diff --git "a/finetuned_codet5large_512/trainer_state.json" "b/finetuned_codet5large_512/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/finetuned_codet5large_512/trainer_state.json"
@@ -0,0 +1,23176 @@
+{
+  "best_metric": 0.5263,
+  "best_model_checkpoint": "/repo_data/finetuning_checkpoints/codet5-base-ntp-java/train_large_tAll_v10k_sl512_nep1_bspd12_dn4_graccs3_lr1e-4_wup100_wd005/checkpoint-558000",
+  "epoch": 0.7634825882455328,
+  "global_step": 579000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "learning_rate": 9.998707579256384e-05,
+      "loss": 1.8045,
+      "step": 200
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 9.996069985902063e-05,
+      "loss": 1.1549,
+      "step": 400
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 9.993432392547745e-05,
+      "loss": 1.1114,
+      "step": 600
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 9.990794799193425e-05,
+      "loss": 1.0857,
+      "step": 800
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 9.988157205839104e-05,
+      "loss": 1.062,
+      "step": 1000
+    },
+    {
+      "epoch": 0.0,
+      "eval_em_first_line_ratio": 0.4423,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3314,
+      "eval_loss": 0.9850248098373413,
+      "eval_runtime": 241.9983,
+      "eval_samples_per_second": 41.323,
+      "eval_steps_per_second": 0.219,
+      "step": 1000
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 9.985519612484785e-05,
+      "loss": 1.0572,
+      "step": 1200
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 9.982882019130465e-05,
+      "loss": 1.0382,
+      "step": 1400
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 9.980244425776146e-05,
+      "loss": 1.0261,
+      "step": 1600
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 9.977606832421825e-05,
+      "loss": 1.0173,
+      "step": 1800
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 9.974969239067505e-05,
+      "loss": 1.0129,
+      "step": 2000
+    },
+    {
+      "epoch": 0.0,
+      "eval_em_first_line_ratio": 0.4524,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3405,
+      "eval_loss": 0.9412904381752014,
+      "eval_runtime": 240.9453,
+      "eval_samples_per_second": 41.503,
+      "eval_steps_per_second": 0.22,
+      "step": 2000
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 9.972331645713186e-05,
+      "loss": 0.9976,
+      "step": 2200
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 9.969694052358867e-05,
+      "loss": 0.9921,
+      "step": 2400
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 9.967056459004546e-05,
+      "loss": 0.9905,
+      "step": 2600
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 9.964418865650227e-05,
+      "loss": 0.9886,
+      "step": 2800
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 9.961781272295907e-05,
+      "loss": 0.9821,
+      "step": 3000
+    },
+    {
+      "epoch": 0.0,
+      "eval_em_first_line_ratio": 0.4524,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3414,
+      "eval_loss": 0.9193007946014404,
+      "eval_runtime": 240.9308,
+      "eval_samples_per_second": 41.506,
+      "eval_steps_per_second": 0.22,
+      "step": 3000
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 9.959143678941588e-05,
+      "loss": 0.9741,
+      "step": 3200
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 9.956506085587267e-05,
+      "loss": 0.9742,
+      "step": 3400
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 9.953868492232947e-05,
+      "loss": 0.9648,
+      "step": 3600
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 9.951230898878628e-05,
+      "loss": 0.9589,
+      "step": 3800
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 9.948593305524307e-05,
+      "loss": 0.9611,
+      "step": 4000
+    },
+    {
+      "epoch": 0.01,
+      "eval_em_first_line_ratio": 0.4553,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3447,
+      "eval_loss": 0.902411699295044,
+      "eval_runtime": 241.1503,
+      "eval_samples_per_second": 41.468,
+      "eval_steps_per_second": 0.22,
+      "step": 4000
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 9.945955712169987e-05,
+      "loss": 0.9532,
+      "step": 4200
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 9.94333130678244e-05,
+      "loss": 0.9521,
+      "step": 4400
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 9.940720089361663e-05,
+      "loss": 0.9622,
+      "step": 4600
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 9.938082496007344e-05,
+      "loss": 0.9484,
+      "step": 4800
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 9.935444902653024e-05,
+      "loss": 0.9428,
+      "step": 5000
+    },
+    {
+      "epoch": 0.01,
+      "eval_em_first_line_ratio": 0.4554,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3442,
+      "eval_loss": 0.8882227540016174,
+      "eval_runtime": 241.3322,
+      "eval_samples_per_second": 41.437,
+      "eval_steps_per_second": 0.22,
+      "step": 5000
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 9.932807309298703e-05,
+      "loss": 0.9456,
+      "step": 5200
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 9.930169715944384e-05,
+      "loss": 0.9361,
+      "step": 5400
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 9.927532122590064e-05,
+      "loss": 0.9373,
+      "step": 5600
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 9.924894529235745e-05,
+      "loss": 0.9246,
+      "step": 5800
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 9.922256935881424e-05,
+      "loss": 0.9264,
+      "step": 6000
+    },
+    {
+      "epoch": 0.01,
+      "eval_em_first_line_ratio": 0.4615,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3496,
+      "eval_loss": 0.878157913684845,
+      "eval_runtime": 239.0396,
+      "eval_samples_per_second": 41.834,
+      "eval_steps_per_second": 0.222,
+      "step": 6000
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 9.919619342527106e-05,
+      "loss": 0.9218,
+      "step": 6200
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 9.916981749172786e-05,
+      "loss": 0.9245,
+      "step": 6400
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 9.914344155818465e-05,
+      "loss": 0.9226,
+      "step": 6600
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 9.911706562464146e-05,
+      "loss": 0.9224,
+      "step": 6800
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 9.909068969109826e-05,
+      "loss": 0.9183,
+      "step": 7000
+    },
+    {
+      "epoch": 0.01,
+      "eval_em_first_line_ratio": 0.4624,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3514,
+      "eval_loss": 0.8701657056808472,
+      "eval_runtime": 241.2289,
+      "eval_samples_per_second": 41.454,
+      "eval_steps_per_second": 0.22,
+      "step": 7000
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 9.906431375755507e-05,
+      "loss": 0.9172,
+      "step": 7200
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 9.903793782401186e-05,
+      "loss": 0.9125,
+      "step": 7400
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 9.901156189046866e-05,
+      "loss": 0.9165,
+      "step": 7600
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 9.898518595692546e-05,
+      "loss": 0.9112,
+      "step": 7800
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 9.895881002338227e-05,
+      "loss": 0.9028,
+      "step": 8000
+    },
+    {
+      "epoch": 0.01,
+      "eval_em_first_line_ratio": 0.4634,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3513,
+      "eval_loss": 0.8628376126289368,
+      "eval_runtime": 241.1445,
+      "eval_samples_per_second": 41.469,
+      "eval_steps_per_second": 0.22,
+      "step": 8000
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 9.893243408983907e-05,
+      "loss": 0.9072,
+      "step": 8200
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 9.890605815629588e-05,
+      "loss": 0.9072,
+      "step": 8400
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 9.887968222275268e-05,
+      "loss": 0.907,
+      "step": 8600
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 9.885330628920947e-05,
+      "loss": 0.9036,
+      "step": 8800
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 9.882693035566628e-05,
+      "loss": 0.9015,
+      "step": 9000
+    },
+    {
+      "epoch": 0.01,
+      "eval_em_first_line_ratio": 0.4655,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3535,
+      "eval_loss": 0.8549745082855225,
+      "eval_runtime": 241.484,
+      "eval_samples_per_second": 41.411,
+      "eval_steps_per_second": 0.219,
+      "step": 9000
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 9.880055442212308e-05,
+      "loss": 0.8963,
+      "step": 9200
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 9.877417848857989e-05,
+      "loss": 0.8949,
+      "step": 9400
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 9.874780255503668e-05,
+      "loss": 0.8922,
+      "step": 9600
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 9.872142662149348e-05,
+      "loss": 0.8931,
+      "step": 9800
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 9.86950506879503e-05,
+      "loss": 0.8906,
+      "step": 10000
+    },
+    {
+      "epoch": 0.01,
+      "eval_em_first_line_ratio": 0.4645,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3528,
+      "eval_loss": 0.8475026488304138,
+      "eval_runtime": 240.689,
+      "eval_samples_per_second": 41.547,
+      "eval_steps_per_second": 0.22,
+      "step": 10000
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 9.86686747544071e-05,
+      "loss": 0.8857,
+      "step": 10200
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 9.86422988208639e-05,
+      "loss": 0.8885,
+      "step": 10400
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 9.86159228873207e-05,
+      "loss": 0.8865,
+      "step": 10600
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 9.85895469537775e-05,
+      "loss": 0.8818,
+      "step": 10800
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 9.856317102023431e-05,
+      "loss": 0.8795,
+      "step": 11000
+    },
+    {
+      "epoch": 0.01,
+      "eval_em_first_line_ratio": 0.4714,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3584,
+      "eval_loss": 0.8434772491455078,
+      "eval_runtime": 238.79,
+      "eval_samples_per_second": 41.878,
+      "eval_steps_per_second": 0.222,
+      "step": 11000
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 9.85367950866911e-05,
+      "loss": 0.8802,
+      "step": 11200
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 9.85104191531479e-05,
+      "loss": 0.8811,
+      "step": 11400
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 9.848404321960471e-05,
+      "loss": 0.8774,
+      "step": 11600
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 9.84576672860615e-05,
+      "loss": 0.8776,
+      "step": 11800
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 9.843129135251832e-05,
+      "loss": 0.872,
+      "step": 12000
+    },
+    {
+      "epoch": 0.02,
+      "eval_em_first_line_ratio": 0.4711,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3592,
+      "eval_loss": 0.8353820443153381,
+      "eval_runtime": 238.1829,
+      "eval_samples_per_second": 41.985,
+      "eval_steps_per_second": 0.223,
+      "step": 12000
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 9.840491541897512e-05,
+      "loss": 0.8751,
+      "step": 12200
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 9.837867136509963e-05,
+      "loss": 0.8713,
+      "step": 12400
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 9.835242731122415e-05,
+      "loss": 0.8761,
+      "step": 12600
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 9.832605137768095e-05,
+      "loss": 0.8694,
+      "step": 12800
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 9.829967544413776e-05,
+      "loss": 0.8705,
+      "step": 13000
+    },
+    {
+      "epoch": 0.02,
+      "eval_em_first_line_ratio": 0.4696,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3582,
+      "eval_loss": 0.8334267735481262,
+      "eval_runtime": 240.2182,
+      "eval_samples_per_second": 41.629,
+      "eval_steps_per_second": 0.221,
+      "step": 13000
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 9.827329951059456e-05,
+      "loss": 0.8715,
+      "step": 13200
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 9.824692357705137e-05,
+      "loss": 0.8714,
+      "step": 13400
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 9.822054764350817e-05,
+      "loss": 0.8713,
+      "step": 13600
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 9.819417170996496e-05,
+      "loss": 0.8701,
+      "step": 13800
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 9.816779577642177e-05,
+      "loss": 0.8658,
+      "step": 14000
+    },
+    {
+      "epoch": 0.02,
+      "eval_em_first_line_ratio": 0.4738,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3611,
+      "eval_loss": 0.8282851576805115,
+      "eval_runtime": 236.6363,
+      "eval_samples_per_second": 42.259,
+      "eval_steps_per_second": 0.224,
+      "step": 14000
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 9.814141984287857e-05,
+      "loss": 0.8697,
+      "step": 14200
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 9.811504390933537e-05,
+      "loss": 0.8653,
+      "step": 14400
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 9.808866797579216e-05,
+      "loss": 0.8566,
+      "step": 14600
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 9.806229204224897e-05,
+      "loss": 0.8643,
+      "step": 14800
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 9.803591610870577e-05,
+      "loss": 0.863,
+      "step": 15000
+    },
+    {
+      "epoch": 0.02,
+      "eval_em_first_line_ratio": 0.4733,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3616,
+      "eval_loss": 0.8259214162826538,
+      "eval_runtime": 237.7167,
+      "eval_samples_per_second": 42.067,
+      "eval_steps_per_second": 0.223,
+      "step": 15000
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 9.800954017516258e-05,
+      "loss": 0.8634,
+      "step": 15200
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 9.798316424161938e-05,
+      "loss": 0.8606,
+      "step": 15400
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 9.795678830807619e-05,
+      "loss": 0.8669,
+      "step": 15600
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 9.793041237453299e-05,
+      "loss": 0.8588,
+      "step": 15800
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 9.790403644098978e-05,
+      "loss": 0.8602,
+      "step": 16000
+    },
+    {
+      "epoch": 0.02,
+      "eval_em_first_line_ratio": 0.4704,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3586,
+      "eval_loss": 0.821124792098999,
+      "eval_runtime": 238.7817,
+      "eval_samples_per_second": 41.879,
+      "eval_steps_per_second": 0.222,
+      "step": 16000
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 9.787766050744659e-05,
+      "loss": 0.851,
+      "step": 16200
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 9.785128457390339e-05,
+      "loss": 0.8497,
+      "step": 16400
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 9.78249086403602e-05,
+      "loss": 0.8564,
+      "step": 16600
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 9.779853270681699e-05,
+      "loss": 0.8542,
+      "step": 16800
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 9.77721567732738e-05,
+      "loss": 0.8522,
+      "step": 17000
+    },
+    {
+      "epoch": 0.02,
+      "eval_em_first_line_ratio": 0.4729,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3612,
+      "eval_loss": 0.8176316618919373,
+      "eval_runtime": 232.9709,
+      "eval_samples_per_second": 42.924,
+      "eval_steps_per_second": 0.227,
+      "step": 17000
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 9.774578083973061e-05,
+      "loss": 0.8557,
+      "step": 17200
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 9.77194049061874e-05,
+      "loss": 0.8497,
+      "step": 17400
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 9.76930289726442e-05,
+      "loss": 0.8477,
+      "step": 17600
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 9.766665303910101e-05,
+      "loss": 0.8458,
+      "step": 17800
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 9.764027710555781e-05,
+      "loss": 0.8491,
+      "step": 18000
+    },
+    {
+      "epoch": 0.02,
+      "eval_em_first_line_ratio": 0.477,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3643,
+      "eval_loss": 0.8149089217185974,
+      "eval_runtime": 236.8549,
+      "eval_samples_per_second": 42.22,
+      "eval_steps_per_second": 0.224,
+      "step": 18000
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 9.76139011720146e-05,
+      "loss": 0.8515,
+      "step": 18200
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 9.758752523847141e-05,
+      "loss": 0.8438,
+      "step": 18400
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 9.756114930492821e-05,
+      "loss": 0.8481,
+      "step": 18600
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 9.753477337138502e-05,
+      "loss": 0.8491,
+      "step": 18800
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 9.750839743784182e-05,
+      "loss": 0.8419,
+      "step": 19000
+    },
+    {
+      "epoch": 0.03,
+      "eval_em_first_line_ratio": 0.4738,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3621,
+      "eval_loss": 0.8104076981544495,
+      "eval_runtime": 240.2128,
+      "eval_samples_per_second": 41.63,
+      "eval_steps_per_second": 0.221,
+      "step": 19000
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 9.748202150429863e-05,
+      "loss": 0.8442,
+      "step": 19200
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 9.745564557075543e-05,
+      "loss": 0.8442,
+      "step": 19400
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 9.742926963721223e-05,
+      "loss": 0.8412,
+      "step": 19600
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 9.740289370366903e-05,
+      "loss": 0.8452,
+      "step": 19800
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 9.737651777012583e-05,
+      "loss": 0.8448,
+      "step": 20000
+    },
+    {
+      "epoch": 0.03,
+      "eval_em_first_line_ratio": 0.4741,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3618,
+      "eval_loss": 0.8083608150482178,
+      "eval_runtime": 238.5921,
+      "eval_samples_per_second": 41.913,
+      "eval_steps_per_second": 0.222,
+      "step": 20000
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 9.735014183658263e-05,
+      "loss": 0.8427,
+      "step": 20200
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 9.732376590303943e-05,
+      "loss": 0.8368,
+      "step": 20400
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 9.729738996949623e-05,
+      "loss": 0.8318,
+      "step": 20600
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 9.727101403595303e-05,
+      "loss": 0.8387,
+      "step": 20800
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 9.724463810240985e-05,
+      "loss": 0.8367,
+      "step": 21000
+    },
+    {
+      "epoch": 0.03,
+      "eval_em_first_line_ratio": 0.4731,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3611,
+      "eval_loss": 0.8046004176139832,
+      "eval_runtime": 240.9703,
+      "eval_samples_per_second": 41.499,
+      "eval_steps_per_second": 0.22,
+      "step": 21000
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 9.721826216886664e-05,
+      "loss": 0.8371,
+      "step": 21200
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 9.719188623532345e-05,
+      "loss": 0.836,
+      "step": 21400
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 9.716551030178025e-05,
+      "loss": 0.8371,
+      "step": 21600
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 9.713913436823706e-05,
+      "loss": 0.8342,
+      "step": 21800
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 9.711289031436157e-05,
+      "loss": 0.8327,
+      "step": 22000
+    },
+    {
+      "epoch": 0.03,
+      "eval_em_first_line_ratio": 0.4785,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3661,
+      "eval_loss": 0.8026383519172668,
+      "eval_runtime": 237.6618,
+      "eval_samples_per_second": 42.077,
+      "eval_steps_per_second": 0.223,
+      "step": 22000
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 9.708651438081837e-05,
+      "loss": 0.8298,
+      "step": 22200
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 9.706013844727518e-05,
+      "loss": 0.831,
+      "step": 22400
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 9.703376251373197e-05,
+      "loss": 0.8287,
+      "step": 22600
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 9.700738658018877e-05,
+      "loss": 0.8306,
+      "step": 22800
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 9.698101064664558e-05,
+      "loss": 0.831,
+      "step": 23000
+    },
+    {
+      "epoch": 0.03,
+      "eval_em_first_line_ratio": 0.4754,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3634,
+      "eval_loss": 0.8006582260131836,
+      "eval_runtime": 237.9785,
+      "eval_samples_per_second": 42.021,
+      "eval_steps_per_second": 0.223,
+      "step": 23000
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 9.695463471310238e-05,
+      "loss": 0.8325,
+      "step": 23200
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 9.692825877955919e-05,
+      "loss": 0.8294,
+      "step": 23400
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 9.690188284601599e-05,
+      "loss": 0.8292,
+      "step": 23600
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 9.68755069124728e-05,
+      "loss": 0.8327,
+      "step": 23800
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 9.684913097892959e-05,
+      "loss": 0.8262,
+      "step": 24000
+    },
+    {
+      "epoch": 0.03,
+      "eval_em_first_line_ratio": 0.4797,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3664,
+      "eval_loss": 0.7977368831634521,
+      "eval_runtime": 234.7242,
+      "eval_samples_per_second": 42.603,
+      "eval_steps_per_second": 0.226,
+      "step": 24000
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 9.682275504538639e-05,
+      "loss": 0.8269,
+      "step": 24200
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 9.679651099151092e-05,
+      "loss": 0.8263,
+      "step": 24400
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 9.677026693763542e-05,
+      "loss": 0.834,
+      "step": 24600
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 9.674389100409224e-05,
+      "loss": 0.8274,
+      "step": 24800
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 9.671764695021675e-05,
+      "loss": 0.8304,
+      "step": 25000
+    },
+    {
+      "epoch": 0.03,
+      "eval_em_first_line_ratio": 0.478,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3657,
+      "eval_loss": 0.7961474061012268,
+      "eval_runtime": 235.3367,
+      "eval_samples_per_second": 42.492,
+      "eval_steps_per_second": 0.225,
+      "step": 25000
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 9.669127101667354e-05,
+      "loss": 0.8235,
+      "step": 25200
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 9.666489508313036e-05,
+      "loss": 0.8281,
+      "step": 25400
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 9.663851914958716e-05,
+      "loss": 0.8339,
+      "step": 25600
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 9.661214321604397e-05,
+      "loss": 0.8232,
+      "step": 25800
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 9.658576728250076e-05,
+      "loss": 0.8247,
+      "step": 26000
+    },
+    {
+      "epoch": 0.03,
+      "eval_em_first_line_ratio": 0.4789,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3659,
+      "eval_loss": 0.7938370108604431,
+      "eval_runtime": 233.7373,
+      "eval_samples_per_second": 42.783,
+      "eval_steps_per_second": 0.227,
+      "step": 26000
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 9.655939134895756e-05,
+      "loss": 0.8267,
+      "step": 26200
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 9.653301541541437e-05,
+      "loss": 0.8231,
+      "step": 26400
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 9.650663948187117e-05,
+      "loss": 0.8206,
+      "step": 26600
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 9.648026354832796e-05,
+      "loss": 0.8231,
+      "step": 26800
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 9.645401949445249e-05,
+      "loss": 0.8217,
+      "step": 27000
+    },
+    {
+      "epoch": 0.04,
+      "eval_em_first_line_ratio": 0.4798,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3682,
+      "eval_loss": 0.7917524576187134,
+      "eval_runtime": 238.4685,
+      "eval_samples_per_second": 41.934,
+      "eval_steps_per_second": 0.222,
+      "step": 27000
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 9.642764356090929e-05,
+      "loss": 0.8207,
+      "step": 27200
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 9.640126762736608e-05,
+      "loss": 0.8256,
+      "step": 27400
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 9.637489169382289e-05,
+      "loss": 0.823,
+      "step": 27600
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 9.634851576027969e-05,
+      "loss": 0.8221,
+      "step": 27800
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 9.63221398267365e-05,
+      "loss": 0.8167,
+      "step": 28000
+    },
+    {
+      "epoch": 0.04,
+      "eval_em_first_line_ratio": 0.4843,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.371,
+      "eval_loss": 0.7904202342033386,
+      "eval_runtime": 231.6037,
+      "eval_samples_per_second": 43.177,
+      "eval_steps_per_second": 0.229,
+      "step": 28000
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 9.62957638931933e-05,
+      "loss": 0.8181,
+      "step": 28200
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 9.62693879596501e-05,
+      "loss": 0.812,
+      "step": 28400
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 9.624301202610691e-05,
+      "loss": 0.8181,
+      "step": 28600
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 9.62166360925637e-05,
+      "loss": 0.8162,
+      "step": 28800
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 9.61902601590205e-05,
+      "loss": 0.8162,
+      "step": 29000
+    },
+    {
+      "epoch": 0.04,
+      "eval_em_first_line_ratio": 0.4835,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3704,
+      "eval_loss": 0.7884989380836487,
+      "eval_runtime": 230.2546,
+      "eval_samples_per_second": 43.43,
+      "eval_steps_per_second": 0.23,
+      "step": 29000
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 9.616388422547731e-05,
+      "loss": 0.8172,
+      "step": 29200
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 9.613750829193411e-05,
+      "loss": 0.8113,
+      "step": 29400
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 9.61111323583909e-05,
+      "loss": 0.821,
+      "step": 29600
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 9.608475642484772e-05,
+      "loss": 0.8108,
+      "step": 29800
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 9.605838049130453e-05,
+      "loss": 0.8145,
+      "step": 30000
+    },
+    {
+      "epoch": 0.04,
+      "eval_em_first_line_ratio": 0.4807,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3687,
+      "eval_loss": 0.7866184115409851,
+      "eval_runtime": 238.2679,
+      "eval_samples_per_second": 41.97,
+      "eval_steps_per_second": 0.222,
+      "step": 30000
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 9.603200455776132e-05,
+      "loss": 0.8174,
+      "step": 30200
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 9.600562862421812e-05,
+      "loss": 0.8094,
+      "step": 30400
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 9.597925269067493e-05,
+      "loss": 0.8127,
+      "step": 30600
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 9.595287675713173e-05,
+      "loss": 0.8125,
+      "step": 30800
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 9.592650082358852e-05,
+      "loss": 0.8105,
+      "step": 31000
+    },
+    {
+      "epoch": 0.04,
+      "eval_em_first_line_ratio": 0.4837,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3717,
+      "eval_loss": 0.7838398814201355,
+      "eval_runtime": 233.2327,
+      "eval_samples_per_second": 42.876,
+      "eval_steps_per_second": 0.227,
+      "step": 31000
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 9.590012489004533e-05,
+      "loss": 0.8104,
+      "step": 31200
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 9.587388083616985e-05,
+      "loss": 0.8043,
+      "step": 31400
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 9.584750490262664e-05,
+      "loss": 0.8136,
+      "step": 31600
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 9.582112896908345e-05,
+      "loss": 0.8143,
+      "step": 31800
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 9.579475303554025e-05,
+      "loss": 0.8067,
+      "step": 32000
+    },
+    {
+      "epoch": 0.04,
+      "eval_em_first_line_ratio": 0.4821,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3694,
+      "eval_loss": 0.7813404202461243,
+      "eval_runtime": 239.1849,
+      "eval_samples_per_second": 41.809,
+      "eval_steps_per_second": 0.222,
+      "step": 32000
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 9.576837710199706e-05,
+      "loss": 0.8105,
+      "step": 32200
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 9.574200116845386e-05,
+      "loss": 0.8093,
+      "step": 32400
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 9.571562523491067e-05,
+      "loss": 0.8043,
+      "step": 32600
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 9.568924930136747e-05,
+      "loss": 0.8079,
+      "step": 32800
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 9.566287336782426e-05,
+      "loss": 0.8074,
+      "step": 33000
+    },
+    {
+      "epoch": 0.04,
+      "eval_em_first_line_ratio": 0.4834,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3714,
+      "eval_loss": 0.7801371812820435,
+      "eval_runtime": 238.7743,
+      "eval_samples_per_second": 41.881,
+      "eval_steps_per_second": 0.222,
+      "step": 33000
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 9.563649743428107e-05,
+      "loss": 0.802,
+      "step": 33200
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 9.561012150073787e-05,
+      "loss": 0.8051,
+      "step": 33400
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 9.558374556719468e-05,
+      "loss": 0.8057,
+      "step": 33600
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 9.555736963365147e-05,
+      "loss": 0.8021,
+      "step": 33800
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 9.553099370010827e-05,
+      "loss": 0.804,
+      "step": 34000
+    },
+    {
+      "epoch": 0.04,
+      "eval_em_first_line_ratio": 0.4833,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3713,
+      "eval_loss": 0.778603196144104,
+      "eval_runtime": 236.8801,
+      "eval_samples_per_second": 42.215,
+      "eval_steps_per_second": 0.224,
+      "step": 34000
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 9.550461776656509e-05,
+      "loss": 0.7972,
+      "step": 34200
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 9.547824183302189e-05,
+      "loss": 0.8098,
+      "step": 34400
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 9.545186589947868e-05,
+      "loss": 0.8054,
+      "step": 34600
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 9.542548996593549e-05,
+      "loss": 0.8021,
+      "step": 34800
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 9.539911403239229e-05,
+      "loss": 0.7987,
+      "step": 35000
+    },
+    {
+      "epoch": 0.05,
+      "eval_em_first_line_ratio": 0.4832,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3705,
+      "eval_loss": 0.7769062519073486,
+      "eval_runtime": 236.0286,
+      "eval_samples_per_second": 42.368,
+      "eval_steps_per_second": 0.225,
+      "step": 35000
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 9.53727380988491e-05,
+      "loss": 0.7999,
+      "step": 35200
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 9.534636216530589e-05,
+      "loss": 0.8043,
+      "step": 35400
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 9.531998623176269e-05,
+      "loss": 0.7998,
+      "step": 35600
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 9.52936102982195e-05,
+      "loss": 0.7993,
+      "step": 35800
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 9.526723436467629e-05,
+      "loss": 0.8004,
+      "step": 36000
+    },
+    {
+      "epoch": 0.05,
+      "eval_em_first_line_ratio": 0.4815,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3697,
+      "eval_loss": 0.7781192064285278,
+      "eval_runtime": 239.1984,
+      "eval_samples_per_second": 41.806,
+      "eval_steps_per_second": 0.222,
+      "step": 36000
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 9.52408584311331e-05,
+      "loss": 0.798,
+      "step": 36200
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 9.521448249758991e-05,
+      "loss": 0.8013,
+      "step": 36400
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 9.518810656404671e-05,
+      "loss": 0.8011,
+      "step": 36600
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 9.51617306305035e-05,
+      "loss": 0.8009,
+      "step": 36800
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 9.513535469696031e-05,
+      "loss": 0.7947,
+      "step": 37000
+    },
+    {
+      "epoch": 0.05,
+      "eval_em_first_line_ratio": 0.4848,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3718,
+      "eval_loss": 0.7750248312950134,
+      "eval_runtime": 233.8702,
+      "eval_samples_per_second": 42.759,
+      "eval_steps_per_second": 0.227,
+      "step": 37000
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 9.510897876341711e-05,
+      "loss": 0.8007,
+      "step": 37200
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 9.508260282987392e-05,
+      "loss": 0.7991,
+      "step": 37400
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 9.505622689633071e-05,
+      "loss": 0.8017,
+      "step": 37600
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 9.502985096278751e-05,
+      "loss": 0.7997,
+      "step": 37800
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 9.500347502924432e-05,
+      "loss": 0.7968,
+      "step": 38000
+    },
+    {
+      "epoch": 0.05,
+      "eval_em_first_line_ratio": 0.4838,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3709,
+      "eval_loss": 0.772707462310791,
+      "eval_runtime": 237.8575,
+      "eval_samples_per_second": 42.042,
+      "eval_steps_per_second": 0.223,
+      "step": 38000
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 9.497709909570112e-05,
+      "loss": 0.7955,
+      "step": 38200
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 9.495085504182564e-05,
+      "loss": 0.7956,
+      "step": 38400
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 9.492447910828244e-05,
+      "loss": 0.7971,
+      "step": 38600
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 9.489810317473924e-05,
+      "loss": 0.7911,
+      "step": 38800
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 9.487172724119605e-05,
+      "loss": 0.8009,
+      "step": 39000
+    },
+    {
+      "epoch": 0.05,
+      "eval_em_first_line_ratio": 0.4819,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3696,
+      "eval_loss": 0.7710398435592651,
+      "eval_runtime": 239.4966,
+      "eval_samples_per_second": 41.754,
+      "eval_steps_per_second": 0.221,
+      "step": 39000
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 9.484535130765285e-05,
+      "loss": 0.7962,
+      "step": 39200
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 9.481897537410966e-05,
+      "loss": 0.7985,
+      "step": 39400
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 9.479259944056645e-05,
+      "loss": 0.7882,
+      "step": 39600
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 9.476622350702325e-05,
+      "loss": 0.7892,
+      "step": 39800
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 9.473984757348006e-05,
+      "loss": 0.7955,
+      "step": 40000
+    },
+    {
+      "epoch": 0.05,
+      "eval_em_first_line_ratio": 0.4842,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3716,
+      "eval_loss": 0.7706788182258606,
+      "eval_runtime": 321.756,
+      "eval_samples_per_second": 31.079,
+      "eval_steps_per_second": 0.165,
+      "step": 40000
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 9.471347163993686e-05,
+      "loss": 0.7974,
+      "step": 40200
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 9.468709570639365e-05,
+      "loss": 0.7948,
+      "step": 40400
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 9.466071977285047e-05,
+      "loss": 0.792,
+      "step": 40600
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 9.463434383930728e-05,
+      "loss": 0.792,
+      "step": 40800
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 9.46082316650995e-05,
+      "loss": 0.7975,
+      "step": 41000
+    },
+    {
+      "epoch": 0.05,
+      "eval_em_first_line_ratio": 0.483,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3706,
+      "eval_loss": 0.7681934237480164,
+      "eval_runtime": 329.3455,
+      "eval_samples_per_second": 30.363,
+      "eval_steps_per_second": 0.161,
+      "step": 41000
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 9.45818557315563e-05,
+      "loss": 0.7984,
+      "step": 41200
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 9.45554797980131e-05,
+      "loss": 0.7946,
+      "step": 41400
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 9.45291038644699e-05,
+      "loss": 0.7885,
+      "step": 41600
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 9.45027279309267e-05,
+      "loss": 0.793,
+      "step": 41800
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 9.447635199738352e-05,
+      "loss": 0.7889,
+      "step": 42000
+    },
+    {
+      "epoch": 0.06,
+      "eval_em_first_line_ratio": 0.4798,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3685,
+      "eval_loss": 0.7692549824714661,
+      "eval_runtime": 314.8124,
+      "eval_samples_per_second": 31.765,
+      "eval_steps_per_second": 0.168,
+      "step": 42000
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 9.444997606384032e-05,
+      "loss": 0.7849,
+      "step": 42200
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 9.442360013029711e-05,
+      "loss": 0.7894,
+      "step": 42400
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 9.439722419675392e-05,
+      "loss": 0.7935,
+      "step": 42600
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 9.437084826321072e-05,
+      "loss": 0.7909,
+      "step": 42800
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 9.434447232966753e-05,
+      "loss": 0.7921,
+      "step": 43000
+    },
+    {
+      "epoch": 0.06,
+      "eval_em_first_line_ratio": 0.4816,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3692,
+      "eval_loss": 0.7668728232383728,
+      "eval_runtime": 316.9263,
+      "eval_samples_per_second": 31.553,
+      "eval_steps_per_second": 0.167,
+      "step": 43000
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 9.431809639612432e-05,
+      "loss": 0.793,
+      "step": 43200
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 9.429172046258112e-05,
+      "loss": 0.7907,
+      "step": 43400
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 9.426534452903793e-05,
+      "loss": 0.7884,
+      "step": 43600
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 9.423896859549473e-05,
+      "loss": 0.7871,
+      "step": 43800
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 9.421259266195154e-05,
+      "loss": 0.7859,
+      "step": 44000
+    },
+    {
+      "epoch": 0.06,
+      "eval_em_first_line_ratio": 0.4859,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3726,
+      "eval_loss": 0.7652727365493774,
+      "eval_runtime": 303.3051,
+      "eval_samples_per_second": 32.97,
+      "eval_steps_per_second": 0.175,
+      "step": 44000
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 9.418621672840834e-05,
+      "loss": 0.7879,
+      "step": 44200
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 9.415997267453285e-05,
+      "loss": 0.7818,
+      "step": 44400
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 9.413359674098966e-05,
+      "loss": 0.7884,
+      "step": 44600
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 9.410722080744646e-05,
+      "loss": 0.7886,
+      "step": 44800
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 9.408084487390327e-05,
+      "loss": 0.7846,
+      "step": 45000
+    },
+    {
+      "epoch": 0.06,
+      "eval_em_first_line_ratio": 0.4874,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3748,
+      "eval_loss": 0.7626891732215881,
+      "eval_runtime": 332.9525,
+      "eval_samples_per_second": 30.034,
+      "eval_steps_per_second": 0.159,
+      "step": 45000
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 9.405446894036006e-05,
+      "loss": 0.7848,
+      "step": 45200
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 9.402809300681686e-05,
+      "loss": 0.7856,
+      "step": 45400
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 9.400171707327367e-05,
+      "loss": 0.7811,
+      "step": 45600
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 9.397534113973047e-05,
+      "loss": 0.7875,
+      "step": 45800
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 9.394896520618726e-05,
+      "loss": 0.783,
+      "step": 46000
+    },
+    {
+      "epoch": 0.06,
+      "eval_em_first_line_ratio": 0.4863,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3743,
+      "eval_loss": 0.7623755931854248,
+      "eval_runtime": 312.5491,
+      "eval_samples_per_second": 31.995,
+      "eval_steps_per_second": 0.17,
+      "step": 46000
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 9.392258927264407e-05,
+      "loss": 0.7841,
+      "step": 46200
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 9.389621333910089e-05,
+      "loss": 0.7837,
+      "step": 46400
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 9.386996928522538e-05,
+      "loss": 0.7821,
+      "step": 46600
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 9.384359335168219e-05,
+      "loss": 0.78,
+      "step": 46800
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 9.381721741813901e-05,
+      "loss": 0.7809,
+      "step": 47000
+    },
+    {
+      "epoch": 0.06,
+      "eval_em_first_line_ratio": 0.49,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.377,
+      "eval_loss": 0.7606080770492554,
+      "eval_runtime": 295.8672,
+      "eval_samples_per_second": 33.799,
+      "eval_steps_per_second": 0.179,
+      "step": 47000
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 9.37908414845958e-05,
+      "loss": 0.7829,
+      "step": 47200
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 9.37644655510526e-05,
+      "loss": 0.7813,
+      "step": 47400
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 9.373808961750941e-05,
+      "loss": 0.7822,
+      "step": 47600
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 9.371171368396621e-05,
+      "loss": 0.7815,
+      "step": 47800
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 9.3685337750423e-05,
+      "loss": 0.7835,
+      "step": 48000
+    },
+    {
+      "epoch": 0.06,
+      "eval_em_first_line_ratio": 0.4868,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3737,
+      "eval_loss": 0.7599393725395203,
+      "eval_runtime": 282.0663,
+      "eval_samples_per_second": 35.453,
+      "eval_steps_per_second": 0.188,
+      "step": 48000
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 9.36589618168798e-05,
+      "loss": 0.7822,
+      "step": 48200
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 9.363258588333661e-05,
+      "loss": 0.7834,
+      "step": 48400
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 9.360620994979342e-05,
+      "loss": 0.7834,
+      "step": 48600
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 9.35798340162502e-05,
+      "loss": 0.7796,
+      "step": 48800
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 9.355345808270702e-05,
+      "loss": 0.7741,
+      "step": 49000
+    },
+    {
+      "epoch": 0.06,
+      "eval_em_first_line_ratio": 0.487,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3734,
+      "eval_loss": 0.7583541870117188,
+      "eval_runtime": 289.555,
+      "eval_samples_per_second": 34.536,
+      "eval_steps_per_second": 0.183,
+      "step": 49000
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 9.352708214916383e-05,
+      "loss": 0.7758,
+      "step": 49200
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 9.350070621562062e-05,
+      "loss": 0.7782,
+      "step": 49400
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 9.347433028207742e-05,
+      "loss": 0.7771,
+      "step": 49600
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 9.344795434853423e-05,
+      "loss": 0.7746,
+      "step": 49800
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 9.342157841499103e-05,
+      "loss": 0.7811,
+      "step": 50000
+    },
+    {
+      "epoch": 0.07,
+      "eval_em_first_line_ratio": 0.4861,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3728,
+      "eval_loss": 0.7574751377105713,
+      "eval_runtime": 289.5873,
+      "eval_samples_per_second": 34.532,
+      "eval_steps_per_second": 0.183,
+      "step": 50000
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 9.339520248144782e-05,
+      "loss": 0.7777,
+      "step": 50200
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 9.336882654790463e-05,
+      "loss": 0.7776,
+      "step": 50400
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 9.334245061436143e-05,
+      "loss": 0.781,
+      "step": 50600
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 9.331620656048596e-05,
+      "loss": 0.7728,
+      "step": 50800
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 9.328983062694275e-05,
+      "loss": 0.7761,
+      "step": 51000
+    },
+    {
+      "epoch": 0.07,
+      "eval_em_first_line_ratio": 0.4883,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3749,
+      "eval_loss": 0.7560361623764038,
+      "eval_runtime": 286.4122,
+      "eval_samples_per_second": 34.915,
+      "eval_steps_per_second": 0.185,
+      "step": 51000
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 9.326345469339955e-05,
+      "loss": 0.7784,
+      "step": 51200
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 9.323707875985637e-05,
+      "loss": 0.7739,
+      "step": 51400
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 9.321070282631316e-05,
+      "loss": 0.7777,
+      "step": 51600
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 9.318432689276997e-05,
+      "loss": 0.7798,
+      "step": 51800
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 9.315795095922677e-05,
+      "loss": 0.7804,
+      "step": 52000
+    },
+    {
+      "epoch": 0.07,
+      "eval_em_first_line_ratio": 0.4833,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3714,
+      "eval_loss": 0.7572079300880432,
+      "eval_runtime": 282.4675,
+      "eval_samples_per_second": 35.402,
+      "eval_steps_per_second": 0.188,
+      "step": 52000
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 9.313157502568358e-05,
+      "loss": 0.7772,
+      "step": 52200
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 9.310519909214037e-05,
+      "loss": 0.7758,
+      "step": 52400
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 9.307882315859717e-05,
+      "loss": 0.7749,
+      "step": 52600
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 9.30525791047217e-05,
+      "loss": 0.7722,
+      "step": 52800
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 9.302620317117849e-05,
+      "loss": 0.7743,
+      "step": 53000
+    },
+    {
+      "epoch": 0.07,
+      "eval_em_first_line_ratio": 0.4865,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3734,
+      "eval_loss": 0.7539185285568237,
+      "eval_runtime": 275.8337,
+      "eval_samples_per_second": 36.254,
+      "eval_steps_per_second": 0.192,
+      "step": 53000
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 9.29998272376353e-05,
+      "loss": 0.7709,
+      "step": 53200
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 9.29734513040921e-05,
+      "loss": 0.7761,
+      "step": 53400
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 9.29470753705489e-05,
+      "loss": 0.7742,
+      "step": 53600
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 9.29206994370057e-05,
+      "loss": 0.7731,
+      "step": 53800
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 9.289432350346251e-05,
+      "loss": 0.7725,
+      "step": 54000
+    },
+    {
+      "epoch": 0.07,
+      "eval_em_first_line_ratio": 0.4888,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3755,
+      "eval_loss": 0.7536767721176147,
+      "eval_runtime": 271.7713,
+      "eval_samples_per_second": 36.796,
+      "eval_steps_per_second": 0.195,
+      "step": 54000
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 9.286794756991932e-05,
+      "loss": 0.7731,
+      "step": 54200
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 9.284157163637611e-05,
+      "loss": 0.7739,
+      "step": 54400
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 9.281532758250063e-05,
+      "loss": 0.7736,
+      "step": 54600
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 9.278895164895744e-05,
+      "loss": 0.7723,
+      "step": 54800
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 9.276257571541423e-05,
+      "loss": 0.7714,
+      "step": 55000
+    },
+    {
+      "epoch": 0.07,
+      "eval_em_first_line_ratio": 0.4895,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3767,
+      "eval_loss": 0.7519065141677856,
+      "eval_runtime": 288.3489,
+      "eval_samples_per_second": 34.68,
+      "eval_steps_per_second": 0.184,
+      "step": 55000
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 9.273619978187103e-05,
+      "loss": 0.7699,
+      "step": 55200
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 9.270982384832784e-05,
+      "loss": 0.7741,
+      "step": 55400
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 9.268344791478464e-05,
+      "loss": 0.7786,
+      "step": 55600
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 9.265707198124143e-05,
+      "loss": 0.7724,
+      "step": 55800
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 9.263069604769824e-05,
+      "loss": 0.7743,
+      "step": 56000
+    },
+    {
+      "epoch": 0.07,
+      "eval_em_first_line_ratio": 0.4881,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3756,
+      "eval_loss": 0.7522148489952087,
+      "eval_runtime": 287.4147,
+      "eval_samples_per_second": 34.793,
+      "eval_steps_per_second": 0.184,
+      "step": 56000
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 9.260432011415504e-05,
+      "loss": 0.7711,
+      "step": 56200
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 9.257794418061185e-05,
+      "loss": 0.7686,
+      "step": 56400
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 9.255156824706865e-05,
+      "loss": 0.7671,
+      "step": 56600
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 9.252519231352546e-05,
+      "loss": 0.7699,
+      "step": 56800
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 9.249881637998226e-05,
+      "loss": 0.7642,
+      "step": 57000
+    },
+    {
+      "epoch": 0.08,
+      "eval_em_first_line_ratio": 0.4861,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3732,
+      "eval_loss": 0.7505404353141785,
+      "eval_runtime": 273.9789,
+      "eval_samples_per_second": 36.499,
+      "eval_steps_per_second": 0.193,
+      "step": 57000
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 9.247257232610677e-05,
+      "loss": 0.7688,
+      "step": 57200
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 9.244619639256358e-05,
+      "loss": 0.7717,
+      "step": 57400
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 9.241982045902038e-05,
+      "loss": 0.7697,
+      "step": 57600
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 9.239344452547719e-05,
+      "loss": 0.7653,
+      "step": 57800
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 9.236706859193398e-05,
+      "loss": 0.7703,
+      "step": 58000
+    },
+    {
+      "epoch": 0.08,
+      "eval_em_first_line_ratio": 0.4901,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3765,
+      "eval_loss": 0.7501883506774902,
+      "eval_runtime": 276.0115,
+      "eval_samples_per_second": 36.23,
+      "eval_steps_per_second": 0.192,
+      "step": 58000
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 9.234069265839078e-05,
+      "loss": 0.7672,
+      "step": 58200
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 9.231431672484759e-05,
+      "loss": 0.7682,
+      "step": 58400
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 9.228794079130439e-05,
+      "loss": 0.769,
+      "step": 58600
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 9.226156485776118e-05,
+      "loss": 0.7741,
+      "step": 58800
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 9.223518892421799e-05,
+      "loss": 0.7677,
+      "step": 59000
+    },
+    {
+      "epoch": 0.08,
+      "eval_em_first_line_ratio": 0.4877,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3743,
+      "eval_loss": 0.7498610615730286,
+      "eval_runtime": 268.6376,
+      "eval_samples_per_second": 37.225,
+      "eval_steps_per_second": 0.197,
+      "step": 59000
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 9.22088129906748e-05,
+      "loss": 0.769,
+      "step": 59200
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 9.21824370571316e-05,
+      "loss": 0.7689,
+      "step": 59400
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 9.21560611235884e-05,
+      "loss": 0.7681,
+      "step": 59600
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 9.21296851900452e-05,
+      "loss": 0.7647,
+      "step": 59800
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 9.210330925650201e-05,
+      "loss": 0.7696,
+      "step": 60000
+    },
+    {
+      "epoch": 0.08,
+      "eval_em_first_line_ratio": 0.4869,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3744,
+      "eval_loss": 0.747885525226593,
+      "eval_runtime": 280.1175,
+      "eval_samples_per_second": 35.699,
+      "eval_steps_per_second": 0.189,
+      "step": 60000
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 9.20769333229588e-05,
+      "loss": 0.7641,
+      "step": 60200
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 9.20505573894156e-05,
+      "loss": 0.7573,
+      "step": 60400
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 9.202418145587241e-05,
+      "loss": 0.762,
+      "step": 60600
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 9.199780552232921e-05,
+      "loss": 0.7704,
+      "step": 60800
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 9.197142958878602e-05,
+      "loss": 0.7678,
+      "step": 61000
+    },
+    {
+      "epoch": 0.08,
+      "eval_em_first_line_ratio": 0.4867,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3748,
+      "eval_loss": 0.748175859451294,
+      "eval_runtime": 276.6696,
+      "eval_samples_per_second": 36.144,
+      "eval_steps_per_second": 0.192,
+      "step": 61000
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 9.194505365524282e-05,
+      "loss": 0.7691,
+      "step": 61200
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 9.191880960136733e-05,
+      "loss": 0.7679,
+      "step": 61400
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 9.189243366782414e-05,
+      "loss": 0.7701,
+      "step": 61600
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 9.186605773428094e-05,
+      "loss": 0.7614,
+      "step": 61800
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 9.183968180073775e-05,
+      "loss": 0.7634,
+      "step": 62000
+    },
+    {
+      "epoch": 0.08,
+      "eval_em_first_line_ratio": 0.4897,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3765,
+      "eval_loss": 0.7467879056930542,
+      "eval_runtime": 269.373,
+      "eval_samples_per_second": 37.123,
+      "eval_steps_per_second": 0.197,
+      "step": 62000
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 9.181330586719454e-05,
+      "loss": 0.7678,
+      "step": 62200
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 9.178692993365134e-05,
+      "loss": 0.7664,
+      "step": 62400
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 9.176055400010815e-05,
+      "loss": 0.759,
+      "step": 62600
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 9.173417806656495e-05,
+      "loss": 0.768,
+      "step": 62800
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 9.170780213302174e-05,
+      "loss": 0.7629,
+      "step": 63000
+    },
+    {
+      "epoch": 0.08,
+      "eval_em_first_line_ratio": 0.4879,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3751,
+      "eval_loss": 0.7455302476882935,
+      "eval_runtime": 276.0562,
+      "eval_samples_per_second": 36.225,
+      "eval_steps_per_second": 0.192,
+      "step": 63000
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 9.168142619947855e-05,
+      "loss": 0.7614,
+      "step": 63200
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 9.165505026593535e-05,
+      "loss": 0.7645,
+      "step": 63400
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 9.162867433239216e-05,
+      "loss": 0.7597,
+      "step": 63600
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 9.160229839884896e-05,
+      "loss": 0.7668,
+      "step": 63800
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 9.157592246530576e-05,
+      "loss": 0.7622,
+      "step": 64000
+    },
+    {
+      "epoch": 0.08,
+      "eval_em_first_line_ratio": 0.4896,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3771,
+      "eval_loss": 0.7450270652770996,
+      "eval_runtime": 264.7095,
+      "eval_samples_per_second": 37.777,
+      "eval_steps_per_second": 0.2,
+      "step": 64000
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 9.154954653176257e-05,
+      "loss": 0.762,
+      "step": 64200
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 9.152317059821936e-05,
+      "loss": 0.7611,
+      "step": 64400
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 9.149679466467616e-05,
+      "loss": 0.7616,
+      "step": 64600
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 9.147055061080069e-05,
+      "loss": 0.7604,
+      "step": 64800
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 9.144417467725748e-05,
+      "loss": 0.7657,
+      "step": 65000
+    },
+    {
+      "epoch": 0.09,
+      "eval_em_first_line_ratio": 0.4905,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3774,
+      "eval_loss": 0.744494616985321,
+      "eval_runtime": 254.6849,
+      "eval_samples_per_second": 39.264,
+      "eval_steps_per_second": 0.208,
+      "step": 65000
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 9.141779874371429e-05,
+      "loss": 0.7596,
+      "step": 65200
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 9.139142281017109e-05,
+      "loss": 0.7605,
+      "step": 65400
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 9.13650468766279e-05,
+      "loss": 0.7554,
+      "step": 65600
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 9.133867094308469e-05,
+      "loss": 0.7623,
+      "step": 65800
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 9.131229500954149e-05,
+      "loss": 0.7619,
+      "step": 66000
+    },
+    {
+      "epoch": 0.09,
+      "eval_em_first_line_ratio": 0.4852,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3729,
+      "eval_loss": 0.7440447807312012,
+      "eval_runtime": 265.4883,
+      "eval_samples_per_second": 37.666,
+      "eval_steps_per_second": 0.2,
+      "step": 66000
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 9.128591907599831e-05,
+      "loss": 0.7577,
+      "step": 66200
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 9.125954314245511e-05,
+      "loss": 0.7637,
+      "step": 66400
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 9.12331672089119e-05,
+      "loss": 0.7618,
+      "step": 66600
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 9.120679127536871e-05,
+      "loss": 0.7654,
+      "step": 66800
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 9.118041534182551e-05,
+      "loss": 0.7607,
+      "step": 67000
+    },
+    {
+      "epoch": 0.09,
+      "eval_em_first_line_ratio": 0.4889,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3761,
+      "eval_loss": 0.743151843547821,
+      "eval_runtime": 262.6838,
+      "eval_samples_per_second": 38.069,
+      "eval_steps_per_second": 0.202,
+      "step": 67000
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 9.115403940828232e-05,
+      "loss": 0.7596,
+      "step": 67200
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 9.112779535440683e-05,
+      "loss": 0.7576,
+      "step": 67400
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 9.110141942086363e-05,
+      "loss": 0.7628,
+      "step": 67600
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 9.107504348732044e-05,
+      "loss": 0.7609,
+      "step": 67800
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 9.104879943344495e-05,
+      "loss": 0.7576,
+      "step": 68000
+    },
+    {
+      "epoch": 0.09,
+      "eval_em_first_line_ratio": 0.4901,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3768,
+      "eval_loss": 0.7415655255317688,
+      "eval_runtime": 255.376,
+      "eval_samples_per_second": 39.158,
+      "eval_steps_per_second": 0.208,
+      "step": 68000
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 9.102242349990176e-05,
+      "loss": 0.7603,
+      "step": 68200
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 9.099604756635856e-05,
+      "loss": 0.7556,
+      "step": 68400
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 9.096967163281535e-05,
+      "loss": 0.7526,
+      "step": 68600
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 9.094329569927216e-05,
+      "loss": 0.7576,
+      "step": 68800
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 9.091691976572896e-05,
+      "loss": 0.7576,
+      "step": 69000
+    },
+    {
+      "epoch": 0.09,
+      "eval_em_first_line_ratio": 0.4927,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3796,
+      "eval_loss": 0.739445686340332,
+      "eval_runtime": 248.1443,
+      "eval_samples_per_second": 40.299,
+      "eval_steps_per_second": 0.214,
+      "step": 69000
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 9.089054383218577e-05,
+      "loss": 0.756,
+      "step": 69200
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 9.086416789864257e-05,
+      "loss": 0.7576,
+      "step": 69400
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 9.083779196509937e-05,
+      "loss": 0.7585,
+      "step": 69600
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 9.081141603155618e-05,
+      "loss": 0.7625,
+      "step": 69800
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 9.078517197768069e-05,
+      "loss": 0.7642,
+      "step": 70000
+    },
+    {
+      "epoch": 0.09,
+      "eval_em_first_line_ratio": 0.4858,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3726,
+      "eval_loss": 0.742847740650177,
+      "eval_runtime": 255.7614,
+      "eval_samples_per_second": 39.099,
+      "eval_steps_per_second": 0.207,
+      "step": 70000
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 9.07587960441375e-05,
+      "loss": 0.7597,
+      "step": 70200
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 9.07324201105943e-05,
+      "loss": 0.7546,
+      "step": 70400
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 9.070604417705109e-05,
+      "loss": 0.7607,
+      "step": 70600
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 9.06796682435079e-05,
+      "loss": 0.7538,
+      "step": 70800
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 9.06532923099647e-05,
+      "loss": 0.7532,
+      "step": 71000
+    },
+    {
+      "epoch": 0.09,
+      "eval_em_first_line_ratio": 0.4927,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3791,
+      "eval_loss": 0.7391790747642517,
+      "eval_runtime": 253.5667,
+      "eval_samples_per_second": 39.437,
+      "eval_steps_per_second": 0.209,
+      "step": 71000
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 9.06269163764215e-05,
+      "loss": 0.7542,
+      "step": 71200
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 9.06005404428783e-05,
+      "loss": 0.7541,
+      "step": 71400
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 9.05741645093351e-05,
+      "loss": 0.7554,
+      "step": 71600
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 9.054778857579192e-05,
+      "loss": 0.7553,
+      "step": 71800
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 9.052141264224871e-05,
+      "loss": 0.7535,
+      "step": 72000
+    },
+    {
+      "epoch": 0.09,
+      "eval_em_first_line_ratio": 0.49,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3769,
+      "eval_loss": 0.73856520652771,
+      "eval_runtime": 240.2356,
+      "eval_samples_per_second": 41.626,
+      "eval_steps_per_second": 0.221,
+      "step": 72000
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 9.049503670870551e-05,
+      "loss": 0.7555,
+      "step": 72200
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 9.046866077516232e-05,
+      "loss": 0.7511,
+      "step": 72400
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 9.044228484161912e-05,
+      "loss": 0.7496,
+      "step": 72600
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 9.041590890807591e-05,
+      "loss": 0.758,
+      "step": 72800
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 9.038953297453272e-05,
+      "loss": 0.7581,
+      "step": 73000
+    },
+    {
+      "epoch": 0.1,
+      "eval_em_first_line_ratio": 0.4898,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.377,
+      "eval_loss": 0.7394999265670776,
+      "eval_runtime": 226.1336,
+      "eval_samples_per_second": 44.222,
+      "eval_steps_per_second": 0.234,
+      "step": 73000
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 9.036315704098952e-05,
+      "loss": 0.7568,
+      "step": 73200
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 9.033678110744633e-05,
+      "loss": 0.7557,
+      "step": 73400
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 9.031040517390312e-05,
+      "loss": 0.753,
+      "step": 73600
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 9.028402924035993e-05,
+      "loss": 0.754,
+      "step": 73800
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 9.025765330681674e-05,
+      "loss": 0.7534,
+      "step": 74000
+    },
+    {
+      "epoch": 0.1,
+      "eval_em_first_line_ratio": 0.4913,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3782,
+      "eval_loss": 0.7374927997589111,
+      "eval_runtime": 235.637,
+      "eval_samples_per_second": 42.438,
+      "eval_steps_per_second": 0.225,
+      "step": 74000
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 9.023127737327354e-05,
+      "loss": 0.7546,
+      "step": 74200
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 9.020490143973033e-05,
+      "loss": 0.7524,
+      "step": 74400
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 9.017852550618714e-05,
+      "loss": 0.7553,
+      "step": 74600
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 9.015214957264394e-05,
+      "loss": 0.7545,
+      "step": 74800
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 9.012577363910075e-05,
+      "loss": 0.7537,
+      "step": 75000
+    },
+    {
+      "epoch": 0.1,
+      "eval_em_first_line_ratio": 0.4908,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3779,
+      "eval_loss": 0.7363139986991882,
+      "eval_runtime": 236.4387,
+      "eval_samples_per_second": 42.294,
+      "eval_steps_per_second": 0.224,
+      "step": 75000
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 9.009939770555754e-05,
+      "loss": 0.7542,
+      "step": 75200
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 9.007302177201434e-05,
+      "loss": 0.7474,
+      "step": 75400
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 9.004664583847115e-05,
+      "loss": 0.7528,
+      "step": 75600
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 9.002026990492795e-05,
+      "loss": 0.7481,
+      "step": 75800
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 8.999389397138476e-05,
+      "loss": 0.7483,
+      "step": 76000
+    },
+    {
+      "epoch": 0.1,
+      "eval_em_first_line_ratio": 0.4939,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3803,
+      "eval_loss": 0.7362607717514038,
+      "eval_runtime": 233.2654,
+      "eval_samples_per_second": 42.87,
+      "eval_steps_per_second": 0.227,
+      "step": 76000
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 8.996751803784156e-05,
+      "loss": 0.7499,
+      "step": 76200
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 8.994114210429837e-05,
+      "loss": 0.7517,
+      "step": 76400
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 8.991489805042288e-05,
+      "loss": 0.7462,
+      "step": 76600
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 8.988852211687968e-05,
+      "loss": 0.7443,
+      "step": 76800
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 8.986214618333649e-05,
+      "loss": 0.7456,
+      "step": 77000
+    },
+    {
+      "epoch": 0.1,
+      "eval_em_first_line_ratio": 0.4896,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3769,
+      "eval_loss": 0.7355839610099792,
+      "eval_runtime": 231.1313,
+      "eval_samples_per_second": 43.265,
+      "eval_steps_per_second": 0.229,
+      "step": 77000
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 8.9835902129461e-05,
+      "loss": 0.7473,
+      "step": 77200
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 8.98095261959178e-05,
+      "loss": 0.7487,
+      "step": 77400
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 8.978315026237461e-05,
+      "loss": 0.7475,
+      "step": 77600
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 8.97567743288314e-05,
+      "loss": 0.744,
+      "step": 77800
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 8.97303983952882e-05,
+      "loss": 0.7452,
+      "step": 78000
+    },
+    {
+      "epoch": 0.1,
+      "eval_em_first_line_ratio": 0.4928,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3794,
+      "eval_loss": 0.7345384955406189,
+      "eval_runtime": 229.2778,
+      "eval_samples_per_second": 43.615,
+      "eval_steps_per_second": 0.231,
+      "step": 78000
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 8.970402246174501e-05,
+      "loss": 0.7478,
+      "step": 78200
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 8.967764652820181e-05,
+      "loss": 0.7455,
+      "step": 78400
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 8.96512705946586e-05,
+      "loss": 0.7487,
+      "step": 78600
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 8.962489466111542e-05,
+      "loss": 0.7462,
+      "step": 78800
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 8.959851872757223e-05,
+      "loss": 0.749,
+      "step": 79000
+    },
+    {
+      "epoch": 0.1,
+      "eval_em_first_line_ratio": 0.4939,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3801,
+      "eval_loss": 0.7336086630821228,
+      "eval_runtime": 223.8,
+      "eval_samples_per_second": 44.683,
+      "eval_steps_per_second": 0.237,
+      "step": 79000
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 8.957227467369673e-05,
+      "loss": 0.752,
+      "step": 79200
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 8.954589874015353e-05,
+      "loss": 0.7432,
+      "step": 79400
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 8.951952280661035e-05,
+      "loss": 0.7514,
+      "step": 79600
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 8.949314687306714e-05,
+      "loss": 0.7488,
+      "step": 79800
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 8.946677093952394e-05,
+      "loss": 0.7505,
+      "step": 80000
+    },
+    {
+      "epoch": 0.11,
+      "eval_em_first_line_ratio": 0.4894,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.377,
+      "eval_loss": 0.7347418665885925,
+      "eval_runtime": 229.419,
+      "eval_samples_per_second": 43.588,
+      "eval_steps_per_second": 0.231,
+      "step": 80000
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 8.944052688564847e-05,
+      "loss": 0.7431,
+      "step": 80200
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 8.941415095210528e-05,
+      "loss": 0.7447,
+      "step": 80400
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 8.938777501856207e-05,
+      "loss": 0.7459,
+      "step": 80600
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 8.936139908501887e-05,
+      "loss": 0.7453,
+      "step": 80800
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 8.93351550311434e-05,
+      "loss": 0.746,
+      "step": 81000
+    },
+    {
+      "epoch": 0.11,
+      "eval_em_first_line_ratio": 0.4939,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3804,
+      "eval_loss": 0.732323408126831,
+      "eval_runtime": 226.544,
+      "eval_samples_per_second": 44.142,
+      "eval_steps_per_second": 0.234,
+      "step": 81000
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 8.930877909760019e-05,
+      "loss": 0.7428,
+      "step": 81200
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 8.928240316405699e-05,
+      "loss": 0.7444,
+      "step": 81400
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 8.92560272305138e-05,
+      "loss": 0.7534,
+      "step": 81600
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 8.92296512969706e-05,
+      "loss": 0.748,
+      "step": 81800
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 8.920327536342739e-05,
+      "loss": 0.7524,
+      "step": 82000
+    },
+    {
+      "epoch": 0.11,
+      "eval_em_first_line_ratio": 0.4922,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3791,
+      "eval_loss": 0.7321872711181641,
+      "eval_runtime": 229.8878,
+      "eval_samples_per_second": 43.499,
+      "eval_steps_per_second": 0.231,
+      "step": 82000
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 8.91768994298842e-05,
+      "loss": 0.7488,
+      "step": 82200
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 8.9150523496341e-05,
+      "loss": 0.749,
+      "step": 82400
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 8.91241475627978e-05,
+      "loss": 0.7442,
+      "step": 82600
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 8.909777162925461e-05,
+      "loss": 0.742,
+      "step": 82800
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 8.907139569571141e-05,
+      "loss": 0.7471,
+      "step": 83000
+    },
+    {
+      "epoch": 0.11,
+      "eval_em_first_line_ratio": 0.4908,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3781,
+      "eval_loss": 0.73191237449646,
+      "eval_runtime": 236.3687,
+      "eval_samples_per_second": 42.307,
+      "eval_steps_per_second": 0.224,
+      "step": 83000
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 8.904501976216822e-05,
+      "loss": 0.7452,
+      "step": 83200
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 8.901864382862501e-05,
+      "loss": 0.7461,
+      "step": 83400
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 8.899226789508181e-05,
+      "loss": 0.7513,
+      "step": 83600
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 8.896589196153862e-05,
+      "loss": 0.743,
+      "step": 83800
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 8.893951602799542e-05,
+      "loss": 0.7424,
+      "step": 84000
+    },
+    {
+      "epoch": 0.11,
+      "eval_em_first_line_ratio": 0.4912,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3782,
+      "eval_loss": 0.7312846183776855,
+      "eval_runtime": 231.3432,
+      "eval_samples_per_second": 43.226,
+      "eval_steps_per_second": 0.229,
+      "step": 84000
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 8.891314009445221e-05,
+      "loss": 0.7504,
+      "step": 84200
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 8.888676416090902e-05,
+      "loss": 0.7448,
+      "step": 84400
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 8.886038822736584e-05,
+      "loss": 0.7401,
+      "step": 84600
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 8.883401229382263e-05,
+      "loss": 0.7425,
+      "step": 84800
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 8.880763636027943e-05,
+      "loss": 0.7425,
+      "step": 85000
+    },
+    {
+      "epoch": 0.11,
+      "eval_em_first_line_ratio": 0.4929,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3802,
+      "eval_loss": 0.7295014262199402,
+      "eval_runtime": 236.022,
+      "eval_samples_per_second": 42.369,
+      "eval_steps_per_second": 0.225,
+      "step": 85000
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 8.878126042673624e-05,
+      "loss": 0.7417,
+      "step": 85200
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 8.875488449319304e-05,
+      "loss": 0.7432,
+      "step": 85400
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 8.872850855964983e-05,
+      "loss": 0.7469,
+      "step": 85600
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 8.870213262610664e-05,
+      "loss": 0.7444,
+      "step": 85800
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 8.867575669256344e-05,
+      "loss": 0.7416,
+      "step": 86000
+    },
+    {
+      "epoch": 0.11,
+      "eval_em_first_line_ratio": 0.4914,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3782,
+      "eval_loss": 0.7297839522361755,
+      "eval_runtime": 229.9174,
+      "eval_samples_per_second": 43.494,
+      "eval_steps_per_second": 0.231,
+      "step": 86000
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 8.864938075902024e-05,
+      "loss": 0.7415,
+      "step": 86200
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 8.862300482547704e-05,
+      "loss": 0.7417,
+      "step": 86400
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 8.859662889193385e-05,
+      "loss": 0.7466,
+      "step": 86600
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 8.857025295839066e-05,
+      "loss": 0.7433,
+      "step": 86800
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 8.854387702484745e-05,
+      "loss": 0.7426,
+      "step": 87000
+    },
+    {
+      "epoch": 0.11,
+      "eval_em_first_line_ratio": 0.4942,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3805,
+      "eval_loss": 0.7285897731781006,
+      "eval_runtime": 233.5931,
+      "eval_samples_per_second": 42.809,
+      "eval_steps_per_second": 0.227,
+      "step": 87000
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 8.851750109130425e-05,
+      "loss": 0.7503,
+      "step": 87200
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 8.849112515776106e-05,
+      "loss": 0.7423,
+      "step": 87400
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 8.846488110388557e-05,
+      "loss": 0.7478,
+      "step": 87600
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 8.843850517034238e-05,
+      "loss": 0.7433,
+      "step": 87800
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 8.841212923679918e-05,
+      "loss": 0.7395,
+      "step": 88000
+    },
+    {
+      "epoch": 0.12,
+      "eval_em_first_line_ratio": 0.4913,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3784,
+      "eval_loss": 0.7287758588790894,
+      "eval_runtime": 230.7854,
+      "eval_samples_per_second": 43.33,
+      "eval_steps_per_second": 0.23,
+      "step": 88000
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 8.838575330325598e-05,
+      "loss": 0.7448,
+      "step": 88200
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 8.835937736971277e-05,
+      "loss": 0.7454,
+      "step": 88400
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 8.833300143616958e-05,
+      "loss": 0.7384,
+      "step": 88600
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 8.830662550262638e-05,
+      "loss": 0.7376,
+      "step": 88800
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 8.828038144875091e-05,
+      "loss": 0.7367,
+      "step": 89000
+    },
+    {
+      "epoch": 0.12,
+      "eval_em_first_line_ratio": 0.4928,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3792,
+      "eval_loss": 0.7280991077423096,
+      "eval_runtime": 236.2705,
+      "eval_samples_per_second": 42.324,
+      "eval_steps_per_second": 0.224,
+      "step": 89000
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 8.82540055152077e-05,
+      "loss": 0.7385,
+      "step": 89200
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 8.82276295816645e-05,
+      "loss": 0.7376,
+      "step": 89400
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 8.820125364812131e-05,
+      "loss": 0.7454,
+      "step": 89600
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 8.817487771457811e-05,
+      "loss": 0.7375,
+      "step": 89800
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 8.814850178103492e-05,
+      "loss": 0.7382,
+      "step": 90000
+    },
+    {
+      "epoch": 0.12,
+      "eval_em_first_line_ratio": 0.4944,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3809,
+      "eval_loss": 0.7268965840339661,
+      "eval_runtime": 225.9428,
+      "eval_samples_per_second": 44.259,
+      "eval_steps_per_second": 0.235,
+      "step": 90000
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 8.812212584749172e-05,
+      "loss": 0.739,
+      "step": 90200
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 8.809574991394853e-05,
+      "loss": 0.7422,
+      "step": 90400
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 8.806937398040532e-05,
+      "loss": 0.7411,
+      "step": 90600
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 8.804299804686212e-05,
+      "loss": 0.7441,
+      "step": 90800
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 8.801662211331893e-05,
+      "loss": 0.7406,
+      "step": 91000
+    },
+    {
+      "epoch": 0.12,
+      "eval_em_first_line_ratio": 0.4964,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.383,
+      "eval_loss": 0.7255858182907104,
+      "eval_runtime": 232.3196,
+      "eval_samples_per_second": 43.044,
+      "eval_steps_per_second": 0.228,
+      "step": 91000
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 8.799024617977573e-05,
+      "loss": 0.738,
+      "step": 91200
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 8.796400212590025e-05,
+      "loss": 0.7429,
+      "step": 91400
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 8.793762619235705e-05,
+      "loss": 0.7375,
+      "step": 91600
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 8.791125025881385e-05,
+      "loss": 0.7394,
+      "step": 91800
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 8.788487432527065e-05,
+      "loss": 0.7368,
+      "step": 92000
+    },
+    {
+      "epoch": 0.12,
+      "eval_em_first_line_ratio": 0.496,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3825,
+      "eval_loss": 0.7258254885673523,
+      "eval_runtime": 231.3679,
+      "eval_samples_per_second": 43.221,
+      "eval_steps_per_second": 0.229,
+      "step": 92000
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 8.785849839172746e-05,
+      "loss": 0.7377,
+      "step": 92200
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 8.783212245818427e-05,
+      "loss": 0.7404,
+      "step": 92400
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 8.780574652464106e-05,
+      "loss": 0.7404,
+      "step": 92600
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 8.777937059109786e-05,
+      "loss": 0.7354,
+      "step": 92800
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 8.775299465755467e-05,
+      "loss": 0.7437,
+      "step": 93000
+    },
+    {
+      "epoch": 0.12,
+      "eval_em_first_line_ratio": 0.4953,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3812,
+      "eval_loss": 0.7254249453544617,
+      "eval_runtime": 225.2018,
+      "eval_samples_per_second": 44.405,
+      "eval_steps_per_second": 0.235,
+      "step": 93000
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 8.772661872401147e-05,
+      "loss": 0.7377,
+      "step": 93200
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 8.770024279046826e-05,
+      "loss": 0.7336,
+      "step": 93400
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 8.767386685692507e-05,
+      "loss": 0.7378,
+      "step": 93600
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 8.764749092338187e-05,
+      "loss": 0.7412,
+      "step": 93800
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 8.762111498983868e-05,
+      "loss": 0.7379,
+      "step": 94000
+    },
+    {
+      "epoch": 0.12,
+      "eval_em_first_line_ratio": 0.4939,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3807,
+      "eval_loss": 0.7243178486824036,
+      "eval_runtime": 235.4294,
+      "eval_samples_per_second": 42.476,
+      "eval_steps_per_second": 0.225,
+      "step": 94000
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 8.759473905629548e-05,
+      "loss": 0.7366,
+      "step": 94200
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 8.756836312275228e-05,
+      "loss": 0.7384,
+      "step": 94400
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 8.754198718920909e-05,
+      "loss": 0.7351,
+      "step": 94600
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 8.751561125566588e-05,
+      "loss": 0.7295,
+      "step": 94800
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 8.748923532212268e-05,
+      "loss": 0.7316,
+      "step": 95000
+    },
+    {
+      "epoch": 0.13,
+      "eval_em_first_line_ratio": 0.4914,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3779,
+      "eval_loss": 0.7241101264953613,
+      "eval_runtime": 236.1448,
+      "eval_samples_per_second": 42.347,
+      "eval_steps_per_second": 0.224,
+      "step": 95000
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 8.746299126824721e-05,
+      "loss": 0.738,
+      "step": 95200
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 8.7436615334704e-05,
+      "loss": 0.7423,
+      "step": 95400
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 8.74102394011608e-05,
+      "loss": 0.7441,
+      "step": 95600
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 8.738386346761761e-05,
+      "loss": 0.734,
+      "step": 95800
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 8.735748753407442e-05,
+      "loss": 0.7364,
+      "step": 96000
+    },
+    {
+      "epoch": 0.13,
+      "eval_em_first_line_ratio": 0.4905,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3777,
+      "eval_loss": 0.7244338989257812,
+      "eval_runtime": 235.4868,
+      "eval_samples_per_second": 42.465,
+      "eval_steps_per_second": 0.225,
+      "step": 96000
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 8.73311116005312e-05,
+      "loss": 0.7326,
+      "step": 96200
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 8.730473566698801e-05,
+      "loss": 0.7322,
+      "step": 96400
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 8.727835973344481e-05,
+      "loss": 0.7373,
+      "step": 96600
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 8.725198379990163e-05,
+      "loss": 0.732,
+      "step": 96800
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 8.722560786635842e-05,
+      "loss": 0.735,
+      "step": 97000
+    },
+    {
+      "epoch": 0.13,
+      "eval_em_first_line_ratio": 0.4927,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3794,
+      "eval_loss": 0.7244437336921692,
+      "eval_runtime": 229.7125,
+      "eval_samples_per_second": 43.533,
+      "eval_steps_per_second": 0.231,
+      "step": 97000
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 8.719936381248294e-05,
+      "loss": 0.7412,
+      "step": 97200
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 8.717298787893975e-05,
+      "loss": 0.7364,
+      "step": 97400
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 8.714661194539655e-05,
+      "loss": 0.7324,
+      "step": 97600
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 8.712023601185335e-05,
+      "loss": 0.736,
+      "step": 97800
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 8.709386007831015e-05,
+      "loss": 0.7342,
+      "step": 98000
+    },
+    {
+      "epoch": 0.13,
+      "eval_em_first_line_ratio": 0.4929,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3803,
+      "eval_loss": 0.7228650450706482,
+      "eval_runtime": 233.077,
+      "eval_samples_per_second": 42.904,
+      "eval_steps_per_second": 0.227,
+      "step": 98000
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 8.706748414476696e-05,
+      "loss": 0.7351,
+      "step": 98200
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 8.704110821122375e-05,
+      "loss": 0.7297,
+      "step": 98400
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 8.701473227768055e-05,
+      "loss": 0.7383,
+      "step": 98600
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 8.698835634413736e-05,
+      "loss": 0.7384,
+      "step": 98800
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 8.696198041059416e-05,
+      "loss": 0.734,
+      "step": 99000
+    },
+    {
+      "epoch": 0.13,
+      "eval_em_first_line_ratio": 0.4933,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3797,
+      "eval_loss": 0.7225985527038574,
+      "eval_runtime": 230.8248,
+      "eval_samples_per_second": 43.323,
+      "eval_steps_per_second": 0.23,
+      "step": 99000
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 8.693560447705097e-05,
+      "loss": 0.7341,
+      "step": 99200
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 8.690922854350777e-05,
+      "loss": 0.7341,
+      "step": 99400
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 8.688285260996458e-05,
+      "loss": 0.7362,
+      "step": 99600
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 8.685647667642137e-05,
+      "loss": 0.7366,
+      "step": 99800
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 8.683010074287817e-05,
+      "loss": 0.7302,
+      "step": 100000
+    },
+    {
+      "epoch": 0.13,
+      "eval_em_first_line_ratio": 0.4954,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3817,
+      "eval_loss": 0.7220028042793274,
+      "eval_runtime": 231.3209,
+      "eval_samples_per_second": 43.23,
+      "eval_steps_per_second": 0.229,
+      "step": 100000
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 8.680372480933498e-05,
+      "loss": 0.7335,
+      "step": 100200
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 8.677734887579178e-05,
+      "loss": 0.731,
+      "step": 100400
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 8.675097294224857e-05,
+      "loss": 0.7352,
+      "step": 100600
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 8.672459700870538e-05,
+      "loss": 0.7291,
+      "step": 100800
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 8.669822107516218e-05,
+      "loss": 0.7329,
+      "step": 101000
+    },
+    {
+      "epoch": 0.13,
+      "eval_em_first_line_ratio": 0.4942,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3811,
+      "eval_loss": 0.7217410206794739,
+      "eval_runtime": 237.3553,
+      "eval_samples_per_second": 42.131,
+      "eval_steps_per_second": 0.223,
+      "step": 101000
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 8.667184514161898e-05,
+      "loss": 0.7307,
+      "step": 101200
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 8.664546920807579e-05,
+      "loss": 0.7388,
+      "step": 101400
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 8.66190932745326e-05,
+      "loss": 0.7305,
+      "step": 101600
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 8.65927173409894e-05,
+      "loss": 0.7298,
+      "step": 101800
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 8.656647328711391e-05,
+      "loss": 0.7293,
+      "step": 102000
+    },
+    {
+      "epoch": 0.13,
+      "eval_em_first_line_ratio": 0.4959,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3821,
+      "eval_loss": 0.7212633490562439,
+      "eval_runtime": 239.0527,
+      "eval_samples_per_second": 41.832,
+      "eval_steps_per_second": 0.222,
+      "step": 102000
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 8.654009735357072e-05,
+      "loss": 0.7326,
+      "step": 102200
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 8.651372142002752e-05,
+      "loss": 0.7272,
+      "step": 102400
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 8.648734548648431e-05,
+      "loss": 0.7284,
+      "step": 102600
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 8.646096955294112e-05,
+      "loss": 0.7321,
+      "step": 102800
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 8.643459361939792e-05,
+      "loss": 0.7312,
+      "step": 103000
+    },
+    {
+      "epoch": 0.14,
+      "eval_em_first_line_ratio": 0.4965,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3831,
+      "eval_loss": 0.7213169932365417,
+      "eval_runtime": 231.9466,
+      "eval_samples_per_second": 43.113,
+      "eval_steps_per_second": 0.229,
+      "step": 103000
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 8.640821768585472e-05,
+      "loss": 0.73,
+      "step": 103200
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 8.638184175231152e-05,
+      "loss": 0.7273,
+      "step": 103400
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 8.635546581876832e-05,
+      "loss": 0.7285,
+      "step": 103600
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 8.632908988522514e-05,
+      "loss": 0.7294,
+      "step": 103800
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 8.630271395168193e-05,
+      "loss": 0.7341,
+      "step": 104000
+    },
+    {
+      "epoch": 0.14,
+      "eval_em_first_line_ratio": 0.4969,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3834,
+      "eval_loss": 0.7200750708580017,
+      "eval_runtime": 237.2452,
+      "eval_samples_per_second": 42.15,
+      "eval_steps_per_second": 0.223,
+      "step": 104000
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 8.627633801813873e-05,
+      "loss": 0.7319,
+      "step": 104200
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 8.624996208459554e-05,
+      "loss": 0.7313,
+      "step": 104400
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 8.622358615105234e-05,
+      "loss": 0.7266,
+      "step": 104600
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 8.619721021750913e-05,
+      "loss": 0.7313,
+      "step": 104800
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 8.617083428396594e-05,
+      "loss": 0.7332,
+      "step": 105000
+    },
+    {
+      "epoch": 0.14,
+      "eval_em_first_line_ratio": 0.4938,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3807,
+      "eval_loss": 0.7197936773300171,
+      "eval_runtime": 236.6786,
+      "eval_samples_per_second": 42.251,
+      "eval_steps_per_second": 0.224,
+      "step": 105000
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 8.614445835042274e-05,
+      "loss": 0.7306,
+      "step": 105200
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 8.611808241687955e-05,
+      "loss": 0.7315,
+      "step": 105400
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 8.609170648333635e-05,
+      "loss": 0.726,
+      "step": 105600
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 8.606546242946086e-05,
+      "loss": 0.7258,
+      "step": 105800
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 8.603908649591767e-05,
+      "loss": 0.7284,
+      "step": 106000
+    },
+    {
+      "epoch": 0.14,
+      "eval_em_first_line_ratio": 0.4937,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3809,
+      "eval_loss": 0.7198227643966675,
+      "eval_runtime": 231.6365,
+      "eval_samples_per_second": 43.171,
+      "eval_steps_per_second": 0.229,
+      "step": 106000
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 8.601271056237447e-05,
+      "loss": 0.7245,
+      "step": 106200
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 8.598633462883128e-05,
+      "loss": 0.7313,
+      "step": 106400
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 8.595995869528808e-05,
+      "loss": 0.7273,
+      "step": 106600
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 8.593358276174489e-05,
+      "loss": 0.7301,
+      "step": 106800
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 8.590720682820168e-05,
+      "loss": 0.7288,
+      "step": 107000
+    },
+    {
+      "epoch": 0.14,
+      "eval_em_first_line_ratio": 0.497,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3838,
+      "eval_loss": 0.7190430760383606,
+      "eval_runtime": 232.4182,
+      "eval_samples_per_second": 43.026,
+      "eval_steps_per_second": 0.228,
+      "step": 107000
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 8.588083089465848e-05,
+      "loss": 0.734,
+      "step": 107200
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 8.585445496111529e-05,
+      "loss": 0.7332,
+      "step": 107400
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 8.582807902757209e-05,
+      "loss": 0.7313,
+      "step": 107600
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 8.580170309402888e-05,
+      "loss": 0.7273,
+      "step": 107800
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 8.577532716048568e-05,
+      "loss": 0.7293,
+      "step": 108000
+    },
+    {
+      "epoch": 0.14,
+      "eval_em_first_line_ratio": 0.4937,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3805,
+      "eval_loss": 0.7188526391983032,
+      "eval_runtime": 238.3914,
+      "eval_samples_per_second": 41.948,
+      "eval_steps_per_second": 0.222,
+      "step": 108000
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 8.57489512269425e-05,
+      "loss": 0.7283,
+      "step": 108200
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 8.57225752933993e-05,
+      "loss": 0.7256,
+      "step": 108400
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 8.56961993598561e-05,
+      "loss": 0.7268,
+      "step": 108600
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 8.56698234263129e-05,
+      "loss": 0.729,
+      "step": 108800
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 8.564344749276971e-05,
+      "loss": 0.7266,
+      "step": 109000
+    },
+    {
+      "epoch": 0.14,
+      "eval_em_first_line_ratio": 0.4981,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3846,
+      "eval_loss": 0.7174707055091858,
+      "eval_runtime": 239.6201,
+      "eval_samples_per_second": 41.733,
+      "eval_steps_per_second": 0.221,
+      "step": 109000
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 8.561720343889422e-05,
+      "loss": 0.7246,
+      "step": 109200
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 8.559082750535102e-05,
+      "loss": 0.7271,
+      "step": 109400
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 8.556445157180783e-05,
+      "loss": 0.7241,
+      "step": 109600
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 8.553807563826462e-05,
+      "loss": 0.733,
+      "step": 109800
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 8.551169970472142e-05,
+      "loss": 0.7254,
+      "step": 110000
+    },
+    {
+      "epoch": 0.15,
+      "eval_em_first_line_ratio": 0.496,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3831,
+      "eval_loss": 0.7165535688400269,
+      "eval_runtime": 236.6449,
+      "eval_samples_per_second": 42.257,
+      "eval_steps_per_second": 0.224,
+      "step": 110000
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 8.548532377117823e-05,
+      "loss": 0.7233,
+      "step": 110200
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 8.545894783763503e-05,
+      "loss": 0.7269,
+      "step": 110400
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 8.543270378375955e-05,
+      "loss": 0.729,
+      "step": 110600
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 8.540632785021635e-05,
+      "loss": 0.7319,
+      "step": 110800
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 8.537995191667316e-05,
+      "loss": 0.7254,
+      "step": 111000
+    },
+    {
+      "epoch": 0.15,
+      "eval_em_first_line_ratio": 0.4915,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3786,
+      "eval_loss": 0.7179467082023621,
+      "eval_runtime": 250.0553,
+      "eval_samples_per_second": 39.991,
+      "eval_steps_per_second": 0.212,
+      "step": 111000
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 8.535357598312995e-05,
+      "loss": 0.7287,
+      "step": 111200
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 8.532720004958676e-05,
+      "loss": 0.7313,
+      "step": 111400
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 8.530082411604357e-05,
+      "loss": 0.7216,
+      "step": 111600
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 8.527444818250036e-05,
+      "loss": 0.7229,
+      "step": 111800
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 8.524807224895716e-05,
+      "loss": 0.7307,
+      "step": 112000
+    },
+    {
+      "epoch": 0.15,
+      "eval_em_first_line_ratio": 0.4973,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3846,
+      "eval_loss": 0.7168014645576477,
+      "eval_runtime": 242.3638,
+      "eval_samples_per_second": 41.26,
+      "eval_steps_per_second": 0.219,
+      "step": 112000
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 8.522169631541397e-05,
+      "loss": 0.7202,
+      "step": 112200
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 8.519532038187077e-05,
+      "loss": 0.7285,
+      "step": 112400
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 8.516894444832756e-05,
+      "loss": 0.7247,
+      "step": 112600
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 8.514256851478437e-05,
+      "loss": 0.7212,
+      "step": 112800
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 8.511619258124117e-05,
+      "loss": 0.723,
+      "step": 113000
+    },
+    {
+      "epoch": 0.15,
+      "eval_em_first_line_ratio": 0.5002,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3861,
+      "eval_loss": 0.7160351276397705,
+      "eval_runtime": 242.3169,
+      "eval_samples_per_second": 41.268,
+      "eval_steps_per_second": 0.219,
+      "step": 113000
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 8.508981664769798e-05,
+      "loss": 0.7239,
+      "step": 113200
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 8.506344071415478e-05,
+      "loss": 0.723,
+      "step": 113400
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 8.503706478061159e-05,
+      "loss": 0.7245,
+      "step": 113600
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 8.501068884706839e-05,
+      "loss": 0.7228,
+      "step": 113800
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 8.49843129135252e-05,
+      "loss": 0.721,
+      "step": 114000
+    },
+    {
+      "epoch": 0.15,
+      "eval_em_first_line_ratio": 0.4961,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3828,
+      "eval_loss": 0.7165052890777588,
+      "eval_runtime": 248.2937,
+      "eval_samples_per_second": 40.275,
+      "eval_steps_per_second": 0.213,
+      "step": 114000
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 8.495793697998199e-05,
+      "loss": 0.7225,
+      "step": 114200
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 8.493156104643879e-05,
+      "loss": 0.726,
+      "step": 114400
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 8.49051851128956e-05,
+      "loss": 0.7248,
+      "step": 114600
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 8.48788091793524e-05,
+      "loss": 0.7277,
+      "step": 114800
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 8.485243324580919e-05,
+      "loss": 0.7239,
+      "step": 115000
+    },
+    {
+      "epoch": 0.15,
+      "eval_em_first_line_ratio": 0.4978,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3846,
+      "eval_loss": 0.7142688632011414,
+      "eval_runtime": 239.9703,
+      "eval_samples_per_second": 41.672,
+      "eval_steps_per_second": 0.221,
+      "step": 115000
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 8.482605731226601e-05,
+      "loss": 0.7245,
+      "step": 115200
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 8.479968137872281e-05,
+      "loss": 0.7249,
+      "step": 115400
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 8.47733054451796e-05,
+      "loss": 0.7218,
+      "step": 115600
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 8.474692951163641e-05,
+      "loss": 0.7247,
+      "step": 115800
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 8.472055357809321e-05,
+      "loss": 0.7215,
+      "step": 116000
+    },
+    {
+      "epoch": 0.15,
+      "eval_em_first_line_ratio": 0.4975,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3841,
+      "eval_loss": 0.7153100371360779,
+      "eval_runtime": 250.1549,
+      "eval_samples_per_second": 39.975,
+      "eval_steps_per_second": 0.212,
+      "step": 116000
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 8.469417764455002e-05,
+      "loss": 0.719,
+      "step": 116200
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 8.466780171100681e-05,
+      "loss": 0.7228,
+      "step": 116400
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 8.464142577746361e-05,
+      "loss": 0.7229,
+      "step": 116600
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 8.461504984392042e-05,
+      "loss": 0.7225,
+      "step": 116800
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 8.458867391037722e-05,
+      "loss": 0.7203,
+      "step": 117000
+    },
+    {
+      "epoch": 0.15,
+      "eval_em_first_line_ratio": 0.4976,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3839,
+      "eval_loss": 0.7144935131072998,
+      "eval_runtime": 244.4768,
+      "eval_samples_per_second": 40.904,
+      "eval_steps_per_second": 0.217,
+      "step": 117000
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 8.456242985650173e-05,
+      "loss": 0.7238,
+      "step": 117200
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 8.453605392295854e-05,
+      "loss": 0.7197,
+      "step": 117400
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 8.450967798941534e-05,
+      "loss": 0.7198,
+      "step": 117600
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 8.448330205587215e-05,
+      "loss": 0.7201,
+      "step": 117800
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 8.445705800199666e-05,
+      "loss": 0.717,
+      "step": 118000
+    },
+    {
+      "epoch": 0.16,
+      "eval_em_first_line_ratio": 0.4983,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3848,
+      "eval_loss": 0.7141348123550415,
+      "eval_runtime": 243.5142,
+      "eval_samples_per_second": 41.065,
+      "eval_steps_per_second": 0.218,
+      "step": 118000
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 8.443068206845346e-05,
+      "loss": 0.7222,
+      "step": 118200
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 8.440430613491027e-05,
+      "loss": 0.7235,
+      "step": 118400
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 8.437793020136707e-05,
+      "loss": 0.7211,
+      "step": 118600
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 8.435168614749159e-05,
+      "loss": 0.7182,
+      "step": 118800
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 8.432531021394839e-05,
+      "loss": 0.7257,
+      "step": 119000
+    },
+    {
+      "epoch": 0.16,
+      "eval_em_first_line_ratio": 0.4978,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3853,
+      "eval_loss": 0.7131052017211914,
+      "eval_runtime": 242.7576,
+      "eval_samples_per_second": 41.193,
+      "eval_steps_per_second": 0.218,
+      "step": 119000
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 8.42989342804052e-05,
+      "loss": 0.722,
+      "step": 119200
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 8.4272558346862e-05,
+      "loss": 0.7232,
+      "step": 119400
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 8.424618241331879e-05,
+      "loss": 0.723,
+      "step": 119600
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 8.42198064797756e-05,
+      "loss": 0.7196,
+      "step": 119800
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 8.41934305462324e-05,
+      "loss": 0.7187,
+      "step": 120000
+    },
+    {
+      "epoch": 0.16,
+      "eval_em_first_line_ratio": 0.5005,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3872,
+      "eval_loss": 0.7123618721961975,
+      "eval_runtime": 244.8823,
+      "eval_samples_per_second": 40.836,
+      "eval_steps_per_second": 0.216,
+      "step": 120000
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 8.41670546126892e-05,
+      "loss": 0.7256,
+      "step": 120200
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 8.4140678679146e-05,
+      "loss": 0.7206,
+      "step": 120400
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 8.41143027456028e-05,
+      "loss": 0.7173,
+      "step": 120600
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 8.40879268120596e-05,
+      "loss": 0.7177,
+      "step": 120800
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 8.406155087851642e-05,
+      "loss": 0.7265,
+      "step": 121000
+    },
+    {
+      "epoch": 0.16,
+      "eval_em_first_line_ratio": 0.496,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3835,
+      "eval_loss": 0.7120243310928345,
+      "eval_runtime": 240.6891,
+      "eval_samples_per_second": 41.547,
+      "eval_steps_per_second": 0.22,
+      "step": 121000
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 8.403517494497321e-05,
+      "loss": 0.7196,
+      "step": 121200
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 8.400879901143002e-05,
+      "loss": 0.7127,
+      "step": 121400
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 8.398242307788682e-05,
+      "loss": 0.7191,
+      "step": 121600
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 8.395604714434363e-05,
+      "loss": 0.721,
+      "step": 121800
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 8.392967121080042e-05,
+      "loss": 0.7209,
+      "step": 122000
+    },
+    {
+      "epoch": 0.16,
+      "eval_em_first_line_ratio": 0.4964,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3834,
+      "eval_loss": 0.7130839824676514,
+      "eval_runtime": 249.9411,
+      "eval_samples_per_second": 40.009,
+      "eval_steps_per_second": 0.212,
+      "step": 122000
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 8.390329527725722e-05,
+      "loss": 0.7202,
+      "step": 122200
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 8.387691934371403e-05,
+      "loss": 0.7189,
+      "step": 122400
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 8.385054341017083e-05,
+      "loss": 0.7197,
+      "step": 122600
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 8.382416747662763e-05,
+      "loss": 0.7197,
+      "step": 122800
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 8.379779154308444e-05,
+      "loss": 0.7175,
+      "step": 123000
+    },
+    {
+      "epoch": 0.16,
+      "eval_em_first_line_ratio": 0.4974,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3842,
+      "eval_loss": 0.7118337154388428,
+      "eval_runtime": 240.2365,
+      "eval_samples_per_second": 41.626,
+      "eval_steps_per_second": 0.221,
+      "step": 123000
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 8.377141560954124e-05,
+      "loss": 0.7169,
+      "step": 123200
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 8.374503967599803e-05,
+      "loss": 0.7169,
+      "step": 123400
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 8.371879562212256e-05,
+      "loss": 0.7232,
+      "step": 123600
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 8.369241968857937e-05,
+      "loss": 0.7244,
+      "step": 123800
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 8.366604375503616e-05,
+      "loss": 0.7168,
+      "step": 124000
+    },
+    {
+      "epoch": 0.16,
+      "eval_em_first_line_ratio": 0.5006,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3869,
+      "eval_loss": 0.7112560868263245,
+      "eval_runtime": 230.4458,
+      "eval_samples_per_second": 43.394,
+      "eval_steps_per_second": 0.23,
+      "step": 124000
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 8.363966782149296e-05,
+      "loss": 0.723,
+      "step": 124200
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 8.361329188794976e-05,
+      "loss": 0.7154,
+      "step": 124400
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 8.358691595440657e-05,
+      "loss": 0.7173,
+      "step": 124600
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 8.356054002086336e-05,
+      "loss": 0.718,
+      "step": 124800
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 8.353416408732016e-05,
+      "loss": 0.7138,
+      "step": 125000
+    },
+    {
+      "epoch": 0.16,
+      "eval_em_first_line_ratio": 0.4962,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3825,
+      "eval_loss": 0.7125546932220459,
+      "eval_runtime": 232.7962,
+      "eval_samples_per_second": 42.956,
+      "eval_steps_per_second": 0.228,
+      "step": 125000
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 8.350778815377697e-05,
+      "loss": 0.7216,
+      "step": 125200
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 8.348141222023377e-05,
+      "loss": 0.7175,
+      "step": 125400
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 8.345503628669058e-05,
+      "loss": 0.7269,
+      "step": 125600
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 8.342879223281509e-05,
+      "loss": 0.7171,
+      "step": 125800
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 8.34024162992719e-05,
+      "loss": 0.7189,
+      "step": 126000
+    },
+    {
+      "epoch": 0.17,
+      "eval_em_first_line_ratio": 0.4963,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3832,
+      "eval_loss": 0.7101957201957703,
+      "eval_runtime": 235.737,
+      "eval_samples_per_second": 42.42,
+      "eval_steps_per_second": 0.225,
+      "step": 126000
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 8.33760403657287e-05,
+      "loss": 0.7197,
+      "step": 126200
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 8.33496644321855e-05,
+      "loss": 0.7213,
+      "step": 126400
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 8.332328849864231e-05,
+      "loss": 0.7175,
+      "step": 126600
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 8.329704444476682e-05,
+      "loss": 0.7141,
+      "step": 126800
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 8.327066851122363e-05,
+      "loss": 0.7154,
+      "step": 127000
+    },
+    {
+      "epoch": 0.17,
+      "eval_em_first_line_ratio": 0.4966,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3832,
+      "eval_loss": 0.7098414897918701,
+      "eval_runtime": 231.5607,
+      "eval_samples_per_second": 43.185,
+      "eval_steps_per_second": 0.229,
+      "step": 127000
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 8.324429257768043e-05,
+      "loss": 0.7168,
+      "step": 127200
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 8.321791664413722e-05,
+      "loss": 0.7204,
+      "step": 127400
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 8.319154071059403e-05,
+      "loss": 0.7199,
+      "step": 127600
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 8.316516477705083e-05,
+      "loss": 0.7146,
+      "step": 127800
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 8.313878884350763e-05,
+      "loss": 0.7191,
+      "step": 128000
+    },
+    {
+      "epoch": 0.17,
+      "eval_em_first_line_ratio": 0.4961,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3831,
+      "eval_loss": 0.7086435556411743,
+      "eval_runtime": 228.2302,
+      "eval_samples_per_second": 43.815,
+      "eval_steps_per_second": 0.232,
+      "step": 128000
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 8.311241290996443e-05,
+      "loss": 0.7192,
+      "step": 128200
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 8.308603697642123e-05,
+      "loss": 0.7138,
+      "step": 128400
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 8.305966104287805e-05,
+      "loss": 0.7144,
+      "step": 128600
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 8.303328510933485e-05,
+      "loss": 0.7158,
+      "step": 128800
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 8.300690917579164e-05,
+      "loss": 0.7109,
+      "step": 129000
+    },
+    {
+      "epoch": 0.17,
+      "eval_em_first_line_ratio": 0.4962,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3831,
+      "eval_loss": 0.7079706192016602,
+      "eval_runtime": 233.9355,
+      "eval_samples_per_second": 42.747,
+      "eval_steps_per_second": 0.227,
+      "step": 129000
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 8.298053324224845e-05,
+      "loss": 0.7128,
+      "step": 129200
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 8.295415730870525e-05,
+      "loss": 0.715,
+      "step": 129400
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 8.292778137516206e-05,
+      "loss": 0.7181,
+      "step": 129600
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 8.290140544161885e-05,
+      "loss": 0.7173,
+      "step": 129800
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 8.287516138774337e-05,
+      "loss": 0.7152,
+      "step": 130000
+    },
+    {
+      "epoch": 0.17,
+      "eval_em_first_line_ratio": 0.5,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3864,
+      "eval_loss": 0.7089020609855652,
+      "eval_runtime": 236.7923,
+      "eval_samples_per_second": 42.231,
+      "eval_steps_per_second": 0.224,
+      "step": 130000
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 8.284878545420018e-05,
+      "loss": 0.712,
+      "step": 130200
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 8.282240952065697e-05,
+      "loss": 0.7184,
+      "step": 130400
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 8.279603358711377e-05,
+      "loss": 0.7146,
+      "step": 130600
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 8.276965765357058e-05,
+      "loss": 0.7165,
+      "step": 130800
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 8.274328172002738e-05,
+      "loss": 0.7153,
+      "step": 131000
+    },
+    {
+      "epoch": 0.17,
+      "eval_em_first_line_ratio": 0.5004,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3866,
+      "eval_loss": 0.7078763246536255,
+      "eval_runtime": 230.8956,
+      "eval_samples_per_second": 43.31,
+      "eval_steps_per_second": 0.23,
+      "step": 131000
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 8.271690578648419e-05,
+      "loss": 0.7152,
+      "step": 131200
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 8.269052985294099e-05,
+      "loss": 0.7196,
+      "step": 131400
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 8.26641539193978e-05,
+      "loss": 0.7144,
+      "step": 131600
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 8.263777798585459e-05,
+      "loss": 0.7116,
+      "step": 131800
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 8.261140205231139e-05,
+      "loss": 0.7185,
+      "step": 132000
+    },
+    {
+      "epoch": 0.17,
+      "eval_em_first_line_ratio": 0.4978,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3851,
+      "eval_loss": 0.708818256855011,
+      "eval_runtime": 238.9376,
+      "eval_samples_per_second": 41.852,
+      "eval_steps_per_second": 0.222,
+      "step": 132000
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 8.25850261187682e-05,
+      "loss": 0.7134,
+      "step": 132200
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 8.2558650185225e-05,
+      "loss": 0.7154,
+      "step": 132400
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 8.253227425168179e-05,
+      "loss": 0.7159,
+      "step": 132600
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 8.25058983181386e-05,
+      "loss": 0.7135,
+      "step": 132800
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 8.247952238459541e-05,
+      "loss": 0.7191,
+      "step": 133000
+    },
+    {
+      "epoch": 0.18,
+      "eval_em_first_line_ratio": 0.4996,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.386,
+      "eval_loss": 0.7074763774871826,
+      "eval_runtime": 233.7892,
+      "eval_samples_per_second": 42.774,
+      "eval_steps_per_second": 0.227,
+      "step": 133000
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 8.24531464510522e-05,
+      "loss": 0.7201,
+      "step": 133200
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 8.242677051750901e-05,
+      "loss": 0.7099,
+      "step": 133400
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 8.240039458396581e-05,
+      "loss": 0.7196,
+      "step": 133600
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 8.237401865042262e-05,
+      "loss": 0.7132,
+      "step": 133800
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 8.234764271687941e-05,
+      "loss": 0.7145,
+      "step": 134000
+    },
+    {
+      "epoch": 0.18,
+      "eval_em_first_line_ratio": 0.5022,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3882,
+      "eval_loss": 0.7070466876029968,
+      "eval_runtime": 234.2098,
+      "eval_samples_per_second": 42.697,
+      "eval_steps_per_second": 0.226,
+      "step": 134000
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 8.232126678333621e-05,
+      "loss": 0.7135,
+      "step": 134200
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 8.229489084979302e-05,
+      "loss": 0.7111,
+      "step": 134400
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 8.226851491624982e-05,
+      "loss": 0.7147,
+      "step": 134600
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 8.224213898270661e-05,
+      "loss": 0.7161,
+      "step": 134800
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 8.221589492883114e-05,
+      "loss": 0.7129,
+      "step": 135000
+    },
+    {
+      "epoch": 0.18,
+      "eval_em_first_line_ratio": 0.4997,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3862,
+      "eval_loss": 0.7070465683937073,
+      "eval_runtime": 236.4586,
+      "eval_samples_per_second": 42.291,
+      "eval_steps_per_second": 0.224,
+      "step": 135000
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 8.218951899528794e-05,
+      "loss": 0.7117,
+      "step": 135200
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 8.216314306174473e-05,
+      "loss": 0.7144,
+      "step": 135400
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 8.213676712820155e-05,
+      "loss": 0.7132,
+      "step": 135600
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 8.211039119465836e-05,
+      "loss": 0.7156,
+      "step": 135800
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 8.208401526111515e-05,
+      "loss": 0.7129,
+      "step": 136000
+    },
+    {
+      "epoch": 0.18,
+      "eval_em_first_line_ratio": 0.4991,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3858,
+      "eval_loss": 0.7068952918052673,
+      "eval_runtime": 233.4532,
+      "eval_samples_per_second": 42.835,
+      "eval_steps_per_second": 0.227,
+      "step": 136000
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 8.205777120723967e-05,
+      "loss": 0.7112,
+      "step": 136200
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 8.203139527369648e-05,
+      "loss": 0.7097,
+      "step": 136400
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 8.200501934015328e-05,
+      "loss": 0.7125,
+      "step": 136600
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 8.197864340661007e-05,
+      "loss": 0.7135,
+      "step": 136800
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 8.195226747306688e-05,
+      "loss": 0.7129,
+      "step": 137000
+    },
+    {
+      "epoch": 0.18,
+      "eval_em_first_line_ratio": 0.5002,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3873,
+      "eval_loss": 0.706534743309021,
+      "eval_runtime": 231.5389,
+      "eval_samples_per_second": 43.189,
+      "eval_steps_per_second": 0.229,
+      "step": 137000
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 8.192589153952368e-05,
+      "loss": 0.7185,
+      "step": 137200
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 8.189951560598049e-05,
+      "loss": 0.717,
+      "step": 137400
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 8.187313967243728e-05,
+      "loss": 0.7156,
+      "step": 137600
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 8.184676373889408e-05,
+      "loss": 0.7153,
+      "step": 137800
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 8.182038780535089e-05,
+      "loss": 0.7113,
+      "step": 138000
+    },
+    {
+      "epoch": 0.18,
+      "eval_em_first_line_ratio": 0.5018,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3883,
+      "eval_loss": 0.7059966325759888,
+      "eval_runtime": 231.133,
+      "eval_samples_per_second": 43.265,
+      "eval_steps_per_second": 0.229,
+      "step": 138000
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 8.179401187180769e-05,
+      "loss": 0.7077,
+      "step": 138200
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 8.17676359382645e-05,
+      "loss": 0.712,
+      "step": 138400
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 8.17412600047213e-05,
+      "loss": 0.7113,
+      "step": 138600
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 8.17148840711781e-05,
+      "loss": 0.7075,
+      "step": 138800
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 8.16885081376349e-05,
+      "loss": 0.713,
+      "step": 139000
+    },
+    {
+      "epoch": 0.18,
+      "eval_em_first_line_ratio": 0.5031,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3898,
+      "eval_loss": 0.7059535980224609,
+      "eval_runtime": 234.1355,
+      "eval_samples_per_second": 42.71,
+      "eval_steps_per_second": 0.226,
+      "step": 139000
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 8.16621322040917e-05,
+      "loss": 0.716,
+      "step": 139200
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 8.16357562705485e-05,
+      "loss": 0.7109,
+      "step": 139400
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 8.160938033700531e-05,
+      "loss": 0.7106,
+      "step": 139600
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 8.15830044034621e-05,
+      "loss": 0.7132,
+      "step": 139800
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 8.155662846991892e-05,
+      "loss": 0.709,
+      "step": 140000
+    },
+    {
+      "epoch": 0.18,
+      "eval_em_first_line_ratio": 0.4998,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3869,
+      "eval_loss": 0.7049171924591064,
+      "eval_runtime": 228.2164,
+      "eval_samples_per_second": 43.818,
+      "eval_steps_per_second": 0.232,
+      "step": 140000
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 8.153025253637572e-05,
+      "loss": 0.7117,
+      "step": 140200
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 8.150387660283251e-05,
+      "loss": 0.7094,
+      "step": 140400
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 8.147750066928932e-05,
+      "loss": 0.7079,
+      "step": 140600
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 8.145112473574612e-05,
+      "loss": 0.7143,
+      "step": 140800
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 8.142474880220293e-05,
+      "loss": 0.7092,
+      "step": 141000
+    },
+    {
+      "epoch": 0.19,
+      "eval_em_first_line_ratio": 0.4993,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3858,
+      "eval_loss": 0.7038989067077637,
+      "eval_runtime": 232.5269,
+      "eval_samples_per_second": 43.006,
+      "eval_steps_per_second": 0.228,
+      "step": 141000
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 8.139837286865972e-05,
+      "loss": 0.7083,
+      "step": 141200
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 8.137199693511652e-05,
+      "loss": 0.7108,
+      "step": 141400
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 8.134562100157333e-05,
+      "loss": 0.7121,
+      "step": 141600
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 8.131937694769784e-05,
+      "loss": 0.7109,
+      "step": 141800
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 8.129313289382237e-05,
+      "loss": 0.7123,
+      "step": 142000
+    },
+    {
+      "epoch": 0.19,
+      "eval_em_first_line_ratio": 0.5052,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3917,
+      "eval_loss": 0.7040054202079773,
+      "eval_runtime": 226.983,
+      "eval_samples_per_second": 44.056,
+      "eval_steps_per_second": 0.233,
+      "step": 142000
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 8.126675696027917e-05,
+      "loss": 0.7044,
+      "step": 142200
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 8.124038102673596e-05,
+      "loss": 0.7069,
+      "step": 142400
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 8.121400509319277e-05,
+      "loss": 0.7122,
+      "step": 142600
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 8.118762915964957e-05,
+      "loss": 0.709,
+      "step": 142800
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 8.116125322610637e-05,
+      "loss": 0.7107,
+      "step": 143000
+    },
+    {
+      "epoch": 0.19,
+      "eval_em_first_line_ratio": 0.5022,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3889,
+      "eval_loss": 0.7032320499420166,
+      "eval_runtime": 231.8452,
+      "eval_samples_per_second": 43.132,
+      "eval_steps_per_second": 0.229,
+      "step": 143000
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 8.113487729256318e-05,
+      "loss": 0.7108,
+      "step": 143200
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 8.110850135901998e-05,
+      "loss": 0.7073,
+      "step": 143400
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 8.108212542547679e-05,
+      "loss": 0.7133,
+      "step": 143600
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 8.105574949193358e-05,
+      "loss": 0.7102,
+      "step": 143800
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 8.102937355839038e-05,
+      "loss": 0.7041,
+      "step": 144000
+    },
+    {
+      "epoch": 0.19,
+      "eval_em_first_line_ratio": 0.5016,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3883,
+      "eval_loss": 0.7035794854164124,
+      "eval_runtime": 232.8718,
+      "eval_samples_per_second": 42.942,
+      "eval_steps_per_second": 0.228,
+      "step": 144000
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 8.100299762484719e-05,
+      "loss": 0.7064,
+      "step": 144200
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 8.097662169130399e-05,
+      "loss": 0.711,
+      "step": 144400
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 8.095024575776078e-05,
+      "loss": 0.7078,
+      "step": 144600
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 8.092386982421759e-05,
+      "loss": 0.7117,
+      "step": 144800
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 8.089749389067439e-05,
+      "loss": 0.7076,
+      "step": 145000
+    },
+    {
+      "epoch": 0.19,
+      "eval_em_first_line_ratio": 0.5035,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3902,
+      "eval_loss": 0.7037900686264038,
+      "eval_runtime": 227.2206,
+      "eval_samples_per_second": 44.01,
+      "eval_steps_per_second": 0.233,
+      "step": 145000
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 8.087111795713121e-05,
+      "loss": 0.71,
+      "step": 145200
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 8.0844742023588e-05,
+      "loss": 0.7158,
+      "step": 145400
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 8.081849796971251e-05,
+      "loss": 0.7049,
+      "step": 145600
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 8.079212203616933e-05,
+      "loss": 0.7078,
+      "step": 145800
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 8.076574610262612e-05,
+      "loss": 0.7101,
+      "step": 146000
+    },
+    {
+      "epoch": 0.19,
+      "eval_em_first_line_ratio": 0.5027,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3894,
+      "eval_loss": 0.7034583687782288,
+      "eval_runtime": 231.9289,
+      "eval_samples_per_second": 43.117,
+      "eval_steps_per_second": 0.229,
+      "step": 146000
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 8.073937016908293e-05,
+      "loss": 0.7083,
+      "step": 146200
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 8.071299423553973e-05,
+      "loss": 0.7042,
+      "step": 146400
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 8.068661830199654e-05,
+      "loss": 0.708,
+      "step": 146600
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 8.066024236845333e-05,
+      "loss": 0.7099,
+      "step": 146800
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 8.063386643491013e-05,
+      "loss": 0.7109,
+      "step": 147000
+    },
+    {
+      "epoch": 0.19,
+      "eval_em_first_line_ratio": 0.5016,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3887,
+      "eval_loss": 0.7034907341003418,
+      "eval_runtime": 223.7008,
+      "eval_samples_per_second": 44.703,
+      "eval_steps_per_second": 0.237,
+      "step": 147000
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 8.060749050136694e-05,
+      "loss": 0.7071,
+      "step": 147200
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 8.058111456782374e-05,
+      "loss": 0.7067,
+      "step": 147400
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 8.055473863428053e-05,
+      "loss": 0.7075,
+      "step": 147600
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 8.052836270073735e-05,
+      "loss": 0.7027,
+      "step": 147800
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 8.050211864686186e-05,
+      "loss": 0.7094,
+      "step": 148000
+    },
+    {
+      "epoch": 0.2,
+      "eval_em_first_line_ratio": 0.5007,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3876,
+      "eval_loss": 0.7013224363327026,
+      "eval_runtime": 232.0723,
+      "eval_samples_per_second": 43.09,
+      "eval_steps_per_second": 0.228,
+      "step": 148000
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 8.047574271331865e-05,
+      "loss": 0.6977,
+      "step": 148200
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 8.044936677977547e-05,
+      "loss": 0.7058,
+      "step": 148400
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 8.042299084623228e-05,
+      "loss": 0.7104,
+      "step": 148600
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 8.039661491268907e-05,
+      "loss": 0.71,
+      "step": 148800
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 8.037023897914587e-05,
+      "loss": 0.7069,
+      "step": 149000
+    },
+    {
+      "epoch": 0.2,
+      "eval_em_first_line_ratio": 0.4972,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.385,
+      "eval_loss": 0.70399409532547,
+      "eval_runtime": 234.9365,
+      "eval_samples_per_second": 42.565,
+      "eval_steps_per_second": 0.226,
+      "step": 149000
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 8.034386304560268e-05,
+      "loss": 0.7064,
+      "step": 149200
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 8.031748711205948e-05,
+      "loss": 0.7099,
+      "step": 149400
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 8.029111117851627e-05,
+      "loss": 0.7076,
+      "step": 149600
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 8.026473524497308e-05,
+      "loss": 0.7104,
+      "step": 149800
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 8.023835931142988e-05,
+      "loss": 0.7118,
+      "step": 150000
+    },
+    {
+      "epoch": 0.2,
+      "eval_em_first_line_ratio": 0.4975,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3843,
+      "eval_loss": 0.7021588683128357,
+      "eval_runtime": 235.2843,
+      "eval_samples_per_second": 42.502,
+      "eval_steps_per_second": 0.225,
+      "step": 150000
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 8.021198337788668e-05,
+      "loss": 0.7066,
+      "step": 150200
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 8.018560744434349e-05,
+      "loss": 0.7103,
+      "step": 150400
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 8.015923151080029e-05,
+      "loss": 0.7048,
+      "step": 150600
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 8.01328555772571e-05,
+      "loss": 0.7084,
+      "step": 150800
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 8.010647964371389e-05,
+      "loss": 0.7039,
+      "step": 151000
+    },
+    {
+      "epoch": 0.2,
+      "eval_em_first_line_ratio": 0.5023,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3889,
+      "eval_loss": 0.7010955810546875,
+      "eval_runtime": 230.5877,
+      "eval_samples_per_second": 43.367,
+      "eval_steps_per_second": 0.23,
+      "step": 151000
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 8.008010371017069e-05,
+      "loss": 0.6992,
+      "step": 151200
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 8.00537277766275e-05,
+      "loss": 0.7021,
+      "step": 151400
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 8.00273518430843e-05,
+      "loss": 0.7036,
+      "step": 151600
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 8.000097590954109e-05,
+      "loss": 0.7078,
+      "step": 151800
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 7.99745999759979e-05,
+      "loss": 0.7037,
+      "step": 152000
+    },
+    {
+      "epoch": 0.2,
+      "eval_em_first_line_ratio": 0.5011,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.388,
+      "eval_loss": 0.7010334730148315,
+      "eval_runtime": 228.1832,
+      "eval_samples_per_second": 43.824,
+      "eval_steps_per_second": 0.232,
+      "step": 152000
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 7.994822404245471e-05,
+      "loss": 0.7057,
+      "step": 152200
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 7.99218481089115e-05,
+      "loss": 0.7049,
+      "step": 152400
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 7.989547217536831e-05,
+      "loss": 0.7072,
+      "step": 152600
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 7.986909624182511e-05,
+      "loss": 0.7074,
+      "step": 152800
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 7.984272030828192e-05,
+      "loss": 0.7001,
+      "step": 153000
+    },
+    {
+      "epoch": 0.2,
+      "eval_em_first_line_ratio": 0.5014,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3879,
+      "eval_loss": 0.700883686542511,
+      "eval_runtime": 233.2631,
+      "eval_samples_per_second": 42.87,
+      "eval_steps_per_second": 0.227,
+      "step": 153000
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 7.981634437473871e-05,
+      "loss": 0.7067,
+      "step": 153200
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 7.978996844119551e-05,
+      "loss": 0.7023,
+      "step": 153400
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 7.976359250765232e-05,
+      "loss": 0.7036,
+      "step": 153600
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 7.973721657410912e-05,
+      "loss": 0.7066,
+      "step": 153800
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 7.971097252023364e-05,
+      "loss": 0.7051,
+      "step": 154000
+    },
+    {
+      "epoch": 0.2,
+      "eval_em_first_line_ratio": 0.5002,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3868,
+      "eval_loss": 0.699952244758606,
+      "eval_runtime": 225.2464,
+      "eval_samples_per_second": 44.396,
+      "eval_steps_per_second": 0.235,
+      "step": 154000
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 7.968459658669044e-05,
+      "loss": 0.707,
+      "step": 154200
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 7.965822065314725e-05,
+      "loss": 0.6979,
+      "step": 154400
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 7.963184471960405e-05,
+      "loss": 0.7045,
+      "step": 154600
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 7.960546878606085e-05,
+      "loss": 0.708,
+      "step": 154800
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 7.957909285251766e-05,
+      "loss": 0.7051,
+      "step": 155000
+    },
+    {
+      "epoch": 0.2,
+      "eval_em_first_line_ratio": 0.5012,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.388,
+      "eval_loss": 0.69964599609375,
+      "eval_runtime": 233.6453,
+      "eval_samples_per_second": 42.8,
+      "eval_steps_per_second": 0.227,
+      "step": 155000
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 7.955271691897446e-05,
+      "loss": 0.7023,
+      "step": 155200
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 7.952634098543125e-05,
+      "loss": 0.7055,
+      "step": 155400
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 7.949996505188806e-05,
+      "loss": 0.7065,
+      "step": 155600
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 7.947358911834486e-05,
+      "loss": 0.704,
+      "step": 155800
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 7.944721318480167e-05,
+      "loss": 0.7023,
+      "step": 156000
+    },
+    {
+      "epoch": 0.21,
+      "eval_em_first_line_ratio": 0.5026,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3892,
+      "eval_loss": 0.699143648147583,
+      "eval_runtime": 232.8123,
+      "eval_samples_per_second": 42.953,
+      "eval_steps_per_second": 0.228,
+      "step": 156000
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 7.942083725125846e-05,
+      "loss": 0.7088,
+      "step": 156200
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 7.939446131771526e-05,
+      "loss": 0.7102,
+      "step": 156400
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 7.936808538417208e-05,
+      "loss": 0.7052,
+      "step": 156600
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 7.934170945062887e-05,
+      "loss": 0.7054,
+      "step": 156800
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 7.931533351708568e-05,
+      "loss": 0.709,
+      "step": 157000
+    },
+    {
+      "epoch": 0.21,
+      "eval_em_first_line_ratio": 0.5008,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.388,
+      "eval_loss": 0.6988568902015686,
+      "eval_runtime": 231.759,
+      "eval_samples_per_second": 43.148,
+      "eval_steps_per_second": 0.229,
+      "step": 157000
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 7.928895758354248e-05,
+      "loss": 0.7023,
+      "step": 157200
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 7.926271352966699e-05,
+      "loss": 0.7004,
+      "step": 157400
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 7.92363375961238e-05,
+      "loss": 0.7017,
+      "step": 157600
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 7.92099616625806e-05,
+      "loss": 0.7026,
+      "step": 157800
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 7.91835857290374e-05,
+      "loss": 0.7033,
+      "step": 158000
+    },
+    {
+      "epoch": 0.21,
+      "eval_em_first_line_ratio": 0.5001,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3873,
+      "eval_loss": 0.6988629698753357,
+      "eval_runtime": 232.5442,
+      "eval_samples_per_second": 43.003,
+      "eval_steps_per_second": 0.228,
+      "step": 158000
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 7.91572097954942e-05,
+      "loss": 0.7012,
+      "step": 158200
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 7.9130833861951e-05,
+      "loss": 0.6995,
+      "step": 158400
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 7.91044579284078e-05,
+      "loss": 0.7067,
+      "step": 158600
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 7.907808199486461e-05,
+      "loss": 0.701,
+      "step": 158800
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 7.90517060613214e-05,
+      "loss": 0.7004,
+      "step": 159000
+    },
+    {
+      "epoch": 0.21,
+      "eval_em_first_line_ratio": 0.505,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.392,
+      "eval_loss": 0.6979176998138428,
+      "eval_runtime": 237.0989,
+      "eval_samples_per_second": 42.176,
+      "eval_steps_per_second": 0.224,
+      "step": 159000
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 7.902533012777822e-05,
+      "loss": 0.7025,
+      "step": 159200
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 7.899895419423502e-05,
+      "loss": 0.6996,
+      "step": 159400
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 7.897257826069181e-05,
+      "loss": 0.7064,
+      "step": 159600
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 7.894620232714862e-05,
+      "loss": 0.7012,
+      "step": 159800
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 7.891982639360542e-05,
+      "loss": 0.7017,
+      "step": 160000
+    },
+    {
+      "epoch": 0.21,
+      "eval_em_first_line_ratio": 0.5032,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3901,
+      "eval_loss": 0.6980948448181152,
+      "eval_runtime": 225.8709,
+      "eval_samples_per_second": 44.273,
+      "eval_steps_per_second": 0.235,
+      "step": 160000
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 7.889358233972994e-05,
+      "loss": 0.7064,
+      "step": 160200
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 7.886720640618674e-05,
+      "loss": 0.7014,
+      "step": 160400
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 7.884083047264355e-05,
+      "loss": 0.7006,
+      "step": 160600
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 7.881445453910035e-05,
+      "loss": 0.7027,
+      "step": 160800
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 7.878807860555714e-05,
+      "loss": 0.703,
+      "step": 161000
+    },
+    {
+      "epoch": 0.21,
+      "eval_em_first_line_ratio": 0.5043,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3914,
+      "eval_loss": 0.6985665559768677,
+      "eval_runtime": 232.319,
+      "eval_samples_per_second": 43.044,
+      "eval_steps_per_second": 0.228,
+      "step": 161000
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 7.876170267201395e-05,
+      "loss": 0.7066,
+      "step": 161200
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 7.873532673847075e-05,
+      "loss": 0.6993,
+      "step": 161400
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 7.870895080492755e-05,
+      "loss": 0.7016,
+      "step": 161600
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 7.868257487138436e-05,
+      "loss": 0.7038,
+      "step": 161800
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 7.865619893784116e-05,
+      "loss": 0.6988,
+      "step": 162000
+    },
+    {
+      "epoch": 0.21,
+      "eval_em_first_line_ratio": 0.5055,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3922,
+      "eval_loss": 0.6972260475158691,
+      "eval_runtime": 227.5379,
+      "eval_samples_per_second": 43.949,
+      "eval_steps_per_second": 0.233,
+      "step": 162000
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 7.862982300429797e-05,
+      "loss": 0.701,
+      "step": 162200
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 7.860344707075477e-05,
+      "loss": 0.7008,
+      "step": 162400
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 7.857720301687928e-05,
+      "loss": 0.7029,
+      "step": 162600
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 7.855082708333609e-05,
+      "loss": 0.6948,
+      "step": 162800
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 7.85244511497929e-05,
+      "loss": 0.7008,
+      "step": 163000
+    },
+    {
+      "epoch": 0.21,
+      "eval_em_first_line_ratio": 0.5031,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3902,
+      "eval_loss": 0.6980459690093994,
+      "eval_runtime": 231.3662,
+      "eval_samples_per_second": 43.222,
+      "eval_steps_per_second": 0.229,
+      "step": 163000
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 7.849807521624968e-05,
+      "loss": 0.7044,
+      "step": 163200
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 7.847169928270649e-05,
+      "loss": 0.7002,
+      "step": 163400
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 7.84453233491633e-05,
+      "loss": 0.6992,
+      "step": 163600
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 7.84189474156201e-05,
+      "loss": 0.7032,
+      "step": 163800
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 7.839257148207689e-05,
+      "loss": 0.7033,
+      "step": 164000
+    },
+    {
+      "epoch": 0.22,
+      "eval_em_first_line_ratio": 0.4997,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3865,
+      "eval_loss": 0.6971564292907715,
+      "eval_runtime": 230.1828,
+      "eval_samples_per_second": 43.444,
+      "eval_steps_per_second": 0.23,
+      "step": 164000
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 7.83661955485337e-05,
+      "loss": 0.6984,
+      "step": 164200
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 7.833981961499051e-05,
+      "loss": 0.7007,
+      "step": 164400
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 7.83134436814473e-05,
+      "loss": 0.7061,
+      "step": 164600
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 7.828719962757182e-05,
+      "loss": 0.7014,
+      "step": 164800
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 7.826082369402863e-05,
+      "loss": 0.7019,
+      "step": 165000
+    },
+    {
+      "epoch": 0.22,
+      "eval_em_first_line_ratio": 0.5018,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3883,
+      "eval_loss": 0.6965271234512329,
+      "eval_runtime": 232.9682,
+      "eval_samples_per_second": 42.924,
+      "eval_steps_per_second": 0.227,
+      "step": 165000
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 7.823444776048542e-05,
+      "loss": 0.6982,
+      "step": 165200
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 7.820807182694223e-05,
+      "loss": 0.6994,
+      "step": 165400
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 7.818169589339903e-05,
+      "loss": 0.6988,
+      "step": 165600
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 7.815531995985584e-05,
+      "loss": 0.6958,
+      "step": 165800
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 7.812894402631263e-05,
+      "loss": 0.7023,
+      "step": 166000
+    },
+    {
+      "epoch": 0.22,
+      "eval_em_first_line_ratio": 0.5015,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3876,
+      "eval_loss": 0.6968411207199097,
+      "eval_runtime": 237.6796,
+      "eval_samples_per_second": 42.073,
+      "eval_steps_per_second": 0.223,
+      "step": 166000
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 7.810256809276943e-05,
+      "loss": 0.7022,
+      "step": 166200
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 7.807619215922624e-05,
+      "loss": 0.7029,
+      "step": 166400
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 7.804981622568304e-05,
+      "loss": 0.6954,
+      "step": 166600
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 7.802344029213985e-05,
+      "loss": 0.7025,
+      "step": 166800
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 7.799719623826436e-05,
+      "loss": 0.6995,
+      "step": 167000
+    },
+    {
+      "epoch": 0.22,
+      "eval_em_first_line_ratio": 0.5019,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3888,
+      "eval_loss": 0.6961172223091125,
+      "eval_runtime": 236.1294,
+      "eval_samples_per_second": 42.35,
+      "eval_steps_per_second": 0.224,
+      "step": 167000
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 7.797082030472116e-05,
+      "loss": 0.7029,
+      "step": 167200
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 7.794444437117797e-05,
+      "loss": 0.6982,
+      "step": 167400
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 7.791806843763477e-05,
+      "loss": 0.6973,
+      "step": 167600
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 7.789169250409158e-05,
+      "loss": 0.6982,
+      "step": 167800
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 7.786531657054837e-05,
+      "loss": 0.7,
+      "step": 168000
+    },
+    {
+      "epoch": 0.22,
+      "eval_em_first_line_ratio": 0.5022,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.389,
+      "eval_loss": 0.6953169703483582,
+      "eval_runtime": 230.6932,
+      "eval_samples_per_second": 43.348,
+      "eval_steps_per_second": 0.23,
+      "step": 168000
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 7.783894063700517e-05,
+      "loss": 0.7001,
+      "step": 168200
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 7.781256470346198e-05,
+      "loss": 0.6983,
+      "step": 168400
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 7.778618876991878e-05,
+      "loss": 0.6943,
+      "step": 168600
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 7.775981283637557e-05,
+      "loss": 0.6948,
+      "step": 168800
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 7.773343690283238e-05,
+      "loss": 0.7004,
+      "step": 169000
+    },
+    {
+      "epoch": 0.22,
+      "eval_em_first_line_ratio": 0.5029,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3899,
+      "eval_loss": 0.6944438219070435,
+      "eval_runtime": 231.5738,
+      "eval_samples_per_second": 43.183,
+      "eval_steps_per_second": 0.229,
+      "step": 169000
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 7.770706096928918e-05,
+      "loss": 0.701,
+      "step": 169200
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 7.7680685035746e-05,
+      "loss": 0.7004,
+      "step": 169400
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 7.76544409818705e-05,
+      "loss": 0.6954,
+      "step": 169600
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 7.76280650483273e-05,
+      "loss": 0.6982,
+      "step": 169800
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 7.760168911478412e-05,
+      "loss": 0.6986,
+      "step": 170000
+    },
+    {
+      "epoch": 0.22,
+      "eval_em_first_line_ratio": 0.5011,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3884,
+      "eval_loss": 0.6947265863418579,
+      "eval_runtime": 230.0486,
+      "eval_samples_per_second": 43.469,
+      "eval_steps_per_second": 0.23,
+      "step": 170000
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 7.757531318124091e-05,
+      "loss": 0.6995,
+      "step": 170200
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 7.754893724769772e-05,
+      "loss": 0.6973,
+      "step": 170400
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 7.752256131415452e-05,
+      "loss": 0.6964,
+      "step": 170600
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 7.749618538061132e-05,
+      "loss": 0.6977,
+      "step": 170800
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 7.746980944706812e-05,
+      "loss": 0.6993,
+      "step": 171000
+    },
+    {
+      "epoch": 0.23,
+      "eval_em_first_line_ratio": 0.5031,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3899,
+      "eval_loss": 0.6949790120124817,
+      "eval_runtime": 234.8174,
+      "eval_samples_per_second": 42.586,
+      "eval_steps_per_second": 0.226,
+      "step": 171000
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 7.744343351352492e-05,
+      "loss": 0.6964,
+      "step": 171200
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 7.741705757998172e-05,
+      "loss": 0.6968,
+      "step": 171400
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 7.739068164643853e-05,
+      "loss": 0.6996,
+      "step": 171600
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 7.736430571289532e-05,
+      "loss": 0.6947,
+      "step": 171800
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 7.733792977935214e-05,
+      "loss": 0.6999,
+      "step": 172000
+    },
+    {
+      "epoch": 0.23,
+      "eval_em_first_line_ratio": 0.5028,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3901,
+      "eval_loss": 0.6935190558433533,
+      "eval_runtime": 230.5251,
+      "eval_samples_per_second": 43.379,
+      "eval_steps_per_second": 0.23,
+      "step": 172000
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 7.731155384580894e-05,
+      "loss": 0.6993,
+      "step": 172200
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 7.728517791226573e-05,
+      "loss": 0.7008,
+      "step": 172400
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 7.725893385839026e-05,
+      "loss": 0.6998,
+      "step": 172600
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 7.723255792484706e-05,
+      "loss": 0.7011,
+      "step": 172800
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 7.720618199130386e-05,
+      "loss": 0.6993,
+      "step": 173000
+    },
+    {
+      "epoch": 0.23,
+      "eval_em_first_line_ratio": 0.5019,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.389,
+      "eval_loss": 0.6951680779457092,
+      "eval_runtime": 222.8993,
+      "eval_samples_per_second": 44.863,
+      "eval_steps_per_second": 0.238,
+      "step": 173000
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 7.717980605776066e-05,
+      "loss": 0.6997,
+      "step": 173200
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 7.715343012421746e-05,
+      "loss": 0.7005,
+      "step": 173400
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 7.712705419067427e-05,
+      "loss": 0.6911,
+      "step": 173600
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 7.710067825713106e-05,
+      "loss": 0.692,
+      "step": 173800
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 7.707430232358786e-05,
+      "loss": 0.6966,
+      "step": 174000
+    },
+    {
+      "epoch": 0.23,
+      "eval_em_first_line_ratio": 0.5012,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3883,
+      "eval_loss": 0.6939478516578674,
+      "eval_runtime": 227.5557,
+      "eval_samples_per_second": 43.945,
+      "eval_steps_per_second": 0.233,
+      "step": 174000
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 7.704792639004467e-05,
+      "loss": 0.697,
+      "step": 174200
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 7.702155045650147e-05,
+      "loss": 0.7004,
+      "step": 174400
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 7.699517452295828e-05,
+      "loss": 0.7014,
+      "step": 174600
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 7.696879858941508e-05,
+      "loss": 0.6973,
+      "step": 174800
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 7.69425545355396e-05,
+      "loss": 0.6927,
+      "step": 175000
+    },
+    {
+      "epoch": 0.23,
+      "eval_em_first_line_ratio": 0.5,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3874,
+      "eval_loss": 0.693142294883728,
+      "eval_runtime": 233.6114,
+      "eval_samples_per_second": 42.806,
+      "eval_steps_per_second": 0.227,
+      "step": 175000
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 7.69161786019964e-05,
+      "loss": 0.6971,
+      "step": 175200
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 7.68898026684532e-05,
+      "loss": 0.6983,
+      "step": 175400
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 7.686355861457772e-05,
+      "loss": 0.6947,
+      "step": 175600
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 7.683718268103452e-05,
+      "loss": 0.6961,
+      "step": 175800
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 7.681080674749133e-05,
+      "loss": 0.6965,
+      "step": 176000
+    },
+    {
+      "epoch": 0.23,
+      "eval_em_first_line_ratio": 0.5034,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3902,
+      "eval_loss": 0.6924988627433777,
+      "eval_runtime": 228.8086,
+      "eval_samples_per_second": 43.705,
+      "eval_steps_per_second": 0.232,
+      "step": 176000
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 7.678443081394813e-05,
+      "loss": 0.6968,
+      "step": 176200
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 7.675805488040493e-05,
+      "loss": 0.6971,
+      "step": 176400
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 7.673167894686173e-05,
+      "loss": 0.6957,
+      "step": 176600
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 7.670530301331853e-05,
+      "loss": 0.6933,
+      "step": 176800
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 7.667892707977533e-05,
+      "loss": 0.6911,
+      "step": 177000
+    },
+    {
+      "epoch": 0.23,
+      "eval_em_first_line_ratio": 0.5026,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.39,
+      "eval_loss": 0.6927745342254639,
+      "eval_runtime": 231.6135,
+      "eval_samples_per_second": 43.175,
+      "eval_steps_per_second": 0.229,
+      "step": 177000
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 7.665255114623214e-05,
+      "loss": 0.6936,
+      "step": 177200
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 7.662617521268893e-05,
+      "loss": 0.6969,
+      "step": 177400
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 7.659979927914575e-05,
+      "loss": 0.6901,
+      "step": 177600
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 7.657342334560255e-05,
+      "loss": 0.694,
+      "step": 177800
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 7.654704741205934e-05,
+      "loss": 0.695,
+      "step": 178000
+    },
+    {
+      "epoch": 0.23,
+      "eval_em_first_line_ratio": 0.5044,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3919,
+      "eval_loss": 0.6929817199707031,
+      "eval_runtime": 234.4459,
+      "eval_samples_per_second": 42.654,
+      "eval_steps_per_second": 0.226,
+      "step": 178000
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 7.652067147851615e-05,
+      "loss": 0.6996,
+      "step": 178200
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 7.649429554497295e-05,
+      "loss": 0.6981,
+      "step": 178400
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 7.646791961142976e-05,
+      "loss": 0.6942,
+      "step": 178600
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 7.644167555755427e-05,
+      "loss": 0.6924,
+      "step": 178800
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 7.64154315036788e-05,
+      "loss": 0.695,
+      "step": 179000
+    },
+    {
+      "epoch": 0.24,
+      "eval_em_first_line_ratio": 0.5025,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3896,
+      "eval_loss": 0.6926127076148987,
+      "eval_runtime": 229.108,
+      "eval_samples_per_second": 43.648,
+      "eval_steps_per_second": 0.231,
+      "step": 179000
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 7.638905557013559e-05,
+      "loss": 0.6953,
+      "step": 179200
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 7.636267963659239e-05,
+      "loss": 0.693,
+      "step": 179400
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 7.63363037030492e-05,
+      "loss": 0.6963,
+      "step": 179600
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 7.6309927769506e-05,
+      "loss": 0.6944,
+      "step": 179800
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 7.628355183596279e-05,
+      "loss": 0.6909,
+      "step": 180000
+    },
+    {
+      "epoch": 0.24,
+      "eval_em_first_line_ratio": 0.5033,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3907,
+      "eval_loss": 0.6916074752807617,
+      "eval_runtime": 226.9397,
+      "eval_samples_per_second": 44.065,
+      "eval_steps_per_second": 0.234,
+      "step": 180000
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 7.62571759024196e-05,
+      "loss": 0.698,
+      "step": 180200
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 7.62307999688764e-05,
+      "loss": 0.6971,
+      "step": 180400
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 7.62044240353332e-05,
+      "loss": 0.6954,
+      "step": 180600
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 7.617817998145772e-05,
+      "loss": 0.6986,
+      "step": 180800
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 7.615180404791452e-05,
+      "loss": 0.6931,
+      "step": 181000
+    },
+    {
+      "epoch": 0.24,
+      "eval_em_first_line_ratio": 0.5018,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3889,
+      "eval_loss": 0.6931282877922058,
+      "eval_runtime": 224.2909,
+      "eval_samples_per_second": 44.585,
+      "eval_steps_per_second": 0.236,
+      "step": 181000
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 7.612542811437133e-05,
+      "loss": 0.6905,
+      "step": 181200
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 7.609905218082813e-05,
+      "loss": 0.6932,
+      "step": 181400
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 7.607267624728494e-05,
+      "loss": 0.6942,
+      "step": 181600
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 7.604630031374174e-05,
+      "loss": 0.688,
+      "step": 181800
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 7.601992438019854e-05,
+      "loss": 0.6916,
+      "step": 182000
+    },
+    {
+      "epoch": 0.24,
+      "eval_em_first_line_ratio": 0.5018,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3894,
+      "eval_loss": 0.6921393871307373,
+      "eval_runtime": 227.7389,
+      "eval_samples_per_second": 43.91,
+      "eval_steps_per_second": 0.233,
+      "step": 182000
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 7.599354844665533e-05,
+      "loss": 0.6909,
+      "step": 182200
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 7.596717251311214e-05,
+      "loss": 0.6935,
+      "step": 182400
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 7.594079657956894e-05,
+      "loss": 0.6941,
+      "step": 182600
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 7.591442064602573e-05,
+      "loss": 0.6912,
+      "step": 182800
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 7.588804471248254e-05,
+      "loss": 0.693,
+      "step": 183000
+    },
+    {
+      "epoch": 0.24,
+      "eval_em_first_line_ratio": 0.5042,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3911,
+      "eval_loss": 0.6916182041168213,
+      "eval_runtime": 230.4851,
+      "eval_samples_per_second": 43.387,
+      "eval_steps_per_second": 0.23,
+      "step": 183000
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 7.586166877893934e-05,
+      "loss": 0.6974,
+      "step": 183200
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 7.583529284539616e-05,
+      "loss": 0.694,
+      "step": 183400
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 7.580891691185295e-05,
+      "loss": 0.6987,
+      "step": 183600
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 7.578254097830976e-05,
+      "loss": 0.6912,
+      "step": 183800
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 7.575616504476656e-05,
+      "loss": 0.6922,
+      "step": 184000
+    },
+    {
+      "epoch": 0.24,
+      "eval_em_first_line_ratio": 0.5036,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3904,
+      "eval_loss": 0.6919209361076355,
+      "eval_runtime": 228.7244,
+      "eval_samples_per_second": 43.721,
+      "eval_steps_per_second": 0.232,
+      "step": 184000
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 7.572978911122337e-05,
+      "loss": 0.6923,
+      "step": 184200
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 7.570341317768016e-05,
+      "loss": 0.6964,
+      "step": 184400
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 7.567703724413696e-05,
+      "loss": 0.6925,
+      "step": 184600
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 7.565066131059377e-05,
+      "loss": 0.6926,
+      "step": 184800
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 7.562428537705057e-05,
+      "loss": 0.6954,
+      "step": 185000
+    },
+    {
+      "epoch": 0.24,
+      "eval_em_first_line_ratio": 0.503,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3899,
+      "eval_loss": 0.6922071576118469,
+      "eval_runtime": 233.4024,
+      "eval_samples_per_second": 42.844,
+      "eval_steps_per_second": 0.227,
+      "step": 185000
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 7.559790944350736e-05,
+      "loss": 0.6914,
+      "step": 185200
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 7.557153350996418e-05,
+      "loss": 0.6929,
+      "step": 185400
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 7.554515757642098e-05,
+      "loss": 0.6916,
+      "step": 185600
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 7.551878164287777e-05,
+      "loss": 0.6964,
+      "step": 185800
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 7.549240570933458e-05,
+      "loss": 0.6918,
+      "step": 186000
+    },
+    {
+      "epoch": 0.25,
+      "eval_em_first_line_ratio": 0.5052,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3915,
+      "eval_loss": 0.689504861831665,
+      "eval_runtime": 238.7665,
+      "eval_samples_per_second": 41.882,
+      "eval_steps_per_second": 0.222,
+      "step": 186000
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 7.546602977579138e-05,
+      "loss": 0.6935,
+      "step": 186200
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 7.543965384224819e-05,
+      "loss": 0.6932,
+      "step": 186400
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 7.541327790870498e-05,
+      "loss": 0.6904,
+      "step": 186600
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 7.538690197516178e-05,
+      "loss": 0.6943,
+      "step": 186800
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 7.536052604161859e-05,
+      "loss": 0.6902,
+      "step": 187000
+    },
+    {
+      "epoch": 0.25,
+      "eval_em_first_line_ratio": 0.5012,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3887,
+      "eval_loss": 0.6916294693946838,
+      "eval_runtime": 235.6106,
+      "eval_samples_per_second": 42.443,
+      "eval_steps_per_second": 0.225,
+      "step": 187000
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 7.533415010807539e-05,
+      "loss": 0.6964,
+      "step": 187200
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 7.53077741745322e-05,
+      "loss": 0.6874,
+      "step": 187400
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 7.5281398240989e-05,
+      "loss": 0.6915,
+      "step": 187600
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 7.52550223074458e-05,
+      "loss": 0.6895,
+      "step": 187800
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 7.52286463739026e-05,
+      "loss": 0.689,
+      "step": 188000
+    },
+    {
+      "epoch": 0.25,
+      "eval_em_first_line_ratio": 0.5024,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3897,
+      "eval_loss": 0.6902817487716675,
+      "eval_runtime": 234.9246,
+      "eval_samples_per_second": 42.567,
+      "eval_steps_per_second": 0.226,
+      "step": 188000
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 7.52022704403594e-05,
+      "loss": 0.6931,
+      "step": 188200
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 7.51758945068162e-05,
+      "loss": 0.6931,
+      "step": 188400
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 7.514951857327301e-05,
+      "loss": 0.6908,
+      "step": 188600
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 7.512327451939752e-05,
+      "loss": 0.6917,
+      "step": 188800
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 7.509689858585433e-05,
+      "loss": 0.6906,
+      "step": 189000
+    },
+    {
+      "epoch": 0.25,
+      "eval_em_first_line_ratio": 0.5037,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3899,
+      "eval_loss": 0.6903719305992126,
+      "eval_runtime": 230.5001,
+      "eval_samples_per_second": 43.384,
+      "eval_steps_per_second": 0.23,
+      "step": 189000
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 7.507052265231113e-05,
+      "loss": 0.6904,
+      "step": 189200
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 7.504414671876792e-05,
+      "loss": 0.6917,
+      "step": 189400
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 7.501777078522473e-05,
+      "loss": 0.6965,
+      "step": 189600
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 7.499152673134925e-05,
+      "loss": 0.6893,
+      "step": 189800
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 7.496528267747377e-05,
+      "loss": 0.6945,
+      "step": 190000
+    },
+    {
+      "epoch": 0.25,
+      "eval_em_first_line_ratio": 0.4993,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3862,
+      "eval_loss": 0.6902524828910828,
+      "eval_runtime": 225.6459,
+      "eval_samples_per_second": 44.317,
+      "eval_steps_per_second": 0.235,
+      "step": 190000
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 7.493890674393057e-05,
+      "loss": 0.6911,
+      "step": 190200
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 7.491253081038737e-05,
+      "loss": 0.6896,
+      "step": 190400
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 7.488615487684417e-05,
+      "loss": 0.6957,
+      "step": 190600
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 7.485977894330097e-05,
+      "loss": 0.6916,
+      "step": 190800
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 7.483340300975779e-05,
+      "loss": 0.6942,
+      "step": 191000
+    },
+    {
+      "epoch": 0.25,
+      "eval_em_first_line_ratio": 0.5022,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3892,
+      "eval_loss": 0.6894727945327759,
+      "eval_runtime": 232.5796,
+      "eval_samples_per_second": 42.996,
+      "eval_steps_per_second": 0.228,
+      "step": 191000
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 7.480702707621459e-05,
+      "loss": 0.6895,
+      "step": 191200
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 7.478065114267138e-05,
+      "loss": 0.6932,
+      "step": 191400
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 7.475427520912819e-05,
+      "loss": 0.6903,
+      "step": 191600
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 7.472789927558499e-05,
+      "loss": 0.6898,
+      "step": 191800
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 7.47015233420418e-05,
+      "loss": 0.6915,
+      "step": 192000
+    },
+    {
+      "epoch": 0.25,
+      "eval_em_first_line_ratio": 0.5004,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3875,
+      "eval_loss": 0.6885014772415161,
+      "eval_runtime": 229.6192,
+      "eval_samples_per_second": 43.55,
+      "eval_steps_per_second": 0.231,
+      "step": 192000
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 7.467514740849859e-05,
+      "loss": 0.6884,
+      "step": 192200
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 7.464877147495539e-05,
+      "loss": 0.6876,
+      "step": 192400
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 7.46223955414122e-05,
+      "loss": 0.6894,
+      "step": 192600
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 7.4596019607869e-05,
+      "loss": 0.6823,
+      "step": 192800
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 7.45696436743258e-05,
+      "loss": 0.6941,
+      "step": 193000
+    },
+    {
+      "epoch": 0.25,
+      "eval_em_first_line_ratio": 0.504,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.391,
+      "eval_loss": 0.6897890567779541,
+      "eval_runtime": 230.735,
+      "eval_samples_per_second": 43.34,
+      "eval_steps_per_second": 0.23,
+      "step": 193000
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 7.454326774078261e-05,
+      "loss": 0.6862,
+      "step": 193200
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 7.451689180723941e-05,
+      "loss": 0.6922,
+      "step": 193400
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 7.44905158736962e-05,
+      "loss": 0.6947,
+      "step": 193600
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 7.446413994015301e-05,
+      "loss": 0.6873,
+      "step": 193800
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 7.443776400660981e-05,
+      "loss": 0.6893,
+      "step": 194000
+    },
+    {
+      "epoch": 0.26,
+      "eval_em_first_line_ratio": 0.5038,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3909,
+      "eval_loss": 0.6887952089309692,
+      "eval_runtime": 234.3937,
+      "eval_samples_per_second": 42.663,
+      "eval_steps_per_second": 0.226,
+      "step": 194000
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 7.441138807306662e-05,
+      "loss": 0.6908,
+      "step": 194200
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 7.438501213952341e-05,
+      "loss": 0.6903,
+      "step": 194400
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 7.435863620598021e-05,
+      "loss": 0.6928,
+      "step": 194600
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 7.433226027243703e-05,
+      "loss": 0.6869,
+      "step": 194800
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 7.430588433889382e-05,
+      "loss": 0.6919,
+      "step": 195000
+    },
+    {
+      "epoch": 0.26,
+      "eval_em_first_line_ratio": 0.5013,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3884,
+      "eval_loss": 0.6877405643463135,
+      "eval_runtime": 234.2192,
+      "eval_samples_per_second": 42.695,
+      "eval_steps_per_second": 0.226,
+      "step": 195000
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 7.427950840535063e-05,
+      "loss": 0.6889,
+      "step": 195200
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 7.425313247180743e-05,
+      "loss": 0.6937,
+      "step": 195400
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 7.422675653826424e-05,
+      "loss": 0.6896,
+      "step": 195600
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 7.420038060472103e-05,
+      "loss": 0.6898,
+      "step": 195800
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 7.417413655084555e-05,
+      "loss": 0.6897,
+      "step": 196000
+    },
+    {
+      "epoch": 0.26,
+      "eval_em_first_line_ratio": 0.5035,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3907,
+      "eval_loss": 0.6882504224777222,
+      "eval_runtime": 229.4919,
+      "eval_samples_per_second": 43.575,
+      "eval_steps_per_second": 0.231,
+      "step": 196000
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 7.414776061730236e-05,
+      "loss": 0.6941,
+      "step": 196200
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 7.412138468375915e-05,
+      "loss": 0.6917,
+      "step": 196400
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 7.409500875021595e-05,
+      "loss": 0.6952,
+      "step": 196600
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 7.406863281667276e-05,
+      "loss": 0.6853,
+      "step": 196800
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 7.404225688312956e-05,
+      "loss": 0.6923,
+      "step": 197000
+    },
+    {
+      "epoch": 0.26,
+      "eval_em_first_line_ratio": 0.5036,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3904,
+      "eval_loss": 0.6881018280982971,
+      "eval_runtime": 225.5849,
+      "eval_samples_per_second": 44.329,
+      "eval_steps_per_second": 0.235,
+      "step": 197000
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 7.401588094958635e-05,
+      "loss": 0.6918,
+      "step": 197200
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 7.398963689571088e-05,
+      "loss": 0.6872,
+      "step": 197400
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 7.396326096216768e-05,
+      "loss": 0.6903,
+      "step": 197600
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 7.393688502862447e-05,
+      "loss": 0.6905,
+      "step": 197800
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 7.391050909508129e-05,
+      "loss": 0.6882,
+      "step": 198000
+    },
+    {
+      "epoch": 0.26,
+      "eval_em_first_line_ratio": 0.5031,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3895,
+      "eval_loss": 0.6877618432044983,
+      "eval_runtime": 223.7442,
+      "eval_samples_per_second": 44.694,
+      "eval_steps_per_second": 0.237,
+      "step": 198000
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 7.38841331615381e-05,
+      "loss": 0.6862,
+      "step": 198200
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 7.38577572279949e-05,
+      "loss": 0.6885,
+      "step": 198400
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 7.383138129445169e-05,
+      "loss": 0.6861,
+      "step": 198600
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 7.38050053609085e-05,
+      "loss": 0.685,
+      "step": 198800
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 7.37786294273653e-05,
+      "loss": 0.691,
+      "step": 199000
+    },
+    {
+      "epoch": 0.26,
+      "eval_em_first_line_ratio": 0.5029,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3897,
+      "eval_loss": 0.6864962577819824,
+      "eval_runtime": 229.8356,
+      "eval_samples_per_second": 43.509,
+      "eval_steps_per_second": 0.231,
+      "step": 199000
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 7.375225349382209e-05,
+      "loss": 0.6874,
+      "step": 199200
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 7.37258775602789e-05,
+      "loss": 0.6893,
+      "step": 199400
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 7.36995016267357e-05,
+      "loss": 0.6927,
+      "step": 199600
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 7.36731256931925e-05,
+      "loss": 0.6861,
+      "step": 199800
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 7.364674975964931e-05,
+      "loss": 0.6848,
+      "step": 200000
+    },
+    {
+      "epoch": 0.26,
+      "eval_em_first_line_ratio": 0.502,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.389,
+      "eval_loss": 0.6876330971717834,
+      "eval_runtime": 221.9568,
+      "eval_samples_per_second": 45.054,
+      "eval_steps_per_second": 0.239,
+      "step": 200000
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 7.362037382610611e-05,
+      "loss": 0.6867,
+      "step": 200200
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 7.359399789256292e-05,
+      "loss": 0.6871,
+      "step": 200400
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 7.356762195901972e-05,
+      "loss": 0.6843,
+      "step": 200600
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 7.354124602547651e-05,
+      "loss": 0.6864,
+      "step": 200800
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 7.351487009193332e-05,
+      "loss": 0.69,
+      "step": 201000
+    },
+    {
+      "epoch": 0.27,
+      "eval_em_first_line_ratio": 0.5017,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3883,
+      "eval_loss": 0.6865004897117615,
+      "eval_runtime": 232.5619,
+      "eval_samples_per_second": 42.999,
+      "eval_steps_per_second": 0.228,
+      "step": 201000
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 7.348862603805785e-05,
+      "loss": 0.6852,
+      "step": 201200
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 7.346225010451464e-05,
+      "loss": 0.6916,
+      "step": 201400
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 7.343587417097144e-05,
+      "loss": 0.689,
+      "step": 201600
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 7.340949823742824e-05,
+      "loss": 0.6858,
+      "step": 201800
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 7.338312230388505e-05,
+      "loss": 0.6827,
+      "step": 202000
+    },
+    {
+      "epoch": 0.27,
+      "eval_em_first_line_ratio": 0.5053,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3919,
+      "eval_loss": 0.6869927048683167,
+      "eval_runtime": 223.1083,
+      "eval_samples_per_second": 44.821,
+      "eval_steps_per_second": 0.238,
+      "step": 202000
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 7.335674637034184e-05,
+      "loss": 0.692,
+      "step": 202200
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 7.333037043679864e-05,
+      "loss": 0.6829,
+      "step": 202400
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 7.330399450325546e-05,
+      "loss": 0.6852,
+      "step": 202600
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 7.327761856971225e-05,
+      "loss": 0.686,
+      "step": 202800
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 7.325124263616906e-05,
+      "loss": 0.6866,
+      "step": 203000
+    },
+    {
+      "epoch": 0.27,
+      "eval_em_first_line_ratio": 0.5039,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3908,
+      "eval_loss": 0.6865259408950806,
+      "eval_runtime": 232.9531,
+      "eval_samples_per_second": 42.927,
+      "eval_steps_per_second": 0.228,
+      "step": 203000
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 7.322486670262586e-05,
+      "loss": 0.6848,
+      "step": 203200
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 7.319862264875038e-05,
+      "loss": 0.6842,
+      "step": 203400
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 7.317224671520718e-05,
+      "loss": 0.6886,
+      "step": 203600
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 7.314587078166398e-05,
+      "loss": 0.6846,
+      "step": 203800
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 7.311949484812079e-05,
+      "loss": 0.6897,
+      "step": 204000
+    },
+    {
+      "epoch": 0.27,
+      "eval_em_first_line_ratio": 0.5028,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3898,
+      "eval_loss": 0.6874383091926575,
+      "eval_runtime": 235.3167,
+      "eval_samples_per_second": 42.496,
+      "eval_steps_per_second": 0.225,
+      "step": 204000
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 7.309311891457758e-05,
+      "loss": 0.6852,
+      "step": 204200
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 7.306674298103438e-05,
+      "loss": 0.6878,
+      "step": 204400
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 7.304036704749119e-05,
+      "loss": 0.6862,
+      "step": 204600
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 7.301399111394799e-05,
+      "loss": 0.6827,
+      "step": 204800
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 7.29876151804048e-05,
+      "loss": 0.6913,
+      "step": 205000
+    },
+    {
+      "epoch": 0.27,
+      "eval_em_first_line_ratio": 0.5032,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3897,
+      "eval_loss": 0.6848097443580627,
+      "eval_runtime": 234.3568,
+      "eval_samples_per_second": 42.67,
+      "eval_steps_per_second": 0.226,
+      "step": 205000
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 7.29612392468616e-05,
+      "loss": 0.6878,
+      "step": 205200
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 7.29348633133184e-05,
+      "loss": 0.6859,
+      "step": 205400
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 7.29084873797752e-05,
+      "loss": 0.6913,
+      "step": 205600
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 7.2882111446232e-05,
+      "loss": 0.6839,
+      "step": 205800
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 7.28557355126888e-05,
+      "loss": 0.685,
+      "step": 206000
+    },
+    {
+      "epoch": 0.27,
+      "eval_em_first_line_ratio": 0.5064,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3928,
+      "eval_loss": 0.6848724484443665,
+      "eval_runtime": 230.6241,
+      "eval_samples_per_second": 43.361,
+      "eval_steps_per_second": 0.23,
+      "step": 206000
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 7.282935957914561e-05,
+      "loss": 0.6859,
+      "step": 206200
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 7.28029836456024e-05,
+      "loss": 0.6881,
+      "step": 206400
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 7.27766077120592e-05,
+      "loss": 0.685,
+      "step": 206600
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 7.275036365818373e-05,
+      "loss": 0.6833,
+      "step": 206800
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 7.272398772464052e-05,
+      "loss": 0.6853,
+      "step": 207000
+    },
+    {
+      "epoch": 0.27,
+      "eval_em_first_line_ratio": 0.5039,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3906,
+      "eval_loss": 0.6854568123817444,
+      "eval_runtime": 232.1379,
+      "eval_samples_per_second": 43.078,
+      "eval_steps_per_second": 0.228,
+      "step": 207000
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 7.269761179109733e-05,
+      "loss": 0.6857,
+      "step": 207200
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 7.267123585755413e-05,
+      "loss": 0.6856,
+      "step": 207400
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 7.264485992401095e-05,
+      "loss": 0.6845,
+      "step": 207600
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 7.261848399046774e-05,
+      "loss": 0.684,
+      "step": 207800
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 7.259210805692455e-05,
+      "loss": 0.6875,
+      "step": 208000
+    },
+    {
+      "epoch": 0.27,
+      "eval_em_first_line_ratio": 0.5051,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3918,
+      "eval_loss": 0.6844572424888611,
+      "eval_runtime": 232.6854,
+      "eval_samples_per_second": 42.976,
+      "eval_steps_per_second": 0.228,
+      "step": 208000
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 7.256573212338135e-05,
+      "loss": 0.6826,
+      "step": 208200
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 7.253935618983815e-05,
+      "loss": 0.6859,
+      "step": 208400
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 7.251298025629495e-05,
+      "loss": 0.684,
+      "step": 208600
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 7.248660432275175e-05,
+      "loss": 0.6851,
+      "step": 208800
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 7.246022838920855e-05,
+      "loss": 0.6852,
+      "step": 209000
+    },
+    {
+      "epoch": 0.28,
+      "eval_em_first_line_ratio": 0.5048,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3916,
+      "eval_loss": 0.6844390630722046,
+      "eval_runtime": 232.7768,
+      "eval_samples_per_second": 42.96,
+      "eval_steps_per_second": 0.228,
+      "step": 209000
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 7.243385245566536e-05,
+      "loss": 0.6868,
+      "step": 209200
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 7.240747652212215e-05,
+      "loss": 0.6824,
+      "step": 209400
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 7.238110058857897e-05,
+      "loss": 0.6829,
+      "step": 209600
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 7.235472465503577e-05,
+      "loss": 0.6836,
+      "step": 209800
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 7.232834872149256e-05,
+      "loss": 0.6834,
+      "step": 210000
+    },
+    {
+      "epoch": 0.28,
+      "eval_em_first_line_ratio": 0.5095,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3952,
+      "eval_loss": 0.6832934021949768,
+      "eval_runtime": 230.8693,
+      "eval_samples_per_second": 43.315,
+      "eval_steps_per_second": 0.23,
+      "step": 210000
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 7.230197278794937e-05,
+      "loss": 0.6864,
+      "step": 210200
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 7.227559685440617e-05,
+      "loss": 0.6828,
+      "step": 210400
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 7.224922092086298e-05,
+      "loss": 0.6879,
+      "step": 210600
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 7.222284498731977e-05,
+      "loss": 0.6841,
+      "step": 210800
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 7.21966009334443e-05,
+      "loss": 0.6822,
+      "step": 211000
+    },
+    {
+      "epoch": 0.28,
+      "eval_em_first_line_ratio": 0.5057,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3926,
+      "eval_loss": 0.6841102242469788,
+      "eval_runtime": 235.8637,
+      "eval_samples_per_second": 42.397,
+      "eval_steps_per_second": 0.225,
+      "step": 211000
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 7.21702249999011e-05,
+      "loss": 0.6839,
+      "step": 211200
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 7.214384906635789e-05,
+      "loss": 0.6828,
+      "step": 211400
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 7.211747313281469e-05,
+      "loss": 0.6818,
+      "step": 211600
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 7.20910971992715e-05,
+      "loss": 0.6856,
+      "step": 211800
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 7.20647212657283e-05,
+      "loss": 0.6856,
+      "step": 212000
+    },
+    {
+      "epoch": 0.28,
+      "eval_em_first_line_ratio": 0.5076,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3946,
+      "eval_loss": 0.6842368245124817,
+      "eval_runtime": 230.6503,
+      "eval_samples_per_second": 43.356,
+      "eval_steps_per_second": 0.23,
+      "step": 212000
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 7.20383453321851e-05,
+      "loss": 0.6837,
+      "step": 212200
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 7.201196939864191e-05,
+      "loss": 0.685,
+      "step": 212400
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 7.198559346509872e-05,
+      "loss": 0.6837,
+      "step": 212600
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 7.19592175315555e-05,
+      "loss": 0.6827,
+      "step": 212800
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 7.193284159801231e-05,
+      "loss": 0.6818,
+      "step": 213000
+    },
+    {
+      "epoch": 0.28,
+      "eval_em_first_line_ratio": 0.5062,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3925,
+      "eval_loss": 0.6834655404090881,
+      "eval_runtime": 236.0564,
+      "eval_samples_per_second": 42.363,
+      "eval_steps_per_second": 0.225,
+      "step": 213000
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 7.190646566446911e-05,
+      "loss": 0.6828,
+      "step": 213200
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 7.188008973092592e-05,
+      "loss": 0.6878,
+      "step": 213400
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 7.185371379738271e-05,
+      "loss": 0.6804,
+      "step": 213600
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 7.182733786383951e-05,
+      "loss": 0.6839,
+      "step": 213800
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 7.180096193029633e-05,
+      "loss": 0.6769,
+      "step": 214000
+    },
+    {
+      "epoch": 0.28,
+      "eval_em_first_line_ratio": 0.502,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3886,
+      "eval_loss": 0.6838258504867554,
+      "eval_runtime": 231.294,
+      "eval_samples_per_second": 43.235,
+      "eval_steps_per_second": 0.229,
+      "step": 214000
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 7.177458599675312e-05,
+      "loss": 0.6835,
+      "step": 214200
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 7.174834194287764e-05,
+      "loss": 0.68,
+      "step": 214400
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 7.172222976866989e-05,
+      "loss": 0.6838,
+      "step": 214600
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 7.169585383512668e-05,
+      "loss": 0.6793,
+      "step": 214800
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 7.166947790158348e-05,
+      "loss": 0.683,
+      "step": 215000
+    },
+    {
+      "epoch": 0.28,
+      "eval_em_first_line_ratio": 0.5065,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3924,
+      "eval_loss": 0.6839645504951477,
+      "eval_runtime": 230.3455,
+      "eval_samples_per_second": 43.413,
+      "eval_steps_per_second": 0.23,
+      "step": 215000
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 7.164310196804029e-05,
+      "loss": 0.6835,
+      "step": 215200
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 7.161672603449709e-05,
+      "loss": 0.68,
+      "step": 215400
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 7.159035010095388e-05,
+      "loss": 0.6856,
+      "step": 215600
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 7.15639741674107e-05,
+      "loss": 0.6801,
+      "step": 215800
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 7.15375982338675e-05,
+      "loss": 0.6844,
+      "step": 216000
+    },
+    {
+      "epoch": 0.28,
+      "eval_em_first_line_ratio": 0.5052,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.392,
+      "eval_loss": 0.6831923127174377,
+      "eval_runtime": 229.9395,
+      "eval_samples_per_second": 43.49,
+      "eval_steps_per_second": 0.23,
+      "step": 216000
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 7.15112223003243e-05,
+      "loss": 0.6803,
+      "step": 216200
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 7.14848463667811e-05,
+      "loss": 0.6839,
+      "step": 216400
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 7.14584704332379e-05,
+      "loss": 0.68,
+      "step": 216600
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 7.143209449969471e-05,
+      "loss": 0.6827,
+      "step": 216800
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 7.14057185661515e-05,
+      "loss": 0.6872,
+      "step": 217000
+    },
+    {
+      "epoch": 0.29,
+      "eval_em_first_line_ratio": 0.5045,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3916,
+      "eval_loss": 0.6833539009094238,
+      "eval_runtime": 230.3676,
+      "eval_samples_per_second": 43.409,
+      "eval_steps_per_second": 0.23,
+      "step": 217000
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 7.13793426326083e-05,
+      "loss": 0.6859,
+      "step": 217200
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 7.135296669906511e-05,
+      "loss": 0.6836,
+      "step": 217400
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 7.132659076552191e-05,
+      "loss": 0.6813,
+      "step": 217600
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 7.130021483197872e-05,
+      "loss": 0.6835,
+      "step": 217800
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 7.127383889843552e-05,
+      "loss": 0.6773,
+      "step": 218000
+    },
+    {
+      "epoch": 0.29,
+      "eval_em_first_line_ratio": 0.504,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3903,
+      "eval_loss": 0.6829410791397095,
+      "eval_runtime": 226.0626,
+      "eval_samples_per_second": 44.236,
+      "eval_steps_per_second": 0.234,
+      "step": 218000
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 7.124746296489232e-05,
+      "loss": 0.6812,
+      "step": 218200
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 7.122108703134912e-05,
+      "loss": 0.6834,
+      "step": 218400
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 7.119471109780592e-05,
+      "loss": 0.6787,
+      "step": 218600
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 7.116833516426272e-05,
+      "loss": 0.6818,
+      "step": 218800
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 7.114195923071953e-05,
+      "loss": 0.6756,
+      "step": 219000
+    },
+    {
+      "epoch": 0.29,
+      "eval_em_first_line_ratio": 0.5031,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3887,
+      "eval_loss": 0.6822870373725891,
+      "eval_runtime": 235.7215,
+      "eval_samples_per_second": 42.423,
+      "eval_steps_per_second": 0.225,
+      "step": 219000
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 7.111571517684404e-05,
+      "loss": 0.6819,
+      "step": 219200
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 7.108933924330085e-05,
+      "loss": 0.6821,
+      "step": 219400
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 7.106296330975765e-05,
+      "loss": 0.6826,
+      "step": 219600
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 7.103658737621444e-05,
+      "loss": 0.6809,
+      "step": 219800
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 7.101021144267125e-05,
+      "loss": 0.6794,
+      "step": 220000
+    },
+    {
+      "epoch": 0.29,
+      "eval_em_first_line_ratio": 0.5047,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3906,
+      "eval_loss": 0.6822872757911682,
+      "eval_runtime": 225.0168,
+      "eval_samples_per_second": 44.441,
+      "eval_steps_per_second": 0.236,
+      "step": 220000
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 7.098383550912805e-05,
+      "loss": 0.6853,
+      "step": 220200
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 7.095745957558486e-05,
+      "loss": 0.6822,
+      "step": 220400
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 7.093108364204166e-05,
+      "loss": 0.6796,
+      "step": 220600
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 7.090470770849846e-05,
+      "loss": 0.6794,
+      "step": 220800
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 7.087833177495527e-05,
+      "loss": 0.6822,
+      "step": 221000
+    },
+    {
+      "epoch": 0.29,
+      "eval_em_first_line_ratio": 0.5054,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3917,
+      "eval_loss": 0.6826604008674622,
+      "eval_runtime": 224.9791,
+      "eval_samples_per_second": 44.449,
+      "eval_steps_per_second": 0.236,
+      "step": 221000
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 7.085195584141206e-05,
+      "loss": 0.6839,
+      "step": 221200
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 7.082557990786886e-05,
+      "loss": 0.68,
+      "step": 221400
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 7.079920397432567e-05,
+      "loss": 0.6833,
+      "step": 221600
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 7.077282804078247e-05,
+      "loss": 0.6828,
+      "step": 221800
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 7.074645210723926e-05,
+      "loss": 0.6797,
+      "step": 222000
+    },
+    {
+      "epoch": 0.29,
+      "eval_em_first_line_ratio": 0.5025,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3885,
+      "eval_loss": 0.6821091175079346,
+      "eval_runtime": 233.7121,
+      "eval_samples_per_second": 42.788,
+      "eval_steps_per_second": 0.227,
+      "step": 222000
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 7.072020805336379e-05,
+      "loss": 0.6857,
+      "step": 222200
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 7.06938321198206e-05,
+      "loss": 0.6851,
+      "step": 222400
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 7.066745618627739e-05,
+      "loss": 0.679,
+      "step": 222600
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 7.064108025273419e-05,
+      "loss": 0.6803,
+      "step": 222800
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 7.061470431919101e-05,
+      "loss": 0.679,
+      "step": 223000
+    },
+    {
+      "epoch": 0.29,
+      "eval_em_first_line_ratio": 0.503,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3896,
+      "eval_loss": 0.6818556785583496,
+      "eval_runtime": 236.6794,
+      "eval_samples_per_second": 42.251,
+      "eval_steps_per_second": 0.224,
+      "step": 223000
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 7.058832838564781e-05,
+      "loss": 0.6833,
+      "step": 223200
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 7.05619524521046e-05,
+      "loss": 0.6787,
+      "step": 223400
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 7.053557651856141e-05,
+      "loss": 0.6764,
+      "step": 223600
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 7.050920058501821e-05,
+      "loss": 0.6834,
+      "step": 223800
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 7.048282465147502e-05,
+      "loss": 0.6815,
+      "step": 224000
+    },
+    {
+      "epoch": 0.3,
+      "eval_em_first_line_ratio": 0.5051,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3915,
+      "eval_loss": 0.68302983045578,
+      "eval_runtime": 231.4248,
+      "eval_samples_per_second": 43.211,
+      "eval_steps_per_second": 0.229,
+      "step": 224000
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 7.045644871793181e-05,
+      "loss": 0.6832,
+      "step": 224200
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 7.043007278438861e-05,
+      "loss": 0.6808,
+      "step": 224400
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 7.040369685084542e-05,
+      "loss": 0.6778,
+      "step": 224600
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 7.037732091730222e-05,
+      "loss": 0.6831,
+      "step": 224800
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 7.035094498375902e-05,
+      "loss": 0.6771,
+      "step": 225000
+    },
+    {
+      "epoch": 0.3,
+      "eval_em_first_line_ratio": 0.5049,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3917,
+      "eval_loss": 0.6818101406097412,
+      "eval_runtime": 230.9043,
+      "eval_samples_per_second": 43.308,
+      "eval_steps_per_second": 0.23,
+      "step": 225000
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 7.032470092988354e-05,
+      "loss": 0.6757,
+      "step": 225200
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 7.029832499634034e-05,
+      "loss": 0.6806,
+      "step": 225400
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 7.027194906279715e-05,
+      "loss": 0.6783,
+      "step": 225600
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 7.024557312925395e-05,
+      "loss": 0.6809,
+      "step": 225800
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 7.021919719571076e-05,
+      "loss": 0.6793,
+      "step": 226000
+    },
+    {
+      "epoch": 0.3,
+      "eval_em_first_line_ratio": 0.5048,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3919,
+      "eval_loss": 0.6809067130088806,
+      "eval_runtime": 229.071,
+      "eval_samples_per_second": 43.655,
+      "eval_steps_per_second": 0.231,
+      "step": 226000
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 7.019282126216755e-05,
+      "loss": 0.6784,
+      "step": 226200
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 7.016644532862435e-05,
+      "loss": 0.6808,
+      "step": 226400
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 7.014006939508116e-05,
+      "loss": 0.6773,
+      "step": 226600
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 7.011369346153796e-05,
+      "loss": 0.6765,
+      "step": 226800
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 7.008731752799475e-05,
+      "loss": 0.6805,
+      "step": 227000
+    },
+    {
+      "epoch": 0.3,
+      "eval_em_first_line_ratio": 0.507,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3938,
+      "eval_loss": 0.6813559532165527,
+      "eval_runtime": 226.2274,
+      "eval_samples_per_second": 44.203,
+      "eval_steps_per_second": 0.234,
+      "step": 227000
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 7.006094159445156e-05,
+      "loss": 0.6816,
+      "step": 227200
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 7.003469754057608e-05,
+      "loss": 0.6824,
+      "step": 227400
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 7.000832160703287e-05,
+      "loss": 0.6834,
+      "step": 227600
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 6.998194567348968e-05,
+      "loss": 0.6821,
+      "step": 227800
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 6.99555697399465e-05,
+      "loss": 0.6787,
+      "step": 228000
+    },
+    {
+      "epoch": 0.3,
+      "eval_em_first_line_ratio": 0.5059,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3928,
+      "eval_loss": 0.6805691123008728,
+      "eval_runtime": 233.612,
+      "eval_samples_per_second": 42.806,
+      "eval_steps_per_second": 0.227,
+      "step": 228000
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 6.992919380640329e-05,
+      "loss": 0.676,
+      "step": 228200
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 6.990281787286009e-05,
+      "loss": 0.6747,
+      "step": 228400
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 6.98764419393169e-05,
+      "loss": 0.6782,
+      "step": 228600
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 6.98500660057737e-05,
+      "loss": 0.6806,
+      "step": 228800
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 6.982382195189821e-05,
+      "loss": 0.6751,
+      "step": 229000
+    },
+    {
+      "epoch": 0.3,
+      "eval_em_first_line_ratio": 0.5053,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3924,
+      "eval_loss": 0.6796121001243591,
+      "eval_runtime": 227.7996,
+      "eval_samples_per_second": 43.898,
+      "eval_steps_per_second": 0.233,
+      "step": 229000
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 6.979744601835502e-05,
+      "loss": 0.6749,
+      "step": 229200
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 6.977107008481182e-05,
+      "loss": 0.6792,
+      "step": 229400
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 6.974469415126863e-05,
+      "loss": 0.6811,
+      "step": 229600
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 6.971831821772542e-05,
+      "loss": 0.6819,
+      "step": 229800
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 6.969194228418222e-05,
+      "loss": 0.6779,
+      "step": 230000
+    },
+    {
+      "epoch": 0.3,
+      "eval_em_first_line_ratio": 0.5058,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3921,
+      "eval_loss": 0.6792682409286499,
+      "eval_runtime": 229.7768,
+      "eval_samples_per_second": 43.52,
+      "eval_steps_per_second": 0.231,
+      "step": 230000
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 6.966556635063903e-05,
+      "loss": 0.6791,
+      "step": 230200
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 6.963919041709582e-05,
+      "loss": 0.6817,
+      "step": 230400
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 6.961281448355263e-05,
+      "loss": 0.6752,
+      "step": 230600
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 6.958643855000944e-05,
+      "loss": 0.6769,
+      "step": 230800
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 6.956006261646624e-05,
+      "loss": 0.6808,
+      "step": 231000
+    },
+    {
+      "epoch": 0.3,
+      "eval_em_first_line_ratio": 0.507,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3929,
+      "eval_loss": 0.679900586605072,
+      "eval_runtime": 227.6546,
+      "eval_samples_per_second": 43.926,
+      "eval_steps_per_second": 0.233,
+      "step": 231000
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 6.953368668292303e-05,
+      "loss": 0.6764,
+      "step": 231200
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 6.950731074937984e-05,
+      "loss": 0.6763,
+      "step": 231400
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 6.948093481583664e-05,
+      "loss": 0.6744,
+      "step": 231600
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 6.945455888229345e-05,
+      "loss": 0.6737,
+      "step": 231800
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 6.942818294875024e-05,
+      "loss": 0.676,
+      "step": 232000
+    },
+    {
+      "epoch": 0.31,
+      "eval_em_first_line_ratio": 0.5043,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3909,
+      "eval_loss": 0.6792597770690918,
+      "eval_runtime": 231.14,
+      "eval_samples_per_second": 43.264,
+      "eval_steps_per_second": 0.229,
+      "step": 232000
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 6.940180701520704e-05,
+      "loss": 0.6762,
+      "step": 232200
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 6.937543108166386e-05,
+      "loss": 0.6779,
+      "step": 232400
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 6.934905514812065e-05,
+      "loss": 0.6814,
+      "step": 232600
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 6.932267921457746e-05,
+      "loss": 0.6752,
+      "step": 232800
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 6.929643516070198e-05,
+      "loss": 0.6797,
+      "step": 233000
+    },
+    {
+      "epoch": 0.31,
+      "eval_em_first_line_ratio": 0.508,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3953,
+      "eval_loss": 0.6787919402122498,
+      "eval_runtime": 235.6691,
+      "eval_samples_per_second": 42.432,
+      "eval_steps_per_second": 0.225,
+      "step": 233000
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 6.927005922715877e-05,
+      "loss": 0.6769,
+      "step": 233200
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 6.924368329361558e-05,
+      "loss": 0.6748,
+      "step": 233400
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 6.921730736007238e-05,
+      "loss": 0.6829,
+      "step": 233600
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 6.919093142652919e-05,
+      "loss": 0.6748,
+      "step": 233800
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 6.916455549298598e-05,
+      "loss": 0.6758,
+      "step": 234000
+    },
+    {
+      "epoch": 0.31,
+      "eval_em_first_line_ratio": 0.5047,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3917,
+      "eval_loss": 0.6798585057258606,
+      "eval_runtime": 225.4728,
+      "eval_samples_per_second": 44.351,
+      "eval_steps_per_second": 0.235,
+      "step": 234000
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 6.913817955944278e-05,
+      "loss": 0.6769,
+      "step": 234200
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 6.911180362589959e-05,
+      "loss": 0.67,
+      "step": 234400
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 6.908542769235639e-05,
+      "loss": 0.6789,
+      "step": 234600
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 6.905905175881318e-05,
+      "loss": 0.6756,
+      "step": 234800
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 6.903267582527e-05,
+      "loss": 0.6744,
+      "step": 235000
+    },
+    {
+      "epoch": 0.31,
+      "eval_em_first_line_ratio": 0.5061,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3927,
+      "eval_loss": 0.678789496421814,
+      "eval_runtime": 225.1722,
+      "eval_samples_per_second": 44.41,
+      "eval_steps_per_second": 0.235,
+      "step": 235000
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 6.90062998917268e-05,
+      "loss": 0.6733,
+      "step": 235200
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 6.89799239581836e-05,
+      "loss": 0.6785,
+      "step": 235400
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 6.89535480246404e-05,
+      "loss": 0.68,
+      "step": 235600
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 6.89271720910972e-05,
+      "loss": 0.6794,
+      "step": 235800
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 6.890079615755401e-05,
+      "loss": 0.6739,
+      "step": 236000
+    },
+    {
+      "epoch": 0.31,
+      "eval_em_first_line_ratio": 0.5063,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3929,
+      "eval_loss": 0.6798728108406067,
+      "eval_runtime": 234.3418,
+      "eval_samples_per_second": 42.673,
+      "eval_steps_per_second": 0.226,
+      "step": 236000
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 6.88744202240108e-05,
+      "loss": 0.6721,
+      "step": 236200
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 6.88480442904676e-05,
+      "loss": 0.6736,
+      "step": 236400
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 6.882166835692441e-05,
+      "loss": 0.6734,
+      "step": 236600
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 6.879529242338121e-05,
+      "loss": 0.6809,
+      "step": 236800
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 6.876891648983802e-05,
+      "loss": 0.6761,
+      "step": 237000
+    },
+    {
+      "epoch": 0.31,
+      "eval_em_first_line_ratio": 0.5057,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3932,
+      "eval_loss": 0.6787990927696228,
+      "eval_runtime": 226.6748,
+      "eval_samples_per_second": 44.116,
+      "eval_steps_per_second": 0.234,
+      "step": 237000
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 6.874254055629482e-05,
+      "loss": 0.673,
+      "step": 237200
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 6.871616462275163e-05,
+      "loss": 0.6763,
+      "step": 237400
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 6.868978868920842e-05,
+      "loss": 0.6748,
+      "step": 237600
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 6.866341275566522e-05,
+      "loss": 0.6794,
+      "step": 237800
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 6.863703682212203e-05,
+      "loss": 0.677,
+      "step": 238000
+    },
+    {
+      "epoch": 0.31,
+      "eval_em_first_line_ratio": 0.5048,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.392,
+      "eval_loss": 0.6788815259933472,
+      "eval_runtime": 226.4494,
+      "eval_samples_per_second": 44.16,
+      "eval_steps_per_second": 0.234,
+      "step": 238000
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 6.861079276824655e-05,
+      "loss": 0.6714,
+      "step": 238200
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 6.858441683470334e-05,
+      "loss": 0.6772,
+      "step": 238400
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 6.855804090116015e-05,
+      "loss": 0.6787,
+      "step": 238600
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 6.853166496761695e-05,
+      "loss": 0.679,
+      "step": 238800
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 6.850528903407374e-05,
+      "loss": 0.6757,
+      "step": 239000
+    },
+    {
+      "epoch": 0.32,
+      "eval_em_first_line_ratio": 0.5041,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3913,
+      "eval_loss": 0.6794099807739258,
+      "eval_runtime": 232.5831,
+      "eval_samples_per_second": 42.995,
+      "eval_steps_per_second": 0.228,
+      "step": 239000
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 6.847891310053055e-05,
+      "loss": 0.6781,
+      "step": 239200
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 6.845253716698737e-05,
+      "loss": 0.6758,
+      "step": 239400
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 6.842616123344417e-05,
+      "loss": 0.6775,
+      "step": 239600
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 6.839978529990096e-05,
+      "loss": 0.6825,
+      "step": 239800
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 6.837340936635776e-05,
+      "loss": 0.6748,
+      "step": 240000
+    },
+    {
+      "epoch": 0.32,
+      "eval_em_first_line_ratio": 0.5055,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3926,
+      "eval_loss": 0.6787717938423157,
+      "eval_runtime": 232.6463,
+      "eval_samples_per_second": 42.984,
+      "eval_steps_per_second": 0.228,
+      "step": 240000
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 6.834703343281457e-05,
+      "loss": 0.6692,
+      "step": 240200
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 6.832065749927137e-05,
+      "loss": 0.6777,
+      "step": 240400
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 6.829428156572816e-05,
+      "loss": 0.6796,
+      "step": 240600
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 6.826803751185269e-05,
+      "loss": 0.6711,
+      "step": 240800
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 6.82416615783095e-05,
+      "loss": 0.6726,
+      "step": 241000
+    },
+    {
+      "epoch": 0.32,
+      "eval_em_first_line_ratio": 0.5046,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3917,
+      "eval_loss": 0.6775960326194763,
+      "eval_runtime": 229.0399,
+      "eval_samples_per_second": 43.661,
+      "eval_steps_per_second": 0.231,
+      "step": 241000
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 6.821528564476629e-05,
+      "loss": 0.6735,
+      "step": 241200
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 6.818890971122309e-05,
+      "loss": 0.6795,
+      "step": 241400
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 6.81625337776799e-05,
+      "loss": 0.6751,
+      "step": 241600
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 6.81361578441367e-05,
+      "loss": 0.6772,
+      "step": 241800
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 6.81097819105935e-05,
+      "loss": 0.6759,
+      "step": 242000
+    },
+    {
+      "epoch": 0.32,
+      "eval_em_first_line_ratio": 0.5054,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3926,
+      "eval_loss": 0.6780937314033508,
+      "eval_runtime": 226.909,
+      "eval_samples_per_second": 44.071,
+      "eval_steps_per_second": 0.234,
+      "step": 242000
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 6.808340597705031e-05,
+      "loss": 0.6769,
+      "step": 242200
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 6.805716192317482e-05,
+      "loss": 0.6784,
+      "step": 242400
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 6.803078598963163e-05,
+      "loss": 0.6782,
+      "step": 242600
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 6.800441005608843e-05,
+      "loss": 0.6785,
+      "step": 242800
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 6.797803412254524e-05,
+      "loss": 0.6741,
+      "step": 243000
+    },
+    {
+      "epoch": 0.32,
+      "eval_em_first_line_ratio": 0.5073,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3941,
+      "eval_loss": 0.6774539351463318,
+      "eval_runtime": 231.5814,
+      "eval_samples_per_second": 43.181,
+      "eval_steps_per_second": 0.229,
+      "step": 243000
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 6.795165818900203e-05,
+      "loss": 0.677,
+      "step": 243200
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 6.792528225545883e-05,
+      "loss": 0.6764,
+      "step": 243400
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 6.789890632191564e-05,
+      "loss": 0.6754,
+      "step": 243600
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 6.787253038837244e-05,
+      "loss": 0.6695,
+      "step": 243800
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 6.784615445482923e-05,
+      "loss": 0.6756,
+      "step": 244000
+    },
+    {
+      "epoch": 0.32,
+      "eval_em_first_line_ratio": 0.5061,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.393,
+      "eval_loss": 0.6761301755905151,
+      "eval_runtime": 233.8598,
+      "eval_samples_per_second": 42.761,
+      "eval_steps_per_second": 0.227,
+      "step": 244000
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 6.781977852128603e-05,
+      "loss": 0.6779,
+      "step": 244200
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 6.779340258774284e-05,
+      "loss": 0.6779,
+      "step": 244400
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 6.776702665419964e-05,
+      "loss": 0.6755,
+      "step": 244600
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 6.774065072065645e-05,
+      "loss": 0.6798,
+      "step": 244800
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 6.771427478711325e-05,
+      "loss": 0.671,
+      "step": 245000
+    },
+    {
+      "epoch": 0.32,
+      "eval_em_first_line_ratio": 0.5061,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.393,
+      "eval_loss": 0.6761966943740845,
+      "eval_runtime": 232.0614,
+      "eval_samples_per_second": 43.092,
+      "eval_steps_per_second": 0.228,
+      "step": 245000
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 6.768789885357006e-05,
+      "loss": 0.6727,
+      "step": 245200
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 6.766152292002685e-05,
+      "loss": 0.6756,
+      "step": 245400
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 6.763514698648365e-05,
+      "loss": 0.6741,
+      "step": 245600
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 6.760877105294046e-05,
+      "loss": 0.6755,
+      "step": 245800
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 6.758239511939726e-05,
+      "loss": 0.6733,
+      "step": 246000
+    },
+    {
+      "epoch": 0.32,
+      "eval_em_first_line_ratio": 0.5085,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3953,
+      "eval_loss": 0.6763582229614258,
+      "eval_runtime": 226.0313,
+      "eval_samples_per_second": 44.242,
+      "eval_steps_per_second": 0.234,
+      "step": 246000
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 6.755601918585405e-05,
+      "loss": 0.6757,
+      "step": 246200
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 6.752964325231087e-05,
+      "loss": 0.6713,
+      "step": 246400
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 6.750339919843538e-05,
+      "loss": 0.6772,
+      "step": 246600
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 6.747702326489217e-05,
+      "loss": 0.6808,
+      "step": 246800
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 6.745064733134898e-05,
+      "loss": 0.6705,
+      "step": 247000
+    },
+    {
+      "epoch": 0.33,
+      "eval_em_first_line_ratio": 0.506,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3932,
+      "eval_loss": 0.6762988567352295,
+      "eval_runtime": 230.5307,
+      "eval_samples_per_second": 43.378,
+      "eval_steps_per_second": 0.23,
+      "step": 247000
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 6.74242713978058e-05,
+      "loss": 0.6772,
+      "step": 247200
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 6.73978954642626e-05,
+      "loss": 0.676,
+      "step": 247400
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 6.737151953071939e-05,
+      "loss": 0.6752,
+      "step": 247600
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 6.73451435971762e-05,
+      "loss": 0.6741,
+      "step": 247800
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 6.7318767663633e-05,
+      "loss": 0.6753,
+      "step": 248000
+    },
+    {
+      "epoch": 0.33,
+      "eval_em_first_line_ratio": 0.5094,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3964,
+      "eval_loss": 0.6757665872573853,
+      "eval_runtime": 231.6364,
+      "eval_samples_per_second": 43.171,
+      "eval_steps_per_second": 0.229,
+      "step": 248000
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 6.72923917300898e-05,
+      "loss": 0.6726,
+      "step": 248200
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 6.72660157965466e-05,
+      "loss": 0.6725,
+      "step": 248400
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 6.72396398630034e-05,
+      "loss": 0.6708,
+      "step": 248600
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 6.72132639294602e-05,
+      "loss": 0.67,
+      "step": 248800
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 6.718688799591701e-05,
+      "loss": 0.6734,
+      "step": 249000
+    },
+    {
+      "epoch": 0.33,
+      "eval_em_first_line_ratio": 0.5089,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3946,
+      "eval_loss": 0.6757145524024963,
+      "eval_runtime": 225.3153,
+      "eval_samples_per_second": 44.382,
+      "eval_steps_per_second": 0.235,
+      "step": 249000
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 6.716051206237381e-05,
+      "loss": 0.6777,
+      "step": 249200
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 6.713426800849833e-05,
+      "loss": 0.6714,
+      "step": 249400
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 6.710789207495513e-05,
+      "loss": 0.6736,
+      "step": 249600
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 6.708151614141194e-05,
+      "loss": 0.6714,
+      "step": 249800
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 6.705514020786874e-05,
+      "loss": 0.6757,
+      "step": 250000
+    },
+    {
+      "epoch": 0.33,
+      "eval_em_first_line_ratio": 0.5073,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3941,
+      "eval_loss": 0.675456702709198,
+      "eval_runtime": 235.1507,
+      "eval_samples_per_second": 42.526,
+      "eval_steps_per_second": 0.225,
+      "step": 250000
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 6.702876427432554e-05,
+      "loss": 0.6697,
+      "step": 250200
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 6.700238834078234e-05,
+      "loss": 0.6736,
+      "step": 250400
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 6.697601240723914e-05,
+      "loss": 0.6679,
+      "step": 250600
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 6.694963647369594e-05,
+      "loss": 0.6739,
+      "step": 250800
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 6.692326054015275e-05,
+      "loss": 0.6732,
+      "step": 251000
+    },
+    {
+      "epoch": 0.33,
+      "eval_em_first_line_ratio": 0.5101,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.397,
+      "eval_loss": 0.6755709052085876,
+      "eval_runtime": 226.2824,
+      "eval_samples_per_second": 44.193,
+      "eval_steps_per_second": 0.234,
+      "step": 251000
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 6.689688460660954e-05,
+      "loss": 0.6768,
+      "step": 251200
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 6.687050867306634e-05,
+      "loss": 0.6778,
+      "step": 251400
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 6.684413273952316e-05,
+      "loss": 0.6748,
+      "step": 251600
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 6.681788868564766e-05,
+      "loss": 0.6752,
+      "step": 251800
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 6.679151275210447e-05,
+      "loss": 0.6766,
+      "step": 252000
+    },
+    {
+      "epoch": 0.33,
+      "eval_em_first_line_ratio": 0.5072,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3943,
+      "eval_loss": 0.6756595969200134,
+      "eval_runtime": 226.4228,
+      "eval_samples_per_second": 44.165,
+      "eval_steps_per_second": 0.234,
+      "step": 252000
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 6.676513681856128e-05,
+      "loss": 0.6701,
+      "step": 252200
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 6.673876088501807e-05,
+      "loss": 0.67,
+      "step": 252400
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 6.671238495147488e-05,
+      "loss": 0.6762,
+      "step": 252600
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 6.668600901793168e-05,
+      "loss": 0.6748,
+      "step": 252800
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 6.665963308438849e-05,
+      "loss": 0.6711,
+      "step": 253000
+    },
+    {
+      "epoch": 0.33,
+      "eval_em_first_line_ratio": 0.5109,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3972,
+      "eval_loss": 0.6754322648048401,
+      "eval_runtime": 228.5309,
+      "eval_samples_per_second": 43.758,
+      "eval_steps_per_second": 0.232,
+      "step": 253000
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 6.663325715084528e-05,
+      "loss": 0.6683,
+      "step": 253200
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 6.660688121730208e-05,
+      "loss": 0.6713,
+      "step": 253400
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 6.658050528375889e-05,
+      "loss": 0.6693,
+      "step": 253600
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 6.655412935021569e-05,
+      "loss": 0.6721,
+      "step": 253800
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 6.652775341667248e-05,
+      "loss": 0.6668,
+      "step": 254000
+    },
+    {
+      "epoch": 0.33,
+      "eval_em_first_line_ratio": 0.5107,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3967,
+      "eval_loss": 0.6752244830131531,
+      "eval_runtime": 230.872,
+      "eval_samples_per_second": 43.314,
+      "eval_steps_per_second": 0.23,
+      "step": 254000
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 6.65013774831293e-05,
+      "loss": 0.6642,
+      "step": 254200
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 6.64750015495861e-05,
+      "loss": 0.6705,
+      "step": 254400
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 6.644862561604291e-05,
+      "loss": 0.6714,
+      "step": 254600
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 6.64222496824997e-05,
+      "loss": 0.6682,
+      "step": 254800
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 6.63958737489565e-05,
+      "loss": 0.6727,
+      "step": 255000
+    },
+    {
+      "epoch": 0.34,
+      "eval_em_first_line_ratio": 0.5068,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3936,
+      "eval_loss": 0.674951434135437,
+      "eval_runtime": 234.5733,
+      "eval_samples_per_second": 42.631,
+      "eval_steps_per_second": 0.226,
+      "step": 255000
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 6.636949781541331e-05,
+      "loss": 0.6689,
+      "step": 255200
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 6.634312188187011e-05,
+      "loss": 0.6757,
+      "step": 255400
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 6.63167459483269e-05,
+      "loss": 0.6719,
+      "step": 255600
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 6.629037001478371e-05,
+      "loss": 0.6687,
+      "step": 255800
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 6.626399408124053e-05,
+      "loss": 0.6722,
+      "step": 256000
+    },
+    {
+      "epoch": 0.34,
+      "eval_em_first_line_ratio": 0.5064,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3928,
+      "eval_loss": 0.6751517653465271,
+      "eval_runtime": 226.209,
+      "eval_samples_per_second": 44.207,
+      "eval_steps_per_second": 0.234,
+      "step": 256000
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 6.623775002736503e-05,
+      "loss": 0.6736,
+      "step": 256200
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 6.621150597348955e-05,
+      "loss": 0.6785,
+      "step": 256400
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 6.618513003994636e-05,
+      "loss": 0.6658,
+      "step": 256600
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 6.615875410640315e-05,
+      "loss": 0.6698,
+      "step": 256800
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 6.613237817285995e-05,
+      "loss": 0.6712,
+      "step": 257000
+    },
+    {
+      "epoch": 0.34,
+      "eval_em_first_line_ratio": 0.5043,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3914,
+      "eval_loss": 0.6752673983573914,
+      "eval_runtime": 233.1159,
+      "eval_samples_per_second": 42.897,
+      "eval_steps_per_second": 0.227,
+      "step": 257000
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 6.610600223931676e-05,
+      "loss": 0.6711,
+      "step": 257200
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 6.607962630577356e-05,
+      "loss": 0.6737,
+      "step": 257400
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 6.605325037223037e-05,
+      "loss": 0.6733,
+      "step": 257600
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 6.602687443868717e-05,
+      "loss": 0.67,
+      "step": 257800
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 6.600049850514398e-05,
+      "loss": 0.6717,
+      "step": 258000
+    },
+    {
+      "epoch": 0.34,
+      "eval_em_first_line_ratio": 0.5057,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3924,
+      "eval_loss": 0.6736807823181152,
+      "eval_runtime": 233.7985,
+      "eval_samples_per_second": 42.772,
+      "eval_steps_per_second": 0.227,
+      "step": 258000
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 6.597412257160077e-05,
+      "loss": 0.6711,
+      "step": 258200
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 6.594774663805757e-05,
+      "loss": 0.6696,
+      "step": 258400
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 6.59215025841821e-05,
+      "loss": 0.6661,
+      "step": 258600
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 6.589512665063889e-05,
+      "loss": 0.6694,
+      "step": 258800
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 6.586875071709569e-05,
+      "loss": 0.6743,
+      "step": 259000
+    },
+    {
+      "epoch": 0.34,
+      "eval_em_first_line_ratio": 0.5091,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3959,
+      "eval_loss": 0.6738882064819336,
+      "eval_runtime": 232.7683,
+      "eval_samples_per_second": 42.961,
+      "eval_steps_per_second": 0.228,
+      "step": 259000
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 6.58423747835525e-05,
+      "loss": 0.6754,
+      "step": 259200
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 6.58159988500093e-05,
+      "loss": 0.6672,
+      "step": 259400
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 6.578962291646609e-05,
+      "loss": 0.6725,
+      "step": 259600
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 6.576324698292291e-05,
+      "loss": 0.6738,
+      "step": 259800
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 6.573687104937971e-05,
+      "loss": 0.6687,
+      "step": 260000
+    },
+    {
+      "epoch": 0.34,
+      "eval_em_first_line_ratio": 0.5075,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3942,
+      "eval_loss": 0.673912763595581,
+      "eval_runtime": 232.8214,
+      "eval_samples_per_second": 42.951,
+      "eval_steps_per_second": 0.228,
+      "step": 260000
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 6.57104951158365e-05,
+      "loss": 0.6753,
+      "step": 260200
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 6.568411918229331e-05,
+      "loss": 0.6726,
+      "step": 260400
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 6.565774324875011e-05,
+      "loss": 0.6705,
+      "step": 260600
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 6.563136731520692e-05,
+      "loss": 0.6746,
+      "step": 260800
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 6.560499138166371e-05,
+      "loss": 0.6746,
+      "step": 261000
+    },
+    {
+      "epoch": 0.34,
+      "eval_em_first_line_ratio": 0.5106,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3968,
+      "eval_loss": 0.6739708781242371,
+      "eval_runtime": 230.1243,
+      "eval_samples_per_second": 43.455,
+      "eval_steps_per_second": 0.23,
+      "step": 261000
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 6.557861544812051e-05,
+      "loss": 0.6685,
+      "step": 261200
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 6.555223951457732e-05,
+      "loss": 0.6723,
+      "step": 261400
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 6.552586358103412e-05,
+      "loss": 0.6654,
+      "step": 261600
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 6.549948764749093e-05,
+      "loss": 0.6689,
+      "step": 261800
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 6.547311171394773e-05,
+      "loss": 0.669,
+      "step": 262000
+    },
+    {
+      "epoch": 0.35,
+      "eval_em_first_line_ratio": 0.5063,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3926,
+      "eval_loss": 0.6738796830177307,
+      "eval_runtime": 226.7017,
+      "eval_samples_per_second": 44.111,
+      "eval_steps_per_second": 0.234,
+      "step": 262000
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 6.544673578040454e-05,
+      "loss": 0.6739,
+      "step": 262200
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 6.542035984686134e-05,
+      "loss": 0.6678,
+      "step": 262400
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 6.539411579298585e-05,
+      "loss": 0.6693,
+      "step": 262600
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 6.536773985944266e-05,
+      "loss": 0.6678,
+      "step": 262800
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 6.534136392589946e-05,
+      "loss": 0.668,
+      "step": 263000
+    },
+    {
+      "epoch": 0.35,
+      "eval_em_first_line_ratio": 0.51,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3963,
+      "eval_loss": 0.6740802526473999,
+      "eval_runtime": 230.3849,
+      "eval_samples_per_second": 43.406,
+      "eval_steps_per_second": 0.23,
+      "step": 263000
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 6.531498799235625e-05,
+      "loss": 0.6721,
+      "step": 263200
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 6.528861205881306e-05,
+      "loss": 0.6669,
+      "step": 263400
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 6.526223612526986e-05,
+      "loss": 0.6653,
+      "step": 263600
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 6.523586019172667e-05,
+      "loss": 0.6678,
+      "step": 263800
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 6.520948425818346e-05,
+      "loss": 0.6631,
+      "step": 264000
+    },
+    {
+      "epoch": 0.35,
+      "eval_em_first_line_ratio": 0.5065,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3934,
+      "eval_loss": 0.6737580299377441,
+      "eval_runtime": 228.185,
+      "eval_samples_per_second": 43.824,
+      "eval_steps_per_second": 0.232,
+      "step": 264000
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 6.518310832464026e-05,
+      "loss": 0.6665,
+      "step": 264200
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 6.515673239109708e-05,
+      "loss": 0.6731,
+      "step": 264400
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 6.513035645755387e-05,
+      "loss": 0.6679,
+      "step": 264600
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 6.510398052401068e-05,
+      "loss": 0.6674,
+      "step": 264800
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 6.507760459046748e-05,
+      "loss": 0.6715,
+      "step": 265000
+    },
+    {
+      "epoch": 0.35,
+      "eval_em_first_line_ratio": 0.5102,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3968,
+      "eval_loss": 0.674156129360199,
+      "eval_runtime": 234.0593,
+      "eval_samples_per_second": 42.724,
+      "eval_steps_per_second": 0.226,
+      "step": 265000
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 6.505162429592742e-05,
+      "loss": 0.6708,
+      "step": 265200
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 6.502524836238423e-05,
+      "loss": 0.669,
+      "step": 265400
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 6.499887242884103e-05,
+      "loss": 0.6662,
+      "step": 265600
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 6.497249649529782e-05,
+      "loss": 0.6709,
+      "step": 265800
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 6.494612056175463e-05,
+      "loss": 0.6713,
+      "step": 266000
+    },
+    {
+      "epoch": 0.35,
+      "eval_em_first_line_ratio": 0.51,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3961,
+      "eval_loss": 0.6731346845626831,
+      "eval_runtime": 226.6861,
+      "eval_samples_per_second": 44.114,
+      "eval_steps_per_second": 0.234,
+      "step": 266000
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 6.491974462821145e-05,
+      "loss": 0.6699,
+      "step": 266200
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 6.489336869466824e-05,
+      "loss": 0.6683,
+      "step": 266400
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 6.486699276112504e-05,
+      "loss": 0.6687,
+      "step": 266600
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 6.484061682758185e-05,
+      "loss": 0.6634,
+      "step": 266800
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 6.481424089403865e-05,
+      "loss": 0.6669,
+      "step": 267000
+    },
+    {
+      "epoch": 0.35,
+      "eval_em_first_line_ratio": 0.5047,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3913,
+      "eval_loss": 0.6731341481208801,
+      "eval_runtime": 231.947,
+      "eval_samples_per_second": 43.113,
+      "eval_steps_per_second": 0.229,
+      "step": 267000
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 6.478786496049544e-05,
+      "loss": 0.6616,
+      "step": 267200
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 6.476148902695225e-05,
+      "loss": 0.6701,
+      "step": 267400
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 6.473511309340905e-05,
+      "loss": 0.6679,
+      "step": 267600
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 6.470873715986585e-05,
+      "loss": 0.6706,
+      "step": 267800
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 6.468236122632265e-05,
+      "loss": 0.6703,
+      "step": 268000
+    },
+    {
+      "epoch": 0.35,
+      "eval_em_first_line_ratio": 0.5089,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.395,
+      "eval_loss": 0.6730291247367859,
+      "eval_runtime": 233.3394,
+      "eval_samples_per_second": 42.856,
+      "eval_steps_per_second": 0.227,
+      "step": 268000
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 6.465598529277946e-05,
+      "loss": 0.67,
+      "step": 268200
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 6.462960935923627e-05,
+      "loss": 0.6659,
+      "step": 268400
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 6.460323342569307e-05,
+      "loss": 0.6653,
+      "step": 268600
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 6.457685749214986e-05,
+      "loss": 0.6684,
+      "step": 268800
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 6.455048155860667e-05,
+      "loss": 0.6679,
+      "step": 269000
+    },
+    {
+      "epoch": 0.35,
+      "eval_em_first_line_ratio": 0.5072,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3938,
+      "eval_loss": 0.6721344590187073,
+      "eval_runtime": 225.8951,
+      "eval_samples_per_second": 44.268,
+      "eval_steps_per_second": 0.235,
+      "step": 269000
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 6.452410562506347e-05,
+      "loss": 0.6632,
+      "step": 269200
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 6.449772969152028e-05,
+      "loss": 0.6654,
+      "step": 269400
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 6.447135375797707e-05,
+      "loss": 0.6664,
+      "step": 269600
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 6.444497782443387e-05,
+      "loss": 0.6658,
+      "step": 269800
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 6.441860189089069e-05,
+      "loss": 0.6659,
+      "step": 270000
+    },
+    {
+      "epoch": 0.36,
+      "eval_em_first_line_ratio": 0.5106,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3976,
+      "eval_loss": 0.6728203892707825,
+      "eval_runtime": 233.5639,
+      "eval_samples_per_second": 42.815,
+      "eval_steps_per_second": 0.227,
+      "step": 270000
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 6.439222595734748e-05,
+      "loss": 0.6651,
+      "step": 270200
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 6.436585002380429e-05,
+      "loss": 0.6718,
+      "step": 270400
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 6.433947409026109e-05,
+      "loss": 0.6666,
+      "step": 270600
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 6.43132300363856e-05,
+      "loss": 0.6714,
+      "step": 270800
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 6.428685410284241e-05,
+      "loss": 0.6669,
+      "step": 271000
+    },
+    {
+      "epoch": 0.36,
+      "eval_em_first_line_ratio": 0.511,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3975,
+      "eval_loss": 0.6718416213989258,
+      "eval_runtime": 230.1273,
+      "eval_samples_per_second": 43.454,
+      "eval_steps_per_second": 0.23,
+      "step": 271000
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 6.426047816929921e-05,
+      "loss": 0.6674,
+      "step": 271200
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 6.423410223575602e-05,
+      "loss": 0.6663,
+      "step": 271400
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 6.420772630221281e-05,
+      "loss": 0.67,
+      "step": 271600
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 6.418135036866961e-05,
+      "loss": 0.6661,
+      "step": 271800
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 6.415497443512642e-05,
+      "loss": 0.6725,
+      "step": 272000
+    },
+    {
+      "epoch": 0.36,
+      "eval_em_first_line_ratio": 0.5084,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3945,
+      "eval_loss": 0.6723572015762329,
+      "eval_runtime": 234.2513,
+      "eval_samples_per_second": 42.689,
+      "eval_steps_per_second": 0.226,
+      "step": 272000
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 6.412859850158322e-05,
+      "loss": 0.6709,
+      "step": 272200
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 6.410222256804001e-05,
+      "loss": 0.6728,
+      "step": 272400
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 6.407584663449683e-05,
+      "loss": 0.6675,
+      "step": 272600
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 6.404947070095363e-05,
+      "loss": 0.6731,
+      "step": 272800
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 6.402309476741042e-05,
+      "loss": 0.6697,
+      "step": 273000
+    },
+    {
+      "epoch": 0.36,
+      "eval_em_first_line_ratio": 0.5091,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3956,
+      "eval_loss": 0.6715953946113586,
+      "eval_runtime": 231.8048,
+      "eval_samples_per_second": 43.14,
+      "eval_steps_per_second": 0.229,
+      "step": 273000
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 6.399671883386723e-05,
+      "loss": 0.6612,
+      "step": 273200
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 6.397034290032403e-05,
+      "loss": 0.6651,
+      "step": 273400
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 6.394396696678084e-05,
+      "loss": 0.6668,
+      "step": 273600
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 6.391759103323763e-05,
+      "loss": 0.6665,
+      "step": 273800
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 6.389121509969443e-05,
+      "loss": 0.6672,
+      "step": 274000
+    },
+    {
+      "epoch": 0.36,
+      "eval_em_first_line_ratio": 0.5105,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.397,
+      "eval_loss": 0.6707741618156433,
+      "eval_runtime": 230.6702,
+      "eval_samples_per_second": 43.352,
+      "eval_steps_per_second": 0.23,
+      "step": 274000
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 6.386483916615124e-05,
+      "loss": 0.668,
+      "step": 274200
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 6.383846323260804e-05,
+      "loss": 0.6658,
+      "step": 274400
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 6.381208729906485e-05,
+      "loss": 0.6641,
+      "step": 274600
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 6.378571136552165e-05,
+      "loss": 0.6685,
+      "step": 274800
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 6.375946731164616e-05,
+      "loss": 0.6646,
+      "step": 275000
+    },
+    {
+      "epoch": 0.36,
+      "eval_em_first_line_ratio": 0.509,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3957,
+      "eval_loss": 0.6714920997619629,
+      "eval_runtime": 232.1808,
+      "eval_samples_per_second": 43.07,
+      "eval_steps_per_second": 0.228,
+      "step": 275000
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 6.373309137810297e-05,
+      "loss": 0.6646,
+      "step": 275200
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 6.370684732422748e-05,
+      "loss": 0.6655,
+      "step": 275400
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 6.368047139068429e-05,
+      "loss": 0.6666,
+      "step": 275600
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 6.365409545714109e-05,
+      "loss": 0.6736,
+      "step": 275800
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 6.36277195235979e-05,
+      "loss": 0.6659,
+      "step": 276000
+    },
+    {
+      "epoch": 0.36,
+      "eval_em_first_line_ratio": 0.5082,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3953,
+      "eval_loss": 0.6707987785339355,
+      "eval_runtime": 228.547,
+      "eval_samples_per_second": 43.755,
+      "eval_steps_per_second": 0.232,
+      "step": 276000
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 6.36013435900547e-05,
+      "loss": 0.6657,
+      "step": 276200
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 6.35749676565115e-05,
+      "loss": 0.6608,
+      "step": 276400
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 6.35485917229683e-05,
+      "loss": 0.6664,
+      "step": 276600
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 6.35222157894251e-05,
+      "loss": 0.6639,
+      "step": 276800
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 6.34958398558819e-05,
+      "loss": 0.668,
+      "step": 277000
+    },
+    {
+      "epoch": 0.37,
+      "eval_em_first_line_ratio": 0.5064,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3939,
+      "eval_loss": 0.6707022190093994,
+      "eval_runtime": 231.0922,
+      "eval_samples_per_second": 43.273,
+      "eval_steps_per_second": 0.229,
+      "step": 277000
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 6.346946392233871e-05,
+      "loss": 0.6696,
+      "step": 277200
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 6.34430879887955e-05,
+      "loss": 0.6676,
+      "step": 277400
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 6.34167120552523e-05,
+      "loss": 0.6678,
+      "step": 277600
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 6.339033612170912e-05,
+      "loss": 0.674,
+      "step": 277800
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 6.336396018816591e-05,
+      "loss": 0.6642,
+      "step": 278000
+    },
+    {
+      "epoch": 0.37,
+      "eval_em_first_line_ratio": 0.5079,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3951,
+      "eval_loss": 0.6701758503913879,
+      "eval_runtime": 229.4149,
+      "eval_samples_per_second": 43.589,
+      "eval_steps_per_second": 0.231,
+      "step": 278000
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 6.333758425462272e-05,
+      "loss": 0.6657,
+      "step": 278200
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 6.331120832107952e-05,
+      "loss": 0.6664,
+      "step": 278400
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 6.328483238753633e-05,
+      "loss": 0.6711,
+      "step": 278600
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 6.325845645399312e-05,
+      "loss": 0.668,
+      "step": 278800
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 6.323208052044992e-05,
+      "loss": 0.6671,
+      "step": 279000
+    },
+    {
+      "epoch": 0.37,
+      "eval_em_first_line_ratio": 0.51,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3962,
+      "eval_loss": 0.6700901389122009,
+      "eval_runtime": 228.9128,
+      "eval_samples_per_second": 43.685,
+      "eval_steps_per_second": 0.232,
+      "step": 279000
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 6.320570458690672e-05,
+      "loss": 0.665,
+      "step": 279200
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 6.317932865336353e-05,
+      "loss": 0.665,
+      "step": 279400
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 6.315308459948804e-05,
+      "loss": 0.6667,
+      "step": 279600
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 6.312670866594485e-05,
+      "loss": 0.6663,
+      "step": 279800
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 6.310046461206936e-05,
+      "loss": 0.662,
+      "step": 280000
+    },
+    {
+      "epoch": 0.37,
+      "eval_em_first_line_ratio": 0.5098,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3961,
+      "eval_loss": 0.6700763702392578,
+      "eval_runtime": 228.9532,
+      "eval_samples_per_second": 43.677,
+      "eval_steps_per_second": 0.231,
+      "step": 280000
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 6.307408867852616e-05,
+      "loss": 0.6639,
+      "step": 280200
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 6.304784462465069e-05,
+      "loss": 0.6668,
+      "step": 280400
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 6.302146869110748e-05,
+      "loss": 0.6671,
+      "step": 280600
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 6.299509275756429e-05,
+      "loss": 0.667,
+      "step": 280800
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 6.296871682402109e-05,
+      "loss": 0.6641,
+      "step": 281000
+    },
+    {
+      "epoch": 0.37,
+      "eval_em_first_line_ratio": 0.5087,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3954,
+      "eval_loss": 0.6698007583618164,
+      "eval_runtime": 227.8592,
+      "eval_samples_per_second": 43.887,
+      "eval_steps_per_second": 0.233,
+      "step": 281000
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 6.29423408904779e-05,
+      "loss": 0.6622,
+      "step": 281200
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 6.291596495693469e-05,
+      "loss": 0.6673,
+      "step": 281400
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 6.28895890233915e-05,
+      "loss": 0.6631,
+      "step": 281600
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 6.286321308984831e-05,
+      "loss": 0.6683,
+      "step": 281800
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 6.28368371563051e-05,
+      "loss": 0.6632,
+      "step": 282000
+    },
+    {
+      "epoch": 0.37,
+      "eval_em_first_line_ratio": 0.5106,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3978,
+      "eval_loss": 0.6697657108306885,
+      "eval_runtime": 222.5723,
+      "eval_samples_per_second": 44.929,
+      "eval_steps_per_second": 0.238,
+      "step": 282000
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 6.281059310242963e-05,
+      "loss": 0.6629,
+      "step": 282200
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 6.278421716888643e-05,
+      "loss": 0.6623,
+      "step": 282400
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 6.275784123534324e-05,
+      "loss": 0.6599,
+      "step": 282600
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 6.273146530180003e-05,
+      "loss": 0.6645,
+      "step": 282800
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 6.270508936825683e-05,
+      "loss": 0.6668,
+      "step": 283000
+    },
+    {
+      "epoch": 0.37,
+      "eval_em_first_line_ratio": 0.5104,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3972,
+      "eval_loss": 0.670370876789093,
+      "eval_runtime": 233.032,
+      "eval_samples_per_second": 42.913,
+      "eval_steps_per_second": 0.227,
+      "step": 283000
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 6.267871343471363e-05,
+      "loss": 0.664,
+      "step": 283200
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 6.265246938083815e-05,
+      "loss": 0.6658,
+      "step": 283400
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 6.262609344729495e-05,
+      "loss": 0.6658,
+      "step": 283600
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 6.259971751375176e-05,
+      "loss": 0.6672,
+      "step": 283800
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 6.257334158020856e-05,
+      "loss": 0.6616,
+      "step": 284000
+    },
+    {
+      "epoch": 0.37,
+      "eval_em_first_line_ratio": 0.5082,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3948,
+      "eval_loss": 0.6699286699295044,
+      "eval_runtime": 225.6266,
+      "eval_samples_per_second": 44.321,
+      "eval_steps_per_second": 0.235,
+      "step": 284000
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 6.254696564666535e-05,
+      "loss": 0.6642,
+      "step": 284200
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 6.252058971312216e-05,
+      "loss": 0.665,
+      "step": 284400
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 6.249421377957897e-05,
+      "loss": 0.6644,
+      "step": 284600
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 6.246783784603577e-05,
+      "loss": 0.6656,
+      "step": 284800
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 6.244146191249257e-05,
+      "loss": 0.6651,
+      "step": 285000
+    },
+    {
+      "epoch": 0.38,
+      "eval_em_first_line_ratio": 0.5105,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3963,
+      "eval_loss": 0.6698468923568726,
+      "eval_runtime": 232.1428,
+      "eval_samples_per_second": 43.077,
+      "eval_steps_per_second": 0.228,
+      "step": 285000
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 6.241508597894937e-05,
+      "loss": 0.6629,
+      "step": 285200
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 6.238871004540618e-05,
+      "loss": 0.6651,
+      "step": 285400
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 6.236233411186297e-05,
+      "loss": 0.6642,
+      "step": 285600
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 6.233595817831977e-05,
+      "loss": 0.6615,
+      "step": 285800
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 6.230958224477658e-05,
+      "loss": 0.6705,
+      "step": 286000
+    },
+    {
+      "epoch": 0.38,
+      "eval_em_first_line_ratio": 0.5112,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.397,
+      "eval_loss": 0.6694324016571045,
+      "eval_runtime": 225.4811,
+      "eval_samples_per_second": 44.35,
+      "eval_steps_per_second": 0.235,
+      "step": 286000
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 6.228320631123338e-05,
+      "loss": 0.6608,
+      "step": 286200
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 6.225683037769017e-05,
+      "loss": 0.668,
+      "step": 286400
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 6.223045444414699e-05,
+      "loss": 0.6666,
+      "step": 286600
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 6.22040785106038e-05,
+      "loss": 0.6666,
+      "step": 286800
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 6.217770257706059e-05,
+      "loss": 0.6614,
+      "step": 287000
+    },
+    {
+      "epoch": 0.38,
+      "eval_em_first_line_ratio": 0.5132,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4,
+      "eval_loss": 0.6688478589057922,
+      "eval_runtime": 232.5958,
+      "eval_samples_per_second": 42.993,
+      "eval_steps_per_second": 0.228,
+      "step": 287000
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 6.215132664351739e-05,
+      "loss": 0.6636,
+      "step": 287200
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 6.21249507099742e-05,
+      "loss": 0.6634,
+      "step": 287400
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 6.2098574776431e-05,
+      "loss": 0.6615,
+      "step": 287600
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 6.207219884288779e-05,
+      "loss": 0.6684,
+      "step": 287800
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 6.20458229093446e-05,
+      "loss": 0.6589,
+      "step": 288000
+    },
+    {
+      "epoch": 0.38,
+      "eval_em_first_line_ratio": 0.5101,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3967,
+      "eval_loss": 0.6695097088813782,
+      "eval_runtime": 224.1993,
+      "eval_samples_per_second": 44.603,
+      "eval_steps_per_second": 0.236,
+      "step": 288000
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 6.20194469758014e-05,
+      "loss": 0.6675,
+      "step": 288200
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 6.19930710422582e-05,
+      "loss": 0.6657,
+      "step": 288400
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 6.196669510871501e-05,
+      "loss": 0.6644,
+      "step": 288600
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 6.194031917517181e-05,
+      "loss": 0.6614,
+      "step": 288800
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 6.191394324162862e-05,
+      "loss": 0.6653,
+      "step": 289000
+    },
+    {
+      "epoch": 0.38,
+      "eval_em_first_line_ratio": 0.5124,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3987,
+      "eval_loss": 0.6679722666740417,
+      "eval_runtime": 227.665,
+      "eval_samples_per_second": 43.924,
+      "eval_steps_per_second": 0.233,
+      "step": 289000
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 6.188756730808541e-05,
+      "loss": 0.6647,
+      "step": 289200
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 6.186119137454221e-05,
+      "loss": 0.6641,
+      "step": 289400
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 6.183481544099902e-05,
+      "loss": 0.6636,
+      "step": 289600
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 6.180843950745582e-05,
+      "loss": 0.66,
+      "step": 289800
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 6.178206357391261e-05,
+      "loss": 0.6613,
+      "step": 290000
+    },
+    {
+      "epoch": 0.38,
+      "eval_em_first_line_ratio": 0.5114,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3976,
+      "eval_loss": 0.6675393581390381,
+      "eval_runtime": 218.6686,
+      "eval_samples_per_second": 45.731,
+      "eval_steps_per_second": 0.242,
+      "step": 290000
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 6.175568764036942e-05,
+      "loss": 0.6656,
+      "step": 290200
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 6.172931170682624e-05,
+      "loss": 0.6611,
+      "step": 290400
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 6.170293577328303e-05,
+      "loss": 0.6609,
+      "step": 290600
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 6.167655983973983e-05,
+      "loss": 0.6593,
+      "step": 290800
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 6.165018390619663e-05,
+      "loss": 0.6607,
+      "step": 291000
+    },
+    {
+      "epoch": 0.38,
+      "eval_em_first_line_ratio": 0.508,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3947,
+      "eval_loss": 0.6682593822479248,
+      "eval_runtime": 229.1235,
+      "eval_samples_per_second": 43.645,
+      "eval_steps_per_second": 0.231,
+      "step": 291000
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 6.162380797265344e-05,
+      "loss": 0.6576,
+      "step": 291200
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 6.159743203911023e-05,
+      "loss": 0.6637,
+      "step": 291400
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 6.157105610556703e-05,
+      "loss": 0.6685,
+      "step": 291600
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 6.154468017202384e-05,
+      "loss": 0.6661,
+      "step": 291800
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 6.151830423848064e-05,
+      "loss": 0.6604,
+      "step": 292000
+    },
+    {
+      "epoch": 0.39,
+      "eval_em_first_line_ratio": 0.5073,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3936,
+      "eval_loss": 0.6690237522125244,
+      "eval_runtime": 226.8151,
+      "eval_samples_per_second": 44.089,
+      "eval_steps_per_second": 0.234,
+      "step": 292000
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 6.149192830493743e-05,
+      "loss": 0.6628,
+      "step": 292200
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 6.146555237139425e-05,
+      "loss": 0.661,
+      "step": 292400
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 6.143917643785106e-05,
+      "loss": 0.6599,
+      "step": 292600
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 6.141280050430786e-05,
+      "loss": 0.6608,
+      "step": 292800
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 6.138642457076465e-05,
+      "loss": 0.6646,
+      "step": 293000
+    },
+    {
+      "epoch": 0.39,
+      "eval_em_first_line_ratio": 0.5064,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3925,
+      "eval_loss": 0.6679369211196899,
+      "eval_runtime": 232.3895,
+      "eval_samples_per_second": 43.031,
+      "eval_steps_per_second": 0.228,
+      "step": 293000
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 6.136004863722146e-05,
+      "loss": 0.6576,
+      "step": 293200
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 6.133367270367826e-05,
+      "loss": 0.6644,
+      "step": 293400
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 6.13075605294705e-05,
+      "loss": 0.6655,
+      "step": 293600
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 6.12811845959273e-05,
+      "loss": 0.6662,
+      "step": 293800
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 6.12548086623841e-05,
+      "loss": 0.6601,
+      "step": 294000
+    },
+    {
+      "epoch": 0.39,
+      "eval_em_first_line_ratio": 0.5078,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3944,
+      "eval_loss": 0.6683861017227173,
+      "eval_runtime": 225.9037,
+      "eval_samples_per_second": 44.267,
+      "eval_steps_per_second": 0.235,
+      "step": 294000
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 6.12284327288409e-05,
+      "loss": 0.6632,
+      "step": 294200
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 6.12020567952977e-05,
+      "loss": 0.6602,
+      "step": 294400
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 6.11756808617545e-05,
+      "loss": 0.6588,
+      "step": 294600
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 6.114930492821131e-05,
+      "loss": 0.6562,
+      "step": 294800
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 6.11229289946681e-05,
+      "loss": 0.6622,
+      "step": 295000
+    },
+    {
+      "epoch": 0.39,
+      "eval_em_first_line_ratio": 0.5101,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3971,
+      "eval_loss": 0.6679697036743164,
+      "eval_runtime": 231.8367,
+      "eval_samples_per_second": 43.134,
+      "eval_steps_per_second": 0.229,
+      "step": 295000
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 6.10965530611249e-05,
+      "loss": 0.663,
+      "step": 295200
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 6.107017712758171e-05,
+      "loss": 0.6655,
+      "step": 295400
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 6.104380119403851e-05,
+      "loss": 0.6642,
+      "step": 295600
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 6.101742526049532e-05,
+      "loss": 0.6632,
+      "step": 295800
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 6.099104932695212e-05,
+      "loss": 0.6641,
+      "step": 296000
+    },
+    {
+      "epoch": 0.39,
+      "eval_em_first_line_ratio": 0.511,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3972,
+      "eval_loss": 0.6675227880477905,
+      "eval_runtime": 228.0984,
+      "eval_samples_per_second": 43.841,
+      "eval_steps_per_second": 0.232,
+      "step": 296000
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 6.096480527307664e-05,
+      "loss": 0.661,
+      "step": 296200
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 6.093842933953344e-05,
+      "loss": 0.6597,
+      "step": 296400
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 6.0912053405990244e-05,
+      "loss": 0.6623,
+      "step": 296600
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 6.088567747244704e-05,
+      "loss": 0.6609,
+      "step": 296800
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 6.0859301538903846e-05,
+      "loss": 0.6622,
+      "step": 297000
+    },
+    {
+      "epoch": 0.39,
+      "eval_em_first_line_ratio": 0.5106,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3976,
+      "eval_loss": 0.6672832369804382,
+      "eval_runtime": 226.8363,
+      "eval_samples_per_second": 44.085,
+      "eval_steps_per_second": 0.234,
+      "step": 297000
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 6.0832925605360644e-05,
+      "loss": 0.6637,
+      "step": 297200
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 6.080654967181745e-05,
+      "loss": 0.6566,
+      "step": 297400
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 6.0780173738274246e-05,
+      "loss": 0.6611,
+      "step": 297600
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 6.075379780473105e-05,
+      "loss": 0.6628,
+      "step": 297800
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 6.072742187118786e-05,
+      "loss": 0.6597,
+      "step": 298000
+    },
+    {
+      "epoch": 0.39,
+      "eval_em_first_line_ratio": 0.5119,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3987,
+      "eval_loss": 0.6675494313240051,
+      "eval_runtime": 229.1446,
+      "eval_samples_per_second": 43.641,
+      "eval_steps_per_second": 0.231,
+      "step": 298000
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 6.070104593764466e-05,
+      "loss": 0.6636,
+      "step": 298200
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 6.0674670004101464e-05,
+      "loss": 0.6618,
+      "step": 298400
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 6.064829407055826e-05,
+      "loss": 0.6574,
+      "step": 298600
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 6.0621918137015066e-05,
+      "loss": 0.6635,
+      "step": 298800
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 6.0595542203471864e-05,
+      "loss": 0.6619,
+      "step": 299000
+    },
+    {
+      "epoch": 0.39,
+      "eval_em_first_line_ratio": 0.5102,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3974,
+      "eval_loss": 0.6675372123718262,
+      "eval_runtime": 232.0755,
+      "eval_samples_per_second": 43.089,
+      "eval_steps_per_second": 0.228,
+      "step": 299000
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 6.056929814959639e-05,
+      "loss": 0.6605,
+      "step": 299200
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 6.054305409572091e-05,
+      "loss": 0.6628,
+      "step": 299400
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 6.051667816217771e-05,
+      "loss": 0.6611,
+      "step": 299600
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 6.049030222863451e-05,
+      "loss": 0.6574,
+      "step": 299800
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 6.046392629509131e-05,
+      "loss": 0.6619,
+      "step": 300000
+    },
+    {
+      "epoch": 0.4,
+      "eval_em_first_line_ratio": 0.5074,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3949,
+      "eval_loss": 0.6663480997085571,
+      "eval_runtime": 231.3453,
+      "eval_samples_per_second": 43.225,
+      "eval_steps_per_second": 0.229,
+      "step": 300000
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 6.0437550361548115e-05,
+      "loss": 0.6593,
+      "step": 300200
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 6.041117442800491e-05,
+      "loss": 0.6613,
+      "step": 300400
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 6.038493037412943e-05,
+      "loss": 0.6585,
+      "step": 300600
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 6.0358554440586237e-05,
+      "loss": 0.6669,
+      "step": 300800
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 6.0332178507043034e-05,
+      "loss": 0.6641,
+      "step": 301000
+    },
+    {
+      "epoch": 0.4,
+      "eval_em_first_line_ratio": 0.5081,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3949,
+      "eval_loss": 0.6660829186439514,
+      "eval_runtime": 223.6383,
+      "eval_samples_per_second": 44.715,
+      "eval_steps_per_second": 0.237,
+      "step": 301000
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 6.030580257349984e-05,
+      "loss": 0.6624,
+      "step": 301200
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 6.0279426639956636e-05,
+      "loss": 0.6569,
+      "step": 301400
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 6.025305070641344e-05,
+      "loss": 0.6609,
+      "step": 301600
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 6.022667477287024e-05,
+      "loss": 0.6603,
+      "step": 301800
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 6.020029883932705e-05,
+      "loss": 0.6611,
+      "step": 302000
+    },
+    {
+      "epoch": 0.4,
+      "eval_em_first_line_ratio": 0.5103,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3964,
+      "eval_loss": 0.666179895401001,
+      "eval_runtime": 229.6847,
+      "eval_samples_per_second": 43.538,
+      "eval_steps_per_second": 0.231,
+      "step": 302000
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 6.0173922905783854e-05,
+      "loss": 0.6626,
+      "step": 302200
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 6.014754697224065e-05,
+      "loss": 0.6684,
+      "step": 302400
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 6.0121171038697456e-05,
+      "loss": 0.6611,
+      "step": 302600
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 6.0094795105154254e-05,
+      "loss": 0.6603,
+      "step": 302800
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 6.006841917161106e-05,
+      "loss": 0.6627,
+      "step": 303000
+    },
+    {
+      "epoch": 0.4,
+      "eval_em_first_line_ratio": 0.51,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.397,
+      "eval_loss": 0.665461540222168,
+      "eval_runtime": 221.2098,
+      "eval_samples_per_second": 45.206,
+      "eval_steps_per_second": 0.24,
+      "step": 303000
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 6.0042043238067856e-05,
+      "loss": 0.6602,
+      "step": 303200
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 6.001566730452466e-05,
+      "loss": 0.6606,
+      "step": 303400
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 5.998929137098146e-05,
+      "loss": 0.6587,
+      "step": 303600
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 5.996291543743827e-05,
+      "loss": 0.6614,
+      "step": 303800
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 5.9936539503895073e-05,
+      "loss": 0.6609,
+      "step": 304000
+    },
+    {
+      "epoch": 0.4,
+      "eval_em_first_line_ratio": 0.5121,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3994,
+      "eval_loss": 0.6654477715492249,
+      "eval_runtime": 226.9298,
+      "eval_samples_per_second": 44.066,
+      "eval_steps_per_second": 0.234,
+      "step": 304000
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 5.991016357035187e-05,
+      "loss": 0.6621,
+      "step": 304200
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 5.9883787636808676e-05,
+      "loss": 0.6621,
+      "step": 304400
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 5.985741170326547e-05,
+      "loss": 0.6591,
+      "step": 304600
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 5.983103576972228e-05,
+      "loss": 0.6589,
+      "step": 304800
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 5.9804659836179075e-05,
+      "loss": 0.6634,
+      "step": 305000
+    },
+    {
+      "epoch": 0.4,
+      "eval_em_first_line_ratio": 0.5124,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3989,
+      "eval_loss": 0.6652033925056458,
+      "eval_runtime": 224.4893,
+      "eval_samples_per_second": 44.546,
+      "eval_steps_per_second": 0.236,
+      "step": 305000
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 5.977828390263588e-05,
+      "loss": 0.658,
+      "step": 305200
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 5.975190796909268e-05,
+      "loss": 0.6597,
+      "step": 305400
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 5.972553203554948e-05,
+      "loss": 0.6591,
+      "step": 305600
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 5.969915610200629e-05,
+      "loss": 0.6567,
+      "step": 305800
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 5.967278016846309e-05,
+      "loss": 0.6641,
+      "step": 306000
+    },
+    {
+      "epoch": 0.4,
+      "eval_em_first_line_ratio": 0.5116,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3989,
+      "eval_loss": 0.6652944087982178,
+      "eval_runtime": 234.0544,
+      "eval_samples_per_second": 42.725,
+      "eval_steps_per_second": 0.226,
+      "step": 306000
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 5.9646404234919895e-05,
+      "loss": 0.6612,
+      "step": 306200
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 5.962002830137669e-05,
+      "loss": 0.6592,
+      "step": 306400
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 5.95936523678335e-05,
+      "loss": 0.6594,
+      "step": 306600
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 5.9567276434290295e-05,
+      "loss": 0.6603,
+      "step": 306800
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 5.95409005007471e-05,
+      "loss": 0.6564,
+      "step": 307000
+    },
+    {
+      "epoch": 0.4,
+      "eval_em_first_line_ratio": 0.5098,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.397,
+      "eval_loss": 0.6651482582092285,
+      "eval_runtime": 228.4571,
+      "eval_samples_per_second": 43.772,
+      "eval_steps_per_second": 0.232,
+      "step": 307000
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 5.95145245672039e-05,
+      "loss": 0.6619,
+      "step": 307200
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 5.94881486336607e-05,
+      "loss": 0.6595,
+      "step": 307400
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 5.946177270011751e-05,
+      "loss": 0.6587,
+      "step": 307600
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 5.943539676657432e-05,
+      "loss": 0.6599,
+      "step": 307800
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 5.9409020833031114e-05,
+      "loss": 0.657,
+      "step": 308000
+    },
+    {
+      "epoch": 0.41,
+      "eval_em_first_line_ratio": 0.5121,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3981,
+      "eval_loss": 0.6651002764701843,
+      "eval_runtime": 231.0596,
+      "eval_samples_per_second": 43.279,
+      "eval_steps_per_second": 0.229,
+      "step": 308000
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 5.938264489948792e-05,
+      "loss": 0.6562,
+      "step": 308200
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 5.9356268965944717e-05,
+      "loss": 0.6577,
+      "step": 308400
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 5.932989303240152e-05,
+      "loss": 0.6579,
+      "step": 308600
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 5.930351709885832e-05,
+      "loss": 0.6568,
+      "step": 308800
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 5.927714116531512e-05,
+      "loss": 0.6596,
+      "step": 309000
+    },
+    {
+      "epoch": 0.41,
+      "eval_em_first_line_ratio": 0.5129,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3998,
+      "eval_loss": 0.6646019816398621,
+      "eval_runtime": 220.761,
+      "eval_samples_per_second": 45.298,
+      "eval_steps_per_second": 0.24,
+      "step": 309000
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 5.925089711143964e-05,
+      "loss": 0.661,
+      "step": 309200
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 5.922452117789644e-05,
+      "loss": 0.6601,
+      "step": 309400
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 5.9198145244353245e-05,
+      "loss": 0.6659,
+      "step": 309600
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 5.917176931081004e-05,
+      "loss": 0.6597,
+      "step": 309800
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 5.914539337726685e-05,
+      "loss": 0.654,
+      "step": 310000
+    },
+    {
+      "epoch": 0.41,
+      "eval_em_first_line_ratio": 0.5099,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3964,
+      "eval_loss": 0.6642576456069946,
+      "eval_runtime": 225.5297,
+      "eval_samples_per_second": 44.34,
+      "eval_steps_per_second": 0.235,
+      "step": 310000
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 5.911914932339137e-05,
+      "loss": 0.6578,
+      "step": 310200
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 5.9092773389848165e-05,
+      "loss": 0.6634,
+      "step": 310400
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 5.906639745630497e-05,
+      "loss": 0.6575,
+      "step": 310600
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 5.904002152276178e-05,
+      "loss": 0.6583,
+      "step": 310800
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 5.901364558921858e-05,
+      "loss": 0.6591,
+      "step": 311000
+    },
+    {
+      "epoch": 0.41,
+      "eval_em_first_line_ratio": 0.5111,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3975,
+      "eval_loss": 0.6645973920822144,
+      "eval_runtime": 230.5149,
+      "eval_samples_per_second": 43.381,
+      "eval_steps_per_second": 0.23,
+      "step": 311000
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 5.898726965567538e-05,
+      "loss": 0.6579,
+      "step": 311200
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 5.896089372213218e-05,
+      "loss": 0.6633,
+      "step": 311400
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 5.8934517788588985e-05,
+      "loss": 0.6583,
+      "step": 311600
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 5.890814185504578e-05,
+      "loss": 0.6557,
+      "step": 311800
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 5.888176592150259e-05,
+      "loss": 0.6525,
+      "step": 312000
+    },
+    {
+      "epoch": 0.41,
+      "eval_em_first_line_ratio": 0.5094,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3959,
+      "eval_loss": 0.6641407012939453,
+      "eval_runtime": 224.287,
+      "eval_samples_per_second": 44.586,
+      "eval_steps_per_second": 0.236,
+      "step": 312000
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 5.8855389987959384e-05,
+      "loss": 0.6593,
+      "step": 312200
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 5.882901405441619e-05,
+      "loss": 0.6571,
+      "step": 312400
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 5.880277000054071e-05,
+      "loss": 0.6592,
+      "step": 312600
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 5.8776394066997506e-05,
+      "loss": 0.6536,
+      "step": 312800
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 5.875001813345431e-05,
+      "loss": 0.6574,
+      "step": 313000
+    },
+    {
+      "epoch": 0.41,
+      "eval_em_first_line_ratio": 0.5119,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3983,
+      "eval_loss": 0.6652553081512451,
+      "eval_runtime": 224.1971,
+      "eval_samples_per_second": 44.604,
+      "eval_steps_per_second": 0.236,
+      "step": 313000
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 5.872364219991111e-05,
+      "loss": 0.6619,
+      "step": 313200
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 5.869726626636792e-05,
+      "loss": 0.6532,
+      "step": 313400
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 5.8670890332824724e-05,
+      "loss": 0.6561,
+      "step": 313600
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 5.864451439928152e-05,
+      "loss": 0.6519,
+      "step": 313800
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 5.861827034540605e-05,
+      "loss": 0.6601,
+      "step": 314000
+    },
+    {
+      "epoch": 0.41,
+      "eval_em_first_line_ratio": 0.5094,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.396,
+      "eval_loss": 0.6640971899032593,
+      "eval_runtime": 231.9869,
+      "eval_samples_per_second": 43.106,
+      "eval_steps_per_second": 0.228,
+      "step": 314000
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 5.8591894411862846e-05,
+      "loss": 0.6577,
+      "step": 314200
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 5.856551847831965e-05,
+      "loss": 0.6573,
+      "step": 314400
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 5.853914254477645e-05,
+      "loss": 0.6563,
+      "step": 314600
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 5.851276661123325e-05,
+      "loss": 0.6629,
+      "step": 314800
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 5.848639067769005e-05,
+      "loss": 0.6551,
+      "step": 315000
+    },
+    {
+      "epoch": 0.42,
+      "eval_em_first_line_ratio": 0.5097,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3961,
+      "eval_loss": 0.664097249507904,
+      "eval_runtime": 228.6256,
+      "eval_samples_per_second": 43.74,
+      "eval_steps_per_second": 0.232,
+      "step": 315000
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 5.8460014744146855e-05,
+      "loss": 0.6583,
+      "step": 315200
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 5.843363881060365e-05,
+      "loss": 0.653,
+      "step": 315400
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 5.840726287706046e-05,
+      "loss": 0.6587,
+      "step": 315600
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 5.8380886943517254e-05,
+      "loss": 0.6557,
+      "step": 315800
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 5.8354511009974066e-05,
+      "loss": 0.6592,
+      "step": 316000
+    },
+    {
+      "epoch": 0.42,
+      "eval_em_first_line_ratio": 0.5115,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3978,
+      "eval_loss": 0.6637352108955383,
+      "eval_runtime": 231.2625,
+      "eval_samples_per_second": 43.241,
+      "eval_steps_per_second": 0.229,
+      "step": 316000
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 5.832813507643087e-05,
+      "loss": 0.6574,
+      "step": 316200
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 5.830175914288767e-05,
+      "loss": 0.6545,
+      "step": 316400
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 5.827538320934447e-05,
+      "loss": 0.6562,
+      "step": 316600
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 5.824900727580127e-05,
+      "loss": 0.6545,
+      "step": 316800
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 5.8222631342258074e-05,
+      "loss": 0.6574,
+      "step": 317000
+    },
+    {
+      "epoch": 0.42,
+      "eval_em_first_line_ratio": 0.5081,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3942,
+      "eval_loss": 0.6631019115447998,
+      "eval_runtime": 232.8177,
+      "eval_samples_per_second": 42.952,
+      "eval_steps_per_second": 0.228,
+      "step": 317000
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 5.819625540871487e-05,
+      "loss": 0.653,
+      "step": 317200
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 5.8169879475171676e-05,
+      "loss": 0.6585,
+      "step": 317400
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 5.8143503541628474e-05,
+      "loss": 0.6526,
+      "step": 317600
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 5.8117127608085285e-05,
+      "loss": 0.6537,
+      "step": 317800
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 5.809075167454209e-05,
+      "loss": 0.6571,
+      "step": 318000
+    },
+    {
+      "epoch": 0.42,
+      "eval_em_first_line_ratio": 0.5129,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3992,
+      "eval_loss": 0.6627387404441833,
+      "eval_runtime": 226.4571,
+      "eval_samples_per_second": 44.158,
+      "eval_steps_per_second": 0.234,
+      "step": 318000
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 5.806437574099889e-05,
+      "loss": 0.6567,
+      "step": 318200
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 5.803799980745569e-05,
+      "loss": 0.6552,
+      "step": 318400
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 5.801162387391249e-05,
+      "loss": 0.6533,
+      "step": 318600
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 5.7985247940369294e-05,
+      "loss": 0.6593,
+      "step": 318800
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 5.795887200682609e-05,
+      "loss": 0.6531,
+      "step": 319000
+    },
+    {
+      "epoch": 0.42,
+      "eval_em_first_line_ratio": 0.5114,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3982,
+      "eval_loss": 0.662930965423584,
+      "eval_runtime": 233.3909,
+      "eval_samples_per_second": 42.847,
+      "eval_steps_per_second": 0.227,
+      "step": 319000
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 5.7932496073282896e-05,
+      "loss": 0.6623,
+      "step": 319200
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 5.7906252019407416e-05,
+      "loss": 0.6544,
+      "step": 319400
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 5.787987608586421e-05,
+      "loss": 0.6569,
+      "step": 319600
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 5.785350015232102e-05,
+      "loss": 0.6597,
+      "step": 319800
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 5.7827124218777815e-05,
+      "loss": 0.6563,
+      "step": 320000
+    },
+    {
+      "epoch": 0.42,
+      "eval_em_first_line_ratio": 0.5115,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3982,
+      "eval_loss": 0.663051426410675,
+      "eval_runtime": 223.9055,
+      "eval_samples_per_second": 44.662,
+      "eval_steps_per_second": 0.237,
+      "step": 320000
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 5.780074828523462e-05,
+      "loss": 0.6599,
+      "step": 320200
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 5.777437235169143e-05,
+      "loss": 0.6534,
+      "step": 320400
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 5.774799641814823e-05,
+      "loss": 0.6553,
+      "step": 320600
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 5.772162048460503e-05,
+      "loss": 0.6561,
+      "step": 320800
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 5.769524455106183e-05,
+      "loss": 0.6595,
+      "step": 321000
+    },
+    {
+      "epoch": 0.42,
+      "eval_em_first_line_ratio": 0.5102,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3968,
+      "eval_loss": 0.6629048585891724,
+      "eval_runtime": 230.9518,
+      "eval_samples_per_second": 43.299,
+      "eval_steps_per_second": 0.229,
+      "step": 321000
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 5.7668868617518635e-05,
+      "loss": 0.655,
+      "step": 321200
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 5.764249268397543e-05,
+      "loss": 0.6545,
+      "step": 321400
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 5.761611675043224e-05,
+      "loss": 0.6587,
+      "step": 321600
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 5.7589740816889035e-05,
+      "loss": 0.6542,
+      "step": 321800
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 5.756336488334584e-05,
+      "loss": 0.659,
+      "step": 322000
+    },
+    {
+      "epoch": 0.42,
+      "eval_em_first_line_ratio": 0.5105,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3967,
+      "eval_loss": 0.6634783148765564,
+      "eval_runtime": 229.1224,
+      "eval_samples_per_second": 43.645,
+      "eval_steps_per_second": 0.231,
+      "step": 322000
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 5.753698894980264e-05,
+      "loss": 0.6571,
+      "step": 322200
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 5.751061301625945e-05,
+      "loss": 0.6516,
+      "step": 322400
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 5.748423708271625e-05,
+      "loss": 0.6569,
+      "step": 322600
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 5.745786114917305e-05,
+      "loss": 0.6543,
+      "step": 322800
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 5.7431485215629855e-05,
+      "loss": 0.6545,
+      "step": 323000
+    },
+    {
+      "epoch": 0.43,
+      "eval_em_first_line_ratio": 0.5123,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3989,
+      "eval_loss": 0.6624303460121155,
+      "eval_runtime": 228.9864,
+      "eval_samples_per_second": 43.671,
+      "eval_steps_per_second": 0.231,
+      "step": 323000
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 5.740510928208665e-05,
+      "loss": 0.6566,
+      "step": 323200
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 5.737873334854346e-05,
+      "loss": 0.6537,
+      "step": 323400
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 5.7352357415000254e-05,
+      "loss": 0.6519,
+      "step": 323600
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 5.732598148145706e-05,
+      "loss": 0.6557,
+      "step": 323800
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 5.729973742758158e-05,
+      "loss": 0.6586,
+      "step": 324000
+    },
+    {
+      "epoch": 0.43,
+      "eval_em_first_line_ratio": 0.5129,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3993,
+      "eval_loss": 0.6619406938552856,
+      "eval_runtime": 227.7358,
+      "eval_samples_per_second": 43.911,
+      "eval_steps_per_second": 0.233,
+      "step": 324000
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 5.727336149403838e-05,
+      "loss": 0.6555,
+      "step": 324200
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 5.724698556049518e-05,
+      "loss": 0.6518,
+      "step": 324400
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 5.72207415066197e-05,
+      "loss": 0.6566,
+      "step": 324600
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 5.7194365573076505e-05,
+      "loss": 0.6562,
+      "step": 324800
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 5.716825339886874e-05,
+      "loss": 0.6517,
+      "step": 325000
+    },
+    {
+      "epoch": 0.43,
+      "eval_em_first_line_ratio": 0.5097,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3961,
+      "eval_loss": 0.6614932417869568,
+      "eval_runtime": 225.1797,
+      "eval_samples_per_second": 44.409,
+      "eval_steps_per_second": 0.235,
+      "step": 325000
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 5.7141877465325545e-05,
+      "loss": 0.6534,
+      "step": 325200
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 5.711550153178234e-05,
+      "loss": 0.6541,
+      "step": 325400
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 5.708912559823915e-05,
+      "loss": 0.6534,
+      "step": 325600
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 5.7062749664695945e-05,
+      "loss": 0.6552,
+      "step": 325800
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 5.703637373115275e-05,
+      "loss": 0.6573,
+      "step": 326000
+    },
+    {
+      "epoch": 0.43,
+      "eval_em_first_line_ratio": 0.5107,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3967,
+      "eval_loss": 0.6614888906478882,
+      "eval_runtime": 224.5472,
+      "eval_samples_per_second": 44.534,
+      "eval_steps_per_second": 0.236,
+      "step": 326000
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 5.700999779760955e-05,
+      "loss": 0.6518,
+      "step": 326200
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 5.698362186406635e-05,
+      "loss": 0.6523,
+      "step": 326400
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 5.695724593052315e-05,
+      "loss": 0.6517,
+      "step": 326600
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 5.693086999697996e-05,
+      "loss": 0.6571,
+      "step": 326800
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 5.6904494063436765e-05,
+      "loss": 0.6525,
+      "step": 327000
+    },
+    {
+      "epoch": 0.43,
+      "eval_em_first_line_ratio": 0.5096,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3963,
+      "eval_loss": 0.6619186401367188,
+      "eval_runtime": 232.8801,
+      "eval_samples_per_second": 42.941,
+      "eval_steps_per_second": 0.228,
+      "step": 327000
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 5.687811812989356e-05,
+      "loss": 0.6557,
+      "step": 327200
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 5.685174219635037e-05,
+      "loss": 0.6542,
+      "step": 327400
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 5.6825366262807164e-05,
+      "loss": 0.6564,
+      "step": 327600
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 5.679899032926397e-05,
+      "loss": 0.6525,
+      "step": 327800
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 5.6772614395720766e-05,
+      "loss": 0.6498,
+      "step": 328000
+    },
+    {
+      "epoch": 0.43,
+      "eval_em_first_line_ratio": 0.5109,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3976,
+      "eval_loss": 0.6614850759506226,
+      "eval_runtime": 234.3278,
+      "eval_samples_per_second": 42.675,
+      "eval_steps_per_second": 0.226,
+      "step": 328000
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 5.674623846217757e-05,
+      "loss": 0.6522,
+      "step": 328200
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 5.671986252863437e-05,
+      "loss": 0.6523,
+      "step": 328400
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 5.6693486595091186e-05,
+      "loss": 0.6547,
+      "step": 328600
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 5.6667110661547984e-05,
+      "loss": 0.6557,
+      "step": 328800
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 5.664073472800478e-05,
+      "loss": 0.6521,
+      "step": 329000
+    },
+    {
+      "epoch": 0.43,
+      "eval_em_first_line_ratio": 0.5063,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3934,
+      "eval_loss": 0.6629480719566345,
+      "eval_runtime": 230.5463,
+      "eval_samples_per_second": 43.375,
+      "eval_steps_per_second": 0.23,
+      "step": 329000
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 5.6614490674129295e-05,
+      "loss": 0.6566,
+      "step": 329200
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 5.6588114740586106e-05,
+      "loss": 0.6524,
+      "step": 329400
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 5.656173880704291e-05,
+      "loss": 0.6497,
+      "step": 329600
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 5.653536287349971e-05,
+      "loss": 0.6547,
+      "step": 329800
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 5.650898693995651e-05,
+      "loss": 0.6582,
+      "step": 330000
+    },
+    {
+      "epoch": 0.44,
+      "eval_em_first_line_ratio": 0.5109,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3975,
+      "eval_loss": 0.6609156131744385,
+      "eval_runtime": 223.1303,
+      "eval_samples_per_second": 44.817,
+      "eval_steps_per_second": 0.238,
+      "step": 330000
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 5.648261100641331e-05,
+      "loss": 0.6524,
+      "step": 330200
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 5.6456235072870115e-05,
+      "loss": 0.6526,
+      "step": 330400
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 5.642985913932691e-05,
+      "loss": 0.6584,
+      "step": 330600
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 5.640348320578372e-05,
+      "loss": 0.6561,
+      "step": 330800
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 5.6377107272240514e-05,
+      "loss": 0.6536,
+      "step": 331000
+    },
+    {
+      "epoch": 0.44,
+      "eval_em_first_line_ratio": 0.5123,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3987,
+      "eval_loss": 0.6608020663261414,
+      "eval_runtime": 219.9738,
+      "eval_samples_per_second": 45.46,
+      "eval_steps_per_second": 0.241,
+      "step": 331000
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 5.6350731338697326e-05,
+      "loss": 0.6552,
+      "step": 331200
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 5.632435540515413e-05,
+      "loss": 0.658,
+      "step": 331400
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 5.629797947161093e-05,
+      "loss": 0.6595,
+      "step": 331600
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 5.627160353806773e-05,
+      "loss": 0.6549,
+      "step": 331800
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 5.624522760452453e-05,
+      "loss": 0.6573,
+      "step": 332000
+    },
+    {
+      "epoch": 0.44,
+      "eval_em_first_line_ratio": 0.5144,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.401,
+      "eval_loss": 0.6605109572410583,
+      "eval_runtime": 232.4037,
+      "eval_samples_per_second": 43.029,
+      "eval_steps_per_second": 0.228,
+      "step": 332000
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 5.6218851670981334e-05,
+      "loss": 0.6529,
+      "step": 332200
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 5.619247573743813e-05,
+      "loss": 0.6563,
+      "step": 332400
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 5.6166099803894936e-05,
+      "loss": 0.655,
+      "step": 332600
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 5.6139723870351734e-05,
+      "loss": 0.6512,
+      "step": 332800
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 5.611334793680854e-05,
+      "loss": 0.6544,
+      "step": 333000
+    },
+    {
+      "epoch": 0.44,
+      "eval_em_first_line_ratio": 0.5103,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3969,
+      "eval_loss": 0.6605858206748962,
+      "eval_runtime": 232.0715,
+      "eval_samples_per_second": 43.09,
+      "eval_steps_per_second": 0.228,
+      "step": 333000
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 5.608710388293306e-05,
+      "loss": 0.6543,
+      "step": 333200
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 5.6060727949389856e-05,
+      "loss": 0.6575,
+      "step": 333400
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 5.603435201584666e-05,
+      "loss": 0.6531,
+      "step": 333600
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 5.600797608230347e-05,
+      "loss": 0.6527,
+      "step": 333800
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 5.598160014876027e-05,
+      "loss": 0.6535,
+      "step": 334000
+    },
+    {
+      "epoch": 0.44,
+      "eval_em_first_line_ratio": 0.5126,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.399,
+      "eval_loss": 0.661575973033905,
+      "eval_runtime": 246.1693,
+      "eval_samples_per_second": 40.622,
+      "eval_steps_per_second": 0.215,
+      "step": 334000
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 5.5955224215217074e-05,
+      "loss": 0.652,
+      "step": 334200
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 5.592884828167387e-05,
+      "loss": 0.6522,
+      "step": 334400
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 5.5902472348130676e-05,
+      "loss": 0.6529,
+      "step": 334600
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 5.5876096414587473e-05,
+      "loss": 0.6544,
+      "step": 334800
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 5.584972048104428e-05,
+      "loss": 0.6498,
+      "step": 335000
+    },
+    {
+      "epoch": 0.44,
+      "eval_em_first_line_ratio": 0.5105,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3973,
+      "eval_loss": 0.6614336371421814,
+      "eval_runtime": 248.1476,
+      "eval_samples_per_second": 40.299,
+      "eval_steps_per_second": 0.214,
+      "step": 335000
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 5.5823344547501075e-05,
+      "loss": 0.6558,
+      "step": 335200
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 5.579696861395788e-05,
+      "loss": 0.6562,
+      "step": 335400
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 5.577059268041469e-05,
+      "loss": 0.6533,
+      "step": 335600
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 5.574421674687149e-05,
+      "loss": 0.6519,
+      "step": 335800
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 5.5717972692996e-05,
+      "loss": 0.6527,
+      "step": 336000
+    },
+    {
+      "epoch": 0.44,
+      "eval_em_first_line_ratio": 0.5127,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3995,
+      "eval_loss": 0.660659670829773,
+      "eval_runtime": 225.2488,
+      "eval_samples_per_second": 44.395,
+      "eval_steps_per_second": 0.235,
+      "step": 336000
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 5.56915967594528e-05,
+      "loss": 0.6543,
+      "step": 336200
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 5.566522082590962e-05,
+      "loss": 0.6568,
+      "step": 336400
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 5.5638844892366415e-05,
+      "loss": 0.6531,
+      "step": 336600
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 5.561246895882321e-05,
+      "loss": 0.657,
+      "step": 336800
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 5.558609302528002e-05,
+      "loss": 0.6526,
+      "step": 337000
+    },
+    {
+      "epoch": 0.44,
+      "eval_em_first_line_ratio": 0.5114,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.398,
+      "eval_loss": 0.6595136523246765,
+      "eval_runtime": 230.5209,
+      "eval_samples_per_second": 43.38,
+      "eval_steps_per_second": 0.23,
+      "step": 337000
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 5.5559717091736815e-05,
+      "loss": 0.6527,
+      "step": 337200
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 5.553334115819362e-05,
+      "loss": 0.652,
+      "step": 337400
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 5.550696522465042e-05,
+      "loss": 0.6552,
+      "step": 337600
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 5.548058929110722e-05,
+      "loss": 0.6509,
+      "step": 337800
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 5.545421335756402e-05,
+      "loss": 0.6525,
+      "step": 338000
+    },
+    {
+      "epoch": 0.45,
+      "eval_em_first_line_ratio": 0.5105,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3968,
+      "eval_loss": 0.660034716129303,
+      "eval_runtime": 228.7369,
+      "eval_samples_per_second": 43.718,
+      "eval_steps_per_second": 0.232,
+      "step": 338000
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 5.5427969303688546e-05,
+      "loss": 0.6524,
+      "step": 338200
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 5.5401593370145343e-05,
+      "loss": 0.6539,
+      "step": 338400
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 5.537521743660215e-05,
+      "loss": 0.651,
+      "step": 338600
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 5.534884150305896e-05,
+      "loss": 0.6564,
+      "step": 338800
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 5.532246556951576e-05,
+      "loss": 0.6531,
+      "step": 339000
+    },
+    {
+      "epoch": 0.45,
+      "eval_em_first_line_ratio": 0.51,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.397,
+      "eval_loss": 0.6596644520759583,
+      "eval_runtime": 232.0103,
+      "eval_samples_per_second": 43.102,
+      "eval_steps_per_second": 0.228,
+      "step": 339000
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 5.529608963597256e-05,
+      "loss": 0.6526,
+      "step": 339200
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 5.526971370242936e-05,
+      "loss": 0.6518,
+      "step": 339400
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 5.524333776888616e-05,
+      "loss": 0.6517,
+      "step": 339600
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 5.521696183534296e-05,
+      "loss": 0.6482,
+      "step": 339800
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 5.5190585901799765e-05,
+      "loss": 0.6515,
+      "step": 340000
+    },
+    {
+      "epoch": 0.45,
+      "eval_em_first_line_ratio": 0.512,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3984,
+      "eval_loss": 0.6597820520401001,
+      "eval_runtime": 230.3601,
+      "eval_samples_per_second": 43.41,
+      "eval_steps_per_second": 0.23,
+      "step": 340000
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 5.5164341847924285e-05,
+      "loss": 0.6541,
+      "step": 340200
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 5.513796591438108e-05,
+      "loss": 0.6503,
+      "step": 340400
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 5.511158998083789e-05,
+      "loss": 0.6569,
+      "step": 340600
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 5.5085214047294685e-05,
+      "loss": 0.654,
+      "step": 340800
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 5.505883811375149e-05,
+      "loss": 0.6534,
+      "step": 341000
+    },
+    {
+      "epoch": 0.45,
+      "eval_em_first_line_ratio": 0.5147,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4013,
+      "eval_loss": 0.6594851016998291,
+      "eval_runtime": 223.6661,
+      "eval_samples_per_second": 44.71,
+      "eval_steps_per_second": 0.237,
+      "step": 341000
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 5.503246218020829e-05,
+      "loss": 0.6465,
+      "step": 341200
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 5.500621812633281e-05,
+      "loss": 0.6526,
+      "step": 341400
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 5.497984219278961e-05,
+      "loss": 0.6531,
+      "step": 341600
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 5.495346625924641e-05,
+      "loss": 0.6481,
+      "step": 341800
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 5.492709032570322e-05,
+      "loss": 0.6526,
+      "step": 342000
+    },
+    {
+      "epoch": 0.45,
+      "eval_em_first_line_ratio": 0.513,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3995,
+      "eval_loss": 0.6592131853103638,
+      "eval_runtime": 223.0348,
+      "eval_samples_per_second": 44.836,
+      "eval_steps_per_second": 0.238,
+      "step": 342000
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 5.4900714392160025e-05,
+      "loss": 0.6504,
+      "step": 342200
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 5.487433845861682e-05,
+      "loss": 0.654,
+      "step": 342400
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 5.484796252507363e-05,
+      "loss": 0.6528,
+      "step": 342600
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 5.4821586591530424e-05,
+      "loss": 0.6502,
+      "step": 342800
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 5.479521065798723e-05,
+      "loss": 0.6516,
+      "step": 343000
+    },
+    {
+      "epoch": 0.45,
+      "eval_em_first_line_ratio": 0.5131,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4,
+      "eval_loss": 0.6589600443840027,
+      "eval_runtime": 231.8121,
+      "eval_samples_per_second": 43.138,
+      "eval_steps_per_second": 0.229,
+      "step": 343000
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 5.4768834724444027e-05,
+      "loss": 0.6499,
+      "step": 343200
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 5.474245879090083e-05,
+      "loss": 0.6534,
+      "step": 343400
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 5.471608285735763e-05,
+      "loss": 0.6499,
+      "step": 343600
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 5.468970692381443e-05,
+      "loss": 0.6472,
+      "step": 343800
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 5.4663330990271244e-05,
+      "loss": 0.649,
+      "step": 344000
+    },
+    {
+      "epoch": 0.45,
+      "eval_em_first_line_ratio": 0.512,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3986,
+      "eval_loss": 0.6587649583816528,
+      "eval_runtime": 231.8041,
+      "eval_samples_per_second": 43.14,
+      "eval_steps_per_second": 0.229,
+      "step": 344000
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 5.463695505672804e-05,
+      "loss": 0.6491,
+      "step": 344200
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 5.4610579123184846e-05,
+      "loss": 0.6535,
+      "step": 344400
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 5.4584203189641644e-05,
+      "loss": 0.6508,
+      "step": 344600
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 5.455782725609845e-05,
+      "loss": 0.649,
+      "step": 344800
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 5.4531451322555246e-05,
+      "loss": 0.6505,
+      "step": 345000
+    },
+    {
+      "epoch": 0.45,
+      "eval_em_first_line_ratio": 0.5097,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3964,
+      "eval_loss": 0.6586926579475403,
+      "eval_runtime": 237.4326,
+      "eval_samples_per_second": 42.117,
+      "eval_steps_per_second": 0.223,
+      "step": 345000
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 5.450507538901205e-05,
+      "loss": 0.6511,
+      "step": 345200
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 5.447869945546885e-05,
+      "loss": 0.652,
+      "step": 345400
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 5.445232352192565e-05,
+      "loss": 0.6475,
+      "step": 345600
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 5.4425947588382464e-05,
+      "loss": 0.6532,
+      "step": 345800
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 5.439970353450698e-05,
+      "loss": 0.6519,
+      "step": 346000
+    },
+    {
+      "epoch": 0.46,
+      "eval_em_first_line_ratio": 0.51,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3968,
+      "eval_loss": 0.65865558385849,
+      "eval_runtime": 231.0306,
+      "eval_samples_per_second": 43.284,
+      "eval_steps_per_second": 0.229,
+      "step": 346000
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 5.4373327600963775e-05,
+      "loss": 0.6546,
+      "step": 346200
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 5.434695166742058e-05,
+      "loss": 0.6484,
+      "step": 346400
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 5.432057573387739e-05,
+      "loss": 0.6482,
+      "step": 346600
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 5.429419980033419e-05,
+      "loss": 0.6493,
+      "step": 346800
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 5.426782386679099e-05,
+      "loss": 0.6486,
+      "step": 347000
+    },
+    {
+      "epoch": 0.46,
+      "eval_em_first_line_ratio": 0.5136,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4004,
+      "eval_loss": 0.6586874127388,
+      "eval_runtime": 229.2285,
+      "eval_samples_per_second": 43.625,
+      "eval_steps_per_second": 0.231,
+      "step": 347000
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 5.424144793324779e-05,
+      "loss": 0.6497,
+      "step": 347200
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 5.4215071999704594e-05,
+      "loss": 0.6512,
+      "step": 347400
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 5.418869606616139e-05,
+      "loss": 0.6471,
+      "step": 347600
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 5.4162320132618196e-05,
+      "loss": 0.6505,
+      "step": 347800
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 5.4135944199074994e-05,
+      "loss": 0.6533,
+      "step": 348000
+    },
+    {
+      "epoch": 0.46,
+      "eval_em_first_line_ratio": 0.51,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3966,
+      "eval_loss": 0.6582003235816956,
+      "eval_runtime": 227.7136,
+      "eval_samples_per_second": 43.915,
+      "eval_steps_per_second": 0.233,
+      "step": 348000
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 5.41095682655318e-05,
+      "loss": 0.651,
+      "step": 348200
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 5.408319233198861e-05,
+      "loss": 0.6514,
+      "step": 348400
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 5.405681639844541e-05,
+      "loss": 0.6527,
+      "step": 348600
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 5.403044046490221e-05,
+      "loss": 0.648,
+      "step": 348800
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 5.400406453135901e-05,
+      "loss": 0.6534,
+      "step": 349000
+    },
+    {
+      "epoch": 0.46,
+      "eval_em_first_line_ratio": 0.5099,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3968,
+      "eval_loss": 0.6584148406982422,
+      "eval_runtime": 224.6033,
+      "eval_samples_per_second": 44.523,
+      "eval_steps_per_second": 0.236,
+      "step": 349000
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 5.397782047748353e-05,
+      "loss": 0.6493,
+      "step": 349200
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 5.3951444543940334e-05,
+      "loss": 0.6516,
+      "step": 349400
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 5.392506861039713e-05,
+      "loss": 0.6535,
+      "step": 349600
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 5.3898692676853936e-05,
+      "loss": 0.6507,
+      "step": 349800
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 5.3872316743310734e-05,
+      "loss": 0.6517,
+      "step": 350000
+    },
+    {
+      "epoch": 0.46,
+      "eval_em_first_line_ratio": 0.5119,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3979,
+      "eval_loss": 0.6589285135269165,
+      "eval_runtime": 237.2863,
+      "eval_samples_per_second": 42.143,
+      "eval_steps_per_second": 0.223,
+      "step": 350000
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 5.384594080976754e-05,
+      "loss": 0.6481,
+      "step": 350200
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 5.3819564876224336e-05,
+      "loss": 0.6484,
+      "step": 350400
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 5.3793320822348856e-05,
+      "loss": 0.6524,
+      "step": 350600
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 5.376694488880566e-05,
+      "loss": 0.6496,
+      "step": 350800
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 5.374056895526246e-05,
+      "loss": 0.6528,
+      "step": 351000
+    },
+    {
+      "epoch": 0.46,
+      "eval_em_first_line_ratio": 0.5114,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3976,
+      "eval_loss": 0.6588383913040161,
+      "eval_runtime": 226.8303,
+      "eval_samples_per_second": 44.086,
+      "eval_steps_per_second": 0.234,
+      "step": 351000
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 5.371419302171926e-05,
+      "loss": 0.649,
+      "step": 351200
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 5.368781708817606e-05,
+      "loss": 0.6522,
+      "step": 351400
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 5.366144115463288e-05,
+      "loss": 0.6464,
+      "step": 351600
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 5.3635065221089675e-05,
+      "loss": 0.6547,
+      "step": 351800
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 5.360868928754647e-05,
+      "loss": 0.6474,
+      "step": 352000
+    },
+    {
+      "epoch": 0.46,
+      "eval_em_first_line_ratio": 0.5116,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3981,
+      "eval_loss": 0.6583807468414307,
+      "eval_runtime": 231.1769,
+      "eval_samples_per_second": 43.257,
+      "eval_steps_per_second": 0.229,
+      "step": 352000
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 5.358231335400328e-05,
+      "loss": 0.6539,
+      "step": 352200
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 5.3555937420460075e-05,
+      "loss": 0.6494,
+      "step": 352400
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 5.352956148691688e-05,
+      "loss": 0.6528,
+      "step": 352600
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 5.350318555337368e-05,
+      "loss": 0.6525,
+      "step": 352800
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 5.347680961983048e-05,
+      "loss": 0.6533,
+      "step": 353000
+    },
+    {
+      "epoch": 0.47,
+      "eval_em_first_line_ratio": 0.5102,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3968,
+      "eval_loss": 0.6580835580825806,
+      "eval_runtime": 223.4063,
+      "eval_samples_per_second": 44.761,
+      "eval_steps_per_second": 0.237,
+      "step": 353000
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 5.345043368628728e-05,
+      "loss": 0.6549,
+      "step": 353200
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 5.3424057752744084e-05,
+      "loss": 0.6505,
+      "step": 353400
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 5.3397681819200895e-05,
+      "loss": 0.6452,
+      "step": 353600
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 5.33713058856577e-05,
+      "loss": 0.6478,
+      "step": 353800
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 5.33449299521145e-05,
+      "loss": 0.6462,
+      "step": 354000
+    },
+    {
+      "epoch": 0.47,
+      "eval_em_first_line_ratio": 0.5095,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3963,
+      "eval_loss": 0.6579205989837646,
+      "eval_runtime": 226.5815,
+      "eval_samples_per_second": 44.134,
+      "eval_steps_per_second": 0.234,
+      "step": 354000
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 5.33185540185713e-05,
+      "loss": 0.6502,
+      "step": 354200
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 5.32921780850281e-05,
+      "loss": 0.6523,
+      "step": 354400
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 5.326593403115262e-05,
+      "loss": 0.6488,
+      "step": 354600
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 5.323955809760942e-05,
+      "loss": 0.6475,
+      "step": 354800
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 5.321318216406622e-05,
+      "loss": 0.6492,
+      "step": 355000
+    },
+    {
+      "epoch": 0.47,
+      "eval_em_first_line_ratio": 0.5131,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3993,
+      "eval_loss": 0.6579131484031677,
+      "eval_runtime": 228.1333,
+      "eval_samples_per_second": 43.834,
+      "eval_steps_per_second": 0.232,
+      "step": 355000
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 5.3186806230523025e-05,
+      "loss": 0.6526,
+      "step": 355200
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 5.316043029697982e-05,
+      "loss": 0.6489,
+      "step": 355400
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 5.313405436343663e-05,
+      "loss": 0.6483,
+      "step": 355600
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 5.3107678429893425e-05,
+      "loss": 0.6498,
+      "step": 355800
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 5.3081302496350236e-05,
+      "loss": 0.651,
+      "step": 356000
+    },
+    {
+      "epoch": 0.47,
+      "eval_em_first_line_ratio": 0.5123,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3993,
+      "eval_loss": 0.6575753092765808,
+      "eval_runtime": 225.8138,
+      "eval_samples_per_second": 44.284,
+      "eval_steps_per_second": 0.235,
+      "step": 356000
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 5.305492656280704e-05,
+      "loss": 0.6522,
+      "step": 356200
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 5.302855062926384e-05,
+      "loss": 0.6511,
+      "step": 356400
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 5.300230657538835e-05,
+      "loss": 0.6505,
+      "step": 356600
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 5.297593064184516e-05,
+      "loss": 0.654,
+      "step": 356800
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 5.294955470830196e-05,
+      "loss": 0.6482,
+      "step": 357000
+    },
+    {
+      "epoch": 0.47,
+      "eval_em_first_line_ratio": 0.51,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3966,
+      "eval_loss": 0.6570956707000732,
+      "eval_runtime": 224.7219,
+      "eval_samples_per_second": 44.499,
+      "eval_steps_per_second": 0.236,
+      "step": 357000
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 5.2923178774758765e-05,
+      "loss": 0.6487,
+      "step": 357200
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 5.289680284121556e-05,
+      "loss": 0.6528,
+      "step": 357400
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 5.287042690767237e-05,
+      "loss": 0.649,
+      "step": 357600
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 5.2844050974129165e-05,
+      "loss": 0.6528,
+      "step": 357800
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 5.281767504058597e-05,
+      "loss": 0.6508,
+      "step": 358000
+    },
+    {
+      "epoch": 0.47,
+      "eval_em_first_line_ratio": 0.5132,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3996,
+      "eval_loss": 0.6568164229393005,
+      "eval_runtime": 233.9822,
+      "eval_samples_per_second": 42.738,
+      "eval_steps_per_second": 0.227,
+      "step": 358000
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 5.279129910704277e-05,
+      "loss": 0.6462,
+      "step": 358200
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 5.276492317349957e-05,
+      "loss": 0.6491,
+      "step": 358400
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 5.273867911962409e-05,
+      "loss": 0.6474,
+      "step": 358600
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 5.271230318608089e-05,
+      "loss": 0.648,
+      "step": 358800
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 5.268592725253769e-05,
+      "loss": 0.6526,
+      "step": 359000
+    },
+    {
+      "epoch": 0.47,
+      "eval_em_first_line_ratio": 0.5151,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4014,
+      "eval_loss": 0.657264232635498,
+      "eval_runtime": 221.9337,
+      "eval_samples_per_second": 45.059,
+      "eval_steps_per_second": 0.239,
+      "step": 359000
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 5.2659551318994504e-05,
+      "loss": 0.6542,
+      "step": 359200
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 5.263317538545131e-05,
+      "loss": 0.6493,
+      "step": 359400
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 5.2606799451908106e-05,
+      "loss": 0.6487,
+      "step": 359600
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 5.2580423518364904e-05,
+      "loss": 0.6482,
+      "step": 359800
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 5.255404758482171e-05,
+      "loss": 0.651,
+      "step": 360000
+    },
+    {
+      "epoch": 0.47,
+      "eval_em_first_line_ratio": 0.5119,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3987,
+      "eval_loss": 0.6571521759033203,
+      "eval_runtime": 226.1585,
+      "eval_samples_per_second": 44.217,
+      "eval_steps_per_second": 0.234,
+      "step": 360000
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 5.2527671651278506e-05,
+      "loss": 0.6471,
+      "step": 360200
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 5.250129571773531e-05,
+      "loss": 0.6476,
+      "step": 360400
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 5.247491978419211e-05,
+      "loss": 0.6503,
+      "step": 360600
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 5.244854385064891e-05,
+      "loss": 0.6462,
+      "step": 360800
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 5.242216791710571e-05,
+      "loss": 0.6477,
+      "step": 361000
+    },
+    {
+      "epoch": 0.48,
+      "eval_em_first_line_ratio": 0.515,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4014,
+      "eval_loss": 0.6566711664199829,
+      "eval_runtime": 229.299,
+      "eval_samples_per_second": 43.611,
+      "eval_steps_per_second": 0.231,
+      "step": 361000
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 5.239579198356253e-05,
+      "loss": 0.6468,
+      "step": 361200
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 5.2369416050019326e-05,
+      "loss": 0.6476,
+      "step": 361400
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 5.234304011647613e-05,
+      "loss": 0.6483,
+      "step": 361600
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 5.231666418293293e-05,
+      "loss": 0.6486,
+      "step": 361800
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 5.229028824938973e-05,
+      "loss": 0.6504,
+      "step": 362000
+    },
+    {
+      "epoch": 0.48,
+      "eval_em_first_line_ratio": 0.5162,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4026,
+      "eval_loss": 0.6563349366188049,
+      "eval_runtime": 229.0813,
+      "eval_samples_per_second": 43.653,
+      "eval_steps_per_second": 0.231,
+      "step": 362000
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 5.226391231584653e-05,
+      "loss": 0.6498,
+      "step": 362200
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 5.2237536382303334e-05,
+      "loss": 0.6495,
+      "step": 362400
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 5.221116044876013e-05,
+      "loss": 0.6507,
+      "step": 362600
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 5.218491639488465e-05,
+      "loss": 0.6455,
+      "step": 362800
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 5.2158540461341457e-05,
+      "loss": 0.6471,
+      "step": 363000
+    },
+    {
+      "epoch": 0.48,
+      "eval_em_first_line_ratio": 0.5135,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4004,
+      "eval_loss": 0.6558201909065247,
+      "eval_runtime": 225.3072,
+      "eval_samples_per_second": 44.384,
+      "eval_steps_per_second": 0.235,
+      "step": 363000
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 5.2132296407465976e-05,
+      "loss": 0.6472,
+      "step": 363200
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 5.2105920473922774e-05,
+      "loss": 0.6441,
+      "step": 363400
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 5.207954454037958e-05,
+      "loss": 0.6478,
+      "step": 363600
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 5.2053168606836376e-05,
+      "loss": 0.6416,
+      "step": 363800
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 5.202679267329318e-05,
+      "loss": 0.6471,
+      "step": 364000
+    },
+    {
+      "epoch": 0.48,
+      "eval_em_first_line_ratio": 0.5125,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3988,
+      "eval_loss": 0.6561964750289917,
+      "eval_runtime": 225.9369,
+      "eval_samples_per_second": 44.26,
+      "eval_steps_per_second": 0.235,
+      "step": 364000
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 5.200041673974998e-05,
+      "loss": 0.6471,
+      "step": 364200
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 5.197404080620679e-05,
+      "loss": 0.648,
+      "step": 364400
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 5.1947664872663594e-05,
+      "loss": 0.6465,
+      "step": 364600
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 5.192128893912039e-05,
+      "loss": 0.6506,
+      "step": 364800
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 5.189504488524491e-05,
+      "loss": 0.6489,
+      "step": 365000
+    },
+    {
+      "epoch": 0.48,
+      "eval_em_first_line_ratio": 0.5122,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3994,
+      "eval_loss": 0.6555902361869812,
+      "eval_runtime": 223.5453,
+      "eval_samples_per_second": 44.734,
+      "eval_steps_per_second": 0.237,
+      "step": 365000
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 5.1868668951701716e-05,
+      "loss": 0.6448,
+      "step": 365200
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 5.1842293018158514e-05,
+      "loss": 0.6438,
+      "step": 365400
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 5.181591708461532e-05,
+      "loss": 0.6501,
+      "step": 365600
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 5.1789541151072116e-05,
+      "loss": 0.6481,
+      "step": 365800
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 5.176316521752892e-05,
+      "loss": 0.6462,
+      "step": 366000
+    },
+    {
+      "epoch": 0.48,
+      "eval_em_first_line_ratio": 0.5154,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4017,
+      "eval_loss": 0.6561769843101501,
+      "eval_runtime": 227.3377,
+      "eval_samples_per_second": 43.987,
+      "eval_steps_per_second": 0.233,
+      "step": 366000
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 5.173678928398572e-05,
+      "loss": 0.6488,
+      "step": 366200
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 5.171041335044252e-05,
+      "loss": 0.6459,
+      "step": 366400
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 5.168403741689932e-05,
+      "loss": 0.6432,
+      "step": 366600
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 5.165766148335614e-05,
+      "loss": 0.6493,
+      "step": 366800
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 5.1631285549812935e-05,
+      "loss": 0.6474,
+      "step": 367000
+    },
+    {
+      "epoch": 0.48,
+      "eval_em_first_line_ratio": 0.5146,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4017,
+      "eval_loss": 0.6558321118354797,
+      "eval_runtime": 227.6662,
+      "eval_samples_per_second": 43.924,
+      "eval_steps_per_second": 0.233,
+      "step": 367000
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 5.160490961626974e-05,
+      "loss": 0.6452,
+      "step": 367200
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 5.157853368272654e-05,
+      "loss": 0.6461,
+      "step": 367400
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 5.1552157749183335e-05,
+      "loss": 0.6483,
+      "step": 367600
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 5.152578181564014e-05,
+      "loss": 0.6513,
+      "step": 367800
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 5.149940588209694e-05,
+      "loss": 0.6504,
+      "step": 368000
+    },
+    {
+      "epoch": 0.49,
+      "eval_em_first_line_ratio": 0.5139,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4009,
+      "eval_loss": 0.6551929712295532,
+      "eval_runtime": 218.532,
+      "eval_samples_per_second": 45.76,
+      "eval_steps_per_second": 0.243,
+      "step": 368000
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 5.147302994855374e-05,
+      "loss": 0.6452,
+      "step": 368200
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 5.144665401501054e-05,
+      "loss": 0.6501,
+      "step": 368400
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 5.1420278081467344e-05,
+      "loss": 0.6455,
+      "step": 368600
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 5.1393902147924155e-05,
+      "loss": 0.6444,
+      "step": 368800
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 5.136752621438096e-05,
+      "loss": 0.6481,
+      "step": 369000
+    },
+    {
+      "epoch": 0.49,
+      "eval_em_first_line_ratio": 0.5145,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4012,
+      "eval_loss": 0.6557741165161133,
+      "eval_runtime": 230.7158,
+      "eval_samples_per_second": 43.343,
+      "eval_steps_per_second": 0.23,
+      "step": 369000
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 5.134115028083776e-05,
+      "loss": 0.6477,
+      "step": 369200
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 5.131490622696228e-05,
+      "loss": 0.648,
+      "step": 369400
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 5.128853029341908e-05,
+      "loss": 0.644,
+      "step": 369600
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 5.126215435987588e-05,
+      "loss": 0.647,
+      "step": 369800
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 5.1235778426332683e-05,
+      "loss": 0.6437,
+      "step": 370000
+    },
+    {
+      "epoch": 0.49,
+      "eval_em_first_line_ratio": 0.5155,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4018,
+      "eval_loss": 0.6548938751220703,
+      "eval_runtime": 233.6838,
+      "eval_samples_per_second": 42.793,
+      "eval_steps_per_second": 0.227,
+      "step": 370000
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 5.120940249278948e-05,
+      "loss": 0.6469,
+      "step": 370200
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 5.1183026559246286e-05,
+      "loss": 0.6453,
+      "step": 370400
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 5.115665062570308e-05,
+      "loss": 0.6441,
+      "step": 370600
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 5.113027469215989e-05,
+      "loss": 0.6475,
+      "step": 370800
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 5.1103898758616685e-05,
+      "loss": 0.6446,
+      "step": 371000
+    },
+    {
+      "epoch": 0.49,
+      "eval_em_first_line_ratio": 0.5154,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4011,
+      "eval_loss": 0.6553583741188049,
+      "eval_runtime": 230.2184,
+      "eval_samples_per_second": 43.437,
+      "eval_steps_per_second": 0.23,
+      "step": 371000
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 5.107752282507349e-05,
+      "loss": 0.6469,
+      "step": 371200
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 5.10511468915303e-05,
+      "loss": 0.6494,
+      "step": 371400
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 5.10247709579871e-05,
+      "loss": 0.6475,
+      "step": 371600
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 5.09983950244439e-05,
+      "loss": 0.6476,
+      "step": 371800
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 5.09720190909007e-05,
+      "loss": 0.6453,
+      "step": 372000
+    },
+    {
+      "epoch": 0.49,
+      "eval_em_first_line_ratio": 0.5142,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4006,
+      "eval_loss": 0.6551316380500793,
+      "eval_runtime": 259.572,
+      "eval_samples_per_second": 38.525,
+      "eval_steps_per_second": 0.204,
+      "step": 372000
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 5.0945643157357505e-05,
+      "loss": 0.6456,
+      "step": 372200
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 5.09192672238143e-05,
+      "loss": 0.6451,
+      "step": 372400
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 5.089302316993882e-05,
+      "loss": 0.6446,
+      "step": 372600
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 5.086664723639563e-05,
+      "loss": 0.6454,
+      "step": 372800
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 5.0840271302852425e-05,
+      "loss": 0.645,
+      "step": 373000
+    },
+    {
+      "epoch": 0.49,
+      "eval_em_first_line_ratio": 0.5141,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3996,
+      "eval_loss": 0.6534752249717712,
+      "eval_runtime": 225.0425,
+      "eval_samples_per_second": 44.436,
+      "eval_steps_per_second": 0.236,
+      "step": 373000
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 5.081389536930923e-05,
+      "loss": 0.6446,
+      "step": 373200
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 5.078751943576603e-05,
+      "loss": 0.6438,
+      "step": 373400
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 5.076114350222283e-05,
+      "loss": 0.6465,
+      "step": 373600
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 5.073476756867963e-05,
+      "loss": 0.6459,
+      "step": 373800
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 5.070839163513644e-05,
+      "loss": 0.6488,
+      "step": 374000
+    },
+    {
+      "epoch": 0.49,
+      "eval_em_first_line_ratio": 0.5156,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4014,
+      "eval_loss": 0.6547940373420715,
+      "eval_runtime": 222.1198,
+      "eval_samples_per_second": 45.021,
+      "eval_steps_per_second": 0.239,
+      "step": 374000
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 5.0682015701593244e-05,
+      "loss": 0.6448,
+      "step": 374200
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 5.065563976805004e-05,
+      "loss": 0.6475,
+      "step": 374400
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 5.0629263834506847e-05,
+      "loss": 0.6419,
+      "step": 374600
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 5.0602887900963644e-05,
+      "loss": 0.6439,
+      "step": 374800
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 5.057651196742045e-05,
+      "loss": 0.6454,
+      "step": 375000
+    },
+    {
+      "epoch": 0.49,
+      "eval_em_first_line_ratio": 0.5143,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4,
+      "eval_loss": 0.6544922590255737,
+      "eval_runtime": 225.4664,
+      "eval_samples_per_second": 44.352,
+      "eval_steps_per_second": 0.235,
+      "step": 375000
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 5.0550136033877246e-05,
+      "loss": 0.6403,
+      "step": 375200
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 5.052376010033405e-05,
+      "loss": 0.6477,
+      "step": 375400
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 5.049738416679085e-05,
+      "loss": 0.6416,
+      "step": 375600
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 5.0471008233247666e-05,
+      "loss": 0.6436,
+      "step": 375800
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 5.0444632299704464e-05,
+      "loss": 0.6415,
+      "step": 376000
+    },
+    {
+      "epoch": 0.5,
+      "eval_em_first_line_ratio": 0.5125,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3988,
+      "eval_loss": 0.6540663838386536,
+      "eval_runtime": 232.2045,
+      "eval_samples_per_second": 43.065,
+      "eval_steps_per_second": 0.228,
+      "step": 376000
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 5.041825636616127e-05,
+      "loss": 0.645,
+      "step": 376200
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 5.0391880432618066e-05,
+      "loss": 0.6446,
+      "step": 376400
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 5.0365504499074864e-05,
+      "loss": 0.6427,
+      "step": 376600
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 5.033912856553167e-05,
+      "loss": 0.6465,
+      "step": 376800
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 5.0312752631988466e-05,
+      "loss": 0.6453,
+      "step": 377000
+    },
+    {
+      "epoch": 0.5,
+      "eval_em_first_line_ratio": 0.5122,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3985,
+      "eval_loss": 0.6541827917098999,
+      "eval_runtime": 229.9273,
+      "eval_samples_per_second": 43.492,
+      "eval_steps_per_second": 0.231,
+      "step": 377000
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 5.028637669844527e-05,
+      "loss": 0.6442,
+      "step": 377200
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 5.026013264456979e-05,
+      "loss": 0.6426,
+      "step": 377400
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 5.023388859069431e-05,
+      "loss": 0.647,
+      "step": 377600
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 5.0207512657151115e-05,
+      "loss": 0.6415,
+      "step": 377800
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 5.018113672360791e-05,
+      "loss": 0.6472,
+      "step": 378000
+    },
+    {
+      "epoch": 0.5,
+      "eval_em_first_line_ratio": 0.513,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3995,
+      "eval_loss": 0.654106855392456,
+      "eval_runtime": 224.0993,
+      "eval_samples_per_second": 44.623,
+      "eval_steps_per_second": 0.237,
+      "step": 378000
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 5.0154760790064717e-05,
+      "loss": 0.6448,
+      "step": 378200
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 5.0128384856521514e-05,
+      "loss": 0.6445,
+      "step": 378400
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 5.010200892297832e-05,
+      "loss": 0.6437,
+      "step": 378600
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 5.0075632989435116e-05,
+      "loss": 0.6406,
+      "step": 378800
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 5.004925705589193e-05,
+      "loss": 0.6466,
+      "step": 379000
+    },
+    {
+      "epoch": 0.5,
+      "eval_em_first_line_ratio": 0.5148,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4011,
+      "eval_loss": 0.6539368033409119,
+      "eval_runtime": 227.133,
+      "eval_samples_per_second": 44.027,
+      "eval_steps_per_second": 0.233,
+      "step": 379000
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 5.002288112234873e-05,
+      "loss": 0.6444,
+      "step": 379200
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 4.999650518880553e-05,
+      "loss": 0.6449,
+      "step": 379400
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 4.9970129255262334e-05,
+      "loss": 0.6454,
+      "step": 379600
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 4.994375332171913e-05,
+      "loss": 0.6439,
+      "step": 379800
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 4.9917377388175936e-05,
+      "loss": 0.6431,
+      "step": 380000
+    },
+    {
+      "epoch": 0.5,
+      "eval_em_first_line_ratio": 0.5126,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3989,
+      "eval_loss": 0.6537491083145142,
+      "eval_runtime": 227.4224,
+      "eval_samples_per_second": 43.971,
+      "eval_steps_per_second": 0.233,
+      "step": 380000
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 4.9891001454632734e-05,
+      "loss": 0.646,
+      "step": 380200
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 4.9864625521089545e-05,
+      "loss": 0.6406,
+      "step": 380400
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 4.983824958754634e-05,
+      "loss": 0.6488,
+      "step": 380600
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 4.981187365400315e-05,
+      "loss": 0.648,
+      "step": 380800
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 4.9785497720459945e-05,
+      "loss": 0.6418,
+      "step": 381000
+    },
+    {
+      "epoch": 0.5,
+      "eval_em_first_line_ratio": 0.5124,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3991,
+      "eval_loss": 0.6530358195304871,
+      "eval_runtime": 229.6047,
+      "eval_samples_per_second": 43.553,
+      "eval_steps_per_second": 0.231,
+      "step": 381000
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 4.975912178691675e-05,
+      "loss": 0.6443,
+      "step": 381200
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 4.9732745853373553e-05,
+      "loss": 0.6425,
+      "step": 381400
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 4.970636991983035e-05,
+      "loss": 0.6425,
+      "step": 381600
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 4.9679993986287156e-05,
+      "loss": 0.6433,
+      "step": 381800
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 4.9653749932411676e-05,
+      "loss": 0.6425,
+      "step": 382000
+    },
+    {
+      "epoch": 0.5,
+      "eval_em_first_line_ratio": 0.5132,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3991,
+      "eval_loss": 0.6539551019668579,
+      "eval_runtime": 232.4945,
+      "eval_samples_per_second": 43.012,
+      "eval_steps_per_second": 0.228,
+      "step": 382000
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 4.962737399886847e-05,
+      "loss": 0.6437,
+      "step": 382200
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 4.960099806532528e-05,
+      "loss": 0.6464,
+      "step": 382400
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 4.9574622131782075e-05,
+      "loss": 0.6477,
+      "step": 382600
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 4.954824619823888e-05,
+      "loss": 0.6395,
+      "step": 382800
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 4.9521870264695684e-05,
+      "loss": 0.6449,
+      "step": 383000
+    },
+    {
+      "epoch": 0.51,
+      "eval_em_first_line_ratio": 0.5134,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3996,
+      "eval_loss": 0.6535948514938354,
+      "eval_runtime": 232.756,
+      "eval_samples_per_second": 42.963,
+      "eval_steps_per_second": 0.228,
+      "step": 383000
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 4.949549433115249e-05,
+      "loss": 0.6496,
+      "step": 383200
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 4.9469118397609286e-05,
+      "loss": 0.6426,
+      "step": 383400
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 4.944274246406609e-05,
+      "loss": 0.6417,
+      "step": 383600
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 4.941636653052289e-05,
+      "loss": 0.6421,
+      "step": 383800
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 4.939012247664741e-05,
+      "loss": 0.6465,
+      "step": 384000
+    },
+    {
+      "epoch": 0.51,
+      "eval_em_first_line_ratio": 0.5124,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.398,
+      "eval_loss": 0.6530749797821045,
+      "eval_runtime": 229.6434,
+      "eval_samples_per_second": 43.546,
+      "eval_steps_per_second": 0.231,
+      "step": 384000
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 4.936374654310421e-05,
+      "loss": 0.6446,
+      "step": 384200
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 4.933737060956101e-05,
+      "loss": 0.6369,
+      "step": 384400
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 4.931099467601782e-05,
+      "loss": 0.6473,
+      "step": 384600
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 4.928461874247462e-05,
+      "loss": 0.6437,
+      "step": 384800
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 4.9258242808931424e-05,
+      "loss": 0.6401,
+      "step": 385000
+    },
+    {
+      "epoch": 0.51,
+      "eval_em_first_line_ratio": 0.5142,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4002,
+      "eval_loss": 0.6523552536964417,
+      "eval_runtime": 228.1067,
+      "eval_samples_per_second": 43.839,
+      "eval_steps_per_second": 0.232,
+      "step": 385000
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 4.923186687538822e-05,
+      "loss": 0.6491,
+      "step": 385200
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 4.9205490941845026e-05,
+      "loss": 0.6432,
+      "step": 385400
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 4.917911500830183e-05,
+      "loss": 0.6462,
+      "step": 385600
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 4.915273907475863e-05,
+      "loss": 0.6423,
+      "step": 385800
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 4.912636314121543e-05,
+      "loss": 0.6425,
+      "step": 386000
+    },
+    {
+      "epoch": 0.51,
+      "eval_em_first_line_ratio": 0.5151,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4016,
+      "eval_loss": 0.652869462966919,
+      "eval_runtime": 236.2905,
+      "eval_samples_per_second": 42.321,
+      "eval_steps_per_second": 0.224,
+      "step": 386000
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 4.909998720767223e-05,
+      "loss": 0.6462,
+      "step": 386200
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 4.907361127412904e-05,
+      "loss": 0.6469,
+      "step": 386400
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 4.9047367220253554e-05,
+      "loss": 0.6444,
+      "step": 386600
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 4.902099128671035e-05,
+      "loss": 0.6479,
+      "step": 386800
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 4.8994615353167156e-05,
+      "loss": 0.6412,
+      "step": 387000
+    },
+    {
+      "epoch": 0.51,
+      "eval_em_first_line_ratio": 0.5145,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4005,
+      "eval_loss": 0.6527149677276611,
+      "eval_runtime": 232.8741,
+      "eval_samples_per_second": 42.942,
+      "eval_steps_per_second": 0.228,
+      "step": 387000
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 4.896823941962396e-05,
+      "loss": 0.6445,
+      "step": 387200
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 4.8941863486080765e-05,
+      "loss": 0.6417,
+      "step": 387400
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 4.891548755253756e-05,
+      "loss": 0.6462,
+      "step": 387600
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 4.888911161899437e-05,
+      "loss": 0.6432,
+      "step": 387800
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 4.886273568545117e-05,
+      "loss": 0.6407,
+      "step": 388000
+    },
+    {
+      "epoch": 0.51,
+      "eval_em_first_line_ratio": 0.5155,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4018,
+      "eval_loss": 0.6521013379096985,
+      "eval_runtime": 225.514,
+      "eval_samples_per_second": 44.343,
+      "eval_steps_per_second": 0.235,
+      "step": 388000
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 4.8836359751907976e-05,
+      "loss": 0.6409,
+      "step": 388200
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 4.8809983818364774e-05,
+      "loss": 0.6454,
+      "step": 388400
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 4.878360788482158e-05,
+      "loss": 0.645,
+      "step": 388600
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 4.8757231951278376e-05,
+      "loss": 0.6438,
+      "step": 388800
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 4.873085601773518e-05,
+      "loss": 0.642,
+      "step": 389000
+    },
+    {
+      "epoch": 0.51,
+      "eval_em_first_line_ratio": 0.5149,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4006,
+      "eval_loss": 0.6520912647247314,
+      "eval_runtime": 232.2408,
+      "eval_samples_per_second": 43.059,
+      "eval_steps_per_second": 0.228,
+      "step": 389000
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 4.8704480084191985e-05,
+      "loss": 0.645,
+      "step": 389200
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 4.867810415064878e-05,
+      "loss": 0.645,
+      "step": 389400
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 4.865172821710559e-05,
+      "loss": 0.6424,
+      "step": 389600
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 4.8625352283562384e-05,
+      "loss": 0.6432,
+      "step": 389800
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 4.8598976350019195e-05,
+      "loss": 0.6383,
+      "step": 390000
+    },
+    {
+      "epoch": 0.51,
+      "eval_em_first_line_ratio": 0.5154,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4014,
+      "eval_loss": 0.6519700884819031,
+      "eval_runtime": 353.5291,
+      "eval_samples_per_second": 28.286,
+      "eval_steps_per_second": 0.15,
+      "step": 390000
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 4.857260041647599e-05,
+      "loss": 0.6367,
+      "step": 390200
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 4.85462244829328e-05,
+      "loss": 0.6392,
+      "step": 390400
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 4.8519848549389595e-05,
+      "loss": 0.6437,
+      "step": 390600
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 4.8493604495514115e-05,
+      "loss": 0.6432,
+      "step": 390800
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 4.846722856197092e-05,
+      "loss": 0.6423,
+      "step": 391000
+    },
+    {
+      "epoch": 0.52,
+      "eval_em_first_line_ratio": 0.5178,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4039,
+      "eval_loss": 0.6519811749458313,
+      "eval_runtime": 350.5596,
+      "eval_samples_per_second": 28.526,
+      "eval_steps_per_second": 0.151,
+      "step": 391000
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 4.844085262842772e-05,
+      "loss": 0.6449,
+      "step": 391200
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 4.841447669488452e-05,
+      "loss": 0.6427,
+      "step": 391400
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 4.8388100761341326e-05,
+      "loss": 0.6444,
+      "step": 391600
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 4.836172482779813e-05,
+      "loss": 0.6394,
+      "step": 391800
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 4.833534889425493e-05,
+      "loss": 0.6369,
+      "step": 392000
+    },
+    {
+      "epoch": 0.52,
+      "eval_em_first_line_ratio": 0.5124,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3988,
+      "eval_loss": 0.6526699066162109,
+      "eval_runtime": 338.9751,
+      "eval_samples_per_second": 29.501,
+      "eval_steps_per_second": 0.156,
+      "step": 392000
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 4.8308972960711726e-05,
+      "loss": 0.6381,
+      "step": 392200
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 4.828259702716853e-05,
+      "loss": 0.6398,
+      "step": 392400
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 4.825635297329305e-05,
+      "loss": 0.643,
+      "step": 392600
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 4.8229977039749855e-05,
+      "loss": 0.6442,
+      "step": 392800
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 4.820360110620665e-05,
+      "loss": 0.6427,
+      "step": 393000
+    },
+    {
+      "epoch": 0.52,
+      "eval_em_first_line_ratio": 0.5148,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4013,
+      "eval_loss": 0.651681125164032,
+      "eval_runtime": 339.0258,
+      "eval_samples_per_second": 29.496,
+      "eval_steps_per_second": 0.156,
+      "step": 393000
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 4.817722517266346e-05,
+      "loss": 0.6396,
+      "step": 393200
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 4.815084923912026e-05,
+      "loss": 0.6403,
+      "step": 393400
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 4.812447330557706e-05,
+      "loss": 0.641,
+      "step": 393600
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 4.809809737203386e-05,
+      "loss": 0.6382,
+      "step": 393800
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 4.807172143849066e-05,
+      "loss": 0.64,
+      "step": 394000
+    },
+    {
+      "epoch": 0.52,
+      "eval_em_first_line_ratio": 0.5134,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.3999,
+      "eval_loss": 0.6519544720649719,
+      "eval_runtime": 342.5021,
+      "eval_samples_per_second": 29.197,
+      "eval_steps_per_second": 0.155,
+      "step": 394000
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 4.804534550494747e-05,
+      "loss": 0.6397,
+      "step": 394200
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 4.801896957140427e-05,
+      "loss": 0.6396,
+      "step": 394400
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 4.7992593637861074e-05,
+      "loss": 0.6394,
+      "step": 394600
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 4.796621770431787e-05,
+      "loss": 0.6455,
+      "step": 394800
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 4.7939841770774676e-05,
+      "loss": 0.6459,
+      "step": 395000
+    },
+    {
+      "epoch": 0.52,
+      "eval_em_first_line_ratio": 0.5154,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4008,
+      "eval_loss": 0.6515817642211914,
+      "eval_runtime": 308.0943,
+      "eval_samples_per_second": 32.458,
+      "eval_steps_per_second": 0.172,
+      "step": 395000
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 4.791346583723148e-05,
+      "loss": 0.6449,
+      "step": 395200
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 4.788708990368828e-05,
+      "loss": 0.6362,
+      "step": 395400
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 4.786071397014508e-05,
+      "loss": 0.6438,
+      "step": 395600
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 4.783433803660188e-05,
+      "loss": 0.6411,
+      "step": 395800
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 4.780796210305869e-05,
+      "loss": 0.6414,
+      "step": 396000
+    },
+    {
+      "epoch": 0.52,
+      "eval_em_first_line_ratio": 0.5161,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4018,
+      "eval_loss": 0.6517772674560547,
+      "eval_runtime": 315.0672,
+      "eval_samples_per_second": 31.739,
+      "eval_steps_per_second": 0.168,
+      "step": 396000
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 4.778158616951549e-05,
+      "loss": 0.6393,
+      "step": 396200
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 4.7755210235972294e-05,
+      "loss": 0.6418,
+      "step": 396400
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 4.772883430242909e-05,
+      "loss": 0.6425,
+      "step": 396600
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 4.7702458368885896e-05,
+      "loss": 0.6391,
+      "step": 396800
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 4.76760824353427e-05,
+      "loss": 0.6377,
+      "step": 397000
+    },
+    {
+      "epoch": 0.52,
+      "eval_em_first_line_ratio": 0.5195,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4051,
+      "eval_loss": 0.6516339182853699,
+      "eval_runtime": 225.4595,
+      "eval_samples_per_second": 44.354,
+      "eval_steps_per_second": 0.235,
+      "step": 397000
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 4.7649706501799505e-05,
+      "loss": 0.6437,
+      "step": 397200
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 4.76233305682563e-05,
+      "loss": 0.6404,
+      "step": 397400
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 4.759708651438082e-05,
+      "loss": 0.6394,
+      "step": 397600
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 4.7570710580837627e-05,
+      "loss": 0.6376,
+      "step": 397800
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 4.7544334647294424e-05,
+      "loss": 0.6416,
+      "step": 398000
+    },
+    {
+      "epoch": 0.52,
+      "eval_em_first_line_ratio": 0.5163,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4022,
+      "eval_loss": 0.651257336139679,
+      "eval_runtime": 225.4335,
+      "eval_samples_per_second": 44.359,
+      "eval_steps_per_second": 0.235,
+      "step": 398000
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 4.751795871375123e-05,
+      "loss": 0.6423,
+      "step": 398200
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 4.7491582780208026e-05,
+      "loss": 0.6391,
+      "step": 398400
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 4.746520684666483e-05,
+      "loss": 0.64,
+      "step": 398600
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 4.7438830913121635e-05,
+      "loss": 0.6417,
+      "step": 398800
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 4.741245497957843e-05,
+      "loss": 0.6414,
+      "step": 399000
+    },
+    {
+      "epoch": 0.53,
+      "eval_em_first_line_ratio": 0.5165,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4029,
+      "eval_loss": 0.6507397294044495,
+      "eval_runtime": 229.3378,
+      "eval_samples_per_second": 43.604,
+      "eval_steps_per_second": 0.231,
+      "step": 399000
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 4.738607904603524e-05,
+      "loss": 0.6411,
+      "step": 399200
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 4.7359703112492035e-05,
+      "loss": 0.6395,
+      "step": 399400
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 4.7333327178948846e-05,
+      "loss": 0.6395,
+      "step": 399600
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 4.7306951245405644e-05,
+      "loss": 0.6386,
+      "step": 399800
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 4.728057531186245e-05,
+      "loss": 0.6399,
+      "step": 400000
+    },
+    {
+      "epoch": 0.53,
+      "eval_em_first_line_ratio": 0.5157,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4017,
+      "eval_loss": 0.6509023904800415,
+      "eval_runtime": 224.2893,
+      "eval_samples_per_second": 44.585,
+      "eval_steps_per_second": 0.236,
+      "step": 400000
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 4.7254199378319246e-05,
+      "loss": 0.6435,
+      "step": 400200
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 4.722782344477605e-05,
+      "loss": 0.6377,
+      "step": 400400
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 4.7201447511232855e-05,
+      "loss": 0.6399,
+      "step": 400600
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 4.717507157768966e-05,
+      "loss": 0.6414,
+      "step": 400800
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 4.714895940348189e-05,
+      "loss": 0.6379,
+      "step": 401000
+    },
+    {
+      "epoch": 0.53,
+      "eval_em_first_line_ratio": 0.5154,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4011,
+      "eval_loss": 0.6499211192131042,
+      "eval_runtime": 226.6797,
+      "eval_samples_per_second": 44.115,
+      "eval_steps_per_second": 0.234,
+      "step": 401000
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 4.712258346993869e-05,
+      "loss": 0.6408,
+      "step": 401200
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 4.709620753639549e-05,
+      "loss": 0.635,
+      "step": 401400
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 4.7069831602852294e-05,
+      "loss": 0.6424,
+      "step": 401600
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 4.70434556693091e-05,
+      "loss": 0.6407,
+      "step": 401800
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 4.70170797357659e-05,
+      "loss": 0.6386,
+      "step": 402000
+    },
+    {
+      "epoch": 0.53,
+      "eval_em_first_line_ratio": 0.5147,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4009,
+      "eval_loss": 0.6503856182098389,
+      "eval_runtime": 231.6413,
+      "eval_samples_per_second": 43.17,
+      "eval_steps_per_second": 0.229,
+      "step": 402000
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 4.69907038022227e-05,
+      "loss": 0.6433,
+      "step": 402200
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 4.6964327868679505e-05,
+      "loss": 0.6417,
+      "step": 402400
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 4.69379519351363e-05,
+      "loss": 0.6413,
+      "step": 402600
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 4.691157600159311e-05,
+      "loss": 0.6442,
+      "step": 402800
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 4.688520006804991e-05,
+      "loss": 0.6357,
+      "step": 403000
+    },
+    {
+      "epoch": 0.53,
+      "eval_em_first_line_ratio": 0.5172,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4028,
+      "eval_loss": 0.6498135924339294,
+      "eval_runtime": 230.762,
+      "eval_samples_per_second": 43.335,
+      "eval_steps_per_second": 0.23,
+      "step": 403000
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 4.6858956014174425e-05,
+      "loss": 0.6378,
+      "step": 403200
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 4.6832580080631236e-05,
+      "loss": 0.6376,
+      "step": 403400
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 4.6806204147088034e-05,
+      "loss": 0.6446,
+      "step": 403600
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 4.677982821354484e-05,
+      "loss": 0.639,
+      "step": 403800
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 4.6753452280001636e-05,
+      "loss": 0.6363,
+      "step": 404000
+    },
+    {
+      "epoch": 0.53,
+      "eval_em_first_line_ratio": 0.5163,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4025,
+      "eval_loss": 0.6499820947647095,
+      "eval_runtime": 221.6516,
+      "eval_samples_per_second": 45.116,
+      "eval_steps_per_second": 0.239,
+      "step": 404000
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 4.672707634645844e-05,
+      "loss": 0.6357,
+      "step": 404200
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 4.6700700412915245e-05,
+      "loss": 0.639,
+      "step": 404400
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 4.667432447937204e-05,
+      "loss": 0.6429,
+      "step": 404600
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 4.664794854582885e-05,
+      "loss": 0.6419,
+      "step": 404800
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 4.6621572612285644e-05,
+      "loss": 0.6422,
+      "step": 405000
+    },
+    {
+      "epoch": 0.53,
+      "eval_em_first_line_ratio": 0.5165,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4028,
+      "eval_loss": 0.6499873995780945,
+      "eval_runtime": 230.938,
+      "eval_samples_per_second": 43.302,
+      "eval_steps_per_second": 0.229,
+      "step": 405000
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 4.6595196678742456e-05,
+      "loss": 0.6429,
+      "step": 405200
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 4.656895262486697e-05,
+      "loss": 0.6404,
+      "step": 405400
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 4.6542576691323766e-05,
+      "loss": 0.6434,
+      "step": 405600
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 4.651620075778057e-05,
+      "loss": 0.6428,
+      "step": 405800
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 4.6489824824237375e-05,
+      "loss": 0.6432,
+      "step": 406000
+    },
+    {
+      "epoch": 0.54,
+      "eval_em_first_line_ratio": 0.5172,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4033,
+      "eval_loss": 0.6491650342941284,
+      "eval_runtime": 229.6618,
+      "eval_samples_per_second": 43.542,
+      "eval_steps_per_second": 0.231,
+      "step": 406000
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 4.646344889069418e-05,
+      "loss": 0.6361,
+      "step": 406200
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 4.643707295715098e-05,
+      "loss": 0.6405,
+      "step": 406400
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 4.641069702360778e-05,
+      "loss": 0.6376,
+      "step": 406600
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 4.6384321090064586e-05,
+      "loss": 0.6385,
+      "step": 406800
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 4.635794515652139e-05,
+      "loss": 0.6439,
+      "step": 407000
+    },
+    {
+      "epoch": 0.54,
+      "eval_em_first_line_ratio": 0.5177,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4037,
+      "eval_loss": 0.6507595181465149,
+      "eval_runtime": 229.2793,
+      "eval_samples_per_second": 43.615,
+      "eval_steps_per_second": 0.231,
+      "step": 407000
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 4.633156922297819e-05,
+      "loss": 0.6401,
+      "step": 407200
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 4.630519328943499e-05,
+      "loss": 0.6417,
+      "step": 407400
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 4.627881735589179e-05,
+      "loss": 0.6422,
+      "step": 407600
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 4.6252441422348595e-05,
+      "loss": 0.6383,
+      "step": 407800
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 4.62260654888054e-05,
+      "loss": 0.6379,
+      "step": 408000
+    },
+    {
+      "epoch": 0.54,
+      "eval_em_first_line_ratio": 0.517,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4029,
+      "eval_loss": 0.6494207382202148,
+      "eval_runtime": 232.7505,
+      "eval_samples_per_second": 42.964,
+      "eval_steps_per_second": 0.228,
+      "step": 408000
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 4.61996895552622e-05,
+      "loss": 0.6393,
+      "step": 408200
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 4.617344550138672e-05,
+      "loss": 0.6407,
+      "step": 408400
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 4.614706956784352e-05,
+      "loss": 0.6433,
+      "step": 408600
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 4.612069363430032e-05,
+      "loss": 0.6379,
+      "step": 408800
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 4.609431770075712e-05,
+      "loss": 0.64,
+      "step": 409000
+    },
+    {
+      "epoch": 0.54,
+      "eval_em_first_line_ratio": 0.5161,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4023,
+      "eval_loss": 0.64903324842453,
+      "eval_runtime": 226.5198,
+      "eval_samples_per_second": 44.146,
+      "eval_steps_per_second": 0.234,
+      "step": 409000
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 4.606794176721392e-05,
+      "loss": 0.6436,
+      "step": 409200
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 4.604156583367073e-05,
+      "loss": 0.6394,
+      "step": 409400
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 4.601518990012753e-05,
+      "loss": 0.6387,
+      "step": 409600
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 4.5988813966584334e-05,
+      "loss": 0.6356,
+      "step": 409800
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 4.596243803304113e-05,
+      "loss": 0.6382,
+      "step": 410000
+    },
+    {
+      "epoch": 0.54,
+      "eval_em_first_line_ratio": 0.5157,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.402,
+      "eval_loss": 0.6487342119216919,
+      "eval_runtime": 232.172,
+      "eval_samples_per_second": 43.072,
+      "eval_steps_per_second": 0.228,
+      "step": 410000
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 4.5936062099497936e-05,
+      "loss": 0.6417,
+      "step": 410200
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 4.590968616595474e-05,
+      "loss": 0.6396,
+      "step": 410400
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 4.588331023241154e-05,
+      "loss": 0.6388,
+      "step": 410600
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 4.585693429886834e-05,
+      "loss": 0.6352,
+      "step": 410800
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 4.583055836532514e-05,
+      "loss": 0.6376,
+      "step": 411000
+    },
+    {
+      "epoch": 0.54,
+      "eval_em_first_line_ratio": 0.514,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4006,
+      "eval_loss": 0.6481953263282776,
+      "eval_runtime": 228.4487,
+      "eval_samples_per_second": 43.774,
+      "eval_steps_per_second": 0.232,
+      "step": 411000
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 4.5804182431781945e-05,
+      "loss": 0.6355,
+      "step": 411200
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 4.5777938377906465e-05,
+      "loss": 0.6389,
+      "step": 411400
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 4.575156244436327e-05,
+      "loss": 0.6339,
+      "step": 411600
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 4.572518651082007e-05,
+      "loss": 0.6388,
+      "step": 411800
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 4.569881057727687e-05,
+      "loss": 0.6389,
+      "step": 412000
+    },
+    {
+      "epoch": 0.54,
+      "eval_em_first_line_ratio": 0.5168,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4035,
+      "eval_loss": 0.6489261984825134,
+      "eval_runtime": 230.9234,
+      "eval_samples_per_second": 43.304,
+      "eval_steps_per_second": 0.23,
+      "step": 412000
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 4.5672434643733676e-05,
+      "loss": 0.6355,
+      "step": 412200
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 4.5646058710190473e-05,
+      "loss": 0.6438,
+      "step": 412400
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 4.561968277664728e-05,
+      "loss": 0.6378,
+      "step": 412600
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 4.5593306843104075e-05,
+      "loss": 0.6385,
+      "step": 412800
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 4.556693090956089e-05,
+      "loss": 0.6371,
+      "step": 413000
+    },
+    {
+      "epoch": 0.54,
+      "eval_em_first_line_ratio": 0.518,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4042,
+      "eval_loss": 0.6482269763946533,
+      "eval_runtime": 222.9957,
+      "eval_samples_per_second": 44.844,
+      "eval_steps_per_second": 0.238,
+      "step": 413000
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 4.5540554976017684e-05,
+      "loss": 0.6402,
+      "step": 413200
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 4.55143109221422e-05,
+      "loss": 0.6408,
+      "step": 413400
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 4.548793498859901e-05,
+      "loss": 0.6423,
+      "step": 413600
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 4.5461559055055806e-05,
+      "loss": 0.6378,
+      "step": 413800
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 4.543518312151261e-05,
+      "loss": 0.6383,
+      "step": 414000
+    },
+    {
+      "epoch": 0.55,
+      "eval_em_first_line_ratio": 0.5154,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4018,
+      "eval_loss": 0.6483412384986877,
+      "eval_runtime": 224.7499,
+      "eval_samples_per_second": 44.494,
+      "eval_steps_per_second": 0.236,
+      "step": 414000
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 4.540880718796941e-05,
+      "loss": 0.6372,
+      "step": 414200
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 4.538243125442621e-05,
+      "loss": 0.6362,
+      "step": 414400
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 4.535605532088302e-05,
+      "loss": 0.6347,
+      "step": 414600
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 4.532967938733982e-05,
+      "loss": 0.6422,
+      "step": 414800
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 4.530330345379662e-05,
+      "loss": 0.6365,
+      "step": 415000
+    },
+    {
+      "epoch": 0.55,
+      "eval_em_first_line_ratio": 0.5161,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4019,
+      "eval_loss": 0.6485750079154968,
+      "eval_runtime": 220.7892,
+      "eval_samples_per_second": 45.292,
+      "eval_steps_per_second": 0.24,
+      "step": 415000
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 4.527705939992114e-05,
+      "loss": 0.638,
+      "step": 415200
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 4.5250683466377944e-05,
+      "loss": 0.6389,
+      "step": 415400
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 4.522430753283474e-05,
+      "loss": 0.6389,
+      "step": 415600
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 4.5197931599291546e-05,
+      "loss": 0.6339,
+      "step": 415800
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 4.5171555665748344e-05,
+      "loss": 0.6339,
+      "step": 416000
+    },
+    {
+      "epoch": 0.55,
+      "eval_em_first_line_ratio": 0.5186,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4045,
+      "eval_loss": 0.6488232612609863,
+      "eval_runtime": 224.0827,
+      "eval_samples_per_second": 44.626,
+      "eval_steps_per_second": 0.237,
+      "step": 416000
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 4.514517973220515e-05,
+      "loss": 0.6359,
+      "step": 416200
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 4.511880379866195e-05,
+      "loss": 0.6397,
+      "step": 416400
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 4.509242786511875e-05,
+      "loss": 0.6383,
+      "step": 416600
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 4.5066051931575554e-05,
+      "loss": 0.6369,
+      "step": 416800
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 4.503967599803236e-05,
+      "loss": 0.6366,
+      "step": 417000
+    },
+    {
+      "epoch": 0.55,
+      "eval_em_first_line_ratio": 0.5161,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4023,
+      "eval_loss": 0.6483793258666992,
+      "eval_runtime": 224.8069,
+      "eval_samples_per_second": 44.483,
+      "eval_steps_per_second": 0.236,
+      "step": 417000
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 4.501330006448916e-05,
+      "loss": 0.6386,
+      "step": 417200
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 4.498692413094596e-05,
+      "loss": 0.6366,
+      "step": 417400
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 4.4960548197402765e-05,
+      "loss": 0.6369,
+      "step": 417600
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 4.493417226385956e-05,
+      "loss": 0.6388,
+      "step": 417800
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 4.490779633031637e-05,
+      "loss": 0.6354,
+      "step": 418000
+    },
+    {
+      "epoch": 0.55,
+      "eval_em_first_line_ratio": 0.5163,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4026,
+      "eval_loss": 0.6475383639335632,
+      "eval_runtime": 223.8212,
+      "eval_samples_per_second": 44.679,
+      "eval_steps_per_second": 0.237,
+      "step": 418000
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 4.488142039677317e-05,
+      "loss": 0.6349,
+      "step": 418200
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 4.485504446322997e-05,
+      "loss": 0.6354,
+      "step": 418400
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 4.4828668529686774e-05,
+      "loss": 0.6326,
+      "step": 418600
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 4.480229259614357e-05,
+      "loss": 0.6324,
+      "step": 418800
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 4.477591666260038e-05,
+      "loss": 0.6353,
+      "step": 419000
+    },
+    {
+      "epoch": 0.55,
+      "eval_em_first_line_ratio": 0.5176,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4036,
+      "eval_loss": 0.6480697393417358,
+      "eval_runtime": 222.8201,
+      "eval_samples_per_second": 44.879,
+      "eval_steps_per_second": 0.238,
+      "step": 419000
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 4.474954072905718e-05,
+      "loss": 0.6381,
+      "step": 419200
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 4.4723164795513985e-05,
+      "loss": 0.635,
+      "step": 419400
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 4.469678886197078e-05,
+      "loss": 0.64,
+      "step": 419600
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 4.46705448080953e-05,
+      "loss": 0.6371,
+      "step": 419800
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 4.464416887455211e-05,
+      "loss": 0.6401,
+      "step": 420000
+    },
+    {
+      "epoch": 0.55,
+      "eval_em_first_line_ratio": 0.5155,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4037,
+      "eval_loss": 0.648481011390686,
+      "eval_runtime": 228.6172,
+      "eval_samples_per_second": 43.741,
+      "eval_steps_per_second": 0.232,
+      "step": 420000
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 4.4617792941008905e-05,
+      "loss": 0.6369,
+      "step": 420200
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 4.459141700746571e-05,
+      "loss": 0.6359,
+      "step": 420400
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 4.456504107392251e-05,
+      "loss": 0.6348,
+      "step": 420600
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 4.453866514037932e-05,
+      "loss": 0.6373,
+      "step": 420800
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 4.4512289206836115e-05,
+      "loss": 0.6387,
+      "step": 421000
+    },
+    {
+      "epoch": 0.56,
+      "eval_em_first_line_ratio": 0.5178,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4041,
+      "eval_loss": 0.6487793922424316,
+      "eval_runtime": 225.6194,
+      "eval_samples_per_second": 44.322,
+      "eval_steps_per_second": 0.235,
+      "step": 421000
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 4.448591327329292e-05,
+      "loss": 0.6349,
+      "step": 421200
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 4.445953733974972e-05,
+      "loss": 0.6365,
+      "step": 421400
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 4.443316140620652e-05,
+      "loss": 0.637,
+      "step": 421600
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 4.4406785472663326e-05,
+      "loss": 0.634,
+      "step": 421800
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 4.438054141878784e-05,
+      "loss": 0.6397,
+      "step": 422000
+    },
+    {
+      "epoch": 0.56,
+      "eval_em_first_line_ratio": 0.5166,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4023,
+      "eval_loss": 0.647998571395874,
+      "eval_runtime": 229.6812,
+      "eval_samples_per_second": 43.539,
+      "eval_steps_per_second": 0.231,
+      "step": 422000
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 4.435416548524465e-05,
+      "loss": 0.6296,
+      "step": 422200
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 4.432778955170145e-05,
+      "loss": 0.6341,
+      "step": 422400
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 4.430141361815825e-05,
+      "loss": 0.6337,
+      "step": 422600
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 4.427503768461505e-05,
+      "loss": 0.6367,
+      "step": 422800
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 4.4248661751071855e-05,
+      "loss": 0.6353,
+      "step": 423000
+    },
+    {
+      "epoch": 0.56,
+      "eval_em_first_line_ratio": 0.5164,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4022,
+      "eval_loss": 0.6474470496177673,
+      "eval_runtime": 227.8401,
+      "eval_samples_per_second": 43.89,
+      "eval_steps_per_second": 0.233,
+      "step": 423000
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 4.422228581752866e-05,
+      "loss": 0.6367,
+      "step": 423200
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 4.419590988398546e-05,
+      "loss": 0.6383,
+      "step": 423400
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 4.416953395044226e-05,
+      "loss": 0.6378,
+      "step": 423600
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 4.414315801689906e-05,
+      "loss": 0.6356,
+      "step": 423800
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 4.411678208335587e-05,
+      "loss": 0.6391,
+      "step": 424000
+    },
+    {
+      "epoch": 0.56,
+      "eval_em_first_line_ratio": 0.5181,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4041,
+      "eval_loss": 0.6475912928581238,
+      "eval_runtime": 225.2713,
+      "eval_samples_per_second": 44.391,
+      "eval_steps_per_second": 0.235,
+      "step": 424000
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 4.409040614981267e-05,
+      "loss": 0.6417,
+      "step": 424200
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 4.406403021626947e-05,
+      "loss": 0.6419,
+      "step": 424400
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 4.4037786162393986e-05,
+      "loss": 0.6345,
+      "step": 424600
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 4.401141022885079e-05,
+      "loss": 0.6364,
+      "step": 424800
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 4.3985034295307594e-05,
+      "loss": 0.6389,
+      "step": 425000
+    },
+    {
+      "epoch": 0.56,
+      "eval_em_first_line_ratio": 0.5172,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4037,
+      "eval_loss": 0.6464712619781494,
+      "eval_runtime": 224.3412,
+      "eval_samples_per_second": 44.575,
+      "eval_steps_per_second": 0.236,
+      "step": 425000
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 4.395865836176439e-05,
+      "loss": 0.6321,
+      "step": 425200
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 4.3932282428221196e-05,
+      "loss": 0.6342,
+      "step": 425400
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 4.3905906494678e-05,
+      "loss": 0.6368,
+      "step": 425600
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 4.3879662440802514e-05,
+      "loss": 0.6358,
+      "step": 425800
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 4.385328650725932e-05,
+      "loss": 0.6327,
+      "step": 426000
+    },
+    {
+      "epoch": 0.56,
+      "eval_em_first_line_ratio": 0.5148,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4018,
+      "eval_loss": 0.647316038608551,
+      "eval_runtime": 232.3867,
+      "eval_samples_per_second": 43.032,
+      "eval_steps_per_second": 0.228,
+      "step": 426000
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 4.3826910573716116e-05,
+      "loss": 0.6323,
+      "step": 426200
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 4.380053464017293e-05,
+      "loss": 0.6342,
+      "step": 426400
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 4.3774158706629725e-05,
+      "loss": 0.6346,
+      "step": 426600
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 4.374778277308653e-05,
+      "loss": 0.6329,
+      "step": 426800
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 4.372140683954333e-05,
+      "loss": 0.6316,
+      "step": 427000
+    },
+    {
+      "epoch": 0.56,
+      "eval_em_first_line_ratio": 0.5144,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4009,
+      "eval_loss": 0.6470224857330322,
+      "eval_runtime": 230.3823,
+      "eval_samples_per_second": 43.406,
+      "eval_steps_per_second": 0.23,
+      "step": 427000
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 4.369503090600013e-05,
+      "loss": 0.6315,
+      "step": 427200
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 4.3668654972456936e-05,
+      "loss": 0.6369,
+      "step": 427400
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 4.3642279038913734e-05,
+      "loss": 0.6358,
+      "step": 427600
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 4.361590310537054e-05,
+      "loss": 0.6392,
+      "step": 427800
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 4.3589527171827336e-05,
+      "loss": 0.6347,
+      "step": 428000
+    },
+    {
+      "epoch": 0.56,
+      "eval_em_first_line_ratio": 0.5173,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4029,
+      "eval_loss": 0.6464382410049438,
+      "eval_runtime": 225.7366,
+      "eval_samples_per_second": 44.299,
+      "eval_steps_per_second": 0.235,
+      "step": 428000
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 4.356315123828415e-05,
+      "loss": 0.6359,
+      "step": 428200
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 4.3536775304740944e-05,
+      "loss": 0.6344,
+      "step": 428400
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 4.351039937119775e-05,
+      "loss": 0.639,
+      "step": 428600
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 4.3484023437654547e-05,
+      "loss": 0.6354,
+      "step": 428800
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 4.345764750411135e-05,
+      "loss": 0.6306,
+      "step": 429000
+    },
+    {
+      "epoch": 0.57,
+      "eval_em_first_line_ratio": 0.5177,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4036,
+      "eval_loss": 0.6467543840408325,
+      "eval_runtime": 226.8559,
+      "eval_samples_per_second": 44.081,
+      "eval_steps_per_second": 0.234,
+      "step": 429000
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 4.3431271570568155e-05,
+      "loss": 0.6306,
+      "step": 429200
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 4.340502751669267e-05,
+      "loss": 0.6314,
+      "step": 429400
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 4.337865158314947e-05,
+      "loss": 0.6337,
+      "step": 429600
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 4.335227564960628e-05,
+      "loss": 0.6336,
+      "step": 429800
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 4.332589971606308e-05,
+      "loss": 0.6362,
+      "step": 430000
+    },
+    {
+      "epoch": 0.57,
+      "eval_em_first_line_ratio": 0.5172,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4029,
+      "eval_loss": 0.6464583277702332,
+      "eval_runtime": 229.7697,
+      "eval_samples_per_second": 43.522,
+      "eval_steps_per_second": 0.231,
+      "step": 430000
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 4.329952378251988e-05,
+      "loss": 0.6385,
+      "step": 430200
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 4.3273147848976684e-05,
+      "loss": 0.6348,
+      "step": 430400
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 4.324677191543348e-05,
+      "loss": 0.6315,
+      "step": 430600
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 4.3220395981890286e-05,
+      "loss": 0.6347,
+      "step": 430800
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 4.319402004834709e-05,
+      "loss": 0.6327,
+      "step": 431000
+    },
+    {
+      "epoch": 0.57,
+      "eval_em_first_line_ratio": 0.5179,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4035,
+      "eval_loss": 0.6469618082046509,
+      "eval_runtime": 221.3223,
+      "eval_samples_per_second": 45.183,
+      "eval_steps_per_second": 0.239,
+      "step": 431000
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 4.316764411480389e-05,
+      "loss": 0.6361,
+      "step": 431200
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 4.314126818126069e-05,
+      "loss": 0.636,
+      "step": 431400
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 4.311489224771749e-05,
+      "loss": 0.6373,
+      "step": 431600
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 4.30885163141743e-05,
+      "loss": 0.6335,
+      "step": 431800
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 4.30621403806311e-05,
+      "loss": 0.6346,
+      "step": 432000
+    },
+    {
+      "epoch": 0.57,
+      "eval_em_first_line_ratio": 0.5179,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4035,
+      "eval_loss": 0.6469694375991821,
+      "eval_runtime": 231.1769,
+      "eval_samples_per_second": 43.257,
+      "eval_steps_per_second": 0.229,
+      "step": 432000
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 4.303589632675561e-05,
+      "loss": 0.6378,
+      "step": 432200
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 4.300952039321242e-05,
+      "loss": 0.6371,
+      "step": 432400
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 4.298314445966922e-05,
+      "loss": 0.6351,
+      "step": 432600
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 4.2956768526126025e-05,
+      "loss": 0.6318,
+      "step": 432800
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 4.293039259258282e-05,
+      "loss": 0.6331,
+      "step": 433000
+    },
+    {
+      "epoch": 0.57,
+      "eval_em_first_line_ratio": 0.5186,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4045,
+      "eval_loss": 0.6462664604187012,
+      "eval_runtime": 229.5367,
+      "eval_samples_per_second": 43.566,
+      "eval_steps_per_second": 0.231,
+      "step": 433000
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 4.290401665903963e-05,
+      "loss": 0.636,
+      "step": 433200
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 4.287764072549643e-05,
+      "loss": 0.6363,
+      "step": 433400
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 4.285126479195323e-05,
+      "loss": 0.6342,
+      "step": 433600
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 4.2824888858410034e-05,
+      "loss": 0.635,
+      "step": 433800
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 4.279851292486683e-05,
+      "loss": 0.6321,
+      "step": 434000
+    },
+    {
+      "epoch": 0.57,
+      "eval_em_first_line_ratio": 0.5203,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4066,
+      "eval_loss": 0.6460878252983093,
+      "eval_runtime": 226.6975,
+      "eval_samples_per_second": 44.112,
+      "eval_steps_per_second": 0.234,
+      "step": 434000
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 4.277213699132364e-05,
+      "loss": 0.632,
+      "step": 434200
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 4.274576105778044e-05,
+      "loss": 0.6329,
+      "step": 434400
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 4.2719385124237245e-05,
+      "loss": 0.636,
+      "step": 434600
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 4.269300919069404e-05,
+      "loss": 0.6336,
+      "step": 434800
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 4.266663325715085e-05,
+      "loss": 0.6334,
+      "step": 435000
+    },
+    {
+      "epoch": 0.57,
+      "eval_em_first_line_ratio": 0.5207,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4064,
+      "eval_loss": 0.6465334892272949,
+      "eval_runtime": 223.5593,
+      "eval_samples_per_second": 44.731,
+      "eval_steps_per_second": 0.237,
+      "step": 435000
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 4.264025732360765e-05,
+      "loss": 0.6357,
+      "step": 435200
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 4.2613881390064456e-05,
+      "loss": 0.6328,
+      "step": 435400
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 4.2587505456521253e-05,
+      "loss": 0.6344,
+      "step": 435600
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 4.256112952297806e-05,
+      "loss": 0.6348,
+      "step": 435800
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 4.2534753589434856e-05,
+      "loss": 0.6351,
+      "step": 436000
+    },
+    {
+      "epoch": 0.57,
+      "eval_em_first_line_ratio": 0.5179,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4042,
+      "eval_loss": 0.6462960243225098,
+      "eval_runtime": 230.0809,
+      "eval_samples_per_second": 43.463,
+      "eval_steps_per_second": 0.23,
+      "step": 436000
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 4.250837765589166e-05,
+      "loss": 0.6318,
+      "step": 436200
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 4.2482001722348464e-05,
+      "loss": 0.6338,
+      "step": 436400
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 4.245562578880526e-05,
+      "loss": 0.6337,
+      "step": 436600
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 4.2429249855262066e-05,
+      "loss": 0.6354,
+      "step": 436800
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 4.2402873921718864e-05,
+      "loss": 0.6301,
+      "step": 437000
+    },
+    {
+      "epoch": 0.58,
+      "eval_em_first_line_ratio": 0.5203,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4065,
+      "eval_loss": 0.645214855670929,
+      "eval_runtime": 228.9545,
+      "eval_samples_per_second": 43.677,
+      "eval_steps_per_second": 0.231,
+      "step": 437000
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 4.2376629867843384e-05,
+      "loss": 0.6315,
+      "step": 437200
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 4.235025393430019e-05,
+      "loss": 0.6368,
+      "step": 437400
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 4.2323878000756986e-05,
+      "loss": 0.632,
+      "step": 437600
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 4.22975020672138e-05,
+      "loss": 0.6303,
+      "step": 437800
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 4.2271126133670595e-05,
+      "loss": 0.6318,
+      "step": 438000
+    },
+    {
+      "epoch": 0.58,
+      "eval_em_first_line_ratio": 0.5204,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4064,
+      "eval_loss": 0.6453118920326233,
+      "eval_runtime": 223.7903,
+      "eval_samples_per_second": 44.685,
+      "eval_steps_per_second": 0.237,
+      "step": 438000
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 4.22447502001274e-05,
+      "loss": 0.633,
+      "step": 438200
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 4.22183742665842e-05,
+      "loss": 0.6355,
+      "step": 438400
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 4.2191998333041e-05,
+      "loss": 0.6317,
+      "step": 438600
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 4.216575427916552e-05,
+      "loss": 0.6282,
+      "step": 438800
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 4.213937834562232e-05,
+      "loss": 0.6385,
+      "step": 439000
+    },
+    {
+      "epoch": 0.58,
+      "eval_em_first_line_ratio": 0.5204,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4061,
+      "eval_loss": 0.6456355452537537,
+      "eval_runtime": 221.7257,
+      "eval_samples_per_second": 45.101,
+      "eval_steps_per_second": 0.239,
+      "step": 439000
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 4.2113002412079124e-05,
+      "loss": 0.6352,
+      "step": 439200
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 4.208662647853593e-05,
+      "loss": 0.6314,
+      "step": 439400
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 4.206025054499273e-05,
+      "loss": 0.6327,
+      "step": 439600
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 4.203387461144953e-05,
+      "loss": 0.6334,
+      "step": 439800
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 4.2007498677906334e-05,
+      "loss": 0.6331,
+      "step": 440000
+    },
+    {
+      "epoch": 0.58,
+      "eval_em_first_line_ratio": 0.5192,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4057,
+      "eval_loss": 0.6453011631965637,
+      "eval_runtime": 229.2927,
+      "eval_samples_per_second": 43.612,
+      "eval_steps_per_second": 0.231,
+      "step": 440000
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 4.198112274436313e-05,
+      "loss": 0.6309,
+      "step": 440200
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 4.1954746810819937e-05,
+      "loss": 0.6308,
+      "step": 440400
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 4.192837087727674e-05,
+      "loss": 0.6294,
+      "step": 440600
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 4.190199494373354e-05,
+      "loss": 0.6353,
+      "step": 440800
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 4.187561901019034e-05,
+      "loss": 0.6319,
+      "step": 441000
+    },
+    {
+      "epoch": 0.58,
+      "eval_em_first_line_ratio": 0.5175,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4035,
+      "eval_loss": 0.6448260545730591,
+      "eval_runtime": 229.6552,
+      "eval_samples_per_second": 43.544,
+      "eval_steps_per_second": 0.231,
+      "step": 441000
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 4.184924307664715e-05,
+      "loss": 0.6315,
+      "step": 441200
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 4.182286714310395e-05,
+      "loss": 0.6327,
+      "step": 441400
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 4.179649120956075e-05,
+      "loss": 0.6307,
+      "step": 441600
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 4.1770115276017554e-05,
+      "loss": 0.6336,
+      "step": 441800
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 4.174373934247435e-05,
+      "loss": 0.6364,
+      "step": 442000
+    },
+    {
+      "epoch": 0.58,
+      "eval_em_first_line_ratio": 0.5171,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.403,
+      "eval_loss": 0.6453405022621155,
+      "eval_runtime": 231.1328,
+      "eval_samples_per_second": 43.265,
+      "eval_steps_per_second": 0.229,
+      "step": 442000
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 4.171749528859887e-05,
+      "loss": 0.6361,
+      "step": 442200
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 4.1691119355055676e-05,
+      "loss": 0.6323,
+      "step": 442400
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 4.1664743421512474e-05,
+      "loss": 0.6348,
+      "step": 442600
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 4.1638499367636994e-05,
+      "loss": 0.6318,
+      "step": 442800
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 4.16121234340938e-05,
+      "loss": 0.6322,
+      "step": 443000
+    },
+    {
+      "epoch": 0.58,
+      "eval_em_first_line_ratio": 0.5166,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.403,
+      "eval_loss": 0.6444166302680969,
+      "eval_runtime": 222.3229,
+      "eval_samples_per_second": 44.98,
+      "eval_steps_per_second": 0.238,
+      "step": 443000
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 4.1585747500550596e-05,
+      "loss": 0.6363,
+      "step": 443200
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 4.15593715670074e-05,
+      "loss": 0.6326,
+      "step": 443400
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 4.1532995633464205e-05,
+      "loss": 0.6325,
+      "step": 443600
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 4.150661969992101e-05,
+      "loss": 0.6323,
+      "step": 443800
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 4.148024376637781e-05,
+      "loss": 0.6293,
+      "step": 444000
+    },
+    {
+      "epoch": 0.59,
+      "eval_em_first_line_ratio": 0.5161,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4021,
+      "eval_loss": 0.6449143290519714,
+      "eval_runtime": 226.0603,
+      "eval_samples_per_second": 44.236,
+      "eval_steps_per_second": 0.234,
+      "step": 444000
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 4.145386783283461e-05,
+      "loss": 0.6364,
+      "step": 444200
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 4.1427491899291415e-05,
+      "loss": 0.6326,
+      "step": 444400
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 4.140111596574821e-05,
+      "loss": 0.6349,
+      "step": 444600
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 4.137474003220502e-05,
+      "loss": 0.634,
+      "step": 444800
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 4.1348364098661815e-05,
+      "loss": 0.6306,
+      "step": 445000
+    },
+    {
+      "epoch": 0.59,
+      "eval_em_first_line_ratio": 0.516,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4017,
+      "eval_loss": 0.6445021629333496,
+      "eval_runtime": 223.9119,
+      "eval_samples_per_second": 44.66,
+      "eval_steps_per_second": 0.237,
+      "step": 445000
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 4.132198816511862e-05,
+      "loss": 0.6303,
+      "step": 445200
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 4.1295612231575424e-05,
+      "loss": 0.6309,
+      "step": 445400
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 4.126923629803223e-05,
+      "loss": 0.633,
+      "step": 445600
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 4.1242860364489026e-05,
+      "loss": 0.6296,
+      "step": 445800
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 4.121648443094583e-05,
+      "loss": 0.6313,
+      "step": 446000
+    },
+    {
+      "epoch": 0.59,
+      "eval_em_first_line_ratio": 0.5159,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4018,
+      "eval_loss": 0.6441246867179871,
+      "eval_runtime": 228.6261,
+      "eval_samples_per_second": 43.74,
+      "eval_steps_per_second": 0.232,
+      "step": 446000
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 4.119010849740263e-05,
+      "loss": 0.6368,
+      "step": 446200
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 4.116373256385944e-05,
+      "loss": 0.6308,
+      "step": 446400
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 4.113735663031624e-05,
+      "loss": 0.6312,
+      "step": 446600
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 4.111098069677304e-05,
+      "loss": 0.6308,
+      "step": 446800
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 4.108460476322984e-05,
+      "loss": 0.6346,
+      "step": 447000
+    },
+    {
+      "epoch": 0.59,
+      "eval_em_first_line_ratio": 0.5177,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4036,
+      "eval_loss": 0.6440823078155518,
+      "eval_runtime": 231.0059,
+      "eval_samples_per_second": 43.289,
+      "eval_steps_per_second": 0.229,
+      "step": 447000
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 4.1058228829686643e-05,
+      "loss": 0.6304,
+      "step": 447200
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 4.103185289614345e-05,
+      "loss": 0.6314,
+      "step": 447400
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 4.1005476962600246e-05,
+      "loss": 0.6352,
+      "step": 447600
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 4.097910102905705e-05,
+      "loss": 0.6373,
+      "step": 447800
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 4.095272509551385e-05,
+      "loss": 0.6321,
+      "step": 448000
+    },
+    {
+      "epoch": 0.59,
+      "eval_em_first_line_ratio": 0.518,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4034,
+      "eval_loss": 0.6444352865219116,
+      "eval_runtime": 223.9529,
+      "eval_samples_per_second": 44.652,
+      "eval_steps_per_second": 0.237,
+      "step": 448000
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 4.092634916197065e-05,
+      "loss": 0.6345,
+      "step": 448200
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 4.0899973228427456e-05,
+      "loss": 0.6362,
+      "step": 448400
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 4.087359729488426e-05,
+      "loss": 0.6262,
+      "step": 448600
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 4.084722136134106e-05,
+      "loss": 0.6296,
+      "step": 448800
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 4.082084542779786e-05,
+      "loss": 0.632,
+      "step": 449000
+    },
+    {
+      "epoch": 0.59,
+      "eval_em_first_line_ratio": 0.5173,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4033,
+      "eval_loss": 0.6443926095962524,
+      "eval_runtime": 230.3003,
+      "eval_samples_per_second": 43.422,
+      "eval_steps_per_second": 0.23,
+      "step": 449000
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 4.079446949425467e-05,
+      "loss": 0.6313,
+      "step": 449200
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 4.0768093560711465e-05,
+      "loss": 0.6331,
+      "step": 449400
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 4.074171762716827e-05,
+      "loss": 0.6332,
+      "step": 449600
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 4.071534169362507e-05,
+      "loss": 0.6332,
+      "step": 449800
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 4.068896576008187e-05,
+      "loss": 0.6298,
+      "step": 450000
+    },
+    {
+      "epoch": 0.59,
+      "eval_em_first_line_ratio": 0.5188,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4051,
+      "eval_loss": 0.6434319615364075,
+      "eval_runtime": 228.6438,
+      "eval_samples_per_second": 43.736,
+      "eval_steps_per_second": 0.232,
+      "step": 450000
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 4.0662589826538676e-05,
+      "loss": 0.6323,
+      "step": 450200
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 4.063621389299548e-05,
+      "loss": 0.6338,
+      "step": 450400
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 4.060983795945228e-05,
+      "loss": 0.6309,
+      "step": 450600
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 4.058346202590908e-05,
+      "loss": 0.6269,
+      "step": 450800
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 4.055708609236588e-05,
+      "loss": 0.631,
+      "step": 451000
+    },
+    {
+      "epoch": 0.59,
+      "eval_em_first_line_ratio": 0.5192,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4053,
+      "eval_loss": 0.6435883045196533,
+      "eval_runtime": 228.6626,
+      "eval_samples_per_second": 43.733,
+      "eval_steps_per_second": 0.232,
+      "step": 451000
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 4.0530710158822685e-05,
+      "loss": 0.6333,
+      "step": 451200
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 4.050433422527949e-05,
+      "loss": 0.6308,
+      "step": 451400
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 4.0478090171404e-05,
+      "loss": 0.6316,
+      "step": 451600
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 4.045171423786081e-05,
+      "loss": 0.6318,
+      "step": 451800
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 4.042533830431761e-05,
+      "loss": 0.63,
+      "step": 452000
+    },
+    {
+      "epoch": 0.6,
+      "eval_em_first_line_ratio": 0.5189,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4046,
+      "eval_loss": 0.6437413692474365,
+      "eval_runtime": 228.833,
+      "eval_samples_per_second": 43.7,
+      "eval_steps_per_second": 0.232,
+      "step": 452000
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 4.0399094250442124e-05,
+      "loss": 0.6274,
+      "step": 452200
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 4.0372718316898935e-05,
+      "loss": 0.6355,
+      "step": 452400
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 4.034634238335573e-05,
+      "loss": 0.6347,
+      "step": 452600
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 4.031996644981254e-05,
+      "loss": 0.6308,
+      "step": 452800
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 4.0293590516269335e-05,
+      "loss": 0.6311,
+      "step": 453000
+    },
+    {
+      "epoch": 0.6,
+      "eval_em_first_line_ratio": 0.5193,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4053,
+      "eval_loss": 0.643491268157959,
+      "eval_runtime": 219.9514,
+      "eval_samples_per_second": 45.465,
+      "eval_steps_per_second": 0.241,
+      "step": 453000
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 4.026721458272614e-05,
+      "loss": 0.6296,
+      "step": 453200
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 4.0240838649182944e-05,
+      "loss": 0.6326,
+      "step": 453400
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 4.021446271563974e-05,
+      "loss": 0.6341,
+      "step": 453600
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 4.0188086782096546e-05,
+      "loss": 0.6295,
+      "step": 453800
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 4.0161710848553344e-05,
+      "loss": 0.6311,
+      "step": 454000
+    },
+    {
+      "epoch": 0.6,
+      "eval_em_first_line_ratio": 0.517,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.403,
+      "eval_loss": 0.6436184644699097,
+      "eval_runtime": 224.7922,
+      "eval_samples_per_second": 44.486,
+      "eval_steps_per_second": 0.236,
+      "step": 454000
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 4.013533491501015e-05,
+      "loss": 0.6318,
+      "step": 454200
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 4.010895898146695e-05,
+      "loss": 0.6314,
+      "step": 454400
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 4.008258304792376e-05,
+      "loss": 0.6301,
+      "step": 454600
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 4.0056207114380555e-05,
+      "loss": 0.6324,
+      "step": 454800
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 4.002983118083736e-05,
+      "loss": 0.6272,
+      "step": 455000
+    },
+    {
+      "epoch": 0.6,
+      "eval_em_first_line_ratio": 0.5186,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4044,
+      "eval_loss": 0.6431202292442322,
+      "eval_runtime": 221.7531,
+      "eval_samples_per_second": 45.095,
+      "eval_steps_per_second": 0.239,
+      "step": 455000
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 4.000345524729416e-05,
+      "loss": 0.6308,
+      "step": 455200
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 3.997721119341868e-05,
+      "loss": 0.6313,
+      "step": 455400
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 3.995083525987548e-05,
+      "loss": 0.6264,
+      "step": 455600
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 3.992445932633228e-05,
+      "loss": 0.6303,
+      "step": 455800
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 3.989808339278909e-05,
+      "loss": 0.6314,
+      "step": 456000
+    },
+    {
+      "epoch": 0.6,
+      "eval_em_first_line_ratio": 0.5208,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4068,
+      "eval_loss": 0.6435231566429138,
+      "eval_runtime": 228.4957,
+      "eval_samples_per_second": 43.765,
+      "eval_steps_per_second": 0.232,
+      "step": 456000
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 3.987170745924589e-05,
+      "loss": 0.6299,
+      "step": 456200
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 3.98454634053704e-05,
+      "loss": 0.6293,
+      "step": 456400
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 3.981908747182721e-05,
+      "loss": 0.6318,
+      "step": 456600
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 3.979271153828401e-05,
+      "loss": 0.6289,
+      "step": 456800
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 3.9766335604740814e-05,
+      "loss": 0.634,
+      "step": 457000
+    },
+    {
+      "epoch": 0.6,
+      "eval_em_first_line_ratio": 0.5181,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4045,
+      "eval_loss": 0.643288791179657,
+      "eval_runtime": 221.9322,
+      "eval_samples_per_second": 45.059,
+      "eval_steps_per_second": 0.239,
+      "step": 457000
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 3.973995967119761e-05,
+      "loss": 0.6255,
+      "step": 457200
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 3.9713583737654416e-05,
+      "loss": 0.6254,
+      "step": 457400
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 3.968720780411122e-05,
+      "loss": 0.6292,
+      "step": 457600
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 3.966083187056802e-05,
+      "loss": 0.6321,
+      "step": 457800
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 3.963445593702482e-05,
+      "loss": 0.6337,
+      "step": 458000
+    },
+    {
+      "epoch": 0.6,
+      "eval_em_first_line_ratio": 0.5197,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4061,
+      "eval_loss": 0.6430273652076721,
+      "eval_runtime": 229.8251,
+      "eval_samples_per_second": 43.511,
+      "eval_steps_per_second": 0.231,
+      "step": 458000
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 3.960808000348162e-05,
+      "loss": 0.6326,
+      "step": 458200
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 3.9581704069938425e-05,
+      "loss": 0.6313,
+      "step": 458400
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 3.9555460016062945e-05,
+      "loss": 0.6302,
+      "step": 458600
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 3.952908408251975e-05,
+      "loss": 0.6281,
+      "step": 458800
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 3.950270814897655e-05,
+      "loss": 0.6293,
+      "step": 459000
+    },
+    {
+      "epoch": 0.61,
+      "eval_em_first_line_ratio": 0.5187,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4049,
+      "eval_loss": 0.6430788636207581,
+      "eval_runtime": 225.7816,
+      "eval_samples_per_second": 44.291,
+      "eval_steps_per_second": 0.235,
+      "step": 459000
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 3.947633221543335e-05,
+      "loss": 0.6278,
+      "step": 459200
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 3.9449956281890156e-05,
+      "loss": 0.63,
+      "step": 459400
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 3.942358034834695e-05,
+      "loss": 0.6282,
+      "step": 459600
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 3.939720441480376e-05,
+      "loss": 0.6292,
+      "step": 459800
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 3.937082848126056e-05,
+      "loss": 0.627,
+      "step": 460000
+    },
+    {
+      "epoch": 0.61,
+      "eval_em_first_line_ratio": 0.5184,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4048,
+      "eval_loss": 0.6426063179969788,
+      "eval_runtime": 224.4027,
+      "eval_samples_per_second": 44.563,
+      "eval_steps_per_second": 0.236,
+      "step": 460000
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 3.9344452547717366e-05,
+      "loss": 0.6321,
+      "step": 460200
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 3.9318076614174164e-05,
+      "loss": 0.6331,
+      "step": 460400
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 3.929170068063097e-05,
+      "loss": 0.632,
+      "step": 460600
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 3.9265324747087766e-05,
+      "loss": 0.6311,
+      "step": 460800
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 3.923894881354457e-05,
+      "loss": 0.6253,
+      "step": 461000
+    },
+    {
+      "epoch": 0.61,
+      "eval_em_first_line_ratio": 0.5203,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4065,
+      "eval_loss": 0.6422901153564453,
+      "eval_runtime": 223.1656,
+      "eval_samples_per_second": 44.81,
+      "eval_steps_per_second": 0.237,
+      "step": 461000
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 3.9212572880001375e-05,
+      "loss": 0.6281,
+      "step": 461200
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 3.918619694645817e-05,
+      "loss": 0.6284,
+      "step": 461400
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 3.915982101291498e-05,
+      "loss": 0.6299,
+      "step": 461600
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 3.9133445079371775e-05,
+      "loss": 0.6327,
+      "step": 461800
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 3.9107069145828586e-05,
+      "loss": 0.6275,
+      "step": 462000
+    },
+    {
+      "epoch": 0.61,
+      "eval_em_first_line_ratio": 0.5197,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4056,
+      "eval_loss": 0.6427074074745178,
+      "eval_runtime": 219.3938,
+      "eval_samples_per_second": 45.58,
+      "eval_steps_per_second": 0.242,
+      "step": 462000
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 3.9080693212285384e-05,
+      "loss": 0.6292,
+      "step": 462200
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 3.905431727874219e-05,
+      "loss": 0.6303,
+      "step": 462400
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 3.9027941345198986e-05,
+      "loss": 0.6303,
+      "step": 462600
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 3.9001697291323506e-05,
+      "loss": 0.6287,
+      "step": 462800
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 3.897532135778031e-05,
+      "loss": 0.6314,
+      "step": 463000
+    },
+    {
+      "epoch": 0.61,
+      "eval_em_first_line_ratio": 0.517,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4033,
+      "eval_loss": 0.641755223274231,
+      "eval_runtime": 229.4692,
+      "eval_samples_per_second": 43.579,
+      "eval_steps_per_second": 0.231,
+      "step": 463000
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 3.894894542423711e-05,
+      "loss": 0.6287,
+      "step": 463200
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 3.892270137036163e-05,
+      "loss": 0.6288,
+      "step": 463400
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 3.889632543681843e-05,
+      "loss": 0.6293,
+      "step": 463600
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 3.886994950327523e-05,
+      "loss": 0.6299,
+      "step": 463800
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 3.8843573569732034e-05,
+      "loss": 0.6277,
+      "step": 464000
+    },
+    {
+      "epoch": 0.61,
+      "eval_em_first_line_ratio": 0.5168,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.403,
+      "eval_loss": 0.6416076421737671,
+      "eval_runtime": 224.8778,
+      "eval_samples_per_second": 44.469,
+      "eval_steps_per_second": 0.236,
+      "step": 464000
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 3.881719763618884e-05,
+      "loss": 0.6258,
+      "step": 464200
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 3.879082170264564e-05,
+      "loss": 0.6293,
+      "step": 464400
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 3.8764577648770156e-05,
+      "loss": 0.6319,
+      "step": 464600
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 3.873820171522696e-05,
+      "loss": 0.6272,
+      "step": 464800
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 3.8711825781683765e-05,
+      "loss": 0.6314,
+      "step": 465000
+    },
+    {
+      "epoch": 0.61,
+      "eval_em_first_line_ratio": 0.516,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4023,
+      "eval_loss": 0.6419846415519714,
+      "eval_runtime": 222.8603,
+      "eval_samples_per_second": 44.871,
+      "eval_steps_per_second": 0.238,
+      "step": 465000
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 3.868544984814056e-05,
+      "loss": 0.6267,
+      "step": 465200
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 3.865907391459737e-05,
+      "loss": 0.6329,
+      "step": 465400
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 3.8632697981054165e-05,
+      "loss": 0.6278,
+      "step": 465600
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 3.8606322047510976e-05,
+      "loss": 0.6295,
+      "step": 465800
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 3.8579946113967774e-05,
+      "loss": 0.6289,
+      "step": 466000
+    },
+    {
+      "epoch": 0.61,
+      "eval_em_first_line_ratio": 0.5183,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4042,
+      "eval_loss": 0.6415135264396667,
+      "eval_runtime": 224.8238,
+      "eval_samples_per_second": 44.479,
+      "eval_steps_per_second": 0.236,
+      "step": 466000
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 3.855357018042458e-05,
+      "loss": 0.6247,
+      "step": 466200
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 3.8527194246881376e-05,
+      "loss": 0.6285,
+      "step": 466400
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 3.850081831333818e-05,
+      "loss": 0.6295,
+      "step": 466600
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 3.8474442379794985e-05,
+      "loss": 0.6278,
+      "step": 466800
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 3.844806644625178e-05,
+      "loss": 0.628,
+      "step": 467000
+    },
+    {
+      "epoch": 0.62,
+      "eval_em_first_line_ratio": 0.5177,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4035,
+      "eval_loss": 0.6413995027542114,
+      "eval_runtime": 230.3309,
+      "eval_samples_per_second": 43.416,
+      "eval_steps_per_second": 0.23,
+      "step": 467000
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 3.842169051270859e-05,
+      "loss": 0.6261,
+      "step": 467200
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 3.8395314579165384e-05,
+      "loss": 0.6272,
+      "step": 467400
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 3.836893864562219e-05,
+      "loss": 0.6277,
+      "step": 467600
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 3.834256271207899e-05,
+      "loss": 0.6328,
+      "step": 467800
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 3.83161867785358e-05,
+      "loss": 0.6277,
+      "step": 468000
+    },
+    {
+      "epoch": 0.62,
+      "eval_em_first_line_ratio": 0.516,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4021,
+      "eval_loss": 0.6412070393562317,
+      "eval_runtime": 224.1527,
+      "eval_samples_per_second": 44.612,
+      "eval_steps_per_second": 0.236,
+      "step": 468000
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 3.8289810844992595e-05,
+      "loss": 0.6294,
+      "step": 468200
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 3.82634349114494e-05,
+      "loss": 0.6307,
+      "step": 468400
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 3.8237058977906204e-05,
+      "loss": 0.6301,
+      "step": 468600
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 3.8210683044363e-05,
+      "loss": 0.6269,
+      "step": 468800
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 3.8184307110819806e-05,
+      "loss": 0.6314,
+      "step": 469000
+    },
+    {
+      "epoch": 0.62,
+      "eval_em_first_line_ratio": 0.5189,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4046,
+      "eval_loss": 0.6411029696464539,
+      "eval_runtime": 227.5404,
+      "eval_samples_per_second": 43.948,
+      "eval_steps_per_second": 0.233,
+      "step": 469000
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 3.8157931177276604e-05,
+      "loss": 0.6225,
+      "step": 469200
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 3.813155524373341e-05,
+      "loss": 0.6262,
+      "step": 469400
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 3.810517931019021e-05,
+      "loss": 0.6277,
+      "step": 469600
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 3.807880337664702e-05,
+      "loss": 0.6308,
+      "step": 469800
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 3.8052427443103815e-05,
+      "loss": 0.6293,
+      "step": 470000
+    },
+    {
+      "epoch": 0.62,
+      "eval_em_first_line_ratio": 0.5182,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4039,
+      "eval_loss": 0.6413387656211853,
+      "eval_runtime": 222.6191,
+      "eval_samples_per_second": 44.92,
+      "eval_steps_per_second": 0.238,
+      "step": 470000
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 3.802605150956062e-05,
+      "loss": 0.6268,
+      "step": 470200
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 3.799967557601742e-05,
+      "loss": 0.6349,
+      "step": 470400
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 3.797329964247423e-05,
+      "loss": 0.6261,
+      "step": 470600
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 3.7946923708931026e-05,
+      "loss": 0.6295,
+      "step": 470800
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 3.792054777538783e-05,
+      "loss": 0.6284,
+      "step": 471000
+    },
+    {
+      "epoch": 0.62,
+      "eval_em_first_line_ratio": 0.5165,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4024,
+      "eval_loss": 0.641747772693634,
+      "eval_runtime": 225.5474,
+      "eval_samples_per_second": 44.337,
+      "eval_steps_per_second": 0.235,
+      "step": 471000
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 3.789417184184463e-05,
+      "loss": 0.6277,
+      "step": 471200
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 3.786792778796915e-05,
+      "loss": 0.6248,
+      "step": 471400
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 3.784155185442595e-05,
+      "loss": 0.6246,
+      "step": 471600
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 3.781517592088275e-05,
+      "loss": 0.6277,
+      "step": 471800
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 3.7788799987339554e-05,
+      "loss": 0.6221,
+      "step": 472000
+    },
+    {
+      "epoch": 0.62,
+      "eval_em_first_line_ratio": 0.5168,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4026,
+      "eval_loss": 0.6408032178878784,
+      "eval_runtime": 232.5349,
+      "eval_samples_per_second": 43.004,
+      "eval_steps_per_second": 0.228,
+      "step": 472000
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 3.776242405379636e-05,
+      "loss": 0.6297,
+      "step": 472200
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 3.7736048120253156e-05,
+      "loss": 0.6278,
+      "step": 472400
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 3.770967218670996e-05,
+      "loss": 0.6275,
+      "step": 472600
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 3.768329625316676e-05,
+      "loss": 0.6256,
+      "step": 472800
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 3.765692031962356e-05,
+      "loss": 0.6258,
+      "step": 473000
+    },
+    {
+      "epoch": 0.62,
+      "eval_em_first_line_ratio": 0.5179,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.404,
+      "eval_loss": 0.6408575177192688,
+      "eval_runtime": 229.7346,
+      "eval_samples_per_second": 43.528,
+      "eval_steps_per_second": 0.231,
+      "step": 473000
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 3.763054438608037e-05,
+      "loss": 0.6273,
+      "step": 473200
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 3.760416845253717e-05,
+      "loss": 0.6286,
+      "step": 473400
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 3.7577924398661685e-05,
+      "loss": 0.6308,
+      "step": 473600
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 3.755154846511849e-05,
+      "loss": 0.6227,
+      "step": 473800
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 3.7525172531575294e-05,
+      "loss": 0.6266,
+      "step": 474000
+    },
+    {
+      "epoch": 0.63,
+      "eval_em_first_line_ratio": 0.5178,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4038,
+      "eval_loss": 0.6409462690353394,
+      "eval_runtime": 227.7855,
+      "eval_samples_per_second": 43.901,
+      "eval_steps_per_second": 0.233,
+      "step": 474000
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 3.749879659803209e-05,
+      "loss": 0.625,
+      "step": 474200
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 3.7472420664488896e-05,
+      "loss": 0.6251,
+      "step": 474400
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 3.744604473094569e-05,
+      "loss": 0.6284,
+      "step": 474600
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 3.7419668797402505e-05,
+      "loss": 0.6266,
+      "step": 474800
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 3.73932928638593e-05,
+      "loss": 0.6263,
+      "step": 475000
+    },
+    {
+      "epoch": 0.63,
+      "eval_em_first_line_ratio": 0.5168,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4025,
+      "eval_loss": 0.6405984163284302,
+      "eval_runtime": 231.4534,
+      "eval_samples_per_second": 43.205,
+      "eval_steps_per_second": 0.229,
+      "step": 475000
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 3.736691693031611e-05,
+      "loss": 0.6282,
+      "step": 475200
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 3.7340540996772904e-05,
+      "loss": 0.6252,
+      "step": 475400
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 3.731416506322971e-05,
+      "loss": 0.6269,
+      "step": 475600
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 3.728778912968651e-05,
+      "loss": 0.627,
+      "step": 475800
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 3.726141319614331e-05,
+      "loss": 0.6252,
+      "step": 476000
+    },
+    {
+      "epoch": 0.63,
+      "eval_em_first_line_ratio": 0.5188,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4044,
+      "eval_loss": 0.6404222249984741,
+      "eval_runtime": 225.4939,
+      "eval_samples_per_second": 44.347,
+      "eval_steps_per_second": 0.235,
+      "step": 476000
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 3.7235037262600115e-05,
+      "loss": 0.6296,
+      "step": 476200
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 3.720866132905691e-05,
+      "loss": 0.6286,
+      "step": 476400
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 3.7182285395513724e-05,
+      "loss": 0.6279,
+      "step": 476600
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 3.715590946197052e-05,
+      "loss": 0.6273,
+      "step": 476800
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 3.7129533528427326e-05,
+      "loss": 0.6247,
+      "step": 477000
+    },
+    {
+      "epoch": 0.63,
+      "eval_em_first_line_ratio": 0.5206,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4059,
+      "eval_loss": 0.6395808458328247,
+      "eval_runtime": 226.4852,
+      "eval_samples_per_second": 44.153,
+      "eval_steps_per_second": 0.234,
+      "step": 477000
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 3.7103157594884124e-05,
+      "loss": 0.6231,
+      "step": 477200
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 3.707678166134093e-05,
+      "loss": 0.6242,
+      "step": 477400
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 3.705053760746545e-05,
+      "loss": 0.6266,
+      "step": 477600
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 3.7024161673922246e-05,
+      "loss": 0.6244,
+      "step": 477800
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 3.699778574037905e-05,
+      "loss": 0.6234,
+      "step": 478000
+    },
+    {
+      "epoch": 0.63,
+      "eval_em_first_line_ratio": 0.5198,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4058,
+      "eval_loss": 0.6395296454429626,
+      "eval_runtime": 221.4662,
+      "eval_samples_per_second": 45.154,
+      "eval_steps_per_second": 0.239,
+      "step": 478000
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 3.697154168650357e-05,
+      "loss": 0.6236,
+      "step": 478200
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 3.694516575296037e-05,
+      "loss": 0.6227,
+      "step": 478400
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 3.691878981941717e-05,
+      "loss": 0.6278,
+      "step": 478600
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 3.689241388587398e-05,
+      "loss": 0.6228,
+      "step": 478800
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 3.686603795233078e-05,
+      "loss": 0.6279,
+      "step": 479000
+    },
+    {
+      "epoch": 0.63,
+      "eval_em_first_line_ratio": 0.5164,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4028,
+      "eval_loss": 0.6396872401237488,
+      "eval_runtime": 230.1097,
+      "eval_samples_per_second": 43.458,
+      "eval_steps_per_second": 0.23,
+      "step": 479000
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 3.683966201878758e-05,
+      "loss": 0.6271,
+      "step": 479200
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 3.681328608524438e-05,
+      "loss": 0.6262,
+      "step": 479400
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 3.678691015170118e-05,
+      "loss": 0.6253,
+      "step": 479600
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 3.6760534218157985e-05,
+      "loss": 0.625,
+      "step": 479800
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 3.673415828461479e-05,
+      "loss": 0.6247,
+      "step": 480000
+    },
+    {
+      "epoch": 0.63,
+      "eval_em_first_line_ratio": 0.5167,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4028,
+      "eval_loss": 0.6393082141876221,
+      "eval_runtime": 223.4869,
+      "eval_samples_per_second": 44.745,
+      "eval_steps_per_second": 0.237,
+      "step": 480000
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 3.670778235107159e-05,
+      "loss": 0.624,
+      "step": 480200
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 3.668140641752839e-05,
+      "loss": 0.6244,
+      "step": 480400
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 3.665503048398519e-05,
+      "loss": 0.6217,
+      "step": 480600
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 3.6628654550442e-05,
+      "loss": 0.6233,
+      "step": 480800
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 3.66022786168988e-05,
+      "loss": 0.6273,
+      "step": 481000
+    },
+    {
+      "epoch": 0.63,
+      "eval_em_first_line_ratio": 0.516,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4022,
+      "eval_loss": 0.6393333077430725,
+      "eval_runtime": 232.0202,
+      "eval_samples_per_second": 43.1,
+      "eval_steps_per_second": 0.228,
+      "step": 481000
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 3.65759026833556e-05,
+      "loss": 0.623,
+      "step": 481200
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 3.65495267498124e-05,
+      "loss": 0.6322,
+      "step": 481400
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 3.6523150816269205e-05,
+      "loss": 0.6245,
+      "step": 481600
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 3.649677488272601e-05,
+      "loss": 0.6278,
+      "step": 481800
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 3.6470398949182814e-05,
+      "loss": 0.6307,
+      "step": 482000
+    },
+    {
+      "epoch": 0.64,
+      "eval_em_first_line_ratio": 0.5174,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4035,
+      "eval_loss": 0.6385302543640137,
+      "eval_runtime": 228.0274,
+      "eval_samples_per_second": 43.854,
+      "eval_steps_per_second": 0.232,
+      "step": 482000
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 3.644402301563961e-05,
+      "loss": 0.6267,
+      "step": 482200
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 3.641764708209641e-05,
+      "loss": 0.6203,
+      "step": 482400
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 3.639127114855321e-05,
+      "loss": 0.6265,
+      "step": 482600
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 3.636489521501002e-05,
+      "loss": 0.6242,
+      "step": 482800
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 3.633851928146682e-05,
+      "loss": 0.623,
+      "step": 483000
+    },
+    {
+      "epoch": 0.64,
+      "eval_em_first_line_ratio": 0.5189,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4051,
+      "eval_loss": 0.6393818259239197,
+      "eval_runtime": 225.0986,
+      "eval_samples_per_second": 44.425,
+      "eval_steps_per_second": 0.235,
+      "step": 483000
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 3.631214334792362e-05,
+      "loss": 0.6278,
+      "step": 483200
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 3.6285767414380424e-05,
+      "loss": 0.6265,
+      "step": 483400
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 3.625939148083723e-05,
+      "loss": 0.6258,
+      "step": 483600
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 3.623301554729403e-05,
+      "loss": 0.627,
+      "step": 483800
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 3.6206771493418546e-05,
+      "loss": 0.6298,
+      "step": 484000
+    },
+    {
+      "epoch": 0.64,
+      "eval_em_first_line_ratio": 0.5175,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4041,
+      "eval_loss": 0.638744592666626,
+      "eval_runtime": 221.5068,
+      "eval_samples_per_second": 45.145,
+      "eval_steps_per_second": 0.239,
+      "step": 484000
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 3.6180395559875344e-05,
+      "loss": 0.6218,
+      "step": 484200
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 3.6154019626332155e-05,
+      "loss": 0.6265,
+      "step": 484400
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 3.612764369278895e-05,
+      "loss": 0.6312,
+      "step": 484600
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 3.610126775924576e-05,
+      "loss": 0.6254,
+      "step": 484800
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 3.607502370537028e-05,
+      "loss": 0.6224,
+      "step": 485000
+    },
+    {
+      "epoch": 0.64,
+      "eval_em_first_line_ratio": 0.5184,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4048,
+      "eval_loss": 0.6390528082847595,
+      "eval_runtime": 226.8783,
+      "eval_samples_per_second": 44.076,
+      "eval_steps_per_second": 0.234,
+      "step": 485000
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 3.6048647771827075e-05,
+      "loss": 0.6226,
+      "step": 485200
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 3.602227183828388e-05,
+      "loss": 0.6319,
+      "step": 485400
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 3.599589590474068e-05,
+      "loss": 0.6247,
+      "step": 485600
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 3.596951997119749e-05,
+      "loss": 0.6226,
+      "step": 485800
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 3.5943144037654286e-05,
+      "loss": 0.6276,
+      "step": 486000
+    },
+    {
+      "epoch": 0.64,
+      "eval_em_first_line_ratio": 0.5188,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4049,
+      "eval_loss": 0.6389985084533691,
+      "eval_runtime": 226.279,
+      "eval_samples_per_second": 44.193,
+      "eval_steps_per_second": 0.234,
+      "step": 486000
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 3.591676810411109e-05,
+      "loss": 0.6263,
+      "step": 486200
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 3.589039217056789e-05,
+      "loss": 0.6246,
+      "step": 486400
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 3.586401623702469e-05,
+      "loss": 0.6271,
+      "step": 486600
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 3.58376403034815e-05,
+      "loss": 0.6249,
+      "step": 486800
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 3.5811264369938294e-05,
+      "loss": 0.6258,
+      "step": 487000
+    },
+    {
+      "epoch": 0.64,
+      "eval_em_first_line_ratio": 0.5211,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4069,
+      "eval_loss": 0.6388115286827087,
+      "eval_runtime": 225.5277,
+      "eval_samples_per_second": 44.34,
+      "eval_steps_per_second": 0.235,
+      "step": 487000
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 3.5785020316062814e-05,
+      "loss": 0.63,
+      "step": 487200
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 3.575864438251962e-05,
+      "loss": 0.6198,
+      "step": 487400
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 3.5732268448976416e-05,
+      "loss": 0.6264,
+      "step": 487600
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 3.570589251543322e-05,
+      "loss": 0.6258,
+      "step": 487800
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 3.567951658189002e-05,
+      "loss": 0.6228,
+      "step": 488000
+    },
+    {
+      "epoch": 0.64,
+      "eval_em_first_line_ratio": 0.5189,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4048,
+      "eval_loss": 0.6392587423324585,
+      "eval_runtime": 224.1688,
+      "eval_samples_per_second": 44.609,
+      "eval_steps_per_second": 0.236,
+      "step": 488000
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 3.565314064834682e-05,
+      "loss": 0.6241,
+      "step": 488200
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 3.562676471480363e-05,
+      "loss": 0.6267,
+      "step": 488400
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 3.560038878126043e-05,
+      "loss": 0.626,
+      "step": 488600
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 3.557401284771723e-05,
+      "loss": 0.622,
+      "step": 488800
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 3.5547636914174034e-05,
+      "loss": 0.6237,
+      "step": 489000
+    },
+    {
+      "epoch": 0.64,
+      "eval_em_first_line_ratio": 0.5189,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4051,
+      "eval_loss": 0.6382855772972107,
+      "eval_runtime": 225.9761,
+      "eval_samples_per_second": 44.252,
+      "eval_steps_per_second": 0.235,
+      "step": 489000
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 3.552126098063083e-05,
+      "loss": 0.6218,
+      "step": 489200
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 3.549488504708764e-05,
+      "loss": 0.6225,
+      "step": 489400
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 3.5468640993212156e-05,
+      "loss": 0.6252,
+      "step": 489600
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 3.5442265059668953e-05,
+      "loss": 0.6266,
+      "step": 489800
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 3.5415889126125765e-05,
+      "loss": 0.6227,
+      "step": 490000
+    },
+    {
+      "epoch": 0.65,
+      "eval_em_first_line_ratio": 0.5195,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4062,
+      "eval_loss": 0.6387284994125366,
+      "eval_runtime": 220.731,
+      "eval_samples_per_second": 45.304,
+      "eval_steps_per_second": 0.24,
+      "step": 490000
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 3.538951319258256e-05,
+      "loss": 0.6232,
+      "step": 490200
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 3.536313725903937e-05,
+      "loss": 0.6241,
+      "step": 490400
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 3.5336761325496164e-05,
+      "loss": 0.6253,
+      "step": 490600
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 3.531038539195297e-05,
+      "loss": 0.6192,
+      "step": 490800
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 3.528400945840977e-05,
+      "loss": 0.6228,
+      "step": 491000
+    },
+    {
+      "epoch": 0.65,
+      "eval_em_first_line_ratio": 0.5185,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.405,
+      "eval_loss": 0.6381157636642456,
+      "eval_runtime": 227.7216,
+      "eval_samples_per_second": 43.913,
+      "eval_steps_per_second": 0.233,
+      "step": 491000
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 3.525763352486657e-05,
+      "loss": 0.6206,
+      "step": 491200
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 3.5231257591323375e-05,
+      "loss": 0.6237,
+      "step": 491400
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 3.520488165778017e-05,
+      "loss": 0.6194,
+      "step": 491600
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 3.517850572423698e-05,
+      "loss": 0.623,
+      "step": 491800
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 3.515212979069378e-05,
+      "loss": 0.6208,
+      "step": 492000
+    },
+    {
+      "epoch": 0.65,
+      "eval_em_first_line_ratio": 0.5198,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4053,
+      "eval_loss": 0.6379780173301697,
+      "eval_runtime": 218.2341,
+      "eval_samples_per_second": 45.822,
+      "eval_steps_per_second": 0.243,
+      "step": 492000
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 3.5125753857150586e-05,
+      "loss": 0.6304,
+      "step": 492200
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 3.50995098032751e-05,
+      "loss": 0.6266,
+      "step": 492400
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 3.5073133869731904e-05,
+      "loss": 0.6233,
+      "step": 492600
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 3.504675793618871e-05,
+      "loss": 0.6242,
+      "step": 492800
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 3.5020382002645506e-05,
+      "loss": 0.6239,
+      "step": 493000
+    },
+    {
+      "epoch": 0.65,
+      "eval_em_first_line_ratio": 0.5237,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4094,
+      "eval_loss": 0.6378296613693237,
+      "eval_runtime": 218.1042,
+      "eval_samples_per_second": 45.85,
+      "eval_steps_per_second": 0.243,
+      "step": 493000
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 3.499400606910231e-05,
+      "loss": 0.6223,
+      "step": 493200
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 3.496763013555911e-05,
+      "loss": 0.6235,
+      "step": 493400
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 3.494125420201592e-05,
+      "loss": 0.6233,
+      "step": 493600
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 3.491487826847272e-05,
+      "loss": 0.6193,
+      "step": 493800
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 3.488850233492952e-05,
+      "loss": 0.622,
+      "step": 494000
+    },
+    {
+      "epoch": 0.65,
+      "eval_em_first_line_ratio": 0.5206,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4062,
+      "eval_loss": 0.6377866864204407,
+      "eval_runtime": 217.9261,
+      "eval_samples_per_second": 45.887,
+      "eval_steps_per_second": 0.243,
+      "step": 494000
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 3.486212640138632e-05,
+      "loss": 0.6244,
+      "step": 494200
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 3.483575046784312e-05,
+      "loss": 0.6205,
+      "step": 494400
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 3.480937453429993e-05,
+      "loss": 0.6264,
+      "step": 494600
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 3.4782998600756725e-05,
+      "loss": 0.6268,
+      "step": 494800
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 3.475662266721353e-05,
+      "loss": 0.6222,
+      "step": 495000
+    },
+    {
+      "epoch": 0.65,
+      "eval_em_first_line_ratio": 0.5219,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4078,
+      "eval_loss": 0.6381061673164368,
+      "eval_runtime": 219.457,
+      "eval_samples_per_second": 45.567,
+      "eval_steps_per_second": 0.242,
+      "step": 495000
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 3.473024673367033e-05,
+      "loss": 0.6282,
+      "step": 495200
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 3.470387080012714e-05,
+      "loss": 0.624,
+      "step": 495400
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 3.467762674625165e-05,
+      "loss": 0.6312,
+      "step": 495600
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 3.465125081270845e-05,
+      "loss": 0.6259,
+      "step": 495800
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 3.462487487916526e-05,
+      "loss": 0.6221,
+      "step": 496000
+    },
+    {
+      "epoch": 0.65,
+      "eval_em_first_line_ratio": 0.5203,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4063,
+      "eval_loss": 0.6375981569290161,
+      "eval_runtime": 225.4006,
+      "eval_samples_per_second": 44.365,
+      "eval_steps_per_second": 0.235,
+      "step": 496000
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 3.459849894562206e-05,
+      "loss": 0.6222,
+      "step": 496200
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 3.457212301207886e-05,
+      "loss": 0.6264,
+      "step": 496400
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 3.454574707853566e-05,
+      "loss": 0.6208,
+      "step": 496600
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 3.4519371144992465e-05,
+      "loss": 0.6233,
+      "step": 496800
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 3.449299521144927e-05,
+      "loss": 0.6219,
+      "step": 497000
+    },
+    {
+      "epoch": 0.66,
+      "eval_em_first_line_ratio": 0.521,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4065,
+      "eval_loss": 0.6367995142936707,
+      "eval_runtime": 217.9899,
+      "eval_samples_per_second": 45.874,
+      "eval_steps_per_second": 0.243,
+      "step": 497000
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 3.4466619277906074e-05,
+      "loss": 0.6246,
+      "step": 497200
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 3.444024334436287e-05,
+      "loss": 0.6224,
+      "step": 497400
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 3.441386741081967e-05,
+      "loss": 0.6187,
+      "step": 497600
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 3.4387623356944196e-05,
+      "loss": 0.6205,
+      "step": 497800
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 3.4361247423400993e-05,
+      "loss": 0.6263,
+      "step": 498000
+    },
+    {
+      "epoch": 0.66,
+      "eval_em_first_line_ratio": 0.5181,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.404,
+      "eval_loss": 0.6366584897041321,
+      "eval_runtime": 222.545,
+      "eval_samples_per_second": 44.935,
+      "eval_steps_per_second": 0.238,
+      "step": 498000
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 3.43348714898578e-05,
+      "loss": 0.6248,
+      "step": 498200
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 3.430862743598232e-05,
+      "loss": 0.6222,
+      "step": 498400
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 3.4282251502439115e-05,
+      "loss": 0.6242,
+      "step": 498600
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 3.425587556889592e-05,
+      "loss": 0.6223,
+      "step": 498800
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 3.422949963535272e-05,
+      "loss": 0.622,
+      "step": 499000
+    },
+    {
+      "epoch": 0.66,
+      "eval_em_first_line_ratio": 0.5216,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4075,
+      "eval_loss": 0.6371378302574158,
+      "eval_runtime": 226.7455,
+      "eval_samples_per_second": 44.102,
+      "eval_steps_per_second": 0.234,
+      "step": 499000
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 3.420312370180952e-05,
+      "loss": 0.6233,
+      "step": 499200
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 3.4176747768266326e-05,
+      "loss": 0.6198,
+      "step": 499400
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 3.4150371834723124e-05,
+      "loss": 0.6237,
+      "step": 499600
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 3.412399590117993e-05,
+      "loss": 0.6248,
+      "step": 499800
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 3.4097619967636726e-05,
+      "loss": 0.6188,
+      "step": 500000
+    },
+    {
+      "epoch": 0.66,
+      "eval_em_first_line_ratio": 0.5219,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4076,
+      "eval_loss": 0.6368451714515686,
+      "eval_runtime": 227.9707,
+      "eval_samples_per_second": 43.865,
+      "eval_steps_per_second": 0.232,
+      "step": 500000
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 3.407124403409354e-05,
+      "loss": 0.6237,
+      "step": 500200
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 3.4044868100550335e-05,
+      "loss": 0.6212,
+      "step": 500400
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 3.401849216700714e-05,
+      "loss": 0.6191,
+      "step": 500600
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 3.399211623346394e-05,
+      "loss": 0.623,
+      "step": 500800
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 3.396574029992074e-05,
+      "loss": 0.6277,
+      "step": 501000
+    },
+    {
+      "epoch": 0.66,
+      "eval_em_first_line_ratio": 0.5216,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4073,
+      "eval_loss": 0.6367818117141724,
+      "eval_runtime": 219.0149,
+      "eval_samples_per_second": 45.659,
+      "eval_steps_per_second": 0.242,
+      "step": 501000
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 3.3939364366377546e-05,
+      "loss": 0.6226,
+      "step": 501200
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 3.391298843283435e-05,
+      "loss": 0.6237,
+      "step": 501400
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 3.388661249929115e-05,
+      "loss": 0.6219,
+      "step": 501600
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 3.386023656574795e-05,
+      "loss": 0.6208,
+      "step": 501800
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 3.383386063220475e-05,
+      "loss": 0.6221,
+      "step": 502000
+    },
+    {
+      "epoch": 0.66,
+      "eval_em_first_line_ratio": 0.5203,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4066,
+      "eval_loss": 0.6363468766212463,
+      "eval_runtime": 220.2887,
+      "eval_samples_per_second": 45.395,
+      "eval_steps_per_second": 0.241,
+      "step": 502000
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 3.3807484698661554e-05,
+      "loss": 0.623,
+      "step": 502200
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 3.378110876511836e-05,
+      "loss": 0.6238,
+      "step": 502400
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 3.3754732831575156e-05,
+      "loss": 0.6176,
+      "step": 502600
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 3.372835689803196e-05,
+      "loss": 0.6238,
+      "step": 502800
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 3.370198096448876e-05,
+      "loss": 0.6229,
+      "step": 503000
+    },
+    {
+      "epoch": 0.66,
+      "eval_em_first_line_ratio": 0.5185,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.404,
+      "eval_loss": 0.6363696455955505,
+      "eval_runtime": 225.3778,
+      "eval_samples_per_second": 44.37,
+      "eval_steps_per_second": 0.235,
+      "step": 503000
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 3.367560503094557e-05,
+      "loss": 0.6243,
+      "step": 503200
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 3.364922909740237e-05,
+      "loss": 0.6205,
+      "step": 503400
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 3.362285316385917e-05,
+      "loss": 0.6207,
+      "step": 503600
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 3.359647723031597e-05,
+      "loss": 0.6198,
+      "step": 503800
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 3.3570101296772774e-05,
+      "loss": 0.6208,
+      "step": 504000
+    },
+    {
+      "epoch": 0.66,
+      "eval_em_first_line_ratio": 0.522,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4081,
+      "eval_loss": 0.6362546682357788,
+      "eval_runtime": 222.8039,
+      "eval_samples_per_second": 44.883,
+      "eval_steps_per_second": 0.238,
+      "step": 504000
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 3.354372536322958e-05,
+      "loss": 0.6208,
+      "step": 504200
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 3.3517349429686376e-05,
+      "loss": 0.6216,
+      "step": 504400
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 3.349097349614318e-05,
+      "loss": 0.6181,
+      "step": 504600
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 3.34647294422677e-05,
+      "loss": 0.6204,
+      "step": 504800
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 3.3438353508724505e-05,
+      "loss": 0.6247,
+      "step": 505000
+    },
+    {
+      "epoch": 0.67,
+      "eval_em_first_line_ratio": 0.5203,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4068,
+      "eval_loss": 0.6362699270248413,
+      "eval_runtime": 224.3745,
+      "eval_samples_per_second": 44.568,
+      "eval_steps_per_second": 0.236,
+      "step": 505000
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 3.34119775751813e-05,
+      "loss": 0.6235,
+      "step": 505200
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 3.33856016416381e-05,
+      "loss": 0.6206,
+      "step": 505400
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 3.335922570809491e-05,
+      "loss": 0.6233,
+      "step": 505600
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 3.333284977455171e-05,
+      "loss": 0.6209,
+      "step": 505800
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 3.330647384100851e-05,
+      "loss": 0.6248,
+      "step": 506000
+    },
+    {
+      "epoch": 0.67,
+      "eval_em_first_line_ratio": 0.5219,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.407,
+      "eval_loss": 0.6365532279014587,
+      "eval_runtime": 220.5626,
+      "eval_samples_per_second": 45.339,
+      "eval_steps_per_second": 0.24,
+      "step": 506000
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 3.328009790746531e-05,
+      "loss": 0.6162,
+      "step": 506200
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 3.3253721973922115e-05,
+      "loss": 0.6183,
+      "step": 506400
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 3.3227477920046635e-05,
+      "loss": 0.6241,
+      "step": 506600
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 3.320110198650343e-05,
+      "loss": 0.6204,
+      "step": 506800
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 3.317472605296024e-05,
+      "loss": 0.6203,
+      "step": 507000
+    },
+    {
+      "epoch": 0.67,
+      "eval_em_first_line_ratio": 0.5235,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4086,
+      "eval_loss": 0.6361768245697021,
+      "eval_runtime": 233.5205,
+      "eval_samples_per_second": 42.823,
+      "eval_steps_per_second": 0.227,
+      "step": 507000
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 3.314835011941704e-05,
+      "loss": 0.6207,
+      "step": 507200
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 3.3121974185873846e-05,
+      "loss": 0.6165,
+      "step": 507400
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 3.3095598252330644e-05,
+      "loss": 0.6228,
+      "step": 507600
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 3.306922231878745e-05,
+      "loss": 0.622,
+      "step": 507800
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 3.3042846385244246e-05,
+      "loss": 0.6266,
+      "step": 508000
+    },
+    {
+      "epoch": 0.67,
+      "eval_em_first_line_ratio": 0.5203,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4056,
+      "eval_loss": 0.6356428265571594,
+      "eval_runtime": 227.7191,
+      "eval_samples_per_second": 43.914,
+      "eval_steps_per_second": 0.233,
+      "step": 508000
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 3.301647045170105e-05,
+      "loss": 0.6211,
+      "step": 508200
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 3.2990094518157855e-05,
+      "loss": 0.6247,
+      "step": 508400
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 3.296371858461465e-05,
+      "loss": 0.6166,
+      "step": 508600
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 3.293734265107146e-05,
+      "loss": 0.6172,
+      "step": 508800
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 3.291109859719598e-05,
+      "loss": 0.6262,
+      "step": 509000
+    },
+    {
+      "epoch": 0.67,
+      "eval_em_first_line_ratio": 0.5195,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4054,
+      "eval_loss": 0.6359286904335022,
+      "eval_runtime": 223.019,
+      "eval_samples_per_second": 44.839,
+      "eval_steps_per_second": 0.238,
+      "step": 509000
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 3.288485454332049e-05,
+      "loss": 0.6215,
+      "step": 509200
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 3.28584786097773e-05,
+      "loss": 0.6227,
+      "step": 509400
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 3.28321026762341e-05,
+      "loss": 0.6213,
+      "step": 509600
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 3.2805726742690903e-05,
+      "loss": 0.6174,
+      "step": 509800
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 3.27793508091477e-05,
+      "loss": 0.6237,
+      "step": 510000
+    },
+    {
+      "epoch": 0.67,
+      "eval_em_first_line_ratio": 0.5195,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4053,
+      "eval_loss": 0.6358693838119507,
+      "eval_runtime": 232.3908,
+      "eval_samples_per_second": 43.031,
+      "eval_steps_per_second": 0.228,
+      "step": 510000
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 3.2752974875604505e-05,
+      "loss": 0.6197,
+      "step": 510200
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 3.272659894206131e-05,
+      "loss": 0.625,
+      "step": 510400
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 3.270022300851811e-05,
+      "loss": 0.618,
+      "step": 510600
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 3.267384707497491e-05,
+      "loss": 0.6159,
+      "step": 510800
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 3.264747114143171e-05,
+      "loss": 0.6223,
+      "step": 511000
+    },
+    {
+      "epoch": 0.67,
+      "eval_em_first_line_ratio": 0.5202,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4056,
+      "eval_loss": 0.6356240510940552,
+      "eval_runtime": 226.4106,
+      "eval_samples_per_second": 44.168,
+      "eval_steps_per_second": 0.234,
+      "step": 511000
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 3.2621095207888514e-05,
+      "loss": 0.6205,
+      "step": 511200
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 3.259471927434532e-05,
+      "loss": 0.6253,
+      "step": 511400
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 3.256834334080212e-05,
+      "loss": 0.6237,
+      "step": 511600
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 3.2542099286926636e-05,
+      "loss": 0.6124,
+      "step": 511800
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 3.251572335338344e-05,
+      "loss": 0.6216,
+      "step": 512000
+    },
+    {
+      "epoch": 0.68,
+      "eval_em_first_line_ratio": 0.5218,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4072,
+      "eval_loss": 0.6351751685142517,
+      "eval_runtime": 230.8432,
+      "eval_samples_per_second": 43.319,
+      "eval_steps_per_second": 0.23,
+      "step": 512000
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 3.2489347419840245e-05,
+      "loss": 0.6187,
+      "step": 512200
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 3.246297148629704e-05,
+      "loss": 0.6188,
+      "step": 512400
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 3.243659555275385e-05,
+      "loss": 0.625,
+      "step": 512600
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 3.2410219619210645e-05,
+      "loss": 0.6178,
+      "step": 512800
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 3.2383843685667456e-05,
+      "loss": 0.6189,
+      "step": 513000
+    },
+    {
+      "epoch": 0.68,
+      "eval_em_first_line_ratio": 0.5195,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4052,
+      "eval_loss": 0.6350366473197937,
+      "eval_runtime": 226.8132,
+      "eval_samples_per_second": 44.089,
+      "eval_steps_per_second": 0.234,
+      "step": 513000
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 3.2357467752124254e-05,
+      "loss": 0.6192,
+      "step": 513200
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 3.233109181858106e-05,
+      "loss": 0.619,
+      "step": 513400
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 3.2304715885037856e-05,
+      "loss": 0.6242,
+      "step": 513600
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 3.227833995149466e-05,
+      "loss": 0.6191,
+      "step": 513800
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 3.2251964017951464e-05,
+      "loss": 0.6189,
+      "step": 514000
+    },
+    {
+      "epoch": 0.68,
+      "eval_em_first_line_ratio": 0.5221,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.408,
+      "eval_loss": 0.6352951526641846,
+      "eval_runtime": 226.9024,
+      "eval_samples_per_second": 44.072,
+      "eval_steps_per_second": 0.234,
+      "step": 514000
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 3.222558808440826e-05,
+      "loss": 0.6217,
+      "step": 514200
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 3.2199212150865066e-05,
+      "loss": 0.6225,
+      "step": 514400
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 3.2172836217321864e-05,
+      "loss": 0.6204,
+      "step": 514600
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 3.2146460283778675e-05,
+      "loss": 0.6245,
+      "step": 514800
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 3.212008435023547e-05,
+      "loss": 0.6221,
+      "step": 515000
+    },
+    {
+      "epoch": 0.68,
+      "eval_em_first_line_ratio": 0.5215,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4073,
+      "eval_loss": 0.6353150606155396,
+      "eval_runtime": 230.7708,
+      "eval_samples_per_second": 43.333,
+      "eval_steps_per_second": 0.23,
+      "step": 515000
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 3.209370841669228e-05,
+      "loss": 0.6212,
+      "step": 515200
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 3.2067332483149075e-05,
+      "loss": 0.6186,
+      "step": 515400
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 3.204095654960588e-05,
+      "loss": 0.618,
+      "step": 515600
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 3.2014580616062684e-05,
+      "loss": 0.6198,
+      "step": 515800
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 3.19883365621872e-05,
+      "loss": 0.6188,
+      "step": 516000
+    },
+    {
+      "epoch": 0.68,
+      "eval_em_first_line_ratio": 0.5196,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.405,
+      "eval_loss": 0.6352788209915161,
+      "eval_runtime": 221.3119,
+      "eval_samples_per_second": 45.185,
+      "eval_steps_per_second": 0.239,
+      "step": 516000
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 3.1961960628644e-05,
+      "loss": 0.6191,
+      "step": 516200
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 3.1935584695100806e-05,
+      "loss": 0.6188,
+      "step": 516400
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 3.190920876155761e-05,
+      "loss": 0.6191,
+      "step": 516600
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 3.188283282801441e-05,
+      "loss": 0.622,
+      "step": 516800
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 3.185645689447121e-05,
+      "loss": 0.62,
+      "step": 517000
+    },
+    {
+      "epoch": 0.68,
+      "eval_em_first_line_ratio": 0.5217,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4079,
+      "eval_loss": 0.6344280242919922,
+      "eval_runtime": 223.5365,
+      "eval_samples_per_second": 44.735,
+      "eval_steps_per_second": 0.237,
+      "step": 517000
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 3.183008096092801e-05,
+      "loss": 0.6185,
+      "step": 517200
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 3.1803705027384815e-05,
+      "loss": 0.622,
+      "step": 517400
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 3.177732909384162e-05,
+      "loss": 0.6193,
+      "step": 517600
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 3.1750953160298417e-05,
+      "loss": 0.6198,
+      "step": 517800
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 3.172457722675522e-05,
+      "loss": 0.6208,
+      "step": 518000
+    },
+    {
+      "epoch": 0.68,
+      "eval_em_first_line_ratio": 0.5197,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4057,
+      "eval_loss": 0.6347582340240479,
+      "eval_runtime": 224.1784,
+      "eval_samples_per_second": 44.607,
+      "eval_steps_per_second": 0.236,
+      "step": 518000
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 3.169820129321202e-05,
+      "loss": 0.6158,
+      "step": 518200
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 3.167182535966883e-05,
+      "loss": 0.617,
+      "step": 518400
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 3.164544942612563e-05,
+      "loss": 0.6202,
+      "step": 518600
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 3.161907349258243e-05,
+      "loss": 0.6201,
+      "step": 518800
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 3.159269755903923e-05,
+      "loss": 0.6234,
+      "step": 519000
+    },
+    {
+      "epoch": 0.68,
+      "eval_em_first_line_ratio": 0.52,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4061,
+      "eval_loss": 0.6341407299041748,
+      "eval_runtime": 229.5008,
+      "eval_samples_per_second": 43.573,
+      "eval_steps_per_second": 0.231,
+      "step": 519000
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 3.1566321625496034e-05,
+      "loss": 0.6211,
+      "step": 519200
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 3.153994569195284e-05,
+      "loss": 0.622,
+      "step": 519400
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 3.1513569758409636e-05,
+      "loss": 0.6185,
+      "step": 519600
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 3.148719382486644e-05,
+      "loss": 0.6203,
+      "step": 519800
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 3.146081789132324e-05,
+      "loss": 0.6185,
+      "step": 520000
+    },
+    {
+      "epoch": 0.69,
+      "eval_em_first_line_ratio": 0.5214,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4078,
+      "eval_loss": 0.6338856220245361,
+      "eval_runtime": 224.8031,
+      "eval_samples_per_second": 44.483,
+      "eval_steps_per_second": 0.236,
+      "step": 520000
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 3.1434573837447765e-05,
+      "loss": 0.6166,
+      "step": 520200
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 3.140819790390456e-05,
+      "loss": 0.6207,
+      "step": 520400
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 3.138182197036137e-05,
+      "loss": 0.6177,
+      "step": 520600
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 3.1355446036818165e-05,
+      "loss": 0.6159,
+      "step": 520800
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 3.132907010327497e-05,
+      "loss": 0.6229,
+      "step": 521000
+    },
+    {
+      "epoch": 0.69,
+      "eval_em_first_line_ratio": 0.5199,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4056,
+      "eval_loss": 0.6338607668876648,
+      "eval_runtime": 220.6923,
+      "eval_samples_per_second": 45.312,
+      "eval_steps_per_second": 0.24,
+      "step": 521000
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 3.1302694169731773e-05,
+      "loss": 0.6224,
+      "step": 521200
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 3.127631823618857e-05,
+      "loss": 0.6171,
+      "step": 521400
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 3.1249942302645376e-05,
+      "loss": 0.6177,
+      "step": 521600
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 3.1223698248769896e-05,
+      "loss": 0.6167,
+      "step": 521800
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 3.119745419489441e-05,
+      "loss": 0.6191,
+      "step": 522000
+    },
+    {
+      "epoch": 0.69,
+      "eval_em_first_line_ratio": 0.5224,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4074,
+      "eval_loss": 0.6341127157211304,
+      "eval_runtime": 223.0382,
+      "eval_samples_per_second": 44.835,
+      "eval_steps_per_second": 0.238,
+      "step": 522000
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 3.117107826135121e-05,
+      "loss": 0.6149,
+      "step": 522200
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 3.114483420747573e-05,
+      "loss": 0.6155,
+      "step": 522400
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 3.111845827393253e-05,
+      "loss": 0.6155,
+      "step": 522600
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 3.109208234038934e-05,
+      "loss": 0.6211,
+      "step": 522800
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 3.106570640684614e-05,
+      "loss": 0.6237,
+      "step": 523000
+    },
+    {
+      "epoch": 0.69,
+      "eval_em_first_line_ratio": 0.5223,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4084,
+      "eval_loss": 0.6338132619857788,
+      "eval_runtime": 225.5007,
+      "eval_samples_per_second": 44.346,
+      "eval_steps_per_second": 0.235,
+      "step": 523000
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 3.1039330473302944e-05,
+      "loss": 0.6207,
+      "step": 523200
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 3.101295453975974e-05,
+      "loss": 0.6205,
+      "step": 523400
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 3.0986578606216546e-05,
+      "loss": 0.619,
+      "step": 523600
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 3.096020267267335e-05,
+      "loss": 0.6219,
+      "step": 523800
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 3.093382673913015e-05,
+      "loss": 0.62,
+      "step": 524000
+    },
+    {
+      "epoch": 0.69,
+      "eval_em_first_line_ratio": 0.521,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4073,
+      "eval_loss": 0.6333073377609253,
+      "eval_runtime": 220.3846,
+      "eval_samples_per_second": 45.375,
+      "eval_steps_per_second": 0.24,
+      "step": 524000
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 3.090745080558695e-05,
+      "loss": 0.6189,
+      "step": 524200
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 3.088107487204375e-05,
+      "loss": 0.6183,
+      "step": 524400
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 3.0854698938500555e-05,
+      "loss": 0.6198,
+      "step": 524600
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 3.082832300495736e-05,
+      "loss": 0.617,
+      "step": 524800
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 3.0801947071414164e-05,
+      "loss": 0.6172,
+      "step": 525000
+    },
+    {
+      "epoch": 0.69,
+      "eval_em_first_line_ratio": 0.5212,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4067,
+      "eval_loss": 0.6344726085662842,
+      "eval_runtime": 222.135,
+      "eval_samples_per_second": 45.018,
+      "eval_steps_per_second": 0.239,
+      "step": 525000
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 3.077557113787096e-05,
+      "loss": 0.62,
+      "step": 525200
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 3.0749195204327766e-05,
+      "loss": 0.6181,
+      "step": 525400
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 3.072281927078456e-05,
+      "loss": 0.6168,
+      "step": 525600
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 3.069644333724137e-05,
+      "loss": 0.6225,
+      "step": 525800
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 3.067006740369817e-05,
+      "loss": 0.615,
+      "step": 526000
+    },
+    {
+      "epoch": 0.69,
+      "eval_em_first_line_ratio": 0.5253,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4101,
+      "eval_loss": 0.6331153512001038,
+      "eval_runtime": 219.7726,
+      "eval_samples_per_second": 45.502,
+      "eval_steps_per_second": 0.241,
+      "step": 526000
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 3.064369147015497e-05,
+      "loss": 0.6146,
+      "step": 526200
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 3.0617315536611774e-05,
+      "loss": 0.6166,
+      "step": 526400
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 3.059093960306858e-05,
+      "loss": 0.6189,
+      "step": 526600
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 3.056456366952538e-05,
+      "loss": 0.6184,
+      "step": 526800
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 3.053818773598218e-05,
+      "loss": 0.6149,
+      "step": 527000
+    },
+    {
+      "epoch": 0.69,
+      "eval_em_first_line_ratio": 0.5239,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4096,
+      "eval_loss": 0.6339222192764282,
+      "eval_runtime": 227.5369,
+      "eval_samples_per_second": 43.949,
+      "eval_steps_per_second": 0.233,
+      "step": 527000
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 3.051181180243898e-05,
+      "loss": 0.6155,
+      "step": 527200
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 3.0485435868895783e-05,
+      "loss": 0.6165,
+      "step": 527400
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 3.045905993535259e-05,
+      "loss": 0.618,
+      "step": 527600
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 3.043268400180939e-05,
+      "loss": 0.6144,
+      "step": 527800
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 3.0406308068266193e-05,
+      "loss": 0.6189,
+      "step": 528000
+    },
+    {
+      "epoch": 0.7,
+      "eval_em_first_line_ratio": 0.5223,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4084,
+      "eval_loss": 0.6325196027755737,
+      "eval_runtime": 227.0218,
+      "eval_samples_per_second": 44.049,
+      "eval_steps_per_second": 0.233,
+      "step": 528000
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 3.0379932134722994e-05,
+      "loss": 0.616,
+      "step": 528200
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 3.0353556201179795e-05,
+      "loss": 0.6169,
+      "step": 528400
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 3.0327180267636602e-05,
+      "loss": 0.6125,
+      "step": 528600
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 3.0300936213761116e-05,
+      "loss": 0.6169,
+      "step": 528800
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 3.0274560280217917e-05,
+      "loss": 0.6152,
+      "step": 529000
+    },
+    {
+      "epoch": 0.7,
+      "eval_em_first_line_ratio": 0.5224,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4088,
+      "eval_loss": 0.6323828101158142,
+      "eval_runtime": 230.2029,
+      "eval_samples_per_second": 43.44,
+      "eval_steps_per_second": 0.23,
+      "step": 529000
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 3.0248184346674725e-05,
+      "loss": 0.6153,
+      "step": 529200
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 3.0221808413131526e-05,
+      "loss": 0.6148,
+      "step": 529400
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 3.0195432479588327e-05,
+      "loss": 0.6195,
+      "step": 529600
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 3.0169056546045128e-05,
+      "loss": 0.6155,
+      "step": 529800
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 3.014268061250193e-05,
+      "loss": 0.6208,
+      "step": 530000
+    },
+    {
+      "epoch": 0.7,
+      "eval_em_first_line_ratio": 0.5217,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4075,
+      "eval_loss": 0.6325795650482178,
+      "eval_runtime": 220.5889,
+      "eval_samples_per_second": 45.333,
+      "eval_steps_per_second": 0.24,
+      "step": 530000
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 3.0116304678958733e-05,
+      "loss": 0.6238,
+      "step": 530200
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 3.0089928745415534e-05,
+      "loss": 0.618,
+      "step": 530400
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 3.0063552811872335e-05,
+      "loss": 0.6181,
+      "step": 530600
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 3.0037176878329136e-05,
+      "loss": 0.6148,
+      "step": 530800
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 3.0010800944785937e-05,
+      "loss": 0.6167,
+      "step": 531000
+    },
+    {
+      "epoch": 0.7,
+      "eval_em_first_line_ratio": 0.5253,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4105,
+      "eval_loss": 0.6325233578681946,
+      "eval_runtime": 221.8926,
+      "eval_samples_per_second": 45.067,
+      "eval_steps_per_second": 0.239,
+      "step": 531000
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 2.9984425011242745e-05,
+      "loss": 0.6204,
+      "step": 531200
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 2.9958049077699546e-05,
+      "loss": 0.618,
+      "step": 531400
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 2.9931673144156347e-05,
+      "loss": 0.619,
+      "step": 531600
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 2.9905297210613148e-05,
+      "loss": 0.6157,
+      "step": 531800
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 2.9878921277069956e-05,
+      "loss": 0.619,
+      "step": 532000
+    },
+    {
+      "epoch": 0.7,
+      "eval_em_first_line_ratio": 0.5202,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4058,
+      "eval_loss": 0.6328157186508179,
+      "eval_runtime": 222.6619,
+      "eval_samples_per_second": 44.911,
+      "eval_steps_per_second": 0.238,
+      "step": 532000
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 2.9852545343526757e-05,
+      "loss": 0.6125,
+      "step": 532200
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 2.9826169409983558e-05,
+      "loss": 0.6136,
+      "step": 532400
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 2.979979347644036e-05,
+      "loss": 0.6158,
+      "step": 532600
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 2.977341754289716e-05,
+      "loss": 0.6169,
+      "step": 532800
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 2.9747041609353965e-05,
+      "loss": 0.6136,
+      "step": 533000
+    },
+    {
+      "epoch": 0.7,
+      "eval_em_first_line_ratio": 0.5229,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4082,
+      "eval_loss": 0.6328060626983643,
+      "eval_runtime": 225.086,
+      "eval_samples_per_second": 44.427,
+      "eval_steps_per_second": 0.235,
+      "step": 533000
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 2.9720665675810766e-05,
+      "loss": 0.618,
+      "step": 533200
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 2.9694289742267567e-05,
+      "loss": 0.6152,
+      "step": 533400
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 2.9668045688392087e-05,
+      "loss": 0.6148,
+      "step": 533600
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 2.9641669754848888e-05,
+      "loss": 0.6133,
+      "step": 533800
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 2.961529382130569e-05,
+      "loss": 0.6149,
+      "step": 534000
+    },
+    {
+      "epoch": 0.7,
+      "eval_em_first_line_ratio": 0.5219,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4079,
+      "eval_loss": 0.6325260996818542,
+      "eval_runtime": 219.2526,
+      "eval_samples_per_second": 45.61,
+      "eval_steps_per_second": 0.242,
+      "step": 534000
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 2.958891788776249e-05,
+      "loss": 0.6173,
+      "step": 534200
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 2.956254195421929e-05,
+      "loss": 0.6116,
+      "step": 534400
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 2.95361660206761e-05,
+      "loss": 0.6159,
+      "step": 534600
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 2.95097900871329e-05,
+      "loss": 0.6156,
+      "step": 534800
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 2.94834141535897e-05,
+      "loss": 0.6175,
+      "step": 535000
+    },
+    {
+      "epoch": 0.71,
+      "eval_em_first_line_ratio": 0.5234,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4095,
+      "eval_loss": 0.6322922110557556,
+      "eval_runtime": 224.5414,
+      "eval_samples_per_second": 44.535,
+      "eval_steps_per_second": 0.236,
+      "step": 535000
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 2.94570382200465e-05,
+      "loss": 0.6174,
+      "step": 535200
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 2.9430662286503303e-05,
+      "loss": 0.6135,
+      "step": 535400
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 2.9404286352960107e-05,
+      "loss": 0.6166,
+      "step": 535600
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 2.9377910419416908e-05,
+      "loss": 0.614,
+      "step": 535800
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 2.9351666365541425e-05,
+      "loss": 0.6182,
+      "step": 536000
+    },
+    {
+      "epoch": 0.71,
+      "eval_em_first_line_ratio": 0.5224,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4087,
+      "eval_loss": 0.6324431896209717,
+      "eval_runtime": 220.8042,
+      "eval_samples_per_second": 45.289,
+      "eval_steps_per_second": 0.24,
+      "step": 536000
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 2.9325290431998233e-05,
+      "loss": 0.6142,
+      "step": 536200
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 2.9298914498455034e-05,
+      "loss": 0.615,
+      "step": 536400
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 2.9272670444579547e-05,
+      "loss": 0.6145,
+      "step": 536600
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 2.9246294511036355e-05,
+      "loss": 0.614,
+      "step": 536800
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 2.9219918577493156e-05,
+      "loss": 0.6187,
+      "step": 537000
+    },
+    {
+      "epoch": 0.71,
+      "eval_em_first_line_ratio": 0.5236,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4094,
+      "eval_loss": 0.6320123672485352,
+      "eval_runtime": 222.8568,
+      "eval_samples_per_second": 44.872,
+      "eval_steps_per_second": 0.238,
+      "step": 537000
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 2.9193542643949957e-05,
+      "loss": 0.6198,
+      "step": 537200
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 2.9167166710406758e-05,
+      "loss": 0.6156,
+      "step": 537400
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 2.914079077686356e-05,
+      "loss": 0.6154,
+      "step": 537600
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 2.9114414843320363e-05,
+      "loss": 0.6186,
+      "step": 537800
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 2.9088038909777164e-05,
+      "loss": 0.6132,
+      "step": 538000
+    },
+    {
+      "epoch": 0.71,
+      "eval_em_first_line_ratio": 0.5242,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4104,
+      "eval_loss": 0.6311558485031128,
+      "eval_runtime": 222.5665,
+      "eval_samples_per_second": 44.93,
+      "eval_steps_per_second": 0.238,
+      "step": 538000
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 2.9061662976233965e-05,
+      "loss": 0.6124,
+      "step": 538200
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 2.9035287042690766e-05,
+      "loss": 0.6145,
+      "step": 538400
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 2.9008911109147567e-05,
+      "loss": 0.6162,
+      "step": 538600
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 2.898266705527209e-05,
+      "loss": 0.6149,
+      "step": 538800
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 2.8956554881064323e-05,
+      "loss": 0.6222,
+      "step": 539000
+    },
+    {
+      "epoch": 0.71,
+      "eval_em_first_line_ratio": 0.5227,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.409,
+      "eval_loss": 0.6313305497169495,
+      "eval_runtime": 222.7401,
+      "eval_samples_per_second": 44.895,
+      "eval_steps_per_second": 0.238,
+      "step": 539000
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 2.8930178947521124e-05,
+      "loss": 0.6113,
+      "step": 539200
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 2.8903803013977925e-05,
+      "loss": 0.616,
+      "step": 539400
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 2.8877427080434726e-05,
+      "loss": 0.6164,
+      "step": 539600
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 2.8851051146891534e-05,
+      "loss": 0.6167,
+      "step": 539800
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 2.8824675213348335e-05,
+      "loss": 0.6182,
+      "step": 540000
+    },
+    {
+      "epoch": 0.71,
+      "eval_em_first_line_ratio": 0.5224,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4083,
+      "eval_loss": 0.6313698291778564,
+      "eval_runtime": 222.3639,
+      "eval_samples_per_second": 44.971,
+      "eval_steps_per_second": 0.238,
+      "step": 540000
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 2.8798299279805136e-05,
+      "loss": 0.6123,
+      "step": 540200
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 2.8771923346261937e-05,
+      "loss": 0.6159,
+      "step": 540400
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 2.874554741271874e-05,
+      "loss": 0.616,
+      "step": 540600
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 2.8719303358843258e-05,
+      "loss": 0.6186,
+      "step": 540800
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 2.869292742530006e-05,
+      "loss": 0.6176,
+      "step": 541000
+    },
+    {
+      "epoch": 0.71,
+      "eval_em_first_line_ratio": 0.5231,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4091,
+      "eval_loss": 0.6313590407371521,
+      "eval_runtime": 225.1916,
+      "eval_samples_per_second": 44.407,
+      "eval_steps_per_second": 0.235,
+      "step": 541000
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 2.866655149175686e-05,
+      "loss": 0.6138,
+      "step": 541200
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 2.8640175558213668e-05,
+      "loss": 0.6146,
+      "step": 541400
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 2.861379962467047e-05,
+      "loss": 0.6131,
+      "step": 541600
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 2.858742369112727e-05,
+      "loss": 0.6169,
+      "step": 541800
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 2.856104775758407e-05,
+      "loss": 0.6132,
+      "step": 542000
+    },
+    {
+      "epoch": 0.71,
+      "eval_em_first_line_ratio": 0.5212,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4072,
+      "eval_loss": 0.6312960386276245,
+      "eval_runtime": 225.2084,
+      "eval_samples_per_second": 44.403,
+      "eval_steps_per_second": 0.235,
+      "step": 542000
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 2.8534671824040875e-05,
+      "loss": 0.6136,
+      "step": 542200
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 2.8508295890497676e-05,
+      "loss": 0.6158,
+      "step": 542400
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 2.8481919956954477e-05,
+      "loss": 0.6144,
+      "step": 542600
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 2.845554402341128e-05,
+      "loss": 0.616,
+      "step": 542800
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 2.842916808986808e-05,
+      "loss": 0.6185,
+      "step": 543000
+    },
+    {
+      "epoch": 0.72,
+      "eval_em_first_line_ratio": 0.5236,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4091,
+      "eval_loss": 0.6311962008476257,
+      "eval_runtime": 222.4085,
+      "eval_samples_per_second": 44.962,
+      "eval_steps_per_second": 0.238,
+      "step": 543000
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 2.8402792156324887e-05,
+      "loss": 0.6137,
+      "step": 543200
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 2.8376416222781688e-05,
+      "loss": 0.612,
+      "step": 543400
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 2.835004028923849e-05,
+      "loss": 0.6161,
+      "step": 543600
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 2.832366435569529e-05,
+      "loss": 0.6145,
+      "step": 543800
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 2.829728842215209e-05,
+      "loss": 0.6126,
+      "step": 544000
+    },
+    {
+      "epoch": 0.72,
+      "eval_em_first_line_ratio": 0.5225,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.408,
+      "eval_loss": 0.6315240859985352,
+      "eval_runtime": 222.2991,
+      "eval_samples_per_second": 44.984,
+      "eval_steps_per_second": 0.238,
+      "step": 544000
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 2.8270912488608896e-05,
+      "loss": 0.6172,
+      "step": 544200
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 2.8244536555065697e-05,
+      "loss": 0.6159,
+      "step": 544400
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 2.8218160621522498e-05,
+      "loss": 0.6101,
+      "step": 544600
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 2.81917846879793e-05,
+      "loss": 0.6164,
+      "step": 544800
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 2.81654087544361e-05,
+      "loss": 0.6187,
+      "step": 545000
+    },
+    {
+      "epoch": 0.72,
+      "eval_em_first_line_ratio": 0.5234,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.409,
+      "eval_loss": 0.6310444474220276,
+      "eval_runtime": 225.8823,
+      "eval_samples_per_second": 44.271,
+      "eval_steps_per_second": 0.235,
+      "step": 545000
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 2.8139032820892908e-05,
+      "loss": 0.6117,
+      "step": 545200
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 2.811265688734971e-05,
+      "loss": 0.6141,
+      "step": 545400
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 2.808628095380651e-05,
+      "loss": 0.616,
+      "step": 545600
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 2.805990502026331e-05,
+      "loss": 0.6163,
+      "step": 545800
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 2.8033529086720112e-05,
+      "loss": 0.6145,
+      "step": 546000
+    },
+    {
+      "epoch": 0.72,
+      "eval_em_first_line_ratio": 0.5224,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4083,
+      "eval_loss": 0.6307168006896973,
+      "eval_runtime": 226.1266,
+      "eval_samples_per_second": 44.223,
+      "eval_steps_per_second": 0.234,
+      "step": 546000
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 2.800715315317692e-05,
+      "loss": 0.6161,
+      "step": 546200
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 2.798077721963372e-05,
+      "loss": 0.6159,
+      "step": 546400
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 2.7954401286090522e-05,
+      "loss": 0.6185,
+      "step": 546600
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 2.7928025352547323e-05,
+      "loss": 0.6113,
+      "step": 546800
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 2.7901649419004127e-05,
+      "loss": 0.6166,
+      "step": 547000
+    },
+    {
+      "epoch": 0.72,
+      "eval_em_first_line_ratio": 0.5231,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4086,
+      "eval_loss": 0.6302283406257629,
+      "eval_runtime": 224.6283,
+      "eval_samples_per_second": 44.518,
+      "eval_steps_per_second": 0.236,
+      "step": 547000
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 2.7875273485460928e-05,
+      "loss": 0.6121,
+      "step": 547200
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 2.784889755191773e-05,
+      "loss": 0.6131,
+      "step": 547400
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 2.782252161837453e-05,
+      "loss": 0.612,
+      "step": 547600
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 2.779614568483133e-05,
+      "loss": 0.6173,
+      "step": 547800
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 2.776976975128814e-05,
+      "loss": 0.6151,
+      "step": 548000
+    },
+    {
+      "epoch": 0.72,
+      "eval_em_first_line_ratio": 0.5224,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4082,
+      "eval_loss": 0.6307158470153809,
+      "eval_runtime": 223.2939,
+      "eval_samples_per_second": 44.784,
+      "eval_steps_per_second": 0.237,
+      "step": 548000
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 2.774339381774494e-05,
+      "loss": 0.6157,
+      "step": 548200
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 2.771701788420174e-05,
+      "loss": 0.6131,
+      "step": 548400
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 2.769077383032626e-05,
+      "loss": 0.6134,
+      "step": 548600
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 2.7664397896783062e-05,
+      "loss": 0.6089,
+      "step": 548800
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 2.7638021963239863e-05,
+      "loss": 0.6128,
+      "step": 549000
+    },
+    {
+      "epoch": 0.72,
+      "eval_em_first_line_ratio": 0.526,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4112,
+      "eval_loss": 0.6311665773391724,
+      "eval_runtime": 225.7868,
+      "eval_samples_per_second": 44.29,
+      "eval_steps_per_second": 0.235,
+      "step": 549000
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 2.7611646029696664e-05,
+      "loss": 0.6137,
+      "step": 549200
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 2.7585270096153465e-05,
+      "loss": 0.6122,
+      "step": 549400
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 2.755889416261027e-05,
+      "loss": 0.611,
+      "step": 549600
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 2.753251822906707e-05,
+      "loss": 0.6127,
+      "step": 549800
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 2.7506142295523872e-05,
+      "loss": 0.6119,
+      "step": 550000
+    },
+    {
+      "epoch": 0.73,
+      "eval_em_first_line_ratio": 0.5228,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.409,
+      "eval_loss": 0.6309670805931091,
+      "eval_runtime": 223.534,
+      "eval_samples_per_second": 44.736,
+      "eval_steps_per_second": 0.237,
+      "step": 550000
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 2.7479766361980673e-05,
+      "loss": 0.6179,
+      "step": 550200
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 2.7453390428437474e-05,
+      "loss": 0.6157,
+      "step": 550400
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 2.7427014494894282e-05,
+      "loss": 0.6161,
+      "step": 550600
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 2.7400638561351083e-05,
+      "loss": 0.6143,
+      "step": 550800
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 2.7374262627807884e-05,
+      "loss": 0.614,
+      "step": 551000
+    },
+    {
+      "epoch": 0.73,
+      "eval_em_first_line_ratio": 0.5232,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4087,
+      "eval_loss": 0.6307095289230347,
+      "eval_runtime": 227.1218,
+      "eval_samples_per_second": 44.029,
+      "eval_steps_per_second": 0.233,
+      "step": 551000
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 2.7347886694264685e-05,
+      "loss": 0.6111,
+      "step": 551200
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 2.7321510760721486e-05,
+      "loss": 0.6163,
+      "step": 551400
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 2.7295134827178294e-05,
+      "loss": 0.6111,
+      "step": 551600
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 2.7268758893635095e-05,
+      "loss": 0.6119,
+      "step": 551800
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 2.7242382960091896e-05,
+      "loss": 0.6105,
+      "step": 552000
+    },
+    {
+      "epoch": 0.73,
+      "eval_em_first_line_ratio": 0.5238,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.409,
+      "eval_loss": 0.6299654245376587,
+      "eval_runtime": 219.7715,
+      "eval_samples_per_second": 45.502,
+      "eval_steps_per_second": 0.241,
+      "step": 552000
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 2.7216007026548697e-05,
+      "loss": 0.6135,
+      "step": 552200
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 2.71896310930055e-05,
+      "loss": 0.6131,
+      "step": 552400
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 2.7163255159462302e-05,
+      "loss": 0.6127,
+      "step": 552600
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 2.7136879225919103e-05,
+      "loss": 0.6099,
+      "step": 552800
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 2.711063517204362e-05,
+      "loss": 0.6129,
+      "step": 553000
+    },
+    {
+      "epoch": 0.73,
+      "eval_em_first_line_ratio": 0.519,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4054,
+      "eval_loss": 0.6302984356880188,
+      "eval_runtime": 229.1249,
+      "eval_samples_per_second": 43.644,
+      "eval_steps_per_second": 0.231,
+      "step": 553000
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 2.7084259238500424e-05,
+      "loss": 0.6171,
+      "step": 553200
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 2.705801518462494e-05,
+      "loss": 0.6144,
+      "step": 553400
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 2.7031639251081742e-05,
+      "loss": 0.6087,
+      "step": 553600
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 2.700526331753855e-05,
+      "loss": 0.6129,
+      "step": 553800
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 2.697888738399535e-05,
+      "loss": 0.612,
+      "step": 554000
+    },
+    {
+      "epoch": 0.73,
+      "eval_em_first_line_ratio": 0.5236,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4092,
+      "eval_loss": 0.6296770572662354,
+      "eval_runtime": 223.0833,
+      "eval_samples_per_second": 44.826,
+      "eval_steps_per_second": 0.238,
+      "step": 554000
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 2.6952511450452152e-05,
+      "loss": 0.6083,
+      "step": 554200
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 2.6926135516908953e-05,
+      "loss": 0.6128,
+      "step": 554400
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 2.6899759583365754e-05,
+      "loss": 0.6118,
+      "step": 554600
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 2.6873383649822558e-05,
+      "loss": 0.6106,
+      "step": 554800
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 2.684700771627936e-05,
+      "loss": 0.6139,
+      "step": 555000
+    },
+    {
+      "epoch": 0.73,
+      "eval_em_first_line_ratio": 0.5239,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4096,
+      "eval_loss": 0.6295897364616394,
+      "eval_runtime": 219.4598,
+      "eval_samples_per_second": 45.566,
+      "eval_steps_per_second": 0.242,
+      "step": 555000
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 2.682063178273616e-05,
+      "loss": 0.6102,
+      "step": 555200
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 2.679425584919296e-05,
+      "loss": 0.6132,
+      "step": 555400
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 2.676787991564977e-05,
+      "loss": 0.6109,
+      "step": 555600
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 2.674150398210657e-05,
+      "loss": 0.6095,
+      "step": 555800
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 2.671512804856337e-05,
+      "loss": 0.6113,
+      "step": 556000
+    },
+    {
+      "epoch": 0.73,
+      "eval_em_first_line_ratio": 0.5242,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4097,
+      "eval_loss": 0.629651665687561,
+      "eval_runtime": 223.3082,
+      "eval_samples_per_second": 44.781,
+      "eval_steps_per_second": 0.237,
+      "step": 556000
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 2.6688752115020172e-05,
+      "loss": 0.6113,
+      "step": 556200
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 2.6662376181476973e-05,
+      "loss": 0.616,
+      "step": 556400
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 2.6636000247933778e-05,
+      "loss": 0.6097,
+      "step": 556600
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 2.660962431439058e-05,
+      "loss": 0.6109,
+      "step": 556800
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 2.658324838084738e-05,
+      "loss": 0.6118,
+      "step": 557000
+    },
+    {
+      "epoch": 0.73,
+      "eval_em_first_line_ratio": 0.5238,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4095,
+      "eval_loss": 0.6293933391571045,
+      "eval_runtime": 221.3981,
+      "eval_samples_per_second": 45.168,
+      "eval_steps_per_second": 0.239,
+      "step": 557000
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 2.655687244730418e-05,
+      "loss": 0.6145,
+      "step": 557200
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 2.6530496513760982e-05,
+      "loss": 0.6155,
+      "step": 557400
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 2.650412058021779e-05,
+      "loss": 0.612,
+      "step": 557600
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 2.647774464667459e-05,
+      "loss": 0.6125,
+      "step": 557800
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 2.6451368713131392e-05,
+      "loss": 0.6158,
+      "step": 558000
+    },
+    {
+      "epoch": 0.74,
+      "eval_em_first_line_ratio": 0.5263,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4122,
+      "eval_loss": 0.6297279000282288,
+      "eval_runtime": 224.8156,
+      "eval_samples_per_second": 44.481,
+      "eval_steps_per_second": 0.236,
+      "step": 558000
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 2.6424992779588193e-05,
+      "loss": 0.6073,
+      "step": 558200
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 2.6398616846044994e-05,
+      "loss": 0.6092,
+      "step": 558400
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 2.6372240912501798e-05,
+      "loss": 0.6114,
+      "step": 558600
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 2.63458649789586e-05,
+      "loss": 0.609,
+      "step": 558800
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 2.6319620925083116e-05,
+      "loss": 0.6135,
+      "step": 559000
+    },
+    {
+      "epoch": 0.74,
+      "eval_em_first_line_ratio": 0.5245,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4105,
+      "eval_loss": 0.6291429996490479,
+      "eval_runtime": 223.4437,
+      "eval_samples_per_second": 44.754,
+      "eval_steps_per_second": 0.237,
+      "step": 559000
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 2.6293244991539924e-05,
+      "loss": 0.6138,
+      "step": 559200
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 2.6266869057996725e-05,
+      "loss": 0.6089,
+      "step": 559400
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 2.6240493124453526e-05,
+      "loss": 0.6166,
+      "step": 559600
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 2.6214117190910327e-05,
+      "loss": 0.6105,
+      "step": 559800
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 2.6187741257367128e-05,
+      "loss": 0.6105,
+      "step": 560000
+    },
+    {
+      "epoch": 0.74,
+      "eval_em_first_line_ratio": 0.5237,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4094,
+      "eval_loss": 0.6294736266136169,
+      "eval_runtime": 224.6965,
+      "eval_samples_per_second": 44.504,
+      "eval_steps_per_second": 0.236,
+      "step": 560000
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 2.6161365323823932e-05,
+      "loss": 0.6168,
+      "step": 560200
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 2.6134989390280733e-05,
+      "loss": 0.6103,
+      "step": 560400
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 2.6108613456737534e-05,
+      "loss": 0.6136,
+      "step": 560600
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 2.6082237523194335e-05,
+      "loss": 0.6152,
+      "step": 560800
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 2.6055861589651143e-05,
+      "loss": 0.6143,
+      "step": 561000
+    },
+    {
+      "epoch": 0.74,
+      "eval_em_first_line_ratio": 0.523,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.409,
+      "eval_loss": 0.6293506622314453,
+      "eval_runtime": 225.2442,
+      "eval_samples_per_second": 44.396,
+      "eval_steps_per_second": 0.235,
+      "step": 561000
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 2.6029617535775656e-05,
+      "loss": 0.6098,
+      "step": 561200
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 2.6003241602232457e-05,
+      "loss": 0.6085,
+      "step": 561400
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 2.597686566868926e-05,
+      "loss": 0.6095,
+      "step": 561600
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 2.5950489735146066e-05,
+      "loss": 0.6115,
+      "step": 561800
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 2.5924113801602867e-05,
+      "loss": 0.6169,
+      "step": 562000
+    },
+    {
+      "epoch": 0.74,
+      "eval_em_first_line_ratio": 0.524,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4096,
+      "eval_loss": 0.6289793848991394,
+      "eval_runtime": 227.5466,
+      "eval_samples_per_second": 43.947,
+      "eval_steps_per_second": 0.233,
+      "step": 562000
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 2.589773786805967e-05,
+      "loss": 0.6107,
+      "step": 562200
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 2.587136193451647e-05,
+      "loss": 0.6093,
+      "step": 562400
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 2.5844986000973277e-05,
+      "loss": 0.6118,
+      "step": 562600
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 2.5818610067430078e-05,
+      "loss": 0.6106,
+      "step": 562800
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 2.579223413388688e-05,
+      "loss": 0.6147,
+      "step": 563000
+    },
+    {
+      "epoch": 0.74,
+      "eval_em_first_line_ratio": 0.5257,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4113,
+      "eval_loss": 0.628500759601593,
+      "eval_runtime": 231.0996,
+      "eval_samples_per_second": 43.271,
+      "eval_steps_per_second": 0.229,
+      "step": 563000
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 2.576585820034368e-05,
+      "loss": 0.6144,
+      "step": 563200
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 2.573948226680048e-05,
+      "loss": 0.6155,
+      "step": 563400
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 2.5713106333257286e-05,
+      "loss": 0.6078,
+      "step": 563600
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 2.5686730399714087e-05,
+      "loss": 0.6112,
+      "step": 563800
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 2.5660354466170888e-05,
+      "loss": 0.61,
+      "step": 564000
+    },
+    {
+      "epoch": 0.74,
+      "eval_em_first_line_ratio": 0.5247,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4099,
+      "eval_loss": 0.6287592053413391,
+      "eval_runtime": 226.3909,
+      "eval_samples_per_second": 44.171,
+      "eval_steps_per_second": 0.234,
+      "step": 564000
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 2.5634110412295408e-05,
+      "loss": 0.6085,
+      "step": 564200
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 2.560773447875221e-05,
+      "loss": 0.6102,
+      "step": 564400
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 2.558135854520901e-05,
+      "loss": 0.6126,
+      "step": 564600
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 2.555498261166581e-05,
+      "loss": 0.61,
+      "step": 564800
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 2.5528606678122612e-05,
+      "loss": 0.6118,
+      "step": 565000
+    },
+    {
+      "epoch": 0.75,
+      "eval_em_first_line_ratio": 0.5249,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4108,
+      "eval_loss": 0.6282285451889038,
+      "eval_runtime": 227.5625,
+      "eval_samples_per_second": 43.944,
+      "eval_steps_per_second": 0.233,
+      "step": 565000
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 2.550223074457942e-05,
+      "loss": 0.6137,
+      "step": 565200
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 2.5475986690703933e-05,
+      "loss": 0.6077,
+      "step": 565400
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 2.5449610757160734e-05,
+      "loss": 0.6157,
+      "step": 565600
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 2.5423234823617542e-05,
+      "loss": 0.6121,
+      "step": 565800
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 2.5396858890074343e-05,
+      "loss": 0.6062,
+      "step": 566000
+    },
+    {
+      "epoch": 0.75,
+      "eval_em_first_line_ratio": 0.523,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4087,
+      "eval_loss": 0.6283529996871948,
+      "eval_runtime": 224.7322,
+      "eval_samples_per_second": 44.497,
+      "eval_steps_per_second": 0.236,
+      "step": 566000
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 2.5370482956531144e-05,
+      "loss": 0.6035,
+      "step": 566200
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 2.5344107022987945e-05,
+      "loss": 0.6126,
+      "step": 566400
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 2.5317731089444746e-05,
+      "loss": 0.6098,
+      "step": 566600
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 2.5291355155901554e-05,
+      "loss": 0.6144,
+      "step": 566800
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 2.5264979222358355e-05,
+      "loss": 0.6117,
+      "step": 567000
+    },
+    {
+      "epoch": 0.75,
+      "eval_em_first_line_ratio": 0.5249,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4103,
+      "eval_loss": 0.6276270747184753,
+      "eval_runtime": 223.9104,
+      "eval_samples_per_second": 44.661,
+      "eval_steps_per_second": 0.237,
+      "step": 567000
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 2.5238603288815156e-05,
+      "loss": 0.608,
+      "step": 567200
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 2.5212227355271957e-05,
+      "loss": 0.6066,
+      "step": 567400
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 2.5185851421728758e-05,
+      "loss": 0.6131,
+      "step": 567600
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 2.5159475488185562e-05,
+      "loss": 0.6104,
+      "step": 567800
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 2.5133099554642363e-05,
+      "loss": 0.6101,
+      "step": 568000
+    },
+    {
+      "epoch": 0.75,
+      "eval_em_first_line_ratio": 0.5236,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4095,
+      "eval_loss": 0.6282823085784912,
+      "eval_runtime": 222.7759,
+      "eval_samples_per_second": 44.888,
+      "eval_steps_per_second": 0.238,
+      "step": 568000
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 2.5106723621099164e-05,
+      "loss": 0.6045,
+      "step": 568200
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 2.5080347687555965e-05,
+      "loss": 0.6071,
+      "step": 568400
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 2.5053971754012766e-05,
+      "loss": 0.6103,
+      "step": 568600
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 2.5027595820469574e-05,
+      "loss": 0.6081,
+      "step": 568800
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 2.5001219886926375e-05,
+      "loss": 0.6065,
+      "step": 569000
+    },
+    {
+      "epoch": 0.75,
+      "eval_em_first_line_ratio": 0.524,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4096,
+      "eval_loss": 0.6281452178955078,
+      "eval_runtime": 219.078,
+      "eval_samples_per_second": 45.646,
+      "eval_steps_per_second": 0.242,
+      "step": 569000
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 2.4974843953383176e-05,
+      "loss": 0.6135,
+      "step": 569200
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 2.494846801983998e-05,
+      "loss": 0.6131,
+      "step": 569400
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 2.4922223965964497e-05,
+      "loss": 0.6098,
+      "step": 569600
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 2.48958480324213e-05,
+      "loss": 0.606,
+      "step": 569800
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 2.4869472098878103e-05,
+      "loss": 0.6105,
+      "step": 570000
+    },
+    {
+      "epoch": 0.75,
+      "eval_em_first_line_ratio": 0.523,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4089,
+      "eval_loss": 0.628602147102356,
+      "eval_runtime": 223.4462,
+      "eval_samples_per_second": 44.754,
+      "eval_steps_per_second": 0.237,
+      "step": 570000
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 2.4843096165334904e-05,
+      "loss": 0.6071,
+      "step": 570200
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 2.4816720231791705e-05,
+      "loss": 0.6105,
+      "step": 570400
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 2.479034429824851e-05,
+      "loss": 0.6103,
+      "step": 570600
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 2.476396836470531e-05,
+      "loss": 0.6087,
+      "step": 570800
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 2.473759243116211e-05,
+      "loss": 0.6116,
+      "step": 571000
+    },
+    {
+      "epoch": 0.75,
+      "eval_em_first_line_ratio": 0.5245,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4106,
+      "eval_loss": 0.6278237104415894,
+      "eval_runtime": 221.2564,
+      "eval_samples_per_second": 45.196,
+      "eval_steps_per_second": 0.24,
+      "step": 571000
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 2.4711216497618912e-05,
+      "loss": 0.6098,
+      "step": 571200
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 2.4684840564075713e-05,
+      "loss": 0.6059,
+      "step": 571400
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 2.4658464630532518e-05,
+      "loss": 0.6101,
+      "step": 571600
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 2.463208869698932e-05,
+      "loss": 0.6083,
+      "step": 571800
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 2.4605712763446123e-05,
+      "loss": 0.61,
+      "step": 572000
+    },
+    {
+      "epoch": 0.75,
+      "eval_em_first_line_ratio": 0.5231,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4095,
+      "eval_loss": 0.6271325945854187,
+      "eval_runtime": 224.8841,
+      "eval_samples_per_second": 44.467,
+      "eval_steps_per_second": 0.236,
+      "step": 572000
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 2.4579336829902924e-05,
+      "loss": 0.6086,
+      "step": 572200
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 2.455296089635973e-05,
+      "loss": 0.6096,
+      "step": 572400
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 2.452658496281653e-05,
+      "loss": 0.609,
+      "step": 572600
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 2.450020902927333e-05,
+      "loss": 0.6065,
+      "step": 572800
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 2.4473833095730135e-05,
+      "loss": 0.6066,
+      "step": 573000
+    },
+    {
+      "epoch": 0.76,
+      "eval_em_first_line_ratio": 0.5254,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4108,
+      "eval_loss": 0.6275391578674316,
+      "eval_runtime": 219.9207,
+      "eval_samples_per_second": 45.471,
+      "eval_steps_per_second": 0.241,
+      "step": 573000
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 2.4447457162186936e-05,
+      "loss": 0.6074,
+      "step": 573200
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 2.4421081228643737e-05,
+      "loss": 0.6113,
+      "step": 573400
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 2.439470529510054e-05,
+      "loss": 0.609,
+      "step": 573600
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 2.436832936155734e-05,
+      "loss": 0.6123,
+      "step": 573800
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 2.4341953428014144e-05,
+      "loss": 0.6072,
+      "step": 574000
+    },
+    {
+      "epoch": 0.76,
+      "eval_em_first_line_ratio": 0.5237,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4096,
+      "eval_loss": 0.6281174421310425,
+      "eval_runtime": 219.7595,
+      "eval_samples_per_second": 45.504,
+      "eval_steps_per_second": 0.241,
+      "step": 574000
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 2.4315577494470945e-05,
+      "loss": 0.6088,
+      "step": 574200
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 2.428920156092775e-05,
+      "loss": 0.6115,
+      "step": 574400
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 2.4262957507052266e-05,
+      "loss": 0.6093,
+      "step": 574600
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 2.4236581573509067e-05,
+      "loss": 0.6035,
+      "step": 574800
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 2.421020563996587e-05,
+      "loss": 0.6113,
+      "step": 575000
+    },
+    {
+      "epoch": 0.76,
+      "eval_em_first_line_ratio": 0.5243,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.41,
+      "eval_loss": 0.6272754669189453,
+      "eval_runtime": 219.1859,
+      "eval_samples_per_second": 45.623,
+      "eval_steps_per_second": 0.242,
+      "step": 575000
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 2.4183829706422672e-05,
+      "loss": 0.6094,
+      "step": 575200
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 2.4157453772879473e-05,
+      "loss": 0.6137,
+      "step": 575400
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 2.4131077839336278e-05,
+      "loss": 0.6106,
+      "step": 575600
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 2.4104833785460794e-05,
+      "loss": 0.6074,
+      "step": 575800
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 2.4078457851917595e-05,
+      "loss": 0.6069,
+      "step": 576000
+    },
+    {
+      "epoch": 0.76,
+      "eval_em_first_line_ratio": 0.5243,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4096,
+      "eval_loss": 0.6274393796920776,
+      "eval_runtime": 224.4184,
+      "eval_samples_per_second": 44.56,
+      "eval_steps_per_second": 0.236,
+      "step": 576000
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 2.40520819183744e-05,
+      "loss": 0.606,
+      "step": 576200
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 2.40257059848312e-05,
+      "loss": 0.6035,
+      "step": 576400
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 2.3999330051288005e-05,
+      "loss": 0.6111,
+      "step": 576600
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 2.3972954117744806e-05,
+      "loss": 0.6124,
+      "step": 576800
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 2.394657818420161e-05,
+      "loss": 0.6095,
+      "step": 577000
+    },
+    {
+      "epoch": 0.76,
+      "eval_em_first_line_ratio": 0.5229,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4087,
+      "eval_loss": 0.627243161201477,
+      "eval_runtime": 221.5919,
+      "eval_samples_per_second": 45.128,
+      "eval_steps_per_second": 0.239,
+      "step": 577000
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 2.3920334130326127e-05,
+      "loss": 0.6068,
+      "step": 577200
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 2.389395819678293e-05,
+      "loss": 0.6109,
+      "step": 577400
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 2.386758226323973e-05,
+      "loss": 0.6128,
+      "step": 577600
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 2.3841206329696534e-05,
+      "loss": 0.608,
+      "step": 577800
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 2.3814830396153335e-05,
+      "loss": 0.6085,
+      "step": 578000
+    },
+    {
+      "epoch": 0.76,
+      "eval_em_first_line_ratio": 0.5243,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4106,
+      "eval_loss": 0.627017617225647,
+      "eval_runtime": 223.3918,
+      "eval_samples_per_second": 44.764,
+      "eval_steps_per_second": 0.237,
+      "step": 578000
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 2.378845446261014e-05,
+      "loss": 0.6109,
+      "step": 578200
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 2.376207852906694e-05,
+      "loss": 0.6087,
+      "step": 578400
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 2.373570259552374e-05,
+      "loss": 0.6089,
+      "step": 578600
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 2.3709326661980542e-05,
+      "loss": 0.6056,
+      "step": 578800
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 2.3682950728437344e-05,
+      "loss": 0.6116,
+      "step": 579000
+    },
+    {
+      "epoch": 0.76,
+      "eval_em_first_line_ratio": 0.5241,
+      "eval_em_first_line_ratio_wo_empty_matches": 0.4103,
+      "eval_loss": 0.6273497939109802,
+      "eval_runtime": 223.5766,
+      "eval_samples_per_second": 44.727,
+      "eval_steps_per_second": 0.237,
+      "step": 579000
+    }
+  ],
+  "max_steps": 758367,
+  "num_train_epochs": 1,
+  "total_flos": 1.803947371096053e+20,
+  "trial_name": null,
+  "trial_params": null
+}