diff --git "a/checkpoint-4257/trainer_state.json" "b/checkpoint-4257/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/checkpoint-4257/trainer_state.json"
@@ -0,0 +1,4284 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 28.00657894736842,
+  "eval_steps": 500,
+  "global_step": 4257,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.06578947368421052,
+      "grad_norm": 5.228871822357178,
+      "learning_rate": 9e-06,
+      "loss": 3.9986,
+      "memory/device_mem_reserved(gib)": 66.83,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 10
+    },
+    {
+      "epoch": 0.13157894736842105,
+      "grad_norm": 2.1954143047332764,
+      "learning_rate": 1.9e-05,
+      "loss": 0.8823,
+      "memory/device_mem_reserved(gib)": 66.83,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 20
+    },
+    {
+      "epoch": 0.19736842105263158,
+      "grad_norm": 2.786663055419922,
+      "learning_rate": 2.9e-05,
+      "loss": 0.8943,
+      "memory/device_mem_reserved(gib)": 66.83,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 30
+    },
+    {
+      "epoch": 0.2631578947368421,
+      "grad_norm": 1.4439328908920288,
+      "learning_rate": 3.9000000000000006e-05,
+      "loss": 0.6571,
+      "memory/device_mem_reserved(gib)": 66.83,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 40
+    },
+    {
+      "epoch": 0.32894736842105265,
+      "grad_norm": 1.8948026895523071,
+      "learning_rate": 4.9e-05,
+      "loss": 0.7597,
+      "memory/device_mem_reserved(gib)": 66.84,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 50
+    },
+    {
+      "epoch": 0.39473684210526316,
+      "grad_norm": 1.6593940258026123,
+      "learning_rate": 5.9e-05,
+      "loss": 0.7014,
+      "memory/device_mem_reserved(gib)": 66.84,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 60
+    },
+    {
+      "epoch": 0.4605263157894737,
+      "grad_norm": 2.095798969268799,
+      "learning_rate": 6.9e-05,
+      "loss": 0.7159,
+      "memory/device_mem_reserved(gib)": 66.84,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 70
+    },
+    {
+      "epoch": 0.5263157894736842,
+      "grad_norm": 1.5991102457046509,
+      "learning_rate": 7.900000000000001e-05,
+      "loss": 0.6052,
+      "memory/device_mem_reserved(gib)": 66.84,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 80
+    },
+    {
+      "epoch": 0.5921052631578947,
+      "grad_norm": 1.9454739093780518,
+      "learning_rate": 8.900000000000001e-05,
+      "loss": 0.6825,
+      "memory/device_mem_reserved(gib)": 66.84,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 90
+    },
+    {
+      "epoch": 0.6578947368421053,
+      "grad_norm": 2.4410903453826904,
+      "learning_rate": 9.900000000000001e-05,
+      "loss": 0.6352,
+      "memory/device_mem_reserved(gib)": 66.84,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 100
+    },
+    {
+      "epoch": 0.7236842105263158,
+      "grad_norm": 3.553440570831299,
+      "learning_rate": 0.000109,
+      "loss": 0.6316,
+      "memory/device_mem_reserved(gib)": 66.84,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 110
+    },
+    {
+      "epoch": 0.7894736842105263,
+      "grad_norm": 3.3301632404327393,
+      "learning_rate": 0.000119,
+      "loss": 0.6096,
+      "memory/device_mem_reserved(gib)": 66.84,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 120
+    },
+    {
+      "epoch": 0.8552631578947368,
+      "grad_norm": 3.497053384780884,
+      "learning_rate": 0.00012900000000000002,
+      "loss": 0.6476,
+      "memory/device_mem_reserved(gib)": 66.84,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 130
+    },
+    {
+      "epoch": 0.9210526315789473,
+      "grad_norm": 2.0716304779052734,
+      "learning_rate": 0.000139,
+      "loss": 0.6442,
+      "memory/device_mem_reserved(gib)": 66.84,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 140
+    },
+    {
+      "epoch": 0.9868421052631579,
+      "grad_norm": 3.281266450881958,
+      "learning_rate": 0.00014900000000000002,
+      "loss": 0.6593,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 150
+    },
+    {
+      "epoch": 1.0526315789473684,
+      "grad_norm": 3.6694233417510986,
+      "learning_rate": 0.00015900000000000002,
+      "loss": 0.6109,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 160
+    },
+    {
+      "epoch": 1.118421052631579,
+      "grad_norm": 3.227562665939331,
+      "learning_rate": 0.00016900000000000002,
+      "loss": 0.5442,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 170
+    },
+    {
+      "epoch": 1.1842105263157894,
+      "grad_norm": 3.710609197616577,
+      "learning_rate": 0.00017900000000000001,
+      "loss": 0.5785,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 180
+    },
+    {
+      "epoch": 1.25,
+      "grad_norm": 5.2679877281188965,
+      "learning_rate": 0.00018899999999999999,
+      "loss": 0.5875,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 190
+    },
+    {
+      "epoch": 1.3157894736842106,
+      "grad_norm": 3.2948834896087646,
+      "learning_rate": 0.000199,
+      "loss": 0.5709,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 200
+    },
+    {
+      "epoch": 1.381578947368421,
+      "grad_norm": 3.0435378551483154,
+      "learning_rate": 0.00019999830003086864,
+      "loss": 0.612,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 210
+    },
+    {
+      "epoch": 1.4473684210526316,
+      "grad_norm": 5.632095813751221,
+      "learning_rate": 0.00019999242370037052,
+      "loss": 0.5554,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 220
+    },
+    {
+      "epoch": 1.513157894736842,
+      "grad_norm": 2.9409236907958984,
+      "learning_rate": 0.0001999823503735144,
+      "loss": 0.5954,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 230
+    },
+    {
+      "epoch": 1.5789473684210527,
+      "grad_norm": 4.222487926483154,
+      "learning_rate": 0.00019996808065433287,
+      "loss": 0.6102,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 240
+    },
+    {
+      "epoch": 1.6447368421052633,
+      "grad_norm": 3.6207034587860107,
+      "learning_rate": 0.00019994961539848957,
+      "loss": 0.6663,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 250
+    },
+    {
+      "epoch": 1.7105263157894737,
+      "grad_norm": 6.34915828704834,
+      "learning_rate": 0.00019992695571322724,
+      "loss": 0.5956,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 260
+    },
+    {
+      "epoch": 1.776315789473684,
+      "grad_norm": 5.599079608917236,
+      "learning_rate": 0.00019990010295730176,
+      "loss": 0.6437,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 270
+    },
+    {
+      "epoch": 1.8421052631578947,
+      "grad_norm": 3.9812982082366943,
+      "learning_rate": 0.00019986905874090063,
+      "loss": 0.6788,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 280
+    },
+    {
+      "epoch": 1.9078947368421053,
+      "grad_norm": 3.97013783454895,
+      "learning_rate": 0.00019983382492554607,
+      "loss": 0.5463,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 290
+    },
+    {
+      "epoch": 1.973684210526316,
+      "grad_norm": 4.474316120147705,
+      "learning_rate": 0.00019979440362398397,
+      "loss": 0.6562,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 300
+    },
+    {
+      "epoch": 2.039473684210526,
+      "grad_norm": 6.513801574707031,
+      "learning_rate": 0.00019975079720005665,
+      "loss": 0.5757,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 310
+    },
+    {
+      "epoch": 2.1052631578947367,
+      "grad_norm": 4.47428560256958,
+      "learning_rate": 0.00019970300826856145,
+      "loss": 0.4933,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 320
+    },
+    {
+      "epoch": 2.1710526315789473,
+      "grad_norm": 3.6297032833099365,
+      "learning_rate": 0.00019965103969509385,
+      "loss": 0.5623,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 330
+    },
+    {
+      "epoch": 2.236842105263158,
+      "grad_norm": 2.9947564601898193,
+      "learning_rate": 0.00019959489459587558,
+      "loss": 0.5077,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 340
+    },
+    {
+      "epoch": 2.3026315789473686,
+      "grad_norm": 5.7996039390563965,
+      "learning_rate": 0.00019953457633756786,
+      "loss": 0.5758,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 350
+    },
+    {
+      "epoch": 2.3684210526315788,
+      "grad_norm": 7.004776954650879,
+      "learning_rate": 0.0001994700885370694,
+      "loss": 0.5046,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 360
+    },
+    {
+      "epoch": 2.4342105263157894,
+      "grad_norm": 2.668243169784546,
+      "learning_rate": 0.00019940143506129974,
+      "loss": 0.5202,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 370
+    },
+    {
+      "epoch": 2.5,
+      "grad_norm": 5.5414252281188965,
+      "learning_rate": 0.00019932862002696702,
+      "loss": 0.5763,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 380
+    },
+    {
+      "epoch": 2.5657894736842106,
+      "grad_norm": 4.762598514556885,
+      "learning_rate": 0.00019925164780032145,
+      "loss": 0.5329,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 390
+    },
+    {
+      "epoch": 2.6315789473684212,
+      "grad_norm": 5.675711154937744,
+      "learning_rate": 0.00019917052299689344,
+      "loss": 0.5617,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 400
+    },
+    {
+      "epoch": 2.6973684210526314,
+      "grad_norm": 5.017973899841309,
+      "learning_rate": 0.00019908525048121666,
+      "loss": 0.5775,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 410
+    },
+    {
+      "epoch": 2.763157894736842,
+      "grad_norm": 4.787014961242676,
+      "learning_rate": 0.00019899583536653648,
+      "loss": 0.5534,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 420
+    },
+    {
+      "epoch": 2.8289473684210527,
+      "grad_norm": 4.4121856689453125,
+      "learning_rate": 0.0001989022830145034,
+      "loss": 0.5598,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 430
+    },
+    {
+      "epoch": 2.8947368421052633,
+      "grad_norm": 5.0906982421875,
+      "learning_rate": 0.0001988045990348514,
+      "loss": 0.5637,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 440
+    },
+    {
+      "epoch": 2.9605263157894735,
+      "grad_norm": 5.442751884460449,
+      "learning_rate": 0.0001987027892850617,
+      "loss": 0.5529,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 450
+    },
+    {
+      "epoch": 3.026315789473684,
+      "grad_norm": 3.2587528228759766,
+      "learning_rate": 0.0001985968598700115,
+      "loss": 0.5539,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 460
+    },
+    {
+      "epoch": 3.0921052631578947,
+      "grad_norm": 5.267582893371582,
+      "learning_rate": 0.0001984868171416078,
+      "loss": 0.4498,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 470
+    },
+    {
+      "epoch": 3.1578947368421053,
+      "grad_norm": 6.26557731628418,
+      "learning_rate": 0.0001983726676984067,
+      "loss": 0.5543,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 480
+    },
+    {
+      "epoch": 3.223684210526316,
+      "grad_norm": 7.339848518371582,
+      "learning_rate": 0.00019825441838521753,
+      "loss": 0.5068,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 490
+    },
+    {
+      "epoch": 3.2894736842105265,
+      "grad_norm": 7.684560298919678,
+      "learning_rate": 0.0001981320762926925,
+      "loss": 0.4447,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 500
+    },
+    {
+      "epoch": 3.3552631578947367,
+      "grad_norm": 5.45398473739624,
+      "learning_rate": 0.00019800564875690167,
+      "loss": 0.5575,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 510
+    },
+    {
+      "epoch": 3.4210526315789473,
+      "grad_norm": 8.519457817077637,
+      "learning_rate": 0.0001978751433588927,
+      "loss": 0.4763,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 520
+    },
+    {
+      "epoch": 3.486842105263158,
+      "grad_norm": 6.082020282745361,
+      "learning_rate": 0.00019774056792423665,
+      "loss": 0.4817,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 530
+    },
+    {
+      "epoch": 3.5526315789473686,
+      "grad_norm": 5.932907581329346,
+      "learning_rate": 0.0001976019305225584,
+      "loss": 0.5368,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 540
+    },
+    {
+      "epoch": 3.6184210526315788,
+      "grad_norm": 4.9030561447143555,
+      "learning_rate": 0.0001974592394670531,
+      "loss": 0.5254,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 550
+    },
+    {
+      "epoch": 3.6842105263157894,
+      "grad_norm": 8.374964714050293,
+      "learning_rate": 0.00019731250331398733,
+      "loss": 0.5555,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 560
+    },
+    {
+      "epoch": 3.75,
+      "grad_norm": 10.392874717712402,
+      "learning_rate": 0.00019716173086218626,
+      "loss": 0.5351,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 570
+    },
+    {
+      "epoch": 3.8157894736842106,
+      "grad_norm": 5.0130815505981445,
+      "learning_rate": 0.00019700693115250607,
+      "loss": 0.5059,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 580
+    },
+    {
+      "epoch": 3.8815789473684212,
+      "grad_norm": 4.398721694946289,
+      "learning_rate": 0.00019684811346729156,
+      "loss": 0.5064,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 590
+    },
+    {
+      "epoch": 3.9473684210526314,
+      "grad_norm": 5.850337505340576,
+      "learning_rate": 0.0001966852873298199,
+      "loss": 0.5199,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 600
+    },
+    {
+      "epoch": 4.0131578947368425,
+      "grad_norm": 5.074413776397705,
+      "learning_rate": 0.00019651846250372937,
+      "loss": 0.4722,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 610
+    },
+    {
+      "epoch": 4.078947368421052,
+      "grad_norm": 8.62219524383545,
+      "learning_rate": 0.00019634764899243389,
+      "loss": 0.4188,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 620
+    },
+    {
+      "epoch": 4.144736842105263,
+      "grad_norm": 5.095092296600342,
+      "learning_rate": 0.00019617285703852323,
+      "loss": 0.4477,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 630
+    },
+    {
+      "epoch": 4.2105263157894735,
+      "grad_norm": 4.571897506713867,
+      "learning_rate": 0.0001959940971231489,
+      "loss": 0.4309,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 640
+    },
+    {
+      "epoch": 4.276315789473684,
+      "grad_norm": 6.1334075927734375,
+      "learning_rate": 0.00019581137996539552,
+      "loss": 0.4363,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 650
+    },
+    {
+      "epoch": 4.342105263157895,
+      "grad_norm": 5.199427127838135,
+      "learning_rate": 0.00019562471652163815,
+      "loss": 0.4562,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 660
+    },
+    {
+      "epoch": 4.407894736842105,
+      "grad_norm": 5.7770676612854,
+      "learning_rate": 0.00019543411798488532,
+      "loss": 0.4621,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 670
+    },
+    {
+      "epoch": 4.473684210526316,
+      "grad_norm": 5.296002388000488,
+      "learning_rate": 0.00019523959578410772,
+      "loss": 0.4868,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 680
+    },
+    {
+      "epoch": 4.5394736842105265,
+      "grad_norm": 7.707890510559082,
+      "learning_rate": 0.00019504116158355314,
+      "loss": 0.4507,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 690
+    },
+    {
+      "epoch": 4.605263157894737,
+      "grad_norm": 5.631321907043457,
+      "learning_rate": 0.00019483882728204682,
+      "loss": 0.5316,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 700
+    },
+    {
+      "epoch": 4.671052631578947,
+      "grad_norm": 6.96210241317749,
+      "learning_rate": 0.000194632605012278,
+      "loss": 0.4924,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 710
+    },
+    {
+      "epoch": 4.7368421052631575,
+      "grad_norm": 6.5153937339782715,
+      "learning_rate": 0.00019442250714007234,
+      "loss": 0.4652,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 720
+    },
+    {
+      "epoch": 4.802631578947368,
+      "grad_norm": 7.671054363250732,
+      "learning_rate": 0.00019420854626365065,
+      "loss": 0.5399,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 730
+    },
+    {
+      "epoch": 4.868421052631579,
+      "grad_norm": 6.234044075012207,
+      "learning_rate": 0.00019399073521287333,
+      "loss": 0.5311,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 740
+    },
+    {
+      "epoch": 4.934210526315789,
+      "grad_norm": 4.996755599975586,
+      "learning_rate": 0.00019376908704847083,
+      "loss": 0.5329,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 750
+    },
+    {
+      "epoch": 5.0,
+      "grad_norm": 9.81963062286377,
+      "learning_rate": 0.0001935436150612608,
+      "loss": 0.4912,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 760
+    },
+    {
+      "epoch": 5.065789473684211,
+      "grad_norm": 5.93031644821167,
+      "learning_rate": 0.00019331433277135096,
+      "loss": 0.4343,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 770
+    },
+    {
+      "epoch": 5.131578947368421,
+      "grad_norm": 7.9363837242126465,
+      "learning_rate": 0.00019308125392732847,
+      "loss": 0.4316,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 780
+    },
+    {
+      "epoch": 5.197368421052632,
+      "grad_norm": 5.575252532958984,
+      "learning_rate": 0.00019284439250543542,
+      "loss": 0.4041,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 790
+    },
+    {
+      "epoch": 5.2631578947368425,
+      "grad_norm": 6.127432346343994,
+      "learning_rate": 0.00019260376270873087,
+      "loss": 0.4205,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 800
+    },
+    {
+      "epoch": 5.328947368421053,
+      "grad_norm": 8.333614349365234,
+      "learning_rate": 0.00019235937896623905,
+      "loss": 0.4604,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 810
+    },
+    {
+      "epoch": 5.394736842105263,
+      "grad_norm": 6.921279430389404,
+      "learning_rate": 0.0001921112559320843,
+      "loss": 0.416,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 820
+    },
+    {
+      "epoch": 5.4605263157894735,
+      "grad_norm": 8.031194686889648,
+      "learning_rate": 0.00019185940848461225,
+      "loss": 0.4593,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 830
+    },
+    {
+      "epoch": 5.526315789473684,
+      "grad_norm": 5.994675636291504,
+      "learning_rate": 0.00019160385172549757,
+      "loss": 0.5126,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 840
+    },
+    {
+      "epoch": 5.592105263157895,
+      "grad_norm": 8.621824264526367,
+      "learning_rate": 0.0001913446009788388,
+      "loss": 0.4761,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 850
+    },
+    {
+      "epoch": 5.657894736842105,
+      "grad_norm": 7.873433589935303,
+      "learning_rate": 0.00019108167179023893,
+      "loss": 0.4399,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 860
+    },
+    {
+      "epoch": 5.723684210526316,
+      "grad_norm": 7.9271650314331055,
+      "learning_rate": 0.0001908150799258737,
+      "loss": 0.4469,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 870
+    },
+    {
+      "epoch": 5.7894736842105265,
+      "grad_norm": 7.980473041534424,
+      "learning_rate": 0.0001905448413715459,
+      "loss": 0.4781,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 880
+    },
+    {
+      "epoch": 5.855263157894737,
+      "grad_norm": 5.033918380737305,
+      "learning_rate": 0.00019027097233172694,
+      "loss": 0.5237,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 890
+    },
+    {
+      "epoch": 5.921052631578947,
+      "grad_norm": 5.501688003540039,
+      "learning_rate": 0.00018999348922858514,
+      "loss": 0.4594,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 900
+    },
+    {
+      "epoch": 5.9868421052631575,
+      "grad_norm": 5.770474910736084,
+      "learning_rate": 0.00018971240870100094,
+      "loss": 0.4845,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 910
+    },
+    {
+      "epoch": 6.052631578947368,
+      "grad_norm": 6.346399784088135,
+      "learning_rate": 0.00018942774760356934,
+      "loss": 0.4128,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 920
+    },
+    {
+      "epoch": 6.118421052631579,
+      "grad_norm": 6.847306728363037,
+      "learning_rate": 0.00018913952300558907,
+      "loss": 0.3971,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 930
+    },
+    {
+      "epoch": 6.184210526315789,
+      "grad_norm": 4.9199347496032715,
+      "learning_rate": 0.00018884775219003909,
+      "loss": 0.3764,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 940
+    },
+    {
+      "epoch": 6.25,
+      "grad_norm": 8.841605186462402,
+      "learning_rate": 0.00018855245265254227,
+      "loss": 0.4349,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 950
+    },
+    {
+      "epoch": 6.315789473684211,
+      "grad_norm": 7.917416095733643,
+      "learning_rate": 0.00018825364210031635,
+      "loss": 0.4043,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 960
+    },
+    {
+      "epoch": 6.381578947368421,
+      "grad_norm": 9.34174919128418,
+      "learning_rate": 0.00018795133845111206,
+      "loss": 0.3961,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 970
+    },
+    {
+      "epoch": 6.447368421052632,
+      "grad_norm": 7.267455101013184,
+      "learning_rate": 0.00018764555983213864,
+      "loss": 0.436,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 980
+    },
+    {
+      "epoch": 6.5131578947368425,
+      "grad_norm": 8.763919830322266,
+      "learning_rate": 0.00018733632457897708,
+      "loss": 0.4639,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 990
+    },
+    {
+      "epoch": 6.578947368421053,
+      "grad_norm": 6.768899917602539,
+      "learning_rate": 0.00018702365123448046,
+      "loss": 0.4811,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1000
+    },
+    {
+      "epoch": 6.644736842105263,
+      "grad_norm": 6.540329456329346,
+      "learning_rate": 0.0001867075585476622,
+      "loss": 0.4476,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1010
+    },
+    {
+      "epoch": 6.7105263157894735,
+      "grad_norm": 5.938429355621338,
+      "learning_rate": 0.00018638806547257168,
+      "loss": 0.4052,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1020
+    },
+    {
+      "epoch": 6.776315789473684,
+      "grad_norm": 6.720884799957275,
+      "learning_rate": 0.00018606519116715772,
+      "loss": 0.4525,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1030
+    },
+    {
+      "epoch": 6.842105263157895,
+      "grad_norm": 8.531013488769531,
+      "learning_rate": 0.0001857389549921198,
+      "loss": 0.4441,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1040
+    },
+    {
+      "epoch": 6.907894736842105,
+      "grad_norm": 6.624144077301025,
+      "learning_rate": 0.00018540937650974728,
+      "loss": 0.4791,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1050
+    },
+    {
+      "epoch": 6.973684210526316,
+      "grad_norm": 8.459912300109863,
+      "learning_rate": 0.000185076475482746,
+      "loss": 0.4243,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1060
+    },
+    {
+      "epoch": 7.0394736842105265,
+      "grad_norm": 8.084807395935059,
+      "learning_rate": 0.00018474027187305377,
+      "loss": 0.4047,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1070
+    },
+    {
+      "epoch": 7.105263157894737,
+      "grad_norm": 6.393621921539307,
+      "learning_rate": 0.0001844007858406428,
+      "loss": 0.377,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1080
+    },
+    {
+      "epoch": 7.171052631578948,
+      "grad_norm": 9.367650985717773,
+      "learning_rate": 0.0001840580377423113,
+      "loss": 0.3671,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1090
+    },
+    {
+      "epoch": 7.2368421052631575,
+      "grad_norm": 7.191049098968506,
+      "learning_rate": 0.00018371204813046258,
+      "loss": 0.3986,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1100
+    },
+    {
+      "epoch": 7.302631578947368,
+      "grad_norm": 5.127690315246582,
+      "learning_rate": 0.00018336283775187266,
+      "loss": 0.3936,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1110
+    },
+    {
+      "epoch": 7.368421052631579,
+      "grad_norm": 8.498778343200684,
+      "learning_rate": 0.00018301042754644638,
+      "loss": 0.3985,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1120
+    },
+    {
+      "epoch": 7.434210526315789,
+      "grad_norm": 7.945418357849121,
+      "learning_rate": 0.00018265483864596154,
+      "loss": 0.4012,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1130
+    },
+    {
+      "epoch": 7.5,
+      "grad_norm": 9.021563529968262,
+      "learning_rate": 0.00018229609237280196,
+      "loss": 0.4443,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1140
+    },
+    {
+      "epoch": 7.565789473684211,
+      "grad_norm": 14.01905632019043,
+      "learning_rate": 0.00018193421023867878,
+      "loss": 0.4273,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1150
+    },
+    {
+      "epoch": 7.631578947368421,
+      "grad_norm": 7.097718238830566,
+      "learning_rate": 0.0001815692139433406,
+      "loss": 0.3867,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1160
+    },
+    {
+      "epoch": 7.697368421052632,
+      "grad_norm": 6.583855628967285,
+      "learning_rate": 0.00018120112537327234,
+      "loss": 0.4731,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1170
+    },
+    {
+      "epoch": 7.7631578947368425,
+      "grad_norm": 8.54875373840332,
+      "learning_rate": 0.00018082996660038266,
+      "loss": 0.378,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1180
+    },
+    {
+      "epoch": 7.828947368421053,
+      "grad_norm": 7.399291515350342,
+      "learning_rate": 0.00018045575988068072,
+      "loss": 0.4387,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1190
+    },
+    {
+      "epoch": 7.894736842105263,
+      "grad_norm": 9.295549392700195,
+      "learning_rate": 0.00018007852765294135,
+      "loss": 0.4581,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1200
+    },
+    {
+      "epoch": 7.9605263157894735,
+      "grad_norm": 9.612926483154297,
+      "learning_rate": 0.00017969829253735976,
+      "loss": 0.4128,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1210
+    },
+    {
+      "epoch": 8.026315789473685,
+      "grad_norm": 7.819753170013428,
+      "learning_rate": 0.00017931507733419508,
+      "loss": 0.4026,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1220
+    },
+    {
+      "epoch": 8.092105263157896,
+      "grad_norm": 6.5006632804870605,
+      "learning_rate": 0.0001789289050224031,
+      "loss": 0.3644,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1230
+    },
+    {
+      "epoch": 8.157894736842104,
+      "grad_norm": 7.953379154205322,
+      "learning_rate": 0.00017853979875825848,
+      "loss": 0.3859,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1240
+    },
+    {
+      "epoch": 8.223684210526315,
+      "grad_norm": 5.973986625671387,
+      "learning_rate": 0.00017814778187396613,
+      "loss": 0.4247,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1250
+    },
+    {
+      "epoch": 8.289473684210526,
+      "grad_norm": 8.134764671325684,
+      "learning_rate": 0.00017775287787626223,
+      "loss": 0.3536,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1260
+    },
+    {
+      "epoch": 8.355263157894736,
+      "grad_norm": 7.2093963623046875,
+      "learning_rate": 0.00017735511044500454,
+      "loss": 0.3697,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1270
+    },
+    {
+      "epoch": 8.421052631578947,
+      "grad_norm": 5.207829475402832,
+      "learning_rate": 0.0001769545034317526,
+      "loss": 0.376,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1280
+    },
+    {
+      "epoch": 8.486842105263158,
+      "grad_norm": 6.704354763031006,
+      "learning_rate": 0.0001765510808583375,
+      "loss": 0.3702,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1290
+    },
+    {
+      "epoch": 8.552631578947368,
+      "grad_norm": 7.463472366333008,
+      "learning_rate": 0.00017614486691542122,
+      "loss": 0.4381,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1300
+    },
+    {
+      "epoch": 8.618421052631579,
+      "grad_norm": 9.245990753173828,
+      "learning_rate": 0.00017573588596104647,
+      "loss": 0.4001,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1310
+    },
+    {
+      "epoch": 8.68421052631579,
+      "grad_norm": 8.084155082702637,
+      "learning_rate": 0.00017532416251917572,
+      "loss": 0.437,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1320
+    },
+    {
+      "epoch": 8.75,
+      "grad_norm": 8.936637878417969,
+      "learning_rate": 0.00017490972127822088,
+      "loss": 0.4072,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1330
+    },
+    {
+      "epoch": 8.81578947368421,
+      "grad_norm": 8.653999328613281,
+      "learning_rate": 0.00017449258708956283,
+      "loss": 0.3979,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1340
+    },
+    {
+      "epoch": 8.881578947368421,
+      "grad_norm": 6.477110862731934,
+      "learning_rate": 0.00017407278496606122,
+      "loss": 0.4,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1350
+    },
+    {
+      "epoch": 8.947368421052632,
+      "grad_norm": 8.848350524902344,
+      "learning_rate": 0.00017365034008055467,
+      "loss": 0.3951,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1360
+    },
+    {
+      "epoch": 9.013157894736842,
+      "grad_norm": 3.7075564861297607,
+      "learning_rate": 0.00017322527776435115,
+      "loss": 0.3813,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1370
+    },
+    {
+      "epoch": 9.078947368421053,
+      "grad_norm": 7.235719680786133,
+      "learning_rate": 0.00017279762350570925,
+      "loss": 0.3373,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1380
+    },
+    {
+      "epoch": 9.144736842105264,
+      "grad_norm": 11.8607177734375,
+      "learning_rate": 0.00017236740294830969,
+      "loss": 0.3858,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1390
+    },
+    {
+      "epoch": 9.210526315789474,
+      "grad_norm": 9.025496482849121,
+      "learning_rate": 0.00017193464188971767,
+      "loss": 0.3341,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1400
+    },
+    {
+      "epoch": 9.276315789473685,
+      "grad_norm": 6.856828212738037,
+      "learning_rate": 0.00017149936627983595,
+      "loss": 0.3887,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1410
+    },
+    {
+      "epoch": 9.342105263157896,
+      "grad_norm": 10.805776596069336,
+      "learning_rate": 0.0001710616022193487,
+      "loss": 0.3375,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1420
+    },
+    {
+      "epoch": 9.407894736842104,
+      "grad_norm": 6.890063285827637,
+      "learning_rate": 0.00017062137595815657,
+      "loss": 0.3546,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1430
+    },
+    {
+      "epoch": 9.473684210526315,
+      "grad_norm": 9.32254695892334,
+      "learning_rate": 0.00017017871389380255,
+      "loss": 0.3875,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1440
+    },
+    {
+      "epoch": 9.539473684210526,
+      "grad_norm": 8.32424545288086,
+      "learning_rate": 0.00016973364256988918,
+      "loss": 0.416,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1450
+    },
+    {
+      "epoch": 9.605263157894736,
+      "grad_norm": 6.700080871582031,
+      "learning_rate": 0.00016928618867448675,
+      "loss": 0.3982,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1460
+    },
+    {
+      "epoch": 9.671052631578947,
+      "grad_norm": 9.059173583984375,
+      "learning_rate": 0.0001688363790385331,
+      "loss": 0.3322,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1470
+    },
+    {
+      "epoch": 9.736842105263158,
+      "grad_norm": 7.265253067016602,
+      "learning_rate": 0.0001683842406342247,
+      "loss": 0.3597,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1480
+    },
+    {
+      "epoch": 9.802631578947368,
+      "grad_norm": 9.619430541992188,
+      "learning_rate": 0.00016792980057339936,
+      "loss": 0.3976,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1490
+    },
+    {
+      "epoch": 9.868421052631579,
+      "grad_norm": 7.571810245513916,
+      "learning_rate": 0.00016747308610591034,
+      "loss": 0.3728,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1500
+    },
+    {
+      "epoch": 9.93421052631579,
+      "grad_norm": 8.444099426269531,
+      "learning_rate": 0.00016701412461799254,
+      "loss": 0.3842,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1510
+    },
+    {
+      "epoch": 10.0,
+      "grad_norm": 6.8055291175842285,
+      "learning_rate": 0.00016655294363062035,
+      "loss": 0.3844,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1520
+    },
+    {
+      "epoch": 10.06578947368421,
+      "grad_norm": 9.351649284362793,
+      "learning_rate": 0.00016608957079785713,
+      "loss": 0.3382,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1530
+    },
+    {
+      "epoch": 10.131578947368421,
+      "grad_norm": 9.218072891235352,
+      "learning_rate": 0.00016562403390519724,
+      "loss": 0.3175,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1540
+    },
+    {
+      "epoch": 10.197368421052632,
+      "grad_norm": 8.207752227783203,
+      "learning_rate": 0.0001651563608678997,
+      "loss": 0.3422,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1550
+    },
+    {
+      "epoch": 10.263157894736842,
+      "grad_norm": 7.536536693572998,
+      "learning_rate": 0.0001646865797293146,
+      "loss": 0.3548,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1560
+    },
+    {
+      "epoch": 10.328947368421053,
+      "grad_norm": 6.550157070159912,
+      "learning_rate": 0.00016421471865920112,
+      "loss": 0.3501,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1570
+    },
+    {
+      "epoch": 10.394736842105264,
+      "grad_norm": 7.151719570159912,
+      "learning_rate": 0.00016374080595203883,
+      "loss": 0.3693,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1580
+    },
+    {
+      "epoch": 10.460526315789474,
+      "grad_norm": 7.3648810386657715,
+      "learning_rate": 0.00016326487002533058,
+      "loss": 0.3312,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1590
+    },
+    {
+      "epoch": 10.526315789473685,
+      "grad_norm": 7.90741491317749,
+      "learning_rate": 0.00016278693941789877,
+      "loss": 0.3586,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1600
+    },
+    {
+      "epoch": 10.592105263157894,
+      "grad_norm": 6.969632148742676,
+      "learning_rate": 0.00016230704278817406,
+      "loss": 0.3492,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1610
+    },
+    {
+      "epoch": 10.657894736842106,
+      "grad_norm": 9.677424430847168,
+      "learning_rate": 0.00016182520891247685,
+      "loss": 0.3514,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1620
+    },
+    {
+      "epoch": 10.723684210526315,
+      "grad_norm": 7.527509689331055,
+      "learning_rate": 0.00016134146668329166,
+      "loss": 0.3368,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1630
+    },
+    {
+      "epoch": 10.789473684210526,
+      "grad_norm": 7.857253074645996,
+      "learning_rate": 0.00016085584510753474,
+      "loss": 0.3755,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1640
+    },
+    {
+      "epoch": 10.855263157894736,
+      "grad_norm": 5.162316799163818,
+      "learning_rate": 0.00016036837330481477,
+      "loss": 0.3805,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1650
+    },
+    {
+      "epoch": 10.921052631578947,
+      "grad_norm": 7.910579204559326,
+      "learning_rate": 0.0001598790805056866,
+      "loss": 0.3351,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1660
+    },
+    {
+      "epoch": 10.986842105263158,
+      "grad_norm": 7.014622688293457,
+      "learning_rate": 0.00015938799604989852,
+      "loss": 0.3842,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1670
+    },
+    {
+      "epoch": 11.052631578947368,
+      "grad_norm": 9.553169250488281,
+      "learning_rate": 0.00015889514938463304,
+      "loss": 0.308,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1680
+    },
+    {
+      "epoch": 11.118421052631579,
+      "grad_norm": 10.32496452331543,
+      "learning_rate": 0.000158400570062741,
+      "loss": 0.3324,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1690
+    },
+    {
+      "epoch": 11.18421052631579,
+      "grad_norm": 13.280173301696777,
+      "learning_rate": 0.00015790428774096953,
+      "loss": 0.3678,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1700
+    },
+    {
+      "epoch": 11.25,
+      "grad_norm": 10.031278610229492,
+      "learning_rate": 0.00015740633217818382,
+      "loss": 0.3176,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1710
+    },
+    {
+      "epoch": 11.31578947368421,
+      "grad_norm": 6.93234920501709,
+      "learning_rate": 0.00015690673323358253,
+      "loss": 0.3413,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1720
+    },
+    {
+      "epoch": 11.381578947368421,
+      "grad_norm": 8.427581787109375,
+      "learning_rate": 0.0001564055208649073,
+      "loss": 0.3147,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1730
+    },
+    {
+      "epoch": 11.447368421052632,
+      "grad_norm": 7.225602626800537,
+      "learning_rate": 0.00015590272512664655,
+      "loss": 0.2998,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1740
+    },
+    {
+      "epoch": 11.513157894736842,
+      "grad_norm": 8.522515296936035,
+      "learning_rate": 0.00015539837616823326,
+      "loss": 0.2984,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1750
+    },
+    {
+      "epoch": 11.578947368421053,
+      "grad_norm": 11.115621566772461,
+      "learning_rate": 0.0001548925042322369,
+      "loss": 0.3423,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1760
+    },
+    {
+      "epoch": 11.644736842105264,
+      "grad_norm": 6.074847221374512,
+      "learning_rate": 0.00015438513965255023,
+      "loss": 0.3423,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1770
+    },
+    {
+      "epoch": 11.710526315789474,
+      "grad_norm": 9.569510459899902,
+      "learning_rate": 0.00015387631285257021,
+      "loss": 0.3416,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1780
+    },
+    {
+      "epoch": 11.776315789473685,
+      "grad_norm": 8.04273509979248,
+      "learning_rate": 0.00015336605434337387,
+      "loss": 0.3376,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1790
+    },
+    {
+      "epoch": 11.842105263157894,
+      "grad_norm": 9.288858413696289,
+      "learning_rate": 0.0001528543947218886,
+      "loss": 0.3668,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1800
+    },
+    {
+      "epoch": 11.907894736842106,
+      "grad_norm": 9.00793743133545,
+      "learning_rate": 0.0001523413646690574,
+      "loss": 0.322,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1810
+    },
+    {
+      "epoch": 11.973684210526315,
+      "grad_norm": 6.323388576507568,
+      "learning_rate": 0.0001518269949479994,
+      "loss": 0.3419,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1820
+    },
+    {
+      "epoch": 12.039473684210526,
+      "grad_norm": 7.900981903076172,
+      "learning_rate": 0.0001513113164021649,
+      "loss": 0.3268,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1830
+    },
+    {
+      "epoch": 12.105263157894736,
+      "grad_norm": 8.535825729370117,
+      "learning_rate": 0.00015079435995348615,
+      "loss": 0.2922,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1840
+    },
+    {
+      "epoch": 12.171052631578947,
+      "grad_norm": 10.261241912841797,
+      "learning_rate": 0.0001502761566005229,
+      "loss": 0.316,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1850
+    },
+    {
+      "epoch": 12.236842105263158,
+      "grad_norm": 9.959452629089355,
+      "learning_rate": 0.0001497567374166038,
+      "loss": 0.2672,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1860
+    },
+    {
+      "epoch": 12.302631578947368,
+      "grad_norm": 8.756327629089355,
+      "learning_rate": 0.00014923613354796313,
+      "loss": 0.2825,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1870
+    },
+    {
+      "epoch": 12.368421052631579,
+      "grad_norm": 11.042543411254883,
+      "learning_rate": 0.00014871437621187306,
+      "loss": 0.348,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1880
+    },
+    {
+      "epoch": 12.43421052631579,
+      "grad_norm": 8.125487327575684,
+      "learning_rate": 0.00014819149669477178,
+      "loss": 0.2827,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1890
+    },
+    {
+      "epoch": 12.5,
+      "grad_norm": 6.272602081298828,
+      "learning_rate": 0.0001476675263503875,
+      "loss": 0.331,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1900
+    },
+    {
+      "epoch": 12.56578947368421,
+      "grad_norm": 9.739338874816895,
+      "learning_rate": 0.00014714249659785833,
+      "loss": 0.3421,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1910
+    },
+    {
+      "epoch": 12.631578947368421,
+      "grad_norm": 10.085533142089844,
+      "learning_rate": 0.0001466164389198483,
+      "loss": 0.3315,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1920
+    },
+    {
+      "epoch": 12.697368421052632,
+      "grad_norm": 10.818131446838379,
+      "learning_rate": 0.00014608938486065953,
+      "loss": 0.3479,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1930
+    },
+    {
+      "epoch": 12.763157894736842,
+      "grad_norm": 9.922698974609375,
+      "learning_rate": 0.00014556136602434064,
+      "loss": 0.3279,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1940
+    },
+    {
+      "epoch": 12.828947368421053,
+      "grad_norm": 6.283480167388916,
+      "learning_rate": 0.00014503241407279194,
+      "loss": 0.3254,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1950
+    },
+    {
+      "epoch": 12.894736842105264,
+      "grad_norm": 6.850347995758057,
+      "learning_rate": 0.00014450256072386645,
+      "loss": 0.3053,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1960
+    },
+    {
+      "epoch": 12.960526315789474,
+      "grad_norm": 7.839858055114746,
+      "learning_rate": 0.00014397183774946833,
+      "loss": 0.296,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1970
+    },
+    {
+      "epoch": 13.026315789473685,
+      "grad_norm": 8.421372413635254,
+      "learning_rate": 0.0001434402769736476,
+      "loss": 0.3125,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1980
+    },
+    {
+      "epoch": 13.092105263157896,
+      "grad_norm": 8.362040519714355,
+      "learning_rate": 0.00014290791027069176,
+      "loss": 0.2749,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 1990
+    },
+    {
+      "epoch": 13.157894736842104,
+      "grad_norm": 10.712233543395996,
+      "learning_rate": 0.00014237476956321468,
+      "loss": 0.2971,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2000
+    },
+    {
+      "epoch": 13.223684210526315,
+      "grad_norm": 7.288019180297852,
+      "learning_rate": 0.00014184088682024233,
+      "loss": 0.3226,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2010
+    },
+    {
+      "epoch": 13.289473684210526,
+      "grad_norm": 12.341687202453613,
+      "learning_rate": 0.00014130629405529566,
+      "loss": 0.2857,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2020
+    },
+    {
+      "epoch": 13.355263157894736,
+      "grad_norm": 8.097869873046875,
+      "learning_rate": 0.00014077102332447122,
+      "loss": 0.2739,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2030
+    },
+    {
+      "epoch": 13.421052631578947,
+      "grad_norm": 10.796974182128906,
+      "learning_rate": 0.0001402351067245187,
+      "loss": 0.2883,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2040
+    },
+    {
+      "epoch": 13.486842105263158,
+      "grad_norm": 10.561753273010254,
+      "learning_rate": 0.00013969857639091653,
+      "loss": 0.2911,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2050
+    },
+    {
+      "epoch": 13.552631578947368,
+      "grad_norm": 7.319301128387451,
+      "learning_rate": 0.00013916146449594472,
+      "loss": 0.322,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2060
+    },
+    {
+      "epoch": 13.618421052631579,
+      "grad_norm": 6.3976850509643555,
+      "learning_rate": 0.0001386238032467558,
+      "loss": 0.2861,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2070
+    },
+    {
+      "epoch": 13.68421052631579,
+      "grad_norm": 8.518961906433105,
+      "learning_rate": 0.00013808562488344363,
+      "loss": 0.293,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2080
+    },
+    {
+      "epoch": 13.75,
+      "grad_norm": 9.818099975585938,
+      "learning_rate": 0.00013754696167710993,
+      "loss": 0.3276,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2090
+    },
+    {
+      "epoch": 13.81578947368421,
+      "grad_norm": 10.728662490844727,
+      "learning_rate": 0.00013700784592792948,
+      "loss": 0.3067,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2100
+    },
+    {
+      "epoch": 13.881578947368421,
+      "grad_norm": 9.833274841308594,
+      "learning_rate": 0.0001364683099632131,
+      "loss": 0.2844,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2110
+    },
+    {
+      "epoch": 13.947368421052632,
+      "grad_norm": 9.362689971923828,
+      "learning_rate": 0.0001359283861354692,
+      "loss": 0.2611,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2120
+    },
+    {
+      "epoch": 14.013157894736842,
+      "grad_norm": 7.0610857009887695,
+      "learning_rate": 0.0001353881068204639,
+      "loss": 0.2892,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2130
+    },
+    {
+      "epoch": 14.078947368421053,
+      "grad_norm": 6.5400495529174805,
+      "learning_rate": 0.00013484750441527957,
+      "loss": 0.2665,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2140
+    },
+    {
+      "epoch": 14.144736842105264,
+      "grad_norm": 7.0163469314575195,
+      "learning_rate": 0.0001343066113363723,
+      "loss": 0.2517,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2150
+    },
+    {
+      "epoch": 14.210526315789474,
+      "grad_norm": 6.757364273071289,
+      "learning_rate": 0.00013376546001762793,
+      "loss": 0.2383,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2160
+    },
+    {
+      "epoch": 14.276315789473685,
+      "grad_norm": 8.389124870300293,
+      "learning_rate": 0.00013322408290841734,
+      "loss": 0.2944,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2170
+    },
+    {
+      "epoch": 14.342105263157896,
+      "grad_norm": 11.034085273742676,
+      "learning_rate": 0.00013268251247165055,
+      "loss": 0.2619,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2180
+    },
+    {
+      "epoch": 14.407894736842104,
+      "grad_norm": 8.859076499938965,
+      "learning_rate": 0.00013214078118183031,
+      "loss": 0.2655,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2190
+    },
+    {
+      "epoch": 14.473684210526315,
+      "grad_norm": 7.588623046875,
+      "learning_rate": 0.0001315989215231045,
+      "loss": 0.2493,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2200
+    },
+    {
+      "epoch": 14.539473684210526,
+      "grad_norm": 11.013096809387207,
+      "learning_rate": 0.0001310569659873187,
+      "loss": 0.2765,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2210
+    },
+    {
+      "epoch": 14.605263157894736,
+      "grad_norm": 9.266365051269531,
+      "learning_rate": 0.00013051494707206742,
+      "loss": 0.2694,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2220
+    },
+    {
+      "epoch": 14.671052631578947,
+      "grad_norm": 7.71904993057251,
+      "learning_rate": 0.00012997289727874578,
+      "loss": 0.2444,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2230
+    },
+    {
+      "epoch": 14.736842105263158,
+      "grad_norm": 7.048625946044922,
+      "learning_rate": 0.00012943084911060034,
+      "loss": 0.2539,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2240
+    },
+    {
+      "epoch": 14.802631578947368,
+      "grad_norm": 9.421672821044922,
+      "learning_rate": 0.00012888883507078028,
+      "loss": 0.2633,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2250
+    },
+    {
+      "epoch": 14.868421052631579,
+      "grad_norm": 8.31679916381836,
+      "learning_rate": 0.0001283468876603883,
+      "loss": 0.2918,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2260
+    },
+    {
+      "epoch": 14.93421052631579,
+      "grad_norm": 10.346576690673828,
+      "learning_rate": 0.00012780503937653175,
+      "loss": 0.3205,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2270
+    },
+    {
+      "epoch": 15.0,
+      "grad_norm": 10.213838577270508,
+      "learning_rate": 0.00012726332271037407,
+      "loss": 0.2688,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2280
+    },
+    {
+      "epoch": 15.06578947368421,
+      "grad_norm": 7.39166784286499,
+      "learning_rate": 0.00012672177014518628,
+      "loss": 0.2079,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2290
+    },
+    {
+      "epoch": 15.131578947368421,
+      "grad_norm": 9.744684219360352,
+      "learning_rate": 0.0001261804141543995,
+      "loss": 0.2397,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2300
+    },
+    {
+      "epoch": 15.197368421052632,
+      "grad_norm": 8.710470199584961,
+      "learning_rate": 0.00012563928719965743,
+      "loss": 0.2277,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2310
+    },
+    {
+      "epoch": 15.263157894736842,
+      "grad_norm": 7.7291998863220215,
+      "learning_rate": 0.00012509842172886995,
+      "loss": 0.2373,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2320
+    },
+    {
+      "epoch": 15.328947368421053,
+      "grad_norm": 8.500285148620605,
+      "learning_rate": 0.00012455785017426743,
+      "loss": 0.2387,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2330
+    },
+    {
+      "epoch": 15.394736842105264,
+      "grad_norm": 12.7296781539917,
+      "learning_rate": 0.00012401760495045608,
+      "loss": 0.2405,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2340
+    },
+    {
+      "epoch": 15.460526315789474,
+      "grad_norm": 8.660082817077637,
+      "learning_rate": 0.000123477718452474,
+      "loss": 0.2075,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2350
+    },
+    {
+      "epoch": 15.526315789473685,
+      "grad_norm": 7.300625324249268,
+      "learning_rate": 0.00012293822305384886,
+      "loss": 0.2772,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2360
+    },
+    {
+      "epoch": 15.592105263157894,
+      "grad_norm": 8.595913887023926,
+      "learning_rate": 0.00012239915110465668,
+      "loss": 0.2583,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2370
+    },
+    {
+      "epoch": 15.657894736842106,
+      "grad_norm": 7.006585121154785,
+      "learning_rate": 0.00012186053492958178,
+      "loss": 0.2493,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2380
+    },
+    {
+      "epoch": 15.723684210526315,
+      "grad_norm": 7.5067267417907715,
+      "learning_rate": 0.00012132240682597882,
+      "loss": 0.2329,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2390
+    },
+    {
+      "epoch": 15.789473684210526,
+      "grad_norm": 9.166180610656738,
+      "learning_rate": 0.00012078479906193579,
+      "loss": 0.2474,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2400
+    },
+    {
+      "epoch": 15.855263157894736,
+      "grad_norm": 12.06090259552002,
+      "learning_rate": 0.00012024774387433933,
+      "loss": 0.2599,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2410
+    },
+    {
+      "epoch": 15.921052631578947,
+      "grad_norm": 8.838837623596191,
+      "learning_rate": 0.00011971127346694164,
+      "loss": 0.2822,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2420
+    },
+    {
+      "epoch": 15.986842105263158,
+      "grad_norm": 10.12195873260498,
+      "learning_rate": 0.00011917542000842932,
+      "loss": 0.2762,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2430
+    },
+    {
+      "epoch": 16.05263157894737,
+      "grad_norm": 7.118005275726318,
+      "learning_rate": 0.00011864021563049461,
+      "loss": 0.2289,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2440
+    },
+    {
+      "epoch": 16.11842105263158,
+      "grad_norm": 9.466529846191406,
+      "learning_rate": 0.0001181056924259085,
+      "loss": 0.2096,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2450
+    },
+    {
+      "epoch": 16.18421052631579,
+      "grad_norm": 6.640699863433838,
+      "learning_rate": 0.00011757188244659633,
+      "loss": 0.2044,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2460
+    },
+    {
+      "epoch": 16.25,
+      "grad_norm": 6.390682220458984,
+      "learning_rate": 0.000117038817701716,
+      "loss": 0.2199,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2470
+    },
+    {
+      "epoch": 16.31578947368421,
+      "grad_norm": 6.348979949951172,
+      "learning_rate": 0.00011650653015573847,
+      "loss": 0.2124,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2480
+    },
+    {
+      "epoch": 16.38157894736842,
+      "grad_norm": 6.506417274475098,
+      "learning_rate": 0.000115975051726531,
+      "loss": 0.2165,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2490
+    },
+    {
+      "epoch": 16.44736842105263,
+      "grad_norm": 9.003459930419922,
+      "learning_rate": 0.00011544441428344337,
+      "loss": 0.2201,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2500
+    },
+    {
+      "epoch": 16.513157894736842,
+      "grad_norm": 8.156404495239258,
+      "learning_rate": 0.00011491464964539678,
+      "loss": 0.2297,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2510
+    },
+    {
+      "epoch": 16.57894736842105,
+      "grad_norm": 6.955758571624756,
+      "learning_rate": 0.00011438578957897598,
+      "loss": 0.226,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2520
+    },
+    {
+      "epoch": 16.644736842105264,
+      "grad_norm": 5.890193462371826,
+      "learning_rate": 0.00011385786579652431,
+      "loss": 0.2266,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2530
+    },
+    {
+      "epoch": 16.710526315789473,
+      "grad_norm": 9.810389518737793,
+      "learning_rate": 0.0001133309099542422,
+      "loss": 0.2323,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2540
+    },
+    {
+      "epoch": 16.776315789473685,
+      "grad_norm": 8.91415023803711,
+      "learning_rate": 0.00011280495365028885,
+      "loss": 0.226,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2550
+    },
+    {
+      "epoch": 16.842105263157894,
+      "grad_norm": 7.513998985290527,
+      "learning_rate": 0.00011228002842288768,
+      "loss": 0.2223,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2560
+    },
+    {
+      "epoch": 16.907894736842106,
+      "grad_norm": 4.712993144989014,
+      "learning_rate": 0.00011175616574843499,
+      "loss": 0.2319,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2570
+    },
+    {
+      "epoch": 16.973684210526315,
+      "grad_norm": 7.016789436340332,
+      "learning_rate": 0.00011123339703961262,
+      "loss": 0.2256,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2580
+    },
+    {
+      "epoch": 17.039473684210527,
+      "grad_norm": 9.788738250732422,
+      "learning_rate": 0.00011071175364350432,
+      "loss": 0.2108,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2590
+    },
+    {
+      "epoch": 17.105263157894736,
+      "grad_norm": 7.460390567779541,
+      "learning_rate": 0.00011019126683971603,
+      "loss": 0.171,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2600
+    },
+    {
+      "epoch": 17.17105263157895,
+      "grad_norm": 8.930059432983398,
+      "learning_rate": 0.00010967196783850033,
+      "loss": 0.1919,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2610
+    },
+    {
+      "epoch": 17.236842105263158,
+      "grad_norm": 6.457809925079346,
+      "learning_rate": 0.00010915388777888482,
+      "loss": 0.2081,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2620
+    },
+    {
+      "epoch": 17.30263157894737,
+      "grad_norm": 7.385796070098877,
+      "learning_rate": 0.0001086370577268051,
+      "loss": 0.1974,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2630
+    },
+    {
+      "epoch": 17.36842105263158,
+      "grad_norm": 7.0241007804870605,
+      "learning_rate": 0.00010812150867324176,
+      "loss": 0.2199,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2640
+    },
+    {
+      "epoch": 17.43421052631579,
+      "grad_norm": 12.928653717041016,
+      "learning_rate": 0.00010760727153236215,
+      "loss": 0.2327,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2650
+    },
+    {
+      "epoch": 17.5,
+      "grad_norm": 7.8260884284973145,
+      "learning_rate": 0.00010709437713966664,
+      "loss": 0.2002,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2660
+    },
+    {
+      "epoch": 17.56578947368421,
+      "grad_norm": 5.859843730926514,
+      "learning_rate": 0.00010658285625013966,
+      "loss": 0.2215,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2670
+    },
+    {
+      "epoch": 17.63157894736842,
+      "grad_norm": 8.312082290649414,
+      "learning_rate": 0.00010607273953640542,
+      "loss": 0.194,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2680
+    },
+    {
+      "epoch": 17.69736842105263,
+      "grad_norm": 8.437788009643555,
+      "learning_rate": 0.0001055640575868887,
+      "loss": 0.225,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2690
+    },
+    {
+      "epoch": 17.763157894736842,
+      "grad_norm": 6.751978874206543,
+      "learning_rate": 0.00010505684090398072,
+      "loss": 0.189,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2700
+    },
+    {
+      "epoch": 17.82894736842105,
+      "grad_norm": 5.854010581970215,
+      "learning_rate": 0.00010455111990221009,
+      "loss": 0.2041,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2710
+    },
+    {
+      "epoch": 17.894736842105264,
+      "grad_norm": 9.748807907104492,
+      "learning_rate": 0.00010404692490641896,
+      "loss": 0.2077,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2720
+    },
+    {
+      "epoch": 17.960526315789473,
+      "grad_norm": 7.232575416564941,
+      "learning_rate": 0.00010354428614994476,
+      "loss": 0.211,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2730
+    },
+    {
+      "epoch": 18.026315789473685,
+      "grad_norm": 6.184597492218018,
+      "learning_rate": 0.00010304323377280715,
+      "loss": 0.1707,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2740
+    },
+    {
+      "epoch": 18.092105263157894,
+      "grad_norm": 5.879486083984375,
+      "learning_rate": 0.00010254379781990091,
+      "loss": 0.1556,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2750
+    },
+    {
+      "epoch": 18.157894736842106,
+      "grad_norm": 6.144423484802246,
+      "learning_rate": 0.00010204600823919419,
+      "loss": 0.1882,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2760
+    },
+    {
+      "epoch": 18.223684210526315,
+      "grad_norm": 7.17083740234375,
+      "learning_rate": 0.00010154989487993272,
+      "loss": 0.1742,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2770
+    },
+    {
+      "epoch": 18.289473684210527,
+      "grad_norm": 5.142153739929199,
+      "learning_rate": 0.00010105548749085007,
+      "loss": 0.1545,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2780
+    },
+    {
+      "epoch": 18.355263157894736,
+      "grad_norm": 4.805744171142578,
+      "learning_rate": 0.00010056281571838369,
+      "loss": 0.1752,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2790
+    },
+    {
+      "epoch": 18.42105263157895,
+      "grad_norm": 9.351277351379395,
+      "learning_rate": 0.00010007190910489725,
+      "loss": 0.1819,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2800
+    },
+    {
+      "epoch": 18.486842105263158,
+      "grad_norm": 7.704065799713135,
+      "learning_rate": 9.958279708690913e-05,
+      "loss": 0.1963,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2810
+    },
+    {
+      "epoch": 18.55263157894737,
+      "grad_norm": 7.071595191955566,
+      "learning_rate": 9.90955089933274e-05,
+      "loss": 0.1861,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2820
+    },
+    {
+      "epoch": 18.61842105263158,
+      "grad_norm": 6.073451042175293,
+      "learning_rate": 9.861007404369105e-05,
+      "loss": 0.1752,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2830
+    },
+    {
+      "epoch": 18.68421052631579,
+      "grad_norm": 10.440926551818848,
+      "learning_rate": 9.812652134641792e-05,
+      "loss": 0.2065,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2840
+    },
+    {
+      "epoch": 18.75,
+      "grad_norm": 4.4701924324035645,
+      "learning_rate": 9.764487989705931e-05,
+      "loss": 0.1853,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2850
+    },
+    {
+      "epoch": 18.81578947368421,
+      "grad_norm": 9.070950508117676,
+      "learning_rate": 9.716517857656117e-05,
+      "loss": 0.1932,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2860
+    },
+    {
+      "epoch": 18.88157894736842,
+      "grad_norm": 6.689568519592285,
+      "learning_rate": 9.668744614953243e-05,
+      "loss": 0.2051,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2870
+    },
+    {
+      "epoch": 18.94736842105263,
+      "grad_norm": 6.0212883949279785,
+      "learning_rate": 9.62117112625201e-05,
+      "loss": 0.181,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2880
+    },
+    {
+      "epoch": 19.013157894736842,
+      "grad_norm": 7.055004596710205,
+      "learning_rate": 9.573800244229162e-05,
+      "loss": 0.1926,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2890
+    },
+    {
+      "epoch": 19.07894736842105,
+      "grad_norm": 6.019172668457031,
+      "learning_rate": 9.526634809412406e-05,
+      "loss": 0.1522,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2900
+    },
+    {
+      "epoch": 19.144736842105264,
+      "grad_norm": 5.405515193939209,
+      "learning_rate": 9.479677650010104e-05,
+      "loss": 0.1477,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2910
+    },
+    {
+      "epoch": 19.210526315789473,
+      "grad_norm": 9.99826717376709,
+      "learning_rate": 9.432931581741688e-05,
+      "loss": 0.145,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2920
+    },
+    {
+      "epoch": 19.276315789473685,
+      "grad_norm": 5.589150428771973,
+      "learning_rate": 9.3863994076688e-05,
+      "loss": 0.1454,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2930
+    },
+    {
+      "epoch": 19.342105263157894,
+      "grad_norm": 5.293956756591797,
+      "learning_rate": 9.340083918027221e-05,
+      "loss": 0.1687,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2940
+    },
+    {
+      "epoch": 19.407894736842106,
+      "grad_norm": 8.118623733520508,
+      "learning_rate": 9.29398789005956e-05,
+      "loss": 0.1781,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2950
+    },
+    {
+      "epoch": 19.473684210526315,
+      "grad_norm": 5.828381538391113,
+      "learning_rate": 9.248114087848716e-05,
+      "loss": 0.1443,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2960
+    },
+    {
+      "epoch": 19.539473684210527,
+      "grad_norm": 9.651313781738281,
+      "learning_rate": 9.20246526215214e-05,
+      "loss": 0.1752,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2970
+    },
+    {
+      "epoch": 19.605263157894736,
+      "grad_norm": 6.652307510375977,
+      "learning_rate": 9.15704415023688e-05,
+      "loss": 0.1744,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2980
+    },
+    {
+      "epoch": 19.67105263157895,
+      "grad_norm": 6.448312282562256,
+      "learning_rate": 9.111853475715455e-05,
+      "loss": 0.1738,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 2990
+    },
+    {
+      "epoch": 19.736842105263158,
+      "grad_norm": 9.422768592834473,
+      "learning_rate": 9.06689594838253e-05,
+      "loss": 0.1546,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3000
+    },
+    {
+      "epoch": 19.80263157894737,
+      "grad_norm": 12.273077964782715,
+      "learning_rate": 9.022174264052431e-05,
+      "loss": 0.1998,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3010
+    },
+    {
+      "epoch": 19.86842105263158,
+      "grad_norm": 7.415524959564209,
+      "learning_rate": 8.977691104397492e-05,
+      "loss": 0.1873,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3020
+    },
+    {
+      "epoch": 19.93421052631579,
+      "grad_norm": 5.000813961029053,
+      "learning_rate": 8.93344913678725e-05,
+      "loss": 0.1546,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3030
+    },
+    {
+      "epoch": 20.0,
+      "grad_norm": 6.139503479003906,
+      "learning_rate": 8.889451014128516e-05,
+      "loss": 0.1721,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3040
+    },
+    {
+      "epoch": 20.06578947368421,
+      "grad_norm": 7.5522050857543945,
+      "learning_rate": 8.845699374706267e-05,
+      "loss": 0.1317,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3050
+    },
+    {
+      "epoch": 20.13157894736842,
+      "grad_norm": 6.387609958648682,
+      "learning_rate": 8.802196842025477e-05,
+      "loss": 0.1129,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3060
+    },
+    {
+      "epoch": 20.19736842105263,
+      "grad_norm": 9.34554672241211,
+      "learning_rate": 8.758946024653779e-05,
+      "loss": 0.1597,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3070
+    },
+    {
+      "epoch": 20.263157894736842,
+      "grad_norm": 3.530388593673706,
+      "learning_rate": 8.715949516065058e-05,
+      "loss": 0.1303,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3080
+    },
+    {
+      "epoch": 20.32894736842105,
+      "grad_norm": 5.443478107452393,
+      "learning_rate": 8.673209894483935e-05,
+      "loss": 0.1251,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3090
+    },
+    {
+      "epoch": 20.394736842105264,
+      "grad_norm": 5.092613220214844,
+      "learning_rate": 8.630729722731162e-05,
+      "loss": 0.1425,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3100
+    },
+    {
+      "epoch": 20.460526315789473,
+      "grad_norm": 9.21035385131836,
+      "learning_rate": 8.588511548069953e-05,
+      "loss": 0.155,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3110
+    },
+    {
+      "epoch": 20.526315789473685,
+      "grad_norm": 7.470590114593506,
+      "learning_rate": 8.54655790205324e-05,
+      "loss": 0.133,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3120
+    },
+    {
+      "epoch": 20.592105263157894,
+      "grad_norm": 6.920080184936523,
+      "learning_rate": 8.504871300371868e-05,
+      "loss": 0.1545,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3130
+    },
+    {
+      "epoch": 20.657894736842106,
+      "grad_norm": 5.543353080749512,
+      "learning_rate": 8.463454242703748e-05,
+      "loss": 0.1469,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3140
+    },
+    {
+      "epoch": 20.723684210526315,
+      "grad_norm": 7.444575786590576,
+      "learning_rate": 8.422309212563966e-05,
+      "loss": 0.1523,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3150
+    },
+    {
+      "epoch": 20.789473684210527,
+      "grad_norm": 9.51788330078125,
+      "learning_rate": 8.381438677155862e-05,
+      "loss": 0.1535,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3160
+    },
+    {
+      "epoch": 20.855263157894736,
+      "grad_norm": 5.5765862464904785,
+      "learning_rate": 8.34084508722309e-05,
+      "loss": 0.1747,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3170
+    },
+    {
+      "epoch": 20.92105263157895,
+      "grad_norm": 7.758854389190674,
+      "learning_rate": 8.300530876902665e-05,
+      "loss": 0.1366,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3180
+    },
+    {
+      "epoch": 20.986842105263158,
+      "grad_norm": 6.344717502593994,
+      "learning_rate": 8.260498463578997e-05,
+      "loss": 0.1446,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3190
+    },
+    {
+      "epoch": 21.05263157894737,
+      "grad_norm": 5.9779276847839355,
+      "learning_rate": 8.22075024773894e-05,
+      "loss": 0.1268,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3200
+    },
+    {
+      "epoch": 21.11842105263158,
+      "grad_norm": 3.6732325553894043,
+      "learning_rate": 8.181288612827847e-05,
+      "loss": 0.1181,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3210
+    },
+    {
+      "epoch": 21.18421052631579,
+      "grad_norm": 7.801024436950684,
+      "learning_rate": 8.142115925106652e-05,
+      "loss": 0.1125,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3220
+    },
+    {
+      "epoch": 21.25,
+      "grad_norm": 5.0273332595825195,
+      "learning_rate": 8.103234533509988e-05,
+      "loss": 0.1182,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3230
+    },
+    {
+      "epoch": 21.31578947368421,
+      "grad_norm": 10.303738594055176,
+      "learning_rate": 8.064646769505319e-05,
+      "loss": 0.12,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3240
+    },
+    {
+      "epoch": 21.38157894736842,
+      "grad_norm": 3.5069081783294678,
+      "learning_rate": 8.026354946953153e-05,
+      "loss": 0.1156,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3250
+    },
+    {
+      "epoch": 21.44736842105263,
+      "grad_norm": 8.985977172851562,
+      "learning_rate": 7.988361361968288e-05,
+      "loss": 0.1474,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3260
+    },
+    {
+      "epoch": 21.513157894736842,
+      "grad_norm": 7.387016296386719,
+      "learning_rate": 7.95066829278213e-05,
+      "loss": 0.1302,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3270
+    },
+    {
+      "epoch": 21.57894736842105,
+      "grad_norm": 4.838381767272949,
+      "learning_rate": 7.913277999606084e-05,
+      "loss": 0.121,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3280
+    },
+    {
+      "epoch": 21.644736842105264,
+      "grad_norm": 6.857619762420654,
+      "learning_rate": 7.876192724496012e-05,
+      "loss": 0.123,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3290
+    },
+    {
+      "epoch": 21.710526315789473,
+      "grad_norm": 5.610379219055176,
+      "learning_rate": 7.839414691217811e-05,
+      "loss": 0.1352,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3300
+    },
+    {
+      "epoch": 21.776315789473685,
+      "grad_norm": 5.731847286224365,
+      "learning_rate": 7.802946105114052e-05,
+      "loss": 0.1547,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3310
+    },
+    {
+      "epoch": 21.842105263157894,
+      "grad_norm": 6.4344377517700195,
+      "learning_rate": 7.766789152971747e-05,
+      "loss": 0.1416,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3320
+    },
+    {
+      "epoch": 21.907894736842106,
+      "grad_norm": 6.7426934242248535,
+      "learning_rate": 7.73094600289122e-05,
+      "loss": 0.1323,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3330
+    },
+    {
+      "epoch": 21.973684210526315,
+      "grad_norm": 7.127729415893555,
+      "learning_rate": 7.695418804156094e-05,
+      "loss": 0.1355,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3340
+    },
+    {
+      "epoch": 22.039473684210527,
+      "grad_norm": 3.8695497512817383,
+      "learning_rate": 7.66020968710443e-05,
+      "loss": 0.1197,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3350
+    },
+    {
+      "epoch": 22.105263157894736,
+      "grad_norm": 4.851317882537842,
+      "learning_rate": 7.625320763000964e-05,
+      "loss": 0.1108,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3360
+    },
+    {
+      "epoch": 22.17105263157895,
+      "grad_norm": 7.691250324249268,
+      "learning_rate": 7.590754123910515e-05,
+      "loss": 0.1168,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3370
+    },
+    {
+      "epoch": 22.236842105263158,
+      "grad_norm": 5.645074844360352,
+      "learning_rate": 7.556511842572543e-05,
+      "loss": 0.1116,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3380
+    },
+    {
+      "epoch": 22.30263157894737,
+      "grad_norm": 5.763618469238281,
+      "learning_rate": 7.522595972276851e-05,
+      "loss": 0.109,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3390
+    },
+    {
+      "epoch": 22.36842105263158,
+      "grad_norm": 8.13197135925293,
+      "learning_rate": 7.489008546740474e-05,
+      "loss": 0.1044,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3400
+    },
+    {
+      "epoch": 22.43421052631579,
+      "grad_norm": 7.799911022186279,
+      "learning_rate": 7.455751579985714e-05,
+      "loss": 0.1409,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3410
+    },
+    {
+      "epoch": 22.5,
+      "grad_norm": 6.310256004333496,
+      "learning_rate": 7.42282706621939e-05,
+      "loss": 0.1297,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3420
+    },
+    {
+      "epoch": 22.56578947368421,
+      "grad_norm": 6.451694965362549,
+      "learning_rate": 7.390236979713248e-05,
+      "loss": 0.1086,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3430
+    },
+    {
+      "epoch": 22.63157894736842,
+      "grad_norm": 8.012211799621582,
+      "learning_rate": 7.357983274685569e-05,
+      "loss": 0.1312,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3440
+    },
+    {
+      "epoch": 22.69736842105263,
+      "grad_norm": 7.39089298248291,
+      "learning_rate": 7.326067885184007e-05,
+      "loss": 0.1127,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3450
+    },
+    {
+      "epoch": 22.763157894736842,
+      "grad_norm": 7.845798969268799,
+      "learning_rate": 7.294492724969598e-05,
+      "loss": 0.1139,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3460
+    },
+    {
+      "epoch": 22.82894736842105,
+      "grad_norm": 5.291744709014893,
+      "learning_rate": 7.263259687402016e-05,
+      "loss": 0.1191,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3470
+    },
+    {
+      "epoch": 22.894736842105264,
+      "grad_norm": 6.590104103088379,
+      "learning_rate": 7.232370645326036e-05,
+      "loss": 0.1185,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3480
+    },
+    {
+      "epoch": 22.960526315789473,
+      "grad_norm": 5.286604404449463,
+      "learning_rate": 7.201827450959225e-05,
+      "loss": 0.1171,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3490
+    },
+    {
+      "epoch": 23.026315789473685,
+      "grad_norm": 3.8253142833709717,
+      "learning_rate": 7.171631935780896e-05,
+      "loss": 0.0955,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3500
+    },
+    {
+      "epoch": 23.092105263157894,
+      "grad_norm": 6.925575256347656,
+      "learning_rate": 7.141785910422259e-05,
+      "loss": 0.0919,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3510
+    },
+    {
+      "epoch": 23.157894736842106,
+      "grad_norm": 9.26873779296875,
+      "learning_rate": 7.11229116455787e-05,
+      "loss": 0.1031,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3520
+    },
+    {
+      "epoch": 23.223684210526315,
+      "grad_norm": 6.181906223297119,
+      "learning_rate": 7.083149466798312e-05,
+      "loss": 0.1151,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3530
+    },
+    {
+      "epoch": 23.289473684210527,
+      "grad_norm": 6.093035697937012,
+      "learning_rate": 7.054362564584128e-05,
+      "loss": 0.1033,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3540
+    },
+    {
+      "epoch": 23.355263157894736,
+      "grad_norm": 8.938755989074707,
+      "learning_rate": 7.025932184081061e-05,
+      "loss": 0.1059,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3550
+    },
+    {
+      "epoch": 23.42105263157895,
+      "grad_norm": 6.073050022125244,
+      "learning_rate": 6.997860030076529e-05,
+      "loss": 0.1256,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3560
+    },
+    {
+      "epoch": 23.486842105263158,
+      "grad_norm": 5.09712553024292,
+      "learning_rate": 6.97014778587741e-05,
+      "loss": 0.1054,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3570
+    },
+    {
+      "epoch": 23.55263157894737,
+      "grad_norm": 6.039034843444824,
+      "learning_rate": 6.9427971132091e-05,
+      "loss": 0.0938,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3580
+    },
+    {
+      "epoch": 23.61842105263158,
+      "grad_norm": 5.665463924407959,
+      "learning_rate": 6.91580965211587e-05,
+      "loss": 0.1012,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3590
+    },
+    {
+      "epoch": 23.68421052631579,
+      "grad_norm": 4.0321807861328125,
+      "learning_rate": 6.889187020862527e-05,
+      "loss": 0.1135,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3600
+    },
+    {
+      "epoch": 23.75,
+      "grad_norm": 8.287370681762695,
+      "learning_rate": 6.862930815837372e-05,
+      "loss": 0.1013,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3610
+    },
+    {
+      "epoch": 23.81578947368421,
+      "grad_norm": 5.64683723449707,
+      "learning_rate": 6.837042611456477e-05,
+      "loss": 0.1048,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3620
+    },
+    {
+      "epoch": 23.88157894736842,
+      "grad_norm": 3.9935617446899414,
+      "learning_rate": 6.81152396006928e-05,
+      "loss": 0.116,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3630
+    },
+    {
+      "epoch": 23.94736842105263,
+      "grad_norm": 4.570790767669678,
+      "learning_rate": 6.786376391865493e-05,
+      "loss": 0.1142,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3640
+    },
+    {
+      "epoch": 24.013157894736842,
+      "grad_norm": 3.90972900390625,
+      "learning_rate": 6.761601414783363e-05,
+      "loss": 0.1112,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3650
+    },
+    {
+      "epoch": 24.07894736842105,
+      "grad_norm": 4.516672611236572,
+      "learning_rate": 6.737200514419225e-05,
+      "loss": 0.0727,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3660
+    },
+    {
+      "epoch": 24.144736842105264,
+      "grad_norm": 4.735109329223633,
+      "learning_rate": 6.713175153938444e-05,
+      "loss": 0.0726,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3670
+    },
+    {
+      "epoch": 24.210526315789473,
+      "grad_norm": 5.059301376342773,
+      "learning_rate": 6.689526773987667e-05,
+      "loss": 0.0818,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3680
+    },
+    {
+      "epoch": 24.276315789473685,
+      "grad_norm": 4.411863803863525,
+      "learning_rate": 6.666256792608434e-05,
+      "loss": 0.0821,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3690
+    },
+    {
+      "epoch": 24.342105263157894,
+      "grad_norm": 5.8183465003967285,
+      "learning_rate": 6.643366605152156e-05,
+      "loss": 0.0943,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3700
+    },
+    {
+      "epoch": 24.407894736842106,
+      "grad_norm": 5.473435878753662,
+      "learning_rate": 6.620857584196439e-05,
+      "loss": 0.0864,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3710
+    },
+    {
+      "epoch": 24.473684210526315,
+      "grad_norm": 5.164342880249023,
+      "learning_rate": 6.598731079462784e-05,
+      "loss": 0.0847,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3720
+    },
+    {
+      "epoch": 24.539473684210527,
+      "grad_norm": 4.145204067230225,
+      "learning_rate": 6.576988417735645e-05,
+      "loss": 0.0845,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3730
+    },
+    {
+      "epoch": 24.605263157894736,
+      "grad_norm": 8.296338081359863,
+      "learning_rate": 6.555630902782875e-05,
+      "loss": 0.0978,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3740
+    },
+    {
+      "epoch": 24.67105263157895,
+      "grad_norm": 7.629764556884766,
+      "learning_rate": 6.534659815277551e-05,
+      "loss": 0.0995,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3750
+    },
+    {
+      "epoch": 24.736842105263158,
+      "grad_norm": 3.3637771606445312,
+      "learning_rate": 6.514076412721174e-05,
+      "loss": 0.1165,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3760
+    },
+    {
+      "epoch": 24.80263157894737,
+      "grad_norm": 3.8581345081329346,
+      "learning_rate": 6.493881929368267e-05,
+      "loss": 0.1068,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3770
+    },
+    {
+      "epoch": 24.86842105263158,
+      "grad_norm": 4.501940727233887,
+      "learning_rate": 6.474077576152366e-05,
+      "loss": 0.0821,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3780
+    },
+    {
+      "epoch": 24.93421052631579,
+      "grad_norm": 7.215863227844238,
+      "learning_rate": 6.454664540613403e-05,
+      "loss": 0.1116,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3790
+    },
+    {
+      "epoch": 25.0,
+      "grad_norm": 7.456452369689941,
+      "learning_rate": 6.43564398682651e-05,
+      "loss": 0.1105,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3800
+    },
+    {
+      "epoch": 25.06578947368421,
+      "grad_norm": 4.260319232940674,
+      "learning_rate": 6.417017055332196e-05,
+      "loss": 0.0692,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3810
+    },
+    {
+      "epoch": 25.13157894736842,
+      "grad_norm": 6.243800163269043,
+      "learning_rate": 6.39878486306798e-05,
+      "loss": 0.0809,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3820
+    },
+    {
+      "epoch": 25.19736842105263,
+      "grad_norm": 5.938414096832275,
+      "learning_rate": 6.380948503301394e-05,
+      "loss": 0.0672,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3830
+    },
+    {
+      "epoch": 25.263157894736842,
+      "grad_norm": 4.197356700897217,
+      "learning_rate": 6.363509045564438e-05,
+      "loss": 0.0704,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3840
+    },
+    {
+      "epoch": 25.32894736842105,
+      "grad_norm": 6.645064353942871,
+      "learning_rate": 6.346467535589447e-05,
+      "loss": 0.0728,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3850
+    },
+    {
+      "epoch": 25.394736842105264,
+      "grad_norm": 3.9544174671173096,
+      "learning_rate": 6.329824995246385e-05,
+      "loss": 0.0876,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3860
+    },
+    {
+      "epoch": 25.460526315789473,
+      "grad_norm": 7.730473041534424,
+      "learning_rate": 6.313582422481561e-05,
+      "loss": 0.0874,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3870
+    },
+    {
+      "epoch": 25.526315789473685,
+      "grad_norm": 4.0409770011901855,
+      "learning_rate": 6.297740791257803e-05,
+      "loss": 0.0916,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3880
+    },
+    {
+      "epoch": 25.592105263157894,
+      "grad_norm": 6.474958419799805,
+      "learning_rate": 6.282301051496045e-05,
+      "loss": 0.1027,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3890
+    },
+    {
+      "epoch": 25.657894736842106,
+      "grad_norm": 5.480382442474365,
+      "learning_rate": 6.267264129018372e-05,
+      "loss": 0.106,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3900
+    },
+    {
+      "epoch": 25.723684210526315,
+      "grad_norm": 5.7164788246154785,
+      "learning_rate": 6.252630925492498e-05,
+      "loss": 0.0769,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3910
+    },
+    {
+      "epoch": 25.789473684210527,
+      "grad_norm": 4.286190509796143,
+      "learning_rate": 6.23840231837771e-05,
+      "loss": 0.0876,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3920
+    },
+    {
+      "epoch": 25.855263157894736,
+      "grad_norm": 6.741421699523926,
+      "learning_rate": 6.224579160872236e-05,
+      "loss": 0.0915,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3930
+    },
+    {
+      "epoch": 25.92105263157895,
+      "grad_norm": 4.233421325683594,
+      "learning_rate": 6.211162281862105e-05,
+      "loss": 0.0813,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3940
+    },
+    {
+      "epoch": 25.986842105263158,
+      "grad_norm": 7.668642044067383,
+      "learning_rate": 6.19815248587142e-05,
+      "loss": 0.089,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3950
+    },
+    {
+      "epoch": 26.05263157894737,
+      "grad_norm": 3.4035732746124268,
+      "learning_rate": 6.185550553014139e-05,
+      "loss": 0.0747,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3960
+    },
+    {
+      "epoch": 26.11842105263158,
+      "grad_norm": 3.7069666385650635,
+      "learning_rate": 6.173357238947281e-05,
+      "loss": 0.0695,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3970
+    },
+    {
+      "epoch": 26.18421052631579,
+      "grad_norm": 5.699673652648926,
+      "learning_rate": 6.161573274825621e-05,
+      "loss": 0.0588,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3980
+    },
+    {
+      "epoch": 26.25,
+      "grad_norm": 6.198535442352295,
+      "learning_rate": 6.150199367257843e-05,
+      "loss": 0.0832,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 3990
+    },
+    {
+      "epoch": 26.31578947368421,
+      "grad_norm": 6.22232723236084,
+      "learning_rate": 6.139236198264172e-05,
+      "loss": 0.0764,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 4000
+    },
+    {
+      "epoch": 26.38157894736842,
+      "grad_norm": 9.198460578918457,
+      "learning_rate": 6.128684425235482e-05,
+      "loss": 0.0902,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 4010
+    },
+    {
+      "epoch": 26.44736842105263,
+      "grad_norm": 8.103082656860352,
+      "learning_rate": 6.118544680893863e-05,
+      "loss": 0.0855,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 4020
+    },
+    {
+      "epoch": 26.513157894736842,
+      "grad_norm": 5.254090785980225,
+      "learning_rate": 6.1088175732547e-05,
+      "loss": 0.0798,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 4030
+    },
+    {
+      "epoch": 26.57894736842105,
+      "grad_norm": 6.608073711395264,
+      "learning_rate": 6.099503685590196e-05,
+      "loss": 0.0877,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 4040
+    },
+    {
+      "epoch": 26.644736842105264,
+      "grad_norm": 6.865917682647705,
+      "learning_rate": 6.0906035763944055e-05,
+      "loss": 0.0783,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 4050
+    },
+    {
+      "epoch": 26.710526315789473,
+      "grad_norm": 5.2130446434021,
+      "learning_rate": 6.082117779349749e-05,
+      "loss": 0.0878,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 4060
+    },
+    {
+      "epoch": 26.776315789473685,
+      "grad_norm": 6.448189735412598,
+      "learning_rate": 6.074046803295003e-05,
+      "loss": 0.0879,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 4070
+    },
+    {
+      "epoch": 26.842105263157894,
+      "grad_norm": 4.945561408996582,
+      "learning_rate": 6.066391132194791e-05,
+      "loss": 0.0867,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 4080
+    },
+    {
+      "epoch": 26.907894736842106,
+      "grad_norm": 5.888365268707275,
+      "learning_rate": 6.0591512251105665e-05,
+      "loss": 0.0892,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 4090
+    },
+    {
+      "epoch": 26.973684210526315,
+      "grad_norm": 7.48224401473999,
+      "learning_rate": 6.0523275161730853e-05,
+      "loss": 0.0911,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 4100
+    },
+    {
+      "epoch": 27.039473684210527,
+      "grad_norm": 3.253098726272583,
+      "learning_rate": 6.0459204145563686e-05,
+      "loss": 0.0678,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 4110
+    },
+    {
+      "epoch": 27.105263157894736,
+      "grad_norm": 3.074223756790161,
+      "learning_rate": 6.039930304453173e-05,
+      "loss": 0.068,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 4120
+    },
+    {
+      "epoch": 27.17105263157895,
+      "grad_norm": 5.4365458488464355,
+      "learning_rate": 6.0343575450519485e-05,
+      "loss": 0.0682,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 4130
+    },
+    {
+      "epoch": 27.236842105263158,
+      "grad_norm": 4.085286617279053,
+      "learning_rate": 6.029202470515306e-05,
+      "loss": 0.0799,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 4140
+    },
+    {
+      "epoch": 27.30263157894737,
+      "grad_norm": 5.46896505355835,
+      "learning_rate": 6.024465389959973e-05,
+      "loss": 0.0722,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 4150
+    },
+    {
+      "epoch": 27.36842105263158,
+      "grad_norm": 5.356459617614746,
+      "learning_rate": 6.020146587438263e-05,
+      "loss": 0.0644,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 4160
+    },
+    {
+      "epoch": 27.43421052631579,
+      "grad_norm": 5.528810501098633,
+      "learning_rate": 6.0162463219210396e-05,
+      "loss": 0.0653,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 4170
+    },
+    {
+      "epoch": 27.5,
+      "grad_norm": 4.0345892906188965,
+      "learning_rate": 6.0127648272821894e-05,
+      "loss": 0.0615,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 4180
+    },
+    {
+      "epoch": 27.56578947368421,
+      "grad_norm": 4.616658687591553,
+      "learning_rate": 6.009702312284599e-05,
+      "loss": 0.0757,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 4190
+    },
+    {
+      "epoch": 27.63157894736842,
+      "grad_norm": 7.638866901397705,
+      "learning_rate": 6.0070589605676334e-05,
+      "loss": 0.0741,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 4200
+    },
+    {
+      "epoch": 27.69736842105263,
+      "grad_norm": 3.7885446548461914,
+      "learning_rate": 6.004834930636126e-05,
+      "loss": 0.0736,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 4210
+    },
+    {
+      "epoch": 27.763157894736842,
+      "grad_norm": 4.767909526824951,
+      "learning_rate": 6.0030303558508773e-05,
+      "loss": 0.0761,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 4220
+    },
+    {
+      "epoch": 27.82894736842105,
+      "grad_norm": 4.913625717163086,
+      "learning_rate": 6.001645344420652e-05,
+      "loss": 0.0767,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 4230
+    },
+    {
+      "epoch": 27.894736842105264,
+      "grad_norm": 3.778637170791626,
+      "learning_rate": 6.0006799793956955e-05,
+      "loss": 0.0888,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 4240
+    },
+    {
+      "epoch": 27.960526315789473,
+      "grad_norm": 4.147508144378662,
+      "learning_rate": 6.0001343186627523e-05,
+      "loss": 0.0804,
+      "memory/device_mem_reserved(gib)": 66.85,
+      "memory/max_mem_active(gib)": 58.44,
+      "memory/max_mem_allocated(gib)": 58.44,
+      "step": 4250
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 4257,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 29,
+  "save_steps": 100,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.1919564813923123e+18,
+  "train_batch_size": 12,
+  "trial_name": null,
+  "trial_params": null
+}