,
+      "step": 73200
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 3.6564192797015e-05,
+      "loss": 0.9107,
+      "step": 73400
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 3.655481482610312e-05,
+      "loss": 0.9164,
+      "step": 73600
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 3.65454837450458e-05,
+      "loss": 0.9123,
+      "step": 73800
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 3.6536105774133915e-05,
+      "loss": 0.9151,
+      "step": 74000
+    },
+    {
+      "epoch": 0.09,
+      "eval_em_first_line_ratio": 0.4616,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.35,
+      "eval_loss": 0.8584997057914734,
+      "eval_runtime": 265.9739,
+      "eval_samples_per_second": 37.598,
+      "eval_steps_per_second": 0.59,
+      "step": 74000
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 3.652672780322204e-05,
+      "loss": 0.9054,
+      "step": 74200
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 3.6517396722164725e-05,
+      "loss": 0.9125,
+      "step": 74400
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 3.650801875125284e-05,
+      "loss": 0.9096,
+      "step": 74600
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 3.649864078034096e-05,
+      "loss": 0.9114,
+      "step": 74800
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 3.648926280942908e-05,
+      "loss": 0.9181,
+      "step": 75000
+    },
+    {
+      "epoch": 0.09,
+      "eval_em_first_line_ratio": 0.4614,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3494,
+      "eval_loss": 0.8581753969192505,
+      "eval_runtime": 269.9269,
+      "eval_samples_per_second": 37.047,
+      "eval_steps_per_second": 0.582,
+      "step": 75000
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 3.64798848385172e-05,
+      "loss": 0.916,
+      "step": 75200
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 3.647050686760532e-05,
+      "loss": 0.9148,
+      "step": 75400
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 3.6461128896693444e-05,
+      "loss": 0.9117,
+      "step": 75600
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 3.645175092578157e-05,
+      "loss": 0.9116,
+      "step": 75800
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 3.6442372954869686e-05,
+      "loss": 0.9149,
+      "step": 76000
+    },
+    {
+      "epoch": 0.09,
+      "eval_em_first_line_ratio": 0.464,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3519,
+      "eval_loss": 0.8569427728652954,
+      "eval_runtime": 229.5229,
+      "eval_samples_per_second": 43.569,
+      "eval_steps_per_second": 0.684,
+      "step": 76000
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 3.6432994983957814e-05,
+      "loss": 0.9144,
+      "step": 76200
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 3.642361701304593e-05,
+      "loss": 0.9143,
+      "step": 76400
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 3.6414239042134056e-05,
+      "loss": 0.9108,
+      "step": 76600
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 3.640486107122218e-05,
+      "loss": 0.9105,
+      "step": 76800
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 3.639552999016485e-05,
+      "loss": 0.9036,
+      "step": 77000
+    },
+    {
+      "epoch": 0.09,
+      "eval_em_first_line_ratio": 0.4622,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3508,
+      "eval_loss": 0.8557179570198059,
+      "eval_runtime": 215.6999,
+      "eval_samples_per_second": 46.361,
+      "eval_steps_per_second": 0.728,
+      "step": 77000
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 3.638615201925298e-05,
+      "loss": 0.9112,
+      "step": 77200
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 3.6376774048341095e-05,
+      "loss": 0.9059,
+      "step": 77400
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 3.636739607742922e-05,
+      "loss": 0.9118,
+      "step": 77600
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 3.635801810651734e-05,
+      "loss": 0.9112,
+      "step": 77800
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 3.6348640135605464e-05,
+      "loss": 0.9138,
+      "step": 78000
+    },
+    {
+      "epoch": 0.09,
+      "eval_em_first_line_ratio": 0.4608,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3492,
+      "eval_loss": 0.8551828861236572,
+      "eval_runtime": 270.2152,
+      "eval_samples_per_second": 37.008,
+      "eval_steps_per_second": 0.581,
+      "step": 78000
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 3.6339262164693585e-05,
+      "loss": 0.9061,
+      "step": 78200
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 3.6329884193781706e-05,
+      "loss": 0.9151,
+      "step": 78400
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 3.632050622286983e-05,
+      "loss": 0.9128,
+      "step": 78600
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 3.631112825195795e-05,
+      "loss": 0.9048,
+      "step": 78800
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 3.630175028104607e-05,
+      "loss": 0.9058,
+      "step": 79000
+    },
+    {
+      "epoch": 0.09,
+      "eval_em_first_line_ratio": 0.4624,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3507,
+      "eval_loss": 0.8544037342071533,
+      "eval_runtime": 249.7377,
+      "eval_samples_per_second": 40.042,
+      "eval_steps_per_second": 0.629,
+      "step": 79000
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 3.629237231013419e-05,
+      "loss": 0.9078,
+      "step": 79200
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 3.628299433922231e-05,
+      "loss": 0.9097,
+      "step": 79400
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 3.627361636831043e-05,
+      "loss": 0.9018,
+      "step": 79600
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 3.626423839739855e-05,
+      "loss": 0.9012,
+      "step": 79800
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 3.6254860426486674e-05,
+      "loss": 0.904,
+      "step": 80000
+    },
+    {
+      "epoch": 0.09,
+      "eval_em_first_line_ratio": 0.4624,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3512,
+      "eval_loss": 0.8525384664535522,
+      "eval_runtime": 246.8232,
+      "eval_samples_per_second": 40.515,
+      "eval_steps_per_second": 0.636,
+      "step": 80000
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 3.6245482455574795e-05,
+      "loss": 0.9052,
+      "step": 80200
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 3.6236104484662916e-05,
+      "loss": 0.9082,
+      "step": 80400
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 3.6226726513751044e-05,
+      "loss": 0.9087,
+      "step": 80600
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 3.621734854283916e-05,
+      "loss": 0.9144,
+      "step": 80800
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 3.6207970571927286e-05,
+      "loss": 0.9063,
+      "step": 81000
+    },
+    {
+      "epoch": 0.09,
+      "eval_em_first_line_ratio": 0.463,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3508,
+      "eval_loss": 0.8522350788116455,
+      "eval_runtime": 248.1515,
+      "eval_samples_per_second": 40.298,
+      "eval_steps_per_second": 0.633,
+      "step": 81000
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 3.61985926010154e-05,
+      "loss": 0.9033,
+      "step": 81200
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 3.618921463010352e-05,
+      "loss": 0.9045,
+      "step": 81400
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 3.61798835490462e-05,
+      "loss": 0.9048,
+      "step": 81600
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 3.6170505578134324e-05,
+      "loss": 0.9096,
+      "step": 81800
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 3.6161127607222445e-05,
+      "loss": 0.9033,
+      "step": 82000
+    },
+    {
+      "epoch": 0.1,
+      "eval_em_first_line_ratio": 0.4643,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3525,
+      "eval_loss": 0.8522495031356812,
+      "eval_runtime": 253.6367,
+      "eval_samples_per_second": 39.426,
+      "eval_steps_per_second": 0.619,
+      "step": 82000
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 3.6151749636310566e-05,
+      "loss": 0.908,
+      "step": 82200
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 3.614237166539869e-05,
+      "loss": 0.9075,
+      "step": 82400
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 3.6132993694486815e-05,
+      "loss": 0.9056,
+      "step": 82600
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 3.612361572357493e-05,
+      "loss": 0.9041,
+      "step": 82800
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 3.611423775266306e-05,
+      "loss": 0.9078,
+      "step": 83000
+    },
+    {
+      "epoch": 0.1,
+      "eval_em_first_line_ratio": 0.4633,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3517,
+      "eval_loss": 0.8507208824157715,
+      "eval_runtime": 250.1533,
+      "eval_samples_per_second": 39.975,
+      "eval_steps_per_second": 0.628,
+      "step": 83000
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 3.610485978175117e-05,
+      "loss": 0.903,
+      "step": 83200
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 3.60954818108393e-05,
+      "loss": 0.9062,
+      "step": 83400
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 3.608610383992741e-05,
+      "loss": 0.9054,
+      "step": 83600
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 3.607672586901554e-05,
+      "loss": 0.9031,
+      "step": 83800
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 3.606734789810366e-05,
+      "loss": 0.9081,
+      "step": 84000
+    },
+    {
+      "epoch": 0.1,
+      "eval_em_first_line_ratio": 0.4649,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3536,
+      "eval_loss": 0.8500064611434937,
+      "eval_runtime": 244.6911,
+      "eval_samples_per_second": 40.868,
+      "eval_steps_per_second": 0.642,
+      "step": 84000
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 3.605796992719178e-05,
+      "loss": 0.9001,
+      "step": 84200
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 3.6048591956279904e-05,
+      "loss": 0.9053,
+      "step": 84400
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 3.6039213985368025e-05,
+      "loss": 0.9057,
+      "step": 84600
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 3.602988290431071e-05,
+      "loss": 0.9008,
+      "step": 84800
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 3.602050493339883e-05,
+      "loss": 0.9013,
+      "step": 85000
+    },
+    {
+      "epoch": 0.1,
+      "eval_em_first_line_ratio": 0.4618,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3498,
+      "eval_loss": 0.8498925566673279,
+      "eval_runtime": 261.3849,
+      "eval_samples_per_second": 38.258,
+      "eval_steps_per_second": 0.601,
+      "step": 85000
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 3.6011173852341504e-05,
+      "loss": 0.9018,
+      "step": 85200
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 3.600179588142963e-05,
+      "loss": 0.8977,
+      "step": 85400
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 3.5992417910517746e-05,
+      "loss": 0.9007,
+      "step": 85600
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 3.5983039939605874e-05,
+      "loss": 0.9032,
+      "step": 85800
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 3.5973661968693995e-05,
+      "loss": 0.9083,
+      "step": 86000
+    },
+    {
+      "epoch": 0.1,
+      "eval_em_first_line_ratio": 0.4666,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3543,
+      "eval_loss": 0.8492092490196228,
+      "eval_runtime": 249.6421,
+      "eval_samples_per_second": 40.057,
+      "eval_steps_per_second": 0.629,
+      "step": 86000
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 3.5964283997782115e-05,
+      "loss": 0.8969,
+      "step": 86200
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 3.5954906026870236e-05,
+      "loss": 0.9019,
+      "step": 86400
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 3.594552805595836e-05,
+      "loss": 0.8984,
+      "step": 86600
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 3.593615008504648e-05,
+      "loss": 0.8959,
+      "step": 86800
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 3.592677211413459e-05,
+      "loss": 0.8973,
+      "step": 87000
+    },
+    {
+      "epoch": 0.1,
+      "eval_em_first_line_ratio": 0.4662,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.354,
+      "eval_loss": 0.8472490310668945,
+      "eval_runtime": 224.7903,
+      "eval_samples_per_second": 44.486,
+      "eval_steps_per_second": 0.698,
+      "step": 87000
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 3.5917441033077275e-05,
+      "loss": 0.903,
+      "step": 87200
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 3.59080630621654e-05,
+      "loss": 0.8994,
+      "step": 87400
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 3.589868509125352e-05,
+      "loss": 0.8953,
+      "step": 87600
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 3.5889307120341645e-05,
+      "loss": 0.899,
+      "step": 87800
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 3.587992914942976e-05,
+      "loss": 0.9032,
+      "step": 88000
+    },
+    {
+      "epoch": 0.1,
+      "eval_em_first_line_ratio": 0.4664,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.354,
+      "eval_loss": 0.8465429544448853,
+      "eval_runtime": 217.7745,
+      "eval_samples_per_second": 45.919,
+      "eval_steps_per_second": 0.721,
+      "step": 88000
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 3.587055117851789e-05,
+      "loss": 0.8996,
+      "step": 88200
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 3.586117320760601e-05,
+      "loss": 0.9008,
+      "step": 88400
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 3.585179523669413e-05,
+      "loss": 0.8977,
+      "step": 88600
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 3.584241726578225e-05,
+      "loss": 0.8985,
+      "step": 88800
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 3.583303929487037e-05,
+      "loss": 0.9075,
+      "step": 89000
+    },
+    {
+      "epoch": 0.1,
+      "eval_em_first_line_ratio": 0.4639,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3521,
+      "eval_loss": 0.8466284871101379,
+      "eval_runtime": 246.1731,
+      "eval_samples_per_second": 40.622,
+      "eval_steps_per_second": 0.638,
+      "step": 89000
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 3.582366132395849e-05,
+      "loss": 0.9019,
+      "step": 89200
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 3.581428335304661e-05,
+      "loss": 0.9012,
+      "step": 89400
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 3.5804905382134734e-05,
+      "loss": 0.8984,
+      "step": 89600
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 3.5795527411222855e-05,
+      "loss": 0.8952,
+      "step": 89800
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 3.5786149440310976e-05,
+      "loss": 0.8983,
+      "step": 90000
+    },
+    {
+      "epoch": 0.11,
+      "eval_em_first_line_ratio": 0.4658,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3543,
+      "eval_loss": 0.8453260660171509,
+      "eval_runtime": 257.3719,
+      "eval_samples_per_second": 38.854,
+      "eval_steps_per_second": 0.61,
+      "step": 90000
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 3.5776771469399096e-05,
+      "loss": 0.8975,
+      "step": 90200
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 3.576739349848722e-05,
+      "loss": 0.8938,
+      "step": 90400
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 3.575801552757534e-05,
+      "loss": 0.8958,
+      "step": 90600
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 3.5748637556663466e-05,
+      "loss": 0.8922,
+      "step": 90800
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 3.573925958575158e-05,
+      "loss": 0.8964,
+      "step": 91000
+    },
+    {
+      "epoch": 0.11,
+      "eval_em_first_line_ratio": 0.4665,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3546,
+      "eval_loss": 0.844612181186676,
+      "eval_runtime": 252.8193,
+      "eval_samples_per_second": 39.554,
+      "eval_steps_per_second": 0.621,
+      "step": 91000
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 3.572988161483971e-05,
+      "loss": 0.8978,
+      "step": 91200
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 3.5720550533782384e-05,
+      "loss": 0.8954,
+      "step": 91400
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 3.5711172562870505e-05,
+      "loss": 0.8974,
+      "step": 91600
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 3.570179459195863e-05,
+      "loss": 0.8981,
+      "step": 91800
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 3.569241662104675e-05,
+      "loss": 0.8982,
+      "step": 92000
+    },
+    {
+      "epoch": 0.11,
+      "eval_em_first_line_ratio": 0.4667,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3555,
+      "eval_loss": 0.841987669467926,
+      "eval_runtime": 247.9558,
+      "eval_samples_per_second": 40.33,
+      "eval_steps_per_second": 0.633,
+      "step": 92000
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 3.5683038650134875e-05,
+      "loss": 0.9033,
+      "step": 92200
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 3.567366067922299e-05,
+      "loss": 0.9011,
+      "step": 92400
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 3.5664282708311116e-05,
+      "loss": 0.8967,
+      "step": 92600
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 3.565490473739923e-05,
+      "loss": 0.8922,
+      "step": 92800
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 3.564552676648735e-05,
+      "loss": 0.8938,
+      "step": 93000
+    },
+    {
+      "epoch": 0.11,
+      "eval_em_first_line_ratio": 0.464,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3523,
+      "eval_loss": 0.8430120348930359,
+      "eval_runtime": 257.3433,
+      "eval_samples_per_second": 38.859,
+      "eval_steps_per_second": 0.61,
+      "step": 93000
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 3.563614879557548e-05,
+      "loss": 0.8917,
+      "step": 93200
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 3.5626770824663594e-05,
+      "loss": 0.8942,
+      "step": 93400
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 3.561739285375172e-05,
+      "loss": 0.8928,
+      "step": 93600
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 3.5608014882839836e-05,
+      "loss": 0.8896,
+      "step": 93800
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 3.559863691192796e-05,
+      "loss": 0.901,
+      "step": 94000
+    },
+    {
+      "epoch": 0.11,
+      "eval_em_first_line_ratio": 0.4668,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3547,
+      "eval_loss": 0.8413771390914917,
+      "eval_runtime": 251.3552,
+      "eval_samples_per_second": 39.784,
+      "eval_steps_per_second": 0.625,
+      "step": 94000
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 3.558925894101608e-05,
+      "loss": 0.8965,
+      "step": 94200
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 3.5579880970104205e-05,
+      "loss": 0.8967,
+      "step": 94400
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 3.5570502999192326e-05,
+      "loss": 0.8954,
+      "step": 94600
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 3.556112502828045e-05,
+      "loss": 0.8946,
+      "step": 94800
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 3.555174705736857e-05,
+      "loss": 0.8902,
+      "step": 95000
+    },
+    {
+      "epoch": 0.11,
+      "eval_em_first_line_ratio": 0.468,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3558,
+      "eval_loss": 0.8411099314689636,
+      "eval_runtime": 251.3802,
+      "eval_samples_per_second": 39.78,
+      "eval_steps_per_second": 0.625,
+      "step": 95000
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 3.554236908645669e-05,
+      "loss": 0.8928,
+      "step": 95200
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 3.553303800539937e-05,
+      "loss": 0.8957,
+      "step": 95400
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 3.552366003448749e-05,
+      "loss": 0.8971,
+      "step": 95600
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 3.5514282063575614e-05,
+      "loss": 0.8917,
+      "step": 95800
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 3.5504904092663735e-05,
+      "loss": 0.8911,
+      "step": 96000
+    },
+    {
+      "epoch": 0.11,
+      "eval_em_first_line_ratio": 0.4671,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3554,
+      "eval_loss": 0.8402512669563293,
+      "eval_runtime": 246.5911,
+      "eval_samples_per_second": 40.553,
+      "eval_steps_per_second": 0.637,
+      "step": 96000
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 3.5495526121751856e-05,
+      "loss": 0.8976,
+      "step": 96200
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 3.5486148150839976e-05,
+      "loss": 0.8921,
+      "step": 96400
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 3.54767701799281e-05,
+      "loss": 0.8944,
+      "step": 96600
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 3.546739220901622e-05,
+      "loss": 0.8904,
+      "step": 96800
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 3.545801423810434e-05,
+      "loss": 0.8915,
+      "step": 97000
+    },
+    {
+      "epoch": 0.11,
+      "eval_em_first_line_ratio": 0.4678,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3561,
+      "eval_loss": 0.8394470810890198,
+      "eval_runtime": 259.6726,
+      "eval_samples_per_second": 38.51,
+      "eval_steps_per_second": 0.605,
+      "step": 97000
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 3.544863626719246e-05,
+      "loss": 0.8884,
+      "step": 97200
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 3.543925829628058e-05,
+      "loss": 0.8946,
+      "step": 97400
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 3.54298803253687e-05,
+      "loss": 0.8926,
+      "step": 97600
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 3.542050235445682e-05,
+      "loss": 0.8863,
+      "step": 97800
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 3.541112438354495e-05,
+      "loss": 0.8934,
+      "step": 98000
+    },
+    {
+      "epoch": 0.11,
+      "eval_em_first_line_ratio": 0.467,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3553,
+      "eval_loss": 0.8401299118995667,
+      "eval_runtime": 243.9754,
+      "eval_samples_per_second": 40.988,
+      "eval_steps_per_second": 0.644,
+      "step": 98000
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 3.5401746412633065e-05,
+      "loss": 0.8891,
+      "step": 98200
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 3.539236844172119e-05,
+      "loss": 0.892,
+      "step": 98400
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 3.538299047080931e-05,
+      "loss": 0.8909,
+      "step": 98600
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 3.5373612499897435e-05,
+      "loss": 0.892,
+      "step": 98800
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 3.536423452898555e-05,
+      "loss": 0.8875,
+      "step": 99000
+    },
+    {
+      "epoch": 0.12,
+      "eval_em_first_line_ratio": 0.4684,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3567,
+      "eval_loss": 0.8379769325256348,
+      "eval_runtime": 219.6275,
+      "eval_samples_per_second": 45.532,
+      "eval_steps_per_second": 0.715,
+      "step": 99000
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 3.535485655807368e-05,
+      "loss": 0.8896,
+      "step": 99200
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 3.53454785871618e-05,
+      "loss": 0.8938,
+      "step": 99400
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 3.533610061624991e-05,
+      "loss": 0.8902,
+      "step": 99600
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 3.532672264533804e-05,
+      "loss": 0.8882,
+      "step": 99800
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 3.5317344674426154e-05,
+      "loss": 0.8865,
+      "step": 100000
+    },
+    {
+      "epoch": 0.12,
+      "eval_em_first_line_ratio": 0.4664,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3548,
+      "eval_loss": 0.8369919657707214,
+      "eval_runtime": 215.1363,
+      "eval_samples_per_second": 46.482,
+      "eval_steps_per_second": 0.73,
+      "step": 100000
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 3.530796670351428e-05,
+      "loss": 0.8891,
+      "step": 100200
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 3.5298635622456964e-05,
+      "loss": 0.888,
+      "step": 100400
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 3.528925765154508e-05,
+      "loss": 0.8851,
+      "step": 100600
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 3.527992657048776e-05,
+      "loss": 0.8882,
+      "step": 100800
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 3.527054859957588e-05,
+      "loss": 0.8916,
+      "step": 101000
+    },
+    {
+      "epoch": 0.12,
+      "eval_em_first_line_ratio": 0.4652,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3535,
+      "eval_loss": 0.8360356688499451,
+      "eval_runtime": 243.5049,
+      "eval_samples_per_second": 41.067,
+      "eval_steps_per_second": 0.645,
+      "step": 101000
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 3.5261170628664e-05,
+      "loss": 0.8863,
+      "step": 101200
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 3.525179265775213e-05,
+      "loss": 0.8862,
+      "step": 101400
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 3.5242414686840245e-05,
+      "loss": 0.8885,
+      "step": 101600
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 3.523303671592837e-05,
+      "loss": 0.8862,
+      "step": 101800
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 3.522365874501649e-05,
+      "loss": 0.8845,
+      "step": 102000
+    },
+    {
+      "epoch": 0.12,
+      "eval_em_first_line_ratio": 0.4665,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.355,
+      "eval_loss": 0.835463285446167,
+      "eval_runtime": 243.277,
+      "eval_samples_per_second": 41.105,
+      "eval_steps_per_second": 0.645,
+      "step": 102000
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 3.5214280774104615e-05,
+      "loss": 0.8872,
+      "step": 102200
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 3.520490280319273e-05,
+      "loss": 0.8863,
+      "step": 102400
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 3.5195524832280856e-05,
+      "loss": 0.8866,
+      "step": 102600
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 3.518614686136898e-05,
+      "loss": 0.8917,
+      "step": 102800
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 3.51767688904571e-05,
+      "loss": 0.8876,
+      "step": 103000
+    },
+    {
+      "epoch": 0.12,
+      "eval_em_first_line_ratio": 0.4657,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3546,
+      "eval_loss": 0.8348633050918579,
+      "eval_runtime": 240.5795,
+      "eval_samples_per_second": 41.566,
+      "eval_steps_per_second": 0.653,
+      "step": 103000
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 3.516739091954522e-05,
+      "loss": 0.8899,
+      "step": 103200
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 3.515801294863334e-05,
+      "loss": 0.8844,
+      "step": 103400
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 3.514863497772146e-05,
+      "loss": 0.8843,
+      "step": 103600
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 3.513925700680958e-05,
+      "loss": 0.8859,
+      "step": 103800
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 3.51298790358977e-05,
+      "loss": 0.8914,
+      "step": 104000
+    },
+    {
+      "epoch": 0.12,
+      "eval_em_first_line_ratio": 0.4657,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.354,
+      "eval_loss": 0.8342224359512329,
+      "eval_runtime": 249.7635,
+      "eval_samples_per_second": 40.038,
+      "eval_steps_per_second": 0.629,
+      "step": 104000
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 3.5120501064985824e-05,
+      "loss": 0.8866,
+      "step": 104200
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 3.5111123094073945e-05,
+      "loss": 0.8861,
+      "step": 104400
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 3.5101745123162066e-05,
+      "loss": 0.8846,
+      "step": 104600
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 3.509236715225019e-05,
+      "loss": 0.8868,
+      "step": 104800
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 3.508298918133831e-05,
+      "loss": 0.8866,
+      "step": 105000
+    },
+    {
+      "epoch": 0.12,
+      "eval_em_first_line_ratio": 0.4656,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3539,
+      "eval_loss": 0.8325536847114563,
+      "eval_runtime": 246.188,
+      "eval_samples_per_second": 40.619,
+      "eval_steps_per_second": 0.638,
+      "step": 105000
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 3.5073611210426436e-05,
+      "loss": 0.8857,
+      "step": 105200
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 3.506423323951455e-05,
+      "loss": 0.8851,
+      "step": 105400
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 3.505485526860267e-05,
+      "loss": 0.8881,
+      "step": 105600
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 3.504547729769079e-05,
+      "loss": 0.8866,
+      "step": 105800
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 3.5036146216633475e-05,
+      "loss": 0.8824,
+      "step": 106000
+    },
+    {
+      "epoch": 0.12,
+      "eval_em_first_line_ratio": 0.467,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3554,
+      "eval_loss": 0.8334246277809143,
+      "eval_runtime": 254.9854,
+      "eval_samples_per_second": 39.218,
+      "eval_steps_per_second": 0.616,
+      "step": 106000
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 3.5026768245721596e-05,
+      "loss": 0.8882,
+      "step": 106200
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 3.5017390274809717e-05,
+      "loss": 0.8859,
+      "step": 106400
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 3.500801230389784e-05,
+      "loss": 0.8783,
+      "step": 106600
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 3.499863433298596e-05,
+      "loss": 0.8798,
+      "step": 106800
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 3.498925636207408e-05,
+      "loss": 0.8865,
+      "step": 107000
+    },
+    {
+      "epoch": 0.13,
+      "eval_em_first_line_ratio": 0.4695,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3575,
+      "eval_loss": 0.8327417969703674,
+      "eval_runtime": 244.0137,
+      "eval_samples_per_second": 40.981,
+      "eval_steps_per_second": 0.643,
+      "step": 107000
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 3.497992528101676e-05,
+      "loss": 0.886,
+      "step": 107200
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 3.497054731010488e-05,
+      "loss": 0.8861,
+      "step": 107400
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 3.4961169339193004e-05,
+      "loss": 0.8832,
+      "step": 107600
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 3.4951791368281125e-05,
+      "loss": 0.876,
+      "step": 107800
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 3.4942413397369246e-05,
+      "loss": 0.8829,
+      "step": 108000
+    },
+    {
+      "epoch": 0.13,
+      "eval_em_first_line_ratio": 0.4694,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3575,
+      "eval_loss": 0.831598162651062,
+      "eval_runtime": 230.4399,
+      "eval_samples_per_second": 43.395,
+      "eval_steps_per_second": 0.681,
+      "step": 108000
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 3.493303542645737e-05,
+      "loss": 0.8792,
+      "step": 108200
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 3.492365745554549e-05,
+      "loss": 0.8762,
+      "step": 108400
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 3.4914279484633616e-05,
+      "loss": 0.883,
+      "step": 108600
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 3.490490151372173e-05,
+      "loss": 0.8747,
+      "step": 108800
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 3.489552354280986e-05,
+      "loss": 0.874,
+      "step": 109000
+    },
+    {
+      "epoch": 0.13,
+      "eval_em_first_line_ratio": 0.4675,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3554,
+      "eval_loss": 0.8313378095626831,
+      "eval_runtime": 245.2622,
+      "eval_samples_per_second": 40.773,
+      "eval_steps_per_second": 0.64,
+      "step": 109000
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 3.488614557189797e-05,
+      "loss": 0.8772,
+      "step": 109200
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 3.48767676009861e-05,
+      "loss": 0.8851,
+      "step": 109400
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 3.486738963007422e-05,
+      "loss": 0.8818,
+      "step": 109600
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 3.485801165916234e-05,
+      "loss": 0.8771,
+      "step": 109800
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 3.484863368825046e-05,
+      "loss": 0.8812,
+      "step": 110000
+    },
+    {
+      "epoch": 0.13,
+      "eval_em_first_line_ratio": 0.47,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3581,
+      "eval_loss": 0.8297807574272156,
+      "eval_runtime": 214.1605,
+      "eval_samples_per_second": 46.694,
+      "eval_steps_per_second": 0.733,
+      "step": 110000
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 3.483925571733858e-05,
+      "loss": 0.8757,
+      "step": 110200
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 3.4829877746426704e-05,
+      "loss": 0.8783,
+      "step": 110400
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 3.4820499775514825e-05,
+      "loss": 0.8753,
+      "step": 110600
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 3.4811121804602946e-05,
+      "loss": 0.8815,
+      "step": 110800
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 3.480174383369107e-05,
+      "loss": 0.8817,
+      "step": 111000
+    },
+    {
+      "epoch": 0.13,
+      "eval_em_first_line_ratio": 0.4673,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3556,
+      "eval_loss": 0.8297120928764343,
+      "eval_runtime": 218.0149,
+      "eval_samples_per_second": 45.868,
+      "eval_steps_per_second": 0.72,
+      "step": 111000
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 3.479236586277919e-05,
+      "loss": 0.8799,
+      "step": 111200
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 3.478303478172187e-05,
+      "loss": 0.8808,
+      "step": 111400
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 3.4773703700664546e-05,
+      "loss": 0.8818,
+      "step": 111600
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 3.476432572975267e-05,
+      "loss": 0.8784,
+      "step": 111800
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 3.4754947758840795e-05,
+      "loss": 0.8789,
+      "step": 112000
+    },
+    {
+      "epoch": 0.13,
+      "eval_em_first_line_ratio": 0.4695,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3573,
+      "eval_loss": 0.8298242092132568,
+      "eval_runtime": 255.7105,
+      "eval_samples_per_second": 39.107,
+      "eval_steps_per_second": 0.614,
+      "step": 112000
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 3.474556978792891e-05,
+      "loss": 0.8836,
+      "step": 112200
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 3.473619181701704e-05,
+      "loss": 0.8799,
+      "step": 112400
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 3.472681384610515e-05,
+      "loss": 0.8772,
+      "step": 112600
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 3.471743587519328e-05,
+      "loss": 0.88,
+      "step": 112800
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 3.47080579042814e-05,
+      "loss": 0.8749,
+      "step": 113000
+    },
+    {
+      "epoch": 0.13,
+      "eval_em_first_line_ratio": 0.4673,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3559,
+      "eval_loss": 0.8295543193817139,
+      "eval_runtime": 255.5603,
+      "eval_samples_per_second": 39.13,
+      "eval_steps_per_second": 0.614,
+      "step": 113000
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 3.469867993336952e-05,
+      "loss": 0.8801,
+      "step": 113200
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 3.468930196245764e-05,
+      "loss": 0.8754,
+      "step": 113400
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 3.467992399154576e-05,
+      "loss": 0.8785,
+      "step": 113600
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 3.4670546020633884e-05,
+      "loss": 0.8728,
+      "step": 113800
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 3.4661168049722005e-05,
+      "loss": 0.878,
+      "step": 114000
+    },
+    {
+      "epoch": 0.13,
+      "eval_em_first_line_ratio": 0.4694,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3575,
+      "eval_loss": 0.8287162184715271,
+      "eval_runtime": 251.653,
+      "eval_samples_per_second": 39.737,
+      "eval_steps_per_second": 0.624,
+      "step": 114000
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 3.4651790078810126e-05,
+      "loss": 0.8821,
+      "step": 114200
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 3.464241210789825e-05,
+      "loss": 0.8769,
+      "step": 114400
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 3.463303413698637e-05,
+      "loss": 0.8771,
+      "step": 114600
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 3.462365616607449e-05,
+      "loss": 0.8718,
+      "step": 114800
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 3.461427819516261e-05,
+      "loss": 0.8778,
+      "step": 115000
+    },
+    {
+      "epoch": 0.13,
+      "eval_em_first_line_ratio": 0.4706,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3586,
+      "eval_loss": 0.8277656435966492,
+      "eval_runtime": 244.3324,
+      "eval_samples_per_second": 40.928,
+      "eval_steps_per_second": 0.643,
+      "step": 115000
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 3.460490022425073e-05,
+      "loss": 0.8714,
+      "step": 115200
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 3.459552225333886e-05,
+      "loss": 0.8705,
+      "step": 115400
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 3.458614428242697e-05,
+      "loss": 0.8757,
+      "step": 115600
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 3.45767663115151e-05,
+      "loss": 0.8796,
+      "step": 115800
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 3.4567388340603215e-05,
+      "loss": 0.8786,
+      "step": 116000
+    },
+    {
+      "epoch": 0.14,
+      "eval_em_first_line_ratio": 0.4692,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3582,
+      "eval_loss": 0.8271190524101257,
+      "eval_runtime": 252.0957,
+      "eval_samples_per_second": 39.667,
+      "eval_steps_per_second": 0.623,
+      "step": 116000
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 3.45580572595459e-05,
+      "loss": 0.8781,
+      "step": 116200
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 3.454872617848858e-05,
+      "loss": 0.8698,
+      "step": 116400
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 3.45393482075767e-05,
+      "loss": 0.8775,
+      "step": 116600
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 3.452997023666482e-05,
+      "loss": 0.8729,
+      "step": 116800
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 3.452059226575294e-05,
+      "loss": 0.875,
+      "step": 117000
+    },
+    {
+      "epoch": 0.14,
+      "eval_em_first_line_ratio": 0.4686,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3572,
+      "eval_loss": 0.8275221586227417,
+      "eval_runtime": 244.3917,
+      "eval_samples_per_second": 40.918,
+      "eval_steps_per_second": 0.642,
+      "step": 117000
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 3.4511214294841063e-05,
+      "loss": 0.8797,
+      "step": 117200
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 3.4501836323929184e-05,
+      "loss": 0.8771,
+      "step": 117400
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 3.4492458353017305e-05,
+      "loss": 0.872,
+      "step": 117600
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 3.4483080382105426e-05,
+      "loss": 0.8729,
+      "step": 117800
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 3.447370241119355e-05,
+      "loss": 0.8782,
+      "step": 118000
+    },
+    {
+      "epoch": 0.14,
+      "eval_em_first_line_ratio": 0.4688,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3572,
+      "eval_loss": 0.8266319632530212,
+      "eval_runtime": 252.0136,
+      "eval_samples_per_second": 39.68,
+      "eval_steps_per_second": 0.623,
+      "step": 118000
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 3.446432444028167e-05,
+      "loss": 0.878,
+      "step": 118200
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 3.445499335922435e-05,
+      "loss": 0.8764,
+      "step": 118400
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 3.444561538831247e-05,
+      "loss": 0.8735,
+      "step": 118600
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 3.443623741740059e-05,
+      "loss": 0.874,
+      "step": 118800
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 3.4426859446488714e-05,
+      "loss": 0.8676,
+      "step": 119000
+    },
+    {
+      "epoch": 0.14,
+      "eval_em_first_line_ratio": 0.4715,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3592,
+      "eval_loss": 0.8253640532493591,
+      "eval_runtime": 252.725,
+      "eval_samples_per_second": 39.569,
+      "eval_steps_per_second": 0.621,
+      "step": 119000
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 3.4417481475576835e-05,
+      "loss": 0.8757,
+      "step": 119200
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 3.4408103504664956e-05,
+      "loss": 0.8763,
+      "step": 119400
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 3.439872553375308e-05,
+      "loss": 0.8705,
+      "step": 119600
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 3.43893475628412e-05,
+      "loss": 0.8794,
+      "step": 119800
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 3.437996959192932e-05,
+      "loss": 0.8738,
+      "step": 120000
+    },
+    {
+      "epoch": 0.14,
+      "eval_em_first_line_ratio": 0.4696,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3575,
+      "eval_loss": 0.824611246585846,
+      "eval_runtime": 239.1436,
+      "eval_samples_per_second": 41.816,
+      "eval_steps_per_second": 0.657,
+      "step": 120000
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 3.4370591621017446e-05,
+      "loss": 0.8705,
+      "step": 120200
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 3.436121365010556e-05,
+      "loss": 0.8716,
+      "step": 120400
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 3.435183567919369e-05,
+      "loss": 0.8798,
+      "step": 120600
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 3.4342504598136364e-05,
+      "loss": 0.8728,
+      "step": 120800
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 3.4333126627224485e-05,
+      "loss": 0.8732,
+      "step": 121000
+    },
+    {
+      "epoch": 0.14,
+      "eval_em_first_line_ratio": 0.4693,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.358,
+      "eval_loss": 0.8241479396820068,
+      "eval_runtime": 216.6012,
+      "eval_samples_per_second": 46.168,
+      "eval_steps_per_second": 0.725,
+      "step": 121000
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 3.432374865631261e-05,
+      "loss": 0.8727,
+      "step": 121200
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 3.431437068540073e-05,
+      "loss": 0.8786,
+      "step": 121400
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 3.4304992714488855e-05,
+      "loss": 0.8761,
+      "step": 121600
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 3.429561474357697e-05,
+      "loss": 0.8737,
+      "step": 121800
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 3.42862367726651e-05,
+      "loss": 0.8722,
+      "step": 122000
+    },
+    {
+      "epoch": 0.14,
+      "eval_em_first_line_ratio": 0.4726,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3603,
+      "eval_loss": 0.8234519362449646,
+      "eval_runtime": 213.0532,
+      "eval_samples_per_second": 46.937,
+      "eval_steps_per_second": 0.737,
+      "step": 122000
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 3.427685880175322e-05,
+      "loss": 0.8681,
+      "step": 122200
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 3.426752772069589e-05,
+      "loss": 0.8719,
+      "step": 122400
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 3.4258149749784014e-05,
+      "loss": 0.8743,
+      "step": 122600
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 3.4248771778872135e-05,
+      "loss": 0.8738,
+      "step": 122800
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 3.4239393807960256e-05,
+      "loss": 0.8762,
+      "step": 123000
+    },
+    {
+      "epoch": 0.14,
+      "eval_em_first_line_ratio": 0.4703,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3581,
+      "eval_loss": 0.8234682679176331,
+      "eval_runtime": 225.5512,
+      "eval_samples_per_second": 44.336,
+      "eval_steps_per_second": 0.696,
+      "step": 123000
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 3.423001583704838e-05,
+      "loss": 0.8642,
+      "step": 123200
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 3.42206378661365e-05,
+      "loss": 0.8716,
+      "step": 123400
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 3.4211259895224626e-05,
+      "loss": 0.8738,
+      "step": 123600
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 3.420188192431274e-05,
+      "loss": 0.8681,
+      "step": 123800
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 3.419250395340087e-05,
+      "loss": 0.8669,
+      "step": 124000
+    },
+    {
+      "epoch": 0.15,
+      "eval_em_first_line_ratio": 0.473,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3602,
+      "eval_loss": 0.8231344819068909,
+      "eval_runtime": 251.988,
+      "eval_samples_per_second": 39.684,
+      "eval_steps_per_second": 0.623,
+      "step": 124000
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 3.418312598248898e-05,
+      "loss": 0.8743,
+      "step": 124200
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 3.417374801157711e-05,
+      "loss": 0.8712,
+      "step": 124400
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 3.416437004066523e-05,
+      "loss": 0.8707,
+      "step": 124600
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 3.415499206975335e-05,
+      "loss": 0.867,
+      "step": 124800
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 3.414561409884147e-05,
+      "loss": 0.8721,
+      "step": 125000
+    },
+    {
+      "epoch": 0.15,
+      "eval_em_first_line_ratio": 0.4718,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.36,
+      "eval_loss": 0.8221076130867004,
+      "eval_runtime": 262.4901,
+      "eval_samples_per_second": 38.097,
+      "eval_steps_per_second": 0.598,
+      "step": 125000
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 3.4136236127929594e-05,
+      "loss": 0.8676,
+      "step": 125200
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 3.4126858157017715e-05,
+      "loss": 0.872,
+      "step": 125400
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 3.4117480186105836e-05,
+      "loss": 0.8661,
+      "step": 125600
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 3.410810221519396e-05,
+      "loss": 0.8677,
+      "step": 125800
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 3.409872424428208e-05,
+      "loss": 0.8716,
+      "step": 126000
+    },
+    {
+      "epoch": 0.15,
+      "eval_em_first_line_ratio": 0.4708,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3588,
+      "eval_loss": 0.8223549723625183,
+      "eval_runtime": 252.2412,
+      "eval_samples_per_second": 39.645,
+      "eval_steps_per_second": 0.622,
+      "step": 126000
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 3.40893462733702e-05,
+      "loss": 0.8644,
+      "step": 126200
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 3.407996830245832e-05,
+      "loss": 0.8732,
+      "step": 126400
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 3.407059033154644e-05,
+      "loss": 0.8742,
+      "step": 126600
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 3.406121236063456e-05,
+      "loss": 0.8672,
+      "step": 126800
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 3.405183438972269e-05,
+      "loss": 0.8655,
+      "step": 127000
+    },
+    {
+      "epoch": 0.15,
+      "eval_em_first_line_ratio": 0.4718,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3595,
+      "eval_loss": 0.8206213712692261,
+      "eval_runtime": 251.9584,
+      "eval_samples_per_second": 39.689,
+      "eval_steps_per_second": 0.623,
+      "step": 127000
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 3.4042456418810803e-05,
+      "loss": 0.8699,
+      "step": 127200
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 3.403307844789893e-05,
+      "loss": 0.8682,
+      "step": 127400
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 3.4023700476987045e-05,
+      "loss": 0.8678,
+      "step": 127600
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 3.401432250607517e-05,
+      "loss": 0.8727,
+      "step": 127800
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 3.400494453516329e-05,
+      "loss": 0.8703,
+      "step": 128000
+    },
+    {
+      "epoch": 0.15,
+      "eval_em_first_line_ratio": 0.4713,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3595,
+      "eval_loss": 0.8201361298561096,
+      "eval_runtime": 249.155,
+      "eval_samples_per_second": 40.136,
+      "eval_steps_per_second": 0.63,
+      "step": 128000
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 3.3995566564251415e-05,
+      "loss": 0.8695,
+      "step": 128200
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 3.3986188593339536e-05,
+      "loss": 0.865,
+      "step": 128400
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 3.397681062242766e-05,
+      "loss": 0.8682,
+      "step": 128600
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 3.396743265151578e-05,
+      "loss": 0.8657,
+      "step": 128800
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 3.39580546806039e-05,
+      "loss": 0.8667,
+      "step": 129000
+    },
+    {
+      "epoch": 0.15,
+      "eval_em_first_line_ratio": 0.4685,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3569,
+      "eval_loss": 0.8203873038291931,
+      "eval_runtime": 252.3711,
+      "eval_samples_per_second": 39.624,
+      "eval_steps_per_second": 0.622,
+      "step": 129000
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 3.394867670969202e-05,
+      "loss": 0.866,
+      "step": 129200
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 3.393929873878014e-05,
+      "loss": 0.8704,
+      "step": 129400
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 3.392992076786826e-05,
+      "loss": 0.8646,
+      "step": 129600
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 3.3920589686810944e-05,
+      "loss": 0.8678,
+      "step": 129800
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 3.391125860575362e-05,
+      "loss": 0.8661,
+      "step": 130000
+    },
+    {
+      "epoch": 0.15,
+      "eval_em_first_line_ratio": 0.4678,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3558,
+      "eval_loss": 0.8190858364105225,
+      "eval_runtime": 243.8006,
+      "eval_samples_per_second": 41.017,
+      "eval_steps_per_second": 0.644,
+      "step": 130000
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 3.390188063484174e-05,
+      "loss": 0.8609,
+      "step": 130200
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 3.389250266392987e-05,
+      "loss": 0.8614,
+      "step": 130400
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 3.388312469301798e-05,
+      "loss": 0.8671,
+      "step": 130600
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 3.387374672210611e-05,
+      "loss": 0.8597,
+      "step": 130800
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 3.3864368751194225e-05,
+      "loss": 0.8719,
+      "step": 131000
+    },
+    {
+      "epoch": 0.15,
+      "eval_em_first_line_ratio": 0.4705,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3584,
+      "eval_loss": 0.8187699913978577,
+      "eval_runtime": 246.0491,
+      "eval_samples_per_second": 40.642,
+      "eval_steps_per_second": 0.638,
+      "step": 131000
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 3.385499078028235e-05,
+      "loss": 0.8684,
+      "step": 131200
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 3.384561280937047e-05,
+      "loss": 0.8626,
+      "step": 131400
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 3.3836234838458595e-05,
+      "loss": 0.8616,
+      "step": 131600
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 3.3826856867546716e-05,
+      "loss": 0.8693,
+      "step": 131800
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 3.381747889663484e-05,
+      "loss": 0.8597,
+      "step": 132000
+    },
+    {
+      "epoch": 0.15,
+      "eval_em_first_line_ratio": 0.4686,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.357,
+      "eval_loss": 0.8179601430892944,
+      "eval_runtime": 243.6137,
+      "eval_samples_per_second": 41.049,
+      "eval_steps_per_second": 0.644,
+      "step": 132000
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 3.380810092572296e-05,
+      "loss": 0.8621,
+      "step": 132200
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 3.379872295481108e-05,
+      "loss": 0.868,
+      "step": 132400
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 3.378939187375376e-05,
+      "loss": 0.8642,
+      "step": 132600
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 3.378001390284188e-05,
+      "loss": 0.8641,
+      "step": 132800
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 3.377063593193e-05,
+      "loss": 0.8659,
+      "step": 133000
+    },
+    {
+      "epoch": 0.16,
+      "eval_em_first_line_ratio": 0.4699,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3579,
+      "eval_loss": 0.8171144127845764,
+      "eval_runtime": 212.7027,
+      "eval_samples_per_second": 47.014,
+      "eval_steps_per_second": 0.738,
+      "step": 133000
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 3.3761257961018124e-05,
+      "loss": 0.8678,
+      "step": 133200
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 3.3751879990106245e-05,
+      "loss": 0.862,
+      "step": 133400
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 3.3742502019194366e-05,
+      "loss": 0.8655,
+      "step": 133600
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 3.373317093813705e-05,
+      "loss": 0.8638,
+      "step": 133800
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 3.372379296722517e-05,
+      "loss": 0.8634,
+      "step": 134000
+    },
+    {
+      "epoch": 0.16,
+      "eval_em_first_line_ratio": 0.4714,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3592,
+      "eval_loss": 0.8174598813056946,
+      "eval_runtime": 213.4756,
+      "eval_samples_per_second": 46.844,
+      "eval_steps_per_second": 0.735,
+      "step": 134000
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 3.371441499631329e-05,
+      "loss": 0.8624,
+      "step": 134200
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 3.370503702540141e-05,
+      "loss": 0.8653,
+      "step": 134400
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 3.369565905448953e-05,
+      "loss": 0.8632,
+      "step": 134600
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 3.3686281083577647e-05,
+      "loss": 0.8643,
+      "step": 134800
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 3.3676903112665774e-05,
+      "loss": 0.8591,
+      "step": 135000
+    },
+    {
+      "epoch": 0.16,
+      "eval_em_first_line_ratio": 0.4727,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3606,
+      "eval_loss": 0.8165234327316284,
+      "eval_runtime": 256.5033,
+      "eval_samples_per_second": 38.986,
+      "eval_steps_per_second": 0.612,
+      "step": 135000
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 3.3667525141753895e-05,
+      "loss": 0.8664,
+      "step": 135200
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 3.3658147170842016e-05,
+      "loss": 0.8638,
+      "step": 135400
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 3.364876919993014e-05,
+      "loss": 0.8567,
+      "step": 135600
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 3.363939122901826e-05,
+      "loss": 0.8544,
+      "step": 135800
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 3.363001325810638e-05,
+      "loss": 0.8612,
+      "step": 136000
+    },
+    {
+      "epoch": 0.16,
+      "eval_em_first_line_ratio": 0.473,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.361,
+      "eval_loss": 0.8164393305778503,
+      "eval_runtime": 250.26,
+      "eval_samples_per_second": 39.958,
+      "eval_steps_per_second": 0.627,
+      "step": 136000
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 3.36206352871945e-05,
+      "loss": 0.8684,
+      "step": 136200
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 3.361125731628262e-05,
+      "loss": 0.8638,
+      "step": 136400
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 3.360187934537074e-05,
+      "loss": 0.8609,
+      "step": 136600
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 3.359250137445886e-05,
+      "loss": 0.8573,
+      "step": 136800
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 3.3583123403546984e-05,
+      "loss": 0.8588,
+      "step": 137000
+    },
+    {
+      "epoch": 0.16,
+      "eval_em_first_line_ratio": 0.4722,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3603,
+      "eval_loss": 0.8162354230880737,
+      "eval_runtime": 258.3712,
+      "eval_samples_per_second": 38.704,
+      "eval_steps_per_second": 0.608,
+      "step": 137000
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 3.3573745432635105e-05,
+      "loss": 0.8622,
+      "step": 137200
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 3.356441435157779e-05,
+      "loss": 0.8638,
+      "step": 137400
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 3.355503638066591e-05,
+      "loss": 0.8606,
+      "step": 137600
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 3.354565840975403e-05,
+      "loss": 0.8548,
+      "step": 137800
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 3.353628043884215e-05,
+      "loss": 0.8576,
+      "step": 138000
+    },
+    {
+      "epoch": 0.16,
+      "eval_em_first_line_ratio": 0.4717,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3601,
+      "eval_loss": 0.8155922889709473,
+      "eval_runtime": 246.9019,
+      "eval_samples_per_second": 40.502,
+      "eval_steps_per_second": 0.636,
+      "step": 138000
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 3.352690246793027e-05,
+      "loss": 0.8581,
+      "step": 138200
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 3.351752449701839e-05,
+      "loss": 0.8584,
+      "step": 138400
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 3.3508193415961075e-05,
+      "loss": 0.8552,
+      "step": 138600
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 3.3498815445049196e-05,
+      "loss": 0.858,
+      "step": 138800
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 3.348943747413732e-05,
+      "loss": 0.8576,
+      "step": 139000
+    },
+    {
+      "epoch": 0.16,
+      "eval_em_first_line_ratio": 0.4717,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3596,
+      "eval_loss": 0.8149245381355286,
+      "eval_runtime": 234.6285,
+      "eval_samples_per_second": 42.621,
+      "eval_steps_per_second": 0.669,
+      "step": 139000
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 3.348005950322544e-05,
+      "loss": 0.8691,
+      "step": 139200
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 3.347068153231356e-05,
+      "loss": 0.8608,
+      "step": 139400
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 3.346130356140168e-05,
+      "loss": 0.8616,
+      "step": 139600
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 3.34519255904898e-05,
+      "loss": 0.861,
+      "step": 139800
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 3.344254761957793e-05,
+      "loss": 0.8582,
+      "step": 140000
+    },
+    {
+      "epoch": 0.16,
+      "eval_em_first_line_ratio": 0.4689,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3576,
+      "eval_loss": 0.8138406872749329,
+      "eval_runtime": 250.2727,
+      "eval_samples_per_second": 39.956,
+      "eval_steps_per_second": 0.627,
+      "step": 140000
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 3.343316964866604e-05,
+      "loss": 0.8601,
+      "step": 140200
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 3.342379167775417e-05,
+      "loss": 0.8602,
+      "step": 140400
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 3.3414413706842285e-05,
+      "loss": 0.8567,
+      "step": 140600
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 3.3405035735930406e-05,
+      "loss": 0.8639,
+      "step": 140800
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 3.339565776501853e-05,
+      "loss": 0.8568,
+      "step": 141000
+    },
+    {
+      "epoch": 0.17,
+      "eval_em_first_line_ratio": 0.474,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3619,
+      "eval_loss": 0.8144135475158691,
+      "eval_runtime": 242.0497,
+      "eval_samples_per_second": 41.314,
+      "eval_steps_per_second": 0.649,
+      "step": 141000
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 3.338627979410665e-05,
+      "loss": 0.8661,
+      "step": 141200
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 3.3376901823194775e-05,
+      "loss": 0.8609,
+      "step": 141400
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 3.336752385228289e-05,
+      "loss": 0.8582,
+      "step": 141600
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 3.335814588137102e-05,
+      "loss": 0.8583,
+      "step": 141800
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 3.334876791045913e-05,
+      "loss": 0.8608,
+      "step": 142000
+    },
+    {
+      "epoch": 0.17,
+      "eval_em_first_line_ratio": 0.4721,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3601,
+      "eval_loss": 0.8136679530143738,
+      "eval_runtime": 252.126,
+      "eval_samples_per_second": 39.663,
+      "eval_steps_per_second": 0.623,
+      "step": 142000
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 3.333938993954726e-05,
+      "loss": 0.8604,
+      "step": 142200
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 3.333001196863538e-05,
+      "loss": 0.8559,
+      "step": 142400
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 3.33206339977235e-05,
+      "loss": 0.8561,
+      "step": 142600
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 3.331125602681162e-05,
+      "loss": 0.86,
+      "step": 142800
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 3.330187805589974e-05,
+      "loss": 0.856,
+      "step": 143000
+    },
+    {
+      "epoch": 0.17,
+      "eval_em_first_line_ratio": 0.4741,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3618,
+      "eval_loss": 0.8129366040229797,
+      "eval_runtime": 241.558,
+      "eval_samples_per_second": 41.398,
+      "eval_steps_per_second": 0.65,
+      "step": 143000
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 3.3292500084987864e-05,
+      "loss": 0.8572,
+      "step": 143200
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 3.3283122114075985e-05,
+      "loss": 0.8608,
+      "step": 143400
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 3.3273744143164106e-05,
+      "loss": 0.856,
+      "step": 143600
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 3.326436617225223e-05,
+      "loss": 0.8544,
+      "step": 143800
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 3.325498820134035e-05,
+      "loss": 0.8551,
+      "step": 144000
+    },
+    {
+      "epoch": 0.17,
+      "eval_em_first_line_ratio": 0.4721,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3598,
+      "eval_loss": 0.8119027614593506,
+      "eval_runtime": 215.9521,
+      "eval_samples_per_second": 46.307,
+      "eval_steps_per_second": 0.727,
+      "step": 144000
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 3.324561023042847e-05,
+      "loss": 0.8598,
+      "step": 144200
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 3.323623225951659e-05,
+      "loss": 0.855,
+      "step": 144400
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 3.322685428860471e-05,
+      "loss": 0.8564,
+      "step": 144600
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 3.321747631769284e-05,
+      "loss": 0.8578,
+      "step": 144800
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 3.3208145236635514e-05,
+      "loss": 0.857,
+      "step": 145000
+    },
+    {
+      "epoch": 0.17,
+      "eval_em_first_line_ratio": 0.4728,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3606,
+      "eval_loss": 0.811653196811676,
+      "eval_runtime": 221.3083,
+      "eval_samples_per_second": 45.186,
+      "eval_steps_per_second": 0.709,
+      "step": 145000
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 3.3198767265723635e-05,
+      "loss": 0.858,
+      "step": 145200
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 3.3189389294811756e-05,
+      "loss": 0.8558,
+      "step": 145400
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 3.318005821375444e-05,
+      "loss": 0.8564,
+      "step": 145600
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 3.317068024284256e-05,
+      "loss": 0.8611,
+      "step": 145800
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 3.316130227193068e-05,
+      "loss": 0.8586,
+      "step": 146000
+    },
+    {
+      "epoch": 0.17,
+      "eval_em_first_line_ratio": 0.4726,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3608,
+      "eval_loss": 0.812241792678833,
+      "eval_runtime": 250.8062,
+      "eval_samples_per_second": 39.871,
+      "eval_steps_per_second": 0.626,
+      "step": 146000
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 3.31519243010188e-05,
+      "loss": 0.8552,
+      "step": 146200
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 3.314254633010692e-05,
+      "loss": 0.852,
+      "step": 146400
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 3.3133168359195044e-05,
+      "loss": 0.8562,
+      "step": 146600
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 3.3123790388283165e-05,
+      "loss": 0.8618,
+      "step": 146800
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 3.3114412417371286e-05,
+      "loss": 0.8565,
+      "step": 147000
+    },
+    {
+      "epoch": 0.17,
+      "eval_em_first_line_ratio": 0.4732,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.361,
+      "eval_loss": 0.8106628656387329,
+      "eval_runtime": 251.8659,
+      "eval_samples_per_second": 39.704,
+      "eval_steps_per_second": 0.623,
+      "step": 147000
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 3.3105034446459407e-05,
+      "loss": 0.8556,
+      "step": 147200
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 3.309565647554753e-05,
+      "loss": 0.8577,
+      "step": 147400
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 3.308627850463565e-05,
+      "loss": 0.8558,
+      "step": 147600
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 3.307690053372377e-05,
+      "loss": 0.8555,
+      "step": 147800
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 3.306752256281189e-05,
+      "loss": 0.8552,
+      "step": 148000
+    },
+    {
+      "epoch": 0.17,
+      "eval_em_first_line_ratio": 0.4735,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3612,
+      "eval_loss": 0.8102081418037415,
+      "eval_runtime": 240.0538,
+      "eval_samples_per_second": 41.657,
+      "eval_steps_per_second": 0.654,
+      "step": 148000
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 3.305814459190002e-05,
+      "loss": 0.8537,
+      "step": 148200
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 3.304876662098813e-05,
+      "loss": 0.8565,
+      "step": 148400
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 3.303938865007626e-05,
+      "loss": 0.8564,
+      "step": 148600
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 3.3030010679164374e-05,
+      "loss": 0.8504,
+      "step": 148800
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 3.30206327082525e-05,
+      "loss": 0.8616,
+      "step": 149000
+    },
+    {
+      "epoch": 0.17,
+      "eval_em_first_line_ratio": 0.4728,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.361,
+      "eval_loss": 0.809592604637146,
+      "eval_runtime": 243.2073,
+      "eval_samples_per_second": 41.117,
+      "eval_steps_per_second": 0.646,
+      "step": 149000
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 3.301125473734062e-05,
+      "loss": 0.8548,
+      "step": 149200
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 3.3001876766428744e-05,
+      "loss": 0.8567,
+      "step": 149400
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 3.2992498795516865e-05,
+      "loss": 0.852,
+      "step": 149600
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 3.2983120824604986e-05,
+      "loss": 0.8554,
+      "step": 149800
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 3.297374285369311e-05,
+      "loss": 0.8532,
+      "step": 150000
+    },
+    {
+      "epoch": 0.18,
+      "eval_em_first_line_ratio": 0.4707,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3589,
+      "eval_loss": 0.8096473217010498,
+      "eval_runtime": 257.8742,
+      "eval_samples_per_second": 38.779,
+      "eval_steps_per_second": 0.609,
+      "step": 150000
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 3.296436488278123e-05,
+      "loss": 0.8605,
+      "step": 150200
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 3.295498691186935e-05,
+      "loss": 0.8555,
+      "step": 150400
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 3.294560894095747e-05,
+      "loss": 0.8514,
+      "step": 150600
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 3.293623097004559e-05,
+      "loss": 0.8542,
+      "step": 150800
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 3.292685299913371e-05,
+      "loss": 0.8575,
+      "step": 151000
+    },
+    {
+      "epoch": 0.18,
+      "eval_em_first_line_ratio": 0.472,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3596,
+      "eval_loss": 0.8092576861381531,
+      "eval_runtime": 245.3518,
+      "eval_samples_per_second": 40.758,
+      "eval_steps_per_second": 0.64,
+      "step": 151000
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 3.291747502822183e-05,
+      "loss": 0.8534,
+      "step": 151200
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 3.2908097057309954e-05,
+      "loss": 0.8542,
+      "step": 151400
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 3.2898719086398075e-05,
+      "loss": 0.852,
+      "step": 151600
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 3.2889341115486196e-05,
+      "loss": 0.8499,
+      "step": 151800
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 3.288001003442888e-05,
+      "loss": 0.852,
+      "step": 152000
+    },
+    {
+      "epoch": 0.18,
+      "eval_em_first_line_ratio": 0.4707,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3585,
+      "eval_loss": 0.8083478808403015,
+      "eval_runtime": 247.5198,
+      "eval_samples_per_second": 40.401,
+      "eval_steps_per_second": 0.634,
+      "step": 152000
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 3.2870632063517e-05,
+      "loss": 0.857,
+      "step": 152200
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 3.286125409260512e-05,
+      "loss": 0.8505,
+      "step": 152400
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 3.285187612169324e-05,
+      "loss": 0.8558,
+      "step": 152600
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 3.284249815078136e-05,
+      "loss": 0.8593,
+      "step": 152800
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 3.283312017986949e-05,
+      "loss": 0.8548,
+      "step": 153000
+    },
+    {
+      "epoch": 0.18,
+      "eval_em_first_line_ratio": 0.4724,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3598,
+      "eval_loss": 0.8087502121925354,
+      "eval_runtime": 248.8783,
+      "eval_samples_per_second": 40.18,
+      "eval_steps_per_second": 0.631,
+      "step": 153000
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 3.2823742208957604e-05,
+      "loss": 0.8478,
+      "step": 153200
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 3.2814364238045725e-05,
+      "loss": 0.8544,
+      "step": 153400
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 3.2804986267133846e-05,
+      "loss": 0.8518,
+      "step": 153600
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 3.279560829622197e-05,
+      "loss": 0.8549,
+      "step": 153800
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 3.2786230325310095e-05,
+      "loss": 0.8499,
+      "step": 154000
+    },
+    {
+      "epoch": 0.18,
+      "eval_em_first_line_ratio": 0.4727,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3602,
+      "eval_loss": 0.8083200454711914,
+      "eval_runtime": 245.1164,
+      "eval_samples_per_second": 40.797,
+      "eval_steps_per_second": 0.641,
+      "step": 154000
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 3.277685235439821e-05,
+      "loss": 0.8588,
+      "step": 154200
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 3.276752127334089e-05,
+      "loss": 0.8572,
+      "step": 154400
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 3.2758190192283574e-05,
+      "loss": 0.8545,
+      "step": 154600
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 3.2748812221371695e-05,
+      "loss": 0.8559,
+      "step": 154800
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 3.2739434250459816e-05,
+      "loss": 0.8524,
+      "step": 155000
+    },
+    {
+      "epoch": 0.18,
+      "eval_em_first_line_ratio": 0.4691,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3567,
+      "eval_loss": 0.8076086044311523,
+      "eval_runtime": 208.9504,
+      "eval_samples_per_second": 47.858,
+      "eval_steps_per_second": 0.751,
+      "step": 155000
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 3.273005627954794e-05,
+      "loss": 0.856,
+      "step": 155200
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 3.272067830863606e-05,
+      "loss": 0.8483,
+      "step": 155400
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 3.271130033772418e-05,
+      "loss": 0.8512,
+      "step": 155600
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 3.27019223668123e-05,
+      "loss": 0.8494,
+      "step": 155800
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 3.269254439590042e-05,
+      "loss": 0.8522,
+      "step": 156000
+    },
+    {
+      "epoch": 0.18,
+      "eval_em_first_line_ratio": 0.4728,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3602,
+      "eval_loss": 0.8068527579307556,
+      "eval_runtime": 217.3871,
+      "eval_samples_per_second": 46.001,
+      "eval_steps_per_second": 0.722,
+      "step": 156000
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 3.268316642498854e-05,
+      "loss": 0.8556,
+      "step": 156200
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 3.267378845407667e-05,
+      "loss": 0.8512,
+      "step": 156400
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 3.2664410483164784e-05,
+      "loss": 0.8525,
+      "step": 156600
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 3.265503251225291e-05,
+      "loss": 0.8521,
+      "step": 156800
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 3.2645654541341026e-05,
+      "loss": 0.8536,
+      "step": 157000
+    },
+    {
+      "epoch": 0.18,
+      "eval_em_first_line_ratio": 0.4704,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3578,
+      "eval_loss": 0.8072694540023804,
+      "eval_runtime": 246.0695,
+      "eval_samples_per_second": 40.639,
+      "eval_steps_per_second": 0.638,
+      "step": 157000
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 3.263627657042915e-05,
+      "loss": 0.8502,
+      "step": 157200
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 3.2626898599517274e-05,
+      "loss": 0.8528,
+      "step": 157400
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 3.2617520628605395e-05,
+      "loss": 0.8504,
+      "step": 157600
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 3.2608142657693516e-05,
+      "loss": 0.8506,
+      "step": 157800
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 3.259876468678164e-05,
+      "loss": 0.8477,
+      "step": 158000
+    },
+    {
+      "epoch": 0.19,
+      "eval_em_first_line_ratio": 0.4737,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3613,
+      "eval_loss": 0.8060622811317444,
+      "eval_runtime": 245.9526,
+      "eval_samples_per_second": 40.658,
+      "eval_steps_per_second": 0.638,
+      "step": 158000
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 3.258938671586976e-05,
+      "loss": 0.8482,
+      "step": 158200
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 3.258000874495788e-05,
+      "loss": 0.8501,
+      "step": 158400
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 3.257067766390056e-05,
+      "loss": 0.8508,
+      "step": 158600
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 3.256129969298868e-05,
+      "loss": 0.8498,
+      "step": 158800
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 3.25519217220768e-05,
+      "loss": 0.8508,
+      "step": 159000
+    },
+    {
+      "epoch": 0.19,
+      "eval_em_first_line_ratio": 0.4732,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3609,
+      "eval_loss": 0.8056519031524658,
+      "eval_runtime": 241.0944,
+      "eval_samples_per_second": 41.478,
+      "eval_steps_per_second": 0.651,
+      "step": 159000
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 3.2542543751164925e-05,
+      "loss": 0.8523,
+      "step": 159200
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 3.253316578025304e-05,
+      "loss": 0.8535,
+      "step": 159400
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 3.2523787809341167e-05,
+      "loss": 0.8532,
+      "step": 159600
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 3.251440983842929e-05,
+      "loss": 0.856,
+      "step": 159800
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 3.250503186751741e-05,
+      "loss": 0.8441,
+      "step": 160000
+    },
+    {
+      "epoch": 0.19,
+      "eval_em_first_line_ratio": 0.4707,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3582,
+      "eval_loss": 0.8057369589805603,
+      "eval_runtime": 253.8368,
+      "eval_samples_per_second": 39.395,
+      "eval_steps_per_second": 0.619,
+      "step": 160000
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 3.249570078646009e-05,
+      "loss": 0.8462,
+      "step": 160200
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 3.2486322815548205e-05,
+      "loss": 0.8481,
+      "step": 160400
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 3.247694484463633e-05,
+      "loss": 0.8521,
+      "step": 160600
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 3.2467566873724454e-05,
+      "loss": 0.8495,
+      "step": 160800
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 3.2458188902812575e-05,
+      "loss": 0.8528,
+      "step": 161000
+    },
+    {
+      "epoch": 0.19,
+      "eval_em_first_line_ratio": 0.4734,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3608,
+      "eval_loss": 0.8051938414573669,
+      "eval_runtime": 246.3141,
+      "eval_samples_per_second": 40.599,
+      "eval_steps_per_second": 0.637,
+      "step": 161000
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 3.2448810931900696e-05,
+      "loss": 0.8485,
+      "step": 161200
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 3.243943296098882e-05,
+      "loss": 0.8506,
+      "step": 161400
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 3.243005499007694e-05,
+      "loss": 0.8518,
+      "step": 161600
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 3.242067701916506e-05,
+      "loss": 0.8505,
+      "step": 161800
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 3.241129904825318e-05,
+      "loss": 0.846,
+      "step": 162000
+    },
+    {
+      "epoch": 0.19,
+      "eval_em_first_line_ratio": 0.4734,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3612,
+      "eval_loss": 0.8049929141998291,
+      "eval_runtime": 245.4818,
+      "eval_samples_per_second": 40.736,
+      "eval_steps_per_second": 0.64,
+      "step": 162000
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 3.24019210773413e-05,
+      "loss": 0.848,
+      "step": 162200
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 3.239254310642942e-05,
+      "loss": 0.8516,
+      "step": 162400
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 3.238316513551754e-05,
+      "loss": 0.8498,
+      "step": 162600
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 3.2373787164605664e-05,
+      "loss": 0.8516,
+      "step": 162800
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 3.2364409193693785e-05,
+      "loss": 0.844,
+      "step": 163000
+    },
+    {
+      "epoch": 0.19,
+      "eval_em_first_line_ratio": 0.473,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.361,
+      "eval_loss": 0.8048765659332275,
+      "eval_runtime": 234.1293,
+      "eval_samples_per_second": 42.711,
+      "eval_steps_per_second": 0.671,
+      "step": 163000
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 3.2355031222781906e-05,
+      "loss": 0.849,
+      "step": 163200
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 3.2345653251870027e-05,
+      "loss": 0.8513,
+      "step": 163400
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 3.2336275280958154e-05,
+      "loss": 0.849,
+      "step": 163600
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 3.232689731004627e-05,
+      "loss": 0.848,
+      "step": 163800
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 3.231756622898895e-05,
+      "loss": 0.8491,
+      "step": 164000
+    },
+    {
+      "epoch": 0.19,
+      "eval_em_first_line_ratio": 0.4745,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3625,
+      "eval_loss": 0.8042445778846741,
+      "eval_runtime": 248.2532,
+      "eval_samples_per_second": 40.281,
+      "eval_steps_per_second": 0.632,
+      "step": 164000
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 3.230818825807707e-05,
+      "loss": 0.8455,
+      "step": 164200
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 3.229881028716519e-05,
+      "loss": 0.8435,
+      "step": 164400
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 3.228943231625332e-05,
+      "loss": 0.8467,
+      "step": 164600
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 3.2280054345341435e-05,
+      "loss": 0.8461,
+      "step": 164800
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 3.2270676374429556e-05,
+      "loss": 0.852,
+      "step": 165000
+    },
+    {
+      "epoch": 0.19,
+      "eval_em_first_line_ratio": 0.4727,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3605,
+      "eval_loss": 0.8038452863693237,
+      "eval_runtime": 244.7668,
+      "eval_samples_per_second": 40.855,
+      "eval_steps_per_second": 0.641,
+      "step": 165000
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 3.226129840351768e-05,
+      "loss": 0.8456,
+      "step": 165200
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 3.22519204326058e-05,
+      "loss": 0.8487,
+      "step": 165400
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 3.2242542461693926e-05,
+      "loss": 0.8444,
+      "step": 165600
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 3.223316449078204e-05,
+      "loss": 0.8493,
+      "step": 165800
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 3.222378651987017e-05,
+      "loss": 0.8469,
+      "step": 166000
+    },
+    {
+      "epoch": 0.19,
+      "eval_em_first_line_ratio": 0.4726,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3605,
+      "eval_loss": 0.8036503195762634,
+      "eval_runtime": 254.618,
+      "eval_samples_per_second": 39.275,
+      "eval_steps_per_second": 0.617,
+      "step": 166000
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 3.221440854895828e-05,
+      "loss": 0.845,
+      "step": 166200
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 3.220503057804641e-05,
+      "loss": 0.8496,
+      "step": 166400
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 3.2195652607134524e-05,
+      "loss": 0.8379,
+      "step": 166600
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 3.218627463622265e-05,
+      "loss": 0.843,
+      "step": 166800
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 3.217689666531077e-05,
+      "loss": 0.8436,
+      "step": 167000
+    },
+    {
+      "epoch": 0.2,
+      "eval_em_first_line_ratio": 0.4727,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3605,
+      "eval_loss": 0.8027209639549255,
+      "eval_runtime": 223.7552,
+      "eval_samples_per_second": 44.692,
+      "eval_steps_per_second": 0.702,
+      "step": 167000
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 3.2167518694398893e-05,
+      "loss": 0.8445,
+      "step": 167200
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 3.2158140723487014e-05,
+      "loss": 0.8489,
+      "step": 167400
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 3.2148762752575135e-05,
+      "loss": 0.8462,
+      "step": 167600
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 3.2139384781663256e-05,
+      "loss": 0.8456,
+      "step": 167800
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 3.213000681075138e-05,
+      "loss": 0.8484,
+      "step": 168000
+    },
+    {
+      "epoch": 0.2,
+      "eval_em_first_line_ratio": 0.4742,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3619,
+      "eval_loss": 0.8021925687789917,
+      "eval_runtime": 220.7613,
+      "eval_samples_per_second": 45.298,
+      "eval_steps_per_second": 0.711,
+      "step": 168000
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 3.21206288398395e-05,
+      "loss": 0.8422,
+      "step": 168200
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 3.211125086892762e-05,
+      "loss": 0.8438,
+      "step": 168400
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 3.21019197878703e-05,
+      "loss": 0.8432,
+      "step": 168600
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 3.209254181695842e-05,
+      "loss": 0.8485,
+      "step": 168800
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 3.2083163846046544e-05,
+      "loss": 0.8413,
+      "step": 169000
+    },
+    {
+      "epoch": 0.2,
+      "eval_em_first_line_ratio": 0.472,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3604,
+      "eval_loss": 0.8023744225502014,
+      "eval_runtime": 247.4726,
+      "eval_samples_per_second": 40.409,
+      "eval_steps_per_second": 0.634,
+      "step": 169000
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 3.2073785875134665e-05,
+      "loss": 0.8448,
+      "step": 169200
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 3.2064407904222786e-05,
+      "loss": 0.8408,
+      "step": 169400
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 3.2055029933310907e-05,
+      "loss": 0.8534,
+      "step": 169600
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 3.204565196239903e-05,
+      "loss": 0.8414,
+      "step": 169800
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 3.203632088134171e-05,
+      "loss": 0.8391,
+      "step": 170000
+    },
+    {
+      "epoch": 0.2,
+      "eval_em_first_line_ratio": 0.4712,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3588,
+      "eval_loss": 0.8016086220741272,
+      "eval_runtime": 254.0243,
+      "eval_samples_per_second": 39.366,
+      "eval_steps_per_second": 0.618,
+      "step": 170000
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 3.202694291042983e-05,
+      "loss": 0.8466,
+      "step": 170200
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 3.201756493951795e-05,
+      "loss": 0.8416,
+      "step": 170400
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 3.200818696860607e-05,
+      "loss": 0.8481,
+      "step": 170600
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 3.1998808997694194e-05,
+      "loss": 0.8471,
+      "step": 170800
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 3.1989431026782315e-05,
+      "loss": 0.8455,
+      "step": 171000
+    },
+    {
+      "epoch": 0.2,
+      "eval_em_first_line_ratio": 0.4754,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3637,
+      "eval_loss": 0.801468551158905,
+      "eval_runtime": 248.0668,
+      "eval_samples_per_second": 40.312,
+      "eval_steps_per_second": 0.633,
+      "step": 171000
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 3.1980053055870436e-05,
+      "loss": 0.8487,
+      "step": 171200
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 3.197067508495856e-05,
+      "loss": 0.843,
+      "step": 171400
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 3.196129711404668e-05,
+      "loss": 0.8506,
+      "step": 171600
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 3.19519191431348e-05,
+      "loss": 0.8465,
+      "step": 171800
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 3.194254117222292e-05,
+      "loss": 0.8432,
+      "step": 172000
+    },
+    {
+      "epoch": 0.2,
+      "eval_em_first_line_ratio": 0.4739,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.362,
+      "eval_loss": 0.8002371788024902,
+      "eval_runtime": 249.03,
+      "eval_samples_per_second": 40.156,
+      "eval_steps_per_second": 0.63,
+      "step": 172000
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 3.193316320131104e-05,
+      "loss": 0.8437,
+      "step": 172200
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 3.192378523039916e-05,
+      "loss": 0.8445,
+      "step": 172400
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 3.191440725948728e-05,
+      "loss": 0.8421,
+      "step": 172600
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 3.190502928857541e-05,
+      "loss": 0.8431,
+      "step": 172800
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 3.1895651317663525e-05,
+      "loss": 0.8427,
+      "step": 173000
+    },
+    {
+      "epoch": 0.2,
+      "eval_em_first_line_ratio": 0.473,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.361,
+      "eval_loss": 0.8012005090713501,
+      "eval_runtime": 248.9011,
+      "eval_samples_per_second": 40.177,
+      "eval_steps_per_second": 0.631,
+      "step": 173000
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 3.188627334675165e-05,
+      "loss": 0.8444,
+      "step": 173200
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 3.187689537583977e-05,
+      "loss": 0.8438,
+      "step": 173400
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 3.1867517404927894e-05,
+      "loss": 0.8427,
+      "step": 173600
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 3.185813943401601e-05,
+      "loss": 0.8374,
+      "step": 173800
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 3.184880835295869e-05,
+      "loss": 0.8469,
+      "step": 174000
+    },
+    {
+      "epoch": 0.2,
+      "eval_em_first_line_ratio": 0.4709,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3588,
+      "eval_loss": 0.8002498745918274,
+      "eval_runtime": 253.6252,
+      "eval_samples_per_second": 39.428,
+      "eval_steps_per_second": 0.619,
+      "step": 174000
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 3.183943038204682e-05,
+      "loss": 0.8444,
+      "step": 174200
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 3.183005241113493e-05,
+      "loss": 0.8392,
+      "step": 174400
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 3.182067444022306e-05,
+      "loss": 0.8405,
+      "step": 174600
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 3.1811296469311175e-05,
+      "loss": 0.8411,
+      "step": 174800
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 3.180196538825386e-05,
+      "loss": 0.8479,
+      "step": 175000
+    },
+    {
+      "epoch": 0.21,
+      "eval_em_first_line_ratio": 0.4696,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3575,
+      "eval_loss": 0.8004719614982605,
+      "eval_runtime": 258.2788,
+      "eval_samples_per_second": 38.718,
+      "eval_steps_per_second": 0.608,
+      "step": 175000
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 3.1792587417341985e-05,
+      "loss": 0.8408,
+      "step": 175200
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 3.17832094464301e-05,
+      "loss": 0.8434,
+      "step": 175400
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 3.177383147551823e-05,
+      "loss": 0.8459,
+      "step": 175600
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 3.176445350460634e-05,
+      "loss": 0.8447,
+      "step": 175800
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 3.175507553369447e-05,
+      "loss": 0.8468,
+      "step": 176000
+    },
+    {
+      "epoch": 0.21,
+      "eval_em_first_line_ratio": 0.473,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.361,
+      "eval_loss": 0.8006030321121216,
+      "eval_runtime": 237.8112,
+      "eval_samples_per_second": 42.05,
+      "eval_steps_per_second": 0.66,
+      "step": 176000
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 3.174569756278259e-05,
+      "loss": 0.8397,
+      "step": 176200
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 3.173631959187071e-05,
+      "loss": 0.8424,
+      "step": 176400
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 3.172694162095883e-05,
+      "loss": 0.8479,
+      "step": 176600
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 3.171756365004695e-05,
+      "loss": 0.8468,
+      "step": 176800
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 3.1708185679135074e-05,
+      "loss": 0.8398,
+      "step": 177000
+    },
+    {
+      "epoch": 0.21,
+      "eval_em_first_line_ratio": 0.4715,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3597,
+      "eval_loss": 0.7996644377708435,
+      "eval_runtime": 246.1176,
+      "eval_samples_per_second": 40.631,
+      "eval_steps_per_second": 0.638,
+      "step": 177000
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 3.169880770822319e-05,
+      "loss": 0.8424,
+      "step": 177200
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 3.1689429737311316e-05,
+      "loss": 0.8408,
+      "step": 177400
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 3.168005176639944e-05,
+      "loss": 0.8441,
+      "step": 177600
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 3.167067379548756e-05,
+      "loss": 0.8415,
+      "step": 177800
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 3.166129582457568e-05,
+      "loss": 0.8383,
+      "step": 178000
+    },
+    {
+      "epoch": 0.21,
+      "eval_em_first_line_ratio": 0.4722,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.36,
+      "eval_loss": 0.7991567254066467,
+      "eval_runtime": 217.1321,
+      "eval_samples_per_second": 46.055,
+      "eval_steps_per_second": 0.723,
+      "step": 178000
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 3.16519178536638e-05,
+      "loss": 0.839,
+      "step": 178200
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 3.164253988275192e-05,
+      "loss": 0.8426,
+      "step": 178400
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 3.163316191184004e-05,
+      "loss": 0.8359,
+      "step": 178600
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 3.162378394092816e-05,
+      "loss": 0.8403,
+      "step": 178800
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 3.1614405970016284e-05,
+      "loss": 0.8426,
+      "step": 179000
+    },
+    {
+      "epoch": 0.21,
+      "eval_em_first_line_ratio": 0.4727,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3605,
+      "eval_loss": 0.7980639934539795,
+      "eval_runtime": 214.581,
+      "eval_samples_per_second": 46.602,
+      "eval_steps_per_second": 0.732,
+      "step": 179000
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 3.1605027999104405e-05,
+      "loss": 0.8484,
+      "step": 179200
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 3.1595650028192526e-05,
+      "loss": 0.8423,
+      "step": 179400
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 3.158627205728065e-05,
+      "loss": 0.8484,
+      "step": 179600
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 3.157689408636877e-05,
+      "loss": 0.8379,
+      "step": 179800
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 3.1567516115456895e-05,
+      "loss": 0.8396,
+      "step": 180000
+    },
+    {
+      "epoch": 0.21,
+      "eval_em_first_line_ratio": 0.4756,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3631,
+      "eval_loss": 0.7975685596466064,
+      "eval_runtime": 241.2382,
+      "eval_samples_per_second": 41.453,
+      "eval_steps_per_second": 0.651,
+      "step": 180000
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 3.155813814454501e-05,
+      "loss": 0.8483,
+      "step": 180200
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 3.154876017363314e-05,
+      "loss": 0.8389,
+      "step": 180400
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 3.153942909257581e-05,
+      "loss": 0.836,
+      "step": 180600
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 3.1530051121663934e-05,
+      "loss": 0.8415,
+      "step": 180800
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 3.152067315075206e-05,
+      "loss": 0.8435,
+      "step": 181000
+    },
+    {
+      "epoch": 0.21,
+      "eval_em_first_line_ratio": 0.4744,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3619,
+      "eval_loss": 0.7978988885879517,
+      "eval_runtime": 251.2407,
+      "eval_samples_per_second": 39.802,
+      "eval_steps_per_second": 0.625,
+      "step": 181000
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 3.1511295179840176e-05,
+      "loss": 0.8387,
+      "step": 181200
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 3.1501917208928304e-05,
+      "loss": 0.8405,
+      "step": 181400
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 3.149253923801642e-05,
+      "loss": 0.8375,
+      "step": 181600
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 3.1483161267104546e-05,
+      "loss": 0.8447,
+      "step": 181800
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 3.147378329619266e-05,
+      "loss": 0.8396,
+      "step": 182000
+    },
+    {
+      "epoch": 0.21,
+      "eval_em_first_line_ratio": 0.4737,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3608,
+      "eval_loss": 0.7974121570587158,
+      "eval_runtime": 242.8031,
+      "eval_samples_per_second": 41.186,
+      "eval_steps_per_second": 0.647,
+      "step": 182000
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 3.146440532528079e-05,
+      "loss": 0.8395,
+      "step": 182200
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 3.145507424422347e-05,
+      "loss": 0.8381,
+      "step": 182400
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 3.1445696273311584e-05,
+      "loss": 0.8376,
+      "step": 182600
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 3.143631830239971e-05,
+      "loss": 0.8394,
+      "step": 182800
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 3.1426940331487826e-05,
+      "loss": 0.8395,
+      "step": 183000
+    },
+    {
+      "epoch": 0.21,
+      "eval_em_first_line_ratio": 0.4763,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3635,
+      "eval_loss": 0.7981505990028381,
+      "eval_runtime": 241.0977,
+      "eval_samples_per_second": 41.477,
+      "eval_steps_per_second": 0.651,
+      "step": 183000
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 3.141756236057595e-05,
+      "loss": 0.8379,
+      "step": 183200
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 3.1408184389664075e-05,
+      "loss": 0.8377,
+      "step": 183400
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 3.139880641875219e-05,
+      "loss": 0.8428,
+      "step": 183600
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 3.138942844784032e-05,
+      "loss": 0.8365,
+      "step": 183800
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 3.138005047692843e-05,
+      "loss": 0.841,
+      "step": 184000
+    },
+    {
+      "epoch": 0.22,
+      "eval_em_first_line_ratio": 0.4757,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3636,
+      "eval_loss": 0.7963166236877441,
+      "eval_runtime": 240.1704,
+      "eval_samples_per_second": 41.637,
+      "eval_steps_per_second": 0.654,
+      "step": 184000
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 3.137067250601656e-05,
+      "loss": 0.8389,
+      "step": 184200
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 3.136129453510468e-05,
+      "loss": 0.8379,
+      "step": 184400
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 3.1351963454047356e-05,
+      "loss": 0.834,
+      "step": 184600
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 3.134258548313548e-05,
+      "loss": 0.84,
+      "step": 184800
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 3.13332075122236e-05,
+      "loss": 0.8382,
+      "step": 185000
+    },
+    {
+      "epoch": 0.22,
+      "eval_em_first_line_ratio": 0.4769,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3641,
+      "eval_loss": 0.7974162101745605,
+      "eval_runtime": 246.8213,
+      "eval_samples_per_second": 40.515,
+      "eval_steps_per_second": 0.636,
+      "step": 185000
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 3.1323829541311725e-05,
+      "loss": 0.8437,
+      "step": 185200
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 3.131445157039984e-05,
+      "loss": 0.8392,
+      "step": 185400
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 3.130507359948797e-05,
+      "loss": 0.8365,
+      "step": 185600
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 3.129569562857609e-05,
+      "loss": 0.8348,
+      "step": 185800
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 3.128631765766421e-05,
+      "loss": 0.8375,
+      "step": 186000
+    },
+    {
+      "epoch": 0.22,
+      "eval_em_first_line_ratio": 0.4761,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3633,
+      "eval_loss": 0.7956709861755371,
+      "eval_runtime": 240.3479,
+      "eval_samples_per_second": 41.606,
+      "eval_steps_per_second": 0.653,
+      "step": 186000
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 3.127693968675233e-05,
+      "loss": 0.8352,
+      "step": 186200
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 3.126756171584045e-05,
+      "loss": 0.8376,
+      "step": 186400
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 3.125818374492857e-05,
+      "loss": 0.8323,
+      "step": 186600
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 3.124880577401669e-05,
+      "loss": 0.8382,
+      "step": 186800
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 3.1239427803104814e-05,
+      "loss": 0.8362,
+      "step": 187000
+    },
+    {
+      "epoch": 0.22,
+      "eval_em_first_line_ratio": 0.4763,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3639,
+      "eval_loss": 0.7957203984260559,
+      "eval_runtime": 246.6376,
+      "eval_samples_per_second": 40.545,
+      "eval_steps_per_second": 0.637,
+      "step": 187000
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 3.1230049832192935e-05,
+      "loss": 0.84,
+      "step": 187200
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 3.1220671861281056e-05,
+      "loss": 0.8374,
+      "step": 187400
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 3.121129389036918e-05,
+      "loss": 0.835,
+      "step": 187600
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 3.12019159194573e-05,
+      "loss": 0.8407,
+      "step": 187800
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 3.119253794854542e-05,
+      "loss": 0.8402,
+      "step": 188000
+    },
+    {
+      "epoch": 0.22,
+      "eval_em_first_line_ratio": 0.4776,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.365,
+      "eval_loss": 0.7950745820999146,
+      "eval_runtime": 241.0135,
+      "eval_samples_per_second": 41.491,
+      "eval_steps_per_second": 0.651,
+      "step": 188000
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 3.1183159977633547e-05,
+      "loss": 0.8313,
+      "step": 188200
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 3.117378200672166e-05,
+      "loss": 0.8373,
+      "step": 188400
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 3.116440403580979e-05,
+      "loss": 0.8329,
+      "step": 188600
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 3.11550260648979e-05,
+      "loss": 0.8366,
+      "step": 188800
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 3.1145694983840585e-05,
+      "loss": 0.8388,
+      "step": 189000
+    },
+    {
+      "epoch": 0.22,
+      "eval_em_first_line_ratio": 0.4764,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3643,
+      "eval_loss": 0.795382022857666,
+      "eval_runtime": 219.2458,
+      "eval_samples_per_second": 45.611,
+      "eval_steps_per_second": 0.716,
+      "step": 189000
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 3.113636390278327e-05,
+      "loss": 0.835,
+      "step": 189200
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 3.112698593187139e-05,
+      "loss": 0.8385,
+      "step": 189400
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 3.111765485081407e-05,
+      "loss": 0.8373,
+      "step": 189600
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 3.1108276879902185e-05,
+      "loss": 0.834,
+      "step": 189800
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 3.109889890899031e-05,
+      "loss": 0.8353,
+      "step": 190000
+    },
+    {
+      "epoch": 0.22,
+      "eval_em_first_line_ratio": 0.4766,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3643,
+      "eval_loss": 0.7955102324485779,
+      "eval_runtime": 210.2488,
+      "eval_samples_per_second": 47.563,
+      "eval_steps_per_second": 0.747,
+      "step": 190000
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 3.1089520938078434e-05,
+      "loss": 0.8368,
+      "step": 190200
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 3.1080142967166555e-05,
+      "loss": 0.8366,
+      "step": 190400
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 3.1070764996254676e-05,
+      "loss": 0.8378,
+      "step": 190600
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 3.10613870253428e-05,
+      "loss": 0.8324,
+      "step": 190800
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 3.105200905443092e-05,
+      "loss": 0.8379,
+      "step": 191000
+    },
+    {
+      "epoch": 0.22,
+      "eval_em_first_line_ratio": 0.4767,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3643,
+      "eval_loss": 0.7949255704879761,
+      "eval_runtime": 240.6465,
+      "eval_samples_per_second": 41.555,
+      "eval_steps_per_second": 0.652,
+      "step": 191000
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 3.104263108351904e-05,
+      "loss": 0.8318,
+      "step": 191200
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 3.103325311260716e-05,
+      "loss": 0.8367,
+      "step": 191400
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 3.102387514169528e-05,
+      "loss": 0.8355,
+      "step": 191600
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 3.10144971707834e-05,
+      "loss": 0.8345,
+      "step": 191800
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 3.100511919987152e-05,
+      "loss": 0.8364,
+      "step": 192000
+    },
+    {
+      "epoch": 0.23,
+      "eval_em_first_line_ratio": 0.476,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3634,
+      "eval_loss": 0.794501781463623,
+      "eval_runtime": 251.3475,
+      "eval_samples_per_second": 39.786,
+      "eval_steps_per_second": 0.625,
+      "step": 192000
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 3.0995741228959644e-05,
+      "loss": 0.8405,
+      "step": 192200
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 3.0986363258047765e-05,
+      "loss": 0.835,
+      "step": 192400
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 3.097698528713589e-05,
+      "loss": 0.8337,
+      "step": 192600
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 3.096760731622401e-05,
+      "loss": 0.8323,
+      "step": 192800
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 3.0958229345312135e-05,
+      "loss": 0.8339,
+      "step": 193000
+    },
+    {
+      "epoch": 0.23,
+      "eval_em_first_line_ratio": 0.4781,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3654,
+      "eval_loss": 0.7949613332748413,
+      "eval_runtime": 246.7528,
+      "eval_samples_per_second": 40.526,
+      "eval_steps_per_second": 0.636,
+      "step": 193000
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 3.094885137440025e-05,
+      "loss": 0.8356,
+      "step": 193200
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 3.0939473403488376e-05,
+      "loss": 0.8372,
+      "step": 193400
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 3.093009543257649e-05,
+      "loss": 0.8392,
+      "step": 193600
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 3.092071746166462e-05,
+      "loss": 0.8387,
+      "step": 193800
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 3.091133949075274e-05,
+      "loss": 0.834,
+      "step": 194000
+    },
+    {
+      "epoch": 0.23,
+      "eval_em_first_line_ratio": 0.4747,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3623,
+      "eval_loss": 0.7940728664398193,
+      "eval_runtime": 250.6976,
+      "eval_samples_per_second": 39.889,
+      "eval_steps_per_second": 0.626,
+      "step": 194000
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 3.090196151984086e-05,
+      "loss": 0.8342,
+      "step": 194200
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 3.089258354892898e-05,
+      "loss": 0.8356,
+      "step": 194400
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 3.088325246787166e-05,
+      "loss": 0.8373,
+      "step": 194600
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 3.0873874496959785e-05,
+      "loss": 0.8376,
+      "step": 194800
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 3.0864496526047906e-05,
+      "loss": 0.8371,
+      "step": 195000
+    },
+    {
+      "epoch": 0.23,
+      "eval_em_first_line_ratio": 0.4763,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3636,
+      "eval_loss": 0.7943687438964844,
+      "eval_runtime": 241.1152,
+      "eval_samples_per_second": 41.474,
+      "eval_steps_per_second": 0.651,
+      "step": 195000
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 3.085511855513602e-05,
+      "loss": 0.8266,
+      "step": 195200
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 3.084574058422415e-05,
+      "loss": 0.8319,
+      "step": 195400
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 3.083636261331226e-05,
+      "loss": 0.8304,
+      "step": 195600
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 3.082698464240039e-05,
+      "loss": 0.8351,
+      "step": 195800
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 3.081760667148851e-05,
+      "loss": 0.8372,
+      "step": 196000
+    },
+    {
+      "epoch": 0.23,
+      "eval_em_first_line_ratio": 0.4742,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3623,
+      "eval_loss": 0.7937250733375549,
+      "eval_runtime": 258.3984,
+      "eval_samples_per_second": 38.7,
+      "eval_steps_per_second": 0.608,
+      "step": 196000
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 3.080822870057663e-05,
+      "loss": 0.8377,
+      "step": 196200
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 3.079885072966475e-05,
+      "loss": 0.8373,
+      "step": 196400
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 3.0789472758752874e-05,
+      "loss": 0.832,
+      "step": 196600
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 3.0780094787840995e-05,
+      "loss": 0.8322,
+      "step": 196800
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 3.0770716816929116e-05,
+      "loss": 0.834,
+      "step": 197000
+    },
+    {
+      "epoch": 0.23,
+      "eval_em_first_line_ratio": 0.4751,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3631,
+      "eval_loss": 0.7923118472099304,
+      "eval_runtime": 240.1219,
+      "eval_samples_per_second": 41.646,
+      "eval_steps_per_second": 0.654,
+      "step": 197000
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 3.0761338846017237e-05,
+      "loss": 0.836,
+      "step": 197200
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 3.075196087510536e-05,
+      "loss": 0.8329,
+      "step": 197400
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 3.074262979404804e-05,
+      "loss": 0.8278,
+      "step": 197600
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 3.073325182313616e-05,
+      "loss": 0.8313,
+      "step": 197800
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 3.072387385222428e-05,
+      "loss": 0.8332,
+      "step": 198000
+    },
+    {
+      "epoch": 0.23,
+      "eval_em_first_line_ratio": 0.4781,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3658,
+      "eval_loss": 0.7930506467819214,
+      "eval_runtime": 254.3515,
+      "eval_samples_per_second": 39.316,
+      "eval_steps_per_second": 0.617,
+      "step": 198000
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 3.07144958813124e-05,
+      "loss": 0.8334,
+      "step": 198200
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 3.0705117910400524e-05,
+      "loss": 0.8356,
+      "step": 198400
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 3.0695739939488645e-05,
+      "loss": 0.8303,
+      "step": 198600
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 3.0686361968576766e-05,
+      "loss": 0.8265,
+      "step": 198800
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 3.067698399766489e-05,
+      "loss": 0.829,
+      "step": 199000
+    },
+    {
+      "epoch": 0.23,
+      "eval_em_first_line_ratio": 0.4772,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3651,
+      "eval_loss": 0.7928906083106995,
+      "eval_runtime": 252.861,
+      "eval_samples_per_second": 39.547,
+      "eval_steps_per_second": 0.621,
+      "step": 199000
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 3.066760602675301e-05,
+      "loss": 0.8299,
+      "step": 199200
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 3.065822805584113e-05,
+      "loss": 0.8373,
+      "step": 199400
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 3.064885008492925e-05,
+      "loss": 0.8313,
+      "step": 199600
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 3.063947211401738e-05,
+      "loss": 0.8303,
+      "step": 199800
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 3.063014103296005e-05,
+      "loss": 0.8307,
+      "step": 200000
+    },
+    {
+      "epoch": 0.23,
+      "eval_em_first_line_ratio": 0.4771,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3647,
+      "eval_loss": 0.7910295724868774,
+      "eval_runtime": 216.3729,
+      "eval_samples_per_second": 46.216,
+      "eval_steps_per_second": 0.726,
+      "step": 200000
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 3.0620763062048174e-05,
+      "loss": 0.8368,
+      "step": 200200
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 3.0611385091136295e-05,
+      "loss": 0.8399,
+      "step": 200400
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 3.0602007120224416e-05,
+      "loss": 0.8329,
+      "step": 200600
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 3.059262914931254e-05,
+      "loss": 0.8284,
+      "step": 200800
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 3.058325117840066e-05,
+      "loss": 0.8275,
+      "step": 201000
+    },
+    {
+      "epoch": 0.24,
+      "eval_em_first_line_ratio": 0.4772,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3647,
+      "eval_loss": 0.7919970750808716,
+      "eval_runtime": 220.6837,
+      "eval_samples_per_second": 45.314,
+      "eval_steps_per_second": 0.711,
+      "step": 201000
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 3.057387320748878e-05,
+      "loss": 0.8267,
+      "step": 201200
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 3.05644952365769e-05,
+      "loss": 0.8296,
+      "step": 201400
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 3.055511726566502e-05,
+      "loss": 0.8335,
+      "step": 201600
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 3.054573929475314e-05,
+      "loss": 0.8322,
+      "step": 201800
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 3.053636132384126e-05,
+      "loss": 0.834,
+      "step": 202000
+    },
+    {
+      "epoch": 0.24,
+      "eval_em_first_line_ratio": 0.4735,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3618,
+      "eval_loss": 0.7916518449783325,
+      "eval_runtime": 238.5215,
+      "eval_samples_per_second": 41.925,
+      "eval_steps_per_second": 0.658,
+      "step": 202000
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 3.0527030242783945e-05,
+      "loss": 0.8314,
+      "step": 202200
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 3.051769916172663e-05,
+      "loss": 0.8334,
+      "step": 202400
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 3.050832119081475e-05,
+      "loss": 0.8265,
+      "step": 202600
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 3.0498943219902873e-05,
+      "loss": 0.8266,
+      "step": 202800
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 3.048956524899099e-05,
+      "loss": 0.8341,
+      "step": 203000
+    },
+    {
+      "epoch": 0.24,
+      "eval_em_first_line_ratio": 0.4779,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3661,
+      "eval_loss": 0.7912693023681641,
+      "eval_runtime": 250.4124,
+      "eval_samples_per_second": 39.934,
+      "eval_steps_per_second": 0.627,
+      "step": 203000
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 3.0480187278079115e-05,
+      "loss": 0.8315,
+      "step": 203200
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 3.0470809307167233e-05,
+      "loss": 0.8325,
+      "step": 203400
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 3.0461431336255357e-05,
+      "loss": 0.8321,
+      "step": 203600
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 3.0452053365343475e-05,
+      "loss": 0.829,
+      "step": 203800
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 3.04426753944316e-05,
+      "loss": 0.8276,
+      "step": 204000
+    },
+    {
+      "epoch": 0.24,
+      "eval_em_first_line_ratio": 0.4789,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.366,
+      "eval_loss": 0.7903376221656799,
+      "eval_runtime": 242.3395,
+      "eval_samples_per_second": 41.264,
+      "eval_steps_per_second": 0.648,
+      "step": 204000
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 3.043329742351972e-05,
+      "loss": 0.8301,
+      "step": 204200
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 3.042391945260784e-05,
+      "loss": 0.8326,
+      "step": 204400
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 3.0414541481695962e-05,
+      "loss": 0.8275,
+      "step": 204600
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 3.0405163510784083e-05,
+      "loss": 0.8324,
+      "step": 204800
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 3.0395785539872204e-05,
+      "loss": 0.829,
+      "step": 205000
+    },
+    {
+      "epoch": 0.24,
+      "eval_em_first_line_ratio": 0.4776,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3651,
+      "eval_loss": 0.7899261713027954,
+      "eval_runtime": 250.1324,
+      "eval_samples_per_second": 39.979,
+      "eval_steps_per_second": 0.628,
+      "step": 205000
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 3.0386407568960325e-05,
+      "loss": 0.8337,
+      "step": 205200
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 3.0377029598048446e-05,
+      "loss": 0.827,
+      "step": 205400
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 3.036765162713657e-05,
+      "loss": 0.8304,
+      "step": 205600
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 3.0358273656224688e-05,
+      "loss": 0.8308,
+      "step": 205800
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 3.0348895685312812e-05,
+      "loss": 0.838,
+      "step": 206000
+    },
+    {
+      "epoch": 0.24,
+      "eval_em_first_line_ratio": 0.4792,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3665,
+      "eval_loss": 0.7902094721794128,
+      "eval_runtime": 241.6507,
+      "eval_samples_per_second": 41.382,
+      "eval_steps_per_second": 0.65,
+      "step": 206000
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 3.033951771440093e-05,
+      "loss": 0.8263,
+      "step": 206200
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 3.0330139743489054e-05,
+      "loss": 0.8295,
+      "step": 206400
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 3.0320761772577175e-05,
+      "loss": 0.8336,
+      "step": 206600
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 3.0311383801665296e-05,
+      "loss": 0.8275,
+      "step": 206800
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 3.0302005830753417e-05,
+      "loss": 0.8291,
+      "step": 207000
+    },
+    {
+      "epoch": 0.24,
+      "eval_em_first_line_ratio": 0.4759,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3638,
+      "eval_loss": 0.789699375629425,
+      "eval_runtime": 249.8526,
+      "eval_samples_per_second": 40.024,
+      "eval_steps_per_second": 0.628,
+      "step": 207000
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 3.0292627859841538e-05,
+      "loss": 0.8263,
+      "step": 207200
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 3.028329677878422e-05,
+      "loss": 0.83,
+      "step": 207400
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 3.027391880787234e-05,
+      "loss": 0.8267,
+      "step": 207600
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 3.0264540836960463e-05,
+      "loss": 0.8267,
+      "step": 207800
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 3.025520975590314e-05,
+      "loss": 0.8354,
+      "step": 208000
+    },
+    {
+      "epoch": 0.24,
+      "eval_em_first_line_ratio": 0.477,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3644,
+      "eval_loss": 0.789588212966919,
+      "eval_runtime": 257.9823,
+      "eval_samples_per_second": 38.762,
+      "eval_steps_per_second": 0.609,
+      "step": 208000
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 3.0245831784991263e-05,
+      "loss": 0.8325,
+      "step": 208200
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 3.0236453814079384e-05,
+      "loss": 0.8322,
+      "step": 208400
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 3.0227122733022066e-05,
+      "loss": 0.8331,
+      "step": 208600
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 3.0217744762110187e-05,
+      "loss": 0.8257,
+      "step": 208800
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 3.0208366791198308e-05,
+      "loss": 0.8297,
+      "step": 209000
+    },
+    {
+      "epoch": 0.24,
+      "eval_em_first_line_ratio": 0.478,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3655,
+      "eval_loss": 0.789603054523468,
+      "eval_runtime": 242.0324,
+      "eval_samples_per_second": 41.317,
+      "eval_steps_per_second": 0.649,
+      "step": 209000
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 3.019898882028643e-05,
+      "loss": 0.8294,
+      "step": 209200
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 3.018961084937455e-05,
+      "loss": 0.8327,
+      "step": 209400
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 3.018023287846267e-05,
+      "loss": 0.825,
+      "step": 209600
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 3.0170854907550792e-05,
+      "loss": 0.8244,
+      "step": 209800
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 3.0161476936638916e-05,
+      "loss": 0.8248,
+      "step": 210000
+    },
+    {
+      "epoch": 0.25,
+      "eval_em_first_line_ratio": 0.4793,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3672,
+      "eval_loss": 0.7891152501106262,
+      "eval_runtime": 243.1807,
+      "eval_samples_per_second": 41.122,
+      "eval_steps_per_second": 0.646,
+      "step": 210000
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 3.0152098965727034e-05,
+      "loss": 0.8257,
+      "step": 210200
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 3.0142720994815158e-05,
+      "loss": 0.8213,
+      "step": 210400
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 3.0133343023903276e-05,
+      "loss": 0.8344,
+      "step": 210600
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 3.012401194284596e-05,
+      "loss": 0.8258,
+      "step": 210800
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 3.0114633971934083e-05,
+      "loss": 0.8254,
+      "step": 211000
+    },
+    {
+      "epoch": 0.25,
+      "eval_em_first_line_ratio": 0.4787,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3663,
+      "eval_loss": 0.7879025340080261,
+      "eval_runtime": 245.2804,
+      "eval_samples_per_second": 40.77,
+      "eval_steps_per_second": 0.64,
+      "step": 211000
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 3.01052560010222e-05,
+      "loss": 0.8255,
+      "step": 211200
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 3.0095878030110325e-05,
+      "loss": 0.8231,
+      "step": 211400
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 3.0086500059198442e-05,
+      "loss": 0.8309,
+      "step": 211600
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 3.0077122088286567e-05,
+      "loss": 0.8275,
+      "step": 211800
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 3.0067744117374684e-05,
+      "loss": 0.8245,
+      "step": 212000
+    },
+    {
+      "epoch": 0.25,
+      "eval_em_first_line_ratio": 0.4798,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3671,
+      "eval_loss": 0.7887754440307617,
+      "eval_runtime": 222.3774,
+      "eval_samples_per_second": 44.969,
+      "eval_steps_per_second": 0.706,
+      "step": 212000
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 3.005836614646281e-05,
+      "loss": 0.8311,
+      "step": 212200
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 3.004898817555093e-05,
+      "loss": 0.828,
+      "step": 212400
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 3.003961020463905e-05,
+      "loss": 0.8266,
+      "step": 212600
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 3.003023223372717e-05,
+      "loss": 0.8232,
+      "step": 212800
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 3.0020854262815292e-05,
+      "loss": 0.8274,
+      "step": 213000
+    },
+    {
+      "epoch": 0.25,
+      "eval_em_first_line_ratio": 0.4796,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3672,
+      "eval_loss": 0.7872698307037354,
+      "eval_runtime": 214.0003,
+      "eval_samples_per_second": 46.729,
+      "eval_steps_per_second": 0.734,
+      "step": 213000
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 3.0011476291903413e-05,
+      "loss": 0.8278,
+      "step": 213200
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 3.0002098320991538e-05,
+      "loss": 0.8314,
+      "step": 213400
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 2.9992720350079655e-05,
+      "loss": 0.8282,
+      "step": 213600
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 2.998334237916778e-05,
+      "loss": 0.8271,
+      "step": 213800
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 2.9973964408255897e-05,
+      "loss": 0.8305,
+      "step": 214000
+    },
+    {
+      "epoch": 0.25,
+      "eval_em_first_line_ratio": 0.4763,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3637,
+      "eval_loss": 0.7879440188407898,
+      "eval_runtime": 245.1828,
+      "eval_samples_per_second": 40.786,
+      "eval_steps_per_second": 0.64,
+      "step": 214000
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 2.996458643734402e-05,
+      "loss": 0.8271,
+      "step": 214200
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 2.995520846643214e-05,
+      "loss": 0.8284,
+      "step": 214400
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 2.9945830495520264e-05,
+      "loss": 0.8327,
+      "step": 214600
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 2.9936452524608385e-05,
+      "loss": 0.8269,
+      "step": 214800
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 2.9927074553696506e-05,
+      "loss": 0.8299,
+      "step": 215000
+    },
+    {
+      "epoch": 0.25,
+      "eval_em_first_line_ratio": 0.4758,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3636,
+      "eval_loss": 0.7876172661781311,
+      "eval_runtime": 246.723,
+      "eval_samples_per_second": 40.531,
+      "eval_steps_per_second": 0.636,
+      "step": 215000
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 2.9917696582784626e-05,
+      "loss": 0.8324,
+      "step": 215200
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 2.9908318611872747e-05,
+      "loss": 0.8339,
+      "step": 215400
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 2.989894064096087e-05,
+      "loss": 0.8228,
+      "step": 215600
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 2.9889562670048993e-05,
+      "loss": 0.829,
+      "step": 215800
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 2.988018469913711e-05,
+      "loss": 0.8286,
+      "step": 216000
+    },
+    {
+      "epoch": 0.25,
+      "eval_em_first_line_ratio": 0.4773,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3655,
+      "eval_loss": 0.787327229976654,
+      "eval_runtime": 246.0017,
+      "eval_samples_per_second": 40.65,
+      "eval_steps_per_second": 0.638,
+      "step": 216000
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 2.9870806728225235e-05,
+      "loss": 0.8261,
+      "step": 216200
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 2.9861428757313352e-05,
+      "loss": 0.8248,
+      "step": 216400
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 2.9852050786401477e-05,
+      "loss": 0.8219,
+      "step": 216600
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 2.9842672815489594e-05,
+      "loss": 0.8211,
+      "step": 216800
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 2.983329484457772e-05,
+      "loss": 0.8248,
+      "step": 217000
+    },
+    {
+      "epoch": 0.25,
+      "eval_em_first_line_ratio": 0.4806,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.368,
+      "eval_loss": 0.7864044308662415,
+      "eval_runtime": 238.7306,
+      "eval_samples_per_second": 41.888,
+      "eval_steps_per_second": 0.658,
+      "step": 217000
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 2.9823916873665843e-05,
+      "loss": 0.8269,
+      "step": 217200
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 2.981453890275396e-05,
+      "loss": 0.8229,
+      "step": 217400
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 2.980516093184208e-05,
+      "loss": 0.8272,
+      "step": 217600
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 2.9795782960930203e-05,
+      "loss": 0.8288,
+      "step": 217800
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 2.9786404990018324e-05,
+      "loss": 0.8289,
+      "step": 218000
+    },
+    {
+      "epoch": 0.26,
+      "eval_em_first_line_ratio": 0.4774,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3653,
+      "eval_loss": 0.7865781784057617,
+      "eval_runtime": 239.1679,
+      "eval_samples_per_second": 41.812,
+      "eval_steps_per_second": 0.656,
+      "step": 218000
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 2.9777027019106448e-05,
+      "loss": 0.8291,
+      "step": 218200
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 2.9767649048194565e-05,
+      "loss": 0.8241,
+      "step": 218400
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 2.975827107728269e-05,
+      "loss": 0.8287,
+      "step": 218600
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 2.9748893106370807e-05,
+      "loss": 0.8253,
+      "step": 218800
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 2.9739515135458932e-05,
+      "loss": 0.8204,
+      "step": 219000
+    },
+    {
+      "epoch": 0.26,
+      "eval_em_first_line_ratio": 0.4776,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3651,
+      "eval_loss": 0.7870668172836304,
+      "eval_runtime": 239.7613,
+      "eval_samples_per_second": 41.708,
+      "eval_steps_per_second": 0.655,
+      "step": 219000
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 2.973018405440161e-05,
+      "loss": 0.8236,
+      "step": 219200
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 2.9720806083489732e-05,
+      "loss": 0.8231,
+      "step": 219400
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 2.9711428112577856e-05,
+      "loss": 0.8232,
+      "step": 219600
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 2.9702050141665974e-05,
+      "loss": 0.8321,
+      "step": 219800
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 2.9692672170754098e-05,
+      "loss": 0.8271,
+      "step": 220000
+    },
+    {
+      "epoch": 0.26,
+      "eval_em_first_line_ratio": 0.4764,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3636,
+      "eval_loss": 0.7860292792320251,
+      "eval_runtime": 250.1098,
+      "eval_samples_per_second": 39.982,
+      "eval_steps_per_second": 0.628,
+      "step": 220000
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 2.9683341089696774e-05,
+      "loss": 0.8231,
+      "step": 220200
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 2.9673963118784898e-05,
+      "loss": 0.8261,
+      "step": 220400
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 2.9664585147873023e-05,
+      "loss": 0.8266,
+      "step": 220600
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 2.965520717696114e-05,
+      "loss": 0.8269,
+      "step": 220800
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 2.9645829206049265e-05,
+      "loss": 0.8288,
+      "step": 221000
+    },
+    {
+      "epoch": 0.26,
+      "eval_em_first_line_ratio": 0.4788,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3665,
+      "eval_loss": 0.7859458327293396,
+      "eval_runtime": 243.4722,
+      "eval_samples_per_second": 41.072,
+      "eval_steps_per_second": 0.645,
+      "step": 221000
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 2.9636451235137382e-05,
+      "loss": 0.8289,
+      "step": 221200
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 2.9627073264225506e-05,
+      "loss": 0.822,
+      "step": 221400
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 2.9617695293313627e-05,
+      "loss": 0.8272,
+      "step": 221600
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 2.960831732240175e-05,
+      "loss": 0.8233,
+      "step": 221800
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 2.959893935148987e-05,
+      "loss": 0.8248,
+      "step": 222000
+    },
+    {
+      "epoch": 0.26,
+      "eval_em_first_line_ratio": 0.4782,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3655,
+      "eval_loss": 0.7855960726737976,
+      "eval_runtime": 260.2718,
+      "eval_samples_per_second": 38.421,
+      "eval_steps_per_second": 0.603,
+      "step": 222000
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 2.958956138057799e-05,
+      "loss": 0.8269,
+      "step": 222200
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 2.958018340966611e-05,
+      "loss": 0.8265,
+      "step": 222400
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 2.957085232860879e-05,
+      "loss": 0.8212,
+      "step": 222600
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 2.9561474357696915e-05,
+      "loss": 0.8208,
+      "step": 222800
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 2.9552096386785036e-05,
+      "loss": 0.8223,
+      "step": 223000
+    },
+    {
+      "epoch": 0.26,
+      "eval_em_first_line_ratio": 0.4776,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3651,
+      "eval_loss": 0.7853403091430664,
+      "eval_runtime": 211.1164,
+      "eval_samples_per_second": 47.367,
+      "eval_steps_per_second": 0.744,
+      "step": 223000
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 2.9542718415873153e-05,
+      "loss": 0.826,
+      "step": 223200
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 2.9533340444961278e-05,
+      "loss": 0.8197,
+      "step": 223400
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 2.9523962474049395e-05,
+      "loss": 0.8245,
+      "step": 223600
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 2.951458450313752e-05,
+      "loss": 0.8204,
+      "step": 223800
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 2.9505206532225644e-05,
+      "loss": 0.824,
+      "step": 224000
+    },
+    {
+      "epoch": 0.26,
+      "eval_em_first_line_ratio": 0.4779,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3656,
+      "eval_loss": 0.7842656373977661,
+      "eval_runtime": 218.5502,
+      "eval_samples_per_second": 45.756,
+      "eval_steps_per_second": 0.718,
+      "step": 224000
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 2.949582856131376e-05,
+      "loss": 0.8248,
+      "step": 224200
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 2.9486450590401886e-05,
+      "loss": 0.8281,
+      "step": 224400
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 2.9477072619490004e-05,
+      "loss": 0.8249,
+      "step": 224600
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 2.9467694648578128e-05,
+      "loss": 0.8265,
+      "step": 224800
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 2.9458316677666246e-05,
+      "loss": 0.823,
+      "step": 225000
+    },
+    {
+      "epoch": 0.26,
+      "eval_em_first_line_ratio": 0.4784,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3658,
+      "eval_loss": 0.7846024036407471,
+      "eval_runtime": 249.4932,
+      "eval_samples_per_second": 40.081,
+      "eval_steps_per_second": 0.629,
+      "step": 225000
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 2.944893870675437e-05,
+      "loss": 0.8264,
+      "step": 225200
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 2.943956073584249e-05,
+      "loss": 0.8233,
+      "step": 225400
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 2.9430182764930612e-05,
+      "loss": 0.8214,
+      "step": 225600
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 2.9420804794018733e-05,
+      "loss": 0.8214,
+      "step": 225800
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 2.9411426823106854e-05,
+      "loss": 0.8242,
+      "step": 226000
+    },
+    {
+      "epoch": 0.26,
+      "eval_em_first_line_ratio": 0.4781,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3662,
+      "eval_loss": 0.7837600111961365,
+      "eval_runtime": 248.6973,
+      "eval_samples_per_second": 40.21,
+      "eval_steps_per_second": 0.631,
+      "step": 226000
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 2.9402048852194975e-05,
+      "loss": 0.8267,
+      "step": 226200
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 2.93926708812831e-05,
+      "loss": 0.8181,
+      "step": 226400
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 2.9383292910371217e-05,
+      "loss": 0.823,
+      "step": 226600
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 2.937391493945934e-05,
+      "loss": 0.8214,
+      "step": 226800
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 2.936453696854746e-05,
+      "loss": 0.8201,
+      "step": 227000
+    },
+    {
+      "epoch": 0.27,
+      "eval_em_first_line_ratio": 0.4796,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3675,
+      "eval_loss": 0.7835144400596619,
+      "eval_runtime": 246.8671,
+      "eval_samples_per_second": 40.508,
+      "eval_steps_per_second": 0.636,
+      "step": 227000
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 2.9355158997635583e-05,
+      "loss": 0.8172,
+      "step": 227200
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 2.93457810267237e-05,
+      "loss": 0.8235,
+      "step": 227400
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 2.9336403055811825e-05,
+      "loss": 0.8211,
+      "step": 227600
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 2.9327025084899946e-05,
+      "loss": 0.8253,
+      "step": 227800
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 2.9317694003842625e-05,
+      "loss": 0.819,
+      "step": 228000
+    },
+    {
+      "epoch": 0.27,
+      "eval_em_first_line_ratio": 0.4794,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.367,
+      "eval_loss": 0.7834131121635437,
+      "eval_runtime": 255.7626,
+      "eval_samples_per_second": 39.099,
+      "eval_steps_per_second": 0.614,
+      "step": 228000
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 2.930831603293075e-05,
+      "loss": 0.8221,
+      "step": 228200
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 2.9298984951873425e-05,
+      "loss": 0.8266,
+      "step": 228400
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 2.928960698096155e-05,
+      "loss": 0.8223,
+      "step": 228600
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 2.9280229010049674e-05,
+      "loss": 0.818,
+      "step": 228800
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 2.927085103913779e-05,
+      "loss": 0.8227,
+      "step": 229000
+    },
+    {
+      "epoch": 0.27,
+      "eval_em_first_line_ratio": 0.4779,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3654,
+      "eval_loss": 0.783180296421051,
+      "eval_runtime": 234.8084,
+      "eval_samples_per_second": 42.588,
+      "eval_steps_per_second": 0.669,
+      "step": 229000
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 2.9261473068225912e-05,
+      "loss": 0.8219,
+      "step": 229200
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 2.9252095097314033e-05,
+      "loss": 0.8227,
+      "step": 229400
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 2.9242717126402154e-05,
+      "loss": 0.8225,
+      "step": 229600
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 2.923333915549028e-05,
+      "loss": 0.8232,
+      "step": 229800
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 2.9223961184578396e-05,
+      "loss": 0.8187,
+      "step": 230000
+    },
+    {
+      "epoch": 0.27,
+      "eval_em_first_line_ratio": 0.4789,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3662,
+      "eval_loss": 0.7833844423294067,
+      "eval_runtime": 244.1485,
+      "eval_samples_per_second": 40.959,
+      "eval_steps_per_second": 0.643,
+      "step": 230000
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 2.921458321366652e-05,
+      "loss": 0.8214,
+      "step": 230200
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 2.92052521326092e-05,
+      "loss": 0.8205,
+      "step": 230400
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 2.919587416169732e-05,
+      "loss": 0.8255,
+      "step": 230600
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 2.9186496190785442e-05,
+      "loss": 0.8186,
+      "step": 230800
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 2.9177118219873563e-05,
+      "loss": 0.8203,
+      "step": 231000
+    },
+    {
+      "epoch": 0.27,
+      "eval_em_first_line_ratio": 0.4793,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3668,
+      "eval_loss": 0.7824200987815857,
+      "eval_runtime": 244.3897,
+      "eval_samples_per_second": 40.918,
+      "eval_steps_per_second": 0.642,
+      "step": 231000
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 2.9167740248961687e-05,
+      "loss": 0.8263,
+      "step": 231200
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 2.9158362278049805e-05,
+      "loss": 0.8175,
+      "step": 231400
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 2.914898430713793e-05,
+      "loss": 0.8204,
+      "step": 231600
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 2.9139606336226047e-05,
+      "loss": 0.8179,
+      "step": 231800
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 2.913022836531417e-05,
+      "loss": 0.8215,
+      "step": 232000
+    },
+    {
+      "epoch": 0.27,
+      "eval_em_first_line_ratio": 0.478,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3654,
+      "eval_loss": 0.7825027704238892,
+      "eval_runtime": 248.2689,
+      "eval_samples_per_second": 40.279,
+      "eval_steps_per_second": 0.632,
+      "step": 232000
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 2.9120850394402292e-05,
+      "loss": 0.8209,
+      "step": 232200
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 2.9111472423490413e-05,
+      "loss": 0.8216,
+      "step": 232400
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 2.9102141342433095e-05,
+      "loss": 0.8184,
+      "step": 232600
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 2.9092763371521213e-05,
+      "loss": 0.8184,
+      "step": 232800
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 2.9083385400609337e-05,
+      "loss": 0.817,
+      "step": 233000
+    },
+    {
+      "epoch": 0.27,
+      "eval_em_first_line_ratio": 0.4801,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.368,
+      "eval_loss": 0.782241702079773,
+      "eval_runtime": 248.4697,
+      "eval_samples_per_second": 40.246,
+      "eval_steps_per_second": 0.632,
+      "step": 233000
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 2.907400742969746e-05,
+      "loss": 0.8237,
+      "step": 233200
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 2.906462945878558e-05,
+      "loss": 0.8188,
+      "step": 233400
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 2.90552514878737e-05,
+      "loss": 0.8197,
+      "step": 233600
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 2.904587351696182e-05,
+      "loss": 0.8204,
+      "step": 233800
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 2.9036495546049942e-05,
+      "loss": 0.8229,
+      "step": 234000
+    },
+    {
+      "epoch": 0.27,
+      "eval_em_first_line_ratio": 0.4803,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3682,
+      "eval_loss": 0.7817152142524719,
+      "eval_runtime": 224.5574,
+      "eval_samples_per_second": 44.532,
+      "eval_steps_per_second": 0.699,
+      "step": 234000
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 2.9027117575138063e-05,
+      "loss": 0.8228,
+      "step": 234200
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 2.9017739604226184e-05,
+      "loss": 0.818,
+      "step": 234400
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 2.900836163331431e-05,
+      "loss": 0.8197,
+      "step": 234600
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 2.8998983662402426e-05,
+      "loss": 0.8225,
+      "step": 234800
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 2.898960569149055e-05,
+      "loss": 0.8183,
+      "step": 235000
+    },
+    {
+      "epoch": 0.28,
+      "eval_em_first_line_ratio": 0.4806,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3685,
+      "eval_loss": 0.7824001908302307,
+      "eval_runtime": 221.3408,
+      "eval_samples_per_second": 45.179,
+      "eval_steps_per_second": 0.709,
+      "step": 235000
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 2.8980227720578668e-05,
+      "loss": 0.8202,
+      "step": 235200
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 2.8970849749666792e-05,
+      "loss": 0.8199,
+      "step": 235400
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 2.896147177875491e-05,
+      "loss": 0.823,
+      "step": 235600
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 2.8952093807843034e-05,
+      "loss": 0.8199,
+      "step": 235800
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 2.8942715836931155e-05,
+      "loss": 0.8199,
+      "step": 236000
+    },
+    {
+      "epoch": 0.28,
+      "eval_em_first_line_ratio": 0.4805,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3684,
+      "eval_loss": 0.7822141647338867,
+      "eval_runtime": 233.045,
+      "eval_samples_per_second": 42.91,
+      "eval_steps_per_second": 0.674,
+      "step": 236000
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 2.8933337866019276e-05,
+      "loss": 0.8216,
+      "step": 236200
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 2.8923959895107397e-05,
+      "loss": 0.8233,
+      "step": 236400
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 2.8914581924195518e-05,
+      "loss": 0.8159,
+      "step": 236600
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 2.890520395328364e-05,
+      "loss": 0.8158,
+      "step": 236800
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 2.8895872872226322e-05,
+      "loss": 0.818,
+      "step": 237000
+    },
+    {
+      "epoch": 0.28,
+      "eval_em_first_line_ratio": 0.4796,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3677,
+      "eval_loss": 0.7818002104759216,
+      "eval_runtime": 249.4744,
+      "eval_samples_per_second": 40.084,
+      "eval_steps_per_second": 0.629,
+      "step": 237000
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 2.8886494901314443e-05,
+      "loss": 0.8225,
+      "step": 237200
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 2.8877116930402564e-05,
+      "loss": 0.8221,
+      "step": 237400
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 2.8867738959490685e-05,
+      "loss": 0.8179,
+      "step": 237600
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 2.8858360988578806e-05,
+      "loss": 0.8226,
+      "step": 237800
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 2.884898301766693e-05,
+      "loss": 0.8207,
+      "step": 238000
+    },
+    {
+      "epoch": 0.28,
+      "eval_em_first_line_ratio": 0.4811,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3683,
+      "eval_loss": 0.781409740447998,
+      "eval_runtime": 241.0453,
+      "eval_samples_per_second": 41.486,
+      "eval_steps_per_second": 0.651,
+      "step": 238000
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 2.8839605046755048e-05,
+      "loss": 0.8179,
+      "step": 238200
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 2.883027396569773e-05,
+      "loss": 0.8203,
+      "step": 238400
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 2.8820895994785848e-05,
+      "loss": 0.8189,
+      "step": 238600
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 2.8811518023873972e-05,
+      "loss": 0.8186,
+      "step": 238800
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 2.880214005296209e-05,
+      "loss": 0.8209,
+      "step": 239000
+    },
+    {
+      "epoch": 0.28,
+      "eval_em_first_line_ratio": 0.4793,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.367,
+      "eval_loss": 0.7804228067398071,
+      "eval_runtime": 241.5667,
+      "eval_samples_per_second": 41.396,
+      "eval_steps_per_second": 0.65,
+      "step": 239000
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 2.8792762082050214e-05,
+      "loss": 0.8162,
+      "step": 239200
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 2.878338411113834e-05,
+      "loss": 0.8221,
+      "step": 239400
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 2.8774006140226456e-05,
+      "loss": 0.8177,
+      "step": 239600
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 2.876462816931458e-05,
+      "loss": 0.8188,
+      "step": 239800
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 2.8755250198402698e-05,
+      "loss": 0.8156,
+      "step": 240000
+    },
+    {
+      "epoch": 0.28,
+      "eval_em_first_line_ratio": 0.4822,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3698,
+      "eval_loss": 0.7811782956123352,
+      "eval_runtime": 241.4883,
+      "eval_samples_per_second": 41.41,
+      "eval_steps_per_second": 0.65,
+      "step": 240000
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 2.8745872227490822e-05,
+      "loss": 0.8182,
+      "step": 240200
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 2.8736494256578943e-05,
+      "loss": 0.8186,
+      "step": 240400
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 2.8727116285667064e-05,
+      "loss": 0.8173,
+      "step": 240600
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 2.8717785204609743e-05,
+      "loss": 0.8192,
+      "step": 240800
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 2.8708407233697864e-05,
+      "loss": 0.8234,
+      "step": 241000
+    },
+    {
+      "epoch": 0.28,
+      "eval_em_first_line_ratio": 0.4807,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3683,
+      "eval_loss": 0.7805859446525574,
+      "eval_runtime": 242.2763,
+      "eval_samples_per_second": 41.275,
+      "eval_steps_per_second": 0.648,
+      "step": 241000
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 2.8699029262785985e-05,
+      "loss": 0.8156,
+      "step": 241200
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 2.868965129187411e-05,
+      "loss": 0.8154,
+      "step": 241400
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 2.8680273320962227e-05,
+      "loss": 0.8193,
+      "step": 241600
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 2.867089535005035e-05,
+      "loss": 0.8126,
+      "step": 241800
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 2.866151737913847e-05,
+      "loss": 0.8126,
+      "step": 242000
+    },
+    {
+      "epoch": 0.28,
+      "eval_em_first_line_ratio": 0.4817,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.369,
+      "eval_loss": 0.7809672951698303,
+      "eval_runtime": 257.5011,
+      "eval_samples_per_second": 38.835,
+      "eval_steps_per_second": 0.61,
+      "step": 242000
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 2.8652139408226593e-05,
+      "loss": 0.8145,
+      "step": 242200
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 2.864276143731471e-05,
+      "loss": 0.8186,
+      "step": 242400
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 2.8633383466402835e-05,
+      "loss": 0.8174,
+      "step": 242600
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 2.862400549549096e-05,
+      "loss": 0.8196,
+      "step": 242800
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 2.8614627524579077e-05,
+      "loss": 0.8212,
+      "step": 243000
+    },
+    {
+      "epoch": 0.28,
+      "eval_em_first_line_ratio": 0.4805,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3681,
+      "eval_loss": 0.7802003026008606,
+      "eval_runtime": 252.9605,
+      "eval_samples_per_second": 39.532,
+      "eval_steps_per_second": 0.621,
+      "step": 243000
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 2.8605249553667202e-05,
+      "loss": 0.8204,
+      "step": 243200
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 2.859587158275532e-05,
+      "loss": 0.8187,
+      "step": 243400
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 2.8586493611843444e-05,
+      "loss": 0.8162,
+      "step": 243600
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 2.857711564093156e-05,
+      "loss": 0.8245,
+      "step": 243800
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 2.8567737670019686e-05,
+      "loss": 0.8217,
+      "step": 244000
+    },
+    {
+      "epoch": 0.29,
+      "eval_em_first_line_ratio": 0.4831,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3705,
+      "eval_loss": 0.7796353101730347,
+      "eval_runtime": 239.3317,
+      "eval_samples_per_second": 41.783,
+      "eval_steps_per_second": 0.656,
+      "step": 244000
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 2.8558359699107807e-05,
+      "loss": 0.8157,
+      "step": 244200
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 2.8549028618050486e-05,
+      "loss": 0.8111,
+      "step": 244400
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 2.8539650647138607e-05,
+      "loss": 0.8177,
+      "step": 244600
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 2.8530272676226728e-05,
+      "loss": 0.8202,
+      "step": 244800
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 2.852089470531485e-05,
+      "loss": 0.8175,
+      "step": 245000
+    },
+    {
+      "epoch": 0.29,
+      "eval_em_first_line_ratio": 0.4789,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3664,
+      "eval_loss": 0.779350221157074,
+      "eval_runtime": 215.3263,
+      "eval_samples_per_second": 46.441,
+      "eval_steps_per_second": 0.729,
+      "step": 245000
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 2.8511516734402973e-05,
+      "loss": 0.8119,
+      "step": 245200
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 2.850213876349109e-05,
+      "loss": 0.8183,
+      "step": 245400
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 2.8492760792579215e-05,
+      "loss": 0.8185,
+      "step": 245600
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 2.8483382821667333e-05,
+      "loss": 0.8204,
+      "step": 245800
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 2.8474004850755457e-05,
+      "loss": 0.8136,
+      "step": 246000
+    },
+    {
+      "epoch": 0.29,
+      "eval_em_first_line_ratio": 0.4822,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3692,
+      "eval_loss": 0.7799516320228577,
+      "eval_runtime": 207.4287,
+      "eval_samples_per_second": 48.209,
+      "eval_steps_per_second": 0.757,
+      "step": 246000
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 2.846462687984358e-05,
+      "loss": 0.8121,
+      "step": 246200
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 2.84552489089317e-05,
+      "loss": 0.8124,
+      "step": 246400
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 2.8445870938019823e-05,
+      "loss": 0.8169,
+      "step": 246600
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 2.843649296710794e-05,
+      "loss": 0.8157,
+      "step": 246800
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 2.8427114996196065e-05,
+      "loss": 0.8106,
+      "step": 247000
+    },
+    {
+      "epoch": 0.29,
+      "eval_em_first_line_ratio": 0.4813,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3688,
+      "eval_loss": 0.7793726921081543,
+      "eval_runtime": 225.6473,
+      "eval_samples_per_second": 44.317,
+      "eval_steps_per_second": 0.696,
+      "step": 247000
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 2.8417737025284183e-05,
+      "loss": 0.8195,
+      "step": 247200
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 2.8408359054372304e-05,
+      "loss": 0.816,
+      "step": 247400
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 2.8398981083460428e-05,
+      "loss": 0.8111,
+      "step": 247600
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 2.8389603112548546e-05,
+      "loss": 0.8203,
+      "step": 247800
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 2.8380272031491228e-05,
+      "loss": 0.8147,
+      "step": 248000
+    },
+    {
+      "epoch": 0.29,
+      "eval_em_first_line_ratio": 0.482,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3698,
+      "eval_loss": 0.7798734307289124,
+      "eval_runtime": 247.2285,
+      "eval_samples_per_second": 40.448,
+      "eval_steps_per_second": 0.635,
+      "step": 248000
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 2.837089406057935e-05,
+      "loss": 0.8159,
+      "step": 248200
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 2.836151608966747e-05,
+      "loss": 0.8129,
+      "step": 248400
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 2.8352138118755594e-05,
+      "loss": 0.8152,
+      "step": 248600
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 2.8342760147843712e-05,
+      "loss": 0.8155,
+      "step": 248800
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 2.8333382176931836e-05,
+      "loss": 0.8161,
+      "step": 249000
+    },
+    {
+      "epoch": 0.29,
+      "eval_em_first_line_ratio": 0.4811,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3682,
+      "eval_loss": 0.7789192199707031,
+      "eval_runtime": 239.1144,
+      "eval_samples_per_second": 41.821,
+      "eval_steps_per_second": 0.657,
+      "step": 249000
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 2.8324004206019954e-05,
+      "loss": 0.8157,
+      "step": 249200
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 2.831462623510808e-05,
+      "loss": 0.814,
+      "step": 249400
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 2.8305248264196196e-05,
+      "loss": 0.8156,
+      "step": 249600
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 2.829587029328432e-05,
+      "loss": 0.814,
+      "step": 249800
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 2.8286492322372445e-05,
+      "loss": 0.819,
+      "step": 250000
+    },
+    {
+      "epoch": 0.29,
+      "eval_em_first_line_ratio": 0.4815,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3682,
+      "eval_loss": 0.7784441709518433,
+      "eval_runtime": 239.3427,
+      "eval_samples_per_second": 41.781,
+      "eval_steps_per_second": 0.656,
+      "step": 250000
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 2.827716124131512e-05,
+      "loss": 0.8203,
+      "step": 250200
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 2.8267783270403245e-05,
+      "loss": 0.8187,
+      "step": 250400
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 2.8258405299491362e-05,
+      "loss": 0.8148,
+      "step": 250600
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 2.8249027328579487e-05,
+      "loss": 0.8128,
+      "step": 250800
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 2.8239649357667608e-05,
+      "loss": 0.8215,
+      "step": 251000
+    },
+    {
+      "epoch": 0.29,
+      "eval_em_first_line_ratio": 0.4804,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3675,
+      "eval_loss": 0.7778878211975098,
+      "eval_runtime": 226.2682,
+      "eval_samples_per_second": 44.195,
+      "eval_steps_per_second": 0.694,
+      "step": 251000
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 2.823027138675573e-05,
+      "loss": 0.8151,
+      "step": 251200
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 2.822089341584385e-05,
+      "loss": 0.8068,
+      "step": 251400
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 2.821151544493197e-05,
+      "loss": 0.8174,
+      "step": 251600
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 2.820213747402009e-05,
+      "loss": 0.8141,
+      "step": 251800
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 2.8192759503108216e-05,
+      "loss": 0.8143,
+      "step": 252000
+    },
+    {
+      "epoch": 0.3,
+      "eval_em_first_line_ratio": 0.4798,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3673,
+      "eval_loss": 0.7782419323921204,
+      "eval_runtime": 240.9351,
+      "eval_samples_per_second": 41.505,
+      "eval_steps_per_second": 0.652,
+      "step": 252000
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 2.8183381532196333e-05,
+      "loss": 0.8183,
+      "step": 252200
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 2.8174003561284458e-05,
+      "loss": 0.8145,
+      "step": 252400
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 2.8164625590372575e-05,
+      "loss": 0.8138,
+      "step": 252600
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 2.81552476194607e-05,
+      "loss": 0.8145,
+      "step": 252800
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 2.8145869648548817e-05,
+      "loss": 0.8195,
+      "step": 253000
+    },
+    {
+      "epoch": 0.3,
+      "eval_em_first_line_ratio": 0.4823,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3696,
+      "eval_loss": 0.7780371308326721,
+      "eval_runtime": 210.2793,
+      "eval_samples_per_second": 47.556,
+      "eval_steps_per_second": 0.747,
+      "step": 253000
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 2.8136491677636942e-05,
+      "loss": 0.809,
+      "step": 253200
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 2.8127113706725063e-05,
+      "loss": 0.814,
+      "step": 253400
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 2.8117735735813184e-05,
+      "loss": 0.8171,
+      "step": 253600
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 2.8108357764901305e-05,
+      "loss": 0.8068,
+      "step": 253800
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 2.8098979793989426e-05,
+      "loss": 0.8144,
+      "step": 254000
+    },
+    {
+      "epoch": 0.3,
+      "eval_em_first_line_ratio": 0.4824,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3702,
+      "eval_loss": 0.7770723700523376,
+      "eval_runtime": 214.574,
+      "eval_samples_per_second": 46.604,
+      "eval_steps_per_second": 0.732,
+      "step": 254000
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 2.8089601823077547e-05,
+      "loss": 0.8136,
+      "step": 254200
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 2.8080223852165668e-05,
+      "loss": 0.8144,
+      "step": 254400
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 2.807084588125379e-05,
+      "loss": 0.8133,
+      "step": 254600
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 2.806151480019647e-05,
+      "loss": 0.8122,
+      "step": 254800
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 2.8052136829284592e-05,
+      "loss": 0.8166,
+      "step": 255000
+    },
+    {
+      "epoch": 0.3,
+      "eval_em_first_line_ratio": 0.4806,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3683,
+      "eval_loss": 0.7777746319770813,
+      "eval_runtime": 209.5227,
+      "eval_samples_per_second": 47.728,
+      "eval_steps_per_second": 0.749,
+      "step": 255000
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 2.8042758858372713e-05,
+      "loss": 0.8149,
+      "step": 255200
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 2.8033380887460834e-05,
+      "loss": 0.8091,
+      "step": 255400
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 2.8024002916548955e-05,
+      "loss": 0.8104,
+      "step": 255600
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 2.801462494563708e-05,
+      "loss": 0.8176,
+      "step": 255800
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 2.8005246974725197e-05,
+      "loss": 0.8165,
+      "step": 256000
+    },
+    {
+      "epoch": 0.3,
+      "eval_em_first_line_ratio": 0.4812,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3691,
+      "eval_loss": 0.7770427465438843,
+      "eval_runtime": 205.4278,
+      "eval_samples_per_second": 48.679,
+      "eval_steps_per_second": 0.764,
+      "step": 256000
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 2.799591589366788e-05,
+      "loss": 0.8176,
+      "step": 256200
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 2.7986537922755997e-05,
+      "loss": 0.808,
+      "step": 256400
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 2.797720684169868e-05,
+      "loss": 0.814,
+      "step": 256600
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 2.7967828870786804e-05,
+      "loss": 0.8134,
+      "step": 256800
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 2.795845089987492e-05,
+      "loss": 0.8126,
+      "step": 257000
+    },
+    {
+      "epoch": 0.3,
+      "eval_em_first_line_ratio": 0.481,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3684,
+      "eval_loss": 0.776982843875885,
+      "eval_runtime": 209.2025,
+      "eval_samples_per_second": 47.801,
+      "eval_steps_per_second": 0.75,
+      "step": 257000
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 2.7949072928963046e-05,
+      "loss": 0.8146,
+      "step": 257200
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 2.7939694958051163e-05,
+      "loss": 0.8145,
+      "step": 257400
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 2.7930316987139288e-05,
+      "loss": 0.8135,
+      "step": 257600
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 2.7920939016227412e-05,
+      "loss": 0.8151,
+      "step": 257800
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 2.791156104531553e-05,
+      "loss": 0.8162,
+      "step": 258000
+    },
+    {
+      "epoch": 0.3,
+      "eval_em_first_line_ratio": 0.4818,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3693,
+      "eval_loss": 0.7771302461624146,
+      "eval_runtime": 217.5715,
+      "eval_samples_per_second": 45.962,
+      "eval_steps_per_second": 0.722,
+      "step": 258000
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 2.7902183074403654e-05,
+      "loss": 0.8167,
+      "step": 258200
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 2.789280510349177e-05,
+      "loss": 0.816,
+      "step": 258400
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 2.7883427132579896e-05,
+      "loss": 0.8133,
+      "step": 258600
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 2.7874049161668014e-05,
+      "loss": 0.8186,
+      "step": 258800
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 2.7864671190756135e-05,
+      "loss": 0.8152,
+      "step": 259000
+    },
+    {
+      "epoch": 0.3,
+      "eval_em_first_line_ratio": 0.4819,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3687,
+      "eval_loss": 0.7770100831985474,
+      "eval_runtime": 209.7949,
+      "eval_samples_per_second": 47.666,
+      "eval_steps_per_second": 0.748,
+      "step": 259000
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 2.785529321984426e-05,
+      "loss": 0.8134,
+      "step": 259200
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 2.7845915248932377e-05,
+      "loss": 0.8088,
+      "step": 259400
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 2.78365372780205e-05,
+      "loss": 0.8076,
+      "step": 259600
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 2.782715930710862e-05,
+      "loss": 0.8127,
+      "step": 259800
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 2.7817781336196743e-05,
+      "loss": 0.8053,
+      "step": 260000
+    },
+    {
+      "epoch": 0.3,
+      "eval_em_first_line_ratio": 0.4808,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3683,
+      "eval_loss": 0.776253879070282,
+      "eval_runtime": 213.3053,
+      "eval_samples_per_second": 46.881,
+      "eval_steps_per_second": 0.736,
+      "step": 260000
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 2.7808403365284867e-05,
+      "loss": 0.8133,
+      "step": 260200
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 2.7799025394372985e-05,
+      "loss": 0.8099,
+      "step": 260400
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 2.778964742346111e-05,
+      "loss": 0.8104,
+      "step": 260600
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 2.7780269452549227e-05,
+      "loss": 0.8147,
+      "step": 260800
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 2.777093837149191e-05,
+      "loss": 0.8068,
+      "step": 261000
+    },
+    {
+      "epoch": 0.31,
+      "eval_em_first_line_ratio": 0.4797,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3667,
+      "eval_loss": 0.7764716744422913,
+      "eval_runtime": 206.6309,
+      "eval_samples_per_second": 48.395,
+      "eval_steps_per_second": 0.76,
+      "step": 261000
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 2.7761560400580027e-05,
+      "loss": 0.8145,
+      "step": 261200
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 2.775218242966815e-05,
+      "loss": 0.8087,
+      "step": 261400
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 2.7742804458756276e-05,
+      "loss": 0.8139,
+      "step": 261600
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 2.7733426487844393e-05,
+      "loss": 0.8159,
+      "step": 261800
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 2.7724048516932514e-05,
+      "loss": 0.8075,
+      "step": 262000
+    },
+    {
+      "epoch": 0.31,
+      "eval_em_first_line_ratio": 0.481,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3689,
+      "eval_loss": 0.776313066482544,
+      "eval_runtime": 211.3397,
+      "eval_samples_per_second": 47.317,
+      "eval_steps_per_second": 0.743,
+      "step": 262000
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 2.7714670546020635e-05,
+      "loss": 0.8129,
+      "step": 262200
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 2.7705292575108756e-05,
+      "loss": 0.8091,
+      "step": 262400
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 2.769591460419688e-05,
+      "loss": 0.8109,
+      "step": 262600
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 2.7686536633284998e-05,
+      "loss": 0.8148,
+      "step": 262800
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 2.7677158662373122e-05,
+      "loss": 0.8115,
+      "step": 263000
+    },
+    {
+      "epoch": 0.31,
+      "eval_em_first_line_ratio": 0.4816,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3691,
+      "eval_loss": 0.7762570381164551,
+      "eval_runtime": 202.9365,
+      "eval_samples_per_second": 49.277,
+      "eval_steps_per_second": 0.774,
+      "step": 263000
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 2.76678275813158e-05,
+      "loss": 0.8122,
+      "step": 263200
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 2.7658449610403922e-05,
+      "loss": 0.8127,
+      "step": 263400
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 2.7649071639492047e-05,
+      "loss": 0.8119,
+      "step": 263600
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 2.7639693668580164e-05,
+      "loss": 0.8066,
+      "step": 263800
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 2.763031569766829e-05,
+      "loss": 0.8119,
+      "step": 264000
+    },
+    {
+      "epoch": 0.31,
+      "eval_em_first_line_ratio": 0.4825,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3695,
+      "eval_loss": 0.7758576273918152,
+      "eval_runtime": 205.5041,
+      "eval_samples_per_second": 48.661,
+      "eval_steps_per_second": 0.764,
+      "step": 264000
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 2.7620937726756406e-05,
+      "loss": 0.8116,
+      "step": 264200
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 2.761160664569909e-05,
+      "loss": 0.809,
+      "step": 264400
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 2.760222867478721e-05,
+      "loss": 0.8106,
+      "step": 264600
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 2.759285070387533e-05,
+      "loss": 0.8136,
+      "step": 264800
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 2.7583472732963455e-05,
+      "loss": 0.8178,
+      "step": 265000
+    },
+    {
+      "epoch": 0.31,
+      "eval_em_first_line_ratio": 0.4822,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3696,
+      "eval_loss": 0.7751840353012085,
+      "eval_runtime": 198.0466,
+      "eval_samples_per_second": 50.493,
+      "eval_steps_per_second": 0.793,
+      "step": 265000
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 2.7574094762051573e-05,
+      "loss": 0.8129,
+      "step": 265200
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 2.7564716791139697e-05,
+      "loss": 0.8127,
+      "step": 265400
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 2.7555338820227815e-05,
+      "loss": 0.8089,
+      "step": 265600
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 2.754596084931594e-05,
+      "loss": 0.8037,
+      "step": 265800
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 2.753658287840406e-05,
+      "loss": 0.809,
+      "step": 266000
+    },
+    {
+      "epoch": 0.31,
+      "eval_em_first_line_ratio": 0.4832,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3705,
+      "eval_loss": 0.7752153277397156,
+      "eval_runtime": 205.1639,
+      "eval_samples_per_second": 48.742,
+      "eval_steps_per_second": 0.765,
+      "step": 266000
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 2.752720490749218e-05,
+      "loss": 0.808,
+      "step": 266200
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 2.7517826936580302e-05,
+      "loss": 0.8142,
+      "step": 266400
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 2.7508448965668423e-05,
+      "loss": 0.81,
+      "step": 266600
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 2.7499070994756544e-05,
+      "loss": 0.8091,
+      "step": 266800
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 2.7489693023844665e-05,
+      "loss": 0.8128,
+      "step": 267000
+    },
+    {
+      "epoch": 0.31,
+      "eval_em_first_line_ratio": 0.4818,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3691,
+      "eval_loss": 0.7761086821556091,
+      "eval_runtime": 208.5048,
+      "eval_samples_per_second": 47.961,
+      "eval_steps_per_second": 0.753,
+      "step": 267000
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 2.7480315052932786e-05,
+      "loss": 0.8042,
+      "step": 267200
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 2.747093708202091e-05,
+      "loss": 0.8056,
+      "step": 267400
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 2.746160600096359e-05,
+      "loss": 0.8072,
+      "step": 267600
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 2.745222803005171e-05,
+      "loss": 0.8081,
+      "step": 267800
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 2.7442850059139828e-05,
+      "loss": 0.8122,
+      "step": 268000
+    },
+    {
+      "epoch": 0.31,
+      "eval_em_first_line_ratio": 0.4812,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3679,
+      "eval_loss": 0.7741822004318237,
+      "eval_runtime": 209.8766,
+      "eval_samples_per_second": 47.647,
+      "eval_steps_per_second": 0.748,
+      "step": 268000
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 2.7433472088227952e-05,
+      "loss": 0.8094,
+      "step": 268200
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 2.7424094117316077e-05,
+      "loss": 0.8121,
+      "step": 268400
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 2.7414716146404194e-05,
+      "loss": 0.8135,
+      "step": 268600
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 2.740533817549232e-05,
+      "loss": 0.8101,
+      "step": 268800
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 2.7395960204580436e-05,
+      "loss": 0.8104,
+      "step": 269000
+    },
+    {
+      "epoch": 0.32,
+      "eval_em_first_line_ratio": 0.4827,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3697,
+      "eval_loss": 0.7745413780212402,
+      "eval_runtime": 209.4045,
+      "eval_samples_per_second": 47.754,
+      "eval_steps_per_second": 0.75,
+      "step": 269000
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 2.738658223366856e-05,
+      "loss": 0.8136,
+      "step": 269200
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 2.7377204262756678e-05,
+      "loss": 0.8138,
+      "step": 269400
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 2.7367826291844802e-05,
+      "loss": 0.818,
+      "step": 269600
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 2.7358448320932923e-05,
+      "loss": 0.815,
+      "step": 269800
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 2.7349070350021044e-05,
+      "loss": 0.8091,
+      "step": 270000
+    },
+    {
+      "epoch": 0.32,
+      "eval_em_first_line_ratio": 0.4835,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3702,
+      "eval_loss": 0.7742384672164917,
+      "eval_runtime": 209.6801,
+      "eval_samples_per_second": 47.692,
+      "eval_steps_per_second": 0.749,
+      "step": 270000
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 2.7339692379109165e-05,
+      "loss": 0.8055,
+      "step": 270200
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 2.7330361298051844e-05,
+      "loss": 0.8104,
+      "step": 270400
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 2.732098332713997e-05,
+      "loss": 0.8162,
+      "step": 270600
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 2.731160535622809e-05,
+      "loss": 0.8055,
+      "step": 270800
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 2.7302227385316207e-05,
+      "loss": 0.8135,
+      "step": 271000
+    },
+    {
+      "epoch": 0.32,
+      "eval_em_first_line_ratio": 0.4831,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3704,
+      "eval_loss": 0.7749566435813904,
+      "eval_runtime": 209.9585,
+      "eval_samples_per_second": 47.628,
+      "eval_steps_per_second": 0.748,
+      "step": 271000
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 2.7292849414404332e-05,
+      "loss": 0.8077,
+      "step": 271200
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 2.728347144349245e-05,
+      "loss": 0.81,
+      "step": 271400
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 2.7274093472580574e-05,
+      "loss": 0.8101,
+      "step": 271600
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 2.7264715501668698e-05,
+      "loss": 0.8085,
+      "step": 271800
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 2.7255337530756816e-05,
+      "loss": 0.8078,
+      "step": 272000
+    },
+    {
+      "epoch": 0.32,
+      "eval_em_first_line_ratio": 0.4826,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3697,
+      "eval_loss": 0.7732316255569458,
+      "eval_runtime": 214.6709,
+      "eval_samples_per_second": 46.583,
+      "eval_steps_per_second": 0.731,
+      "step": 272000
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 2.724595955984494e-05,
+      "loss": 0.8073,
+      "step": 272200
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 2.7236581588933058e-05,
+      "loss": 0.8102,
+      "step": 272400
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 2.7227203618021182e-05,
+      "loss": 0.8089,
+      "step": 272600
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 2.72178256471093e-05,
+      "loss": 0.8117,
+      "step": 272800
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 2.7208447676197424e-05,
+      "loss": 0.8098,
+      "step": 273000
+    },
+    {
+      "epoch": 0.32,
+      "eval_em_first_line_ratio": 0.4829,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3703,
+      "eval_loss": 0.7733469009399414,
+      "eval_runtime": 209.2069,
+      "eval_samples_per_second": 47.8,
+      "eval_steps_per_second": 0.75,
+      "step": 273000
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 2.7199069705285545e-05,
+      "loss": 0.8083,
+      "step": 273200
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 2.7189691734373666e-05,
+      "loss": 0.8106,
+      "step": 273400
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 2.718036065331635e-05,
+      "loss": 0.8084,
+      "step": 273600
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 2.7170982682404466e-05,
+      "loss": 0.8114,
+      "step": 273800
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 2.7161604711492587e-05,
+      "loss": 0.8076,
+      "step": 274000
+    },
+    {
+      "epoch": 0.32,
+      "eval_em_first_line_ratio": 0.4818,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3689,
+      "eval_loss": 0.7733954787254333,
+      "eval_runtime": 216.5483,
+      "eval_samples_per_second": 46.179,
+      "eval_steps_per_second": 0.725,
+      "step": 274000
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 2.715222674058071e-05,
+      "loss": 0.8042,
+      "step": 274200
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 2.714284876966883e-05,
+      "loss": 0.8074,
+      "step": 274400
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 2.7133470798756953e-05,
+      "loss": 0.8056,
+      "step": 274600
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 2.712409282784507e-05,
+      "loss": 0.8113,
+      "step": 274800
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 2.7114714856933195e-05,
+      "loss": 0.8108,
+      "step": 275000
+    },
+    {
+      "epoch": 0.32,
+      "eval_em_first_line_ratio": 0.4819,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3699,
+      "eval_loss": 0.773350715637207,
+      "eval_runtime": 207.947,
+      "eval_samples_per_second": 48.089,
+      "eval_steps_per_second": 0.755,
+      "step": 275000
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 2.7105336886021313e-05,
+      "loss": 0.8082,
+      "step": 275200
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 2.7095958915109437e-05,
+      "loss": 0.8092,
+      "step": 275400
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 2.708658094419756e-05,
+      "loss": 0.8071,
+      "step": 275600
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 2.707720297328568e-05,
+      "loss": 0.8074,
+      "step": 275800
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 2.7067825002373803e-05,
+      "loss": 0.8047,
+      "step": 276000
+    },
+    {
+      "epoch": 0.32,
+      "eval_em_first_line_ratio": 0.4818,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3688,
+      "eval_loss": 0.7730898857116699,
+      "eval_runtime": 210.841,
+      "eval_samples_per_second": 47.429,
+      "eval_steps_per_second": 0.745,
+      "step": 276000
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 2.705844703146192e-05,
+      "loss": 0.8098,
+      "step": 276200
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 2.7049069060550045e-05,
+      "loss": 0.8138,
+      "step": 276400
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 2.7039691089638166e-05,
+      "loss": 0.8118,
+      "step": 276600
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 2.7030313118726287e-05,
+      "loss": 0.81,
+      "step": 276800
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 2.7020935147814408e-05,
+      "loss": 0.8054,
+      "step": 277000
+    },
+    {
+      "epoch": 0.32,
+      "eval_em_first_line_ratio": 0.4838,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3714,
+      "eval_loss": 0.7733277082443237,
+      "eval_runtime": 209.6578,
+      "eval_samples_per_second": 47.697,
+      "eval_steps_per_second": 0.749,
+      "step": 277000
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 2.701155717690253e-05,
+      "loss": 0.8061,
+      "step": 277200
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 2.700217920599065e-05,
+      "loss": 0.8068,
+      "step": 277400
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 2.6992801235078768e-05,
+      "loss": 0.8114,
+      "step": 277600
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 2.6983423264166892e-05,
+      "loss": 0.8078,
+      "step": 277800
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 2.6974045293255017e-05,
+      "loss": 0.8059,
+      "step": 278000
+    },
+    {
+      "epoch": 0.33,
+      "eval_em_first_line_ratio": 0.4833,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3699,
+      "eval_loss": 0.7723669409751892,
+      "eval_runtime": 200.686,
+      "eval_samples_per_second": 49.829,
+      "eval_steps_per_second": 0.782,
+      "step": 278000
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 2.6964667322343134e-05,
+      "loss": 0.8139,
+      "step": 278200
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 2.695528935143126e-05,
+      "loss": 0.811,
+      "step": 278400
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 2.6945911380519376e-05,
+      "loss": 0.8093,
+      "step": 278600
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 2.69365334096075e-05,
+      "loss": 0.8067,
+      "step": 278800
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 2.692715543869562e-05,
+      "loss": 0.81,
+      "step": 279000
+    },
+    {
+      "epoch": 0.33,
+      "eval_em_first_line_ratio": 0.4824,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3695,
+      "eval_loss": 0.7719414234161377,
+      "eval_runtime": 203.9706,
+      "eval_samples_per_second": 49.027,
+      "eval_steps_per_second": 0.77,
+      "step": 279000
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 2.6917777467783742e-05,
+      "loss": 0.8057,
+      "step": 279200
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 2.6908399496871863e-05,
+      "loss": 0.806,
+      "step": 279400
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 2.6899021525959984e-05,
+      "loss": 0.8105,
+      "step": 279600
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 2.6889643555048105e-05,
+      "loss": 0.8042,
+      "step": 279800
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 2.6880265584136226e-05,
+      "loss": 0.8025,
+      "step": 280000
+    },
+    {
+      "epoch": 0.33,
+      "eval_em_first_line_ratio": 0.4856,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3727,
+      "eval_loss": 0.7719889879226685,
+      "eval_runtime": 203.3947,
+      "eval_samples_per_second": 49.165,
+      "eval_steps_per_second": 0.772,
+      "step": 280000
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 2.6870887613224347e-05,
+      "loss": 0.8022,
+      "step": 280200
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 2.6861603422021588e-05,
+      "loss": 0.8107,
+      "step": 280400
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 2.685222545110971e-05,
+      "loss": 0.8111,
+      "step": 280600
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 2.684284748019783e-05,
+      "loss": 0.8134,
+      "step": 280800
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 2.683346950928595e-05,
+      "loss": 0.8035,
+      "step": 281000
+    },
+    {
+      "epoch": 0.33,
+      "eval_em_first_line_ratio": 0.4827,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3705,
+      "eval_loss": 0.7726406455039978,
+      "eval_runtime": 208.0439,
+      "eval_samples_per_second": 48.067,
+      "eval_steps_per_second": 0.755,
+      "step": 281000
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 2.6824091538374072e-05,
+      "loss": 0.8122,
+      "step": 281200
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 2.6814713567462196e-05,
+      "loss": 0.8083,
+      "step": 281400
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 2.6805335596550314e-05,
+      "loss": 0.8059,
+      "step": 281600
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 2.6795957625638438e-05,
+      "loss": 0.8041,
+      "step": 281800
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 2.6786579654726556e-05,
+      "loss": 0.8077,
+      "step": 282000
+    },
+    {
+      "epoch": 0.33,
+      "eval_em_first_line_ratio": 0.4833,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3705,
+      "eval_loss": 0.7716552019119263,
+      "eval_runtime": 203.9766,
+      "eval_samples_per_second": 49.025,
+      "eval_steps_per_second": 0.77,
+      "step": 282000
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 2.677720168381468e-05,
+      "loss": 0.8078,
+      "step": 282200
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 2.6767823712902804e-05,
+      "loss": 0.8103,
+      "step": 282400
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 2.6758445741990922e-05,
+      "loss": 0.8044,
+      "step": 282600
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 2.6749067771079046e-05,
+      "loss": 0.8107,
+      "step": 282800
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 2.6739689800167164e-05,
+      "loss": 0.809,
+      "step": 283000
+    },
+    {
+      "epoch": 0.33,
+      "eval_em_first_line_ratio": 0.4834,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3709,
+      "eval_loss": 0.7713944911956787,
+      "eval_runtime": 204.3658,
+      "eval_samples_per_second": 48.932,
+      "eval_steps_per_second": 0.768,
+      "step": 283000
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 2.6730311829255288e-05,
+      "loss": 0.8096,
+      "step": 283200
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 2.6720933858343406e-05,
+      "loss": 0.8083,
+      "step": 283400
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 2.6711555887431527e-05,
+      "loss": 0.809,
+      "step": 283600
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 2.670217791651965e-05,
+      "loss": 0.8068,
+      "step": 283800
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 2.669279994560777e-05,
+      "loss": 0.8007,
+      "step": 284000
+    },
+    {
+      "epoch": 0.33,
+      "eval_em_first_line_ratio": 0.4845,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3717,
+      "eval_loss": 0.770988941192627,
+      "eval_runtime": 203.6643,
+      "eval_samples_per_second": 49.1,
+      "eval_steps_per_second": 0.771,
+      "step": 284000
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 2.6683421974695893e-05,
+      "loss": 0.8077,
+      "step": 284200
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 2.667404400378401e-05,
+      "loss": 0.8099,
+      "step": 284400
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 2.6664666032872135e-05,
+      "loss": 0.8059,
+      "step": 284600
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 2.6655334951814818e-05,
+      "loss": 0.8037,
+      "step": 284800
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 2.6646003870757497e-05,
+      "loss": 0.8076,
+      "step": 285000
+    },
+    {
+      "epoch": 0.33,
+      "eval_em_first_line_ratio": 0.4845,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3718,
+      "eval_loss": 0.7711907029151917,
+      "eval_runtime": 205.9873,
+      "eval_samples_per_second": 48.547,
+      "eval_steps_per_second": 0.762,
+      "step": 285000
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 2.6636625899845618e-05,
+      "loss": 0.8044,
+      "step": 285200
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 2.662724792893374e-05,
+      "loss": 0.8033,
+      "step": 285400
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 2.661786995802186e-05,
+      "loss": 0.7979,
+      "step": 285600
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 2.660849198710998e-05,
+      "loss": 0.8021,
+      "step": 285800
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 2.65991140161981e-05,
+      "loss": 0.7978,
+      "step": 286000
+    },
+    {
+      "epoch": 0.34,
+      "eval_em_first_line_ratio": 0.4855,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3726,
+      "eval_loss": 0.770454466342926,
+      "eval_runtime": 208.244,
+      "eval_samples_per_second": 48.021,
+      "eval_steps_per_second": 0.754,
+      "step": 286000
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 2.6589736045286226e-05,
+      "loss": 0.8025,
+      "step": 286200
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 2.6580358074374343e-05,
+      "loss": 0.8012,
+      "step": 286400
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 2.6570980103462468e-05,
+      "loss": 0.8077,
+      "step": 286600
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 2.6561602132550585e-05,
+      "loss": 0.8079,
+      "step": 286800
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 2.655222416163871e-05,
+      "loss": 0.8011,
+      "step": 287000
+    },
+    {
+      "epoch": 0.34,
+      "eval_em_first_line_ratio": 0.4851,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.372,
+      "eval_loss": 0.7714453339576721,
+      "eval_runtime": 207.6954,
+      "eval_samples_per_second": 48.147,
+      "eval_steps_per_second": 0.756,
+      "step": 287000
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 2.654284619072683e-05,
+      "loss": 0.8067,
+      "step": 287200
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 2.6533468219814952e-05,
+      "loss": 0.8061,
+      "step": 287400
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 2.6524090248903073e-05,
+      "loss": 0.8081,
+      "step": 287600
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 2.6514712277991194e-05,
+      "loss": 0.8015,
+      "step": 287800
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 2.6505334307079315e-05,
+      "loss": 0.8057,
+      "step": 288000
+    },
+    {
+      "epoch": 0.34,
+      "eval_em_first_line_ratio": 0.484,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3709,
+      "eval_loss": 0.7700431942939758,
+      "eval_runtime": 208.4784,
+      "eval_samples_per_second": 47.967,
+      "eval_steps_per_second": 0.753,
+      "step": 288000
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 2.6495956336167436e-05,
+      "loss": 0.8062,
+      "step": 288200
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 2.6486578365255557e-05,
+      "loss": 0.8091,
+      "step": 288400
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 2.647720039434368e-05,
+      "loss": 0.8025,
+      "step": 288600
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 2.64678224234318e-05,
+      "loss": 0.8053,
+      "step": 288800
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 2.6458444452519923e-05,
+      "loss": 0.8035,
+      "step": 289000
+    },
+    {
+      "epoch": 0.34,
+      "eval_em_first_line_ratio": 0.4842,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3714,
+      "eval_loss": 0.7695319056510925,
+      "eval_runtime": 205.6982,
+      "eval_samples_per_second": 48.615,
+      "eval_steps_per_second": 0.763,
+      "step": 289000
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 2.64491133714626e-05,
+      "loss": 0.8067,
+      "step": 289200
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 2.643978229040528e-05,
+      "loss": 0.8024,
+      "step": 289400
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 2.6430451209347964e-05,
+      "loss": 0.806,
+      "step": 289600
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 2.6421073238436085e-05,
+      "loss": 0.8052,
+      "step": 289800
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 2.6411695267524206e-05,
+      "loss": 0.8016,
+      "step": 290000
+    },
+    {
+      "epoch": 0.34,
+      "eval_em_first_line_ratio": 0.4849,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3718,
+      "eval_loss": 0.7699447274208069,
+      "eval_runtime": 207.3853,
+      "eval_samples_per_second": 48.219,
+      "eval_steps_per_second": 0.757,
+      "step": 290000
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 2.6402317296612327e-05,
+      "loss": 0.8049,
+      "step": 290200
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 2.6392939325700448e-05,
+      "loss": 0.8071,
+      "step": 290400
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 2.6383561354788572e-05,
+      "loss": 0.8022,
+      "step": 290600
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 2.637418338387669e-05,
+      "loss": 0.8,
+      "step": 290800
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 2.6364805412964814e-05,
+      "loss": 0.7999,
+      "step": 291000
+    },
+    {
+      "epoch": 0.34,
+      "eval_em_first_line_ratio": 0.4831,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3707,
+      "eval_loss": 0.7706863284111023,
+      "eval_runtime": 207.0421,
+      "eval_samples_per_second": 48.299,
+      "eval_steps_per_second": 0.758,
+      "step": 291000
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 2.635542744205293e-05,
+      "loss": 0.8022,
+      "step": 291200
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 2.6346049471141056e-05,
+      "loss": 0.805,
+      "step": 291400
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 2.6336718390083738e-05,
+      "loss": 0.8052,
+      "step": 291600
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 2.6327340419171856e-05,
+      "loss": 0.8031,
+      "step": 291800
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 2.631796244825998e-05,
+      "loss": 0.8059,
+      "step": 292000
+    },
+    {
+      "epoch": 0.34,
+      "eval_em_first_line_ratio": 0.4824,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3699,
+      "eval_loss": 0.7700636982917786,
+      "eval_runtime": 208.2803,
+      "eval_samples_per_second": 48.012,
+      "eval_steps_per_second": 0.754,
+      "step": 292000
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 2.6308584477348098e-05,
+      "loss": 0.8055,
+      "step": 292200
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 2.629925339629078e-05,
+      "loss": 0.8056,
+      "step": 292400
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 2.6289875425378905e-05,
+      "loss": 0.8083,
+      "step": 292600
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 2.6280497454467022e-05,
+      "loss": 0.807,
+      "step": 292800
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 2.6271119483555147e-05,
+      "loss": 0.806,
+      "step": 293000
+    },
+    {
+      "epoch": 0.34,
+      "eval_em_first_line_ratio": 0.4822,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3699,
+      "eval_loss": 0.7690271735191345,
+      "eval_runtime": 215.6839,
+      "eval_samples_per_second": 46.364,
+      "eval_steps_per_second": 0.728,
+      "step": 293000
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 2.6261741512643264e-05,
+      "loss": 0.8019,
+      "step": 293200
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 2.625236354173139e-05,
+      "loss": 0.8072,
+      "step": 293400
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 2.6242985570819506e-05,
+      "loss": 0.8043,
+      "step": 293600
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 2.623360759990763e-05,
+      "loss": 0.7993,
+      "step": 293800
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 2.622422962899575e-05,
+      "loss": 0.8045,
+      "step": 294000
+    },
+    {
+      "epoch": 0.34,
+      "eval_em_first_line_ratio": 0.4839,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3711,
+      "eval_loss": 0.7691730260848999,
+      "eval_runtime": 210.2962,
+      "eval_samples_per_second": 47.552,
+      "eval_steps_per_second": 0.747,
+      "step": 294000
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 2.621485165808387e-05,
+      "loss": 0.8077,
+      "step": 294200
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 2.6205473687171993e-05,
+      "loss": 0.8016,
+      "step": 294400
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 2.619609571626011e-05,
+      "loss": 0.808,
+      "step": 294600
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 2.6186717745348235e-05,
+      "loss": 0.8001,
+      "step": 294800
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 2.617733977443636e-05,
+      "loss": 0.8095,
+      "step": 295000
+    },
+    {
+      "epoch": 0.35,
+      "eval_em_first_line_ratio": 0.483,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3709,
+      "eval_loss": 0.7695897221565247,
+      "eval_runtime": 209.7932,
+      "eval_samples_per_second": 47.666,
+      "eval_steps_per_second": 0.748,
+      "step": 295000
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 2.6167961803524477e-05,
+      "loss": 0.8028,
+      "step": 295200
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 2.6158583832612602e-05,
+      "loss": 0.8023,
+      "step": 295400
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 2.6149252751555277e-05,
+      "loss": 0.8063,
+      "step": 295600
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 2.6139874780643402e-05,
+      "loss": 0.8026,
+      "step": 295800
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 2.613049680973152e-05,
+      "loss": 0.8054,
+      "step": 296000
+    },
+    {
+      "epoch": 0.35,
+      "eval_em_first_line_ratio": 0.4831,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.371,
+      "eval_loss": 0.7693973779678345,
+      "eval_runtime": 212.4573,
+      "eval_samples_per_second": 47.068,
+      "eval_steps_per_second": 0.739,
+      "step": 296000
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 2.6121118838819644e-05,
+      "loss": 0.8043,
+      "step": 296200
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 2.6111740867907768e-05,
+      "loss": 0.8062,
+      "step": 296400
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 2.6102362896995886e-05,
+      "loss": 0.802,
+      "step": 296600
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 2.609298492608401e-05,
+      "loss": 0.8039,
+      "step": 296800
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 2.6083606955172128e-05,
+      "loss": 0.7943,
+      "step": 297000
+    },
+    {
+      "epoch": 0.35,
+      "eval_em_first_line_ratio": 0.4858,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3736,
+      "eval_loss": 0.7691106200218201,
+      "eval_runtime": 205.1448,
+      "eval_samples_per_second": 48.746,
+      "eval_steps_per_second": 0.765,
+      "step": 297000
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 2.607422898426025e-05,
+      "loss": 0.7984,
+      "step": 297200
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 2.6064851013348373e-05,
+      "loss": 0.8082,
+      "step": 297400
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 2.605547304243649e-05,
+      "loss": 0.8045,
+      "step": 297600
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 2.6046095071524615e-05,
+      "loss": 0.8025,
+      "step": 297800
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 2.6036717100612732e-05,
+      "loss": 0.8048,
+      "step": 298000
+    },
+    {
+      "epoch": 0.35,
+      "eval_em_first_line_ratio": 0.483,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3712,
+      "eval_loss": 0.7693120837211609,
+      "eval_runtime": 207.7272,
+      "eval_samples_per_second": 48.14,
+      "eval_steps_per_second": 0.756,
+      "step": 298000
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 2.6027339129700857e-05,
+      "loss": 0.8018,
+      "step": 298200
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 2.6017961158788974e-05,
+      "loss": 0.8031,
+      "step": 298400
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 2.60085831878771e-05,
+      "loss": 0.802,
+      "step": 298600
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 2.5999205216965223e-05,
+      "loss": 0.8023,
+      "step": 298800
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 2.598982724605334e-05,
+      "loss": 0.803,
+      "step": 299000
+    },
+    {
+      "epoch": 0.35,
+      "eval_em_first_line_ratio": 0.4819,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3698,
+      "eval_loss": 0.7684518694877625,
+      "eval_runtime": 204.9765,
+      "eval_samples_per_second": 48.786,
+      "eval_steps_per_second": 0.766,
+      "step": 299000
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 2.5980449275141465e-05,
+      "loss": 0.8079,
+      "step": 299200
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 2.5971071304229583e-05,
+      "loss": 0.8053,
+      "step": 299400
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 2.5961693333317707e-05,
+      "loss": 0.8069,
+      "step": 299600
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 2.5952315362405828e-05,
+      "loss": 0.8009,
+      "step": 299800
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 2.594293739149395e-05,
+      "loss": 0.8035,
+      "step": 300000
+    },
+    {
+      "epoch": 0.35,
+      "eval_em_first_line_ratio": 0.4825,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3701,
+      "eval_loss": 0.7686658501625061,
+      "eval_runtime": 209.1104,
+      "eval_samples_per_second": 47.822,
+      "eval_steps_per_second": 0.751,
+      "step": 300000
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 2.593355942058207e-05,
+      "loss": 0.7977,
+      "step": 300200
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 2.592418144967019e-05,
+      "loss": 0.8006,
+      "step": 300400
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 2.5914803478758312e-05,
+      "loss": 0.7981,
+      "step": 300600
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 2.590542550784643e-05,
+      "loss": 0.8034,
+      "step": 300800
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 2.5896047536934554e-05,
+      "loss": 0.8013,
+      "step": 301000
+    },
+    {
+      "epoch": 0.35,
+      "eval_em_first_line_ratio": 0.4825,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3702,
+      "eval_loss": 0.7683463096618652,
+      "eval_runtime": 204.6416,
+      "eval_samples_per_second": 48.866,
+      "eval_steps_per_second": 0.767,
+      "step": 301000
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 2.5886669566022678e-05,
+      "loss": 0.7993,
+      "step": 301200
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 2.5877291595110796e-05,
+      "loss": 0.8067,
+      "step": 301400
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 2.586791362419892e-05,
+      "loss": 0.804,
+      "step": 301600
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 2.5858535653287038e-05,
+      "loss": 0.7961,
+      "step": 301800
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 2.584920457222972e-05,
+      "loss": 0.7987,
+      "step": 302000
+    },
+    {
+      "epoch": 0.35,
+      "eval_em_first_line_ratio": 0.4827,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3704,
+      "eval_loss": 0.768159806728363,
+      "eval_runtime": 211.5747,
+      "eval_samples_per_second": 47.265,
+      "eval_steps_per_second": 0.742,
+      "step": 302000
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 2.5839826601317845e-05,
+      "loss": 0.7985,
+      "step": 302200
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 2.583049552026052e-05,
+      "loss": 0.8044,
+      "step": 302400
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 2.5821117549348645e-05,
+      "loss": 0.8013,
+      "step": 302600
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 2.5811739578436762e-05,
+      "loss": 0.7996,
+      "step": 302800
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 2.5802361607524887e-05,
+      "loss": 0.8039,
+      "step": 303000
+    },
+    {
+      "epoch": 0.36,
+      "eval_em_first_line_ratio": 0.4827,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3708,
+      "eval_loss": 0.767542839050293,
+      "eval_runtime": 213.4533,
+      "eval_samples_per_second": 46.849,
+      "eval_steps_per_second": 0.736,
+      "step": 303000
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 2.5792983636613008e-05,
+      "loss": 0.8006,
+      "step": 303200
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 2.578360566570113e-05,
+      "loss": 0.8014,
+      "step": 303400
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 2.577422769478925e-05,
+      "loss": 0.8012,
+      "step": 303600
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 2.576484972387737e-05,
+      "loss": 0.8064,
+      "step": 303800
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 2.575547175296549e-05,
+      "loss": 0.802,
+      "step": 304000
+    },
+    {
+      "epoch": 0.36,
+      "eval_em_first_line_ratio": 0.4874,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3753,
+      "eval_loss": 0.768119215965271,
+      "eval_runtime": 209.9307,
+      "eval_samples_per_second": 47.635,
+      "eval_steps_per_second": 0.748,
+      "step": 304000
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 2.5746093782053612e-05,
+      "loss": 0.8018,
+      "step": 304200
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 2.5736715811141733e-05,
+      "loss": 0.8029,
+      "step": 304400
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 2.5727337840229858e-05,
+      "loss": 0.8009,
+      "step": 304600
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 2.5717959869317975e-05,
+      "loss": 0.8012,
+      "step": 304800
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 2.57085818984061e-05,
+      "loss": 0.7971,
+      "step": 305000
+    },
+    {
+      "epoch": 0.36,
+      "eval_em_first_line_ratio": 0.4864,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3741,
+      "eval_loss": 0.7683139443397522,
+      "eval_runtime": 206.3585,
+      "eval_samples_per_second": 48.459,
+      "eval_steps_per_second": 0.761,
+      "step": 305000
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 2.5699203927494217e-05,
+      "loss": 0.7993,
+      "step": 305200
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 2.5689825956582342e-05,
+      "loss": 0.8062,
+      "step": 305400
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 2.5680447985670466e-05,
+      "loss": 0.8033,
+      "step": 305600
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 2.5671070014758584e-05,
+      "loss": 0.7997,
+      "step": 305800
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 2.5661692043846708e-05,
+      "loss": 0.8021,
+      "step": 306000
+    },
+    {
+      "epoch": 0.36,
+      "eval_em_first_line_ratio": 0.4814,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3694,
+      "eval_loss": 0.7673913836479187,
+      "eval_runtime": 209.0823,
+      "eval_samples_per_second": 47.828,
+      "eval_steps_per_second": 0.751,
+      "step": 306000
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 2.5652314072934826e-05,
+      "loss": 0.8052,
+      "step": 306200
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 2.564293610202295e-05,
+      "loss": 0.8101,
+      "step": 306400
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 2.5633558131111068e-05,
+      "loss": 0.8003,
+      "step": 306600
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 2.562418016019919e-05,
+      "loss": 0.8049,
+      "step": 306800
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 2.5614802189287313e-05,
+      "loss": 0.8018,
+      "step": 307000
+    },
+    {
+      "epoch": 0.36,
+      "eval_em_first_line_ratio": 0.484,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3718,
+      "eval_loss": 0.767546534538269,
+      "eval_runtime": 208.2072,
+      "eval_samples_per_second": 48.029,
+      "eval_steps_per_second": 0.754,
+      "step": 307000
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 2.560542421837543e-05,
+      "loss": 0.8037,
+      "step": 307200
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 2.5596046247463555e-05,
+      "loss": 0.8036,
+      "step": 307400
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 2.5586668276551672e-05,
+      "loss": 0.8054,
+      "step": 307600
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 2.5577290305639797e-05,
+      "loss": 0.7985,
+      "step": 307800
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 2.556795922458248e-05,
+      "loss": 0.8005,
+      "step": 308000
+    },
+    {
+      "epoch": 0.36,
+      "eval_em_first_line_ratio": 0.4853,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3724,
+      "eval_loss": 0.7673218846321106,
+      "eval_runtime": 202.9263,
+      "eval_samples_per_second": 49.279,
+      "eval_steps_per_second": 0.774,
+      "step": 308000
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 2.5558581253670597e-05,
+      "loss": 0.8026,
+      "step": 308200
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 2.554920328275872e-05,
+      "loss": 0.7943,
+      "step": 308400
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 2.553982531184684e-05,
+      "loss": 0.803,
+      "step": 308600
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 2.5530447340934963e-05,
+      "loss": 0.8017,
+      "step": 308800
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 2.552106937002308e-05,
+      "loss": 0.7997,
+      "step": 309000
+    },
+    {
+      "epoch": 0.36,
+      "eval_em_first_line_ratio": 0.4841,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3715,
+      "eval_loss": 0.7664623260498047,
+      "eval_runtime": 208.9849,
+      "eval_samples_per_second": 47.85,
+      "eval_steps_per_second": 0.751,
+      "step": 309000
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 2.5511691399111205e-05,
+      "loss": 0.8086,
+      "step": 309200
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 2.550231342819933e-05,
+      "loss": 0.7963,
+      "step": 309400
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 2.5492935457287447e-05,
+      "loss": 0.7988,
+      "step": 309600
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 2.5483557486375568e-05,
+      "loss": 0.7977,
+      "step": 309800
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 2.547417951546369e-05,
+      "loss": 0.7998,
+      "step": 310000
+    },
+    {
+      "epoch": 0.36,
+      "eval_em_first_line_ratio": 0.4843,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3722,
+      "eval_loss": 0.766989529132843,
+      "eval_runtime": 208.4503,
+      "eval_samples_per_second": 47.973,
+      "eval_steps_per_second": 0.753,
+      "step": 310000
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 2.546480154455181e-05,
+      "loss": 0.8034,
+      "step": 310200
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 2.5455470463494492e-05,
+      "loss": 0.8014,
+      "step": 310400
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 2.5446092492582613e-05,
+      "loss": 0.7985,
+      "step": 310600
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 2.5436714521670734e-05,
+      "loss": 0.8066,
+      "step": 310800
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 2.5427336550758855e-05,
+      "loss": 0.7916,
+      "step": 311000
+    },
+    {
+      "epoch": 0.36,
+      "eval_em_first_line_ratio": 0.4833,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3714,
+      "eval_loss": 0.7662694454193115,
+      "eval_runtime": 209.8452,
+      "eval_samples_per_second": 47.654,
+      "eval_steps_per_second": 0.748,
+      "step": 311000
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 2.5417958579846976e-05,
+      "loss": 0.8032,
+      "step": 311200
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 2.5408580608935097e-05,
+      "loss": 0.7978,
+      "step": 311400
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 2.539920263802322e-05,
+      "loss": 0.8046,
+      "step": 311600
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 2.5389824667111343e-05,
+      "loss": 0.8033,
+      "step": 311800
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 2.538044669619946e-05,
+      "loss": 0.7995,
+      "step": 312000
+    },
+    {
+      "epoch": 0.37,
+      "eval_em_first_line_ratio": 0.484,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3716,
+      "eval_loss": 0.7666922211647034,
+      "eval_runtime": 208.0129,
+      "eval_samples_per_second": 48.074,
+      "eval_steps_per_second": 0.755,
+      "step": 312000
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 2.5371068725287585e-05,
+      "loss": 0.7974,
+      "step": 312200
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 2.5361690754375702e-05,
+      "loss": 0.8004,
+      "step": 312400
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 2.5352312783463827e-05,
+      "loss": 0.8029,
+      "step": 312600
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 2.534298170240651e-05,
+      "loss": 0.7943,
+      "step": 312800
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 2.5333603731494627e-05,
+      "loss": 0.8026,
+      "step": 313000
+    },
+    {
+      "epoch": 0.37,
+      "eval_em_first_line_ratio": 0.4835,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3706,
+      "eval_loss": 0.7656763195991516,
+      "eval_runtime": 206.6467,
+      "eval_samples_per_second": 48.392,
+      "eval_steps_per_second": 0.76,
+      "step": 313000
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 2.532422576058275e-05,
+      "loss": 0.7999,
+      "step": 313200
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 2.531484778967087e-05,
+      "loss": 0.8046,
+      "step": 313400
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 2.5305469818758993e-05,
+      "loss": 0.7986,
+      "step": 313600
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 2.5296091847847114e-05,
+      "loss": 0.8009,
+      "step": 313800
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 2.5286713876935235e-05,
+      "loss": 0.7996,
+      "step": 314000
+    },
+    {
+      "epoch": 0.37,
+      "eval_em_first_line_ratio": 0.4847,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3722,
+      "eval_loss": 0.765766978263855,
+      "eval_runtime": 206.8974,
+      "eval_samples_per_second": 48.333,
+      "eval_steps_per_second": 0.759,
+      "step": 314000
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 2.5277335906023356e-05,
+      "loss": 0.7953,
+      "step": 314200
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 2.5267957935111477e-05,
+      "loss": 0.8013,
+      "step": 314400
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 2.5258579964199598e-05,
+      "loss": 0.8012,
+      "step": 314600
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 2.524920199328772e-05,
+      "loss": 0.8041,
+      "step": 314800
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 2.52398709122304e-05,
+      "loss": 0.7933,
+      "step": 315000
+    },
+    {
+      "epoch": 0.37,
+      "eval_em_first_line_ratio": 0.483,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3704,
+      "eval_loss": 0.7653993368148804,
+      "eval_runtime": 202.0104,
+      "eval_samples_per_second": 49.502,
+      "eval_steps_per_second": 0.777,
+      "step": 315000
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 2.5230492941318522e-05,
+      "loss": 0.797,
+      "step": 315200
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 2.5221114970406643e-05,
+      "loss": 0.8025,
+      "step": 315400
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 2.5211736999494764e-05,
+      "loss": 0.7974,
+      "step": 315600
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 2.5202359028582882e-05,
+      "loss": 0.7961,
+      "step": 315800
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 2.5192981057671006e-05,
+      "loss": 0.7965,
+      "step": 316000
+    },
+    {
+      "epoch": 0.37,
+      "eval_em_first_line_ratio": 0.4837,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3712,
+      "eval_loss": 0.7654532790184021,
+      "eval_runtime": 202.726,
+      "eval_samples_per_second": 49.328,
+      "eval_steps_per_second": 0.774,
+      "step": 316000
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 2.518360308675913e-05,
+      "loss": 0.7973,
+      "step": 316200
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 2.5174225115847248e-05,
+      "loss": 0.7958,
+      "step": 316400
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 2.5164847144935372e-05,
+      "loss": 0.7986,
+      "step": 316600
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 2.515546917402349e-05,
+      "loss": 0.7962,
+      "step": 316800
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 2.5146091203111614e-05,
+      "loss": 0.7998,
+      "step": 317000
+    },
+    {
+      "epoch": 0.37,
+      "eval_em_first_line_ratio": 0.4858,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3726,
+      "eval_loss": 0.765465259552002,
+      "eval_runtime": 209.71,
+      "eval_samples_per_second": 47.685,
+      "eval_steps_per_second": 0.749,
+      "step": 317000
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 2.5136713232199732e-05,
+      "loss": 0.8046,
+      "step": 317200
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 2.5127335261287856e-05,
+      "loss": 0.7974,
+      "step": 317400
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 2.5117957290375977e-05,
+      "loss": 0.8022,
+      "step": 317600
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 2.51085793194641e-05,
+      "loss": 0.7982,
+      "step": 317800
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 2.509920134855222e-05,
+      "loss": 0.8,
+      "step": 318000
+    },
+    {
+      "epoch": 0.37,
+      "eval_em_first_line_ratio": 0.4829,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3699,
+      "eval_loss": 0.7653586268424988,
+      "eval_runtime": 207.9296,
+      "eval_samples_per_second": 48.093,
+      "eval_steps_per_second": 0.755,
+      "step": 318000
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 2.50898702674949e-05,
+      "loss": 0.7969,
+      "step": 318200
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 2.5080492296583023e-05,
+      "loss": 0.8001,
+      "step": 318400
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 2.5071114325671144e-05,
+      "loss": 0.7977,
+      "step": 318600
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 2.506173635475926e-05,
+      "loss": 0.8007,
+      "step": 318800
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 2.5052358383847386e-05,
+      "loss": 0.8014,
+      "step": 319000
+    },
+    {
+      "epoch": 0.37,
+      "eval_em_first_line_ratio": 0.4824,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3696,
+      "eval_loss": 0.7649000883102417,
+      "eval_runtime": 211.8623,
+      "eval_samples_per_second": 47.2,
+      "eval_steps_per_second": 0.741,
+      "step": 319000
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 2.5042980412935503e-05,
+      "loss": 0.8019,
+      "step": 319200
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 2.5033602442023628e-05,
+      "loss": 0.7975,
+      "step": 319400
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 2.5024224471111745e-05,
+      "loss": 0.7971,
+      "step": 319600
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 2.501484650019987e-05,
+      "loss": 0.8021,
+      "step": 319800
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 2.5005468529287994e-05,
+      "loss": 0.7945,
+      "step": 320000
+    },
+    {
+      "epoch": 0.38,
+      "eval_em_first_line_ratio": 0.4839,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3711,
+      "eval_loss": 0.7648820281028748,
+      "eval_runtime": 208.3322,
+      "eval_samples_per_second": 48.0,
+      "eval_steps_per_second": 0.754,
+      "step": 320000
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 2.499609055837611e-05,
+      "loss": 0.7961,
+      "step": 320200
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 2.4986712587464236e-05,
+      "loss": 0.7993,
+      "step": 320400
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 2.497738150640691e-05,
+      "loss": 0.7995,
+      "step": 320600
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 2.4968003535495036e-05,
+      "loss": 0.796,
+      "step": 320800
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 2.495862556458316e-05,
+      "loss": 0.798,
+      "step": 321000
+    },
+    {
+      "epoch": 0.38,
+      "eval_em_first_line_ratio": 0.483,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3702,
+      "eval_loss": 0.7644860744476318,
+      "eval_runtime": 210.359,
+      "eval_samples_per_second": 47.538,
+      "eval_steps_per_second": 0.746,
+      "step": 321000
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 2.4949247593671278e-05,
+      "loss": 0.7941,
+      "step": 321200
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 2.493991651261396e-05,
+      "loss": 0.8011,
+      "step": 321400
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 2.4930538541702078e-05,
+      "loss": 0.7993,
+      "step": 321600
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 2.4921160570790202e-05,
+      "loss": 0.7974,
+      "step": 321800
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 2.4911782599878323e-05,
+      "loss": 0.8016,
+      "step": 322000
+    },
+    {
+      "epoch": 0.38,
+      "eval_em_first_line_ratio": 0.4844,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3718,
+      "eval_loss": 0.7652278542518616,
+      "eval_runtime": 196.0393,
+      "eval_samples_per_second": 51.01,
+      "eval_steps_per_second": 0.801,
+      "step": 322000
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 2.4902404628966444e-05,
+      "loss": 0.8018,
+      "step": 322200
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 2.4893026658054565e-05,
+      "loss": 0.7999,
+      "step": 322400
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 2.4883648687142686e-05,
+      "loss": 0.7978,
+      "step": 322600
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 2.4874270716230807e-05,
+      "loss": 0.7961,
+      "step": 322800
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 2.4864892745318928e-05,
+      "loss": 0.7936,
+      "step": 323000
+    },
+    {
+      "epoch": 0.38,
+      "eval_em_first_line_ratio": 0.4836,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3713,
+      "eval_loss": 0.7654072046279907,
+      "eval_runtime": 201.2411,
+      "eval_samples_per_second": 49.692,
+      "eval_steps_per_second": 0.78,
+      "step": 323000
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 2.485551477440705e-05,
+      "loss": 0.7985,
+      "step": 323200
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 2.4846136803495174e-05,
+      "loss": 0.794,
+      "step": 323400
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 2.483675883258329e-05,
+      "loss": 0.7945,
+      "step": 323600
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 2.4827380861671416e-05,
+      "loss": 0.8006,
+      "step": 323800
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 2.4818002890759533e-05,
+      "loss": 0.7922,
+      "step": 324000
+    },
+    {
+      "epoch": 0.38,
+      "eval_em_first_line_ratio": 0.4836,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3707,
+      "eval_loss": 0.7644955515861511,
+      "eval_runtime": 212.6036,
+      "eval_samples_per_second": 47.036,
+      "eval_steps_per_second": 0.738,
+      "step": 324000
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 2.4808624919847657e-05,
+      "loss": 0.802,
+      "step": 324200
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 2.479924694893578e-05,
+      "loss": 0.7973,
+      "step": 324400
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 2.47898689780239e-05,
+      "loss": 0.7979,
+      "step": 324600
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 2.4780537896966582e-05,
+      "loss": 0.7982,
+      "step": 324800
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 2.47711599260547e-05,
+      "loss": 0.7975,
+      "step": 325000
+    },
+    {
+      "epoch": 0.38,
+      "eval_em_first_line_ratio": 0.4844,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3717,
+      "eval_loss": 0.7639663219451904,
+      "eval_runtime": 209.578,
+      "eval_samples_per_second": 47.715,
+      "eval_steps_per_second": 0.749,
+      "step": 325000
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 2.4761781955142824e-05,
+      "loss": 0.8005,
+      "step": 325200
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 2.4752403984230945e-05,
+      "loss": 0.8004,
+      "step": 325400
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 2.4743026013319066e-05,
+      "loss": 0.7999,
+      "step": 325600
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 2.4733648042407187e-05,
+      "loss": 0.7997,
+      "step": 325800
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 2.4724270071495308e-05,
+      "loss": 0.7945,
+      "step": 326000
+    },
+    {
+      "epoch": 0.38,
+      "eval_em_first_line_ratio": 0.4856,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3733,
+      "eval_loss": 0.764077365398407,
+      "eval_runtime": 214.1353,
+      "eval_samples_per_second": 46.699,
+      "eval_steps_per_second": 0.733,
+      "step": 326000
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 2.471489210058343e-05,
+      "loss": 0.7978,
+      "step": 326200
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 2.470551412967155e-05,
+      "loss": 0.7979,
+      "step": 326400
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 2.469613615875967e-05,
+      "loss": 0.7948,
+      "step": 326600
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 2.4686758187847795e-05,
+      "loss": 0.7952,
+      "step": 326800
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 2.4677380216935913e-05,
+      "loss": 0.7992,
+      "step": 327000
+    },
+    {
+      "epoch": 0.38,
+      "eval_em_first_line_ratio": 0.4874,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3747,
+      "eval_loss": 0.7638093829154968,
+      "eval_runtime": 205.1867,
+      "eval_samples_per_second": 48.736,
+      "eval_steps_per_second": 0.765,
+      "step": 327000
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 2.4668002246024037e-05,
+      "loss": 0.7928,
+      "step": 327200
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 2.4658624275112155e-05,
+      "loss": 0.8007,
+      "step": 327400
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 2.464924630420028e-05,
+      "loss": 0.7923,
+      "step": 327600
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 2.46398683332884e-05,
+      "loss": 0.8027,
+      "step": 327800
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 2.463049036237652e-05,
+      "loss": 0.7948,
+      "step": 328000
+    },
+    {
+      "epoch": 0.38,
+      "eval_em_first_line_ratio": 0.4869,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3739,
+      "eval_loss": 0.7635656595230103,
+      "eval_runtime": 226.6175,
+      "eval_samples_per_second": 44.127,
+      "eval_steps_per_second": 0.693,
+      "step": 328000
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 2.4621112391464642e-05,
+      "loss": 0.7947,
+      "step": 328200
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 2.4611734420552763e-05,
+      "loss": 0.7977,
+      "step": 328400
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 2.4602356449640884e-05,
+      "loss": 0.7994,
+      "step": 328600
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 2.4593025368583563e-05,
+      "loss": 0.792,
+      "step": 328800
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 2.4583647397671687e-05,
+      "loss": 0.7975,
+      "step": 329000
+    },
+    {
+      "epoch": 0.39,
+      "eval_em_first_line_ratio": 0.4861,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3734,
+      "eval_loss": 0.7640921473503113,
+      "eval_runtime": 207.9823,
+      "eval_samples_per_second": 48.081,
+      "eval_steps_per_second": 0.755,
+      "step": 329000
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 2.4574269426759808e-05,
+      "loss": 0.7924,
+      "step": 329200
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 2.456489145584793e-05,
+      "loss": 0.7983,
+      "step": 329400
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 2.455551348493605e-05,
+      "loss": 0.8042,
+      "step": 329600
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 2.454613551402417e-05,
+      "loss": 0.7943,
+      "step": 329800
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 2.4536757543112292e-05,
+      "loss": 0.7985,
+      "step": 330000
+    },
+    {
+      "epoch": 0.39,
+      "eval_em_first_line_ratio": 0.4847,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3721,
+      "eval_loss": 0.7635481953620911,
+      "eval_runtime": 211.9631,
+      "eval_samples_per_second": 47.178,
+      "eval_steps_per_second": 0.741,
+      "step": 330000
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 2.4527379572200416e-05,
+      "loss": 0.7995,
+      "step": 330200
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 2.4518001601288534e-05,
+      "loss": 0.7903,
+      "step": 330400
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 2.450862363037666e-05,
+      "loss": 0.7937,
+      "step": 330600
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 2.4499245659464776e-05,
+      "loss": 0.7929,
+      "step": 330800
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 2.448991457840746e-05,
+      "loss": 0.7995,
+      "step": 331000
+    },
+    {
+      "epoch": 0.39,
+      "eval_em_first_line_ratio": 0.4853,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3727,
+      "eval_loss": 0.7632489204406738,
+      "eval_runtime": 209.2918,
+      "eval_samples_per_second": 47.78,
+      "eval_steps_per_second": 0.75,
+      "step": 331000
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 2.4480536607495583e-05,
+      "loss": 0.7909,
+      "step": 331200
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 2.44711586365837e-05,
+      "loss": 0.7936,
+      "step": 331400
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 2.4461780665671825e-05,
+      "loss": 0.7887,
+      "step": 331600
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 2.4452402694759942e-05,
+      "loss": 0.8017,
+      "step": 331800
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 2.4443024723848067e-05,
+      "loss": 0.7986,
+      "step": 332000
+    },
+    {
+      "epoch": 0.39,
+      "eval_em_first_line_ratio": 0.4863,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3735,
+      "eval_loss": 0.7629184722900391,
+      "eval_runtime": 205.8792,
+      "eval_samples_per_second": 48.572,
+      "eval_steps_per_second": 0.763,
+      "step": 332000
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 2.4433646752936184e-05,
+      "loss": 0.7954,
+      "step": 332200
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 2.442426878202431e-05,
+      "loss": 0.7978,
+      "step": 332400
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 2.441489081111243e-05,
+      "loss": 0.7975,
+      "step": 332600
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 2.440551284020055e-05,
+      "loss": 0.7984,
+      "step": 332800
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 2.439613486928867e-05,
+      "loss": 0.7968,
+      "step": 333000
+    },
+    {
+      "epoch": 0.39,
+      "eval_em_first_line_ratio": 0.4834,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3705,
+      "eval_loss": 0.7628031969070435,
+      "eval_runtime": 214.5724,
+      "eval_samples_per_second": 46.604,
+      "eval_steps_per_second": 0.732,
+      "step": 333000
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 2.4386756898376793e-05,
+      "loss": 0.7954,
+      "step": 333200
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 2.4377378927464914e-05,
+      "loss": 0.7928,
+      "step": 333400
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 2.4368000956553035e-05,
+      "loss": 0.7971,
+      "step": 333600
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 2.4358622985641156e-05,
+      "loss": 0.7967,
+      "step": 333800
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 2.4349291904583838e-05,
+      "loss": 0.7914,
+      "step": 334000
+    },
+    {
+      "epoch": 0.39,
+      "eval_em_first_line_ratio": 0.4844,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3718,
+      "eval_loss": 0.7623757719993591,
+      "eval_runtime": 204.4283,
+      "eval_samples_per_second": 48.917,
+      "eval_steps_per_second": 0.768,
+      "step": 334000
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 2.4339913933671956e-05,
+      "loss": 0.7978,
+      "step": 334200
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 2.433053596276008e-05,
+      "loss": 0.8006,
+      "step": 334400
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 2.4321157991848198e-05,
+      "loss": 0.7936,
+      "step": 334600
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 2.4311780020936322e-05,
+      "loss": 0.7933,
+      "step": 334800
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 2.4302402050024446e-05,
+      "loss": 0.7974,
+      "step": 335000
+    },
+    {
+      "epoch": 0.39,
+      "eval_em_first_line_ratio": 0.4835,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.371,
+      "eval_loss": 0.7627648711204529,
+      "eval_runtime": 217.2334,
+      "eval_samples_per_second": 46.033,
+      "eval_steps_per_second": 0.723,
+      "step": 335000
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 2.4293024079112564e-05,
+      "loss": 0.796,
+      "step": 335200
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 2.4283646108200688e-05,
+      "loss": 0.7978,
+      "step": 335400
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 2.4274315027143364e-05,
+      "loss": 0.7952,
+      "step": 335600
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 2.426493705623149e-05,
+      "loss": 0.7935,
+      "step": 335800
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 2.4255559085319613e-05,
+      "loss": 0.7936,
+      "step": 336000
+    },
+    {
+      "epoch": 0.39,
+      "eval_em_first_line_ratio": 0.4853,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3726,
+      "eval_loss": 0.7633440494537354,
+      "eval_runtime": 204.7595,
+      "eval_samples_per_second": 48.838,
+      "eval_steps_per_second": 0.767,
+      "step": 336000
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 2.424618111440773e-05,
+      "loss": 0.7987,
+      "step": 336200
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 2.423680314349585e-05,
+      "loss": 0.7978,
+      "step": 336400
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 2.4227425172583972e-05,
+      "loss": 0.7931,
+      "step": 336600
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 2.4218047201672093e-05,
+      "loss": 0.7961,
+      "step": 336800
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 2.4208716120614776e-05,
+      "loss": 0.7934,
+      "step": 337000
+    },
+    {
+      "epoch": 0.4,
+      "eval_em_first_line_ratio": 0.4847,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.372,
+      "eval_loss": 0.7626174688339233,
+      "eval_runtime": 210.4803,
+      "eval_samples_per_second": 47.51,
+      "eval_steps_per_second": 0.746,
+      "step": 337000
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 2.4199338149702897e-05,
+      "loss": 0.7957,
+      "step": 337200
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 2.4189960178791018e-05,
+      "loss": 0.8013,
+      "step": 337400
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 2.418058220787914e-05,
+      "loss": 0.7946,
+      "step": 337600
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 2.417120423696726e-05,
+      "loss": 0.796,
+      "step": 337800
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 2.416182626605538e-05,
+      "loss": 0.7924,
+      "step": 338000
+    },
+    {
+      "epoch": 0.4,
+      "eval_em_first_line_ratio": 0.4847,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3717,
+      "eval_loss": 0.7622376680374146,
+      "eval_runtime": 212.2487,
+      "eval_samples_per_second": 47.115,
+      "eval_steps_per_second": 0.74,
+      "step": 338000
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 2.41524482951435e-05,
+      "loss": 0.7916,
+      "step": 338200
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 2.4143070324231626e-05,
+      "loss": 0.7983,
+      "step": 338400
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 2.4133692353319743e-05,
+      "loss": 0.7971,
+      "step": 338600
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 2.4124314382407868e-05,
+      "loss": 0.7956,
+      "step": 338800
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 2.4114936411495985e-05,
+      "loss": 0.7939,
+      "step": 339000
+    },
+    {
+      "epoch": 0.4,
+      "eval_em_first_line_ratio": 0.4854,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3726,
+      "eval_loss": 0.7619077563285828,
+      "eval_runtime": 205.9776,
+      "eval_samples_per_second": 48.549,
+      "eval_steps_per_second": 0.762,
+      "step": 339000
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 2.410555844058411e-05,
+      "loss": 0.7952,
+      "step": 339200
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 2.409618046967223e-05,
+      "loss": 0.7929,
+      "step": 339400
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 2.4086802498760352e-05,
+      "loss": 0.7954,
+      "step": 339600
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 2.4077424527848473e-05,
+      "loss": 0.7975,
+      "step": 339800
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 2.4068046556936594e-05,
+      "loss": 0.7988,
+      "step": 340000
+    },
+    {
+      "epoch": 0.4,
+      "eval_em_first_line_ratio": 0.4865,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3732,
+      "eval_loss": 0.7618625164031982,
+      "eval_runtime": 210.019,
+      "eval_samples_per_second": 47.615,
+      "eval_steps_per_second": 0.748,
+      "step": 340000
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 2.4058668586024715e-05,
+      "loss": 0.7987,
+      "step": 340200
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 2.4049290615112836e-05,
+      "loss": 0.7984,
+      "step": 340400
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 2.4039912644200957e-05,
+      "loss": 0.7897,
+      "step": 340600
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 2.403053467328908e-05,
+      "loss": 0.7946,
+      "step": 340800
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 2.40211567023772e-05,
+      "loss": 0.7907,
+      "step": 341000
+    },
+    {
+      "epoch": 0.4,
+      "eval_em_first_line_ratio": 0.4839,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3711,
+      "eval_loss": 0.7622260451316833,
+      "eval_runtime": 207.4549,
+      "eval_samples_per_second": 48.203,
+      "eval_steps_per_second": 0.757,
+      "step": 341000
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 2.4011778731465323e-05,
+      "loss": 0.795,
+      "step": 341200
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 2.400240076055344e-05,
+      "loss": 0.7972,
+      "step": 341400
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 2.3993022789641565e-05,
+      "loss": 0.7924,
+      "step": 341600
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 2.3983644818729682e-05,
+      "loss": 0.7951,
+      "step": 341800
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 2.3974266847817807e-05,
+      "loss": 0.7959,
+      "step": 342000
+    },
+    {
+      "epoch": 0.4,
+      "eval_em_first_line_ratio": 0.4851,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3723,
+      "eval_loss": 0.7615914940834045,
+      "eval_runtime": 202.5882,
+      "eval_samples_per_second": 49.361,
+      "eval_steps_per_second": 0.775,
+      "step": 342000
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 2.396488887690593e-05,
+      "loss": 0.7949,
+      "step": 342200
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 2.395551090599405e-05,
+      "loss": 0.7941,
+      "step": 342400
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 2.3946132935082173e-05,
+      "loss": 0.798,
+      "step": 342600
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 2.393675496417029e-05,
+      "loss": 0.7946,
+      "step": 342800
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 2.392737699325841e-05,
+      "loss": 0.7946,
+      "step": 343000
+    },
+    {
+      "epoch": 0.4,
+      "eval_em_first_line_ratio": 0.4862,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3733,
+      "eval_loss": 0.7616750001907349,
+      "eval_runtime": 206.8802,
+      "eval_samples_per_second": 48.337,
+      "eval_steps_per_second": 0.759,
+      "step": 343000
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 2.3917999022346536e-05,
+      "loss": 0.8037,
+      "step": 343200
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 2.3908621051434654e-05,
+      "loss": 0.7931,
+      "step": 343400
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 2.3899243080522778e-05,
+      "loss": 0.795,
+      "step": 343600
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 2.3889865109610896e-05,
+      "loss": 0.7888,
+      "step": 343800
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 2.388048713869902e-05,
+      "loss": 0.7912,
+      "step": 344000
+    },
+    {
+      "epoch": 0.4,
+      "eval_em_first_line_ratio": 0.4852,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3723,
+      "eval_loss": 0.7616862058639526,
+      "eval_runtime": 206.4405,
+      "eval_samples_per_second": 48.44,
+      "eval_steps_per_second": 0.761,
+      "step": 344000
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 2.3871156057641702e-05,
+      "loss": 0.7956,
+      "step": 344200
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 2.386177808672982e-05,
+      "loss": 0.7923,
+      "step": 344400
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 2.3852400115817944e-05,
+      "loss": 0.7962,
+      "step": 344600
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 2.3843022144906062e-05,
+      "loss": 0.7951,
+      "step": 344800
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 2.3833644173994186e-05,
+      "loss": 0.788,
+      "step": 345000
+    },
+    {
+      "epoch": 0.4,
+      "eval_em_first_line_ratio": 0.4849,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.372,
+      "eval_loss": 0.7611204385757446,
+      "eval_runtime": 205.475,
+      "eval_samples_per_second": 48.668,
+      "eval_steps_per_second": 0.764,
+      "step": 345000
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 2.3824266203082304e-05,
+      "loss": 0.7932,
+      "step": 345200
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 2.3814888232170428e-05,
+      "loss": 0.7928,
+      "step": 345400
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 2.3805510261258553e-05,
+      "loss": 0.7936,
+      "step": 345600
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 2.379613229034667e-05,
+      "loss": 0.7897,
+      "step": 345800
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 2.3786801209289353e-05,
+      "loss": 0.7964,
+      "step": 346000
+    },
+    {
+      "epoch": 0.41,
+      "eval_em_first_line_ratio": 0.4859,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3728,
+      "eval_loss": 0.7610469460487366,
+      "eval_runtime": 204.3448,
+      "eval_samples_per_second": 48.937,
+      "eval_steps_per_second": 0.768,
+      "step": 346000
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 2.377742323837747e-05,
+      "loss": 0.7931,
+      "step": 346200
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 2.3768045267465595e-05,
+      "loss": 0.7858,
+      "step": 346400
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 2.3758667296553716e-05,
+      "loss": 0.7958,
+      "step": 346600
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 2.3749289325641837e-05,
+      "loss": 0.7929,
+      "step": 346800
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 2.3739911354729958e-05,
+      "loss": 0.7888,
+      "step": 347000
+    },
+    {
+      "epoch": 0.41,
+      "eval_em_first_line_ratio": 0.4867,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3745,
+      "eval_loss": 0.7608511447906494,
+      "eval_runtime": 203.8707,
+      "eval_samples_per_second": 49.051,
+      "eval_steps_per_second": 0.77,
+      "step": 347000
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 2.373053338381808e-05,
+      "loss": 0.7907,
+      "step": 347200
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 2.37211554129062e-05,
+      "loss": 0.7923,
+      "step": 347400
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 2.371177744199432e-05,
+      "loss": 0.7933,
+      "step": 347600
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 2.370239947108244e-05,
+      "loss": 0.7969,
+      "step": 347800
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 2.3693021500170566e-05,
+      "loss": 0.7915,
+      "step": 348000
+    },
+    {
+      "epoch": 0.41,
+      "eval_em_first_line_ratio": 0.4848,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3716,
+      "eval_loss": 0.7602121233940125,
+      "eval_runtime": 201.7239,
+      "eval_samples_per_second": 49.573,
+      "eval_steps_per_second": 0.778,
+      "step": 348000
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 2.3683643529258683e-05,
+      "loss": 0.7949,
+      "step": 348200
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 2.3674265558346808e-05,
+      "loss": 0.7947,
+      "step": 348400
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 2.3664887587434925e-05,
+      "loss": 0.788,
+      "step": 348600
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 2.365550961652305e-05,
+      "loss": 0.7914,
+      "step": 348800
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 2.364613164561117e-05,
+      "loss": 0.7921,
+      "step": 349000
+    },
+    {
+      "epoch": 0.41,
+      "eval_em_first_line_ratio": 0.4843,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3712,
+      "eval_loss": 0.7608197331428528,
+      "eval_runtime": 205.4202,
+      "eval_samples_per_second": 48.681,
+      "eval_steps_per_second": 0.764,
+      "step": 349000
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 2.363675367469929e-05,
+      "loss": 0.7977,
+      "step": 349200
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 2.3627375703787413e-05,
+      "loss": 0.787,
+      "step": 349400
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 2.3617997732875534e-05,
+      "loss": 0.7881,
+      "step": 349600
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 2.3608619761963655e-05,
+      "loss": 0.789,
+      "step": 349800
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 2.3599241791051776e-05,
+      "loss": 0.7943,
+      "step": 350000
+    },
+    {
+      "epoch": 0.41,
+      "eval_em_first_line_ratio": 0.4869,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3736,
+      "eval_loss": 0.7606350779533386,
+      "eval_runtime": 215.4149,
+      "eval_samples_per_second": 46.422,
+      "eval_steps_per_second": 0.729,
+      "step": 350000
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 2.3589863820139897e-05,
+      "loss": 0.7913,
+      "step": 350200
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 2.358048584922802e-05,
+      "loss": 0.798,
+      "step": 350400
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 2.357110787831614e-05,
+      "loss": 0.788,
+      "step": 350600
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 2.356177679725882e-05,
+      "loss": 0.7948,
+      "step": 350800
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 2.3552398826346942e-05,
+      "loss": 0.791,
+      "step": 351000
+    },
+    {
+      "epoch": 0.41,
+      "eval_em_first_line_ratio": 0.4854,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3721,
+      "eval_loss": 0.7601962685585022,
+      "eval_runtime": 209.185,
+      "eval_samples_per_second": 47.805,
+      "eval_steps_per_second": 0.751,
+      "step": 351000
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 2.3543020855435063e-05,
+      "loss": 0.7947,
+      "step": 351200
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 2.3533642884523187e-05,
+      "loss": 0.7904,
+      "step": 351400
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 2.3524311803465863e-05,
+      "loss": 0.7938,
+      "step": 351600
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 2.3514933832553987e-05,
+      "loss": 0.7869,
+      "step": 351800
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 2.3505555861642105e-05,
+      "loss": 0.7943,
+      "step": 352000
+    },
+    {
+      "epoch": 0.41,
+      "eval_em_first_line_ratio": 0.4858,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3724,
+      "eval_loss": 0.7608544230461121,
+      "eval_runtime": 200.2672,
+      "eval_samples_per_second": 49.933,
+      "eval_steps_per_second": 0.784,
+      "step": 352000
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 2.349617789073023e-05,
+      "loss": 0.7912,
+      "step": 352200
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 2.3486799919818354e-05,
+      "loss": 0.7941,
+      "step": 352400
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 2.347742194890647e-05,
+      "loss": 0.7899,
+      "step": 352600
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 2.3468043977994596e-05,
+      "loss": 0.7915,
+      "step": 352800
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 2.3458666007082713e-05,
+      "loss": 0.7875,
+      "step": 353000
+    },
+    {
+      "epoch": 0.41,
+      "eval_em_first_line_ratio": 0.4842,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3715,
+      "eval_loss": 0.7600659132003784,
+      "eval_runtime": 200.0709,
+      "eval_samples_per_second": 49.982,
+      "eval_steps_per_second": 0.785,
+      "step": 353000
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 2.3449288036170838e-05,
+      "loss": 0.794,
+      "step": 353200
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 2.3439910065258955e-05,
+      "loss": 0.7937,
+      "step": 353400
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 2.343053209434708e-05,
+      "loss": 0.7978,
+      "step": 353600
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 2.34211541234352e-05,
+      "loss": 0.7911,
+      "step": 353800
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 2.341177615252332e-05,
+      "loss": 0.7916,
+      "step": 354000
+    },
+    {
+      "epoch": 0.41,
+      "eval_em_first_line_ratio": 0.4855,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3724,
+      "eval_loss": 0.760942280292511,
+      "eval_runtime": 213.5243,
+      "eval_samples_per_second": 46.833,
+      "eval_steps_per_second": 0.735,
+      "step": 354000
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 2.3402398181611442e-05,
+      "loss": 0.7938,
+      "step": 354200
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 2.3393020210699563e-05,
+      "loss": 0.7896,
+      "step": 354400
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 2.3383642239787684e-05,
+      "loss": 0.7884,
+      "step": 354600
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 2.337426426887581e-05,
+      "loss": 0.7858,
+      "step": 354800
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 2.3364933187818484e-05,
+      "loss": 0.7951,
+      "step": 355000
+    },
+    {
+      "epoch": 0.42,
+      "eval_em_first_line_ratio": 0.4853,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3728,
+      "eval_loss": 0.7593968510627747,
+      "eval_runtime": 209.9683,
+      "eval_samples_per_second": 47.626,
+      "eval_steps_per_second": 0.748,
+      "step": 355000
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 2.335555521690661e-05,
+      "loss": 0.7918,
+      "step": 355200
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 2.3346177245994726e-05,
+      "loss": 0.7933,
+      "step": 355400
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 2.333679927508285e-05,
+      "loss": 0.7915,
+      "step": 355600
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 2.332742130417097e-05,
+      "loss": 0.7898,
+      "step": 355800
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 2.3318043333259093e-05,
+      "loss": 0.7913,
+      "step": 356000
+    },
+    {
+      "epoch": 0.42,
+      "eval_em_first_line_ratio": 0.4829,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3703,
+      "eval_loss": 0.7599010467529297,
+      "eval_runtime": 209.7138,
+      "eval_samples_per_second": 47.684,
+      "eval_steps_per_second": 0.749,
+      "step": 356000
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 2.3308665362347217e-05,
+      "loss": 0.794,
+      "step": 356200
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 2.3299287391435335e-05,
+      "loss": 0.7891,
+      "step": 356400
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 2.328990942052346e-05,
+      "loss": 0.7858,
+      "step": 356600
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 2.3280531449611577e-05,
+      "loss": 0.7858,
+      "step": 356800
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 2.32711534786997e-05,
+      "loss": 0.7937,
+      "step": 357000
+    },
+    {
+      "epoch": 0.42,
+      "eval_em_first_line_ratio": 0.4822,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3693,
+      "eval_loss": 0.7591649889945984,
+      "eval_runtime": 208.6546,
+      "eval_samples_per_second": 47.926,
+      "eval_steps_per_second": 0.752,
+      "step": 357000
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 2.3261775507787822e-05,
+      "loss": 0.7927,
+      "step": 357200
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 2.3252397536875943e-05,
+      "loss": 0.7869,
+      "step": 357400
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 2.3243019565964064e-05,
+      "loss": 0.7865,
+      "step": 357600
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 2.3233641595052185e-05,
+      "loss": 0.7963,
+      "step": 357800
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 2.3224263624140306e-05,
+      "loss": 0.7905,
+      "step": 358000
+    },
+    {
+      "epoch": 0.42,
+      "eval_em_first_line_ratio": 0.4846,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3721,
+      "eval_loss": 0.7591702342033386,
+      "eval_runtime": 211.0933,
+      "eval_samples_per_second": 47.372,
+      "eval_steps_per_second": 0.744,
+      "step": 358000
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 2.3214885653228423e-05,
+      "loss": 0.79,
+      "step": 358200
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 2.3205507682316548e-05,
+      "loss": 0.7874,
+      "step": 358400
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 2.3196129711404672e-05,
+      "loss": 0.7893,
+      "step": 358600
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 2.3186798630347348e-05,
+      "loss": 0.7926,
+      "step": 358800
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 2.3177420659435472e-05,
+      "loss": 0.7874,
+      "step": 359000
+    },
+    {
+      "epoch": 0.42,
+      "eval_em_first_line_ratio": 0.4856,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3726,
+      "eval_loss": 0.7591784000396729,
+      "eval_runtime": 204.0108,
+      "eval_samples_per_second": 49.017,
+      "eval_steps_per_second": 0.77,
+      "step": 359000
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 2.316804268852359e-05,
+      "loss": 0.7918,
+      "step": 359200
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 2.3158664717611714e-05,
+      "loss": 0.7864,
+      "step": 359400
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 2.314928674669984e-05,
+      "loss": 0.7898,
+      "step": 359600
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 2.3139908775787956e-05,
+      "loss": 0.7957,
+      "step": 359800
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 2.313053080487608e-05,
+      "loss": 0.7887,
+      "step": 360000
+    },
+    {
+      "epoch": 0.42,
+      "eval_em_first_line_ratio": 0.4833,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3709,
+      "eval_loss": 0.7588796615600586,
+      "eval_runtime": 206.9032,
+      "eval_samples_per_second": 48.332,
+      "eval_steps_per_second": 0.759,
+      "step": 360000
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 2.3121152833964198e-05,
+      "loss": 0.7911,
+      "step": 360200
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 2.3111774863052322e-05,
+      "loss": 0.7892,
+      "step": 360400
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 2.310239689214044e-05,
+      "loss": 0.7913,
+      "step": 360600
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 2.3093018921228564e-05,
+      "loss": 0.7896,
+      "step": 360800
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 2.3083640950316685e-05,
+      "loss": 0.7944,
+      "step": 361000
+    },
+    {
+      "epoch": 0.42,
+      "eval_em_first_line_ratio": 0.4857,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3728,
+      "eval_loss": 0.7586925625801086,
+      "eval_runtime": 213.2101,
+      "eval_samples_per_second": 46.902,
+      "eval_steps_per_second": 0.736,
+      "step": 361000
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 2.3074262979404803e-05,
+      "loss": 0.7902,
+      "step": 361200
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 2.3064885008492927e-05,
+      "loss": 0.7878,
+      "step": 361400
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 2.3055507037581045e-05,
+      "loss": 0.7929,
+      "step": 361600
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 2.304612906666917e-05,
+      "loss": 0.7935,
+      "step": 361800
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 2.3036751095757294e-05,
+      "loss": 0.79,
+      "step": 362000
+    },
+    {
+      "epoch": 0.42,
+      "eval_em_first_line_ratio": 0.4837,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3709,
+      "eval_loss": 0.7583368420600891,
+      "eval_runtime": 217.1425,
+      "eval_samples_per_second": 46.053,
+      "eval_steps_per_second": 0.723,
+      "step": 362000
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 2.302737312484541e-05,
+      "loss": 0.7909,
+      "step": 362200
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 2.3017995153933536e-05,
+      "loss": 0.7926,
+      "step": 362400
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 2.3008617183021653e-05,
+      "loss": 0.7875,
+      "step": 362600
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 2.2999239212109777e-05,
+      "loss": 0.7902,
+      "step": 362800
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 2.2989861241197895e-05,
+      "loss": 0.7941,
+      "step": 363000
+    },
+    {
+      "epoch": 0.43,
+      "eval_em_first_line_ratio": 0.4829,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3701,
+      "eval_loss": 0.7580798864364624,
+      "eval_runtime": 216.2773,
+      "eval_samples_per_second": 46.237,
+      "eval_steps_per_second": 0.726,
+      "step": 363000
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 2.298048327028602e-05,
+      "loss": 0.7888,
+      "step": 363200
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 2.297110529937414e-05,
+      "loss": 0.7837,
+      "step": 363400
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 2.296172732846226e-05,
+      "loss": 0.7908,
+      "step": 363600
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 2.2952349357550382e-05,
+      "loss": 0.785,
+      "step": 363800
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 2.2942971386638503e-05,
+      "loss": 0.7902,
+      "step": 364000
+    },
+    {
+      "epoch": 0.43,
+      "eval_em_first_line_ratio": 0.4826,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3698,
+      "eval_loss": 0.7581704258918762,
+      "eval_runtime": 213.4541,
+      "eval_samples_per_second": 46.848,
+      "eval_steps_per_second": 0.736,
+      "step": 364000
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 2.2933593415726624e-05,
+      "loss": 0.7849,
+      "step": 364200
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 2.292421544481475e-05,
+      "loss": 0.7925,
+      "step": 364400
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 2.2914837473902866e-05,
+      "loss": 0.7942,
+      "step": 364600
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 2.2905553282700107e-05,
+      "loss": 0.7902,
+      "step": 364800
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 2.2896175311788228e-05,
+      "loss": 0.7868,
+      "step": 365000
+    },
+    {
+      "epoch": 0.43,
+      "eval_em_first_line_ratio": 0.486,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3736,
+      "eval_loss": 0.7583454251289368,
+      "eval_runtime": 207.9276,
+      "eval_samples_per_second": 48.094,
+      "eval_steps_per_second": 0.755,
+      "step": 365000
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 2.288684423073091e-05,
+      "loss": 0.7956,
+      "step": 365200
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 2.287746625981903e-05,
+      "loss": 0.7927,
+      "step": 365400
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 2.2868088288907152e-05,
+      "loss": 0.7851,
+      "step": 365600
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 2.2858710317995273e-05,
+      "loss": 0.7861,
+      "step": 365800
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 2.2849332347083394e-05,
+      "loss": 0.7943,
+      "step": 366000
+    },
+    {
+      "epoch": 0.43,
+      "eval_em_first_line_ratio": 0.4846,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3724,
+      "eval_loss": 0.7578973770141602,
+      "eval_runtime": 206.2124,
+      "eval_samples_per_second": 48.494,
+      "eval_steps_per_second": 0.761,
+      "step": 366000
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 2.2839954376171515e-05,
+      "loss": 0.7848,
+      "step": 366200
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 2.283057640525964e-05,
+      "loss": 0.7941,
+      "step": 366400
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 2.2821198434347757e-05,
+      "loss": 0.7888,
+      "step": 366600
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 2.281182046343588e-05,
+      "loss": 0.7894,
+      "step": 366800
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 2.2802442492524e-05,
+      "loss": 0.7895,
+      "step": 367000
+    },
+    {
+      "epoch": 0.43,
+      "eval_em_first_line_ratio": 0.4849,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3726,
+      "eval_loss": 0.7581456899642944,
+      "eval_runtime": 204.6745,
+      "eval_samples_per_second": 48.858,
+      "eval_steps_per_second": 0.767,
+      "step": 367000
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 2.2793064521612123e-05,
+      "loss": 0.7894,
+      "step": 367200
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 2.278368655070024e-05,
+      "loss": 0.7881,
+      "step": 367400
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 2.2774308579788365e-05,
+      "loss": 0.7871,
+      "step": 367600
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 2.2764977498731048e-05,
+      "loss": 0.7897,
+      "step": 367800
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 2.2755599527819166e-05,
+      "loss": 0.7886,
+      "step": 368000
+    },
+    {
+      "epoch": 0.43,
+      "eval_em_first_line_ratio": 0.4843,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3718,
+      "eval_loss": 0.7575170397758484,
+      "eval_runtime": 201.3663,
+      "eval_samples_per_second": 49.661,
+      "eval_steps_per_second": 0.78,
+      "step": 368000
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 2.274622155690729e-05,
+      "loss": 0.7851,
+      "step": 368200
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 2.2736843585995407e-05,
+      "loss": 0.7915,
+      "step": 368400
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 2.2727465615083532e-05,
+      "loss": 0.7878,
+      "step": 368600
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 2.2718087644171653e-05,
+      "loss": 0.7884,
+      "step": 368800
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 2.2708709673259774e-05,
+      "loss": 0.7798,
+      "step": 369000
+    },
+    {
+      "epoch": 0.43,
+      "eval_em_first_line_ratio": 0.4842,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3718,
+      "eval_loss": 0.7576271295547485,
+      "eval_runtime": 215.7219,
+      "eval_samples_per_second": 46.356,
+      "eval_steps_per_second": 0.728,
+      "step": 369000
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 2.2699331702347895e-05,
+      "loss": 0.7921,
+      "step": 369200
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 2.2689953731436016e-05,
+      "loss": 0.7913,
+      "step": 369400
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 2.2680575760524137e-05,
+      "loss": 0.7931,
+      "step": 369600
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 2.2671197789612254e-05,
+      "loss": 0.793,
+      "step": 369800
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 2.266181981870038e-05,
+      "loss": 0.7887,
+      "step": 370000
+    },
+    {
+      "epoch": 0.43,
+      "eval_em_first_line_ratio": 0.4863,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3732,
+      "eval_loss": 0.7572408318519592,
+      "eval_runtime": 209.3039,
+      "eval_samples_per_second": 47.777,
+      "eval_steps_per_second": 0.75,
+      "step": 370000
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 2.2652441847788503e-05,
+      "loss": 0.779,
+      "step": 370200
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 2.264306387687662e-05,
+      "loss": 0.7911,
+      "step": 370400
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 2.2633685905964745e-05,
+      "loss": 0.7822,
+      "step": 370600
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 2.2624307935052863e-05,
+      "loss": 0.7788,
+      "step": 370800
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 2.2614929964140987e-05,
+      "loss": 0.7847,
+      "step": 371000
+    },
+    {
+      "epoch": 0.43,
+      "eval_em_first_line_ratio": 0.4833,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3707,
+      "eval_loss": 0.7575092911720276,
+      "eval_runtime": 208.0003,
+      "eval_samples_per_second": 48.077,
+      "eval_steps_per_second": 0.755,
+      "step": 371000
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 2.2605551993229108e-05,
+      "loss": 0.7886,
+      "step": 371200
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 2.2596220912171787e-05,
+      "loss": 0.7841,
+      "step": 371400
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 2.258684294125991e-05,
+      "loss": 0.7898,
+      "step": 371600
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 2.257746497034803e-05,
+      "loss": 0.7859,
+      "step": 371800
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 2.2568086999436153e-05,
+      "loss": 0.7943,
+      "step": 372000
+    },
+    {
+      "epoch": 0.44,
+      "eval_em_first_line_ratio": 0.4847,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3715,
+      "eval_loss": 0.7574893832206726,
+      "eval_runtime": 205.3613,
+      "eval_samples_per_second": 48.695,
+      "eval_steps_per_second": 0.765,
+      "step": 372000
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 2.255870902852427e-05,
+      "loss": 0.7873,
+      "step": 372200
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 2.2549331057612395e-05,
+      "loss": 0.7896,
+      "step": 372400
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 2.2539953086700516e-05,
+      "loss": 0.7904,
+      "step": 372600
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 2.2530575115788634e-05,
+      "loss": 0.7888,
+      "step": 372800
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 2.2521197144876758e-05,
+      "loss": 0.7894,
+      "step": 373000
+    },
+    {
+      "epoch": 0.44,
+      "eval_em_first_line_ratio": 0.4852,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3721,
+      "eval_loss": 0.7570964694023132,
+      "eval_runtime": 209.8697,
+      "eval_samples_per_second": 47.649,
+      "eval_steps_per_second": 0.748,
+      "step": 373000
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 2.2511819173964876e-05,
+      "loss": 0.7885,
+      "step": 373200
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 2.2502441203053e-05,
+      "loss": 0.7903,
+      "step": 373400
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 2.2493063232141124e-05,
+      "loss": 0.7861,
+      "step": 373600
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 2.2483685261229242e-05,
+      "loss": 0.7887,
+      "step": 373800
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 2.2474307290317366e-05,
+      "loss": 0.7913,
+      "step": 374000
+    },
+    {
+      "epoch": 0.44,
+      "eval_em_first_line_ratio": 0.4842,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3717,
+      "eval_loss": 0.7572146058082581,
+      "eval_runtime": 210.6972,
+      "eval_samples_per_second": 47.461,
+      "eval_steps_per_second": 0.745,
+      "step": 374000
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 2.2464929319405484e-05,
+      "loss": 0.7915,
+      "step": 374200
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 2.245555134849361e-05,
+      "loss": 0.7906,
+      "step": 374400
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 2.2446173377581726e-05,
+      "loss": 0.7864,
+      "step": 374600
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 2.243679540666985e-05,
+      "loss": 0.7847,
+      "step": 374800
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 2.242741743575797e-05,
+      "loss": 0.7877,
+      "step": 375000
+    },
+    {
+      "epoch": 0.44,
+      "eval_em_first_line_ratio": 0.4848,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3722,
+      "eval_loss": 0.7574293613433838,
+      "eval_runtime": 206.6971,
+      "eval_samples_per_second": 48.38,
+      "eval_steps_per_second": 0.76,
+      "step": 375000
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 2.2418039464846092e-05,
+      "loss": 0.7877,
+      "step": 375200
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 2.2408661493934213e-05,
+      "loss": 0.7861,
+      "step": 375400
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 2.2399283523022334e-05,
+      "loss": 0.7892,
+      "step": 375600
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 2.2389905552110455e-05,
+      "loss": 0.7924,
+      "step": 375800
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 2.238052758119858e-05,
+      "loss": 0.7871,
+      "step": 376000
+    },
+    {
+      "epoch": 0.44,
+      "eval_em_first_line_ratio": 0.4841,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3715,
+      "eval_loss": 0.7561378479003906,
+      "eval_runtime": 207.2,
+      "eval_samples_per_second": 48.263,
+      "eval_steps_per_second": 0.758,
+      "step": 376000
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 2.2371149610286697e-05,
+      "loss": 0.781,
+      "step": 376200
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 2.236177163937482e-05,
+      "loss": 0.7857,
+      "step": 376400
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 2.235239366846294e-05,
+      "loss": 0.7877,
+      "step": 376600
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 2.2343015697551063e-05,
+      "loss": 0.7869,
+      "step": 376800
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 2.233363772663918e-05,
+      "loss": 0.791,
+      "step": 377000
+    },
+    {
+      "epoch": 0.44,
+      "eval_em_first_line_ratio": 0.4841,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3717,
+      "eval_loss": 0.7565041184425354,
+      "eval_runtime": 206.0233,
+      "eval_samples_per_second": 48.538,
+      "eval_steps_per_second": 0.762,
+      "step": 377000
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 2.2324259755727305e-05,
+      "loss": 0.7886,
+      "step": 377200
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 2.2314881784815426e-05,
+      "loss": 0.7918,
+      "step": 377400
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 2.2305503813903547e-05,
+      "loss": 0.7892,
+      "step": 377600
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 2.2296125842991668e-05,
+      "loss": 0.7825,
+      "step": 377800
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 2.2286794761934347e-05,
+      "loss": 0.7914,
+      "step": 378000
+    },
+    {
+      "epoch": 0.44,
+      "eval_em_first_line_ratio": 0.4831,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3707,
+      "eval_loss": 0.7560235261917114,
+      "eval_runtime": 203.7343,
+      "eval_samples_per_second": 49.084,
+      "eval_steps_per_second": 0.771,
+      "step": 378000
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 2.2277416791022472e-05,
+      "loss": 0.7841,
+      "step": 378200
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 2.2268038820110593e-05,
+      "loss": 0.7939,
+      "step": 378400
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 2.2258660849198714e-05,
+      "loss": 0.7874,
+      "step": 378600
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 2.2249282878286835e-05,
+      "loss": 0.7889,
+      "step": 378800
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 2.2239904907374956e-05,
+      "loss": 0.7895,
+      "step": 379000
+    },
+    {
+      "epoch": 0.44,
+      "eval_em_first_line_ratio": 0.4836,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3713,
+      "eval_loss": 0.7565145492553711,
+      "eval_runtime": 201.9267,
+      "eval_samples_per_second": 49.523,
+      "eval_steps_per_second": 0.778,
+      "step": 379000
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 2.2230526936463077e-05,
+      "loss": 0.7857,
+      "step": 379200
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 2.222119585540576e-05,
+      "loss": 0.7867,
+      "step": 379400
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 2.2211817884493877e-05,
+      "loss": 0.7879,
+      "step": 379600
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 2.2202439913582e-05,
+      "loss": 0.7863,
+      "step": 379800
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 2.219306194267012e-05,
+      "loss": 0.7847,
+      "step": 380000
+    },
+    {
+      "epoch": 0.45,
+      "eval_em_first_line_ratio": 0.4863,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3734,
+      "eval_loss": 0.755949079990387,
+      "eval_runtime": 204.3565,
+      "eval_samples_per_second": 48.934,
+      "eval_steps_per_second": 0.768,
+      "step": 380000
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 2.2183683971758243e-05,
+      "loss": 0.7865,
+      "step": 380200
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 2.217430600084636e-05,
+      "loss": 0.7848,
+      "step": 380400
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 2.2164928029934485e-05,
+      "loss": 0.7824,
+      "step": 380600
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 2.215555005902261e-05,
+      "loss": 0.7871,
+      "step": 380800
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 2.2146172088110727e-05,
+      "loss": 0.7882,
+      "step": 381000
+    },
+    {
+      "epoch": 0.45,
+      "eval_em_first_line_ratio": 0.4862,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.373,
+      "eval_loss": 0.7556912302970886,
+      "eval_runtime": 201.73,
+      "eval_samples_per_second": 49.571,
+      "eval_steps_per_second": 0.778,
+      "step": 381000
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 2.213679411719885e-05,
+      "loss": 0.7883,
+      "step": 381200
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 2.212741614628697e-05,
+      "loss": 0.7943,
+      "step": 381400
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 2.2118038175375093e-05,
+      "loss": 0.7851,
+      "step": 381600
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 2.2108660204463214e-05,
+      "loss": 0.7838,
+      "step": 381800
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 2.2099282233551335e-05,
+      "loss": 0.7854,
+      "step": 382000
+    },
+    {
+      "epoch": 0.45,
+      "eval_em_first_line_ratio": 0.4861,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3731,
+      "eval_loss": 0.7553886771202087,
+      "eval_runtime": 202.5558,
+      "eval_samples_per_second": 49.369,
+      "eval_steps_per_second": 0.775,
+      "step": 382000
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 2.2089904262639456e-05,
+      "loss": 0.7839,
+      "step": 382200
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 2.2080526291727574e-05,
+      "loss": 0.7845,
+      "step": 382400
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 2.2071148320815698e-05,
+      "loss": 0.788,
+      "step": 382600
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 2.2061770349903816e-05,
+      "loss": 0.7933,
+      "step": 382800
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 2.205239237899194e-05,
+      "loss": 0.7886,
+      "step": 383000
+    },
+    {
+      "epoch": 0.45,
+      "eval_em_first_line_ratio": 0.4852,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3724,
+      "eval_loss": 0.7553954124450684,
+      "eval_runtime": 200.4532,
+      "eval_samples_per_second": 49.887,
+      "eval_steps_per_second": 0.783,
+      "step": 383000
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 2.2043014408080064e-05,
+      "loss": 0.7921,
+      "step": 383200
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 2.2033636437168182e-05,
+      "loss": 0.7846,
+      "step": 383400
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 2.2024258466256306e-05,
+      "loss": 0.7874,
+      "step": 383600
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 2.2014880495344424e-05,
+      "loss": 0.7846,
+      "step": 383800
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 2.2005549414287106e-05,
+      "loss": 0.7886,
+      "step": 384000
+    },
+    {
+      "epoch": 0.45,
+      "eval_em_first_line_ratio": 0.4856,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3726,
+      "eval_loss": 0.7555776834487915,
+      "eval_runtime": 201.6808,
+      "eval_samples_per_second": 49.583,
+      "eval_steps_per_second": 0.778,
+      "step": 384000
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 2.199617144337523e-05,
+      "loss": 0.7871,
+      "step": 384200
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 2.198679347246335e-05,
+      "loss": 0.7837,
+      "step": 384400
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 2.1977415501551473e-05,
+      "loss": 0.7844,
+      "step": 384600
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 2.196803753063959e-05,
+      "loss": 0.7887,
+      "step": 384800
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 2.1958659559727715e-05,
+      "loss": 0.7836,
+      "step": 385000
+    },
+    {
+      "epoch": 0.45,
+      "eval_em_first_line_ratio": 0.4868,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3738,
+      "eval_loss": 0.7557660341262817,
+      "eval_runtime": 211.8828,
+      "eval_samples_per_second": 47.196,
+      "eval_steps_per_second": 0.741,
+      "step": 385000
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 2.1949281588815832e-05,
+      "loss": 0.7888,
+      "step": 385200
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 2.1939903617903953e-05,
+      "loss": 0.783,
+      "step": 385400
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 2.1930525646992078e-05,
+      "loss": 0.7855,
+      "step": 385600
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 2.1921194565934757e-05,
+      "loss": 0.7774,
+      "step": 385800
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 2.1911816595022878e-05,
+      "loss": 0.7878,
+      "step": 386000
+    },
+    {
+      "epoch": 0.45,
+      "eval_em_first_line_ratio": 0.4859,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3731,
+      "eval_loss": 0.7553533911705017,
+      "eval_runtime": 208.4299,
+      "eval_samples_per_second": 47.978,
+      "eval_steps_per_second": 0.753,
+      "step": 386000
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 2.1902438624111e-05,
+      "loss": 0.7878,
+      "step": 386200
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 2.189306065319912e-05,
+      "loss": 0.7852,
+      "step": 386400
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 2.1883682682287244e-05,
+      "loss": 0.7855,
+      "step": 386600
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 2.187430471137536e-05,
+      "loss": 0.7792,
+      "step": 386800
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 2.1864926740463486e-05,
+      "loss": 0.7856,
+      "step": 387000
+    },
+    {
+      "epoch": 0.45,
+      "eval_em_first_line_ratio": 0.4824,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.37,
+      "eval_loss": 0.7550241351127625,
+      "eval_runtime": 212.3771,
+      "eval_samples_per_second": 47.086,
+      "eval_steps_per_second": 0.739,
+      "step": 387000
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 2.1855548769551604e-05,
+      "loss": 0.7864,
+      "step": 387200
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 2.1846170798639728e-05,
+      "loss": 0.7847,
+      "step": 387400
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 2.1836792827727845e-05,
+      "loss": 0.7867,
+      "step": 387600
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 2.182741485681597e-05,
+      "loss": 0.7838,
+      "step": 387800
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 2.1818036885904094e-05,
+      "loss": 0.7874,
+      "step": 388000
+    },
+    {
+      "epoch": 0.45,
+      "eval_em_first_line_ratio": 0.4851,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3728,
+      "eval_loss": 0.7546792030334473,
+      "eval_runtime": 206.7248,
+      "eval_samples_per_second": 48.374,
+      "eval_steps_per_second": 0.759,
+      "step": 388000
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.1808658914992212e-05,
+      "loss": 0.7859,
+      "step": 388200
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.1799280944080333e-05,
+      "loss": 0.7878,
+      "step": 388400
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.1789902973168454e-05,
+      "loss": 0.7879,
+      "step": 388600
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.1780525002256575e-05,
+      "loss": 0.7836,
+      "step": 388800
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.17711470313447e-05,
+      "loss": 0.7904,
+      "step": 389000
+    },
+    {
+      "epoch": 0.46,
+      "eval_em_first_line_ratio": 0.4864,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3733,
+      "eval_loss": 0.7545261979103088,
+      "eval_runtime": 210.1872,
+      "eval_samples_per_second": 47.577,
+      "eval_steps_per_second": 0.747,
+      "step": 389000
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.1761769060432817e-05,
+      "loss": 0.7816,
+      "step": 389200
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.175239108952094e-05,
+      "loss": 0.7837,
+      "step": 389400
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.174301311860906e-05,
+      "loss": 0.7888,
+      "step": 389600
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.1733635147697183e-05,
+      "loss": 0.7788,
+      "step": 389800
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.1724304066639865e-05,
+      "loss": 0.7831,
+      "step": 390000
+    },
+    {
+      "epoch": 0.46,
+      "eval_em_first_line_ratio": 0.4859,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3734,
+      "eval_loss": 0.7546401619911194,
+      "eval_runtime": 208.9845,
+      "eval_samples_per_second": 47.85,
+      "eval_steps_per_second": 0.751,
+      "step": 390000
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.1714972985582545e-05,
+      "loss": 0.7822,
+      "step": 390200
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.1705595014670666e-05,
+      "loss": 0.7814,
+      "step": 390400
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.1696217043758787e-05,
+      "loss": 0.7889,
+      "step": 390600
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.1686839072846907e-05,
+      "loss": 0.7853,
+      "step": 390800
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.1677461101935025e-05,
+      "loss": 0.7862,
+      "step": 391000
+    },
+    {
+      "epoch": 0.46,
+      "eval_em_first_line_ratio": 0.4857,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3732,
+      "eval_loss": 0.7545140385627747,
+      "eval_runtime": 203.0508,
+      "eval_samples_per_second": 49.249,
+      "eval_steps_per_second": 0.773,
+      "step": 391000
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.166808313102315e-05,
+      "loss": 0.788,
+      "step": 391200
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.1658705160111274e-05,
+      "loss": 0.7843,
+      "step": 391400
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.164932718919939e-05,
+      "loss": 0.7834,
+      "step": 391600
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.1639949218287516e-05,
+      "loss": 0.7868,
+      "step": 391800
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.1630571247375633e-05,
+      "loss": 0.7808,
+      "step": 392000
+    },
+    {
+      "epoch": 0.46,
+      "eval_em_first_line_ratio": 0.4852,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3727,
+      "eval_loss": 0.7544535398483276,
+      "eval_runtime": 208.172,
+      "eval_samples_per_second": 48.037,
+      "eval_steps_per_second": 0.754,
+      "step": 392000
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.1621193276463758e-05,
+      "loss": 0.7898,
+      "step": 392200
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.161186219540644e-05,
+      "loss": 0.7854,
+      "step": 392400
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.1602484224494558e-05,
+      "loss": 0.7875,
+      "step": 392600
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.1593106253582682e-05,
+      "loss": 0.7792,
+      "step": 392800
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.15837282826708e-05,
+      "loss": 0.788,
+      "step": 393000
+    },
+    {
+      "epoch": 0.46,
+      "eval_em_first_line_ratio": 0.4847,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3723,
+      "eval_loss": 0.7543802261352539,
+      "eval_runtime": 203.701,
+      "eval_samples_per_second": 49.092,
+      "eval_steps_per_second": 0.771,
+      "step": 393000
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.1574350311758924e-05,
+      "loss": 0.7881,
+      "step": 393200
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.1564972340847045e-05,
+      "loss": 0.7877,
+      "step": 393400
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.1555594369935166e-05,
+      "loss": 0.7879,
+      "step": 393600
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.1546216399023287e-05,
+      "loss": 0.7841,
+      "step": 393800
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.1536838428111405e-05,
+      "loss": 0.7796,
+      "step": 394000
+    },
+    {
+      "epoch": 0.46,
+      "eval_em_first_line_ratio": 0.4845,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3718,
+      "eval_loss": 0.75446617603302,
+      "eval_runtime": 207.2631,
+      "eval_samples_per_second": 48.248,
+      "eval_steps_per_second": 0.757,
+      "step": 394000
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.152746045719953e-05,
+      "loss": 0.7832,
+      "step": 394200
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.1518082486287647e-05,
+      "loss": 0.7805,
+      "step": 394400
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.150879829508489e-05,
+      "loss": 0.7792,
+      "step": 394600
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.149942032417301e-05,
+      "loss": 0.7844,
+      "step": 394800
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.1490042353261133e-05,
+      "loss": 0.7825,
+      "step": 395000
+    },
+    {
+      "epoch": 0.46,
+      "eval_em_first_line_ratio": 0.4855,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3729,
+      "eval_loss": 0.7550972700119019,
+      "eval_runtime": 210.2452,
+      "eval_samples_per_second": 47.564,
+      "eval_steps_per_second": 0.747,
+      "step": 395000
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.1480664382349253e-05,
+      "loss": 0.7866,
+      "step": 395200
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.1471286411437374e-05,
+      "loss": 0.7836,
+      "step": 395400
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.1461908440525495e-05,
+      "loss": 0.7825,
+      "step": 395600
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.145253046961362e-05,
+      "loss": 0.7885,
+      "step": 395800
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.1443152498701737e-05,
+      "loss": 0.784,
+      "step": 396000
+    },
+    {
+      "epoch": 0.46,
+      "eval_em_first_line_ratio": 0.4858,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3733,
+      "eval_loss": 0.754244327545166,
+      "eval_runtime": 202.003,
+      "eval_samples_per_second": 49.504,
+      "eval_steps_per_second": 0.777,
+      "step": 396000
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.1433774527789862e-05,
+      "loss": 0.7906,
+      "step": 396200
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.142439655687798e-05,
+      "loss": 0.7886,
+      "step": 396400
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.1415018585966104e-05,
+      "loss": 0.7845,
+      "step": 396600
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.1405640615054225e-05,
+      "loss": 0.7836,
+      "step": 396800
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.1396309533996904e-05,
+      "loss": 0.7809,
+      "step": 397000
+    },
+    {
+      "epoch": 0.47,
+      "eval_em_first_line_ratio": 0.4875,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3746,
+      "eval_loss": 0.7537947297096252,
+      "eval_runtime": 204.0172,
+      "eval_samples_per_second": 49.015,
+      "eval_steps_per_second": 0.77,
+      "step": 397000
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.1386931563085028e-05,
+      "loss": 0.7874,
+      "step": 397200
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.1377600482027704e-05,
+      "loss": 0.7881,
+      "step": 397400
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.1368222511115828e-05,
+      "loss": 0.7827,
+      "step": 397600
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.1358844540203953e-05,
+      "loss": 0.7829,
+      "step": 397800
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.134946656929207e-05,
+      "loss": 0.7841,
+      "step": 398000
+    },
+    {
+      "epoch": 0.47,
+      "eval_em_first_line_ratio": 0.4869,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3738,
+      "eval_loss": 0.7538396120071411,
+      "eval_runtime": 211.6625,
+      "eval_samples_per_second": 47.245,
+      "eval_steps_per_second": 0.742,
+      "step": 398000
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.1340088598380195e-05,
+      "loss": 0.7796,
+      "step": 398200
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.1330710627468312e-05,
+      "loss": 0.7828,
+      "step": 398400
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.1321332656556436e-05,
+      "loss": 0.7849,
+      "step": 398600
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.1311954685644554e-05,
+      "loss": 0.7806,
+      "step": 398800
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.130257671473268e-05,
+      "loss": 0.7815,
+      "step": 399000
+    },
+    {
+      "epoch": 0.47,
+      "eval_em_first_line_ratio": 0.4872,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3746,
+      "eval_loss": 0.7541647553443909,
+      "eval_runtime": 204.889,
+      "eval_samples_per_second": 48.807,
+      "eval_steps_per_second": 0.766,
+      "step": 399000
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.12931987438208e-05,
+      "loss": 0.7841,
+      "step": 399200
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.1283820772908917e-05,
+      "loss": 0.7819,
+      "step": 399400
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.127444280199704e-05,
+      "loss": 0.7882,
+      "step": 399600
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.126506483108516e-05,
+      "loss": 0.7846,
+      "step": 399800
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.1255686860173283e-05,
+      "loss": 0.7855,
+      "step": 400000
+    },
+    {
+      "epoch": 0.47,
+      "eval_em_first_line_ratio": 0.4866,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3733,
+      "eval_loss": 0.7536464333534241,
+      "eval_runtime": 210.2756,
+      "eval_samples_per_second": 47.557,
+      "eval_steps_per_second": 0.747,
+      "step": 400000
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.1246308889261408e-05,
+      "loss": 0.7846,
+      "step": 400200
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.1236930918349525e-05,
+      "loss": 0.7827,
+      "step": 400400
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.122755294743765e-05,
+      "loss": 0.7838,
+      "step": 400600
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.1218174976525767e-05,
+      "loss": 0.7851,
+      "step": 400800
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.120879700561389e-05,
+      "loss": 0.7816,
+      "step": 401000
+    },
+    {
+      "epoch": 0.47,
+      "eval_em_first_line_ratio": 0.4872,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3744,
+      "eval_loss": 0.7535558938980103,
+      "eval_runtime": 209.194,
+      "eval_samples_per_second": 47.803,
+      "eval_steps_per_second": 0.75,
+      "step": 401000
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.119941903470201e-05,
+      "loss": 0.7831,
+      "step": 401200
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.1190041063790133e-05,
+      "loss": 0.7816,
+      "step": 401400
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.1180663092878254e-05,
+      "loss": 0.7858,
+      "step": 401600
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.1171285121966375e-05,
+      "loss": 0.7873,
+      "step": 401800
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.1161907151054496e-05,
+      "loss": 0.7843,
+      "step": 402000
+    },
+    {
+      "epoch": 0.47,
+      "eval_em_first_line_ratio": 0.4852,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3725,
+      "eval_loss": 0.7529597878456116,
+      "eval_runtime": 205.2684,
+      "eval_samples_per_second": 48.717,
+      "eval_steps_per_second": 0.765,
+      "step": 402000
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.1152529180142617e-05,
+      "loss": 0.7828,
+      "step": 402200
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.114315120923074e-05,
+      "loss": 0.7854,
+      "step": 402400
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.113377323831886e-05,
+      "loss": 0.7855,
+      "step": 402600
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.112439526740698e-05,
+      "loss": 0.7836,
+      "step": 402800
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.1115017296495105e-05,
+      "loss": 0.7838,
+      "step": 403000
+    },
+    {
+      "epoch": 0.47,
+      "eval_em_first_line_ratio": 0.4878,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3749,
+      "eval_loss": 0.7532795667648315,
+      "eval_runtime": 209.32,
+      "eval_samples_per_second": 47.774,
+      "eval_steps_per_second": 0.75,
+      "step": 403000
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.1105639325583222e-05,
+      "loss": 0.7864,
+      "step": 403200
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.1096261354671347e-05,
+      "loss": 0.7849,
+      "step": 403400
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.1086883383759464e-05,
+      "loss": 0.7836,
+      "step": 403600
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.1077552302702147e-05,
+      "loss": 0.784,
+      "step": 403800
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.106817433179027e-05,
+      "loss": 0.7862,
+      "step": 404000
+    },
+    {
+      "epoch": 0.47,
+      "eval_em_first_line_ratio": 0.4843,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3714,
+      "eval_loss": 0.7533016800880432,
+      "eval_runtime": 213.3784,
+      "eval_samples_per_second": 46.865,
+      "eval_steps_per_second": 0.736,
+      "step": 404000
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.105879636087839e-05,
+      "loss": 0.785,
+      "step": 404200
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.1049418389966513e-05,
+      "loss": 0.7826,
+      "step": 404400
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.104008730890919e-05,
+      "loss": 0.7819,
+      "step": 404600
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.1030709337997313e-05,
+      "loss": 0.7821,
+      "step": 404800
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.1021331367085437e-05,
+      "loss": 0.7845,
+      "step": 405000
+    },
+    {
+      "epoch": 0.47,
+      "eval_em_first_line_ratio": 0.4843,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3716,
+      "eval_loss": 0.7531434297561646,
+      "eval_runtime": 203.5158,
+      "eval_samples_per_second": 49.136,
+      "eval_steps_per_second": 0.771,
+      "step": 405000
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.1011953396173555e-05,
+      "loss": 0.781,
+      "step": 405200
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.1002575425261676e-05,
+      "loss": 0.7821,
+      "step": 405400
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.0993197454349797e-05,
+      "loss": 0.7846,
+      "step": 405600
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.0983819483437918e-05,
+      "loss": 0.7843,
+      "step": 405800
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.097444151252604e-05,
+      "loss": 0.7825,
+      "step": 406000
+    },
+    {
+      "epoch": 0.48,
+      "eval_em_first_line_ratio": 0.4867,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3738,
+      "eval_loss": 0.7525573968887329,
+      "eval_runtime": 212.1372,
+      "eval_samples_per_second": 47.139,
+      "eval_steps_per_second": 0.74,
+      "step": 406000
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.096506354161416e-05,
+      "loss": 0.7848,
+      "step": 406200
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.0955685570702284e-05,
+      "loss": 0.784,
+      "step": 406400
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.0946307599790402e-05,
+      "loss": 0.7808,
+      "step": 406600
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.0936929628878526e-05,
+      "loss": 0.7826,
+      "step": 406800
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.0927551657966644e-05,
+      "loss": 0.7782,
+      "step": 407000
+    },
+    {
+      "epoch": 0.48,
+      "eval_em_first_line_ratio": 0.4857,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3732,
+      "eval_loss": 0.752923846244812,
+      "eval_runtime": 210.01,
+      "eval_samples_per_second": 47.617,
+      "eval_steps_per_second": 0.748,
+      "step": 407000
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.0918173687054768e-05,
+      "loss": 0.7837,
+      "step": 407200
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.0908795716142893e-05,
+      "loss": 0.7833,
+      "step": 407400
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.089941774523101e-05,
+      "loss": 0.7802,
+      "step": 407600
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.0890039774319134e-05,
+      "loss": 0.7833,
+      "step": 407800
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.0880661803407252e-05,
+      "loss": 0.7824,
+      "step": 408000
+    },
+    {
+      "epoch": 0.48,
+      "eval_em_first_line_ratio": 0.4864,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3735,
+      "eval_loss": 0.7522678375244141,
+      "eval_runtime": 210.1403,
+      "eval_samples_per_second": 47.587,
+      "eval_steps_per_second": 0.747,
+      "step": 408000
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.0871283832495376e-05,
+      "loss": 0.7784,
+      "step": 408200
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.0861905861583494e-05,
+      "loss": 0.7806,
+      "step": 408400
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.0852527890671615e-05,
+      "loss": 0.7876,
+      "step": 408600
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.084314991975974e-05,
+      "loss": 0.7788,
+      "step": 408800
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.083381883870242e-05,
+      "loss": 0.784,
+      "step": 409000
+    },
+    {
+      "epoch": 0.48,
+      "eval_em_first_line_ratio": 0.4844,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3712,
+      "eval_loss": 0.7529871463775635,
+      "eval_runtime": 200.1314,
+      "eval_samples_per_second": 49.967,
+      "eval_steps_per_second": 0.784,
+      "step": 409000
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.082444086779054e-05,
+      "loss": 0.7818,
+      "step": 409200
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.081506289687866e-05,
+      "loss": 0.7766,
+      "step": 409400
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.080568492596678e-05,
+      "loss": 0.7808,
+      "step": 409600
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.0796306955054906e-05,
+      "loss": 0.781,
+      "step": 409800
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.0786928984143023e-05,
+      "loss": 0.7837,
+      "step": 410000
+    },
+    {
+      "epoch": 0.48,
+      "eval_em_first_line_ratio": 0.488,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.375,
+      "eval_loss": 0.7524763941764832,
+      "eval_runtime": 205.4241,
+      "eval_samples_per_second": 48.68,
+      "eval_steps_per_second": 0.764,
+      "step": 410000
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.0777551013231148e-05,
+      "loss": 0.7793,
+      "step": 410200
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.0768173042319265e-05,
+      "loss": 0.7837,
+      "step": 410400
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.075879507140739e-05,
+      "loss": 0.7888,
+      "step": 410600
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.0749417100495507e-05,
+      "loss": 0.7809,
+      "step": 410800
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.074003912958363e-05,
+      "loss": 0.7792,
+      "step": 411000
+    },
+    {
+      "epoch": 0.48,
+      "eval_em_first_line_ratio": 0.4863,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3736,
+      "eval_loss": 0.7526983618736267,
+      "eval_runtime": 204.6995,
+      "eval_samples_per_second": 48.852,
+      "eval_steps_per_second": 0.767,
+      "step": 411000
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.0730708048526314e-05,
+      "loss": 0.7873,
+      "step": 411200
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.072133007761443e-05,
+      "loss": 0.7789,
+      "step": 411400
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.0711952106702556e-05,
+      "loss": 0.7812,
+      "step": 411600
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.0702574135790674e-05,
+      "loss": 0.7813,
+      "step": 411800
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.0693196164878798e-05,
+      "loss": 0.7786,
+      "step": 412000
+    },
+    {
+      "epoch": 0.48,
+      "eval_em_first_line_ratio": 0.485,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3719,
+      "eval_loss": 0.7519584894180298,
+      "eval_runtime": 218.4235,
+      "eval_samples_per_second": 45.783,
+      "eval_steps_per_second": 0.719,
+      "step": 412000
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.068381819396692e-05,
+      "loss": 0.7807,
+      "step": 412200
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.067444022305504e-05,
+      "loss": 0.7781,
+      "step": 412400
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.066506225214316e-05,
+      "loss": 0.7822,
+      "step": 412600
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.0655684281231282e-05,
+      "loss": 0.7789,
+      "step": 412800
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.0646306310319403e-05,
+      "loss": 0.7826,
+      "step": 413000
+    },
+    {
+      "epoch": 0.48,
+      "eval_em_first_line_ratio": 0.4854,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3723,
+      "eval_loss": 0.7517655491828918,
+      "eval_runtime": 209.2442,
+      "eval_samples_per_second": 47.791,
+      "eval_steps_per_second": 0.75,
+      "step": 413000
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.0636928339407527e-05,
+      "loss": 0.775,
+      "step": 413200
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.0627550368495645e-05,
+      "loss": 0.7776,
+      "step": 413400
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.061817239758377e-05,
+      "loss": 0.7873,
+      "step": 413600
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.0608794426671887e-05,
+      "loss": 0.7816,
+      "step": 413800
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.059941645576001e-05,
+      "loss": 0.7841,
+      "step": 414000
+    },
+    {
+      "epoch": 0.49,
+      "eval_em_first_line_ratio": 0.4865,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3736,
+      "eval_loss": 0.7513244152069092,
+      "eval_runtime": 208.4439,
+      "eval_samples_per_second": 47.975,
+      "eval_steps_per_second": 0.753,
+      "step": 414000
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.059008537470269e-05,
+      "loss": 0.7812,
+      "step": 414200
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.058070740379081e-05,
+      "loss": 0.7851,
+      "step": 414400
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.0571329432878936e-05,
+      "loss": 0.7832,
+      "step": 414600
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.0561951461967053e-05,
+      "loss": 0.7802,
+      "step": 414800
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.0552573491055177e-05,
+      "loss": 0.7816,
+      "step": 415000
+    },
+    {
+      "epoch": 0.49,
+      "eval_em_first_line_ratio": 0.4858,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3732,
+      "eval_loss": 0.7511129975318909,
+      "eval_runtime": 205.3016,
+      "eval_samples_per_second": 48.709,
+      "eval_steps_per_second": 0.765,
+      "step": 415000
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.0543195520143295e-05,
+      "loss": 0.7856,
+      "step": 415200
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.053381754923142e-05,
+      "loss": 0.7842,
+      "step": 415400
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.052443957831954e-05,
+      "loss": 0.7789,
+      "step": 415600
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.051510849726222e-05,
+      "loss": 0.7841,
+      "step": 415800
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.0505730526350344e-05,
+      "loss": 0.7796,
+      "step": 416000
+    },
+    {
+      "epoch": 0.49,
+      "eval_em_first_line_ratio": 0.4867,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.374,
+      "eval_loss": 0.7513077855110168,
+      "eval_runtime": 209.0779,
+      "eval_samples_per_second": 47.829,
+      "eval_steps_per_second": 0.751,
+      "step": 416000
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.049635255543846e-05,
+      "loss": 0.7835,
+      "step": 416200
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.0486974584526586e-05,
+      "loss": 0.7771,
+      "step": 416400
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.0477596613614707e-05,
+      "loss": 0.7827,
+      "step": 416600
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.0468218642702828e-05,
+      "loss": 0.7786,
+      "step": 416800
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.045884067179095e-05,
+      "loss": 0.7825,
+      "step": 417000
+    },
+    {
+      "epoch": 0.49,
+      "eval_em_first_line_ratio": 0.4845,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3717,
+      "eval_loss": 0.7511657476425171,
+      "eval_runtime": 205.8481,
+      "eval_samples_per_second": 48.58,
+      "eval_steps_per_second": 0.763,
+      "step": 417000
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.044946270087907e-05,
+      "loss": 0.7823,
+      "step": 417200
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.044008472996719e-05,
+      "loss": 0.7807,
+      "step": 417400
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.0430706759055308e-05,
+      "loss": 0.7788,
+      "step": 417600
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.0421328788143433e-05,
+      "loss": 0.782,
+      "step": 417800
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.0411950817231557e-05,
+      "loss": 0.782,
+      "step": 418000
+    },
+    {
+      "epoch": 0.49,
+      "eval_em_first_line_ratio": 0.4847,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3724,
+      "eval_loss": 0.7512258291244507,
+      "eval_runtime": 206.4324,
+      "eval_samples_per_second": 48.442,
+      "eval_steps_per_second": 0.761,
+      "step": 418000
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.0402572846319675e-05,
+      "loss": 0.7762,
+      "step": 418200
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.0393241765262357e-05,
+      "loss": 0.7804,
+      "step": 418400
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.0383863794350475e-05,
+      "loss": 0.7769,
+      "step": 418600
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.03744858234386e-05,
+      "loss": 0.7814,
+      "step": 418800
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.0365107852526723e-05,
+      "loss": 0.7811,
+      "step": 419000
+    },
+    {
+      "epoch": 0.49,
+      "eval_em_first_line_ratio": 0.4862,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3736,
+      "eval_loss": 0.751089870929718,
+      "eval_runtime": 206.9807,
+      "eval_samples_per_second": 48.314,
+      "eval_steps_per_second": 0.759,
+      "step": 419000
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.035572988161484e-05,
+      "loss": 0.7794,
+      "step": 419200
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.0346351910702965e-05,
+      "loss": 0.7804,
+      "step": 419400
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.0336973939791083e-05,
+      "loss": 0.7782,
+      "step": 419600
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.0327595968879207e-05,
+      "loss": 0.779,
+      "step": 419800
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.0318217997967325e-05,
+      "loss": 0.7866,
+      "step": 420000
+    },
+    {
+      "epoch": 0.49,
+      "eval_em_first_line_ratio": 0.4867,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3738,
+      "eval_loss": 0.7511792778968811,
+      "eval_runtime": 202.0791,
+      "eval_samples_per_second": 49.486,
+      "eval_steps_per_second": 0.777,
+      "step": 420000
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.030884002705545e-05,
+      "loss": 0.7817,
+      "step": 420200
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.029946205614357e-05,
+      "loss": 0.7783,
+      "step": 420400
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.0290084085231688e-05,
+      "loss": 0.7784,
+      "step": 420600
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.0280706114319812e-05,
+      "loss": 0.7863,
+      "step": 420800
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.027137503326249e-05,
+      "loss": 0.786,
+      "step": 421000
+    },
+    {
+      "epoch": 0.49,
+      "eval_em_first_line_ratio": 0.4862,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3736,
+      "eval_loss": 0.7502539753913879,
+      "eval_runtime": 210.6065,
+      "eval_samples_per_second": 47.482,
+      "eval_steps_per_second": 0.745,
+      "step": 421000
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.0261997062350612e-05,
+      "loss": 0.7819,
+      "step": 421200
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.0252619091438737e-05,
+      "loss": 0.7794,
+      "step": 421400
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.0243241120526854e-05,
+      "loss": 0.7797,
+      "step": 421600
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.023386314961498e-05,
+      "loss": 0.7783,
+      "step": 421800
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.0224485178703096e-05,
+      "loss": 0.7743,
+      "step": 422000
+    },
+    {
+      "epoch": 0.49,
+      "eval_em_first_line_ratio": 0.4849,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.372,
+      "eval_loss": 0.7506071925163269,
+      "eval_runtime": 206.5826,
+      "eval_samples_per_second": 48.407,
+      "eval_steps_per_second": 0.76,
+      "step": 422000
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.021510720779122e-05,
+      "loss": 0.7794,
+      "step": 422200
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 2.0205729236879338e-05,
+      "loss": 0.7822,
+      "step": 422400
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 2.0196351265967462e-05,
+      "loss": 0.7788,
+      "step": 422600
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 2.0186973295055587e-05,
+      "loss": 0.7847,
+      "step": 422800
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 2.0177595324143704e-05,
+      "loss": 0.7816,
+      "step": 423000
+    },
+    {
+      "epoch": 0.5,
+      "eval_em_first_line_ratio": 0.487,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3746,
+      "eval_loss": 0.7510306239128113,
+      "eval_runtime": 208.1727,
+      "eval_samples_per_second": 48.037,
+      "eval_steps_per_second": 0.754,
+      "step": 423000
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 2.016821735323183e-05,
+      "loss": 0.781,
+      "step": 423200
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 2.0158839382319946e-05,
+      "loss": 0.7809,
+      "step": 423400
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 2.0149461411408067e-05,
+      "loss": 0.7799,
+      "step": 423600
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 2.014008344049619e-05,
+      "loss": 0.7823,
+      "step": 423800
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 2.013070546958431e-05,
+      "loss": 0.7764,
+      "step": 424000
+    },
+    {
+      "epoch": 0.5,
+      "eval_em_first_line_ratio": 0.4866,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3739,
+      "eval_loss": 0.7502138614654541,
+      "eval_runtime": 206.2545,
+      "eval_samples_per_second": 48.484,
+      "eval_steps_per_second": 0.761,
+      "step": 424000
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 2.0121327498672434e-05,
+      "loss": 0.7779,
+      "step": 424200
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 2.0111996417615113e-05,
+      "loss": 0.779,
+      "step": 424400
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 2.0102618446703234e-05,
+      "loss": 0.7735,
+      "step": 424600
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 2.0093240475791358e-05,
+      "loss": 0.785,
+      "step": 424800
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 2.0083862504879476e-05,
+      "loss": 0.7747,
+      "step": 425000
+    },
+    {
+      "epoch": 0.5,
+      "eval_em_first_line_ratio": 0.4872,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3743,
+      "eval_loss": 0.7503239512443542,
+      "eval_runtime": 209.1286,
+      "eval_samples_per_second": 47.817,
+      "eval_steps_per_second": 0.751,
+      "step": 425000
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 2.0074531423822158e-05,
+      "loss": 0.7809,
+      "step": 425200
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 2.006515345291028e-05,
+      "loss": 0.7797,
+      "step": 425400
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 2.00557754819984e-05,
+      "loss": 0.7779,
+      "step": 425600
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 2.004639751108652e-05,
+      "loss": 0.7798,
+      "step": 425800
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 2.0037019540174642e-05,
+      "loss": 0.7779,
+      "step": 426000
+    },
+    {
+      "epoch": 0.5,
+      "eval_em_first_line_ratio": 0.4855,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3728,
+      "eval_loss": 0.7508491277694702,
+      "eval_runtime": 206.3564,
+      "eval_samples_per_second": 48.46,
+      "eval_steps_per_second": 0.761,
+      "step": 426000
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 2.0027641569262766e-05,
+      "loss": 0.7828,
+      "step": 426200
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 2.0018263598350884e-05,
+      "loss": 0.7804,
+      "step": 426400
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 2.000888562743901e-05,
+      "loss": 0.7805,
+      "step": 426600
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 1.999950765652713e-05,
+      "loss": 0.778,
+      "step": 426800
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 1.999012968561525e-05,
+      "loss": 0.7835,
+      "step": 427000
+    },
+    {
+      "epoch": 0.5,
+      "eval_em_first_line_ratio": 0.4873,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3746,
+      "eval_loss": 0.7498982548713684,
+      "eval_runtime": 205.8198,
+      "eval_samples_per_second": 48.586,
+      "eval_steps_per_second": 0.763,
+      "step": 427000
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 1.998075171470337e-05,
+      "loss": 0.7754,
+      "step": 427200
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 1.9971373743791492e-05,
+      "loss": 0.7825,
+      "step": 427400
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 1.9961995772879613e-05,
+      "loss": 0.7805,
+      "step": 427600
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 1.9952617801967734e-05,
+      "loss": 0.7819,
+      "step": 427800
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 1.9943239831055855e-05,
+      "loss": 0.7792,
+      "step": 428000
+    },
+    {
+      "epoch": 0.5,
+      "eval_em_first_line_ratio": 0.4853,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3737,
+      "eval_loss": 0.7508020997047424,
+      "eval_runtime": 204.8275,
+      "eval_samples_per_second": 48.822,
+      "eval_steps_per_second": 0.766,
+      "step": 428000
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 1.9933861860143976e-05,
+      "loss": 0.7776,
+      "step": 428200
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 1.9924483889232097e-05,
+      "loss": 0.7825,
+      "step": 428400
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 1.9915105918320218e-05,
+      "loss": 0.7813,
+      "step": 428600
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 1.990572794740834e-05,
+      "loss": 0.7819,
+      "step": 428800
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 1.9896349976496463e-05,
+      "loss": 0.7853,
+      "step": 429000
+    },
+    {
+      "epoch": 0.5,
+      "eval_em_first_line_ratio": 0.4869,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3751,
+      "eval_loss": 0.7496957778930664,
+      "eval_runtime": 213.7222,
+      "eval_samples_per_second": 46.79,
+      "eval_steps_per_second": 0.735,
+      "step": 429000
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 1.9886972005584584e-05,
+      "loss": 0.7817,
+      "step": 429200
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 1.9877640924527263e-05,
+      "loss": 0.7768,
+      "step": 429400
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 1.9868262953615384e-05,
+      "loss": 0.7808,
+      "step": 429600
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 1.9858884982703505e-05,
+      "loss": 0.776,
+      "step": 429800
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 1.984950701179163e-05,
+      "loss": 0.7793,
+      "step": 430000
+    },
+    {
+      "epoch": 0.5,
+      "eval_em_first_line_ratio": 0.4866,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.374,
+      "eval_loss": 0.7496193647384644,
+      "eval_runtime": 206.26,
+      "eval_samples_per_second": 48.482,
+      "eval_steps_per_second": 0.761,
+      "step": 430000
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 1.984012904087975e-05,
+      "loss": 0.7801,
+      "step": 430200
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 1.9830751069967872e-05,
+      "loss": 0.7775,
+      "step": 430400
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 1.9821373099055993e-05,
+      "loss": 0.776,
+      "step": 430600
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 1.9811995128144114e-05,
+      "loss": 0.777,
+      "step": 430800
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.9802617157232235e-05,
+      "loss": 0.7884,
+      "step": 431000
+    },
+    {
+      "epoch": 0.51,
+      "eval_em_first_line_ratio": 0.4867,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3739,
+      "eval_loss": 0.7508971691131592,
+      "eval_runtime": 200.5827,
+      "eval_samples_per_second": 49.855,
+      "eval_steps_per_second": 0.783,
+      "step": 431000
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.9793239186320356e-05,
+      "loss": 0.7793,
+      "step": 431200
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.9783861215408477e-05,
+      "loss": 0.7787,
+      "step": 431400
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.9774483244496598e-05,
+      "loss": 0.7794,
+      "step": 431600
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.976510527358472e-05,
+      "loss": 0.7797,
+      "step": 431800
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.975572730267284e-05,
+      "loss": 0.7832,
+      "step": 432000
+    },
+    {
+      "epoch": 0.51,
+      "eval_em_first_line_ratio": 0.4846,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3719,
+      "eval_loss": 0.7496617436408997,
+      "eval_runtime": 196.3182,
+      "eval_samples_per_second": 50.938,
+      "eval_steps_per_second": 0.8,
+      "step": 432000
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.974634933176096e-05,
+      "loss": 0.7784,
+      "step": 432200
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.973697136084908e-05,
+      "loss": 0.7747,
+      "step": 432400
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.9727593389937206e-05,
+      "loss": 0.7774,
+      "step": 432600
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.9718215419025327e-05,
+      "loss": 0.7783,
+      "step": 432800
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.9708837448113448e-05,
+      "loss": 0.7795,
+      "step": 433000
+    },
+    {
+      "epoch": 0.51,
+      "eval_em_first_line_ratio": 0.4875,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3747,
+      "eval_loss": 0.7503668665885925,
+      "eval_runtime": 199.1879,
+      "eval_samples_per_second": 50.204,
+      "eval_steps_per_second": 0.788,
+      "step": 433000
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.969945947720157e-05,
+      "loss": 0.7822,
+      "step": 433200
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.969008150628969e-05,
+      "loss": 0.783,
+      "step": 433400
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.9680750425232372e-05,
+      "loss": 0.7809,
+      "step": 433600
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.9671372454320493e-05,
+      "loss": 0.7839,
+      "step": 433800
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.9661994483408614e-05,
+      "loss": 0.7744,
+      "step": 434000
+    },
+    {
+      "epoch": 0.51,
+      "eval_em_first_line_ratio": 0.4861,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3734,
+      "eval_loss": 0.7493233680725098,
+      "eval_runtime": 204.9493,
+      "eval_samples_per_second": 48.793,
+      "eval_steps_per_second": 0.766,
+      "step": 434000
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.9652616512496735e-05,
+      "loss": 0.7758,
+      "step": 434200
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.9643238541584856e-05,
+      "loss": 0.7792,
+      "step": 434400
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.9633860570672977e-05,
+      "loss": 0.7794,
+      "step": 434600
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.9624482599761098e-05,
+      "loss": 0.7789,
+      "step": 434800
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.961510462884922e-05,
+      "loss": 0.7795,
+      "step": 435000
+    },
+    {
+      "epoch": 0.51,
+      "eval_em_first_line_ratio": 0.4871,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3744,
+      "eval_loss": 0.7499796152114868,
+      "eval_runtime": 203.0086,
+      "eval_samples_per_second": 49.259,
+      "eval_steps_per_second": 0.773,
+      "step": 435000
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.960572665793734e-05,
+      "loss": 0.7782,
+      "step": 435200
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.959634868702546e-05,
+      "loss": 0.7736,
+      "step": 435400
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.9586970716113582e-05,
+      "loss": 0.778,
+      "step": 435600
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.9577592745201703e-05,
+      "loss": 0.7752,
+      "step": 435800
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.9568214774289827e-05,
+      "loss": 0.7864,
+      "step": 436000
+    },
+    {
+      "epoch": 0.51,
+      "eval_em_first_line_ratio": 0.4861,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3734,
+      "eval_loss": 0.7496193051338196,
+      "eval_runtime": 199.6024,
+      "eval_samples_per_second": 50.1,
+      "eval_steps_per_second": 0.787,
+      "step": 436000
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.9558930583087065e-05,
+      "loss": 0.7766,
+      "step": 436200
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.9549552612175186e-05,
+      "loss": 0.7767,
+      "step": 436400
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.9540174641263306e-05,
+      "loss": 0.7754,
+      "step": 436600
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.9530796670351427e-05,
+      "loss": 0.7771,
+      "step": 436800
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.9521418699439552e-05,
+      "loss": 0.7781,
+      "step": 437000
+    },
+    {
+      "epoch": 0.51,
+      "eval_em_first_line_ratio": 0.4886,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3762,
+      "eval_loss": 0.7492749094963074,
+      "eval_runtime": 203.5899,
+      "eval_samples_per_second": 49.118,
+      "eval_steps_per_second": 0.771,
+      "step": 437000
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.9512040728527673e-05,
+      "loss": 0.7802,
+      "step": 437200
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.9502662757615794e-05,
+      "loss": 0.7819,
+      "step": 437400
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.9493284786703915e-05,
+      "loss": 0.7757,
+      "step": 437600
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.9483906815792036e-05,
+      "loss": 0.7777,
+      "step": 437800
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.9474528844880157e-05,
+      "loss": 0.781,
+      "step": 438000
+    },
+    {
+      "epoch": 0.51,
+      "eval_em_first_line_ratio": 0.4855,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3731,
+      "eval_loss": 0.7488881945610046,
+      "eval_runtime": 204.1455,
+      "eval_samples_per_second": 48.985,
+      "eval_steps_per_second": 0.769,
+      "step": 438000
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.9465150873968278e-05,
+      "loss": 0.7745,
+      "step": 438200
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.94557729030564e-05,
+      "loss": 0.7789,
+      "step": 438400
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.944639493214452e-05,
+      "loss": 0.7789,
+      "step": 438600
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.943701696123264e-05,
+      "loss": 0.7721,
+      "step": 438800
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.942763899032076e-05,
+      "loss": 0.7734,
+      "step": 439000
+    },
+    {
+      "epoch": 0.51,
+      "eval_em_first_line_ratio": 0.4878,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3752,
+      "eval_loss": 0.7488340139389038,
+      "eval_runtime": 204.2868,
+      "eval_samples_per_second": 48.951,
+      "eval_steps_per_second": 0.769,
+      "step": 439000
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.9418261019408883e-05,
+      "loss": 0.7801,
+      "step": 439200
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1.9408883048497007e-05,
+      "loss": 0.7794,
+      "step": 439400
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1.9399551967439686e-05,
+      "loss": 0.7765,
+      "step": 439600
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1.9390173996527807e-05,
+      "loss": 0.7795,
+      "step": 439800
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1.9380796025615928e-05,
+      "loss": 0.7751,
+      "step": 440000
+    },
+    {
+      "epoch": 0.52,
+      "eval_em_first_line_ratio": 0.4887,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3758,
+      "eval_loss": 0.7487533092498779,
+      "eval_runtime": 203.3569,
+      "eval_samples_per_second": 49.175,
+      "eval_steps_per_second": 0.772,
+      "step": 440000
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1.937141805470405e-05,
+      "loss": 0.7824,
+      "step": 440200
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1.936204008379217e-05,
+      "loss": 0.778,
+      "step": 440400
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1.9352662112880294e-05,
+      "loss": 0.7743,
+      "step": 440600
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1.9343284141968415e-05,
+      "loss": 0.7736,
+      "step": 440800
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1.9333906171056536e-05,
+      "loss": 0.7769,
+      "step": 441000
+    },
+    {
+      "epoch": 0.52,
+      "eval_em_first_line_ratio": 0.4858,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3736,
+      "eval_loss": 0.7490732669830322,
+      "eval_runtime": 208.2564,
+      "eval_samples_per_second": 48.018,
+      "eval_steps_per_second": 0.754,
+      "step": 441000
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1.9324528200144657e-05,
+      "loss": 0.7742,
+      "step": 441200
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1.9315150229232778e-05,
+      "loss": 0.7771,
+      "step": 441400
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1.93057722583209e-05,
+      "loss": 0.7812,
+      "step": 441600
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1.929639428740902e-05,
+      "loss": 0.7777,
+      "step": 441800
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1.928701631649714e-05,
+      "loss": 0.7761,
+      "step": 442000
+    },
+    {
+      "epoch": 0.52,
+      "eval_em_first_line_ratio": 0.4873,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3752,
+      "eval_loss": 0.7492974400520325,
+      "eval_runtime": 205.4153,
+      "eval_samples_per_second": 48.682,
+      "eval_steps_per_second": 0.764,
+      "step": 442000
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1.9277638345585262e-05,
+      "loss": 0.7772,
+      "step": 442200
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1.9268260374673383e-05,
+      "loss": 0.7731,
+      "step": 442400
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1.9258882403761504e-05,
+      "loss": 0.7752,
+      "step": 442600
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1.9249551322704186e-05,
+      "loss": 0.7773,
+      "step": 442800
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1.9240173351792307e-05,
+      "loss": 0.7763,
+      "step": 443000
+    },
+    {
+      "epoch": 0.52,
+      "eval_em_first_line_ratio": 0.4887,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3757,
+      "eval_loss": 0.7487926483154297,
+      "eval_runtime": 206.4562,
+      "eval_samples_per_second": 48.436,
+      "eval_steps_per_second": 0.76,
+      "step": 443000
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1.923084227073499e-05,
+      "loss": 0.7789,
+      "step": 443200
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1.922146429982311e-05,
+      "loss": 0.7772,
+      "step": 443400
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1.9212086328911232e-05,
+      "loss": 0.7773,
+      "step": 443600
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1.9202708357999353e-05,
+      "loss": 0.7737,
+      "step": 443800
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1.9193330387087474e-05,
+      "loss": 0.7778,
+      "step": 444000
+    },
+    {
+      "epoch": 0.52,
+      "eval_em_first_line_ratio": 0.4893,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3767,
+      "eval_loss": 0.748045027256012,
+      "eval_runtime": 207.4658,
+      "eval_samples_per_second": 48.201,
+      "eval_steps_per_second": 0.757,
+      "step": 444000
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1.9183952416175595e-05,
+      "loss": 0.7767,
+      "step": 444200
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1.9174574445263716e-05,
+      "loss": 0.781,
+      "step": 444400
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1.9165196474351837e-05,
+      "loss": 0.7722,
+      "step": 444600
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1.9155818503439958e-05,
+      "loss": 0.7753,
+      "step": 444800
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1.914644053252808e-05,
+      "loss": 0.7796,
+      "step": 445000
+    },
+    {
+      "epoch": 0.52,
+      "eval_em_first_line_ratio": 0.4885,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.376,
+      "eval_loss": 0.7483962178230286,
+      "eval_runtime": 199.2447,
+      "eval_samples_per_second": 50.19,
+      "eval_steps_per_second": 0.788,
+      "step": 445000
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1.9137062561616203e-05,
+      "loss": 0.7771,
+      "step": 445200
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1.9127684590704324e-05,
+      "loss": 0.7755,
+      "step": 445400
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1.9118306619792445e-05,
+      "loss": 0.7753,
+      "step": 445600
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1.9108928648880566e-05,
+      "loss": 0.7766,
+      "step": 445800
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1.9099550677968687e-05,
+      "loss": 0.7744,
+      "step": 446000
+    },
+    {
+      "epoch": 0.52,
+      "eval_em_first_line_ratio": 0.4876,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3755,
+      "eval_loss": 0.7483774423599243,
+      "eval_runtime": 202.4774,
+      "eval_samples_per_second": 49.388,
+      "eval_steps_per_second": 0.775,
+      "step": 446000
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1.9090172707056808e-05,
+      "loss": 0.7731,
+      "step": 446200
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1.908079473614493e-05,
+      "loss": 0.7784,
+      "step": 446400
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1.907141676523305e-05,
+      "loss": 0.7725,
+      "step": 446600
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1.906203879432117e-05,
+      "loss": 0.7767,
+      "step": 446800
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1.9052660823409292e-05,
+      "loss": 0.7746,
+      "step": 447000
+    },
+    {
+      "epoch": 0.52,
+      "eval_em_first_line_ratio": 0.4885,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3758,
+      "eval_loss": 0.747140645980835,
+      "eval_runtime": 204.0664,
+      "eval_samples_per_second": 49.004,
+      "eval_steps_per_second": 0.769,
+      "step": 447000
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1.9043282852497413e-05,
+      "loss": 0.7743,
+      "step": 447200
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1.9033904881585534e-05,
+      "loss": 0.778,
+      "step": 447400
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1.9024526910673658e-05,
+      "loss": 0.7715,
+      "step": 447600
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1.901514893976178e-05,
+      "loss": 0.7756,
+      "step": 447800
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.90057709688499e-05,
+      "loss": 0.7782,
+      "step": 448000
+    },
+    {
+      "epoch": 0.53,
+      "eval_em_first_line_ratio": 0.4898,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3775,
+      "eval_loss": 0.747670590877533,
+      "eval_runtime": 204.6863,
+      "eval_samples_per_second": 48.855,
+      "eval_steps_per_second": 0.767,
+      "step": 448000
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.899643988779258e-05,
+      "loss": 0.772,
+      "step": 448200
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.89870619168807e-05,
+      "loss": 0.7746,
+      "step": 448400
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.897768394596882e-05,
+      "loss": 0.7745,
+      "step": 448600
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.8968305975056946e-05,
+      "loss": 0.7755,
+      "step": 448800
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.8958928004145066e-05,
+      "loss": 0.7775,
+      "step": 449000
+    },
+    {
+      "epoch": 0.53,
+      "eval_em_first_line_ratio": 0.4874,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3751,
+      "eval_loss": 0.7475641369819641,
+      "eval_runtime": 206.6566,
+      "eval_samples_per_second": 48.389,
+      "eval_steps_per_second": 0.76,
+      "step": 449000
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.8949550033233187e-05,
+      "loss": 0.7743,
+      "step": 449200
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.894017206232131e-05,
+      "loss": 0.7688,
+      "step": 449400
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.893079409140943e-05,
+      "loss": 0.7757,
+      "step": 449600
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.892141612049755e-05,
+      "loss": 0.7731,
+      "step": 449800
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.891203814958567e-05,
+      "loss": 0.776,
+      "step": 450000
+    },
+    {
+      "epoch": 0.53,
+      "eval_em_first_line_ratio": 0.488,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3754,
+      "eval_loss": 0.7473819851875305,
+      "eval_runtime": 195.884,
+      "eval_samples_per_second": 51.051,
+      "eval_steps_per_second": 0.801,
+      "step": 450000
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.8902660178673792e-05,
+      "loss": 0.7812,
+      "step": 450200
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.8893282207761913e-05,
+      "loss": 0.7769,
+      "step": 450400
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.8883904236850034e-05,
+      "loss": 0.7796,
+      "step": 450600
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.8874526265938155e-05,
+      "loss": 0.7783,
+      "step": 450800
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.8865195184880838e-05,
+      "loss": 0.7774,
+      "step": 451000
+    },
+    {
+      "epoch": 0.53,
+      "eval_em_first_line_ratio": 0.4875,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.375,
+      "eval_loss": 0.74737149477005,
+      "eval_runtime": 206.7948,
+      "eval_samples_per_second": 48.357,
+      "eval_steps_per_second": 0.759,
+      "step": 451000
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.885581721396896e-05,
+      "loss": 0.7717,
+      "step": 451200
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.884643924305708e-05,
+      "loss": 0.7735,
+      "step": 451400
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.88370612721452e-05,
+      "loss": 0.774,
+      "step": 451600
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.882768330123332e-05,
+      "loss": 0.7781,
+      "step": 451800
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.8818305330321443e-05,
+      "loss": 0.7751,
+      "step": 452000
+    },
+    {
+      "epoch": 0.53,
+      "eval_em_first_line_ratio": 0.4897,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3772,
+      "eval_loss": 0.7468293905258179,
+      "eval_runtime": 199.6634,
+      "eval_samples_per_second": 50.084,
+      "eval_steps_per_second": 0.786,
+      "step": 452000
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.8808927359409564e-05,
+      "loss": 0.7738,
+      "step": 452200
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.8799549388497688e-05,
+      "loss": 0.7771,
+      "step": 452400
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.879017141758581e-05,
+      "loss": 0.777,
+      "step": 452600
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.878079344667393e-05,
+      "loss": 0.7764,
+      "step": 452800
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.877141547576205e-05,
+      "loss": 0.7781,
+      "step": 453000
+    },
+    {
+      "epoch": 0.53,
+      "eval_em_first_line_ratio": 0.4888,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3767,
+      "eval_loss": 0.7473405003547668,
+      "eval_runtime": 203.7712,
+      "eval_samples_per_second": 49.075,
+      "eval_steps_per_second": 0.77,
+      "step": 453000
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.8762037504850172e-05,
+      "loss": 0.7728,
+      "step": 453200
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.875265953393829e-05,
+      "loss": 0.7747,
+      "step": 453400
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.8743281563026414e-05,
+      "loss": 0.7771,
+      "step": 453600
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.8733903592114535e-05,
+      "loss": 0.7721,
+      "step": 453800
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.8724525621202656e-05,
+      "loss": 0.775,
+      "step": 454000
+    },
+    {
+      "epoch": 0.53,
+      "eval_em_first_line_ratio": 0.4865,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3742,
+      "eval_loss": 0.7471861839294434,
+      "eval_runtime": 199.8195,
+      "eval_samples_per_second": 50.045,
+      "eval_steps_per_second": 0.786,
+      "step": 454000
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.8715194540145335e-05,
+      "loss": 0.7787,
+      "step": 454200
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.8705816569233456e-05,
+      "loss": 0.7747,
+      "step": 454400
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.869643859832158e-05,
+      "loss": 0.7706,
+      "step": 454600
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.868710751726426e-05,
+      "loss": 0.7713,
+      "step": 454800
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.867772954635238e-05,
+      "loss": 0.7697,
+      "step": 455000
+    },
+    {
+      "epoch": 0.53,
+      "eval_em_first_line_ratio": 0.4877,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3746,
+      "eval_loss": 0.7474073171615601,
+      "eval_runtime": 198.4965,
+      "eval_samples_per_second": 50.379,
+      "eval_steps_per_second": 0.791,
+      "step": 455000
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.86683515754405e-05,
+      "loss": 0.7767,
+      "step": 455200
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.8658973604528622e-05,
+      "loss": 0.777,
+      "step": 455400
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.8649595633616743e-05,
+      "loss": 0.7793,
+      "step": 455600
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.8640217662704868e-05,
+      "loss": 0.7811,
+      "step": 455800
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.863083969179299e-05,
+      "loss": 0.7817,
+      "step": 456000
+    },
+    {
+      "epoch": 0.53,
+      "eval_em_first_line_ratio": 0.4881,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3752,
+      "eval_loss": 0.7468832731246948,
+      "eval_runtime": 202.3132,
+      "eval_samples_per_second": 49.428,
+      "eval_steps_per_second": 0.776,
+      "step": 456000
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.862146172088111e-05,
+      "loss": 0.7722,
+      "step": 456200
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.861208374996923e-05,
+      "loss": 0.78,
+      "step": 456400
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.860270577905735e-05,
+      "loss": 0.7751,
+      "step": 456600
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.8593327808145472e-05,
+      "loss": 0.7759,
+      "step": 456800
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.8583949837233593e-05,
+      "loss": 0.7732,
+      "step": 457000
+    },
+    {
+      "epoch": 0.54,
+      "eval_em_first_line_ratio": 0.4868,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3742,
+      "eval_loss": 0.7464658617973328,
+      "eval_runtime": 199.5984,
+      "eval_samples_per_second": 50.101,
+      "eval_steps_per_second": 0.787,
+      "step": 457000
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.8574571866321714e-05,
+      "loss": 0.7756,
+      "step": 457200
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.8565193895409835e-05,
+      "loss": 0.7744,
+      "step": 457400
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.8555815924497956e-05,
+      "loss": 0.777,
+      "step": 457600
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.8546437953586077e-05,
+      "loss": 0.7777,
+      "step": 457800
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.8537059982674198e-05,
+      "loss": 0.7744,
+      "step": 458000
+    },
+    {
+      "epoch": 0.54,
+      "eval_em_first_line_ratio": 0.4867,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3737,
+      "eval_loss": 0.7466756701469421,
+      "eval_runtime": 198.9506,
+      "eval_samples_per_second": 50.264,
+      "eval_steps_per_second": 0.789,
+      "step": 458000
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.8527682011762323e-05,
+      "loss": 0.7764,
+      "step": 458200
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.8518304040850444e-05,
+      "loss": 0.7745,
+      "step": 458400
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.8508926069938565e-05,
+      "loss": 0.7712,
+      "step": 458600
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.8499548099026686e-05,
+      "loss": 0.7751,
+      "step": 458800
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.8490170128114806e-05,
+      "loss": 0.7735,
+      "step": 459000
+    },
+    {
+      "epoch": 0.54,
+      "eval_em_first_line_ratio": 0.4888,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3761,
+      "eval_loss": 0.7478552460670471,
+      "eval_runtime": 201.743,
+      "eval_samples_per_second": 49.568,
+      "eval_steps_per_second": 0.778,
+      "step": 459000
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.8480792157202927e-05,
+      "loss": 0.7784,
+      "step": 459200
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.847141418629105e-05,
+      "loss": 0.7777,
+      "step": 459400
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.846203621537917e-05,
+      "loss": 0.7778,
+      "step": 459600
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.845265824446729e-05,
+      "loss": 0.7785,
+      "step": 459800
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.844328027355541e-05,
+      "loss": 0.7757,
+      "step": 460000
+    },
+    {
+      "epoch": 0.54,
+      "eval_em_first_line_ratio": 0.4884,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3753,
+      "eval_loss": 0.7462421655654907,
+      "eval_runtime": 203.6078,
+      "eval_samples_per_second": 49.114,
+      "eval_steps_per_second": 0.771,
+      "step": 460000
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.8433902302643532e-05,
+      "loss": 0.7786,
+      "step": 460200
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.8424524331731653e-05,
+      "loss": 0.779,
+      "step": 460400
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.8415146360819778e-05,
+      "loss": 0.781,
+      "step": 460600
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.84057683899079e-05,
+      "loss": 0.7703,
+      "step": 460800
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.839639041899602e-05,
+      "loss": 0.7703,
+      "step": 461000
+    },
+    {
+      "epoch": 0.54,
+      "eval_em_first_line_ratio": 0.4888,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3756,
+      "eval_loss": 0.7459765672683716,
+      "eval_runtime": 205.7282,
+      "eval_samples_per_second": 48.608,
+      "eval_steps_per_second": 0.763,
+      "step": 461000
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.838701244808414e-05,
+      "loss": 0.7709,
+      "step": 461200
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.837763447717226e-05,
+      "loss": 0.781,
+      "step": 461400
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.8368256506260383e-05,
+      "loss": 0.7721,
+      "step": 461600
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.8358878535348507e-05,
+      "loss": 0.7772,
+      "step": 461800
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.8349547454291186e-05,
+      "loss": 0.776,
+      "step": 462000
+    },
+    {
+      "epoch": 0.54,
+      "eval_em_first_line_ratio": 0.4883,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3754,
+      "eval_loss": 0.7467986345291138,
+      "eval_runtime": 206.697,
+      "eval_samples_per_second": 48.38,
+      "eval_steps_per_second": 0.76,
+      "step": 462000
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.8340169483379307e-05,
+      "loss": 0.772,
+      "step": 462200
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.8330791512467428e-05,
+      "loss": 0.7742,
+      "step": 462400
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.832141354155555e-05,
+      "loss": 0.7719,
+      "step": 462600
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.831203557064367e-05,
+      "loss": 0.7765,
+      "step": 462800
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.830265759973179e-05,
+      "loss": 0.7708,
+      "step": 463000
+    },
+    {
+      "epoch": 0.54,
+      "eval_em_first_line_ratio": 0.4885,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3754,
+      "eval_loss": 0.7463899850845337,
+      "eval_runtime": 208.0145,
+      "eval_samples_per_second": 48.074,
+      "eval_steps_per_second": 0.755,
+      "step": 463000
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.8293279628819912e-05,
+      "loss": 0.7728,
+      "step": 463200
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.8283901657908033e-05,
+      "loss": 0.7773,
+      "step": 463400
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.8274523686996154e-05,
+      "loss": 0.776,
+      "step": 463600
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.8265145716084275e-05,
+      "loss": 0.7731,
+      "step": 463800
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.8255767745172396e-05,
+      "loss": 0.7792,
+      "step": 464000
+    },
+    {
+      "epoch": 0.54,
+      "eval_em_first_line_ratio": 0.4878,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3748,
+      "eval_loss": 0.7459277510643005,
+      "eval_runtime": 199.2962,
+      "eval_samples_per_second": 50.177,
+      "eval_steps_per_second": 0.788,
+      "step": 464000
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.824638977426052e-05,
+      "loss": 0.7707,
+      "step": 464200
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.823701180334864e-05,
+      "loss": 0.7772,
+      "step": 464400
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.822768072229132e-05,
+      "loss": 0.7744,
+      "step": 464600
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.821830275137944e-05,
+      "loss": 0.7782,
+      "step": 464800
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.8208924780467562e-05,
+      "loss": 0.7786,
+      "step": 465000
+    },
+    {
+      "epoch": 0.55,
+      "eval_em_first_line_ratio": 0.4891,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3762,
+      "eval_loss": 0.7460950613021851,
+      "eval_runtime": 205.6857,
+      "eval_samples_per_second": 48.618,
+      "eval_steps_per_second": 0.763,
+      "step": 465000
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.8199546809555686e-05,
+      "loss": 0.7786,
+      "step": 465200
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.8190168838643807e-05,
+      "loss": 0.7749,
+      "step": 465400
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.8180837757586487e-05,
+      "loss": 0.7738,
+      "step": 465600
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.8171459786674608e-05,
+      "loss": 0.7769,
+      "step": 465800
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.816208181576273e-05,
+      "loss": 0.7665,
+      "step": 466000
+    },
+    {
+      "epoch": 0.55,
+      "eval_em_first_line_ratio": 0.4884,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3754,
+      "eval_loss": 0.7463697791099548,
+      "eval_runtime": 202.2638,
+      "eval_samples_per_second": 49.44,
+      "eval_steps_per_second": 0.776,
+      "step": 466000
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.815270384485085e-05,
+      "loss": 0.771,
+      "step": 466200
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.8143325873938974e-05,
+      "loss": 0.7708,
+      "step": 466400
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.8133947903027095e-05,
+      "loss": 0.7764,
+      "step": 466600
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.8124569932115216e-05,
+      "loss": 0.7748,
+      "step": 466800
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.8115191961203337e-05,
+      "loss": 0.7735,
+      "step": 467000
+    },
+    {
+      "epoch": 0.55,
+      "eval_em_first_line_ratio": 0.4889,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3759,
+      "eval_loss": 0.7458497285842896,
+      "eval_runtime": 200.4939,
+      "eval_samples_per_second": 49.877,
+      "eval_steps_per_second": 0.783,
+      "step": 467000
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.8105813990291458e-05,
+      "loss": 0.7731,
+      "step": 467200
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.809643601937958e-05,
+      "loss": 0.7729,
+      "step": 467400
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.80870580484677e-05,
+      "loss": 0.7811,
+      "step": 467600
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.807768007755582e-05,
+      "loss": 0.7716,
+      "step": 467800
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.806830210664394e-05,
+      "loss": 0.7723,
+      "step": 468000
+    },
+    {
+      "epoch": 0.55,
+      "eval_em_first_line_ratio": 0.4902,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.377,
+      "eval_loss": 0.7452747225761414,
+      "eval_runtime": 197.9399,
+      "eval_samples_per_second": 50.52,
+      "eval_steps_per_second": 0.793,
+      "step": 468000
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.8058924135732063e-05,
+      "loss": 0.7729,
+      "step": 468200
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.8049546164820184e-05,
+      "loss": 0.7766,
+      "step": 468400
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.8040168193908305e-05,
+      "loss": 0.774,
+      "step": 468600
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.803079022299643e-05,
+      "loss": 0.7767,
+      "step": 468800
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.802141225208455e-05,
+      "loss": 0.7775,
+      "step": 469000
+    },
+    {
+      "epoch": 0.55,
+      "eval_em_first_line_ratio": 0.4929,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3797,
+      "eval_loss": 0.7458990812301636,
+      "eval_runtime": 201.2646,
+      "eval_samples_per_second": 49.686,
+      "eval_steps_per_second": 0.78,
+      "step": 469000
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.801203428117267e-05,
+      "loss": 0.7722,
+      "step": 469200
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.8002656310260792e-05,
+      "loss": 0.7743,
+      "step": 469400
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.799332522920347e-05,
+      "loss": 0.7753,
+      "step": 469600
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.7983947258291595e-05,
+      "loss": 0.7741,
+      "step": 469800
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.7974569287379716e-05,
+      "loss": 0.7696,
+      "step": 470000
+    },
+    {
+      "epoch": 0.55,
+      "eval_em_first_line_ratio": 0.4892,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3762,
+      "eval_loss": 0.7457063794136047,
+      "eval_runtime": 205.7848,
+      "eval_samples_per_second": 48.594,
+      "eval_steps_per_second": 0.763,
+      "step": 470000
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.7965191316467837e-05,
+      "loss": 0.7721,
+      "step": 470200
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.7955813345555958e-05,
+      "loss": 0.771,
+      "step": 470400
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.7946482264498637e-05,
+      "loss": 0.7704,
+      "step": 470600
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.793710429358676e-05,
+      "loss": 0.7711,
+      "step": 470800
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.7927726322674883e-05,
+      "loss": 0.7703,
+      "step": 471000
+    },
+    {
+      "epoch": 0.55,
+      "eval_em_first_line_ratio": 0.489,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3763,
+      "eval_loss": 0.74530029296875,
+      "eval_runtime": 201.2189,
+      "eval_samples_per_second": 49.697,
+      "eval_steps_per_second": 0.78,
+      "step": 471000
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.7918348351763e-05,
+      "loss": 0.7737,
+      "step": 471200
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.790897038085112e-05,
+      "loss": 0.7805,
+      "step": 471400
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.7899592409939242e-05,
+      "loss": 0.7719,
+      "step": 471600
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.7890214439027363e-05,
+      "loss": 0.7723,
+      "step": 471800
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.7880836468115484e-05,
+      "loss": 0.7705,
+      "step": 472000
+    },
+    {
+      "epoch": 0.55,
+      "eval_em_first_line_ratio": 0.4897,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3766,
+      "eval_loss": 0.7454162836074829,
+      "eval_runtime": 199.8846,
+      "eval_samples_per_second": 50.029,
+      "eval_steps_per_second": 0.785,
+      "step": 472000
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.787145849720361e-05,
+      "loss": 0.7749,
+      "step": 472200
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.786208052629173e-05,
+      "loss": 0.775,
+      "step": 472400
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.785274944523441e-05,
+      "loss": 0.7722,
+      "step": 472600
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.784337147432253e-05,
+      "loss": 0.7767,
+      "step": 472800
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.783399350341065e-05,
+      "loss": 0.7729,
+      "step": 473000
+    },
+    {
+      "epoch": 0.55,
+      "eval_em_first_line_ratio": 0.4887,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3756,
+      "eval_loss": 0.7452238202095032,
+      "eval_runtime": 204.0702,
+      "eval_samples_per_second": 49.003,
+      "eval_steps_per_second": 0.769,
+      "step": 473000
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.7824615532498775e-05,
+      "loss": 0.7725,
+      "step": 473200
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.7815237561586896e-05,
+      "loss": 0.7677,
+      "step": 473400
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.7805859590675017e-05,
+      "loss": 0.7769,
+      "step": 473600
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.7796481619763138e-05,
+      "loss": 0.7693,
+      "step": 473800
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.778710364885126e-05,
+      "loss": 0.7748,
+      "step": 474000
+    },
+    {
+      "epoch": 0.56,
+      "eval_em_first_line_ratio": 0.4884,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3755,
+      "eval_loss": 0.7447240352630615,
+      "eval_runtime": 205.0412,
+      "eval_samples_per_second": 48.771,
+      "eval_steps_per_second": 0.766,
+      "step": 474000
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.777772567793938e-05,
+      "loss": 0.7718,
+      "step": 474200
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.77683477070275e-05,
+      "loss": 0.7701,
+      "step": 474400
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.7758969736115622e-05,
+      "loss": 0.7707,
+      "step": 474600
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.7749591765203743e-05,
+      "loss": 0.7717,
+      "step": 474800
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.7740213794291864e-05,
+      "loss": 0.7699,
+      "step": 475000
+    },
+    {
+      "epoch": 0.56,
+      "eval_em_first_line_ratio": 0.4889,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3763,
+      "eval_loss": 0.7445083856582642,
+      "eval_runtime": 204.378,
+      "eval_samples_per_second": 48.929,
+      "eval_steps_per_second": 0.768,
+      "step": 475000
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.7730835823379985e-05,
+      "loss": 0.7738,
+      "step": 475200
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.7721457852468106e-05,
+      "loss": 0.7711,
+      "step": 475400
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.7712079881556227e-05,
+      "loss": 0.7729,
+      "step": 475600
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.770270191064435e-05,
+      "loss": 0.7728,
+      "step": 475800
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.7693323939732472e-05,
+      "loss": 0.7741,
+      "step": 476000
+    },
+    {
+      "epoch": 0.56,
+      "eval_em_first_line_ratio": 0.4897,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.377,
+      "eval_loss": 0.7448092699050903,
+      "eval_runtime": 193.9121,
+      "eval_samples_per_second": 51.57,
+      "eval_steps_per_second": 0.81,
+      "step": 476000
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.7683945968820593e-05,
+      "loss": 0.7707,
+      "step": 476200
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.7674567997908714e-05,
+      "loss": 0.7746,
+      "step": 476400
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.7665190026996835e-05,
+      "loss": 0.7687,
+      "step": 476600
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.7655858945939517e-05,
+      "loss": 0.7692,
+      "step": 476800
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.764648097502764e-05,
+      "loss": 0.7787,
+      "step": 477000
+    },
+    {
+      "epoch": 0.56,
+      "eval_em_first_line_ratio": 0.4888,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3762,
+      "eval_loss": 0.7446808815002441,
+      "eval_runtime": 199.0703,
+      "eval_samples_per_second": 50.234,
+      "eval_steps_per_second": 0.789,
+      "step": 477000
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.7637149893970317e-05,
+      "loss": 0.7786,
+      "step": 477200
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.762777192305844e-05,
+      "loss": 0.7723,
+      "step": 477400
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.761839395214656e-05,
+      "loss": 0.7751,
+      "step": 477600
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.760901598123468e-05,
+      "loss": 0.7759,
+      "step": 477800
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.7599638010322805e-05,
+      "loss": 0.7753,
+      "step": 478000
+    },
+    {
+      "epoch": 0.56,
+      "eval_em_first_line_ratio": 0.4871,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3746,
+      "eval_loss": 0.7449805736541748,
+      "eval_runtime": 200.6923,
+      "eval_samples_per_second": 49.828,
+      "eval_steps_per_second": 0.782,
+      "step": 478000
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.7590260039410926e-05,
+      "loss": 0.7713,
+      "step": 478200
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.7580882068499047e-05,
+      "loss": 0.7752,
+      "step": 478400
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.7571504097587168e-05,
+      "loss": 0.7733,
+      "step": 478600
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.756212612667529e-05,
+      "loss": 0.7734,
+      "step": 478800
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.755274815576341e-05,
+      "loss": 0.7691,
+      "step": 479000
+    },
+    {
+      "epoch": 0.56,
+      "eval_em_first_line_ratio": 0.4882,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3755,
+      "eval_loss": 0.7455112934112549,
+      "eval_runtime": 198.1195,
+      "eval_samples_per_second": 50.475,
+      "eval_steps_per_second": 0.792,
+      "step": 479000
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.754337018485153e-05,
+      "loss": 0.774,
+      "step": 479200
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.753399221393965e-05,
+      "loss": 0.7675,
+      "step": 479400
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.7524614243027773e-05,
+      "loss": 0.7664,
+      "step": 479600
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.7515236272115893e-05,
+      "loss": 0.7716,
+      "step": 479800
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.7505858301204014e-05,
+      "loss": 0.7726,
+      "step": 480000
+    },
+    {
+      "epoch": 0.56,
+      "eval_em_first_line_ratio": 0.4884,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3762,
+      "eval_loss": 0.7448305487632751,
+      "eval_runtime": 204.017,
+      "eval_samples_per_second": 49.016,
+      "eval_steps_per_second": 0.77,
+      "step": 480000
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.7496480330292135e-05,
+      "loss": 0.7732,
+      "step": 480200
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.748710235938026e-05,
+      "loss": 0.7703,
+      "step": 480400
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.747772438846838e-05,
+      "loss": 0.7747,
+      "step": 480600
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.7468346417556502e-05,
+      "loss": 0.7732,
+      "step": 480800
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.7458968446644623e-05,
+      "loss": 0.7669,
+      "step": 481000
+    },
+    {
+      "epoch": 0.56,
+      "eval_em_first_line_ratio": 0.489,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3763,
+      "eval_loss": 0.7453780770301819,
+      "eval_runtime": 208.5077,
+      "eval_samples_per_second": 47.96,
+      "eval_steps_per_second": 0.753,
+      "step": 481000
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.7449590475732744e-05,
+      "loss": 0.7772,
+      "step": 481200
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.7440212504820865e-05,
+      "loss": 0.7724,
+      "step": 481400
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.7430881423763547e-05,
+      "loss": 0.7779,
+      "step": 481600
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.7421503452851668e-05,
+      "loss": 0.7713,
+      "step": 481800
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.741212548193979e-05,
+      "loss": 0.7724,
+      "step": 482000
+    },
+    {
+      "epoch": 0.56,
+      "eval_em_first_line_ratio": 0.4892,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3764,
+      "eval_loss": 0.7443947792053223,
+      "eval_runtime": 203.6747,
+      "eval_samples_per_second": 49.098,
+      "eval_steps_per_second": 0.771,
+      "step": 482000
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.740274751102791e-05,
+      "loss": 0.7738,
+      "step": 482200
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.739336954011603e-05,
+      "loss": 0.7697,
+      "step": 482400
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.7383991569204152e-05,
+      "loss": 0.771,
+      "step": 482600
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.7374613598292273e-05,
+      "loss": 0.7732,
+      "step": 482800
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.7365235627380394e-05,
+      "loss": 0.7736,
+      "step": 483000
+    },
+    {
+      "epoch": 0.57,
+      "eval_em_first_line_ratio": 0.488,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3753,
+      "eval_loss": 0.7442800998687744,
+      "eval_runtime": 199.5252,
+      "eval_samples_per_second": 50.119,
+      "eval_steps_per_second": 0.787,
+      "step": 483000
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.7355857656468515e-05,
+      "loss": 0.7658,
+      "step": 483200
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.7346479685556636e-05,
+      "loss": 0.7748,
+      "step": 483400
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.7337148604499315e-05,
+      "loss": 0.7681,
+      "step": 483600
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.732777063358744e-05,
+      "loss": 0.7698,
+      "step": 483800
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.731839266267556e-05,
+      "loss": 0.7765,
+      "step": 484000
+    },
+    {
+      "epoch": 0.57,
+      "eval_em_first_line_ratio": 0.4882,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3755,
+      "eval_loss": 0.7443733215332031,
+      "eval_runtime": 207.1039,
+      "eval_samples_per_second": 48.285,
+      "eval_steps_per_second": 0.758,
+      "step": 484000
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.730901469176368e-05,
+      "loss": 0.7719,
+      "step": 484200
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.7299636720851802e-05,
+      "loss": 0.7732,
+      "step": 484400
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.7290258749939923e-05,
+      "loss": 0.7661,
+      "step": 484600
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.7280880779028044e-05,
+      "loss": 0.7737,
+      "step": 484800
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.727150280811617e-05,
+      "loss": 0.7753,
+      "step": 485000
+    },
+    {
+      "epoch": 0.57,
+      "eval_em_first_line_ratio": 0.4895,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3768,
+      "eval_loss": 0.7447143793106079,
+      "eval_runtime": 206.6158,
+      "eval_samples_per_second": 48.399,
+      "eval_steps_per_second": 0.76,
+      "step": 485000
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.726212483720429e-05,
+      "loss": 0.7677,
+      "step": 485200
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.725274686629241e-05,
+      "loss": 0.7724,
+      "step": 485400
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.724341578523509e-05,
+      "loss": 0.7707,
+      "step": 485600
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.723403781432321e-05,
+      "loss": 0.7687,
+      "step": 485800
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.722465984341133e-05,
+      "loss": 0.7717,
+      "step": 486000
+    },
+    {
+      "epoch": 0.57,
+      "eval_em_first_line_ratio": 0.4884,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3759,
+      "eval_loss": 0.7443506717681885,
+      "eval_runtime": 202.8181,
+      "eval_samples_per_second": 49.305,
+      "eval_steps_per_second": 0.774,
+      "step": 486000
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.7215281872499453e-05,
+      "loss": 0.7689,
+      "step": 486200
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.7205903901587574e-05,
+      "loss": 0.7705,
+      "step": 486400
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.7196525930675695e-05,
+      "loss": 0.7728,
+      "step": 486600
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.7187147959763816e-05,
+      "loss": 0.7703,
+      "step": 486800
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.7177769988851936e-05,
+      "loss": 0.7686,
+      "step": 487000
+    },
+    {
+      "epoch": 0.57,
+      "eval_em_first_line_ratio": 0.4889,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3764,
+      "eval_loss": 0.7437922954559326,
+      "eval_runtime": 203.5751,
+      "eval_samples_per_second": 49.122,
+      "eval_steps_per_second": 0.771,
+      "step": 487000
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.7168392017940057e-05,
+      "loss": 0.7683,
+      "step": 487200
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.7159014047028182e-05,
+      "loss": 0.7731,
+      "step": 487400
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.7149636076116303e-05,
+      "loss": 0.7726,
+      "step": 487600
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.7140258105204424e-05,
+      "loss": 0.7712,
+      "step": 487800
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.7130927024147103e-05,
+      "loss": 0.7699,
+      "step": 488000
+    },
+    {
+      "epoch": 0.57,
+      "eval_em_first_line_ratio": 0.4898,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3767,
+      "eval_loss": 0.7439884543418884,
+      "eval_runtime": 200.4279,
+      "eval_samples_per_second": 49.893,
+      "eval_steps_per_second": 0.783,
+      "step": 488000
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.7121549053235224e-05,
+      "loss": 0.7695,
+      "step": 488200
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.7112171082323348e-05,
+      "loss": 0.7658,
+      "step": 488400
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.710279311141147e-05,
+      "loss": 0.7668,
+      "step": 488600
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.709341514049959e-05,
+      "loss": 0.7736,
+      "step": 488800
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.708403716958771e-05,
+      "loss": 0.7656,
+      "step": 489000
+    },
+    {
+      "epoch": 0.57,
+      "eval_em_first_line_ratio": 0.4896,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3767,
+      "eval_loss": 0.7443256974220276,
+      "eval_runtime": 198.3044,
+      "eval_samples_per_second": 50.428,
+      "eval_steps_per_second": 0.792,
+      "step": 489000
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.7074659198675832e-05,
+      "loss": 0.7757,
+      "step": 489200
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.7065281227763953e-05,
+      "loss": 0.7689,
+      "step": 489400
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.7055903256852074e-05,
+      "loss": 0.7687,
+      "step": 489600
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.7046525285940195e-05,
+      "loss": 0.7694,
+      "step": 489800
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.7037147315028316e-05,
+      "loss": 0.7703,
+      "step": 490000
+    },
+    {
+      "epoch": 0.57,
+      "eval_em_first_line_ratio": 0.4904,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3768,
+      "eval_loss": 0.7443216443061829,
+      "eval_runtime": 195.7036,
+      "eval_samples_per_second": 51.098,
+      "eval_steps_per_second": 0.802,
+      "step": 490000
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.7027769344116437e-05,
+      "loss": 0.7669,
+      "step": 490200
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.7018391373204558e-05,
+      "loss": 0.7732,
+      "step": 490400
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.700901340229268e-05,
+      "loss": 0.771,
+      "step": 490600
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.6999635431380803e-05,
+      "loss": 0.7727,
+      "step": 490800
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.6990257460468924e-05,
+      "loss": 0.7709,
+      "step": 491000
+    },
+    {
+      "epoch": 0.58,
+      "eval_em_first_line_ratio": 0.4876,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3751,
+      "eval_loss": 0.7441647052764893,
+      "eval_runtime": 203.8682,
+      "eval_samples_per_second": 49.051,
+      "eval_steps_per_second": 0.77,
+      "step": 491000
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.6980879489557045e-05,
+      "loss": 0.775,
+      "step": 491200
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.6971501518645166e-05,
+      "loss": 0.7717,
+      "step": 491400
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.6962170437587845e-05,
+      "loss": 0.7706,
+      "step": 491600
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.6952792466675966e-05,
+      "loss": 0.7732,
+      "step": 491800
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.694341449576409e-05,
+      "loss": 0.7677,
+      "step": 492000
+    },
+    {
+      "epoch": 0.58,
+      "eval_em_first_line_ratio": 0.4889,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3756,
+      "eval_loss": 0.743827760219574,
+      "eval_runtime": 198.5695,
+      "eval_samples_per_second": 50.36,
+      "eval_steps_per_second": 0.791,
+      "step": 492000
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.693403652485221e-05,
+      "loss": 0.772,
+      "step": 492200
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.6924658553940333e-05,
+      "loss": 0.7722,
+      "step": 492400
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.6915280583028454e-05,
+      "loss": 0.772,
+      "step": 492600
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.6905902612116575e-05,
+      "loss": 0.7661,
+      "step": 492800
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.6896524641204696e-05,
+      "loss": 0.7733,
+      "step": 493000
+    },
+    {
+      "epoch": 0.58,
+      "eval_em_first_line_ratio": 0.4896,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3769,
+      "eval_loss": 0.7429953217506409,
+      "eval_runtime": 202.0204,
+      "eval_samples_per_second": 49.5,
+      "eval_steps_per_second": 0.777,
+      "step": 493000
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.6887146670292816e-05,
+      "loss": 0.7768,
+      "step": 493200
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.68778155892355e-05,
+      "loss": 0.7708,
+      "step": 493400
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.686843761832362e-05,
+      "loss": 0.7621,
+      "step": 493600
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.685905964741174e-05,
+      "loss": 0.7779,
+      "step": 493800
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.6849681676499862e-05,
+      "loss": 0.7706,
+      "step": 494000
+    },
+    {
+      "epoch": 0.58,
+      "eval_em_first_line_ratio": 0.4904,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3774,
+      "eval_loss": 0.7427064180374146,
+      "eval_runtime": 203.1744,
+      "eval_samples_per_second": 49.219,
+      "eval_steps_per_second": 0.773,
+      "step": 494000
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.6840303705587983e-05,
+      "loss": 0.7707,
+      "step": 494200
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.6830925734676104e-05,
+      "loss": 0.77,
+      "step": 494400
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.6821547763764225e-05,
+      "loss": 0.7693,
+      "step": 494600
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.6812169792852346e-05,
+      "loss": 0.7685,
+      "step": 494800
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.6802791821940467e-05,
+      "loss": 0.7683,
+      "step": 495000
+    },
+    {
+      "epoch": 0.58,
+      "eval_em_first_line_ratio": 0.491,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3779,
+      "eval_loss": 0.7431945204734802,
+      "eval_runtime": 199.4092,
+      "eval_samples_per_second": 50.148,
+      "eval_steps_per_second": 0.787,
+      "step": 495000
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.6793413851028588e-05,
+      "loss": 0.7687,
+      "step": 495200
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.678403588011671e-05,
+      "loss": 0.7684,
+      "step": 495400
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.6774657909204833e-05,
+      "loss": 0.7662,
+      "step": 495600
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.6765279938292954e-05,
+      "loss": 0.7685,
+      "step": 495800
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.6755901967381075e-05,
+      "loss": 0.7711,
+      "step": 496000
+    },
+    {
+      "epoch": 0.58,
+      "eval_em_first_line_ratio": 0.4896,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3766,
+      "eval_loss": 0.7432389259338379,
+      "eval_runtime": 200.853,
+      "eval_samples_per_second": 49.788,
+      "eval_steps_per_second": 0.782,
+      "step": 496000
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.6746523996469196e-05,
+      "loss": 0.7711,
+      "step": 496200
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.6737146025557317e-05,
+      "loss": 0.7675,
+      "step": 496400
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.6727768054645438e-05,
+      "loss": 0.7692,
+      "step": 496600
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.671839008373356e-05,
+      "loss": 0.7678,
+      "step": 496800
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.670901211282168e-05,
+      "loss": 0.7722,
+      "step": 497000
+    },
+    {
+      "epoch": 0.58,
+      "eval_em_first_line_ratio": 0.4897,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3766,
+      "eval_loss": 0.7428988218307495,
+      "eval_runtime": 204.5582,
+      "eval_samples_per_second": 48.886,
+      "eval_steps_per_second": 0.768,
+      "step": 497000
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.66996341419098e-05,
+      "loss": 0.7779,
+      "step": 497200
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.6690256170997922e-05,
+      "loss": 0.766,
+      "step": 497400
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.6680878200086043e-05,
+      "loss": 0.7705,
+      "step": 497600
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.6671500229174164e-05,
+      "loss": 0.7725,
+      "step": 497800
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.6662122258262288e-05,
+      "loss": 0.7689,
+      "step": 498000
+    },
+    {
+      "epoch": 0.58,
+      "eval_em_first_line_ratio": 0.4906,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3776,
+      "eval_loss": 0.7431126236915588,
+      "eval_runtime": 203.6138,
+      "eval_samples_per_second": 49.113,
+      "eval_steps_per_second": 0.771,
+      "step": 498000
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.665274428735041e-05,
+      "loss": 0.7684,
+      "step": 498200
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.664336631643853e-05,
+      "loss": 0.766,
+      "step": 498400
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.663398834552665e-05,
+      "loss": 0.772,
+      "step": 498600
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.6624610374614772e-05,
+      "loss": 0.7698,
+      "step": 498800
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.6615232403702893e-05,
+      "loss": 0.7725,
+      "step": 499000
+    },
+    {
+      "epoch": 0.58,
+      "eval_em_first_line_ratio": 0.492,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3787,
+      "eval_loss": 0.7427988648414612,
+      "eval_runtime": 194.8516,
+      "eval_samples_per_second": 51.321,
+      "eval_steps_per_second": 0.806,
+      "step": 499000
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.6605854432791014e-05,
+      "loss": 0.7677,
+      "step": 499200
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.6596476461879135e-05,
+      "loss": 0.773,
+      "step": 499400
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.6587145380821817e-05,
+      "loss": 0.7701,
+      "step": 499600
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.6577814299764497e-05,
+      "loss": 0.7722,
+      "step": 499800
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.656848321870718e-05,
+      "loss": 0.7682,
+      "step": 500000
+    },
+    {
+      "epoch": 0.59,
+      "eval_em_first_line_ratio": 0.4903,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3771,
+      "eval_loss": 0.7428056001663208,
+      "eval_runtime": 200.1081,
+      "eval_samples_per_second": 49.973,
+      "eval_steps_per_second": 0.785,
+      "step": 500000
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.65591052477953e-05,
+      "loss": 0.7716,
+      "step": 500200
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.654972727688342e-05,
+      "loss": 0.771,
+      "step": 500400
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.6540349305971542e-05,
+      "loss": 0.7687,
+      "step": 500600
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.6530971335059663e-05,
+      "loss": 0.7701,
+      "step": 500800
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.6521593364147784e-05,
+      "loss": 0.7693,
+      "step": 501000
+    },
+    {
+      "epoch": 0.59,
+      "eval_em_first_line_ratio": 0.4912,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3782,
+      "eval_loss": 0.7422392964363098,
+      "eval_runtime": 203.6275,
+      "eval_samples_per_second": 49.109,
+      "eval_steps_per_second": 0.771,
+      "step": 501000
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.6512215393235905e-05,
+      "loss": 0.7688,
+      "step": 501200
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.6502837422324026e-05,
+      "loss": 0.7642,
+      "step": 501400
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.6493459451412147e-05,
+      "loss": 0.77,
+      "step": 501600
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.6484081480500268e-05,
+      "loss": 0.772,
+      "step": 501800
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.647470350958839e-05,
+      "loss": 0.7697,
+      "step": 502000
+    },
+    {
+      "epoch": 0.59,
+      "eval_em_first_line_ratio": 0.4926,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3798,
+      "eval_loss": 0.7418246865272522,
+      "eval_runtime": 199.2923,
+      "eval_samples_per_second": 50.178,
+      "eval_steps_per_second": 0.788,
+      "step": 502000
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.646532553867651e-05,
+      "loss": 0.7704,
+      "step": 502200
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.6455947567764634e-05,
+      "loss": 0.7687,
+      "step": 502400
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.6446569596852755e-05,
+      "loss": 0.7696,
+      "step": 502600
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.6437238515795434e-05,
+      "loss": 0.7708,
+      "step": 502800
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.6427860544883555e-05,
+      "loss": 0.7688,
+      "step": 503000
+    },
+    {
+      "epoch": 0.59,
+      "eval_em_first_line_ratio": 0.4907,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3773,
+      "eval_loss": 0.7416971921920776,
+      "eval_runtime": 208.8594,
+      "eval_samples_per_second": 47.879,
+      "eval_steps_per_second": 0.752,
+      "step": 503000
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.6418482573971676e-05,
+      "loss": 0.7772,
+      "step": 503200
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.6409104603059797e-05,
+      "loss": 0.773,
+      "step": 503400
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.639972663214792e-05,
+      "loss": 0.7686,
+      "step": 503600
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.63903955510906e-05,
+      "loss": 0.7738,
+      "step": 503800
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.638101758017872e-05,
+      "loss": 0.7773,
+      "step": 504000
+    },
+    {
+      "epoch": 0.59,
+      "eval_em_first_line_ratio": 0.4908,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3777,
+      "eval_loss": 0.7416525483131409,
+      "eval_runtime": 203.7295,
+      "eval_samples_per_second": 49.085,
+      "eval_steps_per_second": 0.771,
+      "step": 504000
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.6371639609266843e-05,
+      "loss": 0.7713,
+      "step": 504200
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.6362261638354964e-05,
+      "loss": 0.7679,
+      "step": 504400
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.6352883667443088e-05,
+      "loss": 0.7703,
+      "step": 504600
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.634350569653121e-05,
+      "loss": 0.7677,
+      "step": 504800
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.633412772561933e-05,
+      "loss": 0.768,
+      "step": 505000
+    },
+    {
+      "epoch": 0.59,
+      "eval_em_first_line_ratio": 0.4898,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3769,
+      "eval_loss": 0.7418670654296875,
+      "eval_runtime": 208.8487,
+      "eval_samples_per_second": 47.882,
+      "eval_steps_per_second": 0.752,
+      "step": 505000
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.632474975470745e-05,
+      "loss": 0.7741,
+      "step": 505200
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.6315371783795572e-05,
+      "loss": 0.7731,
+      "step": 505400
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.6305993812883693e-05,
+      "loss": 0.7686,
+      "step": 505600
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.6296615841971814e-05,
+      "loss": 0.7728,
+      "step": 505800
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.6287237871059935e-05,
+      "loss": 0.7696,
+      "step": 506000
+    },
+    {
+      "epoch": 0.59,
+      "eval_em_first_line_ratio": 0.4887,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3755,
+      "eval_loss": 0.7417634725570679,
+      "eval_runtime": 202.6398,
+      "eval_samples_per_second": 49.349,
+      "eval_steps_per_second": 0.775,
+      "step": 506000
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.6277859900148056e-05,
+      "loss": 0.7691,
+      "step": 506200
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.6268481929236177e-05,
+      "loss": 0.7709,
+      "step": 506400
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.6259103958324298e-05,
+      "loss": 0.7725,
+      "step": 506600
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.624972598741242e-05,
+      "loss": 0.7642,
+      "step": 506800
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.624034801650054e-05,
+      "loss": 0.7655,
+      "step": 507000
+    },
+    {
+      "epoch": 0.59,
+      "eval_em_first_line_ratio": 0.4919,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3785,
+      "eval_loss": 0.7420018315315247,
+      "eval_runtime": 196.7902,
+      "eval_samples_per_second": 50.816,
+      "eval_steps_per_second": 0.798,
+      "step": 507000
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.6230970045588664e-05,
+      "loss": 0.7672,
+      "step": 507200
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.6221592074676785e-05,
+      "loss": 0.7683,
+      "step": 507400
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.6212214103764906e-05,
+      "loss": 0.7701,
+      "step": 507600
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.6202836132853027e-05,
+      "loss": 0.7699,
+      "step": 507800
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.6193458161941148e-05,
+      "loss": 0.772,
+      "step": 508000
+    },
+    {
+      "epoch": 0.6,
+      "eval_em_first_line_ratio": 0.4899,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3768,
+      "eval_loss": 0.742087721824646,
+      "eval_runtime": 203.5115,
+      "eval_samples_per_second": 49.137,
+      "eval_steps_per_second": 0.771,
+      "step": 508000
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.618408019102927e-05,
+      "loss": 0.7715,
+      "step": 508200
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.617470222011739e-05,
+      "loss": 0.7672,
+      "step": 508400
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.616532424920551e-05,
+      "loss": 0.7643,
+      "step": 508600
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.6155946278293632e-05,
+      "loss": 0.7708,
+      "step": 508800
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.6146568307381753e-05,
+      "loss": 0.7659,
+      "step": 509000
+    },
+    {
+      "epoch": 0.6,
+      "eval_em_first_line_ratio": 0.4893,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3762,
+      "eval_loss": 0.7424294948577881,
+      "eval_runtime": 203.0867,
+      "eval_samples_per_second": 49.24,
+      "eval_steps_per_second": 0.773,
+      "step": 509000
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.6137190336469874e-05,
+      "loss": 0.7721,
+      "step": 509200
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.6127812365557995e-05,
+      "loss": 0.7717,
+      "step": 509400
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.611843439464612e-05,
+      "loss": 0.7674,
+      "step": 509600
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.610905642373424e-05,
+      "loss": 0.7688,
+      "step": 509800
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.609972534267692e-05,
+      "loss": 0.7748,
+      "step": 510000
+    },
+    {
+      "epoch": 0.6,
+      "eval_em_first_line_ratio": 0.4882,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3753,
+      "eval_loss": 0.7416215538978577,
+      "eval_runtime": 200.5666,
+      "eval_samples_per_second": 49.859,
+      "eval_steps_per_second": 0.783,
+      "step": 510000
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.609034737176504e-05,
+      "loss": 0.7704,
+      "step": 510200
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.608096940085316e-05,
+      "loss": 0.7717,
+      "step": 510400
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.6071591429941285e-05,
+      "loss": 0.7678,
+      "step": 510600
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.6062213459029406e-05,
+      "loss": 0.7671,
+      "step": 510800
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.6052835488117527e-05,
+      "loss": 0.7655,
+      "step": 511000
+    },
+    {
+      "epoch": 0.6,
+      "eval_em_first_line_ratio": 0.4889,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3763,
+      "eval_loss": 0.7411304712295532,
+      "eval_runtime": 200.2449,
+      "eval_samples_per_second": 49.939,
+      "eval_steps_per_second": 0.784,
+      "step": 511000
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.6043504407060206e-05,
+      "loss": 0.7706,
+      "step": 511200
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.6034126436148327e-05,
+      "loss": 0.7651,
+      "step": 511400
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.602474846523645e-05,
+      "loss": 0.7693,
+      "step": 511600
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.6015370494324573e-05,
+      "loss": 0.7711,
+      "step": 511800
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.6005992523412694e-05,
+      "loss": 0.769,
+      "step": 512000
+    },
+    {
+      "epoch": 0.6,
+      "eval_em_first_line_ratio": 0.4909,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3777,
+      "eval_loss": 0.7412983179092407,
+      "eval_runtime": 204.745,
+      "eval_samples_per_second": 48.841,
+      "eval_steps_per_second": 0.767,
+      "step": 512000
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.5996614552500815e-05,
+      "loss": 0.7709,
+      "step": 512200
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.5987236581588936e-05,
+      "loss": 0.7705,
+      "step": 512400
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.5977858610677057e-05,
+      "loss": 0.7624,
+      "step": 512600
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.5968527529619736e-05,
+      "loss": 0.7675,
+      "step": 512800
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.5959149558707857e-05,
+      "loss": 0.7706,
+      "step": 513000
+    },
+    {
+      "epoch": 0.6,
+      "eval_em_first_line_ratio": 0.4904,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3774,
+      "eval_loss": 0.7406914830207825,
+      "eval_runtime": 200.0396,
+      "eval_samples_per_second": 49.99,
+      "eval_steps_per_second": 0.785,
+      "step": 513000
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.5949771587795978e-05,
+      "loss": 0.7718,
+      "step": 513200
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.59403936168841e-05,
+      "loss": 0.7667,
+      "step": 513400
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.593101564597222e-05,
+      "loss": 0.7681,
+      "step": 513600
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.592163767506034e-05,
+      "loss": 0.774,
+      "step": 513800
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.5912259704148465e-05,
+      "loss": 0.7634,
+      "step": 514000
+    },
+    {
+      "epoch": 0.6,
+      "eval_em_first_line_ratio": 0.4898,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3776,
+      "eval_loss": 0.7412101626396179,
+      "eval_runtime": 204.4251,
+      "eval_samples_per_second": 48.918,
+      "eval_steps_per_second": 0.768,
+      "step": 514000
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.5902881733236586e-05,
+      "loss": 0.7726,
+      "step": 514200
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.5893503762324707e-05,
+      "loss": 0.7634,
+      "step": 514400
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.5884125791412828e-05,
+      "loss": 0.7618,
+      "step": 514600
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.587474782050095e-05,
+      "loss": 0.7719,
+      "step": 514800
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.586536984958907e-05,
+      "loss": 0.7737,
+      "step": 515000
+    },
+    {
+      "epoch": 0.6,
+      "eval_em_first_line_ratio": 0.4894,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3761,
+      "eval_loss": 0.7411162853240967,
+      "eval_runtime": 200.4384,
+      "eval_samples_per_second": 49.891,
+      "eval_steps_per_second": 0.783,
+      "step": 515000
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.5855991878677194e-05,
+      "loss": 0.7698,
+      "step": 515200
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.5846613907765315e-05,
+      "loss": 0.7648,
+      "step": 515400
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.5837235936853436e-05,
+      "loss": 0.7704,
+      "step": 515600
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.5827857965941554e-05,
+      "loss": 0.7698,
+      "step": 515800
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.5818479995029675e-05,
+      "loss": 0.7701,
+      "step": 516000
+    },
+    {
+      "epoch": 0.6,
+      "eval_em_first_line_ratio": 0.4878,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3752,
+      "eval_loss": 0.7409109473228455,
+      "eval_runtime": 204.4906,
+      "eval_samples_per_second": 48.902,
+      "eval_steps_per_second": 0.768,
+      "step": 516000
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.5809102024117796e-05,
+      "loss": 0.7645,
+      "step": 516200
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.5799724053205917e-05,
+      "loss": 0.7648,
+      "step": 516400
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.579034608229404e-05,
+      "loss": 0.7656,
+      "step": 516600
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.5780968111382162e-05,
+      "loss": 0.7679,
+      "step": 516800
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.5771590140470283e-05,
+      "loss": 0.769,
+      "step": 517000
+    },
+    {
+      "epoch": 0.61,
+      "eval_em_first_line_ratio": 0.4886,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3754,
+      "eval_loss": 0.741053581237793,
+      "eval_runtime": 197.67,
+      "eval_samples_per_second": 50.589,
+      "eval_steps_per_second": 0.794,
+      "step": 517000
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.5762212169558404e-05,
+      "loss": 0.7727,
+      "step": 517200
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.5752834198646525e-05,
+      "loss": 0.7617,
+      "step": 517400
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.5743456227734646e-05,
+      "loss": 0.7703,
+      "step": 517600
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.573407825682277e-05,
+      "loss": 0.7693,
+      "step": 517800
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.572470028591089e-05,
+      "loss": 0.7681,
+      "step": 518000
+    },
+    {
+      "epoch": 0.61,
+      "eval_em_first_line_ratio": 0.4904,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.377,
+      "eval_loss": 0.7415217757225037,
+      "eval_runtime": 197.0304,
+      "eval_samples_per_second": 50.754,
+      "eval_steps_per_second": 0.797,
+      "step": 518000
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.5715322314999012e-05,
+      "loss": 0.7738,
+      "step": 518200
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.5705944344087133e-05,
+      "loss": 0.7708,
+      "step": 518400
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.5696566373175254e-05,
+      "loss": 0.7629,
+      "step": 518600
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.5687188402263375e-05,
+      "loss": 0.7647,
+      "step": 518800
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.5677810431351496e-05,
+      "loss": 0.7703,
+      "step": 519000
+    },
+    {
+      "epoch": 0.61,
+      "eval_em_first_line_ratio": 0.4895,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3764,
+      "eval_loss": 0.7408283352851868,
+      "eval_runtime": 204.7775,
+      "eval_samples_per_second": 48.833,
+      "eval_steps_per_second": 0.767,
+      "step": 519000
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.5668432460439617e-05,
+      "loss": 0.7642,
+      "step": 519200
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.5659054489527738e-05,
+      "loss": 0.765,
+      "step": 519400
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.564967651861586e-05,
+      "loss": 0.7692,
+      "step": 519600
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.564029854770398e-05,
+      "loss": 0.7708,
+      "step": 519800
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.56309205767921e-05,
+      "loss": 0.7657,
+      "step": 520000
+    },
+    {
+      "epoch": 0.61,
+      "eval_em_first_line_ratio": 0.4894,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3761,
+      "eval_loss": 0.7412403225898743,
+      "eval_runtime": 206.5796,
+      "eval_samples_per_second": 48.407,
+      "eval_steps_per_second": 0.76,
+      "step": 520000
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.5621542605880225e-05,
+      "loss": 0.7701,
+      "step": 520200
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.5612211524822904e-05,
+      "loss": 0.7681,
+      "step": 520400
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.5602833553911025e-05,
+      "loss": 0.768,
+      "step": 520600
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.5593455582999146e-05,
+      "loss": 0.7671,
+      "step": 520800
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.5584077612087267e-05,
+      "loss": 0.7666,
+      "step": 521000
+    },
+    {
+      "epoch": 0.61,
+      "eval_em_first_line_ratio": 0.4911,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3783,
+      "eval_loss": 0.74056077003479,
+      "eval_runtime": 202.1943,
+      "eval_samples_per_second": 49.457,
+      "eval_steps_per_second": 0.776,
+      "step": 521000
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.5574699641175392e-05,
+      "loss": 0.7676,
+      "step": 521200
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.5565321670263513e-05,
+      "loss": 0.7674,
+      "step": 521400
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.5555943699351634e-05,
+      "loss": 0.7653,
+      "step": 521600
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.5546565728439755e-05,
+      "loss": 0.7675,
+      "step": 521800
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.5537187757527876e-05,
+      "loss": 0.7694,
+      "step": 522000
+    },
+    {
+      "epoch": 0.61,
+      "eval_em_first_line_ratio": 0.4895,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3766,
+      "eval_loss": 0.7399112582206726,
+      "eval_runtime": 206.0471,
+      "eval_samples_per_second": 48.533,
+      "eval_steps_per_second": 0.762,
+      "step": 522000
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.5527809786615997e-05,
+      "loss": 0.7662,
+      "step": 522200
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.5518431815704114e-05,
+      "loss": 0.7684,
+      "step": 522400
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.550905384479224e-05,
+      "loss": 0.7718,
+      "step": 522600
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.549967587388036e-05,
+      "loss": 0.7682,
+      "step": 522800
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.549029790296848e-05,
+      "loss": 0.7658,
+      "step": 523000
+    },
+    {
+      "epoch": 0.61,
+      "eval_em_first_line_ratio": 0.4912,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3783,
+      "eval_loss": 0.7406296133995056,
+      "eval_runtime": 203.0944,
+      "eval_samples_per_second": 49.238,
+      "eval_steps_per_second": 0.773,
+      "step": 523000
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.54809199320566e-05,
+      "loss": 0.7688,
+      "step": 523200
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.5471541961144722e-05,
+      "loss": 0.7644,
+      "step": 523400
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.5462163990232843e-05,
+      "loss": 0.7716,
+      "step": 523600
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.5452832909175526e-05,
+      "loss": 0.7645,
+      "step": 523800
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.5443454938263647e-05,
+      "loss": 0.7653,
+      "step": 524000
+    },
+    {
+      "epoch": 0.61,
+      "eval_em_first_line_ratio": 0.4909,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3774,
+      "eval_loss": 0.7397486567497253,
+      "eval_runtime": 199.4293,
+      "eval_samples_per_second": 50.143,
+      "eval_steps_per_second": 0.787,
+      "step": 524000
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.5434076967351768e-05,
+      "loss": 0.7673,
+      "step": 524200
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.542469899643989e-05,
+      "loss": 0.7691,
+      "step": 524400
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.541532102552801e-05,
+      "loss": 0.7683,
+      "step": 524600
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.5405943054616134e-05,
+      "loss": 0.7654,
+      "step": 524800
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.5396565083704255e-05,
+      "loss": 0.7685,
+      "step": 525000
+    },
+    {
+      "epoch": 0.62,
+      "eval_em_first_line_ratio": 0.4873,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3745,
+      "eval_loss": 0.7394555807113647,
+      "eval_runtime": 204.0956,
+      "eval_samples_per_second": 48.997,
+      "eval_steps_per_second": 0.769,
+      "step": 525000
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.5387187112792376e-05,
+      "loss": 0.7653,
+      "step": 525200
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.5377809141880494e-05,
+      "loss": 0.7668,
+      "step": 525400
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.5368431170968615e-05,
+      "loss": 0.7664,
+      "step": 525600
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.5359053200056736e-05,
+      "loss": 0.7717,
+      "step": 525800
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.534967522914486e-05,
+      "loss": 0.7642,
+      "step": 526000
+    },
+    {
+      "epoch": 0.62,
+      "eval_em_first_line_ratio": 0.489,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.376,
+      "eval_loss": 0.7405940294265747,
+      "eval_runtime": 195.6556,
+      "eval_samples_per_second": 51.11,
+      "eval_steps_per_second": 0.802,
+      "step": 526000
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.534029725823298e-05,
+      "loss": 0.7729,
+      "step": 526200
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.5330919287321102e-05,
+      "loss": 0.7655,
+      "step": 526400
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.532158820626378e-05,
+      "loss": 0.7646,
+      "step": 526600
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.5312210235351902e-05,
+      "loss": 0.7687,
+      "step": 526800
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.5302832264440023e-05,
+      "loss": 0.7659,
+      "step": 527000
+    },
+    {
+      "epoch": 0.62,
+      "eval_em_first_line_ratio": 0.4903,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3771,
+      "eval_loss": 0.7399801015853882,
+      "eval_runtime": 202.9885,
+      "eval_samples_per_second": 49.264,
+      "eval_steps_per_second": 0.773,
+      "step": 527000
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.5293454293528147e-05,
+      "loss": 0.7669,
+      "step": 527200
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.528407632261627e-05,
+      "loss": 0.7665,
+      "step": 527400
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.527469835170439e-05,
+      "loss": 0.7621,
+      "step": 527600
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.526532038079251e-05,
+      "loss": 0.7641,
+      "step": 527800
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.5255942409880631e-05,
+      "loss": 0.7624,
+      "step": 528000
+    },
+    {
+      "epoch": 0.62,
+      "eval_em_first_line_ratio": 0.4913,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.378,
+      "eval_loss": 0.7392761707305908,
+      "eval_runtime": 204.733,
+      "eval_samples_per_second": 48.844,
+      "eval_steps_per_second": 0.767,
+      "step": 528000
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.5246611328823314e-05,
+      "loss": 0.7632,
+      "step": 528200
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.5237280247765993e-05,
+      "loss": 0.7681,
+      "step": 528400
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.5227902276854114e-05,
+      "loss": 0.7657,
+      "step": 528600
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.5218524305942235e-05,
+      "loss": 0.7674,
+      "step": 528800
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.5209146335030356e-05,
+      "loss": 0.7706,
+      "step": 529000
+    },
+    {
+      "epoch": 0.62,
+      "eval_em_first_line_ratio": 0.4893,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3766,
+      "eval_loss": 0.7398356199264526,
+      "eval_runtime": 201.1192,
+      "eval_samples_per_second": 49.722,
+      "eval_steps_per_second": 0.781,
+      "step": 529000
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.5199768364118477e-05,
+      "loss": 0.7716,
+      "step": 529200
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.51903903932066e-05,
+      "loss": 0.765,
+      "step": 529400
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.518101242229472e-05,
+      "loss": 0.767,
+      "step": 529600
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.5171634451382841e-05,
+      "loss": 0.7608,
+      "step": 529800
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.5162256480470962e-05,
+      "loss": 0.767,
+      "step": 530000
+    },
+    {
+      "epoch": 0.62,
+      "eval_em_first_line_ratio": 0.4914,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.378,
+      "eval_loss": 0.7392562031745911,
+      "eval_runtime": 202.7132,
+      "eval_samples_per_second": 49.331,
+      "eval_steps_per_second": 0.774,
+      "step": 530000
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.5152925399413643e-05,
+      "loss": 0.7666,
+      "step": 530200
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.5143547428501766e-05,
+      "loss": 0.7623,
+      "step": 530400
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.5134169457589887e-05,
+      "loss": 0.7613,
+      "step": 530600
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.5124791486678008e-05,
+      "loss": 0.7609,
+      "step": 530800
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.5115413515766129e-05,
+      "loss": 0.7601,
+      "step": 531000
+    },
+    {
+      "epoch": 0.62,
+      "eval_em_first_line_ratio": 0.4906,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3772,
+      "eval_loss": 0.7401056885719299,
+      "eval_runtime": 199.4951,
+      "eval_samples_per_second": 50.127,
+      "eval_steps_per_second": 0.787,
+      "step": 531000
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.510603554485425e-05,
+      "loss": 0.7659,
+      "step": 531200
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.509665757394237e-05,
+      "loss": 0.7655,
+      "step": 531400
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.5087279603030493e-05,
+      "loss": 0.7667,
+      "step": 531600
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.5077901632118614e-05,
+      "loss": 0.7668,
+      "step": 531800
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.5068523661206735e-05,
+      "loss": 0.7673,
+      "step": 532000
+    },
+    {
+      "epoch": 0.62,
+      "eval_em_first_line_ratio": 0.4886,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3757,
+      "eval_loss": 0.7393961548805237,
+      "eval_runtime": 200.4335,
+      "eval_samples_per_second": 49.892,
+      "eval_steps_per_second": 0.783,
+      "step": 532000
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.5059145690294856e-05,
+      "loss": 0.7662,
+      "step": 532200
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.5049767719382977e-05,
+      "loss": 0.7629,
+      "step": 532400
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.5040389748471098e-05,
+      "loss": 0.7664,
+      "step": 532600
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.5031011777559221e-05,
+      "loss": 0.7642,
+      "step": 532800
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.5021633806647342e-05,
+      "loss": 0.7639,
+      "step": 533000
+    },
+    {
+      "epoch": 0.62,
+      "eval_em_first_line_ratio": 0.4898,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3764,
+      "eval_loss": 0.739612877368927,
+      "eval_runtime": 204.3967,
+      "eval_samples_per_second": 48.924,
+      "eval_steps_per_second": 0.768,
+      "step": 533000
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.5012255835735463e-05,
+      "loss": 0.7648,
+      "step": 533200
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.5002877864823584e-05,
+      "loss": 0.7618,
+      "step": 533400
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.4993499893911705e-05,
+      "loss": 0.764,
+      "step": 533600
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.4984121922999826e-05,
+      "loss": 0.7652,
+      "step": 533800
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.4974743952087947e-05,
+      "loss": 0.7628,
+      "step": 534000
+    },
+    {
+      "epoch": 0.63,
+      "eval_em_first_line_ratio": 0.4897,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3767,
+      "eval_loss": 0.7386098504066467,
+      "eval_runtime": 197.4399,
+      "eval_samples_per_second": 50.648,
+      "eval_steps_per_second": 0.795,
+      "step": 534000
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.496536598117607e-05,
+      "loss": 0.7625,
+      "step": 534200
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.495598801026419e-05,
+      "loss": 0.7702,
+      "step": 534400
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.4946610039352311e-05,
+      "loss": 0.7659,
+      "step": 534600
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.4937232068440432e-05,
+      "loss": 0.7675,
+      "step": 534800
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.4927854097528553e-05,
+      "loss": 0.7666,
+      "step": 535000
+    },
+    {
+      "epoch": 0.63,
+      "eval_em_first_line_ratio": 0.4889,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.376,
+      "eval_loss": 0.7384722232818604,
+      "eval_runtime": 193.2372,
+      "eval_samples_per_second": 51.75,
+      "eval_steps_per_second": 0.812,
+      "step": 535000
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.4918476126616674e-05,
+      "loss": 0.7637,
+      "step": 535200
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.4909098155704797e-05,
+      "loss": 0.766,
+      "step": 535400
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.4899720184792918e-05,
+      "loss": 0.7691,
+      "step": 535600
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.4890342213881039e-05,
+      "loss": 0.7663,
+      "step": 535800
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.488096424296916e-05,
+      "loss": 0.7661,
+      "step": 536000
+    },
+    {
+      "epoch": 0.63,
+      "eval_em_first_line_ratio": 0.4888,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.376,
+      "eval_loss": 0.7386866211891174,
+      "eval_runtime": 203.346,
+      "eval_samples_per_second": 49.177,
+      "eval_steps_per_second": 0.772,
+      "step": 536000
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.487158627205728e-05,
+      "loss": 0.7681,
+      "step": 536200
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.4862208301145402e-05,
+      "loss": 0.7655,
+      "step": 536400
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.4852830330233524e-05,
+      "loss": 0.7647,
+      "step": 536600
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.4843452359321645e-05,
+      "loss": 0.7636,
+      "step": 536800
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.4834121278264326e-05,
+      "loss": 0.7613,
+      "step": 537000
+    },
+    {
+      "epoch": 0.63,
+      "eval_em_first_line_ratio": 0.4874,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3747,
+      "eval_loss": 0.7393508553504944,
+      "eval_runtime": 202.0748,
+      "eval_samples_per_second": 49.487,
+      "eval_steps_per_second": 0.777,
+      "step": 537000
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.4824743307352447e-05,
+      "loss": 0.7629,
+      "step": 537200
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.4815365336440568e-05,
+      "loss": 0.7675,
+      "step": 537400
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.4805987365528691e-05,
+      "loss": 0.7659,
+      "step": 537600
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.4796609394616812e-05,
+      "loss": 0.7639,
+      "step": 537800
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.4787231423704933e-05,
+      "loss": 0.763,
+      "step": 538000
+    },
+    {
+      "epoch": 0.63,
+      "eval_em_first_line_ratio": 0.4907,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3777,
+      "eval_loss": 0.7385621070861816,
+      "eval_runtime": 198.3402,
+      "eval_samples_per_second": 50.418,
+      "eval_steps_per_second": 0.792,
+      "step": 538000
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.4777853452793054e-05,
+      "loss": 0.7649,
+      "step": 538200
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.4768475481881175e-05,
+      "loss": 0.7722,
+      "step": 538400
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.4759097510969296e-05,
+      "loss": 0.7617,
+      "step": 538600
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.4749719540057418e-05,
+      "loss": 0.7651,
+      "step": 538800
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.474034156914554e-05,
+      "loss": 0.7678,
+      "step": 539000
+    },
+    {
+      "epoch": 0.63,
+      "eval_em_first_line_ratio": 0.4893,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3764,
+      "eval_loss": 0.7387046813964844,
+      "eval_runtime": 203.4091,
+      "eval_samples_per_second": 49.162,
+      "eval_steps_per_second": 0.772,
+      "step": 539000
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.473096359823366e-05,
+      "loss": 0.7679,
+      "step": 539200
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.4721585627321781e-05,
+      "loss": 0.7703,
+      "step": 539400
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.4712207656409902e-05,
+      "loss": 0.7645,
+      "step": 539600
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.4702829685498023e-05,
+      "loss": 0.7648,
+      "step": 539800
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.4693451714586144e-05,
+      "loss": 0.7648,
+      "step": 540000
+    },
+    {
+      "epoch": 0.63,
+      "eval_em_first_line_ratio": 0.489,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3756,
+      "eval_loss": 0.7387475371360779,
+      "eval_runtime": 198.0881,
+      "eval_samples_per_second": 50.483,
+      "eval_steps_per_second": 0.793,
+      "step": 540000
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.4684073743674267e-05,
+      "loss": 0.7644,
+      "step": 540200
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.4674695772762388e-05,
+      "loss": 0.7667,
+      "step": 540400
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.4665317801850509e-05,
+      "loss": 0.7678,
+      "step": 540600
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.465593983093863e-05,
+      "loss": 0.7614,
+      "step": 540800
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.464660874988131e-05,
+      "loss": 0.7669,
+      "step": 541000
+    },
+    {
+      "epoch": 0.63,
+      "eval_em_first_line_ratio": 0.4895,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3765,
+      "eval_loss": 0.7394345998764038,
+      "eval_runtime": 201.0597,
+      "eval_samples_per_second": 49.736,
+      "eval_steps_per_second": 0.781,
+      "step": 541000
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.4637230778969433e-05,
+      "loss": 0.7662,
+      "step": 541200
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.4627852808057554e-05,
+      "loss": 0.7667,
+      "step": 541400
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.4618474837145675e-05,
+      "loss": 0.7652,
+      "step": 541600
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.4609096866233796e-05,
+      "loss": 0.767,
+      "step": 541800
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.4599765785176477e-05,
+      "loss": 0.7674,
+      "step": 542000
+    },
+    {
+      "epoch": 0.64,
+      "eval_em_first_line_ratio": 0.4887,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3756,
+      "eval_loss": 0.7390217781066895,
+      "eval_runtime": 205.5949,
+      "eval_samples_per_second": 48.639,
+      "eval_steps_per_second": 0.764,
+      "step": 542000
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.45903878142646e-05,
+      "loss": 0.7693,
+      "step": 542200
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.4581056733207279e-05,
+      "loss": 0.7667,
+      "step": 542400
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.45716787622954e-05,
+      "loss": 0.7687,
+      "step": 542600
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.456230079138352e-05,
+      "loss": 0.7658,
+      "step": 542800
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.4552922820471642e-05,
+      "loss": 0.7641,
+      "step": 543000
+    },
+    {
+      "epoch": 0.64,
+      "eval_em_first_line_ratio": 0.4887,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3756,
+      "eval_loss": 0.7393035888671875,
+      "eval_runtime": 208.3569,
+      "eval_samples_per_second": 47.995,
+      "eval_steps_per_second": 0.754,
+      "step": 543000
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.4543544849559763e-05,
+      "loss": 0.7654,
+      "step": 543200
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.4534213768502445e-05,
+      "loss": 0.7627,
+      "step": 543400
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.4524835797590566e-05,
+      "loss": 0.7614,
+      "step": 543600
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.4515457826678687e-05,
+      "loss": 0.7667,
+      "step": 543800
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.4506079855766808e-05,
+      "loss": 0.7657,
+      "step": 544000
+    },
+    {
+      "epoch": 0.64,
+      "eval_em_first_line_ratio": 0.4891,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3766,
+      "eval_loss": 0.7388086915016174,
+      "eval_runtime": 205.5445,
+      "eval_samples_per_second": 48.651,
+      "eval_steps_per_second": 0.764,
+      "step": 544000
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.4496701884854929e-05,
+      "loss": 0.7649,
+      "step": 544200
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.4487323913943052e-05,
+      "loss": 0.767,
+      "step": 544400
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.4477945943031173e-05,
+      "loss": 0.7678,
+      "step": 544600
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.4468567972119294e-05,
+      "loss": 0.7602,
+      "step": 544800
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.4459190001207415e-05,
+      "loss": 0.766,
+      "step": 545000
+    },
+    {
+      "epoch": 0.64,
+      "eval_em_first_line_ratio": 0.4903,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.377,
+      "eval_loss": 0.7387192249298096,
+      "eval_runtime": 201.8138,
+      "eval_samples_per_second": 49.551,
+      "eval_steps_per_second": 0.778,
+      "step": 545000
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.4449812030295536e-05,
+      "loss": 0.768,
+      "step": 545200
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.4440434059383657e-05,
+      "loss": 0.7633,
+      "step": 545400
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.443105608847178e-05,
+      "loss": 0.7603,
+      "step": 545600
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.44216781175599e-05,
+      "loss": 0.7629,
+      "step": 545800
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.4412300146648021e-05,
+      "loss": 0.7671,
+      "step": 546000
+    },
+    {
+      "epoch": 0.64,
+      "eval_em_first_line_ratio": 0.4889,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3758,
+      "eval_loss": 0.7383544445037842,
+      "eval_runtime": 207.5636,
+      "eval_samples_per_second": 48.178,
+      "eval_steps_per_second": 0.756,
+      "step": 546000
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.4402969065590702e-05,
+      "loss": 0.7657,
+      "step": 546200
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.4393591094678823e-05,
+      "loss": 0.7637,
+      "step": 546400
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.4384213123766944e-05,
+      "loss": 0.766,
+      "step": 546600
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.4374835152855067e-05,
+      "loss": 0.7681,
+      "step": 546800
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.4365457181943188e-05,
+      "loss": 0.7677,
+      "step": 547000
+    },
+    {
+      "epoch": 0.64,
+      "eval_em_first_line_ratio": 0.4864,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3734,
+      "eval_loss": 0.7391924858093262,
+      "eval_runtime": 207.1389,
+      "eval_samples_per_second": 48.277,
+      "eval_steps_per_second": 0.758,
+      "step": 547000
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.4356079211031309e-05,
+      "loss": 0.7629,
+      "step": 547200
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.434670124011943e-05,
+      "loss": 0.7645,
+      "step": 547400
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.433732326920755e-05,
+      "loss": 0.7641,
+      "step": 547600
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.4327945298295672e-05,
+      "loss": 0.7619,
+      "step": 547800
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.4318567327383794e-05,
+      "loss": 0.7743,
+      "step": 548000
+    },
+    {
+      "epoch": 0.64,
+      "eval_em_first_line_ratio": 0.4884,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3754,
+      "eval_loss": 0.7387399673461914,
+      "eval_runtime": 197.3653,
+      "eval_samples_per_second": 50.667,
+      "eval_steps_per_second": 0.795,
+      "step": 548000
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.4309189356471915e-05,
+      "loss": 0.765,
+      "step": 548200
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.4299811385560036e-05,
+      "loss": 0.7677,
+      "step": 548400
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.4290433414648157e-05,
+      "loss": 0.7638,
+      "step": 548600
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.4281055443736278e-05,
+      "loss": 0.763,
+      "step": 548800
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.4271677472824399e-05,
+      "loss": 0.7634,
+      "step": 549000
+    },
+    {
+      "epoch": 0.64,
+      "eval_em_first_line_ratio": 0.4879,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3754,
+      "eval_loss": 0.7385572195053101,
+      "eval_runtime": 205.914,
+      "eval_samples_per_second": 48.564,
+      "eval_steps_per_second": 0.762,
+      "step": 549000
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.4262299501912522e-05,
+      "loss": 0.7671,
+      "step": 549200
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.4252921531000643e-05,
+      "loss": 0.7678,
+      "step": 549400
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.4243543560088764e-05,
+      "loss": 0.7613,
+      "step": 549600
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.4234165589176885e-05,
+      "loss": 0.7616,
+      "step": 549800
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.4224787618265006e-05,
+      "loss": 0.7605,
+      "step": 550000
+    },
+    {
+      "epoch": 0.64,
+      "eval_em_first_line_ratio": 0.4902,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3772,
+      "eval_loss": 0.738932192325592,
+      "eval_runtime": 207.446,
+      "eval_samples_per_second": 48.205,
+      "eval_steps_per_second": 0.757,
+      "step": 550000
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.4215409647353127e-05,
+      "loss": 0.763,
+      "step": 550200
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.420603167644125e-05,
+      "loss": 0.7635,
+      "step": 550400
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.419665370552937e-05,
+      "loss": 0.766,
+      "step": 550600
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.4187275734617491e-05,
+      "loss": 0.7666,
+      "step": 550800
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.4177897763705612e-05,
+      "loss": 0.7673,
+      "step": 551000
+    },
+    {
+      "epoch": 0.65,
+      "eval_em_first_line_ratio": 0.4896,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.377,
+      "eval_loss": 0.7381096482276917,
+      "eval_runtime": 204.1008,
+      "eval_samples_per_second": 48.995,
+      "eval_steps_per_second": 0.769,
+      "step": 551000
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.4168519792793733e-05,
+      "loss": 0.7627,
+      "step": 551200
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.4159141821881854e-05,
+      "loss": 0.7649,
+      "step": 551400
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.4149763850969977e-05,
+      "loss": 0.7628,
+      "step": 551600
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.4140385880058098e-05,
+      "loss": 0.7627,
+      "step": 551800
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.4131007909146219e-05,
+      "loss": 0.7658,
+      "step": 552000
+    },
+    {
+      "epoch": 0.65,
+      "eval_em_first_line_ratio": 0.4915,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.379,
+      "eval_loss": 0.7377377152442932,
+      "eval_runtime": 206.3883,
+      "eval_samples_per_second": 48.452,
+      "eval_steps_per_second": 0.761,
+      "step": 552000
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.412162993823434e-05,
+      "loss": 0.7612,
+      "step": 552200
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.411225196732246e-05,
+      "loss": 0.7625,
+      "step": 552400
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.4102873996410582e-05,
+      "loss": 0.7615,
+      "step": 552600
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.4093496025498703e-05,
+      "loss": 0.7605,
+      "step": 552800
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.4084118054586825e-05,
+      "loss": 0.7612,
+      "step": 553000
+    },
+    {
+      "epoch": 0.65,
+      "eval_em_first_line_ratio": 0.4922,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3791,
+      "eval_loss": 0.7379783987998962,
+      "eval_runtime": 202.0478,
+      "eval_samples_per_second": 49.493,
+      "eval_steps_per_second": 0.777,
+      "step": 553000
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.4074740083674946e-05,
+      "loss": 0.7626,
+      "step": 553200
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.4065362112763067e-05,
+      "loss": 0.7628,
+      "step": 553400
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.4055984141851188e-05,
+      "loss": 0.7655,
+      "step": 553600
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.404660617093931e-05,
+      "loss": 0.7697,
+      "step": 553800
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.403722820002743e-05,
+      "loss": 0.7664,
+      "step": 554000
+    },
+    {
+      "epoch": 0.65,
+      "eval_em_first_line_ratio": 0.4917,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3788,
+      "eval_loss": 0.7379279136657715,
+      "eval_runtime": 200.5787,
+      "eval_samples_per_second": 49.856,
+      "eval_steps_per_second": 0.783,
+      "step": 554000
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.4027850229115555e-05,
+      "loss": 0.7599,
+      "step": 554200
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.4018566037912793e-05,
+      "loss": 0.766,
+      "step": 554400
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.4009188067000914e-05,
+      "loss": 0.7651,
+      "step": 554600
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.3999810096089035e-05,
+      "loss": 0.7623,
+      "step": 554800
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.3990432125177158e-05,
+      "loss": 0.7647,
+      "step": 555000
+    },
+    {
+      "epoch": 0.65,
+      "eval_em_first_line_ratio": 0.4892,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3767,
+      "eval_loss": 0.73763507604599,
+      "eval_runtime": 204.7803,
+      "eval_samples_per_second": 48.833,
+      "eval_steps_per_second": 0.767,
+      "step": 555000
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.3981054154265279e-05,
+      "loss": 0.7654,
+      "step": 555200
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.39716761833534e-05,
+      "loss": 0.7605,
+      "step": 555400
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.3962298212441521e-05,
+      "loss": 0.7629,
+      "step": 555600
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.3952920241529642e-05,
+      "loss": 0.7618,
+      "step": 555800
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.3943542270617763e-05,
+      "loss": 0.7622,
+      "step": 556000
+    },
+    {
+      "epoch": 0.65,
+      "eval_em_first_line_ratio": 0.4914,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.379,
+      "eval_loss": 0.7368842959403992,
+      "eval_runtime": 210.2635,
+      "eval_samples_per_second": 47.559,
+      "eval_steps_per_second": 0.747,
+      "step": 556000
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.3934164299705884e-05,
+      "loss": 0.7633,
+      "step": 556200
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.3924786328794007e-05,
+      "loss": 0.7645,
+      "step": 556400
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.3915408357882128e-05,
+      "loss": 0.7671,
+      "step": 556600
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.3906030386970249e-05,
+      "loss": 0.7663,
+      "step": 556800
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.389665241605837e-05,
+      "loss": 0.7642,
+      "step": 557000
+    },
+    {
+      "epoch": 0.65,
+      "eval_em_first_line_ratio": 0.4925,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3795,
+      "eval_loss": 0.7379512190818787,
+      "eval_runtime": 201.6035,
+      "eval_samples_per_second": 49.602,
+      "eval_steps_per_second": 0.779,
+      "step": 557000
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.388727444514649e-05,
+      "loss": 0.7637,
+      "step": 557200
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.3877896474234611e-05,
+      "loss": 0.7631,
+      "step": 557400
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.3868518503322734e-05,
+      "loss": 0.7689,
+      "step": 557600
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.3859140532410855e-05,
+      "loss": 0.7629,
+      "step": 557800
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.3849762561498976e-05,
+      "loss": 0.7604,
+      "step": 558000
+    },
+    {
+      "epoch": 0.65,
+      "eval_em_first_line_ratio": 0.4919,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3789,
+      "eval_loss": 0.7376495003700256,
+      "eval_runtime": 204.6991,
+      "eval_samples_per_second": 48.852,
+      "eval_steps_per_second": 0.767,
+      "step": 558000
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.3840384590587097e-05,
+      "loss": 0.7643,
+      "step": 558200
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.3831006619675218e-05,
+      "loss": 0.7663,
+      "step": 558400
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.3821628648763339e-05,
+      "loss": 0.7649,
+      "step": 558600
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.3812250677851462e-05,
+      "loss": 0.7608,
+      "step": 558800
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.3802872706939583e-05,
+      "loss": 0.7672,
+      "step": 559000
+    },
+    {
+      "epoch": 0.66,
+      "eval_em_first_line_ratio": 0.4906,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.378,
+      "eval_loss": 0.7376105189323425,
+      "eval_runtime": 197.2082,
+      "eval_samples_per_second": 50.708,
+      "eval_steps_per_second": 0.796,
+      "step": 559000
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.3793494736027704e-05,
+      "loss": 0.7667,
+      "step": 559200
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.3784116765115825e-05,
+      "loss": 0.7661,
+      "step": 559400
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.3774738794203946e-05,
+      "loss": 0.7612,
+      "step": 559600
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.3765360823292067e-05,
+      "loss": 0.76,
+      "step": 559800
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.3756029742234749e-05,
+      "loss": 0.7603,
+      "step": 560000
+    },
+    {
+      "epoch": 0.66,
+      "eval_em_first_line_ratio": 0.4924,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.38,
+      "eval_loss": 0.7376967072486877,
+      "eval_runtime": 197.0366,
+      "eval_samples_per_second": 50.752,
+      "eval_steps_per_second": 0.797,
+      "step": 560000
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.374665177132287e-05,
+      "loss": 0.7666,
+      "step": 560200
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.3737273800410991e-05,
+      "loss": 0.7659,
+      "step": 560400
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.3727895829499112e-05,
+      "loss": 0.761,
+      "step": 560600
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.3718517858587233e-05,
+      "loss": 0.7618,
+      "step": 560800
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.3709139887675354e-05,
+      "loss": 0.7558,
+      "step": 561000
+    },
+    {
+      "epoch": 0.66,
+      "eval_em_first_line_ratio": 0.4919,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3798,
+      "eval_loss": 0.7369788289070129,
+      "eval_runtime": 204.3079,
+      "eval_samples_per_second": 48.946,
+      "eval_steps_per_second": 0.768,
+      "step": 561000
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.3699761916763477e-05,
+      "loss": 0.7659,
+      "step": 561200
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.3690383945851598e-05,
+      "loss": 0.7601,
+      "step": 561400
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.3681005974939719e-05,
+      "loss": 0.7652,
+      "step": 561600
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.3671674893882398e-05,
+      "loss": 0.7605,
+      "step": 561800
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.3662296922970519e-05,
+      "loss": 0.7641,
+      "step": 562000
+    },
+    {
+      "epoch": 0.66,
+      "eval_em_first_line_ratio": 0.4925,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3798,
+      "eval_loss": 0.7367867231369019,
+      "eval_runtime": 213.1015,
+      "eval_samples_per_second": 46.926,
+      "eval_steps_per_second": 0.737,
+      "step": 562000
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.3652918952058643e-05,
+      "loss": 0.7627,
+      "step": 562200
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.3643540981146764e-05,
+      "loss": 0.7644,
+      "step": 562400
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.3634163010234885e-05,
+      "loss": 0.7605,
+      "step": 562600
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.3624785039323006e-05,
+      "loss": 0.7657,
+      "step": 562800
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.3615407068411127e-05,
+      "loss": 0.7676,
+      "step": 563000
+    },
+    {
+      "epoch": 0.66,
+      "eval_em_first_line_ratio": 0.4915,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3793,
+      "eval_loss": 0.7371493577957153,
+      "eval_runtime": 203.4559,
+      "eval_samples_per_second": 49.151,
+      "eval_steps_per_second": 0.772,
+      "step": 563000
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.3606029097499248e-05,
+      "loss": 0.7576,
+      "step": 563200
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.359665112658737e-05,
+      "loss": 0.7606,
+      "step": 563400
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.3587273155675491e-05,
+      "loss": 0.7627,
+      "step": 563600
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.3577895184763612e-05,
+      "loss": 0.7632,
+      "step": 563800
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.3568517213851733e-05,
+      "loss": 0.7644,
+      "step": 564000
+    },
+    {
+      "epoch": 0.66,
+      "eval_em_first_line_ratio": 0.4903,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3772,
+      "eval_loss": 0.7360977530479431,
+      "eval_runtime": 207.2317,
+      "eval_samples_per_second": 48.255,
+      "eval_steps_per_second": 0.758,
+      "step": 564000
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.3559139242939854e-05,
+      "loss": 0.7688,
+      "step": 564200
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.3549761272027975e-05,
+      "loss": 0.7635,
+      "step": 564400
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.3540383301116098e-05,
+      "loss": 0.7645,
+      "step": 564600
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.3531005330204219e-05,
+      "loss": 0.76,
+      "step": 564800
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.352162735929234e-05,
+      "loss": 0.7624,
+      "step": 565000
+    },
+    {
+      "epoch": 0.66,
+      "eval_em_first_line_ratio": 0.4892,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.376,
+      "eval_loss": 0.7370457053184509,
+      "eval_runtime": 206.4513,
+      "eval_samples_per_second": 48.438,
+      "eval_steps_per_second": 0.76,
+      "step": 565000
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.3512249388380461e-05,
+      "loss": 0.7635,
+      "step": 565200
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.3502871417468582e-05,
+      "loss": 0.7567,
+      "step": 565400
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.3493493446556703e-05,
+      "loss": 0.7649,
+      "step": 565600
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.3484115475644826e-05,
+      "loss": 0.7609,
+      "step": 565800
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.3474784394587506e-05,
+      "loss": 0.7605,
+      "step": 566000
+    },
+    {
+      "epoch": 0.66,
+      "eval_em_first_line_ratio": 0.4908,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3778,
+      "eval_loss": 0.7367969751358032,
+      "eval_runtime": 201.5538,
+      "eval_samples_per_second": 49.615,
+      "eval_steps_per_second": 0.779,
+      "step": 566000
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.3465406423675627e-05,
+      "loss": 0.7639,
+      "step": 566200
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.3456028452763748e-05,
+      "loss": 0.7598,
+      "step": 566400
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.3446650481851868e-05,
+      "loss": 0.7591,
+      "step": 566600
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.3437272510939989e-05,
+      "loss": 0.7611,
+      "step": 566800
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.3427894540028113e-05,
+      "loss": 0.759,
+      "step": 567000
+    },
+    {
+      "epoch": 0.66,
+      "eval_em_first_line_ratio": 0.4904,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3771,
+      "eval_loss": 0.7364742159843445,
+      "eval_runtime": 198.4481,
+      "eval_samples_per_second": 50.391,
+      "eval_steps_per_second": 0.791,
+      "step": 567000
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.3418516569116234e-05,
+      "loss": 0.7635,
+      "step": 567200
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.3409138598204355e-05,
+      "loss": 0.7647,
+      "step": 567400
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.3399760627292476e-05,
+      "loss": 0.7616,
+      "step": 567600
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.3390382656380597e-05,
+      "loss": 0.7604,
+      "step": 567800
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.3381004685468718e-05,
+      "loss": 0.7599,
+      "step": 568000
+    },
+    {
+      "epoch": 0.67,
+      "eval_em_first_line_ratio": 0.4904,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3777,
+      "eval_loss": 0.7363232374191284,
+      "eval_runtime": 202.0047,
+      "eval_samples_per_second": 49.504,
+      "eval_steps_per_second": 0.777,
+      "step": 568000
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.337162671455684e-05,
+      "loss": 0.7669,
+      "step": 568200
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.336229563349952e-05,
+      "loss": 0.7646,
+      "step": 568400
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.335291766258764e-05,
+      "loss": 0.7618,
+      "step": 568600
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.3343539691675762e-05,
+      "loss": 0.7662,
+      "step": 568800
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.3334161720763882e-05,
+      "loss": 0.7648,
+      "step": 569000
+    },
+    {
+      "epoch": 0.67,
+      "eval_em_first_line_ratio": 0.4905,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3776,
+      "eval_loss": 0.73649001121521,
+      "eval_runtime": 199.663,
+      "eval_samples_per_second": 50.084,
+      "eval_steps_per_second": 0.786,
+      "step": 569000
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.3324783749852007e-05,
+      "loss": 0.7641,
+      "step": 569200
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.3315405778940128e-05,
+      "loss": 0.7595,
+      "step": 569400
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.3306027808028247e-05,
+      "loss": 0.756,
+      "step": 569600
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.3296649837116368e-05,
+      "loss": 0.7617,
+      "step": 569800
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.3287271866204489e-05,
+      "loss": 0.7618,
+      "step": 570000
+    },
+    {
+      "epoch": 0.67,
+      "eval_em_first_line_ratio": 0.4925,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3793,
+      "eval_loss": 0.7368417978286743,
+      "eval_runtime": 202.2006,
+      "eval_samples_per_second": 49.456,
+      "eval_steps_per_second": 0.776,
+      "step": 570000
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.327789389529261e-05,
+      "loss": 0.7667,
+      "step": 570200
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.3268515924380731e-05,
+      "loss": 0.7603,
+      "step": 570400
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.3259137953468855e-05,
+      "loss": 0.7641,
+      "step": 570600
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.3249759982556976e-05,
+      "loss": 0.7612,
+      "step": 570800
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.3240382011645097e-05,
+      "loss": 0.7622,
+      "step": 571000
+    },
+    {
+      "epoch": 0.67,
+      "eval_em_first_line_ratio": 0.4906,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3771,
+      "eval_loss": 0.7369229793548584,
+      "eval_runtime": 205.6731,
+      "eval_samples_per_second": 48.621,
+      "eval_steps_per_second": 0.763,
+      "step": 571000
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.3231004040733218e-05,
+      "loss": 0.7605,
+      "step": 571200
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.3221626069821338e-05,
+      "loss": 0.7665,
+      "step": 571400
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.3212248098909459e-05,
+      "loss": 0.7685,
+      "step": 571600
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.3202870127997583e-05,
+      "loss": 0.7608,
+      "step": 571800
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.3193492157085704e-05,
+      "loss": 0.7613,
+      "step": 572000
+    },
+    {
+      "epoch": 0.67,
+      "eval_em_first_line_ratio": 0.4906,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3772,
+      "eval_loss": 0.7361516356468201,
+      "eval_runtime": 203.7777,
+      "eval_samples_per_second": 49.073,
+      "eval_steps_per_second": 0.77,
+      "step": 572000
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.3184114186173825e-05,
+      "loss": 0.7627,
+      "step": 572200
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.3174736215261946e-05,
+      "loss": 0.7565,
+      "step": 572400
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.3165358244350067e-05,
+      "loss": 0.7669,
+      "step": 572600
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.3156027163292748e-05,
+      "loss": 0.7628,
+      "step": 572800
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.3146649192380869e-05,
+      "loss": 0.7637,
+      "step": 573000
+    },
+    {
+      "epoch": 0.67,
+      "eval_em_first_line_ratio": 0.4927,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3793,
+      "eval_loss": 0.736573338508606,
+      "eval_runtime": 201.9342,
+      "eval_samples_per_second": 49.521,
+      "eval_steps_per_second": 0.777,
+      "step": 573000
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.313727122146899e-05,
+      "loss": 0.7625,
+      "step": 573200
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.312789325055711e-05,
+      "loss": 0.763,
+      "step": 573400
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.3118515279645231e-05,
+      "loss": 0.7599,
+      "step": 573600
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.3109137308733352e-05,
+      "loss": 0.7668,
+      "step": 573800
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.3099759337821477e-05,
+      "loss": 0.7622,
+      "step": 574000
+    },
+    {
+      "epoch": 0.67,
+      "eval_em_first_line_ratio": 0.4912,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.378,
+      "eval_loss": 0.7361474633216858,
+      "eval_runtime": 195.7519,
+      "eval_samples_per_second": 51.085,
+      "eval_steps_per_second": 0.802,
+      "step": 574000
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.3090381366909596e-05,
+      "loss": 0.766,
+      "step": 574200
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.3081003395997717e-05,
+      "loss": 0.7598,
+      "step": 574400
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.3071625425085838e-05,
+      "loss": 0.7584,
+      "step": 574600
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.3062294344028519e-05,
+      "loss": 0.7618,
+      "step": 574800
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.305291637311664e-05,
+      "loss": 0.7576,
+      "step": 575000
+    },
+    {
+      "epoch": 0.67,
+      "eval_em_first_line_ratio": 0.4929,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3797,
+      "eval_loss": 0.735771656036377,
+      "eval_runtime": 200.8746,
+      "eval_samples_per_second": 49.782,
+      "eval_steps_per_second": 0.782,
+      "step": 575000
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.3043538402204762e-05,
+      "loss": 0.7687,
+      "step": 575200
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.3034160431292883e-05,
+      "loss": 0.765,
+      "step": 575400
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.3024782460381004e-05,
+      "loss": 0.759,
+      "step": 575600
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.3015404489469125e-05,
+      "loss": 0.7546,
+      "step": 575800
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.3006026518557246e-05,
+      "loss": 0.7626,
+      "step": 576000
+    },
+    {
+      "epoch": 0.68,
+      "eval_em_first_line_ratio": 0.4916,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3786,
+      "eval_loss": 0.7362521290779114,
+      "eval_runtime": 200.0982,
+      "eval_samples_per_second": 49.975,
+      "eval_steps_per_second": 0.785,
+      "step": 576000
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.2996648547645367e-05,
+      "loss": 0.7581,
+      "step": 576200
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.298727057673349e-05,
+      "loss": 0.7603,
+      "step": 576400
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.2977892605821611e-05,
+      "loss": 0.7619,
+      "step": 576600
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.2968514634909732e-05,
+      "loss": 0.7607,
+      "step": 576800
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.2959136663997853e-05,
+      "loss": 0.7654,
+      "step": 577000
+    },
+    {
+      "epoch": 0.68,
+      "eval_em_first_line_ratio": 0.4922,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3791,
+      "eval_loss": 0.7357962131500244,
+      "eval_runtime": 196.6622,
+      "eval_samples_per_second": 50.849,
+      "eval_steps_per_second": 0.798,
+      "step": 577000
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.2949758693085974e-05,
+      "loss": 0.7645,
+      "step": 577200
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.2940380722174095e-05,
+      "loss": 0.7622,
+      "step": 577400
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.2931002751262218e-05,
+      "loss": 0.7601,
+      "step": 577600
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.2921624780350339e-05,
+      "loss": 0.7686,
+      "step": 577800
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.291224680943846e-05,
+      "loss": 0.7654,
+      "step": 578000
+    },
+    {
+      "epoch": 0.68,
+      "eval_em_first_line_ratio": 0.4908,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3779,
+      "eval_loss": 0.7358078360557556,
+      "eval_runtime": 205.3564,
+      "eval_samples_per_second": 48.696,
+      "eval_steps_per_second": 0.765,
+      "step": 578000
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.290291572838114e-05,
+      "loss": 0.7576,
+      "step": 578200
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.2893537757469261e-05,
+      "loss": 0.759,
+      "step": 578400
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.2884159786557384e-05,
+      "loss": 0.7629,
+      "step": 578600
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.2874781815645505e-05,
+      "loss": 0.7604,
+      "step": 578800
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.2865403844733626e-05,
+      "loss": 0.7567,
+      "step": 579000
+    },
+    {
+      "epoch": 0.68,
+      "eval_em_first_line_ratio": 0.4938,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3808,
+      "eval_loss": 0.7358170747756958,
+      "eval_runtime": 198.5307,
+      "eval_samples_per_second": 50.37,
+      "eval_steps_per_second": 0.791,
+      "step": 579000
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.2856025873821747e-05,
+      "loss": 0.7665,
+      "step": 579200
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.2846647902909868e-05,
+      "loss": 0.7631,
+      "step": 579400
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.2837269931997989e-05,
+      "loss": 0.7608,
+      "step": 579600
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.282789196108611e-05,
+      "loss": 0.7654,
+      "step": 579800
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.2818513990174232e-05,
+      "loss": 0.7634,
+      "step": 580000
+    },
+    {
+      "epoch": 0.68,
+      "eval_em_first_line_ratio": 0.4919,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3786,
+      "eval_loss": 0.7362378239631653,
+      "eval_runtime": 204.4921,
+      "eval_samples_per_second": 48.902,
+      "eval_steps_per_second": 0.768,
+      "step": 580000
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.2809136019262353e-05,
+      "loss": 0.7621,
+      "step": 580200
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.2799804938205034e-05,
+      "loss": 0.7649,
+      "step": 580400
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.2790426967293155e-05,
+      "loss": 0.7613,
+      "step": 580600
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.2781048996381276e-05,
+      "loss": 0.7617,
+      "step": 580800
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.2771671025469399e-05,
+      "loss": 0.7643,
+      "step": 581000
+    },
+    {
+      "epoch": 0.68,
+      "eval_em_first_line_ratio": 0.4935,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3803,
+      "eval_loss": 0.7358880043029785,
+      "eval_runtime": 203.998,
+      "eval_samples_per_second": 49.02,
+      "eval_steps_per_second": 0.77,
+      "step": 581000
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.276229305455752e-05,
+      "loss": 0.7627,
+      "step": 581200
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.275291508364564e-05,
+      "loss": 0.7686,
+      "step": 581400
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.2743537112733762e-05,
+      "loss": 0.7602,
+      "step": 581600
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.2734159141821883e-05,
+      "loss": 0.7572,
+      "step": 581800
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.2724781170910004e-05,
+      "loss": 0.7682,
+      "step": 582000
+    },
+    {
+      "epoch": 0.68,
+      "eval_em_first_line_ratio": 0.4918,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3788,
+      "eval_loss": 0.7352127432823181,
+      "eval_runtime": 206.3594,
+      "eval_samples_per_second": 48.459,
+      "eval_steps_per_second": 0.761,
+      "step": 582000
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.2715403199998126e-05,
+      "loss": 0.7646,
+      "step": 582200
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.2706025229086247e-05,
+      "loss": 0.7579,
+      "step": 582400
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.2696647258174368e-05,
+      "loss": 0.7625,
+      "step": 582600
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.268726928726249e-05,
+      "loss": 0.7597,
+      "step": 582800
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.267789131635061e-05,
+      "loss": 0.7542,
+      "step": 583000
+    },
+    {
+      "epoch": 0.68,
+      "eval_em_first_line_ratio": 0.4914,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3783,
+      "eval_loss": 0.7354664206504822,
+      "eval_runtime": 200.8735,
+      "eval_samples_per_second": 49.783,
+      "eval_steps_per_second": 0.782,
+      "step": 583000
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.266856023529329e-05,
+      "loss": 0.758,
+      "step": 583200
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.2659182264381414e-05,
+      "loss": 0.7587,
+      "step": 583400
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.2649804293469535e-05,
+      "loss": 0.7626,
+      "step": 583600
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.2640426322557656e-05,
+      "loss": 0.7606,
+      "step": 583800
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.2631048351645777e-05,
+      "loss": 0.764,
+      "step": 584000
+    },
+    {
+      "epoch": 0.68,
+      "eval_em_first_line_ratio": 0.4903,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3771,
+      "eval_loss": 0.7355663776397705,
+      "eval_runtime": 202.7799,
+      "eval_samples_per_second": 49.315,
+      "eval_steps_per_second": 0.774,
+      "step": 584000
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.2621670380733898e-05,
+      "loss": 0.7581,
+      "step": 584200
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.2612292409822019e-05,
+      "loss": 0.7603,
+      "step": 584400
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.2602914438910141e-05,
+      "loss": 0.762,
+      "step": 584600
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.2593536467998262e-05,
+      "loss": 0.758,
+      "step": 584800
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.2584158497086383e-05,
+      "loss": 0.7645,
+      "step": 585000
+    },
+    {
+      "epoch": 0.69,
+      "eval_em_first_line_ratio": 0.4927,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3796,
+      "eval_loss": 0.7355904579162598,
+      "eval_runtime": 204.9069,
+      "eval_samples_per_second": 48.803,
+      "eval_steps_per_second": 0.766,
+      "step": 585000
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.2574780526174504e-05,
+      "loss": 0.7586,
+      "step": 585200
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.2565402555262625e-05,
+      "loss": 0.7632,
+      "step": 585400
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.2556024584350746e-05,
+      "loss": 0.7619,
+      "step": 585600
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.2546646613438869e-05,
+      "loss": 0.7614,
+      "step": 585800
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.2537315532381548e-05,
+      "loss": 0.7578,
+      "step": 586000
+    },
+    {
+      "epoch": 0.69,
+      "eval_em_first_line_ratio": 0.4914,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.378,
+      "eval_loss": 0.7351023554801941,
+      "eval_runtime": 206.5825,
+      "eval_samples_per_second": 48.407,
+      "eval_steps_per_second": 0.76,
+      "step": 586000
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.2527937561469669e-05,
+      "loss": 0.7632,
+      "step": 586200
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.251855959055779e-05,
+      "loss": 0.7579,
+      "step": 586400
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.250918161964591e-05,
+      "loss": 0.7557,
+      "step": 586600
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.2499803648734035e-05,
+      "loss": 0.7597,
+      "step": 586800
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.2490425677822156e-05,
+      "loss": 0.7565,
+      "step": 587000
+    },
+    {
+      "epoch": 0.69,
+      "eval_em_first_line_ratio": 0.4914,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3785,
+      "eval_loss": 0.7351300120353699,
+      "eval_runtime": 198.5931,
+      "eval_samples_per_second": 50.354,
+      "eval_steps_per_second": 0.791,
+      "step": 587000
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.2481047706910277e-05,
+      "loss": 0.7627,
+      "step": 587200
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.2471716625852956e-05,
+      "loss": 0.7607,
+      "step": 587400
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.2462338654941077e-05,
+      "loss": 0.7563,
+      "step": 587600
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.2452960684029198e-05,
+      "loss": 0.7591,
+      "step": 587800
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.2443582713117321e-05,
+      "loss": 0.7579,
+      "step": 588000
+    },
+    {
+      "epoch": 0.69,
+      "eval_em_first_line_ratio": 0.4922,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3789,
+      "eval_loss": 0.7350081205368042,
+      "eval_runtime": 191.7583,
+      "eval_samples_per_second": 52.149,
+      "eval_steps_per_second": 0.819,
+      "step": 588000
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.2434204742205442e-05,
+      "loss": 0.7618,
+      "step": 588200
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.2424826771293563e-05,
+      "loss": 0.7628,
+      "step": 588400
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.2415448800381684e-05,
+      "loss": 0.7672,
+      "step": 588600
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.2406070829469805e-05,
+      "loss": 0.7614,
+      "step": 588800
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.2396692858557926e-05,
+      "loss": 0.7599,
+      "step": 589000
+    },
+    {
+      "epoch": 0.69,
+      "eval_em_first_line_ratio": 0.4908,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3774,
+      "eval_loss": 0.7351235151290894,
+      "eval_runtime": 198.1938,
+      "eval_samples_per_second": 50.456,
+      "eval_steps_per_second": 0.792,
+      "step": 589000
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.2387314887646048e-05,
+      "loss": 0.7612,
+      "step": 589200
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.237793691673417e-05,
+      "loss": 0.7614,
+      "step": 589400
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.236860583567685e-05,
+      "loss": 0.761,
+      "step": 589600
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.2359227864764971e-05,
+      "loss": 0.7607,
+      "step": 589800
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.2349849893853092e-05,
+      "loss": 0.7623,
+      "step": 590000
+    },
+    {
+      "epoch": 0.69,
+      "eval_em_first_line_ratio": 0.4902,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3771,
+      "eval_loss": 0.7346815466880798,
+      "eval_runtime": 200.8541,
+      "eval_samples_per_second": 49.787,
+      "eval_steps_per_second": 0.782,
+      "step": 590000
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.2340471922941215e-05,
+      "loss": 0.7644,
+      "step": 590200
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.2331093952029336e-05,
+      "loss": 0.7566,
+      "step": 590400
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.2321715981117457e-05,
+      "loss": 0.757,
+      "step": 590600
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.2312338010205578e-05,
+      "loss": 0.7632,
+      "step": 590800
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.2302960039293699e-05,
+      "loss": 0.7589,
+      "step": 591000
+    },
+    {
+      "epoch": 0.69,
+      "eval_em_first_line_ratio": 0.4898,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3769,
+      "eval_loss": 0.7350162863731384,
+      "eval_runtime": 197.552,
+      "eval_samples_per_second": 50.62,
+      "eval_steps_per_second": 0.795,
+      "step": 591000
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.229358206838182e-05,
+      "loss": 0.7618,
+      "step": 591200
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.228420409746994e-05,
+      "loss": 0.7605,
+      "step": 591400
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.2274826126558063e-05,
+      "loss": 0.7651,
+      "step": 591600
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.2265448155646184e-05,
+      "loss": 0.7564,
+      "step": 591800
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.2256070184734305e-05,
+      "loss": 0.7576,
+      "step": 592000
+    },
+    {
+      "epoch": 0.69,
+      "eval_em_first_line_ratio": 0.4899,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3768,
+      "eval_loss": 0.7348424792289734,
+      "eval_runtime": 198.4324,
+      "eval_samples_per_second": 50.395,
+      "eval_steps_per_second": 0.791,
+      "step": 592000
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.2246692213822426e-05,
+      "loss": 0.7626,
+      "step": 592200
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.2237314242910547e-05,
+      "loss": 0.7601,
+      "step": 592400
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.2227936271998668e-05,
+      "loss": 0.759,
+      "step": 592600
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.221855830108679e-05,
+      "loss": 0.7615,
+      "step": 592800
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.2209180330174912e-05,
+      "loss": 0.7536,
+      "step": 593000
+    },
+    {
+      "epoch": 0.7,
+      "eval_em_first_line_ratio": 0.4914,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3783,
+      "eval_loss": 0.7349381446838379,
+      "eval_runtime": 203.7546,
+      "eval_samples_per_second": 49.079,
+      "eval_steps_per_second": 0.771,
+      "step": 593000
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.2199802359263033e-05,
+      "loss": 0.7573,
+      "step": 593200
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.2190424388351154e-05,
+      "loss": 0.7594,
+      "step": 593400
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.2181046417439275e-05,
+      "loss": 0.7591,
+      "step": 593600
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.2171668446527396e-05,
+      "loss": 0.758,
+      "step": 593800
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.2162337365470078e-05,
+      "loss": 0.7597,
+      "step": 594000
+    },
+    {
+      "epoch": 0.7,
+      "eval_em_first_line_ratio": 0.493,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3795,
+      "eval_loss": 0.7344464063644409,
+      "eval_runtime": 204.3732,
+      "eval_samples_per_second": 48.93,
+      "eval_steps_per_second": 0.768,
+      "step": 594000
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.21529593945582e-05,
+      "loss": 0.7606,
+      "step": 594200
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.214358142364632e-05,
+      "loss": 0.76,
+      "step": 594400
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.2134203452734441e-05,
+      "loss": 0.7585,
+      "step": 594600
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.2124825481822562e-05,
+      "loss": 0.7574,
+      "step": 594800
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.2115447510910685e-05,
+      "loss": 0.7605,
+      "step": 595000
+    },
+    {
+      "epoch": 0.7,
+      "eval_em_first_line_ratio": 0.4904,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3771,
+      "eval_loss": 0.7344605326652527,
+      "eval_runtime": 202.3883,
+      "eval_samples_per_second": 49.41,
+      "eval_steps_per_second": 0.776,
+      "step": 595000
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.2106069539998806e-05,
+      "loss": 0.7583,
+      "step": 595200
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.2096691569086927e-05,
+      "loss": 0.7574,
+      "step": 595400
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.2087313598175048e-05,
+      "loss": 0.7611,
+      "step": 595600
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.2077935627263169e-05,
+      "loss": 0.7587,
+      "step": 595800
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.206855765635129e-05,
+      "loss": 0.7537,
+      "step": 596000
+    },
+    {
+      "epoch": 0.7,
+      "eval_em_first_line_ratio": 0.4921,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3785,
+      "eval_loss": 0.7343356609344482,
+      "eval_runtime": 200.5386,
+      "eval_samples_per_second": 49.866,
+      "eval_steps_per_second": 0.783,
+      "step": 596000
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.2059226575293972e-05,
+      "loss": 0.7602,
+      "step": 596200
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.2049848604382093e-05,
+      "loss": 0.7621,
+      "step": 596400
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.2040517523324772e-05,
+      "loss": 0.7599,
+      "step": 596600
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.2031139552412893e-05,
+      "loss": 0.7584,
+      "step": 596800
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.2021761581501014e-05,
+      "loss": 0.7579,
+      "step": 597000
+    },
+    {
+      "epoch": 0.7,
+      "eval_em_first_line_ratio": 0.4917,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3784,
+      "eval_loss": 0.7344061136245728,
+      "eval_runtime": 205.1822,
+      "eval_samples_per_second": 48.737,
+      "eval_steps_per_second": 0.765,
+      "step": 597000
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.2012383610589139e-05,
+      "loss": 0.7593,
+      "step": 597200
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.200300563967726e-05,
+      "loss": 0.76,
+      "step": 597400
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.199362766876538e-05,
+      "loss": 0.7634,
+      "step": 597600
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.19842496978535e-05,
+      "loss": 0.7592,
+      "step": 597800
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.197487172694162e-05,
+      "loss": 0.7557,
+      "step": 598000
+    },
+    {
+      "epoch": 0.7,
+      "eval_em_first_line_ratio": 0.4922,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3789,
+      "eval_loss": 0.7342783212661743,
+      "eval_runtime": 202.5235,
+      "eval_samples_per_second": 49.377,
+      "eval_steps_per_second": 0.775,
+      "step": 598000
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.1965493756029742e-05,
+      "loss": 0.7556,
+      "step": 598200
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.1956115785117866e-05,
+      "loss": 0.7594,
+      "step": 598400
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.1946737814205987e-05,
+      "loss": 0.7617,
+      "step": 598600
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.1937359843294108e-05,
+      "loss": 0.759,
+      "step": 598800
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.1927981872382229e-05,
+      "loss": 0.7618,
+      "step": 599000
+    },
+    {
+      "epoch": 0.7,
+      "eval_em_first_line_ratio": 0.4925,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3789,
+      "eval_loss": 0.7345101833343506,
+      "eval_runtime": 201.6275,
+      "eval_samples_per_second": 49.596,
+      "eval_steps_per_second": 0.779,
+      "step": 599000
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.191860390147035e-05,
+      "loss": 0.7558,
+      "step": 599200
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.1909225930558471e-05,
+      "loss": 0.7615,
+      "step": 599400
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.1899847959646594e-05,
+      "loss": 0.7605,
+      "step": 599600
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.1890469988734715e-05,
+      "loss": 0.7602,
+      "step": 599800
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.1881092017822836e-05,
+      "loss": 0.7559,
+      "step": 600000
+    },
+    {
+      "epoch": 0.7,
+      "eval_em_first_line_ratio": 0.4914,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3784,
+      "eval_loss": 0.7341764569282532,
+      "eval_runtime": 196.9609,
+      "eval_samples_per_second": 50.771,
+      "eval_steps_per_second": 0.797,
+      "step": 600000
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.1871714046910957e-05,
+      "loss": 0.7545,
+      "step": 600200
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.1862336075999077e-05,
+      "loss": 0.7557,
+      "step": 600400
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.1852958105087198e-05,
+      "loss": 0.7582,
+      "step": 600600
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.184358013417532e-05,
+      "loss": 0.7599,
+      "step": 600800
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.1834202163263442e-05,
+      "loss": 0.7593,
+      "step": 601000
+    },
+    {
+      "epoch": 0.7,
+      "eval_em_first_line_ratio": 0.4928,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3793,
+      "eval_loss": 0.7344468235969543,
+      "eval_runtime": 196.4037,
+      "eval_samples_per_second": 50.916,
+      "eval_steps_per_second": 0.799,
+      "step": 601000
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.1824871082206121e-05,
+      "loss": 0.7553,
+      "step": 601200
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.1815493111294242e-05,
+      "loss": 0.7591,
+      "step": 601400
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.1806115140382363e-05,
+      "loss": 0.7577,
+      "step": 601600
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.1796737169470484e-05,
+      "loss": 0.7613,
+      "step": 601800
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.1787359198558608e-05,
+      "loss": 0.7596,
+      "step": 602000
+    },
+    {
+      "epoch": 0.71,
+      "eval_em_first_line_ratio": 0.4911,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3779,
+      "eval_loss": 0.7339087128639221,
+      "eval_runtime": 206.5118,
+      "eval_samples_per_second": 48.423,
+      "eval_steps_per_second": 0.76,
+      "step": 602000
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.177798122764673e-05,
+      "loss": 0.7569,
+      "step": 602200
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.176860325673485e-05,
+      "loss": 0.755,
+      "step": 602400
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.175922528582297e-05,
+      "loss": 0.7546,
+      "step": 602600
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.174984731491109e-05,
+      "loss": 0.7611,
+      "step": 602800
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.1740469343999212e-05,
+      "loss": 0.7581,
+      "step": 603000
+    },
+    {
+      "epoch": 0.71,
+      "eval_em_first_line_ratio": 0.4921,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3789,
+      "eval_loss": 0.7338855266571045,
+      "eval_runtime": 203.0004,
+      "eval_samples_per_second": 49.261,
+      "eval_steps_per_second": 0.773,
+      "step": 603000
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.1731091373087336e-05,
+      "loss": 0.7602,
+      "step": 603200
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.1721713402175457e-05,
+      "loss": 0.7637,
+      "step": 603400
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.1712382321118136e-05,
+      "loss": 0.7534,
+      "step": 603600
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.1703004350206257e-05,
+      "loss": 0.7592,
+      "step": 603800
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.1693626379294378e-05,
+      "loss": 0.7612,
+      "step": 604000
+    },
+    {
+      "epoch": 0.71,
+      "eval_em_first_line_ratio": 0.4911,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3781,
+      "eval_loss": 0.7340179085731506,
+      "eval_runtime": 202.612,
+      "eval_samples_per_second": 49.355,
+      "eval_steps_per_second": 0.775,
+      "step": 604000
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.1684248408382499e-05,
+      "loss": 0.7671,
+      "step": 604200
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.1674870437470622e-05,
+      "loss": 0.7632,
+      "step": 604400
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.1665492466558743e-05,
+      "loss": 0.7564,
+      "step": 604600
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.1656114495646864e-05,
+      "loss": 0.7589,
+      "step": 604800
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.1646736524734985e-05,
+      "loss": 0.7588,
+      "step": 605000
+    },
+    {
+      "epoch": 0.71,
+      "eval_em_first_line_ratio": 0.4919,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3784,
+      "eval_loss": 0.7339382767677307,
+      "eval_runtime": 199.0459,
+      "eval_samples_per_second": 50.24,
+      "eval_steps_per_second": 0.789,
+      "step": 605000
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.1637358553823106e-05,
+      "loss": 0.7591,
+      "step": 605200
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.1627980582911227e-05,
+      "loss": 0.7602,
+      "step": 605400
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.161860261199935e-05,
+      "loss": 0.7591,
+      "step": 605600
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.160922464108747e-05,
+      "loss": 0.761,
+      "step": 605800
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.1599893560030151e-05,
+      "loss": 0.7585,
+      "step": 606000
+    },
+    {
+      "epoch": 0.71,
+      "eval_em_first_line_ratio": 0.491,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3781,
+      "eval_loss": 0.7336276173591614,
+      "eval_runtime": 200.1967,
+      "eval_samples_per_second": 49.951,
+      "eval_steps_per_second": 0.784,
+      "step": 606000
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.1590515589118272e-05,
+      "loss": 0.7633,
+      "step": 606200
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.1581137618206393e-05,
+      "loss": 0.7651,
+      "step": 606400
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.1571759647294516e-05,
+      "loss": 0.7549,
+      "step": 606600
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.1562428566237196e-05,
+      "loss": 0.7598,
+      "step": 606800
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.1553050595325317e-05,
+      "loss": 0.758,
+      "step": 607000
+    },
+    {
+      "epoch": 0.71,
+      "eval_em_first_line_ratio": 0.4926,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3791,
+      "eval_loss": 0.7331206202507019,
+      "eval_runtime": 200.9434,
+      "eval_samples_per_second": 49.765,
+      "eval_steps_per_second": 0.781,
+      "step": 607000
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.1543672624413438e-05,
+      "loss": 0.7607,
+      "step": 607200
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.153429465350156e-05,
+      "loss": 0.7609,
+      "step": 607400
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.152491668258968e-05,
+      "loss": 0.7621,
+      "step": 607600
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.1515538711677803e-05,
+      "loss": 0.7555,
+      "step": 607800
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.1506160740765924e-05,
+      "loss": 0.7614,
+      "step": 608000
+    },
+    {
+      "epoch": 0.71,
+      "eval_em_first_line_ratio": 0.4906,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3776,
+      "eval_loss": 0.7338914275169373,
+      "eval_runtime": 194.2535,
+      "eval_samples_per_second": 51.479,
+      "eval_steps_per_second": 0.808,
+      "step": 608000
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.1496782769854045e-05,
+      "loss": 0.7548,
+      "step": 608200
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.1487404798942166e-05,
+      "loss": 0.7677,
+      "step": 608400
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.1478026828030287e-05,
+      "loss": 0.7613,
+      "step": 608600
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.1468648857118408e-05,
+      "loss": 0.7557,
+      "step": 608800
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.145931777606109e-05,
+      "loss": 0.7572,
+      "step": 609000
+    },
+    {
+      "epoch": 0.71,
+      "eval_em_first_line_ratio": 0.4921,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3789,
+      "eval_loss": 0.7334185838699341,
+      "eval_runtime": 204.2335,
+      "eval_samples_per_second": 48.964,
+      "eval_steps_per_second": 0.769,
+      "step": 609000
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.1449939805149211e-05,
+      "loss": 0.7545,
+      "step": 609200
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.144056183423733e-05,
+      "loss": 0.758,
+      "step": 609400
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.1431183863325452e-05,
+      "loss": 0.7585,
+      "step": 609600
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.1421805892413573e-05,
+      "loss": 0.7581,
+      "step": 609800
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.1412474811356255e-05,
+      "loss": 0.7629,
+      "step": 610000
+    },
+    {
+      "epoch": 0.71,
+      "eval_em_first_line_ratio": 0.4916,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3782,
+      "eval_loss": 0.7332810759544373,
+      "eval_runtime": 203.1048,
+      "eval_samples_per_second": 49.236,
+      "eval_steps_per_second": 0.773,
+      "step": 610000
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.1403096840444376e-05,
+      "loss": 0.7532,
+      "step": 610200
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.1393718869532497e-05,
+      "loss": 0.7562,
+      "step": 610400
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.1384340898620618e-05,
+      "loss": 0.7556,
+      "step": 610600
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.1374962927708739e-05,
+      "loss": 0.7599,
+      "step": 610800
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.136558495679686e-05,
+      "loss": 0.759,
+      "step": 611000
+    },
+    {
+      "epoch": 0.72,
+      "eval_em_first_line_ratio": 0.4908,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3776,
+      "eval_loss": 0.7334024906158447,
+      "eval_runtime": 202.4093,
+      "eval_samples_per_second": 49.405,
+      "eval_steps_per_second": 0.776,
+      "step": 611000
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.1356206985884983e-05,
+      "loss": 0.761,
+      "step": 611200
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.1346829014973104e-05,
+      "loss": 0.7528,
+      "step": 611400
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.1337451044061225e-05,
+      "loss": 0.7586,
+      "step": 611600
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.1328073073149346e-05,
+      "loss": 0.7543,
+      "step": 611800
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.1318695102237466e-05,
+      "loss": 0.7572,
+      "step": 612000
+    },
+    {
+      "epoch": 0.72,
+      "eval_em_first_line_ratio": 0.4918,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3787,
+      "eval_loss": 0.7333288788795471,
+      "eval_runtime": 196.4215,
+      "eval_samples_per_second": 50.911,
+      "eval_steps_per_second": 0.799,
+      "step": 612000
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.1309317131325587e-05,
+      "loss": 0.7604,
+      "step": 612200
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.129993916041371e-05,
+      "loss": 0.762,
+      "step": 612400
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.1290561189501831e-05,
+      "loss": 0.7519,
+      "step": 612600
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.1281183218589952e-05,
+      "loss": 0.7575,
+      "step": 612800
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.1271805247678073e-05,
+      "loss": 0.7556,
+      "step": 613000
+    },
+    {
+      "epoch": 0.72,
+      "eval_em_first_line_ratio": 0.492,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3792,
+      "eval_loss": 0.7332608699798584,
+      "eval_runtime": 204.4587,
+      "eval_samples_per_second": 48.91,
+      "eval_steps_per_second": 0.768,
+      "step": 613000
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.1262427276766194e-05,
+      "loss": 0.7638,
+      "step": 613200
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.1253049305854315e-05,
+      "loss": 0.7566,
+      "step": 613400
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.124367133494244e-05,
+      "loss": 0.7615,
+      "step": 613600
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.123429336403056e-05,
+      "loss": 0.7562,
+      "step": 613800
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.1224915393118681e-05,
+      "loss": 0.7594,
+      "step": 614000
+    },
+    {
+      "epoch": 0.72,
+      "eval_em_first_line_ratio": 0.4949,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3813,
+      "eval_loss": 0.7333053946495056,
+      "eval_runtime": 199.7561,
+      "eval_samples_per_second": 50.061,
+      "eval_steps_per_second": 0.786,
+      "step": 614000
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.12155374222068e-05,
+      "loss": 0.7547,
+      "step": 614200
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.1206159451294922e-05,
+      "loss": 0.7658,
+      "step": 614400
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.1196781480383043e-05,
+      "loss": 0.7581,
+      "step": 614600
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.1187403509471167e-05,
+      "loss": 0.7587,
+      "step": 614800
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.1178025538559288e-05,
+      "loss": 0.7606,
+      "step": 615000
+    },
+    {
+      "epoch": 0.72,
+      "eval_em_first_line_ratio": 0.4935,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.38,
+      "eval_loss": 0.7327088117599487,
+      "eval_runtime": 199.6845,
+      "eval_samples_per_second": 50.079,
+      "eval_steps_per_second": 0.786,
+      "step": 615000
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.1168647567647409e-05,
+      "loss": 0.7557,
+      "step": 615200
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.1159316486590088e-05,
+      "loss": 0.755,
+      "step": 615400
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.1149938515678209e-05,
+      "loss": 0.7517,
+      "step": 615600
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.114056054476633e-05,
+      "loss": 0.7566,
+      "step": 615800
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.1131182573854453e-05,
+      "loss": 0.7546,
+      "step": 616000
+    },
+    {
+      "epoch": 0.72,
+      "eval_em_first_line_ratio": 0.4927,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3795,
+      "eval_loss": 0.7333514094352722,
+      "eval_runtime": 201.5596,
+      "eval_samples_per_second": 49.613,
+      "eval_steps_per_second": 0.779,
+      "step": 616000
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.1121804602942574e-05,
+      "loss": 0.7593,
+      "step": 616200
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.1112426632030695e-05,
+      "loss": 0.7608,
+      "step": 616400
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.1103048661118815e-05,
+      "loss": 0.7602,
+      "step": 616600
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.1093670690206936e-05,
+      "loss": 0.754,
+      "step": 616800
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.1084292719295057e-05,
+      "loss": 0.7592,
+      "step": 617000
+    },
+    {
+      "epoch": 0.72,
+      "eval_em_first_line_ratio": 0.4923,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3788,
+      "eval_loss": 0.7331268787384033,
+      "eval_runtime": 199.332,
+      "eval_samples_per_second": 50.168,
+      "eval_steps_per_second": 0.788,
+      "step": 617000
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.107491474838318e-05,
+      "loss": 0.7583,
+      "step": 617200
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.1065536777471301e-05,
+      "loss": 0.7529,
+      "step": 617400
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.1056158806559422e-05,
+      "loss": 0.7603,
+      "step": 617600
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.1046780835647543e-05,
+      "loss": 0.7626,
+      "step": 617800
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.1037402864735664e-05,
+      "loss": 0.7615,
+      "step": 618000
+    },
+    {
+      "epoch": 0.72,
+      "eval_em_first_line_ratio": 0.4913,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3785,
+      "eval_loss": 0.7325000762939453,
+      "eval_runtime": 199.4968,
+      "eval_samples_per_second": 50.126,
+      "eval_steps_per_second": 0.787,
+      "step": 618000
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.1028024893823785e-05,
+      "loss": 0.7594,
+      "step": 618200
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.101864692291191e-05,
+      "loss": 0.7576,
+      "step": 618400
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1.100926895200003e-05,
+      "loss": 0.7559,
+      "step": 618600
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1.0999890981088151e-05,
+      "loss": 0.7551,
+      "step": 618800
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1.099055990003083e-05,
+      "loss": 0.7591,
+      "step": 619000
+    },
+    {
+      "epoch": 0.73,
+      "eval_em_first_line_ratio": 0.493,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3795,
+      "eval_loss": 0.7327407002449036,
+      "eval_runtime": 197.1415,
+      "eval_samples_per_second": 50.725,
+      "eval_steps_per_second": 0.796,
+      "step": 619000
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1.0981181929118951e-05,
+      "loss": 0.7575,
+      "step": 619200
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1.0971803958207074e-05,
+      "loss": 0.7638,
+      "step": 619400
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1.0962425987295195e-05,
+      "loss": 0.7605,
+      "step": 619600
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1.0953048016383316e-05,
+      "loss": 0.7576,
+      "step": 619800
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1.0943670045471437e-05,
+      "loss": 0.76,
+      "step": 620000
+    },
+    {
+      "epoch": 0.73,
+      "eval_em_first_line_ratio": 0.4897,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3768,
+      "eval_loss": 0.7329027652740479,
+      "eval_runtime": 201.4077,
+      "eval_samples_per_second": 49.651,
+      "eval_steps_per_second": 0.78,
+      "step": 620000
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1.0934292074559558e-05,
+      "loss": 0.7546,
+      "step": 620200
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1.0924914103647679e-05,
+      "loss": 0.7607,
+      "step": 620400
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1.0915536132735802e-05,
+      "loss": 0.7543,
+      "step": 620600
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1.0906158161823923e-05,
+      "loss": 0.7542,
+      "step": 620800
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1.0896780190912044e-05,
+      "loss": 0.7502,
+      "step": 621000
+    },
+    {
+      "epoch": 0.73,
+      "eval_em_first_line_ratio": 0.4922,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3788,
+      "eval_loss": 0.7325568199157715,
+      "eval_runtime": 198.2591,
+      "eval_samples_per_second": 50.439,
+      "eval_steps_per_second": 0.792,
+      "step": 621000
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1.0887402220000164e-05,
+      "loss": 0.7572,
+      "step": 621200
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1.0878024249088285e-05,
+      "loss": 0.7591,
+      "step": 621400
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1.0868646278176406e-05,
+      "loss": 0.7578,
+      "step": 621600
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1.0859268307264527e-05,
+      "loss": 0.7595,
+      "step": 621800
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1.084989033635265e-05,
+      "loss": 0.7571,
+      "step": 622000
+    },
+    {
+      "epoch": 0.73,
+      "eval_em_first_line_ratio": 0.4931,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3801,
+      "eval_loss": 0.7330873012542725,
+      "eval_runtime": 202.0203,
+      "eval_samples_per_second": 49.5,
+      "eval_steps_per_second": 0.777,
+      "step": 622000
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1.0840512365440771e-05,
+      "loss": 0.7538,
+      "step": 622200
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1.0831134394528892e-05,
+      "loss": 0.7606,
+      "step": 622400
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1.0821756423617013e-05,
+      "loss": 0.7524,
+      "step": 622600
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1.0812378452705134e-05,
+      "loss": 0.7555,
+      "step": 622800
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1.0803000481793255e-05,
+      "loss": 0.7587,
+      "step": 623000
+    },
+    {
+      "epoch": 0.73,
+      "eval_em_first_line_ratio": 0.4922,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3787,
+      "eval_loss": 0.7327051758766174,
+      "eval_runtime": 198.0599,
+      "eval_samples_per_second": 50.49,
+      "eval_steps_per_second": 0.793,
+      "step": 623000
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1.079362251088138e-05,
+      "loss": 0.7533,
+      "step": 623200
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1.07842445399695e-05,
+      "loss": 0.7571,
+      "step": 623400
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1.077496034876674e-05,
+      "loss": 0.7558,
+      "step": 623600
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1.076558237785486e-05,
+      "loss": 0.7501,
+      "step": 623800
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1.0756204406942983e-05,
+      "loss": 0.76,
+      "step": 624000
+    },
+    {
+      "epoch": 0.73,
+      "eval_em_first_line_ratio": 0.4946,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3812,
+      "eval_loss": 0.7324507832527161,
+      "eval_runtime": 201.7742,
+      "eval_samples_per_second": 49.56,
+      "eval_steps_per_second": 0.778,
+      "step": 624000
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1.0746826436031104e-05,
+      "loss": 0.7572,
+      "step": 624200
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1.0737495354973783e-05,
+      "loss": 0.7533,
+      "step": 624400
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1.0728117384061904e-05,
+      "loss": 0.7518,
+      "step": 624600
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1.0718739413150025e-05,
+      "loss": 0.7544,
+      "step": 624800
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1.0709361442238146e-05,
+      "loss": 0.7543,
+      "step": 625000
+    },
+    {
+      "epoch": 0.73,
+      "eval_em_first_line_ratio": 0.4914,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3781,
+      "eval_loss": 0.7328659892082214,
+      "eval_runtime": 202.1254,
+      "eval_samples_per_second": 49.474,
+      "eval_steps_per_second": 0.777,
+      "step": 625000
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1.069998347132627e-05,
+      "loss": 0.7566,
+      "step": 625200
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1.0690605500414391e-05,
+      "loss": 0.759,
+      "step": 625400
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1.0681227529502512e-05,
+      "loss": 0.7546,
+      "step": 625600
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1.0671849558590633e-05,
+      "loss": 0.7581,
+      "step": 625800
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1.0662471587678752e-05,
+      "loss": 0.756,
+      "step": 626000
+    },
+    {
+      "epoch": 0.73,
+      "eval_em_first_line_ratio": 0.4929,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3796,
+      "eval_loss": 0.7322419285774231,
+      "eval_runtime": 203.9146,
+      "eval_samples_per_second": 49.04,
+      "eval_steps_per_second": 0.77,
+      "step": 626000
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1.0653093616766873e-05,
+      "loss": 0.7574,
+      "step": 626200
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1.0643715645854998e-05,
+      "loss": 0.7546,
+      "step": 626400
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1.0634337674943119e-05,
+      "loss": 0.7529,
+      "step": 626600
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1.062495970403124e-05,
+      "loss": 0.7612,
+      "step": 626800
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1.061558173311936e-05,
+      "loss": 0.7575,
+      "step": 627000
+    },
+    {
+      "epoch": 0.73,
+      "eval_em_first_line_ratio": 0.4922,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.379,
+      "eval_loss": 0.7320851683616638,
+      "eval_runtime": 201.4565,
+      "eval_samples_per_second": 49.639,
+      "eval_steps_per_second": 0.779,
+      "step": 627000
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.0606203762207482e-05,
+      "loss": 0.762,
+      "step": 627200
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.0596825791295603e-05,
+      "loss": 0.7544,
+      "step": 627400
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.0587494710238283e-05,
+      "loss": 0.7583,
+      "step": 627600
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.0578116739326404e-05,
+      "loss": 0.7557,
+      "step": 627800
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.0568738768414525e-05,
+      "loss": 0.7546,
+      "step": 628000
+    },
+    {
+      "epoch": 0.74,
+      "eval_em_first_line_ratio": 0.4934,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3807,
+      "eval_loss": 0.7325658798217773,
+      "eval_runtime": 204.712,
+      "eval_samples_per_second": 48.849,
+      "eval_steps_per_second": 0.767,
+      "step": 628000
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.0559360797502646e-05,
+      "loss": 0.7538,
+      "step": 628200
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.0549982826590767e-05,
+      "loss": 0.7565,
+      "step": 628400
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.0540604855678888e-05,
+      "loss": 0.7551,
+      "step": 628600
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.0531226884767013e-05,
+      "loss": 0.7568,
+      "step": 628800
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.0521848913855132e-05,
+      "loss": 0.7602,
+      "step": 629000
+    },
+    {
+      "epoch": 0.74,
+      "eval_em_first_line_ratio": 0.493,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3799,
+      "eval_loss": 0.7325807809829712,
+      "eval_runtime": 200.4549,
+      "eval_samples_per_second": 49.887,
+      "eval_steps_per_second": 0.783,
+      "step": 629000
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.0512470942943253e-05,
+      "loss": 0.7548,
+      "step": 629200
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.0503092972031374e-05,
+      "loss": 0.757,
+      "step": 629400
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.0493715001119495e-05,
+      "loss": 0.7607,
+      "step": 629600
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.0484337030207616e-05,
+      "loss": 0.7579,
+      "step": 629800
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.047495905929574e-05,
+      "loss": 0.7552,
+      "step": 630000
+    },
+    {
+      "epoch": 0.74,
+      "eval_em_first_line_ratio": 0.4933,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3803,
+      "eval_loss": 0.731943666934967,
+      "eval_runtime": 196.7045,
+      "eval_samples_per_second": 50.838,
+      "eval_steps_per_second": 0.798,
+      "step": 630000
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.0465581088383861e-05,
+      "loss": 0.7563,
+      "step": 630200
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.0456203117471982e-05,
+      "loss": 0.7578,
+      "step": 630400
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.0446825146560103e-05,
+      "loss": 0.7566,
+      "step": 630600
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.0437447175648222e-05,
+      "loss": 0.7581,
+      "step": 630800
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.0428069204736343e-05,
+      "loss": 0.7576,
+      "step": 631000
+    },
+    {
+      "epoch": 0.74,
+      "eval_em_first_line_ratio": 0.4941,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3813,
+      "eval_loss": 0.7317689657211304,
+      "eval_runtime": 199.2419,
+      "eval_samples_per_second": 50.19,
+      "eval_steps_per_second": 0.788,
+      "step": 631000
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.0418691233824468e-05,
+      "loss": 0.7606,
+      "step": 631200
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.0409313262912589e-05,
+      "loss": 0.7538,
+      "step": 631400
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.039993529200071e-05,
+      "loss": 0.7522,
+      "step": 631600
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.039055732108883e-05,
+      "loss": 0.7552,
+      "step": 631800
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.0381179350176952e-05,
+      "loss": 0.7591,
+      "step": 632000
+    },
+    {
+      "epoch": 0.74,
+      "eval_em_first_line_ratio": 0.4945,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3813,
+      "eval_loss": 0.7318770289421082,
+      "eval_runtime": 200.3267,
+      "eval_samples_per_second": 49.918,
+      "eval_steps_per_second": 0.784,
+      "step": 632000
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.0371801379265073e-05,
+      "loss": 0.7591,
+      "step": 632200
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.0362423408353195e-05,
+      "loss": 0.7553,
+      "step": 632400
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.0353045437441316e-05,
+      "loss": 0.7613,
+      "step": 632600
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.0343667466529437e-05,
+      "loss": 0.7593,
+      "step": 632800
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.0334289495617558e-05,
+      "loss": 0.7605,
+      "step": 633000
+    },
+    {
+      "epoch": 0.74,
+      "eval_em_first_line_ratio": 0.4934,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3801,
+      "eval_loss": 0.7317349314689636,
+      "eval_runtime": 203.2195,
+      "eval_samples_per_second": 49.208,
+      "eval_steps_per_second": 0.773,
+      "step": 633000
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.0324911524705679e-05,
+      "loss": 0.7571,
+      "step": 633200
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.03155335537938e-05,
+      "loss": 0.7576,
+      "step": 633400
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.0306155582881923e-05,
+      "loss": 0.7612,
+      "step": 633600
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.0296777611970044e-05,
+      "loss": 0.7595,
+      "step": 633800
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.0287399641058165e-05,
+      "loss": 0.7563,
+      "step": 634000
+    },
+    {
+      "epoch": 0.74,
+      "eval_em_first_line_ratio": 0.4927,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3797,
+      "eval_loss": 0.7320752143859863,
+      "eval_runtime": 200.5688,
+      "eval_samples_per_second": 49.858,
+      "eval_steps_per_second": 0.783,
+      "step": 634000
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.0278068560000844e-05,
+      "loss": 0.7601,
+      "step": 634200
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.0268690589088965e-05,
+      "loss": 0.7542,
+      "step": 634400
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.0259312618177086e-05,
+      "loss": 0.7532,
+      "step": 634600
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.024993464726521e-05,
+      "loss": 0.7505,
+      "step": 634800
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.0240556676353331e-05,
+      "loss": 0.7553,
+      "step": 635000
+    },
+    {
+      "epoch": 0.74,
+      "eval_em_first_line_ratio": 0.4917,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3788,
+      "eval_loss": 0.7315892577171326,
+      "eval_runtime": 201.3461,
+      "eval_samples_per_second": 49.666,
+      "eval_steps_per_second": 0.78,
+      "step": 635000
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.0231178705441452e-05,
+      "loss": 0.7564,
+      "step": 635200
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.0221800734529571e-05,
+      "loss": 0.7541,
+      "step": 635400
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.0212422763617692e-05,
+      "loss": 0.7591,
+      "step": 635600
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 1.0203044792705813e-05,
+      "loss": 0.7599,
+      "step": 635800
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 1.0193666821793938e-05,
+      "loss": 0.7518,
+      "step": 636000
+    },
+    {
+      "epoch": 0.75,
+      "eval_em_first_line_ratio": 0.4917,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3783,
+      "eval_loss": 0.7315619587898254,
+      "eval_runtime": 209.089,
+      "eval_samples_per_second": 47.827,
+      "eval_steps_per_second": 0.751,
+      "step": 636000
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 1.0184288850882059e-05,
+      "loss": 0.7596,
+      "step": 636200
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 1.017491087997018e-05,
+      "loss": 0.7596,
+      "step": 636400
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 1.01655329090583e-05,
+      "loss": 0.7543,
+      "step": 636600
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 1.0156154938146422e-05,
+      "loss": 0.752,
+      "step": 636800
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 1.0146823857089102e-05,
+      "loss": 0.7511,
+      "step": 637000
+    },
+    {
+      "epoch": 0.75,
+      "eval_em_first_line_ratio": 0.4912,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3783,
+      "eval_loss": 0.7317458391189575,
+      "eval_runtime": 203.0644,
+      "eval_samples_per_second": 49.245,
+      "eval_steps_per_second": 0.773,
+      "step": 637000
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 1.0137445886177223e-05,
+      "loss": 0.7556,
+      "step": 637200
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 1.0128067915265344e-05,
+      "loss": 0.7553,
+      "step": 637400
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 1.0118689944353465e-05,
+      "loss": 0.755,
+      "step": 637600
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 1.0109311973441586e-05,
+      "loss": 0.7554,
+      "step": 637800
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 1.0099934002529707e-05,
+      "loss": 0.7568,
+      "step": 638000
+    },
+    {
+      "epoch": 0.75,
+      "eval_em_first_line_ratio": 0.4926,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3793,
+      "eval_loss": 0.7315488457679749,
+      "eval_runtime": 205.5407,
+      "eval_samples_per_second": 48.652,
+      "eval_steps_per_second": 0.764,
+      "step": 638000
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 1.0090556031617832e-05,
+      "loss": 0.7513,
+      "step": 638200
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 1.008122495056051e-05,
+      "loss": 0.7571,
+      "step": 638400
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 1.0071846979648632e-05,
+      "loss": 0.7565,
+      "step": 638600
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 1.0062469008736753e-05,
+      "loss": 0.7552,
+      "step": 638800
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 1.0053091037824874e-05,
+      "loss": 0.7526,
+      "step": 639000
+    },
+    {
+      "epoch": 0.75,
+      "eval_em_first_line_ratio": 0.4911,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3779,
+      "eval_loss": 0.7312169671058655,
+      "eval_runtime": 194.7597,
+      "eval_samples_per_second": 51.345,
+      "eval_steps_per_second": 0.806,
+      "step": 639000
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 1.0043713066912995e-05,
+      "loss": 0.7513,
+      "step": 639200
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 1.0034335096001117e-05,
+      "loss": 0.7616,
+      "step": 639400
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 1.0024957125089238e-05,
+      "loss": 0.7554,
+      "step": 639600
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 1.001557915417736e-05,
+      "loss": 0.7578,
+      "step": 639800
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 1.000620118326548e-05,
+      "loss": 0.7532,
+      "step": 640000
+    },
+    {
+      "epoch": 0.75,
+      "eval_em_first_line_ratio": 0.4926,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3794,
+      "eval_loss": 0.7313681840896606,
+      "eval_runtime": 197.2836,
+      "eval_samples_per_second": 50.688,
+      "eval_steps_per_second": 0.796,
+      "step": 640000
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 9.996823212353603e-06,
+      "loss": 0.7544,
+      "step": 640200
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 9.987445241441724e-06,
+      "loss": 0.7571,
+      "step": 640400
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 9.978067270529845e-06,
+      "loss": 0.7574,
+      "step": 640600
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 9.968689299617966e-06,
+      "loss": 0.7491,
+      "step": 640800
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 9.959311328706087e-06,
+      "loss": 0.7538,
+      "step": 641000
+    },
+    {
+      "epoch": 0.75,
+      "eval_em_first_line_ratio": 0.4926,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3796,
+      "eval_loss": 0.7314398288726807,
+      "eval_runtime": 200.9573,
+      "eval_samples_per_second": 49.762,
+      "eval_steps_per_second": 0.781,
+      "step": 641000
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 9.949933357794208e-06,
+      "loss": 0.755,
+      "step": 641200
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 9.940602276736889e-06,
+      "loss": 0.7527,
+      "step": 641400
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 9.93122430582501e-06,
+      "loss": 0.7565,
+      "step": 641600
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 9.92184633491313e-06,
+      "loss": 0.7528,
+      "step": 641800
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 9.912515253855813e-06,
+      "loss": 0.7536,
+      "step": 642000
+    },
+    {
+      "epoch": 0.75,
+      "eval_em_first_line_ratio": 0.4925,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3792,
+      "eval_loss": 0.7314226627349854,
+      "eval_runtime": 196.253,
+      "eval_samples_per_second": 50.955,
+      "eval_steps_per_second": 0.8,
+      "step": 642000
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 9.903137282943934e-06,
+      "loss": 0.7538,
+      "step": 642200
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 9.893759312032055e-06,
+      "loss": 0.7603,
+      "step": 642400
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 9.884381341120176e-06,
+      "loss": 0.7503,
+      "step": 642600
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 9.875003370208297e-06,
+      "loss": 0.7506,
+      "step": 642800
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 9.86562539929642e-06,
+      "loss": 0.7567,
+      "step": 643000
+    },
+    {
+      "epoch": 0.75,
+      "eval_em_first_line_ratio": 0.493,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3794,
+      "eval_loss": 0.7310846447944641,
+      "eval_runtime": 197.0866,
+      "eval_samples_per_second": 50.739,
+      "eval_steps_per_second": 0.797,
+      "step": 643000
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 9.85624742838454e-06,
+      "loss": 0.7569,
+      "step": 643200
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 9.846869457472662e-06,
+      "loss": 0.7531,
+      "step": 643400
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 9.837491486560782e-06,
+      "loss": 0.7577,
+      "step": 643600
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 9.828113515648903e-06,
+      "loss": 0.7583,
+      "step": 643800
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 9.818735544737024e-06,
+      "loss": 0.7573,
+      "step": 644000
+    },
+    {
+      "epoch": 0.75,
+      "eval_em_first_line_ratio": 0.4926,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3796,
+      "eval_loss": 0.7312988042831421,
+      "eval_runtime": 199.0764,
+      "eval_samples_per_second": 50.232,
+      "eval_steps_per_second": 0.789,
+      "step": 644000
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 9.809357573825145e-06,
+      "loss": 0.7542,
+      "step": 644200
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 9.799979602913268e-06,
+      "loss": 0.7582,
+      "step": 644400
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 9.790601632001389e-06,
+      "loss": 0.7528,
+      "step": 644600
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 9.78122366108951e-06,
+      "loss": 0.7556,
+      "step": 644800
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 9.771845690177631e-06,
+      "loss": 0.76,
+      "step": 645000
+    },
+    {
+      "epoch": 0.76,
+      "eval_em_first_line_ratio": 0.491,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3778,
+      "eval_loss": 0.7311636805534363,
+      "eval_runtime": 198.3051,
+      "eval_samples_per_second": 50.427,
+      "eval_steps_per_second": 0.792,
+      "step": 645000
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 9.762467719265752e-06,
+      "loss": 0.7551,
+      "step": 645200
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 9.753089748353873e-06,
+      "loss": 0.755,
+      "step": 645400
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 9.743711777441996e-06,
+      "loss": 0.7535,
+      "step": 645600
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 9.734333806530117e-06,
+      "loss": 0.7521,
+      "step": 645800
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 9.724955835618238e-06,
+      "loss": 0.7555,
+      "step": 646000
+    },
+    {
+      "epoch": 0.76,
+      "eval_em_first_line_ratio": 0.4911,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3777,
+      "eval_loss": 0.7312288284301758,
+      "eval_runtime": 201.4001,
+      "eval_samples_per_second": 49.652,
+      "eval_steps_per_second": 0.78,
+      "step": 646000
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 9.715577864706359e-06,
+      "loss": 0.7544,
+      "step": 646200
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 9.70619989379448e-06,
+      "loss": 0.7579,
+      "step": 646400
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 9.6968219228826e-06,
+      "loss": 0.7497,
+      "step": 646600
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 9.687443951970723e-06,
+      "loss": 0.7514,
+      "step": 646800
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 9.678065981058844e-06,
+      "loss": 0.7601,
+      "step": 647000
+    },
+    {
+      "epoch": 0.76,
+      "eval_em_first_line_ratio": 0.4939,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3806,
+      "eval_loss": 0.7311167120933533,
+      "eval_runtime": 196.5104,
+      "eval_samples_per_second": 50.888,
+      "eval_steps_per_second": 0.799,
+      "step": 647000
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 9.668688010146965e-06,
+      "loss": 0.756,
+      "step": 647200
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 9.659310039235088e-06,
+      "loss": 0.7542,
+      "step": 647400
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 9.649932068323207e-06,
+      "loss": 0.7562,
+      "step": 647600
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 9.640554097411328e-06,
+      "loss": 0.7485,
+      "step": 647800
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 9.63117612649945e-06,
+      "loss": 0.7578,
+      "step": 648000
+    },
+    {
+      "epoch": 0.76,
+      "eval_em_first_line_ratio": 0.4924,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3794,
+      "eval_loss": 0.7312117218971252,
+      "eval_runtime": 199.9313,
+      "eval_samples_per_second": 50.017,
+      "eval_steps_per_second": 0.785,
+      "step": 648000
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 9.621798155587572e-06,
+      "loss": 0.7562,
+      "step": 648200
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 9.612420184675693e-06,
+      "loss": 0.7523,
+      "step": 648400
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 9.603042213763815e-06,
+      "loss": 0.7525,
+      "step": 648600
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 9.593664242851936e-06,
+      "loss": 0.7596,
+      "step": 648800
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 9.584333161794617e-06,
+      "loss": 0.7546,
+      "step": 649000
+    },
+    {
+      "epoch": 0.76,
+      "eval_em_first_line_ratio": 0.4914,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3783,
+      "eval_loss": 0.7310243248939514,
+      "eval_runtime": 204.8154,
+      "eval_samples_per_second": 48.824,
+      "eval_steps_per_second": 0.767,
+      "step": 649000
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 9.574955190882738e-06,
+      "loss": 0.754,
+      "step": 649200
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 9.565577219970859e-06,
+      "loss": 0.754,
+      "step": 649400
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 9.55619924905898e-06,
+      "loss": 0.7529,
+      "step": 649600
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 9.546821278147101e-06,
+      "loss": 0.7535,
+      "step": 649800
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 9.537443307235222e-06,
+      "loss": 0.755,
+      "step": 650000
+    },
+    {
+      "epoch": 0.76,
+      "eval_em_first_line_ratio": 0.4921,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.379,
+      "eval_loss": 0.7300795316696167,
+      "eval_runtime": 194.9222,
+      "eval_samples_per_second": 51.303,
+      "eval_steps_per_second": 0.805,
+      "step": 650000
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 9.528065336323343e-06,
+      "loss": 0.756,
+      "step": 650200
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 9.518687365411466e-06,
+      "loss": 0.7577,
+      "step": 650400
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 9.509309394499587e-06,
+      "loss": 0.7572,
+      "step": 650600
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 9.499931423587708e-06,
+      "loss": 0.7581,
+      "step": 650800
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 9.490553452675828e-06,
+      "loss": 0.7544,
+      "step": 651000
+    },
+    {
+      "epoch": 0.76,
+      "eval_em_first_line_ratio": 0.4918,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3786,
+      "eval_loss": 0.7306298017501831,
+      "eval_runtime": 203.9243,
+      "eval_samples_per_second": 49.038,
+      "eval_steps_per_second": 0.77,
+      "step": 651000
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 9.48117548176395e-06,
+      "loss": 0.756,
+      "step": 651200
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 9.471844400706632e-06,
+      "loss": 0.7552,
+      "step": 651400
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 9.462466429794753e-06,
+      "loss": 0.7546,
+      "step": 651600
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 9.453088458882874e-06,
+      "loss": 0.7551,
+      "step": 651800
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 9.443710487970995e-06,
+      "loss": 0.7491,
+      "step": 652000
+    },
+    {
+      "epoch": 0.76,
+      "eval_em_first_line_ratio": 0.4925,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3796,
+      "eval_loss": 0.7314202785491943,
+      "eval_runtime": 201.696,
+      "eval_samples_per_second": 49.58,
+      "eval_steps_per_second": 0.778,
+      "step": 652000
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 9.434332517059116e-06,
+      "loss": 0.7516,
+      "step": 652200
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 9.424954546147237e-06,
+      "loss": 0.7554,
+      "step": 652400
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 9.41557657523536e-06,
+      "loss": 0.7581,
+      "step": 652600
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 9.40619860432348e-06,
+      "loss": 0.7587,
+      "step": 652800
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 9.396867523266161e-06,
+      "loss": 0.7579,
+      "step": 653000
+    },
+    {
+      "epoch": 0.77,
+      "eval_em_first_line_ratio": 0.4947,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3814,
+      "eval_loss": 0.7306681871414185,
+      "eval_runtime": 199.1866,
+      "eval_samples_per_second": 50.204,
+      "eval_steps_per_second": 0.788,
+      "step": 653000
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 9.387489552354282e-06,
+      "loss": 0.7496,
+      "step": 653200
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 9.378111581442403e-06,
+      "loss": 0.7543,
+      "step": 653400
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 9.368733610530524e-06,
+      "loss": 0.7508,
+      "step": 653600
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 9.359355639618645e-06,
+      "loss": 0.755,
+      "step": 653800
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 9.349977668706766e-06,
+      "loss": 0.7555,
+      "step": 654000
+    },
+    {
+      "epoch": 0.77,
+      "eval_em_first_line_ratio": 0.4933,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3799,
+      "eval_loss": 0.7304254770278931,
+      "eval_runtime": 200.6017,
+      "eval_samples_per_second": 49.85,
+      "eval_steps_per_second": 0.783,
+      "step": 654000
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 9.340599697794887e-06,
+      "loss": 0.7555,
+      "step": 654200
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 9.33122172688301e-06,
+      "loss": 0.7492,
+      "step": 654400
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 9.32184375597113e-06,
+      "loss": 0.7596,
+      "step": 654600
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 9.312465785059252e-06,
+      "loss": 0.7558,
+      "step": 654800
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 9.303087814147373e-06,
+      "loss": 0.7533,
+      "step": 655000
+    },
+    {
+      "epoch": 0.77,
+      "eval_em_first_line_ratio": 0.4941,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3808,
+      "eval_loss": 0.7308037877082825,
+      "eval_runtime": 198.7823,
+      "eval_samples_per_second": 50.306,
+      "eval_steps_per_second": 0.79,
+      "step": 655000
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 9.293709843235494e-06,
+      "loss": 0.7568,
+      "step": 655200
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 9.284331872323615e-06,
+      "loss": 0.7538,
+      "step": 655400
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 9.274953901411737e-06,
+      "loss": 0.7506,
+      "step": 655600
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 9.265575930499858e-06,
+      "loss": 0.7586,
+      "step": 655800
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 9.25619795958798e-06,
+      "loss": 0.7519,
+      "step": 656000
+    },
+    {
+      "epoch": 0.77,
+      "eval_em_first_line_ratio": 0.4959,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3826,
+      "eval_loss": 0.7304233312606812,
+      "eval_runtime": 196.7781,
+      "eval_samples_per_second": 50.819,
+      "eval_steps_per_second": 0.798,
+      "step": 656000
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 9.246819988676102e-06,
+      "loss": 0.7547,
+      "step": 656200
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 9.237442017764223e-06,
+      "loss": 0.7542,
+      "step": 656400
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 9.228064046852344e-06,
+      "loss": 0.7521,
+      "step": 656600
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 9.218686075940465e-06,
+      "loss": 0.7539,
+      "step": 656800
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 9.209308105028586e-06,
+      "loss": 0.7566,
+      "step": 657000
+    },
+    {
+      "epoch": 0.77,
+      "eval_em_first_line_ratio": 0.4922,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.379,
+      "eval_loss": 0.7303532361984253,
+      "eval_runtime": 199.7847,
+      "eval_samples_per_second": 50.054,
+      "eval_steps_per_second": 0.786,
+      "step": 657000
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 9.200023913825826e-06,
+      "loss": 0.764,
+      "step": 657200
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 9.190645942913947e-06,
+      "loss": 0.7542,
+      "step": 657400
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 9.181267972002068e-06,
+      "loss": 0.7523,
+      "step": 657600
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 9.17189000109019e-06,
+      "loss": 0.7527,
+      "step": 657800
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 9.16251203017831e-06,
+      "loss": 0.7584,
+      "step": 658000
+    },
+    {
+      "epoch": 0.77,
+      "eval_em_first_line_ratio": 0.4936,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3801,
+      "eval_loss": 0.7303171157836914,
+      "eval_runtime": 202.1955,
+      "eval_samples_per_second": 49.457,
+      "eval_steps_per_second": 0.776,
+      "step": 658000
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 9.153134059266431e-06,
+      "loss": 0.7545,
+      "step": 658200
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 9.143756088354554e-06,
+      "loss": 0.7534,
+      "step": 658400
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 9.134378117442675e-06,
+      "loss": 0.7548,
+      "step": 658600
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 9.125000146530796e-06,
+      "loss": 0.7551,
+      "step": 658800
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 9.115622175618919e-06,
+      "loss": 0.7566,
+      "step": 659000
+    },
+    {
+      "epoch": 0.77,
+      "eval_em_first_line_ratio": 0.4936,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3805,
+      "eval_loss": 0.7301334738731384,
+      "eval_runtime": 199.7471,
+      "eval_samples_per_second": 50.063,
+      "eval_steps_per_second": 0.786,
+      "step": 659000
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 9.10624420470704e-06,
+      "loss": 0.7518,
+      "step": 659200
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 9.096866233795159e-06,
+      "loss": 0.7607,
+      "step": 659400
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 9.087488262883282e-06,
+      "loss": 0.7577,
+      "step": 659600
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 9.078110291971402e-06,
+      "loss": 0.751,
+      "step": 659800
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 9.068732321059523e-06,
+      "loss": 0.7504,
+      "step": 660000
+    },
+    {
+      "epoch": 0.77,
+      "eval_em_first_line_ratio": 0.4939,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3806,
+      "eval_loss": 0.7302184700965881,
+      "eval_runtime": 195.5803,
+      "eval_samples_per_second": 51.13,
+      "eval_steps_per_second": 0.803,
+      "step": 660000
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 9.059354350147646e-06,
+      "loss": 0.756,
+      "step": 660200
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 9.049976379235767e-06,
+      "loss": 0.7613,
+      "step": 660400
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 9.040598408323888e-06,
+      "loss": 0.7547,
+      "step": 660600
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 9.031220437412009e-06,
+      "loss": 0.7548,
+      "step": 660800
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 9.02184246650013e-06,
+      "loss": 0.7515,
+      "step": 661000
+    },
+    {
+      "epoch": 0.77,
+      "eval_em_first_line_ratio": 0.4926,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3797,
+      "eval_loss": 0.7301636934280396,
+      "eval_runtime": 205.0661,
+      "eval_samples_per_second": 48.765,
+      "eval_steps_per_second": 0.766,
+      "step": 661000
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 9.012464495588251e-06,
+      "loss": 0.7564,
+      "step": 661200
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 9.003133414530932e-06,
+      "loss": 0.7533,
+      "step": 661400
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 8.993755443619053e-06,
+      "loss": 0.761,
+      "step": 661600
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 8.984377472707175e-06,
+      "loss": 0.7521,
+      "step": 661800
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 8.974999501795296e-06,
+      "loss": 0.7576,
+      "step": 662000
+    },
+    {
+      "epoch": 0.78,
+      "eval_em_first_line_ratio": 0.4925,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3798,
+      "eval_loss": 0.7302247285842896,
+      "eval_runtime": 200.2376,
+      "eval_samples_per_second": 49.941,
+      "eval_steps_per_second": 0.784,
+      "step": 662000
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 8.965621530883417e-06,
+      "loss": 0.753,
+      "step": 662200
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 8.956243559971538e-06,
+      "loss": 0.7584,
+      "step": 662400
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 8.94686558905966e-06,
+      "loss": 0.7478,
+      "step": 662600
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 8.93748761814778e-06,
+      "loss": 0.7521,
+      "step": 662800
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 8.928109647235901e-06,
+      "loss": 0.7512,
+      "step": 663000
+    },
+    {
+      "epoch": 0.78,
+      "eval_em_first_line_ratio": 0.4935,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3801,
+      "eval_loss": 0.730348527431488,
+      "eval_runtime": 197.7758,
+      "eval_samples_per_second": 50.562,
+      "eval_steps_per_second": 0.794,
+      "step": 663000
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 8.918731676324024e-06,
+      "loss": 0.7509,
+      "step": 663200
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 8.909353705412145e-06,
+      "loss": 0.7491,
+      "step": 663400
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 8.899975734500266e-06,
+      "loss": 0.7534,
+      "step": 663600
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 8.890597763588389e-06,
+      "loss": 0.7554,
+      "step": 663800
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 8.881266682531068e-06,
+      "loss": 0.7545,
+      "step": 664000
+    },
+    {
+      "epoch": 0.78,
+      "eval_em_first_line_ratio": 0.4933,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3801,
+      "eval_loss": 0.7300431132316589,
+      "eval_runtime": 195.8206,
+      "eval_samples_per_second": 51.067,
+      "eval_steps_per_second": 0.802,
+      "step": 664000
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 8.87188871161919e-06,
+      "loss": 0.7537,
+      "step": 664200
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 8.86255763056187e-06,
+      "loss": 0.7563,
+      "step": 664400
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 8.853179659649992e-06,
+      "loss": 0.7523,
+      "step": 664600
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 8.843801688738113e-06,
+      "loss": 0.754,
+      "step": 664800
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 8.834423717826234e-06,
+      "loss": 0.7512,
+      "step": 665000
+    },
+    {
+      "epoch": 0.78,
+      "eval_em_first_line_ratio": 0.4925,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.379,
+      "eval_loss": 0.7302541732788086,
+      "eval_runtime": 196.7027,
+      "eval_samples_per_second": 50.838,
+      "eval_steps_per_second": 0.798,
+      "step": 665000
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 8.825045746914355e-06,
+      "loss": 0.7574,
+      "step": 665200
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 8.815667776002476e-06,
+      "loss": 0.7567,
+      "step": 665400
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 8.806289805090597e-06,
+      "loss": 0.7578,
+      "step": 665600
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 8.796911834178718e-06,
+      "loss": 0.7489,
+      "step": 665800
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 8.7875807531214e-06,
+      "loss": 0.7596,
+      "step": 666000
+    },
+    {
+      "epoch": 0.78,
+      "eval_em_first_line_ratio": 0.4945,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3815,
+      "eval_loss": 0.729703426361084,
+      "eval_runtime": 199.4381,
+      "eval_samples_per_second": 50.141,
+      "eval_steps_per_second": 0.787,
+      "step": 666000
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 8.778202782209521e-06,
+      "loss": 0.7513,
+      "step": 666200
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 8.768824811297642e-06,
+      "loss": 0.7557,
+      "step": 666400
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 8.759446840385763e-06,
+      "loss": 0.7513,
+      "step": 666600
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 8.750068869473884e-06,
+      "loss": 0.7541,
+      "step": 666800
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 8.740690898562007e-06,
+      "loss": 0.7553,
+      "step": 667000
+    },
+    {
+      "epoch": 0.78,
+      "eval_em_first_line_ratio": 0.4923,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3793,
+      "eval_loss": 0.7302016615867615,
+      "eval_runtime": 195.8342,
+      "eval_samples_per_second": 51.064,
+      "eval_steps_per_second": 0.802,
+      "step": 667000
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 8.731312927650128e-06,
+      "loss": 0.7494,
+      "step": 667200
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 8.721934956738249e-06,
+      "loss": 0.7495,
+      "step": 667400
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 8.71255698582637e-06,
+      "loss": 0.7498,
+      "step": 667600
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 8.703179014914491e-06,
+      "loss": 0.7544,
+      "step": 667800
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 8.693801044002612e-06,
+      "loss": 0.7553,
+      "step": 668000
+    },
+    {
+      "epoch": 0.78,
+      "eval_em_first_line_ratio": 0.4924,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3795,
+      "eval_loss": 0.7298839688301086,
+      "eval_runtime": 200.7102,
+      "eval_samples_per_second": 49.823,
+      "eval_steps_per_second": 0.782,
+      "step": 668000
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 8.684423073090735e-06,
+      "loss": 0.7547,
+      "step": 668200
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 8.675045102178856e-06,
+      "loss": 0.7535,
+      "step": 668400
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 8.665667131266977e-06,
+      "loss": 0.7541,
+      "step": 668600
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 8.656289160355097e-06,
+      "loss": 0.7516,
+      "step": 668800
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 8.646911189443218e-06,
+      "loss": 0.7571,
+      "step": 669000
+    },
+    {
+      "epoch": 0.78,
+      "eval_em_first_line_ratio": 0.4941,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3812,
+      "eval_loss": 0.7295217514038086,
+      "eval_runtime": 196.3145,
+      "eval_samples_per_second": 50.939,
+      "eval_steps_per_second": 0.8,
+      "step": 669000
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 8.63753321853134e-06,
+      "loss": 0.7504,
+      "step": 669200
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 8.628155247619462e-06,
+      "loss": 0.7552,
+      "step": 669400
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 8.618777276707583e-06,
+      "loss": 0.7577,
+      "step": 669600
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 8.609446195650262e-06,
+      "loss": 0.7542,
+      "step": 669800
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 8.600068224738385e-06,
+      "loss": 0.7488,
+      "step": 670000
+    },
+    {
+      "epoch": 0.79,
+      "eval_em_first_line_ratio": 0.4939,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.381,
+      "eval_loss": 0.7300624847412109,
+      "eval_runtime": 196.2559,
+      "eval_samples_per_second": 50.954,
+      "eval_steps_per_second": 0.8,
+      "step": 670000
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 8.590690253826506e-06,
+      "loss": 0.7572,
+      "step": 670200
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 8.581312282914627e-06,
+      "loss": 0.7519,
+      "step": 670400
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 8.57193431200275e-06,
+      "loss": 0.7588,
+      "step": 670600
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 8.56255634109087e-06,
+      "loss": 0.7576,
+      "step": 670800
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 8.55317837017899e-06,
+      "loss": 0.7554,
+      "step": 671000
+    },
+    {
+      "epoch": 0.79,
+      "eval_em_first_line_ratio": 0.492,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3786,
+      "eval_loss": 0.7297177910804749,
+      "eval_runtime": 204.1513,
+      "eval_samples_per_second": 48.983,
+      "eval_steps_per_second": 0.769,
+      "step": 671000
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 8.543800399267112e-06,
+      "loss": 0.7553,
+      "step": 671200
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 8.534422428355233e-06,
+      "loss": 0.754,
+      "step": 671400
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 8.525044457443354e-06,
+      "loss": 0.7535,
+      "step": 671600
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 8.515666486531477e-06,
+      "loss": 0.7514,
+      "step": 671800
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 8.506288515619598e-06,
+      "loss": 0.7568,
+      "step": 672000
+    },
+    {
+      "epoch": 0.79,
+      "eval_em_first_line_ratio": 0.4929,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3798,
+      "eval_loss": 0.7293977737426758,
+      "eval_runtime": 199.173,
+      "eval_samples_per_second": 50.208,
+      "eval_steps_per_second": 0.788,
+      "step": 672000
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 8.496910544707719e-06,
+      "loss": 0.7484,
+      "step": 672200
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 8.48753257379584e-06,
+      "loss": 0.7548,
+      "step": 672400
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 8.478154602883961e-06,
+      "loss": 0.7525,
+      "step": 672600
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 8.468776631972082e-06,
+      "loss": 0.755,
+      "step": 672800
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 8.459398661060205e-06,
+      "loss": 0.7507,
+      "step": 673000
+    },
+    {
+      "epoch": 0.79,
+      "eval_em_first_line_ratio": 0.4942,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3809,
+      "eval_loss": 0.7295095920562744,
+      "eval_runtime": 194.2493,
+      "eval_samples_per_second": 51.48,
+      "eval_steps_per_second": 0.808,
+      "step": 673000
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 8.450020690148326e-06,
+      "loss": 0.7559,
+      "step": 673200
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 8.440642719236446e-06,
+      "loss": 0.7525,
+      "step": 673400
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 8.431264748324567e-06,
+      "loss": 0.7551,
+      "step": 673600
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 8.421886777412688e-06,
+      "loss": 0.7459,
+      "step": 673800
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 8.41255569635537e-06,
+      "loss": 0.7477,
+      "step": 674000
+    },
+    {
+      "epoch": 0.79,
+      "eval_em_first_line_ratio": 0.4917,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3785,
+      "eval_loss": 0.7296448945999146,
+      "eval_runtime": 201.4451,
+      "eval_samples_per_second": 49.641,
+      "eval_steps_per_second": 0.779,
+      "step": 674000
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 8.40317772544349e-06,
+      "loss": 0.752,
+      "step": 674200
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 8.393799754531611e-06,
+      "loss": 0.7518,
+      "step": 674400
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 8.384421783619732e-06,
+      "loss": 0.7533,
+      "step": 674600
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 8.375043812707855e-06,
+      "loss": 0.7484,
+      "step": 674800
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 8.365665841795976e-06,
+      "loss": 0.755,
+      "step": 675000
+    },
+    {
+      "epoch": 0.79,
+      "eval_em_first_line_ratio": 0.4924,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.379,
+      "eval_loss": 0.72948157787323,
+      "eval_runtime": 202.6846,
+      "eval_samples_per_second": 49.338,
+      "eval_steps_per_second": 0.775,
+      "step": 675000
+    }
+  ],
+  "max_steps": 853163,
+  "num_train_epochs": 1,
+  "total_flos": 5.261396039737488e+19,
+  "trial_name": null,
+  "trial_params": null
+}