diff --git "a/checkpoint-6613/trainer_state.json" "b/checkpoint-6613/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/checkpoint-6613/trainer_state.json"
@@ -0,0 +1,4788 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 7.998790810157194,
+  "eval_steps": 414,
+  "global_step": 6613,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0006049606775559589,
+      "eval_loss": 1.4577645063400269,
+      "eval_runtime": 63.3298,
+      "eval_samples_per_second": 3.0,
+      "eval_steps_per_second": 3.0,
+      "step": 1
+    },
+    {
+      "epoch": 0.006049606775559589,
+      "grad_norm": 1.2785438299179077,
+      "learning_rate": 3e-06,
+      "loss": 1.374,
+      "step": 10
+    },
+    {
+      "epoch": 0.012099213551119177,
+      "grad_norm": 0.7358320355415344,
+      "learning_rate": 6e-06,
+      "loss": 1.2219,
+      "step": 20
+    },
+    {
+      "epoch": 0.018148820326678767,
+      "grad_norm": 3.1359198093414307,
+      "learning_rate": 9e-06,
+      "loss": 1.06,
+      "step": 30
+    },
+    {
+      "epoch": 0.024198427102238355,
+      "grad_norm": 1.704594612121582,
+      "learning_rate": 1.2e-05,
+      "loss": 1.4456,
+      "step": 40
+    },
+    {
+      "epoch": 0.030248033877797943,
+      "grad_norm": 1.833866834640503,
+      "learning_rate": 1.5e-05,
+      "loss": 1.1583,
+      "step": 50
+    },
+    {
+      "epoch": 0.036297640653357534,
+      "grad_norm": 1.3696911334991455,
+      "learning_rate": 1.8e-05,
+      "loss": 1.3531,
+      "step": 60
+    },
+    {
+      "epoch": 0.04234724742891712,
+      "grad_norm": 2.034799575805664,
+      "learning_rate": 2.1e-05,
+      "loss": 1.3092,
+      "step": 70
+    },
+    {
+      "epoch": 0.04839685420447671,
+      "grad_norm": 1.2183700799942017,
+      "learning_rate": 2.4e-05,
+      "loss": 1.3778,
+      "step": 80
+    },
+    {
+      "epoch": 0.0544464609800363,
+      "grad_norm": 2.252204418182373,
+      "learning_rate": 2.7000000000000002e-05,
+      "loss": 1.354,
+      "step": 90
+    },
+    {
+      "epoch": 0.060496067755595885,
+      "grad_norm": 2.8017024993896484,
+      "learning_rate": 3e-05,
+      "loss": 1.0258,
+      "step": 100
+    },
+    {
+      "epoch": 0.06654567453115548,
+      "grad_norm": 1.4654804468154907,
+      "learning_rate": 2.9999957023765772e-05,
+      "loss": 1.0931,
+      "step": 110
+    },
+    {
+      "epoch": 0.07259528130671507,
+      "grad_norm": 1.917343258857727,
+      "learning_rate": 2.999982809530935e-05,
+      "loss": 1.2745,
+      "step": 120
+    },
+    {
+      "epoch": 0.07864488808227466,
+      "grad_norm": 1.5257304906845093,
+      "learning_rate": 2.9999613215369516e-05,
+      "loss": 0.9593,
+      "step": 130
+    },
+    {
+      "epoch": 0.08469449485783424,
+      "grad_norm": 0.9834230542182922,
+      "learning_rate": 2.9999312385177566e-05,
+      "loss": 0.9462,
+      "step": 140
+    },
+    {
+      "epoch": 0.09074410163339383,
+      "grad_norm": 1.3462891578674316,
+      "learning_rate": 2.9998925606457308e-05,
+      "loss": 0.9853,
+      "step": 150
+    },
+    {
+      "epoch": 0.09679370840895342,
+      "grad_norm": 1.9642091989517212,
+      "learning_rate": 2.999845288142505e-05,
+      "loss": 1.1006,
+      "step": 160
+    },
+    {
+      "epoch": 0.102843315184513,
+      "grad_norm": 1.897597312927246,
+      "learning_rate": 2.9997894212789577e-05,
+      "loss": 1.1229,
+      "step": 170
+    },
+    {
+      "epoch": 0.1088929219600726,
+      "grad_norm": 1.2907172441482544,
+      "learning_rate": 2.9997249603752158e-05,
+      "loss": 0.9851,
+      "step": 180
+    },
+    {
+      "epoch": 0.11494252873563218,
+      "grad_norm": 1.7979787588119507,
+      "learning_rate": 2.9996519058006505e-05,
+      "loss": 1.0547,
+      "step": 190
+    },
+    {
+      "epoch": 0.12099213551119177,
+      "grad_norm": 2.189631462097168,
+      "learning_rate": 2.999570257973877e-05,
+      "loss": 1.1659,
+      "step": 200
+    },
+    {
+      "epoch": 0.12704174228675136,
+      "grad_norm": 1.6810208559036255,
+      "learning_rate": 2.9994800173627507e-05,
+      "loss": 0.8911,
+      "step": 210
+    },
+    {
+      "epoch": 0.13309134906231096,
+      "grad_norm": 1.385323405265808,
+      "learning_rate": 2.9993811844843646e-05,
+      "loss": 1.0701,
+      "step": 220
+    },
+    {
+      "epoch": 0.13914095583787053,
+      "grad_norm": 1.4298688173294067,
+      "learning_rate": 2.999273759905048e-05,
+      "loss": 1.0507,
+      "step": 230
+    },
+    {
+      "epoch": 0.14519056261343014,
+      "grad_norm": 1.5514591932296753,
+      "learning_rate": 2.9991577442403607e-05,
+      "loss": 1.1145,
+      "step": 240
+    },
+    {
+      "epoch": 0.1512401693889897,
+      "grad_norm": 1.8914713859558105,
+      "learning_rate": 2.999033138155093e-05,
+      "loss": 0.7448,
+      "step": 250
+    },
+    {
+      "epoch": 0.1572897761645493,
+      "grad_norm": 1.1297534704208374,
+      "learning_rate": 2.9988999423632562e-05,
+      "loss": 1.0863,
+      "step": 260
+    },
+    {
+      "epoch": 0.16333938294010888,
+      "grad_norm": 2.141075849533081,
+      "learning_rate": 2.998758157628086e-05,
+      "loss": 1.0313,
+      "step": 270
+    },
+    {
+      "epoch": 0.1693889897156685,
+      "grad_norm": 1.38814115524292,
+      "learning_rate": 2.998607784762031e-05,
+      "loss": 1.1764,
+      "step": 280
+    },
+    {
+      "epoch": 0.17543859649122806,
+      "grad_norm": 1.4910017251968384,
+      "learning_rate": 2.9984488246267533e-05,
+      "loss": 1.2147,
+      "step": 290
+    },
+    {
+      "epoch": 0.18148820326678766,
+      "grad_norm": 2.273433208465576,
+      "learning_rate": 2.99828127813312e-05,
+      "loss": 1.1703,
+      "step": 300
+    },
+    {
+      "epoch": 0.18753781004234724,
+      "grad_norm": 2.22914981842041,
+      "learning_rate": 2.9981051462412002e-05,
+      "loss": 1.1059,
+      "step": 310
+    },
+    {
+      "epoch": 0.19358741681790684,
+      "grad_norm": 1.5342557430267334,
+      "learning_rate": 2.997920429960259e-05,
+      "loss": 1.0767,
+      "step": 320
+    },
+    {
+      "epoch": 0.1996370235934664,
+      "grad_norm": 1.8331257104873657,
+      "learning_rate": 2.9977271303487513e-05,
+      "loss": 1.2202,
+      "step": 330
+    },
+    {
+      "epoch": 0.205686630369026,
+      "grad_norm": 3.037759304046631,
+      "learning_rate": 2.9975252485143145e-05,
+      "loss": 0.9495,
+      "step": 340
+    },
+    {
+      "epoch": 0.21173623714458562,
+      "grad_norm": 1.4192079305648804,
+      "learning_rate": 2.9973147856137658e-05,
+      "loss": 1.1555,
+      "step": 350
+    },
+    {
+      "epoch": 0.2177858439201452,
+      "grad_norm": 2.3369932174682617,
+      "learning_rate": 2.997095742853092e-05,
+      "loss": 1.073,
+      "step": 360
+    },
+    {
+      "epoch": 0.2238354506957048,
+      "grad_norm": 2.341386556625366,
+      "learning_rate": 2.9968681214874442e-05,
+      "loss": 1.2284,
+      "step": 370
+    },
+    {
+      "epoch": 0.22988505747126436,
+      "grad_norm": 1.4051792621612549,
+      "learning_rate": 2.9966319228211296e-05,
+      "loss": 1.2316,
+      "step": 380
+    },
+    {
+      "epoch": 0.23593466424682397,
+      "grad_norm": 2.0274314880371094,
+      "learning_rate": 2.996387148207607e-05,
+      "loss": 1.1257,
+      "step": 390
+    },
+    {
+      "epoch": 0.24198427102238354,
+      "grad_norm": 1.4432406425476074,
+      "learning_rate": 2.9961337990494733e-05,
+      "loss": 1.0545,
+      "step": 400
+    },
+    {
+      "epoch": 0.24803387779794314,
+      "grad_norm": 1.7953746318817139,
+      "learning_rate": 2.995871876798462e-05,
+      "loss": 0.8797,
+      "step": 410
+    },
+    {
+      "epoch": 0.25045372050816694,
+      "eval_loss": 1.0715683698654175,
+      "eval_runtime": 64.6596,
+      "eval_samples_per_second": 2.938,
+      "eval_steps_per_second": 2.938,
+      "step": 414
+    },
+    {
+      "epoch": 0.2540834845735027,
+      "grad_norm": 1.0973708629608154,
+      "learning_rate": 2.995601382955431e-05,
+      "loss": 0.9573,
+      "step": 420
+    },
+    {
+      "epoch": 0.2601330913490623,
+      "grad_norm": 1.390816569328308,
+      "learning_rate": 2.9953223190703535e-05,
+      "loss": 1.1024,
+      "step": 430
+    },
+    {
+      "epoch": 0.2661826981246219,
+      "grad_norm": 1.5752558708190918,
+      "learning_rate": 2.9950346867423124e-05,
+      "loss": 1.2828,
+      "step": 440
+    },
+    {
+      "epoch": 0.27223230490018147,
+      "grad_norm": 3.318897247314453,
+      "learning_rate": 2.9947384876194876e-05,
+      "loss": 0.9254,
+      "step": 450
+    },
+    {
+      "epoch": 0.27828191167574107,
+      "grad_norm": 2.932630777359009,
+      "learning_rate": 2.994433723399149e-05,
+      "loss": 1.3966,
+      "step": 460
+    },
+    {
+      "epoch": 0.28433151845130067,
+      "grad_norm": 1.24905264377594,
+      "learning_rate": 2.9941203958276463e-05,
+      "loss": 1.0706,
+      "step": 470
+    },
+    {
+      "epoch": 0.29038112522686027,
+      "grad_norm": 1.6373075246810913,
+      "learning_rate": 2.993798506700397e-05,
+      "loss": 1.0972,
+      "step": 480
+    },
+    {
+      "epoch": 0.2964307320024198,
+      "grad_norm": 1.8339027166366577,
+      "learning_rate": 2.9934680578618796e-05,
+      "loss": 1.2781,
+      "step": 490
+    },
+    {
+      "epoch": 0.3024803387779794,
+      "grad_norm": 1.5762040615081787,
+      "learning_rate": 2.9931290512056206e-05,
+      "loss": 0.9826,
+      "step": 500
+    },
+    {
+      "epoch": 0.308529945553539,
+      "grad_norm": 1.3458601236343384,
+      "learning_rate": 2.9927814886741828e-05,
+      "loss": 1.128,
+      "step": 510
+    },
+    {
+      "epoch": 0.3145795523290986,
+      "grad_norm": 2.5347025394439697,
+      "learning_rate": 2.9924253722591575e-05,
+      "loss": 1.0372,
+      "step": 520
+    },
+    {
+      "epoch": 0.3206291591046582,
+      "grad_norm": 1.329092025756836,
+      "learning_rate": 2.99206070400115e-05,
+      "loss": 0.9917,
+      "step": 530
+    },
+    {
+      "epoch": 0.32667876588021777,
+      "grad_norm": 1.4007670879364014,
+      "learning_rate": 2.9916874859897704e-05,
+      "loss": 0.9245,
+      "step": 540
+    },
+    {
+      "epoch": 0.33272837265577737,
+      "grad_norm": 1.8608225584030151,
+      "learning_rate": 2.9913057203636176e-05,
+      "loss": 1.0985,
+      "step": 550
+    },
+    {
+      "epoch": 0.338777979431337,
+      "grad_norm": 1.2260061502456665,
+      "learning_rate": 2.990915409310273e-05,
+      "loss": 1.1108,
+      "step": 560
+    },
+    {
+      "epoch": 0.3448275862068966,
+      "grad_norm": 4.154330730438232,
+      "learning_rate": 2.990516555066282e-05,
+      "loss": 0.9983,
+      "step": 570
+    },
+    {
+      "epoch": 0.3508771929824561,
+      "grad_norm": 1.5533702373504639,
+      "learning_rate": 2.990109159917146e-05,
+      "loss": 0.9014,
+      "step": 580
+    },
+    {
+      "epoch": 0.3569267997580157,
+      "grad_norm": 1.5335776805877686,
+      "learning_rate": 2.9896932261973056e-05,
+      "loss": 1.0543,
+      "step": 590
+    },
+    {
+      "epoch": 0.3629764065335753,
+      "grad_norm": 5.594316482543945,
+      "learning_rate": 2.98926875629013e-05,
+      "loss": 1.0405,
+      "step": 600
+    },
+    {
+      "epoch": 0.3690260133091349,
+      "grad_norm": 1.4797818660736084,
+      "learning_rate": 2.9888357526279008e-05,
+      "loss": 0.9019,
+      "step": 610
+    },
+    {
+      "epoch": 0.37507562008469447,
+      "grad_norm": 0.40816277265548706,
+      "learning_rate": 2.9883942176918015e-05,
+      "loss": 0.8362,
+      "step": 620
+    },
+    {
+      "epoch": 0.3811252268602541,
+      "grad_norm": 1.1441352367401123,
+      "learning_rate": 2.9879441540118988e-05,
+      "loss": 1.0765,
+      "step": 630
+    },
+    {
+      "epoch": 0.3871748336358137,
+      "grad_norm": 1.28082275390625,
+      "learning_rate": 2.9874855641671325e-05,
+      "loss": 0.8888,
+      "step": 640
+    },
+    {
+      "epoch": 0.3932244404113733,
+      "grad_norm": 1.6251568794250488,
+      "learning_rate": 2.987018450785297e-05,
+      "loss": 1.2763,
+      "step": 650
+    },
+    {
+      "epoch": 0.3992740471869328,
+      "grad_norm": 3.480515956878662,
+      "learning_rate": 2.9865428165430292e-05,
+      "loss": 1.191,
+      "step": 660
+    },
+    {
+      "epoch": 0.4053236539624924,
+      "grad_norm": 1.7192935943603516,
+      "learning_rate": 2.9860586641657922e-05,
+      "loss": 1.1641,
+      "step": 670
+    },
+    {
+      "epoch": 0.411373260738052,
+      "grad_norm": 3.2280263900756836,
+      "learning_rate": 2.9855659964278584e-05,
+      "loss": 1.108,
+      "step": 680
+    },
+    {
+      "epoch": 0.41742286751361163,
+      "grad_norm": 3.256509304046631,
+      "learning_rate": 2.9850648161522945e-05,
+      "loss": 1.0619,
+      "step": 690
+    },
+    {
+      "epoch": 0.42347247428917123,
+      "grad_norm": 0.8531071543693542,
+      "learning_rate": 2.984555126210946e-05,
+      "loss": 0.9536,
+      "step": 700
+    },
+    {
+      "epoch": 0.4295220810647308,
+      "grad_norm": 1.469575047492981,
+      "learning_rate": 2.984036929524421e-05,
+      "loss": 1.0315,
+      "step": 710
+    },
+    {
+      "epoch": 0.4355716878402904,
+      "grad_norm": 1.530128836631775,
+      "learning_rate": 2.983510229062071e-05,
+      "loss": 0.9523,
+      "step": 720
+    },
+    {
+      "epoch": 0.44162129461585,
+      "grad_norm": 1.4602179527282715,
+      "learning_rate": 2.982975027841977e-05,
+      "loss": 0.9123,
+      "step": 730
+    },
+    {
+      "epoch": 0.4476709013914096,
+      "grad_norm": 1.4414767026901245,
+      "learning_rate": 2.9824313289309294e-05,
+      "loss": 0.8937,
+      "step": 740
+    },
+    {
+      "epoch": 0.4537205081669691,
+      "grad_norm": 1.5764180421829224,
+      "learning_rate": 2.9818791354444127e-05,
+      "loss": 0.8378,
+      "step": 750
+    },
+    {
+      "epoch": 0.45977011494252873,
+      "grad_norm": 1.798797369003296,
+      "learning_rate": 2.981318450546587e-05,
+      "loss": 1.1008,
+      "step": 760
+    },
+    {
+      "epoch": 0.46581972171808833,
+      "grad_norm": 2.5538880825042725,
+      "learning_rate": 2.9807492774502675e-05,
+      "loss": 1.0576,
+      "step": 770
+    },
+    {
+      "epoch": 0.47186932849364793,
+      "grad_norm": 1.5585602521896362,
+      "learning_rate": 2.9801716194169113e-05,
+      "loss": 1.0796,
+      "step": 780
+    },
+    {
+      "epoch": 0.4779189352692075,
+      "grad_norm": 1.1259243488311768,
+      "learning_rate": 2.9795854797565933e-05,
+      "loss": 0.8746,
+      "step": 790
+    },
+    {
+      "epoch": 0.4839685420447671,
+      "grad_norm": 2.3184139728546143,
+      "learning_rate": 2.97899086182799e-05,
+      "loss": 0.9604,
+      "step": 800
+    },
+    {
+      "epoch": 0.4900181488203267,
+      "grad_norm": 2.184471845626831,
+      "learning_rate": 2.9783877690383607e-05,
+      "loss": 1.1267,
+      "step": 810
+    },
+    {
+      "epoch": 0.4960677555958863,
+      "grad_norm": 1.2299617528915405,
+      "learning_rate": 2.9777762048435256e-05,
+      "loss": 1.1073,
+      "step": 820
+    },
+    {
+      "epoch": 0.5009074410163339,
+      "eval_loss": 1.0542737245559692,
+      "eval_runtime": 64.7633,
+      "eval_samples_per_second": 2.934,
+      "eval_steps_per_second": 2.934,
+      "step": 828
+    },
+    {
+      "epoch": 0.5021173623714459,
+      "grad_norm": 1.2418574094772339,
+      "learning_rate": 2.9771561727478487e-05,
+      "loss": 0.9079,
+      "step": 830
+    },
+    {
+      "epoch": 0.5081669691470054,
+      "grad_norm": 1.4546465873718262,
+      "learning_rate": 2.976527676304216e-05,
+      "loss": 0.9962,
+      "step": 840
+    },
+    {
+      "epoch": 0.514216575922565,
+      "grad_norm": 1.7091883420944214,
+      "learning_rate": 2.9758907191140147e-05,
+      "loss": 1.0415,
+      "step": 850
+    },
+    {
+      "epoch": 0.5202661826981246,
+      "grad_norm": 1.6179261207580566,
+      "learning_rate": 2.9752453048271152e-05,
+      "loss": 1.008,
+      "step": 860
+    },
+    {
+      "epoch": 0.5263157894736842,
+      "grad_norm": 1.8634966611862183,
+      "learning_rate": 2.974591437141847e-05,
+      "loss": 1.1791,
+      "step": 870
+    },
+    {
+      "epoch": 0.5323653962492438,
+      "grad_norm": 1.787575125694275,
+      "learning_rate": 2.9739291198049802e-05,
+      "loss": 1.1416,
+      "step": 880
+    },
+    {
+      "epoch": 0.5384150030248034,
+      "grad_norm": 0.6841387152671814,
+      "learning_rate": 2.9732583566117016e-05,
+      "loss": 1.0713,
+      "step": 890
+    },
+    {
+      "epoch": 0.5444646098003629,
+      "grad_norm": 1.1443063020706177,
+      "learning_rate": 2.972579151405595e-05,
+      "loss": 0.9909,
+      "step": 900
+    },
+    {
+      "epoch": 0.5505142165759226,
+      "grad_norm": 1.3013931512832642,
+      "learning_rate": 2.9718915080786173e-05,
+      "loss": 1.035,
+      "step": 910
+    },
+    {
+      "epoch": 0.5565638233514821,
+      "grad_norm": 3.635468006134033,
+      "learning_rate": 2.971195430571079e-05,
+      "loss": 1.1064,
+      "step": 920
+    },
+    {
+      "epoch": 0.5626134301270418,
+      "grad_norm": 2.095000743865967,
+      "learning_rate": 2.970490922871618e-05,
+      "loss": 1.0936,
+      "step": 930
+    },
+    {
+      "epoch": 0.5686630369026013,
+      "grad_norm": 1.4885168075561523,
+      "learning_rate": 2.9697779890171798e-05,
+      "loss": 1.0447,
+      "step": 940
+    },
+    {
+      "epoch": 0.5747126436781609,
+      "grad_norm": 1.1731853485107422,
+      "learning_rate": 2.969056633092992e-05,
+      "loss": 1.0816,
+      "step": 950
+    },
+    {
+      "epoch": 0.5807622504537205,
+      "grad_norm": 1.6034612655639648,
+      "learning_rate": 2.9683268592325437e-05,
+      "loss": 1.1776,
+      "step": 960
+    },
+    {
+      "epoch": 0.5868118572292801,
+      "grad_norm": 1.821656346321106,
+      "learning_rate": 2.967588671617559e-05,
+      "loss": 0.8754,
+      "step": 970
+    },
+    {
+      "epoch": 0.5928614640048396,
+      "grad_norm": 0.9973627328872681,
+      "learning_rate": 2.9668420744779733e-05,
+      "loss": 1.0242,
+      "step": 980
+    },
+    {
+      "epoch": 0.5989110707803993,
+      "grad_norm": 1.5308775901794434,
+      "learning_rate": 2.9660870720919127e-05,
+      "loss": 1.0948,
+      "step": 990
+    },
+    {
+      "epoch": 0.6049606775559588,
+      "grad_norm": 1.8341882228851318,
+      "learning_rate": 2.965323668785664e-05,
+      "loss": 1.0485,
+      "step": 1000
+    },
+    {
+      "epoch": 0.6110102843315185,
+      "grad_norm": 1.383766531944275,
+      "learning_rate": 2.9645518689336543e-05,
+      "loss": 0.9551,
+      "step": 1010
+    },
+    {
+      "epoch": 0.617059891107078,
+      "grad_norm": 1.5288898944854736,
+      "learning_rate": 2.9637716769584235e-05,
+      "loss": 0.7649,
+      "step": 1020
+    },
+    {
+      "epoch": 0.6231094978826376,
+      "grad_norm": 1.0173107385635376,
+      "learning_rate": 2.9629830973306e-05,
+      "loss": 1.2116,
+      "step": 1030
+    },
+    {
+      "epoch": 0.6291591046581972,
+      "grad_norm": 19.645925521850586,
+      "learning_rate": 2.9621861345688754e-05,
+      "loss": 1.0067,
+      "step": 1040
+    },
+    {
+      "epoch": 0.6352087114337568,
+      "grad_norm": 1.1569609642028809,
+      "learning_rate": 2.9613807932399767e-05,
+      "loss": 0.9409,
+      "step": 1050
+    },
+    {
+      "epoch": 0.6412583182093164,
+      "grad_norm": 4.655813694000244,
+      "learning_rate": 2.9605670779586426e-05,
+      "loss": 1.0169,
+      "step": 1060
+    },
+    {
+      "epoch": 0.647307924984876,
+      "grad_norm": 1.9381617307662964,
+      "learning_rate": 2.9597449933875955e-05,
+      "loss": 1.1012,
+      "step": 1070
+    },
+    {
+      "epoch": 0.6533575317604355,
+      "grad_norm": 1.3556163311004639,
+      "learning_rate": 2.958914544237515e-05,
+      "loss": 0.8116,
+      "step": 1080
+    },
+    {
+      "epoch": 0.6594071385359952,
+      "grad_norm": 1.453025460243225,
+      "learning_rate": 2.9580757352670125e-05,
+      "loss": 1.0269,
+      "step": 1090
+    },
+    {
+      "epoch": 0.6654567453115547,
+      "grad_norm": 1.458400011062622,
+      "learning_rate": 2.9572285712825998e-05,
+      "loss": 1.0251,
+      "step": 1100
+    },
+    {
+      "epoch": 0.6715063520871143,
+      "grad_norm": 2.5511817932128906,
+      "learning_rate": 2.9563730571386676e-05,
+      "loss": 1.0311,
+      "step": 1110
+    },
+    {
+      "epoch": 0.677555958862674,
+      "grad_norm": 1.1598625183105469,
+      "learning_rate": 2.9555091977374512e-05,
+      "loss": 0.9025,
+      "step": 1120
+    },
+    {
+      "epoch": 0.6836055656382335,
+      "grad_norm": 1.7414718866348267,
+      "learning_rate": 2.9546369980290082e-05,
+      "loss": 0.9769,
+      "step": 1130
+    },
+    {
+      "epoch": 0.6896551724137931,
+      "grad_norm": 1.4089117050170898,
+      "learning_rate": 2.953756463011186e-05,
+      "loss": 0.9785,
+      "step": 1140
+    },
+    {
+      "epoch": 0.6957047791893527,
+      "grad_norm": 1.2085853815078735,
+      "learning_rate": 2.9528675977295954e-05,
+      "loss": 1.071,
+      "step": 1150
+    },
+    {
+      "epoch": 0.7017543859649122,
+      "grad_norm": 1.2800265550613403,
+      "learning_rate": 2.9519704072775806e-05,
+      "loss": 1.0059,
+      "step": 1160
+    },
+    {
+      "epoch": 0.7078039927404719,
+      "grad_norm": 1.5352321863174438,
+      "learning_rate": 2.9510648967961903e-05,
+      "loss": 1.0462,
+      "step": 1170
+    },
+    {
+      "epoch": 0.7138535995160314,
+      "grad_norm": 1.59364652633667,
+      "learning_rate": 2.9501510714741494e-05,
+      "loss": 1.0539,
+      "step": 1180
+    },
+    {
+      "epoch": 0.719903206291591,
+      "grad_norm": 0.7905394434928894,
+      "learning_rate": 2.9492289365478263e-05,
+      "loss": 1.085,
+      "step": 1190
+    },
+    {
+      "epoch": 0.7259528130671506,
+      "grad_norm": 1.8049817085266113,
+      "learning_rate": 2.9482984973012064e-05,
+      "loss": 0.9335,
+      "step": 1200
+    },
+    {
+      "epoch": 0.7320024198427102,
+      "grad_norm": 1.6577391624450684,
+      "learning_rate": 2.9473597590658595e-05,
+      "loss": 1.062,
+      "step": 1210
+    },
+    {
+      "epoch": 0.7380520266182699,
+      "grad_norm": 2.00862717628479,
+      "learning_rate": 2.9464127272209107e-05,
+      "loss": 1.265,
+      "step": 1220
+    },
+    {
+      "epoch": 0.7441016333938294,
+      "grad_norm": 1.7227792739868164,
+      "learning_rate": 2.9454574071930075e-05,
+      "loss": 1.198,
+      "step": 1230
+    },
+    {
+      "epoch": 0.7501512401693889,
+      "grad_norm": 2.7444136142730713,
+      "learning_rate": 2.9444938044562916e-05,
+      "loss": 0.9352,
+      "step": 1240
+    },
+    {
+      "epoch": 0.7513611615245009,
+      "eval_loss": 1.0343616008758545,
+      "eval_runtime": 64.7966,
+      "eval_samples_per_second": 2.932,
+      "eval_steps_per_second": 2.932,
+      "step": 1242
+    },
+    {
+      "epoch": 0.7562008469449486,
+      "grad_norm": 2.819312572479248,
+      "learning_rate": 2.9435219245323653e-05,
+      "loss": 1.0105,
+      "step": 1250
+    },
+    {
+      "epoch": 0.7622504537205081,
+      "grad_norm": 1.6756153106689453,
+      "learning_rate": 2.9425417729902596e-05,
+      "loss": 0.9951,
+      "step": 1260
+    },
+    {
+      "epoch": 0.7683000604960678,
+      "grad_norm": 2.266688823699951,
+      "learning_rate": 2.941553355446405e-05,
+      "loss": 1.2856,
+      "step": 1270
+    },
+    {
+      "epoch": 0.7743496672716274,
+      "grad_norm": 1.9197466373443604,
+      "learning_rate": 2.940556677564596e-05,
+      "loss": 1.025,
+      "step": 1280
+    },
+    {
+      "epoch": 0.7803992740471869,
+      "grad_norm": 1.2381973266601562,
+      "learning_rate": 2.939551745055962e-05,
+      "loss": 0.9614,
+      "step": 1290
+    },
+    {
+      "epoch": 0.7864488808227466,
+      "grad_norm": 1.3409382104873657,
+      "learning_rate": 2.9385385636789308e-05,
+      "loss": 1.0783,
+      "step": 1300
+    },
+    {
+      "epoch": 0.7924984875983061,
+      "grad_norm": 1.3915735483169556,
+      "learning_rate": 2.937517139239198e-05,
+      "loss": 0.9862,
+      "step": 1310
+    },
+    {
+      "epoch": 0.7985480943738656,
+      "grad_norm": 3.1286187171936035,
+      "learning_rate": 2.936487477589695e-05,
+      "loss": 1.0044,
+      "step": 1320
+    },
+    {
+      "epoch": 0.8045977011494253,
+      "grad_norm": 2.222550868988037,
+      "learning_rate": 2.9354495846305513e-05,
+      "loss": 0.9816,
+      "step": 1330
+    },
+    {
+      "epoch": 0.8106473079249849,
+      "grad_norm": 1.239647626876831,
+      "learning_rate": 2.934403466309065e-05,
+      "loss": 1.1207,
+      "step": 1340
+    },
+    {
+      "epoch": 0.8166969147005445,
+      "grad_norm": 1.5290794372558594,
+      "learning_rate": 2.933349128619666e-05,
+      "loss": 1.2575,
+      "step": 1350
+    },
+    {
+      "epoch": 0.822746521476104,
+      "grad_norm": 1.5801492929458618,
+      "learning_rate": 2.932286577603883e-05,
+      "loss": 0.9342,
+      "step": 1360
+    },
+    {
+      "epoch": 0.8287961282516636,
+      "grad_norm": 2.460664749145508,
+      "learning_rate": 2.9312158193503076e-05,
+      "loss": 0.9571,
+      "step": 1370
+    },
+    {
+      "epoch": 0.8348457350272233,
+      "grad_norm": 1.5209215879440308,
+      "learning_rate": 2.9301368599945616e-05,
+      "loss": 1.1786,
+      "step": 1380
+    },
+    {
+      "epoch": 0.8408953418027828,
+      "grad_norm": 1.3077425956726074,
+      "learning_rate": 2.929049705719259e-05,
+      "loss": 0.8198,
+      "step": 1390
+    },
+    {
+      "epoch": 0.8469449485783425,
+      "grad_norm": 1.7702195644378662,
+      "learning_rate": 2.927954362753973e-05,
+      "loss": 0.978,
+      "step": 1400
+    },
+    {
+      "epoch": 0.852994555353902,
+      "grad_norm": 2.0915708541870117,
+      "learning_rate": 2.926850837375199e-05,
+      "loss": 1.0862,
+      "step": 1410
+    },
+    {
+      "epoch": 0.8590441621294616,
+      "grad_norm": 1.815843105316162,
+      "learning_rate": 2.9257391359063196e-05,
+      "loss": 1.1928,
+      "step": 1420
+    },
+    {
+      "epoch": 0.8650937689050212,
+      "grad_norm": 2.303849697113037,
+      "learning_rate": 2.924619264717566e-05,
+      "loss": 1.0549,
+      "step": 1430
+    },
+    {
+      "epoch": 0.8711433756805808,
+      "grad_norm": 1.3767309188842773,
+      "learning_rate": 2.9234912302259855e-05,
+      "loss": 0.7791,
+      "step": 1440
+    },
+    {
+      "epoch": 0.8771929824561403,
+      "grad_norm": 1.6787207126617432,
+      "learning_rate": 2.9223550388954005e-05,
+      "loss": 1.0006,
+      "step": 1450
+    },
+    {
+      "epoch": 0.8832425892317,
+      "grad_norm": 1.1419081687927246,
+      "learning_rate": 2.9212106972363748e-05,
+      "loss": 1.1581,
+      "step": 1460
+    },
+    {
+      "epoch": 0.8892921960072595,
+      "grad_norm": 1.349205732345581,
+      "learning_rate": 2.9200582118061743e-05,
+      "loss": 0.9921,
+      "step": 1470
+    },
+    {
+      "epoch": 0.8953418027828192,
+      "grad_norm": 1.474130392074585,
+      "learning_rate": 2.9188975892087303e-05,
+      "loss": 0.8283,
+      "step": 1480
+    },
+    {
+      "epoch": 0.9013914095583787,
+      "grad_norm": 1.0713589191436768,
+      "learning_rate": 2.917728836094601e-05,
+      "loss": 1.0369,
+      "step": 1490
+    },
+    {
+      "epoch": 0.9074410163339383,
+      "grad_norm": 1.2351691722869873,
+      "learning_rate": 2.9165519591609345e-05,
+      "loss": 0.9484,
+      "step": 1500
+    },
+    {
+      "epoch": 0.9134906231094979,
+      "grad_norm": 0.9272758364677429,
+      "learning_rate": 2.9153669651514287e-05,
+      "loss": 0.9043,
+      "step": 1510
+    },
+    {
+      "epoch": 0.9195402298850575,
+      "grad_norm": 2.782837390899658,
+      "learning_rate": 2.9141738608562947e-05,
+      "loss": 1.0876,
+      "step": 1520
+    },
+    {
+      "epoch": 0.925589836660617,
+      "grad_norm": 1.687536358833313,
+      "learning_rate": 2.9129726531122165e-05,
+      "loss": 0.9161,
+      "step": 1530
+    },
+    {
+      "epoch": 0.9316394434361767,
+      "grad_norm": 2.1093292236328125,
+      "learning_rate": 2.9117633488023116e-05,
+      "loss": 0.9388,
+      "step": 1540
+    },
+    {
+      "epoch": 0.9376890502117362,
+      "grad_norm": 1.5271527767181396,
+      "learning_rate": 2.9105459548560938e-05,
+      "loss": 0.9626,
+      "step": 1550
+    },
+    {
+      "epoch": 0.9437386569872959,
+      "grad_norm": 1.899366021156311,
+      "learning_rate": 2.9093204782494297e-05,
+      "loss": 0.9612,
+      "step": 1560
+    },
+    {
+      "epoch": 0.9497882637628554,
+      "grad_norm": 2.3809289932250977,
+      "learning_rate": 2.908086926004502e-05,
+      "loss": 1.1344,
+      "step": 1570
+    },
+    {
+      "epoch": 0.955837870538415,
+      "grad_norm": 2.101499557495117,
+      "learning_rate": 2.9068453051897688e-05,
+      "loss": 0.9209,
+      "step": 1580
+    },
+    {
+      "epoch": 0.9618874773139746,
+      "grad_norm": 1.0858937501907349,
+      "learning_rate": 2.905595622919921e-05,
+      "loss": 0.9232,
+      "step": 1590
+    },
+    {
+      "epoch": 0.9679370840895342,
+      "grad_norm": 1.7263002395629883,
+      "learning_rate": 2.9043378863558442e-05,
+      "loss": 1.0186,
+      "step": 1600
+    },
+    {
+      "epoch": 0.9739866908650938,
+      "grad_norm": 1.759886622428894,
+      "learning_rate": 2.9030721027045752e-05,
+      "loss": 0.9193,
+      "step": 1610
+    },
+    {
+      "epoch": 0.9800362976406534,
+      "grad_norm": 2.344141960144043,
+      "learning_rate": 2.9017982792192636e-05,
+      "loss": 1.0178,
+      "step": 1620
+    },
+    {
+      "epoch": 0.9860859044162129,
+      "grad_norm": 2.1581854820251465,
+      "learning_rate": 2.9005164231991265e-05,
+      "loss": 0.9483,
+      "step": 1630
+    },
+    {
+      "epoch": 0.9921355111917726,
+      "grad_norm": 1.9486889839172363,
+      "learning_rate": 2.8992265419894105e-05,
+      "loss": 1.0735,
+      "step": 1640
+    },
+    {
+      "epoch": 0.9981851179673321,
+      "grad_norm": 1.6225651502609253,
+      "learning_rate": 2.8979286429813473e-05,
+      "loss": 1.0419,
+      "step": 1650
+    },
+    {
+      "epoch": 2.0024183796856105,
+      "eval_loss": 1.0315272808074951,
+      "eval_runtime": 61.5654,
+      "eval_samples_per_second": 3.086,
+      "eval_steps_per_second": 1.543,
+      "step": 1656
+    },
+    {
+      "epoch": 2.007255139056832,
+      "grad_norm": 0.9425010085105896,
+      "learning_rate": 2.5953521207704197e-05,
+      "loss": 1.029,
+      "step": 1660
+    },
+    {
+      "epoch": 2.019347037484885,
+      "grad_norm": 0.900601863861084,
+      "learning_rate": 2.5903985145302143e-05,
+      "loss": 0.8985,
+      "step": 1670
+    },
+    {
+      "epoch": 2.0314389359129383,
+      "grad_norm": 1.3217060565948486,
+      "learning_rate": 2.5854195615561937e-05,
+      "loss": 0.8201,
+      "step": 1680
+    },
+    {
+      "epoch": 2.0435308343409915,
+      "grad_norm": 1.2372468709945679,
+      "learning_rate": 2.5804153775860385e-05,
+      "loss": 0.9249,
+      "step": 1690
+    },
+    {
+      "epoch": 2.0556227327690446,
+      "grad_norm": 1.5520892143249512,
+      "learning_rate": 2.5753860789439343e-05,
+      "loss": 0.9418,
+      "step": 1700
+    },
+    {
+      "epoch": 2.0677146311970978,
+      "grad_norm": 1.664322853088379,
+      "learning_rate": 2.5703317825378677e-05,
+      "loss": 0.8835,
+      "step": 1710
+    },
+    {
+      "epoch": 2.0798065296251513,
+      "grad_norm": 1.084291934967041,
+      "learning_rate": 2.5652526058569055e-05,
+      "loss": 0.9336,
+      "step": 1720
+    },
+    {
+      "epoch": 2.0918984280532045,
+      "grad_norm": 1.3585351705551147,
+      "learning_rate": 2.5601486669684686e-05,
+      "loss": 0.9808,
+      "step": 1730
+    },
+    {
+      "epoch": 2.1039903264812576,
+      "grad_norm": 1.5963308811187744,
+      "learning_rate": 2.5550200845155842e-05,
+      "loss": 1.0655,
+      "step": 1740
+    },
+    {
+      "epoch": 2.1160822249093107,
+      "grad_norm": 1.2812840938568115,
+      "learning_rate": 2.549866977714128e-05,
+      "loss": 1.0266,
+      "step": 1750
+    },
+    {
+      "epoch": 2.128174123337364,
+      "grad_norm": 1.5718134641647339,
+      "learning_rate": 2.5446894663500544e-05,
+      "loss": 0.7775,
+      "step": 1760
+    },
+    {
+      "epoch": 2.140266021765417,
+      "grad_norm": 1.943947434425354,
+      "learning_rate": 2.5394876707766105e-05,
+      "loss": 0.9393,
+      "step": 1770
+    },
+    {
+      "epoch": 2.15235792019347,
+      "grad_norm": 1.304726004600525,
+      "learning_rate": 2.5342617119115404e-05,
+      "loss": 0.8872,
+      "step": 1780
+    },
+    {
+      "epoch": 2.1644498186215237,
+      "grad_norm": 1.2231858968734741,
+      "learning_rate": 2.529011711234272e-05,
+      "loss": 0.9708,
+      "step": 1790
+    },
+    {
+      "epoch": 2.176541717049577,
+      "grad_norm": 1.0328049659729004,
+      "learning_rate": 2.5237377907830947e-05,
+      "loss": 0.8888,
+      "step": 1800
+    },
+    {
+      "epoch": 2.18863361547763,
+      "grad_norm": 1.471614122390747,
+      "learning_rate": 2.5184400731523233e-05,
+      "loss": 0.8714,
+      "step": 1810
+    },
+    {
+      "epoch": 2.200725513905683,
+      "grad_norm": 1.3779953718185425,
+      "learning_rate": 2.5131186814894456e-05,
+      "loss": 0.9406,
+      "step": 1820
+    },
+    {
+      "epoch": 2.2128174123337363,
+      "grad_norm": 1.3803290128707886,
+      "learning_rate": 2.5077737394922627e-05,
+      "loss": 0.8063,
+      "step": 1830
+    },
+    {
+      "epoch": 2.2249093107617894,
+      "grad_norm": 1.749267578125,
+      "learning_rate": 2.5024053714060116e-05,
+      "loss": 0.9785,
+      "step": 1840
+    },
+    {
+      "epoch": 2.2370012091898426,
+      "grad_norm": 2.1147618293762207,
+      "learning_rate": 2.4970137020204773e-05,
+      "loss": 0.9254,
+      "step": 1850
+    },
+    {
+      "epoch": 2.249093107617896,
+      "grad_norm": 1.076285719871521,
+      "learning_rate": 2.4915988566670944e-05,
+      "loss": 0.9499,
+      "step": 1860
+    },
+    {
+      "epoch": 2.2611850060459493,
+      "grad_norm": 1.9593374729156494,
+      "learning_rate": 2.4861609612160297e-05,
+      "loss": 0.9589,
+      "step": 1870
+    },
+    {
+      "epoch": 2.2732769044740024,
+      "grad_norm": 1.759669542312622,
+      "learning_rate": 2.4807001420732592e-05,
+      "loss": 0.9817,
+      "step": 1880
+    },
+    {
+      "epoch": 2.2853688029020556,
+      "grad_norm": 1.2852345705032349,
+      "learning_rate": 2.475216526177629e-05,
+      "loss": 1.1129,
+      "step": 1890
+    },
+    {
+      "epoch": 2.2974607013301087,
+      "grad_norm": 1.3759500980377197,
+      "learning_rate": 2.4697102409979054e-05,
+      "loss": 0.7167,
+      "step": 1900
+    },
+    {
+      "epoch": 2.309552599758162,
+      "grad_norm": 1.2431224584579468,
+      "learning_rate": 2.464181414529809e-05,
+      "loss": 0.8352,
+      "step": 1910
+    },
+    {
+      "epoch": 2.321644498186215,
+      "grad_norm": 1.6261695623397827,
+      "learning_rate": 2.458630175293043e-05,
+      "loss": 0.9874,
+      "step": 1920
+    },
+    {
+      "epoch": 2.3337363966142686,
+      "grad_norm": 1.4598480463027954,
+      "learning_rate": 2.4530566523283032e-05,
+      "loss": 0.9489,
+      "step": 1930
+    },
+    {
+      "epoch": 2.3458282950423217,
+      "grad_norm": 1.9565013647079468,
+      "learning_rate": 2.4474609751942798e-05,
+      "loss": 1.0428,
+      "step": 1940
+    },
+    {
+      "epoch": 2.357920193470375,
+      "grad_norm": 1.8509588241577148,
+      "learning_rate": 2.4418432739646443e-05,
+      "loss": 0.9298,
+      "step": 1950
+    },
+    {
+      "epoch": 2.370012091898428,
+      "grad_norm": 1.634786605834961,
+      "learning_rate": 2.4362036792250275e-05,
+      "loss": 0.9312,
+      "step": 1960
+    },
+    {
+      "epoch": 2.382103990326481,
+      "grad_norm": 1.9874886274337769,
+      "learning_rate": 2.430542322069984e-05,
+      "loss": 1.0114,
+      "step": 1970
+    },
+    {
+      "epoch": 2.3941958887545347,
+      "grad_norm": 1.8052939176559448,
+      "learning_rate": 2.4248593340999423e-05,
+      "loss": 0.8945,
+      "step": 1980
+    },
+    {
+      "epoch": 2.406287787182588,
+      "grad_norm": 1.6896426677703857,
+      "learning_rate": 2.4191548474181494e-05,
+      "loss": 0.9642,
+      "step": 1990
+    },
+    {
+      "epoch": 2.418379685610641,
+      "grad_norm": 1.6607788801193237,
+      "learning_rate": 2.4134289946275963e-05,
+      "loss": 0.9253,
+      "step": 2000
+    },
+    {
+      "epoch": 2.430471584038694,
+      "grad_norm": 2.010333299636841,
+      "learning_rate": 2.4076819088279395e-05,
+      "loss": 0.9328,
+      "step": 2010
+    },
+    {
+      "epoch": 2.4425634824667473,
+      "grad_norm": 1.9691773653030396,
+      "learning_rate": 2.4019137236124027e-05,
+      "loss": 1.0397,
+      "step": 2020
+    },
+    {
+      "epoch": 2.4546553808948004,
+      "grad_norm": 1.8054171800613403,
+      "learning_rate": 2.3961245730646758e-05,
+      "loss": 0.8893,
+      "step": 2030
+    },
+    {
+      "epoch": 2.4667472793228535,
+      "grad_norm": 1.0990610122680664,
+      "learning_rate": 2.3903145917557945e-05,
+      "loss": 0.9129,
+      "step": 2040
+    },
+    {
+      "epoch": 2.478839177750907,
+      "grad_norm": 1.6305997371673584,
+      "learning_rate": 2.3844839147410146e-05,
+      "loss": 0.9494,
+      "step": 2050
+    },
+    {
+      "epoch": 2.4909310761789603,
+      "grad_norm": 2.0180811882019043,
+      "learning_rate": 2.3786326775566705e-05,
+      "loss": 0.8741,
+      "step": 2060
+    },
+    {
+      "epoch": 2.5030229746070134,
+      "grad_norm": 1.390211820602417,
+      "learning_rate": 2.3727610162170262e-05,
+      "loss": 0.9242,
+      "step": 2070
+    },
+    {
+      "epoch": 2.5030229746070134,
+      "eval_loss": 1.027010202407837,
+      "eval_runtime": 62.1115,
+      "eval_samples_per_second": 3.059,
+      "eval_steps_per_second": 1.53,
+      "step": 2070
+    },
+    {
+      "epoch": 2.5151148730350665,
+      "grad_norm": 2.0877339839935303,
+      "learning_rate": 2.3668690672111135e-05,
+      "loss": 0.9463,
+      "step": 2080
+    },
+    {
+      "epoch": 2.5272067714631197,
+      "grad_norm": 1.8376761674880981,
+      "learning_rate": 2.3609569674995575e-05,
+      "loss": 0.9761,
+      "step": 2090
+    },
+    {
+      "epoch": 2.539298669891173,
+      "grad_norm": 1.2929617166519165,
+      "learning_rate": 2.355024854511396e-05,
+      "loss": 0.9552,
+      "step": 2100
+    },
+    {
+      "epoch": 2.551390568319226,
+      "grad_norm": 1.428189992904663,
+      "learning_rate": 2.3490728661408823e-05,
+      "loss": 0.9305,
+      "step": 2110
+    },
+    {
+      "epoch": 2.5634824667472795,
+      "grad_norm": 1.873153567314148,
+      "learning_rate": 2.3431011407442793e-05,
+      "loss": 0.9285,
+      "step": 2120
+    },
+    {
+      "epoch": 2.5755743651753327,
+      "grad_norm": 1.3220646381378174,
+      "learning_rate": 2.3371098171366473e-05,
+      "loss": 0.9198,
+      "step": 2130
+    },
+    {
+      "epoch": 2.587666263603386,
+      "grad_norm": 1.2938573360443115,
+      "learning_rate": 2.3310990345886124e-05,
+      "loss": 0.7408,
+      "step": 2140
+    },
+    {
+      "epoch": 2.599758162031439,
+      "grad_norm": 1.5691282749176025,
+      "learning_rate": 2.325068932823132e-05,
+      "loss": 1.085,
+      "step": 2150
+    },
+    {
+      "epoch": 2.611850060459492,
+      "grad_norm": 1.8494837284088135,
+      "learning_rate": 2.3190196520122468e-05,
+      "loss": 0.9864,
+      "step": 2160
+    },
+    {
+      "epoch": 2.6239419588875452,
+      "grad_norm": 2.0735583305358887,
+      "learning_rate": 2.312951332773821e-05,
+      "loss": 0.9551,
+      "step": 2170
+    },
+    {
+      "epoch": 2.6360338573155984,
+      "grad_norm": 2.0026185512542725,
+      "learning_rate": 2.3068641161682747e-05,
+      "loss": 0.8341,
+      "step": 2180
+    },
+    {
+      "epoch": 2.648125755743652,
+      "grad_norm": 1.0748159885406494,
+      "learning_rate": 2.3007581436953043e-05,
+      "loss": 0.8707,
+      "step": 2190
+    },
+    {
+      "epoch": 2.660217654171705,
+      "grad_norm": 1.2758897542953491,
+      "learning_rate": 2.294633557290594e-05,
+      "loss": 0.9761,
+      "step": 2200
+    },
+    {
+      "epoch": 2.672309552599758,
+      "grad_norm": 1.5994787216186523,
+      "learning_rate": 2.288490499322515e-05,
+      "loss": 0.9284,
+      "step": 2210
+    },
+    {
+      "epoch": 2.6844014510278114,
+      "grad_norm": 1.913944125175476,
+      "learning_rate": 2.282329112588819e-05,
+      "loss": 0.9475,
+      "step": 2220
+    },
+    {
+      "epoch": 2.6964933494558645,
+      "grad_norm": 1.6219775676727295,
+      "learning_rate": 2.2761495403133156e-05,
+      "loss": 0.8979,
+      "step": 2230
+    },
+    {
+      "epoch": 2.7085852478839176,
+      "grad_norm": 2.0113205909729004,
+      "learning_rate": 2.2699519261425437e-05,
+      "loss": 1.0181,
+      "step": 2240
+    },
+    {
+      "epoch": 2.7206771463119708,
+      "grad_norm": 1.56183922290802,
+      "learning_rate": 2.2637364141424347e-05,
+      "loss": 1.027,
+      "step": 2250
+    },
+    {
+      "epoch": 2.7327690447400244,
+      "grad_norm": 1.4773386716842651,
+      "learning_rate": 2.2575031487949607e-05,
+      "loss": 0.9299,
+      "step": 2260
+    },
+    {
+      "epoch": 2.7448609431680775,
+      "grad_norm": 1.8613166809082031,
+      "learning_rate": 2.251252274994778e-05,
+      "loss": 0.9851,
+      "step": 2270
+    },
+    {
+      "epoch": 2.7569528415961306,
+      "grad_norm": 1.6664507389068604,
+      "learning_rate": 2.244983938045858e-05,
+      "loss": 0.8041,
+      "step": 2280
+    },
+    {
+      "epoch": 2.7690447400241838,
+      "grad_norm": 1.3423569202423096,
+      "learning_rate": 2.2386982836581084e-05,
+      "loss": 0.8549,
+      "step": 2290
+    },
+    {
+      "epoch": 2.781136638452237,
+      "grad_norm": 1.7735357284545898,
+      "learning_rate": 2.232395457943989e-05,
+      "loss": 0.997,
+      "step": 2300
+    },
+    {
+      "epoch": 2.79322853688029,
+      "grad_norm": 2.512533664703369,
+      "learning_rate": 2.226075607415114e-05,
+      "loss": 0.8865,
+      "step": 2310
+    },
+    {
+      "epoch": 2.805320435308343,
+      "grad_norm": 2.101529121398926,
+      "learning_rate": 2.2197388789788444e-05,
+      "loss": 0.9071,
+      "step": 2320
+    },
+    {
+      "epoch": 2.8174123337363968,
+      "grad_norm": 1.0863012075424194,
+      "learning_rate": 2.213385419934876e-05,
+      "loss": 0.7935,
+      "step": 2330
+    },
+    {
+      "epoch": 2.82950423216445,
+      "grad_norm": 1.9452869892120361,
+      "learning_rate": 2.2070153779718127e-05,
+      "loss": 0.8955,
+      "step": 2340
+    },
+    {
+      "epoch": 2.841596130592503,
+      "grad_norm": 2.226400136947632,
+      "learning_rate": 2.200628901163737e-05,
+      "loss": 0.8089,
+      "step": 2350
+    },
+    {
+      "epoch": 2.853688029020556,
+      "grad_norm": 1.274936318397522,
+      "learning_rate": 2.1942261379667626e-05,
+      "loss": 0.9588,
+      "step": 2360
+    },
+    {
+      "epoch": 2.8657799274486093,
+      "grad_norm": 1.3687050342559814,
+      "learning_rate": 2.18780723721559e-05,
+      "loss": 0.9048,
+      "step": 2370
+    },
+    {
+      "epoch": 2.877871825876663,
+      "grad_norm": 1.7625335454940796,
+      "learning_rate": 2.1813723481200394e-05,
+      "loss": 1.0745,
+      "step": 2380
+    },
+    {
+      "epoch": 2.8899637243047156,
+      "grad_norm": 1.4267995357513428,
+      "learning_rate": 2.1749216202615907e-05,
+      "loss": 1.0377,
+      "step": 2390
+    },
+    {
+      "epoch": 2.902055622732769,
+      "grad_norm": 1.885882019996643,
+      "learning_rate": 2.1684552035898982e-05,
+      "loss": 0.8572,
+      "step": 2400
+    },
+    {
+      "epoch": 2.9141475211608223,
+      "grad_norm": 1.9667267799377441,
+      "learning_rate": 2.1619732484193113e-05,
+      "loss": 0.8767,
+      "step": 2410
+    },
+    {
+      "epoch": 2.9262394195888755,
+      "grad_norm": 1.962418556213379,
+      "learning_rate": 2.1554759054253758e-05,
+      "loss": 0.9886,
+      "step": 2420
+    },
+    {
+      "epoch": 2.9383313180169286,
+      "grad_norm": 2.5986828804016113,
+      "learning_rate": 2.148963325641335e-05,
+      "loss": 0.9386,
+      "step": 2430
+    },
+    {
+      "epoch": 2.9504232164449817,
+      "grad_norm": 2.477590560913086,
+      "learning_rate": 2.142435660454616e-05,
+      "loss": 0.9533,
+      "step": 2440
+    },
+    {
+      "epoch": 2.9625151148730353,
+      "grad_norm": 1.5716705322265625,
+      "learning_rate": 2.135893061603313e-05,
+      "loss": 0.9436,
+      "step": 2450
+    },
+    {
+      "epoch": 2.974607013301088,
+      "grad_norm": 2.0462253093719482,
+      "learning_rate": 2.129335681172658e-05,
+      "loss": 0.8393,
+      "step": 2460
+    },
+    {
+      "epoch": 2.9866989117291416,
+      "grad_norm": 1.807516098022461,
+      "learning_rate": 2.122763671591487e-05,
+      "loss": 0.8802,
+      "step": 2470
+    },
+    {
+      "epoch": 2.9987908101571947,
+      "grad_norm": 1.7970890998840332,
+      "learning_rate": 2.1161771856286964e-05,
+      "loss": 0.8121,
+      "step": 2480
+    },
+    {
+      "epoch": 3.0024183796856105,
+      "eval_loss": 1.025052547454834,
+      "eval_runtime": 62.1601,
+      "eval_samples_per_second": 3.057,
+      "eval_steps_per_second": 1.528,
+      "step": 2484
+    },
+    {
+      "epoch": 3.0096735187424426,
+      "grad_norm": 1.9102685451507568,
+      "learning_rate": 2.1095763763896915e-05,
+      "loss": 1.0081,
+      "step": 2490
+    },
+    {
+      "epoch": 3.0217654171704957,
+      "grad_norm": 1.673042893409729,
+      "learning_rate": 2.1029613973128274e-05,
+      "loss": 0.9355,
+      "step": 2500
+    },
+    {
+      "epoch": 3.033857315598549,
+      "grad_norm": 3.155738115310669,
+      "learning_rate": 2.0963324021658435e-05,
+      "loss": 0.7849,
+      "step": 2510
+    },
+    {
+      "epoch": 3.045949214026602,
+      "grad_norm": 1.792328953742981,
+      "learning_rate": 2.089689545042286e-05,
+      "loss": 0.8839,
+      "step": 2520
+    },
+    {
+      "epoch": 3.0580411124546556,
+      "grad_norm": 2.182460069656372,
+      "learning_rate": 2.0830329803579312e-05,
+      "loss": 0.8556,
+      "step": 2530
+    },
+    {
+      "epoch": 3.0701330108827087,
+      "grad_norm": 1.5327653884887695,
+      "learning_rate": 2.07636286284719e-05,
+      "loss": 0.9156,
+      "step": 2540
+    },
+    {
+      "epoch": 3.082224909310762,
+      "grad_norm": 1.9906131029129028,
+      "learning_rate": 2.0696793475595162e-05,
+      "loss": 0.78,
+      "step": 2550
+    },
+    {
+      "epoch": 3.094316807738815,
+      "grad_norm": 2.0401036739349365,
+      "learning_rate": 2.0629825898557984e-05,
+      "loss": 0.9573,
+      "step": 2560
+    },
+    {
+      "epoch": 3.106408706166868,
+      "grad_norm": 2.0492191314697266,
+      "learning_rate": 2.056272745404751e-05,
+      "loss": 0.851,
+      "step": 2570
+    },
+    {
+      "epoch": 3.1185006045949213,
+      "grad_norm": 2.2973692417144775,
+      "learning_rate": 2.0495499701792954e-05,
+      "loss": 0.8719,
+      "step": 2580
+    },
+    {
+      "epoch": 3.1305925030229744,
+      "grad_norm": 2.114962577819824,
+      "learning_rate": 2.0428144204529338e-05,
+      "loss": 0.8664,
+      "step": 2590
+    },
+    {
+      "epoch": 3.142684401451028,
+      "grad_norm": 1.581209659576416,
+      "learning_rate": 2.0360662527961144e-05,
+      "loss": 0.8698,
+      "step": 2600
+    },
+    {
+      "epoch": 3.154776299879081,
+      "grad_norm": 2.2733969688415527,
+      "learning_rate": 2.0293056240725978e-05,
+      "loss": 0.8688,
+      "step": 2610
+    },
+    {
+      "epoch": 3.1668681983071343,
+      "grad_norm": 1.9230784177780151,
+      "learning_rate": 2.022532691435803e-05,
+      "loss": 0.9029,
+      "step": 2620
+    },
+    {
+      "epoch": 3.1789600967351874,
+      "grad_norm": 1.8609906435012817,
+      "learning_rate": 2.0157476123251618e-05,
+      "loss": 0.8416,
+      "step": 2630
+    },
+    {
+      "epoch": 3.1910519951632406,
+      "grad_norm": 3.587183713912964,
+      "learning_rate": 2.0089505444624523e-05,
+      "loss": 0.8078,
+      "step": 2640
+    },
+    {
+      "epoch": 3.2031438935912937,
+      "grad_norm": 1.5747348070144653,
+      "learning_rate": 2.0021416458481382e-05,
+      "loss": 0.8362,
+      "step": 2650
+    },
+    {
+      "epoch": 3.215235792019347,
+      "grad_norm": 2.3190407752990723,
+      "learning_rate": 1.9953210747576925e-05,
+      "loss": 0.8644,
+      "step": 2660
+    },
+    {
+      "epoch": 3.2273276904474004,
+      "grad_norm": 2.1347427368164062,
+      "learning_rate": 1.988488989737919e-05,
+      "loss": 0.8559,
+      "step": 2670
+    },
+    {
+      "epoch": 3.2394195888754536,
+      "grad_norm": 1.6797113418579102,
+      "learning_rate": 1.9816455496032678e-05,
+      "loss": 0.6892,
+      "step": 2680
+    },
+    {
+      "epoch": 3.2515114873035067,
+      "grad_norm": 2.676992416381836,
+      "learning_rate": 1.9747909134321433e-05,
+      "loss": 0.9424,
+      "step": 2690
+    },
+    {
+      "epoch": 3.26360338573156,
+      "grad_norm": 2.5450267791748047,
+      "learning_rate": 1.9679252405632056e-05,
+      "loss": 1.0132,
+      "step": 2700
+    },
+    {
+      "epoch": 3.275695284159613,
+      "grad_norm": 1.7200360298156738,
+      "learning_rate": 1.9610486905916672e-05,
+      "loss": 0.7815,
+      "step": 2710
+    },
+    {
+      "epoch": 3.287787182587666,
+      "grad_norm": 1.8882542848587036,
+      "learning_rate": 1.9541614233655832e-05,
+      "loss": 0.7036,
+      "step": 2720
+    },
+    {
+      "epoch": 3.2998790810157193,
+      "grad_norm": 2.0537140369415283,
+      "learning_rate": 1.9472635989821346e-05,
+      "loss": 0.8662,
+      "step": 2730
+    },
+    {
+      "epoch": 3.311970979443773,
+      "grad_norm": 2.3131861686706543,
+      "learning_rate": 1.9403553777839097e-05,
+      "loss": 0.8448,
+      "step": 2740
+    },
+    {
+      "epoch": 3.324062877871826,
+      "grad_norm": 2.300368309020996,
+      "learning_rate": 1.9334369203551722e-05,
+      "loss": 0.8958,
+      "step": 2750
+    },
+    {
+      "epoch": 3.336154776299879,
+      "grad_norm": 2.186077117919922,
+      "learning_rate": 1.926508387518133e-05,
+      "loss": 0.8127,
+      "step": 2760
+    },
+    {
+      "epoch": 3.3482466747279322,
+      "grad_norm": 2.2586021423339844,
+      "learning_rate": 1.9195699403292072e-05,
+      "loss": 0.8263,
+      "step": 2770
+    },
+    {
+      "epoch": 3.3603385731559854,
+      "grad_norm": 2.485656499862671,
+      "learning_rate": 1.912621740075276e-05,
+      "loss": 0.9142,
+      "step": 2780
+    },
+    {
+      "epoch": 3.3724304715840385,
+      "grad_norm": 2.753103494644165,
+      "learning_rate": 1.9056639482699323e-05,
+      "loss": 0.8142,
+      "step": 2790
+    },
+    {
+      "epoch": 3.3845223700120917,
+      "grad_norm": 2.5374977588653564,
+      "learning_rate": 1.8986967266497293e-05,
+      "loss": 0.7762,
+      "step": 2800
+    },
+    {
+      "epoch": 3.3966142684401452,
+      "grad_norm": 2.3932859897613525,
+      "learning_rate": 1.8917202371704184e-05,
+      "loss": 0.8581,
+      "step": 2810
+    },
+    {
+      "epoch": 3.4087061668681984,
+      "grad_norm": 2.2613675594329834,
+      "learning_rate": 1.8847346420031876e-05,
+      "loss": 0.9741,
+      "step": 2820
+    },
+    {
+      "epoch": 3.4207980652962515,
+      "grad_norm": 1.7016990184783936,
+      "learning_rate": 1.877740103530889e-05,
+      "loss": 0.8193,
+      "step": 2830
+    },
+    {
+      "epoch": 3.4328899637243047,
+      "grad_norm": 2.1311845779418945,
+      "learning_rate": 1.8707367843442672e-05,
+      "loss": 0.7398,
+      "step": 2840
+    },
+    {
+      "epoch": 3.444981862152358,
+      "grad_norm": 2.0460548400878906,
+      "learning_rate": 1.8637248472381743e-05,
+      "loss": 0.7596,
+      "step": 2850
+    },
+    {
+      "epoch": 3.457073760580411,
+      "grad_norm": 2.456862688064575,
+      "learning_rate": 1.8567044552077932e-05,
+      "loss": 0.9296,
+      "step": 2860
+    },
+    {
+      "epoch": 3.4691656590084645,
+      "grad_norm": 1.7966513633728027,
+      "learning_rate": 1.8496757714448413e-05,
+      "loss": 0.8596,
+      "step": 2870
+    },
+    {
+      "epoch": 3.4812575574365177,
+      "grad_norm": 2.129179000854492,
+      "learning_rate": 1.8426389593337842e-05,
+      "loss": 0.8115,
+      "step": 2880
+    },
+    {
+      "epoch": 3.493349455864571,
+      "grad_norm": 1.8818175792694092,
+      "learning_rate": 1.835594182448031e-05,
+      "loss": 0.7811,
+      "step": 2890
+    },
+    {
+      "epoch": 3.5030229746070134,
+      "eval_loss": 1.0463489294052124,
+      "eval_runtime": 62.151,
+      "eval_samples_per_second": 3.057,
+      "eval_steps_per_second": 1.529,
+      "step": 2898
+    },
+    {
+      "epoch": 3.505441354292624,
+      "grad_norm": 2.0948948860168457,
+      "learning_rate": 1.828541604546136e-05,
+      "loss": 0.8436,
+      "step": 2900
+    },
+    {
+      "epoch": 3.517533252720677,
+      "grad_norm": 1.82291579246521,
+      "learning_rate": 1.8214813895679916e-05,
+      "loss": 0.7953,
+      "step": 2910
+    },
+    {
+      "epoch": 3.52962515114873,
+      "grad_norm": 2.061556100845337,
+      "learning_rate": 1.8144137016310164e-05,
+      "loss": 0.8741,
+      "step": 2920
+    },
+    {
+      "epoch": 3.541717049576784,
+      "grad_norm": 3.4671363830566406,
+      "learning_rate": 1.8073387050263416e-05,
+      "loss": 0.9057,
+      "step": 2930
+    },
+    {
+      "epoch": 3.5538089480048365,
+      "grad_norm": 3.073554515838623,
+      "learning_rate": 1.80025656421499e-05,
+      "loss": 0.9202,
+      "step": 2940
+    },
+    {
+      "epoch": 3.56590084643289,
+      "grad_norm": 2.2745471000671387,
+      "learning_rate": 1.7931674438240554e-05,
+      "loss": 0.869,
+      "step": 2950
+    },
+    {
+      "epoch": 3.577992744860943,
+      "grad_norm": 2.1967475414276123,
+      "learning_rate": 1.786071508642874e-05,
+      "loss": 0.8464,
+      "step": 2960
+    },
+    {
+      "epoch": 3.5900846432889963,
+      "grad_norm": 3.3642377853393555,
+      "learning_rate": 1.7789689236191942e-05,
+      "loss": 0.7493,
+      "step": 2970
+    },
+    {
+      "epoch": 3.6021765417170495,
+      "grad_norm": 2.1377601623535156,
+      "learning_rate": 1.7718598538553436e-05,
+      "loss": 0.8941,
+      "step": 2980
+    },
+    {
+      "epoch": 3.6142684401451026,
+      "grad_norm": 2.482862710952759,
+      "learning_rate": 1.7647444646043887e-05,
+      "loss": 0.9424,
+      "step": 2990
+    },
+    {
+      "epoch": 3.626360338573156,
+      "grad_norm": 2.38171648979187,
+      "learning_rate": 1.7576229212662955e-05,
+      "loss": 0.7499,
+      "step": 3000
+    },
+    {
+      "epoch": 3.6384522370012093,
+      "grad_norm": 2.8328399658203125,
+      "learning_rate": 1.750495389384085e-05,
+      "loss": 0.8494,
+      "step": 3010
+    },
+    {
+      "epoch": 3.6505441354292625,
+      "grad_norm": 2.6801421642303467,
+      "learning_rate": 1.7433620346399833e-05,
+      "loss": 0.8353,
+      "step": 3020
+    },
+    {
+      "epoch": 3.6626360338573156,
+      "grad_norm": 2.9922213554382324,
+      "learning_rate": 1.7362230228515706e-05,
+      "loss": 0.7786,
+      "step": 3030
+    },
+    {
+      "epoch": 3.6747279322853688,
+      "grad_norm": 1.9845452308654785,
+      "learning_rate": 1.729078519967929e-05,
+      "loss": 0.8261,
+      "step": 3040
+    },
+    {
+      "epoch": 3.686819830713422,
+      "grad_norm": 2.896380662918091,
+      "learning_rate": 1.721928692065781e-05,
+      "loss": 0.8031,
+      "step": 3050
+    },
+    {
+      "epoch": 3.698911729141475,
+      "grad_norm": 2.3239529132843018,
+      "learning_rate": 1.714773705345633e-05,
+      "loss": 0.8602,
+      "step": 3060
+    },
+    {
+      "epoch": 3.7110036275695286,
+      "grad_norm": 1.6386851072311401,
+      "learning_rate": 1.7076137261279105e-05,
+      "loss": 0.7999,
+      "step": 3070
+    },
+    {
+      "epoch": 3.7230955259975818,
+      "grad_norm": 2.6418874263763428,
+      "learning_rate": 1.700448920849089e-05,
+      "loss": 0.8791,
+      "step": 3080
+    },
+    {
+      "epoch": 3.735187424425635,
+      "grad_norm": 2.601285696029663,
+      "learning_rate": 1.6932794560578302e-05,
+      "loss": 0.8067,
+      "step": 3090
+    },
+    {
+      "epoch": 3.747279322853688,
+      "grad_norm": 2.0848560333251953,
+      "learning_rate": 1.6861054984111065e-05,
+      "loss": 0.7878,
+      "step": 3100
+    },
+    {
+      "epoch": 3.759371221281741,
+      "grad_norm": 2.478421449661255,
+      "learning_rate": 1.6789272146703296e-05,
+      "loss": 0.6949,
+      "step": 3110
+    },
+    {
+      "epoch": 3.7714631197097943,
+      "grad_norm": 2.1585545539855957,
+      "learning_rate": 1.6717447716974713e-05,
+      "loss": 0.9221,
+      "step": 3120
+    },
+    {
+      "epoch": 3.7835550181378474,
+      "grad_norm": 1.5959584712982178,
+      "learning_rate": 1.664558336451188e-05,
+      "loss": 0.774,
+      "step": 3130
+    },
+    {
+      "epoch": 3.795646916565901,
+      "grad_norm": 2.674715757369995,
+      "learning_rate": 1.657368075982937e-05,
+      "loss": 0.842,
+      "step": 3140
+    },
+    {
+      "epoch": 3.807738814993954,
+      "grad_norm": 2.4131863117218018,
+      "learning_rate": 1.6501741574330948e-05,
+      "loss": 0.8084,
+      "step": 3150
+    },
+    {
+      "epoch": 3.8198307134220073,
+      "grad_norm": 2.0096678733825684,
+      "learning_rate": 1.6429767480270713e-05,
+      "loss": 0.8851,
+      "step": 3160
+    },
+    {
+      "epoch": 3.8319226118500604,
+      "grad_norm": 2.3412530422210693,
+      "learning_rate": 1.635776015071423e-05,
+      "loss": 0.8997,
+      "step": 3170
+    },
+    {
+      "epoch": 3.8440145102781136,
+      "grad_norm": 2.2029662132263184,
+      "learning_rate": 1.628572125949963e-05,
+      "loss": 0.8425,
+      "step": 3180
+    },
+    {
+      "epoch": 3.8561064087061667,
+      "grad_norm": 2.2945916652679443,
+      "learning_rate": 1.6213652481198715e-05,
+      "loss": 0.8195,
+      "step": 3190
+    },
+    {
+      "epoch": 3.86819830713422,
+      "grad_norm": 3.0929417610168457,
+      "learning_rate": 1.6141555491078012e-05,
+      "loss": 0.78,
+      "step": 3200
+    },
+    {
+      "epoch": 3.8802902055622734,
+      "grad_norm": 3.166088581085205,
+      "learning_rate": 1.6069431965059854e-05,
+      "loss": 1.0325,
+      "step": 3210
+    },
+    {
+      "epoch": 3.8923821039903266,
+      "grad_norm": 2.1898717880249023,
+      "learning_rate": 1.5997283579683403e-05,
+      "loss": 0.7925,
+      "step": 3220
+    },
+    {
+      "epoch": 3.9044740024183797,
+      "grad_norm": 2.5372607707977295,
+      "learning_rate": 1.5925112012065697e-05,
+      "loss": 0.8772,
+      "step": 3230
+    },
+    {
+      "epoch": 3.916565900846433,
+      "grad_norm": 1.7813209295272827,
+      "learning_rate": 1.5852918939862637e-05,
+      "loss": 0.8521,
+      "step": 3240
+    },
+    {
+      "epoch": 3.928657799274486,
+      "grad_norm": 4.24637508392334,
+      "learning_rate": 1.578070604123003e-05,
+      "loss": 0.872,
+      "step": 3250
+    },
+    {
+      "epoch": 3.940749697702539,
+      "grad_norm": 2.9286880493164062,
+      "learning_rate": 1.5708474994784525e-05,
+      "loss": 0.9763,
+      "step": 3260
+    },
+    {
+      "epoch": 3.9528415961305923,
+      "grad_norm": 2.25107479095459,
+      "learning_rate": 1.563622747956466e-05,
+      "loss": 0.7783,
+      "step": 3270
+    },
+    {
+      "epoch": 3.964933494558646,
+      "grad_norm": 3.030318021774292,
+      "learning_rate": 1.5563965174991762e-05,
+      "loss": 0.9837,
+      "step": 3280
+    },
+    {
+      "epoch": 3.977025392986699,
+      "grad_norm": 2.7458813190460205,
+      "learning_rate": 1.5491689760830966e-05,
+      "loss": 0.7787,
+      "step": 3290
+    },
+    {
+      "epoch": 3.989117291414752,
+      "grad_norm": 2.2329845428466797,
+      "learning_rate": 1.5419402917152133e-05,
+      "loss": 0.9339,
+      "step": 3300
+    },
+    {
+      "epoch": 4.0024183796856105,
+      "grad_norm": 1.8471683263778687,
+      "learning_rate": 1.5347106324290817e-05,
+      "loss": 0.8205,
+      "step": 3310
+    },
+    {
+      "epoch": 4.004836759371221,
+      "eval_loss": 1.0431386232376099,
+      "eval_runtime": 62.1184,
+      "eval_samples_per_second": 3.059,
+      "eval_steps_per_second": 1.529,
+      "step": 3312
+    },
+    {
+      "epoch": 4.014510278113664,
+      "grad_norm": 1.8121156692504883,
+      "learning_rate": 1.5274801662809178e-05,
+      "loss": 0.7915,
+      "step": 3320
+    },
+    {
+      "epoch": 4.026602176541717,
+      "grad_norm": 2.4977829456329346,
+      "learning_rate": 1.5202490613456963e-05,
+      "loss": 0.6828,
+      "step": 3330
+    },
+    {
+      "epoch": 4.03869407496977,
+      "grad_norm": 2.696383237838745,
+      "learning_rate": 1.5130174857132367e-05,
+      "loss": 0.7595,
+      "step": 3340
+    },
+    {
+      "epoch": 4.050785973397823,
+      "grad_norm": 2.9162650108337402,
+      "learning_rate": 1.5057856074843042e-05,
+      "loss": 0.8107,
+      "step": 3350
+    },
+    {
+      "epoch": 4.062877871825877,
+      "grad_norm": 2.7059521675109863,
+      "learning_rate": 1.4985535947666946e-05,
+      "loss": 0.8078,
+      "step": 3360
+    },
+    {
+      "epoch": 4.07496977025393,
+      "grad_norm": 2.8959505558013916,
+      "learning_rate": 1.4913216156713333e-05,
+      "loss": 0.8511,
+      "step": 3370
+    },
+    {
+      "epoch": 4.087061668681983,
+      "grad_norm": 2.2846384048461914,
+      "learning_rate": 1.4840898383083603e-05,
+      "loss": 0.7364,
+      "step": 3380
+    },
+    {
+      "epoch": 4.0991535671100365,
+      "grad_norm": 3.3403897285461426,
+      "learning_rate": 1.47685843078323e-05,
+      "loss": 0.9103,
+      "step": 3390
+    },
+    {
+      "epoch": 4.111245465538089,
+      "grad_norm": 1.9266448020935059,
+      "learning_rate": 1.4696275611927962e-05,
+      "loss": 0.6803,
+      "step": 3400
+    },
+    {
+      "epoch": 4.123337363966143,
+      "grad_norm": 2.2451562881469727,
+      "learning_rate": 1.4623973976214117e-05,
+      "loss": 0.7053,
+      "step": 3410
+    },
+    {
+      "epoch": 4.1354292623941955,
+      "grad_norm": 3.214365005493164,
+      "learning_rate": 1.4551681081370149e-05,
+      "loss": 0.7481,
+      "step": 3420
+    },
+    {
+      "epoch": 4.147521160822249,
+      "grad_norm": 2.8689422607421875,
+      "learning_rate": 1.447939860787227e-05,
+      "loss": 0.8524,
+      "step": 3430
+    },
+    {
+      "epoch": 4.159613059250303,
+      "grad_norm": 2.400942087173462,
+      "learning_rate": 1.440712823595444e-05,
+      "loss": 0.8388,
+      "step": 3440
+    },
+    {
+      "epoch": 4.171704957678355,
+      "grad_norm": 2.9148495197296143,
+      "learning_rate": 1.433487164556932e-05,
+      "loss": 0.7776,
+      "step": 3450
+    },
+    {
+      "epoch": 4.183796856106409,
+      "grad_norm": 3.156839370727539,
+      "learning_rate": 1.42626305163492e-05,
+      "loss": 0.7788,
+      "step": 3460
+    },
+    {
+      "epoch": 4.195888754534462,
+      "grad_norm": 3.3853509426116943,
+      "learning_rate": 1.4190406527566984e-05,
+      "loss": 0.768,
+      "step": 3470
+    },
+    {
+      "epoch": 4.207980652962515,
+      "grad_norm": 2.504469871520996,
+      "learning_rate": 1.4118201358097127e-05,
+      "loss": 0.8062,
+      "step": 3480
+    },
+    {
+      "epoch": 4.220072551390568,
+      "grad_norm": 3.1411614418029785,
+      "learning_rate": 1.4046016686376636e-05,
+      "loss": 0.7413,
+      "step": 3490
+    },
+    {
+      "epoch": 4.2321644498186215,
+      "grad_norm": 3.1822478771209717,
+      "learning_rate": 1.3973854190366023e-05,
+      "loss": 0.7172,
+      "step": 3500
+    },
+    {
+      "epoch": 4.244256348246675,
+      "grad_norm": 2.4940245151519775,
+      "learning_rate": 1.3901715547510332e-05,
+      "loss": 0.7663,
+      "step": 3510
+    },
+    {
+      "epoch": 4.256348246674728,
+      "grad_norm": 3.3821640014648438,
+      "learning_rate": 1.3829602434700128e-05,
+      "loss": 0.7442,
+      "step": 3520
+    },
+    {
+      "epoch": 4.268440145102781,
+      "grad_norm": 2.120893955230713,
+      "learning_rate": 1.375751652823251e-05,
+      "loss": 0.7644,
+      "step": 3530
+    },
+    {
+      "epoch": 4.280532043530834,
+      "grad_norm": 3.397862672805786,
+      "learning_rate": 1.368545950377217e-05,
+      "loss": 0.7631,
+      "step": 3540
+    },
+    {
+      "epoch": 4.292623941958888,
+      "grad_norm": 3.241807222366333,
+      "learning_rate": 1.3613433036312415e-05,
+      "loss": 0.8655,
+      "step": 3550
+    },
+    {
+      "epoch": 4.30471584038694,
+      "grad_norm": 2.9539334774017334,
+      "learning_rate": 1.3541438800136251e-05,
+      "loss": 0.7683,
+      "step": 3560
+    },
+    {
+      "epoch": 4.316807738814994,
+      "grad_norm": 3.136444568634033,
+      "learning_rate": 1.346947846877744e-05,
+      "loss": 0.8393,
+      "step": 3570
+    },
+    {
+      "epoch": 4.3288996372430475,
+      "grad_norm": 3.4799044132232666,
+      "learning_rate": 1.3397553714981646e-05,
+      "loss": 0.8738,
+      "step": 3580
+    },
+    {
+      "epoch": 4.3409915356711,
+      "grad_norm": 3.240767240524292,
+      "learning_rate": 1.3325666210667476e-05,
+      "loss": 0.9027,
+      "step": 3590
+    },
+    {
+      "epoch": 4.353083434099154,
+      "grad_norm": 3.1625399589538574,
+      "learning_rate": 1.3253817626887698e-05,
+      "loss": 0.7002,
+      "step": 3600
+    },
+    {
+      "epoch": 4.3651753325272065,
+      "grad_norm": 1.863317608833313,
+      "learning_rate": 1.318200963379032e-05,
+      "loss": 0.7383,
+      "step": 3610
+    },
+    {
+      "epoch": 4.37726723095526,
+      "grad_norm": 3.0041935443878174,
+      "learning_rate": 1.3110243900579846e-05,
+      "loss": 0.7579,
+      "step": 3620
+    },
+    {
+      "epoch": 4.389359129383314,
+      "grad_norm": 2.631523370742798,
+      "learning_rate": 1.303852209547839e-05,
+      "loss": 0.8294,
+      "step": 3630
+    },
+    {
+      "epoch": 4.401451027811366,
+      "grad_norm": 2.0372538566589355,
+      "learning_rate": 1.2966845885686976e-05,
+      "loss": 0.6767,
+      "step": 3640
+    },
+    {
+      "epoch": 4.41354292623942,
+      "grad_norm": 3.182499647140503,
+      "learning_rate": 1.2895216937346716e-05,
+      "loss": 0.7367,
+      "step": 3650
+    },
+    {
+      "epoch": 4.425634824667473,
+      "grad_norm": 3.0209364891052246,
+      "learning_rate": 1.2823636915500135e-05,
+      "loss": 0.7486,
+      "step": 3660
+    },
+    {
+      "epoch": 4.437726723095526,
+      "grad_norm": 3.3213560581207275,
+      "learning_rate": 1.275210748405244e-05,
+      "loss": 0.7267,
+      "step": 3670
+    },
+    {
+      "epoch": 4.449818621523579,
+      "grad_norm": 3.588071346282959,
+      "learning_rate": 1.2680630305732831e-05,
+      "loss": 0.7972,
+      "step": 3680
+    },
+    {
+      "epoch": 4.4619105199516325,
+      "grad_norm": 3.1862246990203857,
+      "learning_rate": 1.2609207042055879e-05,
+      "loss": 0.7063,
+      "step": 3690
+    },
+    {
+      "epoch": 4.474002418379685,
+      "grad_norm": 4.018402099609375,
+      "learning_rate": 1.2537839353282872e-05,
+      "loss": 0.8669,
+      "step": 3700
+    },
+    {
+      "epoch": 4.486094316807739,
+      "grad_norm": 4.134286403656006,
+      "learning_rate": 1.2466528898383259e-05,
+      "loss": 0.8608,
+      "step": 3710
+    },
+    {
+      "epoch": 4.498186215235792,
+      "grad_norm": 2.790292978286743,
+      "learning_rate": 1.2395277334996045e-05,
+      "loss": 0.7505,
+      "step": 3720
+    },
+    {
+      "epoch": 4.505441354292624,
+      "eval_loss": 1.0653315782546997,
+      "eval_runtime": 62.244,
+      "eval_samples_per_second": 3.053,
+      "eval_steps_per_second": 1.526,
+      "step": 3726
+    },
+    {
+      "epoch": 4.510278113663845,
+      "grad_norm": 5.061868667602539,
+      "learning_rate": 1.2324086319391297e-05,
+      "loss": 0.6894,
+      "step": 3730
+    },
+    {
+      "epoch": 4.522370012091899,
+      "grad_norm": 3.3010201454162598,
+      "learning_rate": 1.2252957506431606e-05,
+      "loss": 0.7144,
+      "step": 3740
+    },
+    {
+      "epoch": 4.534461910519951,
+      "grad_norm": 3.672531843185425,
+      "learning_rate": 1.2181892549533657e-05,
+      "loss": 0.7517,
+      "step": 3750
+    },
+    {
+      "epoch": 4.546553808948005,
+      "grad_norm": 2.89050030708313,
+      "learning_rate": 1.2110893100629763e-05,
+      "loss": 0.7112,
+      "step": 3760
+    },
+    {
+      "epoch": 4.5586457073760585,
+      "grad_norm": 3.2547507286071777,
+      "learning_rate": 1.2039960810129491e-05,
+      "loss": 0.8639,
+      "step": 3770
+    },
+    {
+      "epoch": 4.570737605804111,
+      "grad_norm": 3.547781229019165,
+      "learning_rate": 1.1969097326881272e-05,
+      "loss": 0.7602,
+      "step": 3780
+    },
+    {
+      "epoch": 4.582829504232165,
+      "grad_norm": 3.2739574909210205,
+      "learning_rate": 1.189830429813409e-05,
+      "loss": 0.6657,
+      "step": 3790
+    },
+    {
+      "epoch": 4.594921402660217,
+      "grad_norm": 4.094822883605957,
+      "learning_rate": 1.1827583369499199e-05,
+      "loss": 0.8112,
+      "step": 3800
+    },
+    {
+      "epoch": 4.607013301088271,
+      "grad_norm": 3.00482177734375,
+      "learning_rate": 1.1756936184911834e-05,
+      "loss": 0.7148,
+      "step": 3810
+    },
+    {
+      "epoch": 4.619105199516324,
+      "grad_norm": 3.754075288772583,
+      "learning_rate": 1.1686364386593047e-05,
+      "loss": 0.7634,
+      "step": 3820
+    },
+    {
+      "epoch": 4.631197097944377,
+      "grad_norm": 1.9571914672851562,
+      "learning_rate": 1.1615869615011487e-05,
+      "loss": 0.8019,
+      "step": 3830
+    },
+    {
+      "epoch": 4.64328899637243,
+      "grad_norm": 4.519535064697266,
+      "learning_rate": 1.1545453508845303e-05,
+      "loss": 0.7501,
+      "step": 3840
+    },
+    {
+      "epoch": 4.655380894800484,
+      "grad_norm": 3.0004842281341553,
+      "learning_rate": 1.147511770494402e-05,
+      "loss": 0.9003,
+      "step": 3850
+    },
+    {
+      "epoch": 4.667472793228537,
+      "grad_norm": 3.110866069793701,
+      "learning_rate": 1.1404863838290519e-05,
+      "loss": 0.7781,
+      "step": 3860
+    },
+    {
+      "epoch": 4.67956469165659,
+      "grad_norm": 3.056727647781372,
+      "learning_rate": 1.1334693541963009e-05,
+      "loss": 0.782,
+      "step": 3870
+    },
+    {
+      "epoch": 4.691656590084643,
+      "grad_norm": 5.728208065032959,
+      "learning_rate": 1.1264608447097082e-05,
+      "loss": 0.7219,
+      "step": 3880
+    },
+    {
+      "epoch": 4.703748488512696,
+      "grad_norm": 4.19158935546875,
+      "learning_rate": 1.1194610182847785e-05,
+      "loss": 0.7901,
+      "step": 3890
+    },
+    {
+      "epoch": 4.71584038694075,
+      "grad_norm": 4.694254398345947,
+      "learning_rate": 1.1124700376351755e-05,
+      "loss": 0.799,
+      "step": 3900
+    },
+    {
+      "epoch": 4.727932285368803,
+      "grad_norm": 2.9105007648468018,
+      "learning_rate": 1.105488065268939e-05,
+      "loss": 0.7363,
+      "step": 3910
+    },
+    {
+      "epoch": 4.740024183796856,
+      "grad_norm": 2.6782641410827637,
+      "learning_rate": 1.0985152634847087e-05,
+      "loss": 0.749,
+      "step": 3920
+    },
+    {
+      "epoch": 4.7521160822249096,
+      "grad_norm": 4.835862159729004,
+      "learning_rate": 1.0915517943679495e-05,
+      "loss": 0.7918,
+      "step": 3930
+    },
+    {
+      "epoch": 4.764207980652962,
+      "grad_norm": 3.066210985183716,
+      "learning_rate": 1.084597819787186e-05,
+      "loss": 0.76,
+      "step": 3940
+    },
+    {
+      "epoch": 4.776299879081016,
+      "grad_norm": 3.978726863861084,
+      "learning_rate": 1.0776535013902382e-05,
+      "loss": 0.7956,
+      "step": 3950
+    },
+    {
+      "epoch": 4.788391777509069,
+      "grad_norm": 3.4121222496032715,
+      "learning_rate": 1.0707190006004634e-05,
+      "loss": 0.772,
+      "step": 3960
+    },
+    {
+      "epoch": 4.800483675937122,
+      "grad_norm": 4.126001358032227,
+      "learning_rate": 1.0637944786130064e-05,
+      "loss": 0.7623,
+      "step": 3970
+    },
+    {
+      "epoch": 4.812575574365176,
+      "grad_norm": 3.501007556915283,
+      "learning_rate": 1.0568800963910495e-05,
+      "loss": 0.7658,
+      "step": 3980
+    },
+    {
+      "epoch": 4.824667472793228,
+      "grad_norm": 3.7841763496398926,
+      "learning_rate": 1.0499760146620745e-05,
+      "loss": 0.8207,
+      "step": 3990
+    },
+    {
+      "epoch": 4.836759371221282,
+      "grad_norm": 2.9045231342315674,
+      "learning_rate": 1.0430823939141204e-05,
+      "loss": 0.8224,
+      "step": 4000
+    },
+    {
+      "epoch": 4.848851269649335,
+      "grad_norm": 2.3933026790618896,
+      "learning_rate": 1.036199394392061e-05,
+      "loss": 0.894,
+      "step": 4010
+    },
+    {
+      "epoch": 4.860943168077388,
+      "grad_norm": 3.316516876220703,
+      "learning_rate": 1.0293271760938719e-05,
+      "loss": 0.7822,
+      "step": 4020
+    },
+    {
+      "epoch": 4.873035066505441,
+      "grad_norm": 3.8151934146881104,
+      "learning_rate": 1.0224658987669182e-05,
+      "loss": 0.8381,
+      "step": 4030
+    },
+    {
+      "epoch": 4.8851269649334945,
+      "grad_norm": 3.1684491634368896,
+      "learning_rate": 1.0156157219042351e-05,
+      "loss": 0.7151,
+      "step": 4040
+    },
+    {
+      "epoch": 4.897218863361548,
+      "grad_norm": 2.5580480098724365,
+      "learning_rate": 1.0087768047408257e-05,
+      "loss": 0.7844,
+      "step": 4050
+    },
+    {
+      "epoch": 4.909310761789601,
+      "grad_norm": 3.3300466537475586,
+      "learning_rate": 1.0019493062499541e-05,
+      "loss": 0.8607,
+      "step": 4060
+    },
+    {
+      "epoch": 4.921402660217654,
+      "grad_norm": 3.0954511165618896,
+      "learning_rate": 9.951333851394563e-06,
+      "loss": 0.7207,
+      "step": 4070
+    },
+    {
+      "epoch": 4.933494558645707,
+      "grad_norm": 2.838153123855591,
+      "learning_rate": 9.883291998480459e-06,
+      "loss": 0.7848,
+      "step": 4080
+    },
+    {
+      "epoch": 4.945586457073761,
+      "grad_norm": 2.8606677055358887,
+      "learning_rate": 9.815369085416321e-06,
+      "loss": 0.7297,
+      "step": 4090
+    },
+    {
+      "epoch": 4.957678355501814,
+      "grad_norm": 2.995903491973877,
+      "learning_rate": 9.747566691096455e-06,
+      "loss": 0.7815,
+      "step": 4100
+    },
+    {
+      "epoch": 4.969770253929867,
+      "grad_norm": 2.5223402976989746,
+      "learning_rate": 9.679886391613649e-06,
+      "loss": 0.764,
+      "step": 4110
+    },
+    {
+      "epoch": 4.9818621523579205,
+      "grad_norm": 2.362399101257324,
+      "learning_rate": 9.612329760222562e-06,
+      "loss": 0.6716,
+      "step": 4120
+    },
+    {
+      "epoch": 4.993954050785973,
+      "grad_norm": 4.546547889709473,
+      "learning_rate": 9.544898367303126e-06,
+      "loss": 0.9023,
+      "step": 4130
+    },
+    {
+      "epoch": 5.008464328899637,
+      "grad_norm": 2.1150896549224854,
+      "learning_rate": 9.47759378032407e-06,
+      "loss": 0.6997,
+      "step": 4140
+    },
+    {
+      "epoch": 5.008464328899637,
+      "eval_loss": 1.070144534111023,
+      "eval_runtime": 62.259,
+      "eval_samples_per_second": 3.052,
+      "eval_steps_per_second": 1.526,
+      "step": 4140
+    },
+    {
+      "epoch": 5.02055622732769,
+      "grad_norm": 2.231945514678955,
+      "learning_rate": 9.410417563806457e-06,
+      "loss": 0.7199,
+      "step": 4150
+    },
+    {
+      "epoch": 5.032648125755744,
+      "grad_norm": 2.773264169692993,
+      "learning_rate": 9.343371279287346e-06,
+      "loss": 0.7082,
+      "step": 4160
+    },
+    {
+      "epoch": 5.044740024183797,
+      "grad_norm": 4.111907482147217,
+      "learning_rate": 9.276456485283454e-06,
+      "loss": 0.7083,
+      "step": 4170
+    },
+    {
+      "epoch": 5.05683192261185,
+      "grad_norm": 4.149149417877197,
+      "learning_rate": 9.209674737254977e-06,
+      "loss": 0.6057,
+      "step": 4180
+    },
+    {
+      "epoch": 5.0689238210399035,
+      "grad_norm": 3.5627682209014893,
+      "learning_rate": 9.143027587569384e-06,
+      "loss": 0.7398,
+      "step": 4190
+    },
+    {
+      "epoch": 5.081015719467956,
+      "grad_norm": 2.7134649753570557,
+      "learning_rate": 9.076516585465364e-06,
+      "loss": 0.8352,
+      "step": 4200
+    },
+    {
+      "epoch": 5.09310761789601,
+      "grad_norm": 3.4274790287017822,
+      "learning_rate": 9.010143277016797e-06,
+      "loss": 0.789,
+      "step": 4210
+    },
+    {
+      "epoch": 5.105199516324063,
+      "grad_norm": 2.505985736846924,
+      "learning_rate": 8.943909205096825e-06,
+      "loss": 0.7339,
+      "step": 4220
+    },
+    {
+      "epoch": 5.117291414752116,
+      "grad_norm": 2.784248113632202,
+      "learning_rate": 8.877815909341995e-06,
+      "loss": 0.6168,
+      "step": 4230
+    },
+    {
+      "epoch": 5.12938331318017,
+      "grad_norm": 3.0708258152008057,
+      "learning_rate": 8.811864926116423e-06,
+      "loss": 0.681,
+      "step": 4240
+    },
+    {
+      "epoch": 5.141475211608222,
+      "grad_norm": 2.7348103523254395,
+      "learning_rate": 8.746057788476165e-06,
+      "loss": 0.6212,
+      "step": 4250
+    },
+    {
+      "epoch": 5.153567110036276,
+      "grad_norm": 3.076500177383423,
+      "learning_rate": 8.680396026133482e-06,
+      "loss": 0.6856,
+      "step": 4260
+    },
+    {
+      "epoch": 5.165659008464329,
+      "grad_norm": 5.6594085693359375,
+      "learning_rate": 8.61488116542136e-06,
+      "loss": 0.8085,
+      "step": 4270
+    },
+    {
+      "epoch": 5.177750906892382,
+      "grad_norm": 5.323330402374268,
+      "learning_rate": 8.549514729257991e-06,
+      "loss": 0.6922,
+      "step": 4280
+    },
+    {
+      "epoch": 5.189842805320436,
+      "grad_norm": 3.687112331390381,
+      "learning_rate": 8.484298237111392e-06,
+      "loss": 0.8099,
+      "step": 4290
+    },
+    {
+      "epoch": 5.201934703748488,
+      "grad_norm": 3.3872568607330322,
+      "learning_rate": 8.419233204964046e-06,
+      "loss": 0.682,
+      "step": 4300
+    },
+    {
+      "epoch": 5.214026602176542,
+      "grad_norm": 5.219661235809326,
+      "learning_rate": 8.354321145277717e-06,
+      "loss": 0.8211,
+      "step": 4310
+    },
+    {
+      "epoch": 5.226118500604595,
+      "grad_norm": 2.866403341293335,
+      "learning_rate": 8.289563566958256e-06,
+      "loss": 0.6943,
+      "step": 4320
+    },
+    {
+      "epoch": 5.238210399032648,
+      "grad_norm": 5.183192729949951,
+      "learning_rate": 8.224961975320539e-06,
+      "loss": 0.5939,
+      "step": 4330
+    },
+    {
+      "epoch": 5.250302297460701,
+      "grad_norm": 3.7528531551361084,
+      "learning_rate": 8.16051787205345e-06,
+      "loss": 0.7301,
+      "step": 4340
+    },
+    {
+      "epoch": 5.262394195888755,
+      "grad_norm": 3.930178642272949,
+      "learning_rate": 8.096232755185033e-06,
+      "loss": 0.7815,
+      "step": 4350
+    },
+    {
+      "epoch": 5.274486094316808,
+      "grad_norm": 2.9726829528808594,
+      "learning_rate": 8.03210811904761e-06,
+      "loss": 0.6982,
+      "step": 4360
+    },
+    {
+      "epoch": 5.286577992744861,
+      "grad_norm": 5.79806661605835,
+      "learning_rate": 7.968145454243052e-06,
+      "loss": 0.7827,
+      "step": 4370
+    },
+    {
+      "epoch": 5.298669891172914,
+      "grad_norm": 4.123749256134033,
+      "learning_rate": 7.904346247608172e-06,
+      "loss": 0.6476,
+      "step": 4380
+    },
+    {
+      "epoch": 5.310761789600967,
+      "grad_norm": 2.9389238357543945,
+      "learning_rate": 7.840711982180118e-06,
+      "loss": 0.7254,
+      "step": 4390
+    },
+    {
+      "epoch": 5.322853688029021,
+      "grad_norm": 3.596727132797241,
+      "learning_rate": 7.77724413716193e-06,
+      "loss": 0.716,
+      "step": 4400
+    },
+    {
+      "epoch": 5.334945586457073,
+      "grad_norm": 2.72917103767395,
+      "learning_rate": 7.713944187888116e-06,
+      "loss": 0.6552,
+      "step": 4410
+    },
+    {
+      "epoch": 5.347037484885127,
+      "grad_norm": 3.597003221511841,
+      "learning_rate": 7.650813605790409e-06,
+      "loss": 0.7794,
+      "step": 4420
+    },
+    {
+      "epoch": 5.359129383313181,
+      "grad_norm": 3.1855111122131348,
+      "learning_rate": 7.587853858363524e-06,
+      "loss": 0.716,
+      "step": 4430
+    },
+    {
+      "epoch": 5.371221281741233,
+      "grad_norm": 4.469402313232422,
+      "learning_rate": 7.525066409131065e-06,
+      "loss": 0.746,
+      "step": 4440
+    },
+    {
+      "epoch": 5.383313180169287,
+      "grad_norm": 2.5034403800964355,
+      "learning_rate": 7.462452717611488e-06,
+      "loss": 0.8737,
+      "step": 4450
+    },
+    {
+      "epoch": 5.3954050785973395,
+      "grad_norm": 4.369159698486328,
+      "learning_rate": 7.40001423928419e-06,
+      "loss": 0.7226,
+      "step": 4460
+    },
+    {
+      "epoch": 5.407496977025393,
+      "grad_norm": 3.4538373947143555,
+      "learning_rate": 7.337752425555673e-06,
+      "loss": 0.7042,
+      "step": 4470
+    },
+    {
+      "epoch": 5.419588875453446,
+      "grad_norm": 3.6851110458374023,
+      "learning_rate": 7.275668723725802e-06,
+      "loss": 0.7007,
+      "step": 4480
+    },
+    {
+      "epoch": 5.431680773881499,
+      "grad_norm": 3.3171651363372803,
+      "learning_rate": 7.213764576954148e-06,
+      "loss": 0.6816,
+      "step": 4490
+    },
+    {
+      "epoch": 5.443772672309553,
+      "grad_norm": 3.6243672370910645,
+      "learning_rate": 7.152041424226469e-06,
+      "loss": 0.5568,
+      "step": 4500
+    },
+    {
+      "epoch": 5.455864570737606,
+      "grad_norm": 3.998467206954956,
+      "learning_rate": 7.09050070032126e-06,
+      "loss": 0.673,
+      "step": 4510
+    },
+    {
+      "epoch": 5.467956469165659,
+      "grad_norm": 3.8511223793029785,
+      "learning_rate": 7.029143835776353e-06,
+      "loss": 0.6581,
+      "step": 4520
+    },
+    {
+      "epoch": 5.480048367593712,
+      "grad_norm": 3.5216002464294434,
+      "learning_rate": 6.967972256855732e-06,
+      "loss": 0.6585,
+      "step": 4530
+    },
+    {
+      "epoch": 5.4921402660217655,
+      "grad_norm": 3.133366346359253,
+      "learning_rate": 6.906987385516311e-06,
+      "loss": 0.6854,
+      "step": 4540
+    },
+    {
+      "epoch": 5.504232164449819,
+      "grad_norm": 3.453603506088257,
+      "learning_rate": 6.846190639374957e-06,
+      "loss": 0.78,
+      "step": 4550
+    },
+    {
+      "epoch": 5.50906892382104,
+      "eval_loss": 1.0946567058563232,
+      "eval_runtime": 62.2311,
+      "eval_samples_per_second": 3.053,
+      "eval_steps_per_second": 1.527,
+      "step": 4554
+    },
+    {
+      "epoch": 5.516324062877872,
+      "grad_norm": 3.6046810150146484,
+      "learning_rate": 6.785583431675454e-06,
+      "loss": 0.7816,
+      "step": 4560
+    },
+    {
+      "epoch": 5.528415961305925,
+      "grad_norm": 4.014889240264893,
+      "learning_rate": 6.725167171255721e-06,
+      "loss": 0.7841,
+      "step": 4570
+    },
+    {
+      "epoch": 5.540507859733978,
+      "grad_norm": 2.989539384841919,
+      "learning_rate": 6.664943262515004e-06,
+      "loss": 0.6477,
+      "step": 4580
+    },
+    {
+      "epoch": 5.552599758162032,
+      "grad_norm": 3.522326707839966,
+      "learning_rate": 6.604913105381298e-06,
+      "loss": 0.7645,
+      "step": 4590
+    },
+    {
+      "epoch": 5.564691656590084,
+      "grad_norm": 3.36456561088562,
+      "learning_rate": 6.545078095278736e-06,
+      "loss": 0.6827,
+      "step": 4600
+    },
+    {
+      "epoch": 5.576783555018138,
+      "grad_norm": 3.4892289638519287,
+      "learning_rate": 6.485439623095195e-06,
+      "loss": 0.6528,
+      "step": 4610
+    },
+    {
+      "epoch": 5.588875453446191,
+      "grad_norm": 2.6480884552001953,
+      "learning_rate": 6.425999075149952e-06,
+      "loss": 0.7137,
+      "step": 4620
+    },
+    {
+      "epoch": 5.600967351874244,
+      "grad_norm": 2.971666097640991,
+      "learning_rate": 6.366757833161464e-06,
+      "loss": 0.7081,
+      "step": 4630
+    },
+    {
+      "epoch": 5.613059250302298,
+      "grad_norm": 2.607954263687134,
+      "learning_rate": 6.30771727421523e-06,
+      "loss": 0.6816,
+      "step": 4640
+    },
+    {
+      "epoch": 5.6251511487303505,
+      "grad_norm": 3.269277811050415,
+      "learning_rate": 6.248878770731804e-06,
+      "loss": 0.7183,
+      "step": 4650
+    },
+    {
+      "epoch": 5.637243047158404,
+      "grad_norm": 4.291733741760254,
+      "learning_rate": 6.190243690434882e-06,
+      "loss": 0.7328,
+      "step": 4660
+    },
+    {
+      "epoch": 5.649334945586457,
+      "grad_norm": 4.875344276428223,
+      "learning_rate": 6.131813396319514e-06,
+      "loss": 0.7915,
+      "step": 4670
+    },
+    {
+      "epoch": 5.66142684401451,
+      "grad_norm": 2.9836928844451904,
+      "learning_rate": 6.073589246620411e-06,
+      "loss": 0.728,
+      "step": 4680
+    },
+    {
+      "epoch": 5.673518742442564,
+      "grad_norm": 3.9223244190216064,
+      "learning_rate": 6.015572594780369e-06,
+      "loss": 0.7083,
+      "step": 4690
+    },
+    {
+      "epoch": 5.685610640870617,
+      "grad_norm": 4.437389850616455,
+      "learning_rate": 5.957764789418833e-06,
+      "loss": 0.7338,
+      "step": 4700
+    },
+    {
+      "epoch": 5.69770253929867,
+      "grad_norm": 3.038198709487915,
+      "learning_rate": 5.900167174300521e-06,
+      "loss": 0.6842,
+      "step": 4710
+    },
+    {
+      "epoch": 5.709794437726723,
+      "grad_norm": 3.954749345779419,
+      "learning_rate": 5.842781088304205e-06,
+      "loss": 0.7737,
+      "step": 4720
+    },
+    {
+      "epoch": 5.7218863361547765,
+      "grad_norm": 2.9807608127593994,
+      "learning_rate": 5.785607865391567e-06,
+      "loss": 0.6652,
+      "step": 4730
+    },
+    {
+      "epoch": 5.733978234582829,
+      "grad_norm": 3.8414597511291504,
+      "learning_rate": 5.728648834576219e-06,
+      "loss": 0.8281,
+      "step": 4740
+    },
+    {
+      "epoch": 5.746070133010883,
+      "grad_norm": 4.835539817810059,
+      "learning_rate": 5.671905319892789e-06,
+      "loss": 0.6251,
+      "step": 4750
+    },
+    {
+      "epoch": 5.7581620314389355,
+      "grad_norm": 4.228201389312744,
+      "learning_rate": 5.6153786403661555e-06,
+      "loss": 0.684,
+      "step": 4760
+    },
+    {
+      "epoch": 5.770253929866989,
+      "grad_norm": 5.972825050354004,
+      "learning_rate": 5.55907010998076e-06,
+      "loss": 0.6215,
+      "step": 4770
+    },
+    {
+      "epoch": 5.782345828295043,
+      "grad_norm": 5.1888322830200195,
+      "learning_rate": 5.502981037650113e-06,
+      "loss": 0.671,
+      "step": 4780
+    },
+    {
+      "epoch": 5.794437726723095,
+      "grad_norm": 2.67165470123291,
+      "learning_rate": 5.4471127271863235e-06,
+      "loss": 0.7918,
+      "step": 4790
+    },
+    {
+      "epoch": 5.806529625151149,
+      "grad_norm": 3.7494444847106934,
+      "learning_rate": 5.3914664772697955e-06,
+      "loss": 0.7763,
+      "step": 4800
+    },
+    {
+      "epoch": 5.818621523579202,
+      "grad_norm": 2.512258768081665,
+      "learning_rate": 5.336043581419064e-06,
+      "loss": 0.6537,
+      "step": 4810
+    },
+    {
+      "epoch": 5.830713422007255,
+      "grad_norm": 4.046555042266846,
+      "learning_rate": 5.280845327960708e-06,
+      "loss": 0.713,
+      "step": 4820
+    },
+    {
+      "epoch": 5.842805320435309,
+      "grad_norm": 3.8291842937469482,
+      "learning_rate": 5.2258729999994135e-06,
+      "loss": 0.6669,
+      "step": 4830
+    },
+    {
+      "epoch": 5.8548972188633615,
+      "grad_norm": 4.035644054412842,
+      "learning_rate": 5.171127875388126e-06,
+      "loss": 0.7758,
+      "step": 4840
+    },
+    {
+      "epoch": 5.866989117291415,
+      "grad_norm": 4.429537296295166,
+      "learning_rate": 5.116611226698376e-06,
+      "loss": 0.645,
+      "step": 4850
+    },
+    {
+      "epoch": 5.879081015719468,
+      "grad_norm": 4.3502888679504395,
+      "learning_rate": 5.0623243211906775e-06,
+      "loss": 0.786,
+      "step": 4860
+    },
+    {
+      "epoch": 5.891172914147521,
+      "grad_norm": 3.3320226669311523,
+      "learning_rate": 5.008268420785084e-06,
+      "loss": 0.7135,
+      "step": 4870
+    },
+    {
+      "epoch": 5.903264812575574,
+      "grad_norm": 4.4856414794921875,
+      "learning_rate": 4.954444782031824e-06,
+      "loss": 0.7089,
+      "step": 4880
+    },
+    {
+      "epoch": 5.915356711003628,
+      "grad_norm": 3.6500167846679688,
+      "learning_rate": 4.900854656082139e-06,
+      "loss": 0.834,
+      "step": 4890
+    },
+    {
+      "epoch": 5.92744860943168,
+      "grad_norm": 3.233738899230957,
+      "learning_rate": 4.847499288659163e-06,
+      "loss": 0.6637,
+      "step": 4900
+    },
+    {
+      "epoch": 5.939540507859734,
+      "grad_norm": 5.101102352142334,
+      "learning_rate": 4.7943799200289865e-06,
+      "loss": 0.6458,
+      "step": 4910
+    },
+    {
+      "epoch": 5.9516324062877874,
+      "grad_norm": 3.624640941619873,
+      "learning_rate": 4.7414977849718e-06,
+      "loss": 0.72,
+      "step": 4920
+    },
+    {
+      "epoch": 5.96372430471584,
+      "grad_norm": 4.193915843963623,
+      "learning_rate": 4.688854112753217e-06,
+      "loss": 0.81,
+      "step": 4930
+    },
+    {
+      "epoch": 5.975816203143894,
+      "grad_norm": 3.672713279724121,
+      "learning_rate": 4.636450127095709e-06,
+      "loss": 0.7575,
+      "step": 4940
+    },
+    {
+      "epoch": 5.987908101571946,
+      "grad_norm": 4.2874627113342285,
+      "learning_rate": 4.5842870461501104e-06,
+      "loss": 0.7376,
+      "step": 4950
+    },
+    {
+      "epoch": 6.0,
+      "grad_norm": 3.7344837188720703,
+      "learning_rate": 4.532366082467356e-06,
+      "loss": 0.6445,
+      "step": 4960
+    },
+    {
+      "epoch": 6.008464328899637,
+      "eval_loss": 1.1057277917861938,
+      "eval_runtime": 62.2006,
+      "eval_samples_per_second": 3.055,
+      "eval_steps_per_second": 1.527,
+      "step": 4968
+    },
+    {
+      "epoch": 6.010882708585248,
+      "grad_norm": 3.557671546936035,
+      "learning_rate": 4.480688442970244e-06,
+      "loss": 0.7159,
+      "step": 4970
+    },
+    {
+      "epoch": 6.0229746070133015,
+      "grad_norm": 3.6378707885742188,
+      "learning_rate": 4.429255328925449e-06,
+      "loss": 0.6371,
+      "step": 4980
+    },
+    {
+      "epoch": 6.035066505441354,
+      "grad_norm": 5.355988502502441,
+      "learning_rate": 4.378067935915519e-06,
+      "loss": 0.6781,
+      "step": 4990
+    },
+    {
+      "epoch": 6.047158403869408,
+      "grad_norm": 3.7519631385803223,
+      "learning_rate": 4.327127453811147e-06,
+      "loss": 0.7401,
+      "step": 5000
+    },
+    {
+      "epoch": 6.05925030229746,
+      "grad_norm": 4.124637126922607,
+      "learning_rate": 4.2764350667434605e-06,
+      "loss": 0.7264,
+      "step": 5010
+    },
+    {
+      "epoch": 6.071342200725514,
+      "grad_norm": 3.4475553035736084,
+      "learning_rate": 4.225991953076558e-06,
+      "loss": 0.6487,
+      "step": 5020
+    },
+    {
+      "epoch": 6.083434099153567,
+      "grad_norm": 4.422087669372559,
+      "learning_rate": 4.1757992853800516e-06,
+      "loss": 0.7784,
+      "step": 5030
+    },
+    {
+      "epoch": 6.09552599758162,
+      "grad_norm": 4.515448570251465,
+      "learning_rate": 4.125858230401853e-06,
+      "loss": 0.6513,
+      "step": 5040
+    },
+    {
+      "epoch": 6.107617896009674,
+      "grad_norm": 3.828434944152832,
+      "learning_rate": 4.076169949041037e-06,
+      "loss": 0.6473,
+      "step": 5050
+    },
+    {
+      "epoch": 6.119709794437727,
+      "grad_norm": 3.175715684890747,
+      "learning_rate": 4.02673559632086e-06,
+      "loss": 0.6962,
+      "step": 5060
+    },
+    {
+      "epoch": 6.13180169286578,
+      "grad_norm": 3.969517946243286,
+      "learning_rate": 3.977556321361913e-06,
+      "loss": 0.6401,
+      "step": 5070
+    },
+    {
+      "epoch": 6.143893591293833,
+      "grad_norm": 4.145517349243164,
+      "learning_rate": 3.928633267355388e-06,
+      "loss": 0.6098,
+      "step": 5080
+    },
+    {
+      "epoch": 6.155985489721886,
+      "grad_norm": 3.487616539001465,
+      "learning_rate": 3.879967571536543e-06,
+      "loss": 0.6851,
+      "step": 5090
+    },
+    {
+      "epoch": 6.168077388149939,
+      "grad_norm": 4.306605815887451,
+      "learning_rate": 3.831560365158237e-06,
+      "loss": 0.5277,
+      "step": 5100
+    },
+    {
+      "epoch": 6.180169286577993,
+      "grad_norm": 4.123151779174805,
+      "learning_rate": 3.783412773464647e-06,
+      "loss": 0.6657,
+      "step": 5110
+    },
+    {
+      "epoch": 6.192261185006046,
+      "grad_norm": 4.117605209350586,
+      "learning_rate": 3.735525915665094e-06,
+      "loss": 0.7622,
+      "step": 5120
+    },
+    {
+      "epoch": 6.204353083434099,
+      "grad_norm": 3.730079174041748,
+      "learning_rate": 3.687900904908053e-06,
+      "loss": 0.6035,
+      "step": 5130
+    },
+    {
+      "epoch": 6.2164449818621526,
+      "grad_norm": 4.194562911987305,
+      "learning_rate": 3.6405388482552564e-06,
+      "loss": 0.6141,
+      "step": 5140
+    },
+    {
+      "epoch": 6.228536880290205,
+      "grad_norm": 4.338625431060791,
+      "learning_rate": 3.593440846655976e-06,
+      "loss": 0.6479,
+      "step": 5150
+    },
+    {
+      "epoch": 6.240628778718259,
+      "grad_norm": 3.5001437664031982,
+      "learning_rate": 3.5466079949214046e-06,
+      "loss": 0.6212,
+      "step": 5160
+    },
+    {
+      "epoch": 6.252720677146312,
+      "grad_norm": 5.18671989440918,
+      "learning_rate": 3.50004138169924e-06,
+      "loss": 0.6163,
+      "step": 5170
+    },
+    {
+      "epoch": 6.264812575574365,
+      "grad_norm": 3.782536029815674,
+      "learning_rate": 3.4537420894483555e-06,
+      "loss": 0.6026,
+      "step": 5180
+    },
+    {
+      "epoch": 6.276904474002419,
+      "grad_norm": 3.839702844619751,
+      "learning_rate": 3.407711194413655e-06,
+      "loss": 0.6264,
+      "step": 5190
+    },
+    {
+      "epoch": 6.288996372430471,
+      "grad_norm": 3.8614888191223145,
+      "learning_rate": 3.3619497666010256e-06,
+      "loss": 0.6775,
+      "step": 5200
+    },
+    {
+      "epoch": 6.301088270858525,
+      "grad_norm": 2.7845358848571777,
+      "learning_rate": 3.3164588697525084e-06,
+      "loss": 0.5879,
+      "step": 5210
+    },
+    {
+      "epoch": 6.313180169286578,
+      "grad_norm": 3.5601367950439453,
+      "learning_rate": 3.271239561321538e-06,
+      "loss": 0.6745,
+      "step": 5220
+    },
+    {
+      "epoch": 6.325272067714631,
+      "grad_norm": 4.521182537078857,
+      "learning_rate": 3.2262928924483625e-06,
+      "loss": 0.7594,
+      "step": 5230
+    },
+    {
+      "epoch": 6.337363966142684,
+      "grad_norm": 3.8560290336608887,
+      "learning_rate": 3.18161990793563e-06,
+      "loss": 0.6327,
+      "step": 5240
+    },
+    {
+      "epoch": 6.3494558645707375,
+      "grad_norm": 4.2319207191467285,
+      "learning_rate": 3.1372216462240887e-06,
+      "loss": 0.7454,
+      "step": 5250
+    },
+    {
+      "epoch": 6.361547762998791,
+      "grad_norm": 4.699012279510498,
+      "learning_rate": 3.0930991393684505e-06,
+      "loss": 0.688,
+      "step": 5260
+    },
+    {
+      "epoch": 6.373639661426844,
+      "grad_norm": 5.492476940155029,
+      "learning_rate": 3.0492534130133874e-06,
+      "loss": 0.6861,
+      "step": 5270
+    },
+    {
+      "epoch": 6.385731559854897,
+      "grad_norm": 4.874731063842773,
+      "learning_rate": 3.0056854863697175e-06,
+      "loss": 0.5878,
+      "step": 5280
+    },
+    {
+      "epoch": 6.39782345828295,
+      "grad_norm": 4.2611775398254395,
+      "learning_rate": 2.962396372190688e-06,
+      "loss": 0.7257,
+      "step": 5290
+    },
+    {
+      "epoch": 6.409915356711004,
+      "grad_norm": 3.473050594329834,
+      "learning_rate": 2.9193870767484502e-06,
+      "loss": 0.6894,
+      "step": 5300
+    },
+    {
+      "epoch": 6.422007255139057,
+      "grad_norm": 3.451538562774658,
+      "learning_rate": 2.876658599810651e-06,
+      "loss": 0.662,
+      "step": 5310
+    },
+    {
+      "epoch": 6.43409915356711,
+      "grad_norm": 7.277347087860107,
+      "learning_rate": 2.8342119346172102e-06,
+      "loss": 0.7509,
+      "step": 5320
+    },
+    {
+      "epoch": 6.4461910519951635,
+      "grad_norm": 3.73040509223938,
+      "learning_rate": 2.7920480678572237e-06,
+      "loss": 0.6465,
+      "step": 5330
+    },
+    {
+      "epoch": 6.458282950423216,
+      "grad_norm": 3.874638319015503,
+      "learning_rate": 2.7501679796460332e-06,
+      "loss": 0.7057,
+      "step": 5340
+    },
+    {
+      "epoch": 6.47037484885127,
+      "grad_norm": 4.644408226013184,
+      "learning_rate": 2.7085726435024267e-06,
+      "loss": 0.7528,
+      "step": 5350
+    },
+    {
+      "epoch": 6.4824667472793225,
+      "grad_norm": 4.328802585601807,
+      "learning_rate": 2.667263026326024e-06,
+      "loss": 0.6572,
+      "step": 5360
+    },
+    {
+      "epoch": 6.494558645707376,
+      "grad_norm": 5.580451488494873,
+      "learning_rate": 2.626240088374817e-06,
+      "loss": 0.7067,
+      "step": 5370
+    },
+    {
+      "epoch": 6.50665054413543,
+      "grad_norm": 4.495114803314209,
+      "learning_rate": 2.5855047832427988e-06,
+      "loss": 0.6848,
+      "step": 5380
+    },
+    {
+      "epoch": 6.50906892382104,
+      "eval_loss": 1.1272510290145874,
+      "eval_runtime": 62.2526,
+      "eval_samples_per_second": 3.052,
+      "eval_steps_per_second": 1.526,
+      "step": 5382
+    },
+    {
+      "epoch": 6.518742442563482,
+      "grad_norm": 5.930830478668213,
+      "learning_rate": 2.5450580578378484e-06,
+      "loss": 0.6359,
+      "step": 5390
+    },
+    {
+      "epoch": 6.530834340991536,
+      "grad_norm": 4.322328567504883,
+      "learning_rate": 2.5049008523596778e-06,
+      "loss": 0.6232,
+      "step": 5400
+    },
+    {
+      "epoch": 6.542926239419589,
+      "grad_norm": 3.9086663722991943,
+      "learning_rate": 2.4650341002780264e-06,
+      "loss": 0.6725,
+      "step": 5410
+    },
+    {
+      "epoch": 6.555018137847642,
+      "grad_norm": 3.7998921871185303,
+      "learning_rate": 2.4254587283109014e-06,
+      "loss": 0.6814,
+      "step": 5420
+    },
+    {
+      "epoch": 6.567110036275695,
+      "grad_norm": 3.609553575515747,
+      "learning_rate": 2.386175656403085e-06,
+      "loss": 0.599,
+      "step": 5430
+    },
+    {
+      "epoch": 6.5792019347037485,
+      "grad_norm": 4.122236251831055,
+      "learning_rate": 2.347185797704726e-06,
+      "loss": 0.6335,
+      "step": 5440
+    },
+    {
+      "epoch": 6.591293833131802,
+      "grad_norm": 3.8587522506713867,
+      "learning_rate": 2.3084900585501273e-06,
+      "loss": 0.6883,
+      "step": 5450
+    },
+    {
+      "epoch": 6.603385731559855,
+      "grad_norm": 6.455262184143066,
+      "learning_rate": 2.2700893384366516e-06,
+      "loss": 0.6836,
+      "step": 5460
+    },
+    {
+      "epoch": 6.615477629987908,
+      "grad_norm": 4.4655442237854,
+      "learning_rate": 2.231984530003845e-06,
+      "loss": 0.6687,
+      "step": 5470
+    },
+    {
+      "epoch": 6.627569528415961,
+      "grad_norm": 5.083980560302734,
+      "learning_rate": 2.1941765190126718e-06,
+      "loss": 0.6486,
+      "step": 5480
+    },
+    {
+      "epoch": 6.639661426844015,
+      "grad_norm": 3.8917980194091797,
+      "learning_rate": 2.1566661843249203e-06,
+      "loss": 0.6143,
+      "step": 5490
+    },
+    {
+      "epoch": 6.651753325272068,
+      "grad_norm": 5.429650783538818,
+      "learning_rate": 2.119454397882784e-06,
+      "loss": 0.7289,
+      "step": 5500
+    },
+    {
+      "epoch": 6.663845223700121,
+      "grad_norm": 5.140111446380615,
+      "learning_rate": 2.082542024688578e-06,
+      "loss": 0.8017,
+      "step": 5510
+    },
+    {
+      "epoch": 6.6759371221281745,
+      "grad_norm": 5.339375972747803,
+      "learning_rate": 2.045929922784649e-06,
+      "loss": 0.7102,
+      "step": 5520
+    },
+    {
+      "epoch": 6.688029020556227,
+      "grad_norm": 3.4869794845581055,
+      "learning_rate": 2.0096189432334194e-06,
+      "loss": 0.6714,
+      "step": 5530
+    },
+    {
+      "epoch": 6.700120918984281,
+      "grad_norm": 4.189328670501709,
+      "learning_rate": 1.9736099300976124e-06,
+      "loss": 0.7404,
+      "step": 5540
+    },
+    {
+      "epoch": 6.712212817412333,
+      "grad_norm": 4.462172031402588,
+      "learning_rate": 1.937903720420609e-06,
+      "loss": 0.6742,
+      "step": 5550
+    },
+    {
+      "epoch": 6.724304715840387,
+      "grad_norm": 5.784437656402588,
+      "learning_rate": 1.9025011442070245e-06,
+      "loss": 0.6307,
+      "step": 5560
+    },
+    {
+      "epoch": 6.73639661426844,
+      "grad_norm": 4.614073276519775,
+      "learning_rate": 1.8674030244033896e-06,
+      "loss": 0.5992,
+      "step": 5570
+    },
+    {
+      "epoch": 6.748488512696493,
+      "grad_norm": 3.2273013591766357,
+      "learning_rate": 1.8326101768790331e-06,
+      "loss": 0.6459,
+      "step": 5580
+    },
+    {
+      "epoch": 6.760580411124547,
+      "grad_norm": 3.989841938018799,
+      "learning_rate": 1.798123410407097e-06,
+      "loss": 0.6755,
+      "step": 5590
+    },
+    {
+      "epoch": 6.7726723095526,
+      "grad_norm": 5.584028720855713,
+      "learning_rate": 1.7639435266457759e-06,
+      "loss": 0.6293,
+      "step": 5600
+    },
+    {
+      "epoch": 6.784764207980653,
+      "grad_norm": 5.334152698516846,
+      "learning_rate": 1.7300713201196344e-06,
+      "loss": 0.7763,
+      "step": 5610
+    },
+    {
+      "epoch": 6.796856106408706,
+      "grad_norm": 3.9883923530578613,
+      "learning_rate": 1.6965075782011796e-06,
+      "loss": 0.7426,
+      "step": 5620
+    },
+    {
+      "epoch": 6.808948004836759,
+      "grad_norm": 5.658714771270752,
+      "learning_rate": 1.6632530810925178e-06,
+      "loss": 0.7003,
+      "step": 5630
+    },
+    {
+      "epoch": 6.821039903264813,
+      "grad_norm": 4.874862194061279,
+      "learning_rate": 1.6303086018072654e-06,
+      "loss": 0.6416,
+      "step": 5640
+    },
+    {
+      "epoch": 6.833131801692866,
+      "grad_norm": 5.249368190765381,
+      "learning_rate": 1.5976749061525436e-06,
+      "loss": 0.695,
+      "step": 5650
+    },
+    {
+      "epoch": 6.845223700120919,
+      "grad_norm": 3.46927809715271,
+      "learning_rate": 1.5653527527111822e-06,
+      "loss": 0.6439,
+      "step": 5660
+    },
+    {
+      "epoch": 6.857315598548972,
+      "grad_norm": 4.2240309715271,
+      "learning_rate": 1.5333428928240988e-06,
+      "loss": 0.6616,
+      "step": 5670
+    },
+    {
+      "epoch": 6.869407496977026,
+      "grad_norm": 6.947134494781494,
+      "learning_rate": 1.5016460705728285e-06,
+      "loss": 0.7057,
+      "step": 5680
+    },
+    {
+      "epoch": 6.881499395405078,
+      "grad_norm": 6.7013654708862305,
+      "learning_rate": 1.4702630227622249e-06,
+      "loss": 0.6777,
+      "step": 5690
+    },
+    {
+      "epoch": 6.893591293833132,
+      "grad_norm": 3.3105950355529785,
+      "learning_rate": 1.4391944789033278e-06,
+      "loss": 0.6382,
+      "step": 5700
+    },
+    {
+      "epoch": 6.9056831922611845,
+      "grad_norm": 4.079921245574951,
+      "learning_rate": 1.4084411611964187e-06,
+      "loss": 0.6842,
+      "step": 5710
+    },
+    {
+      "epoch": 6.917775090689238,
+      "grad_norm": 4.0758209228515625,
+      "learning_rate": 1.3780037845142218e-06,
+      "loss": 0.6355,
+      "step": 5720
+    },
+    {
+      "epoch": 6.929866989117292,
+      "grad_norm": 4.716675758361816,
+      "learning_rate": 1.3478830563852962e-06,
+      "loss": 0.5975,
+      "step": 5730
+    },
+    {
+      "epoch": 6.941958887545344,
+      "grad_norm": 3.8192057609558105,
+      "learning_rate": 1.318079676977572e-06,
+      "loss": 0.6477,
+      "step": 5740
+    },
+    {
+      "epoch": 6.954050785973398,
+      "grad_norm": 3.5390541553497314,
+      "learning_rate": 1.2885943390820959e-06,
+      "loss": 0.6842,
+      "step": 5750
+    },
+    {
+      "epoch": 6.966142684401451,
+      "grad_norm": 2.6589369773864746,
+      "learning_rate": 1.2594277280969124e-06,
+      "loss": 0.7192,
+      "step": 5760
+    },
+    {
+      "epoch": 6.978234582829504,
+      "grad_norm": 4.020471096038818,
+      "learning_rate": 1.23058052201114e-06,
+      "loss": 0.7203,
+      "step": 5770
+    },
+    {
+      "epoch": 6.990326481257558,
+      "grad_norm": 3.9986772537231445,
+      "learning_rate": 1.2020533913892056e-06,
+      "loss": 0.6348,
+      "step": 5780
+    },
+    {
+      "epoch": 7.003627569528416,
+      "grad_norm": 4.454996109008789,
+      "learning_rate": 1.173846999355252e-06,
+      "loss": 0.6173,
+      "step": 5790
+    },
+    {
+      "epoch": 7.010882708585248,
+      "eval_loss": 1.1261601448059082,
+      "eval_runtime": 62.3195,
+      "eval_samples_per_second": 3.049,
+      "eval_steps_per_second": 1.524,
+      "step": 5796
+    },
+    {
+      "epoch": 7.015719467956469,
+      "grad_norm": 3.0522561073303223,
+      "learning_rate": 1.1459620015777417e-06,
+      "loss": 0.5881,
+      "step": 5800
+    },
+    {
+      "epoch": 7.0278113663845225,
+      "grad_norm": 6.7532477378845215,
+      "learning_rate": 1.1183990462541938e-06,
+      "loss": 0.6966,
+      "step": 5810
+    },
+    {
+      "epoch": 7.039903264812575,
+      "grad_norm": 4.594633102416992,
+      "learning_rate": 1.091158774096132e-06,
+      "loss": 0.6317,
+      "step": 5820
+    },
+    {
+      "epoch": 7.051995163240629,
+      "grad_norm": 5.636754035949707,
+      "learning_rate": 1.064241818314179e-06,
+      "loss": 0.6814,
+      "step": 5830
+    },
+    {
+      "epoch": 7.064087061668682,
+      "grad_norm": 4.056423664093018,
+      "learning_rate": 1.0376488046033583e-06,
+      "loss": 0.6659,
+      "step": 5840
+    },
+    {
+      "epoch": 7.076178960096735,
+      "grad_norm": 5.472976207733154,
+      "learning_rate": 1.0113803511285147e-06,
+      "loss": 0.6233,
+      "step": 5850
+    },
+    {
+      "epoch": 7.088270858524789,
+      "grad_norm": 4.934895992279053,
+      "learning_rate": 9.854370685099807e-07,
+      "loss": 0.6452,
+      "step": 5860
+    },
+    {
+      "epoch": 7.100362756952841,
+      "grad_norm": 6.726385593414307,
+      "learning_rate": 9.598195598093618e-07,
+      "loss": 0.6035,
+      "step": 5870
+    },
+    {
+      "epoch": 7.112454655380895,
+      "grad_norm": 3.287743091583252,
+      "learning_rate": 9.345284205155247e-07,
+      "loss": 0.6692,
+      "step": 5880
+    },
+    {
+      "epoch": 7.124546553808948,
+      "grad_norm": 4.694064617156982,
+      "learning_rate": 9.095642385307479e-07,
+      "loss": 0.667,
+      "step": 5890
+    },
+    {
+      "epoch": 7.136638452237001,
+      "grad_norm": 5.595080375671387,
+      "learning_rate": 8.849275941570662e-07,
+      "loss": 0.7453,
+      "step": 5900
+    },
+    {
+      "epoch": 7.148730350665055,
+      "grad_norm": 3.937896728515625,
+      "learning_rate": 8.606190600827763e-07,
+      "loss": 0.5106,
+      "step": 5910
+    },
+    {
+      "epoch": 7.1608222490931075,
+      "grad_norm": 3.6209468841552734,
+      "learning_rate": 8.366392013691243e-07,
+      "loss": 0.7206,
+      "step": 5920
+    },
+    {
+      "epoch": 7.172914147521161,
+      "grad_norm": 6.286159515380859,
+      "learning_rate": 8.12988575437173e-07,
+      "loss": 0.6356,
+      "step": 5930
+    },
+    {
+      "epoch": 7.185006045949214,
+      "grad_norm": 5.586234092712402,
+      "learning_rate": 7.896677320548318e-07,
+      "loss": 0.5613,
+      "step": 5940
+    },
+    {
+      "epoch": 7.197097944377267,
+      "grad_norm": 4.128060817718506,
+      "learning_rate": 7.666772133241007e-07,
+      "loss": 0.7011,
+      "step": 5950
+    },
+    {
+      "epoch": 7.20918984280532,
+      "grad_norm": 5.194136619567871,
+      "learning_rate": 7.440175536684485e-07,
+      "loss": 0.7093,
+      "step": 5960
+    },
+    {
+      "epoch": 7.221281741233374,
+      "grad_norm": 7.493109703063965,
+      "learning_rate": 7.216892798204027e-07,
+      "loss": 0.6268,
+      "step": 5970
+    },
+    {
+      "epoch": 7.233373639661427,
+      "grad_norm": 4.296862602233887,
+      "learning_rate": 6.996929108092897e-07,
+      "loss": 0.5973,
+      "step": 5980
+    },
+    {
+      "epoch": 7.24546553808948,
+      "grad_norm": 4.065130233764648,
+      "learning_rate": 6.780289579491921e-07,
+      "loss": 0.6123,
+      "step": 5990
+    },
+    {
+      "epoch": 7.2575574365175335,
+      "grad_norm": 3.5596015453338623,
+      "learning_rate": 6.566979248270438e-07,
+      "loss": 0.5583,
+      "step": 6000
+    },
+    {
+      "epoch": 7.269649334945586,
+      "grad_norm": 3.1637520790100098,
+      "learning_rate": 6.35700307290934e-07,
+      "loss": 0.6135,
+      "step": 6010
+    },
+    {
+      "epoch": 7.28174123337364,
+      "grad_norm": 3.459122896194458,
+      "learning_rate": 6.150365934385715e-07,
+      "loss": 0.5554,
+      "step": 6020
+    },
+    {
+      "epoch": 7.2938331318016925,
+      "grad_norm": 3.151329278945923,
+      "learning_rate": 5.947072636059592e-07,
+      "loss": 0.7477,
+      "step": 6030
+    },
+    {
+      "epoch": 7.305925030229746,
+      "grad_norm": 3.4382164478302,
+      "learning_rate": 5.747127903562005e-07,
+      "loss": 0.6141,
+      "step": 6040
+    },
+    {
+      "epoch": 7.3180169286578,
+      "grad_norm": 3.598646879196167,
+      "learning_rate": 5.55053638468534e-07,
+      "loss": 0.5852,
+      "step": 6050
+    },
+    {
+      "epoch": 7.330108827085852,
+      "grad_norm": 3.409090280532837,
+      "learning_rate": 5.357302649275248e-07,
+      "loss": 0.6218,
+      "step": 6060
+    },
+    {
+      "epoch": 7.342200725513906,
+      "grad_norm": 4.669748783111572,
+      "learning_rate": 5.1674311891244e-07,
+      "loss": 0.6777,
+      "step": 6070
+    },
+    {
+      "epoch": 7.354292623941959,
+      "grad_norm": 5.575507164001465,
+      "learning_rate": 4.980926417868103e-07,
+      "loss": 0.6274,
+      "step": 6080
+    },
+    {
+      "epoch": 7.366384522370012,
+      "grad_norm": 4.644615650177002,
+      "learning_rate": 4.797792670881617e-07,
+      "loss": 0.7325,
+      "step": 6090
+    },
+    {
+      "epoch": 7.378476420798066,
+      "grad_norm": 3.4255263805389404,
+      "learning_rate": 4.6180342051795163e-07,
+      "loss": 0.6754,
+      "step": 6100
+    },
+    {
+      "epoch": 7.3905683192261185,
+      "grad_norm": 3.252048969268799,
+      "learning_rate": 4.4416551993166354e-07,
+      "loss": 0.6608,
+      "step": 6110
+    },
+    {
+      "epoch": 7.402660217654172,
+      "grad_norm": 4.5329365730285645,
+      "learning_rate": 4.2686597532909657e-07,
+      "loss": 0.7829,
+      "step": 6120
+    },
+    {
+      "epoch": 7.414752116082225,
+      "grad_norm": 5.146721839904785,
+      "learning_rate": 4.099051888448296e-07,
+      "loss": 0.5754,
+      "step": 6130
+    },
+    {
+      "epoch": 7.426844014510278,
+      "grad_norm": 4.085226058959961,
+      "learning_rate": 3.9328355473888745e-07,
+      "loss": 0.6844,
+      "step": 6140
+    },
+    {
+      "epoch": 7.438935912938331,
+      "grad_norm": 4.377448081970215,
+      "learning_rate": 3.7700145938755924e-07,
+      "loss": 0.7265,
+      "step": 6150
+    },
+    {
+      "epoch": 7.451027811366385,
+      "grad_norm": 3.7620866298675537,
+      "learning_rate": 3.6105928127443123e-07,
+      "loss": 0.6712,
+      "step": 6160
+    },
+    {
+      "epoch": 7.463119709794437,
+      "grad_norm": 4.5110626220703125,
+      "learning_rate": 3.4545739098157834e-07,
+      "loss": 0.6265,
+      "step": 6170
+    },
+    {
+      "epoch": 7.475211608222491,
+      "grad_norm": 4.606451988220215,
+      "learning_rate": 3.301961511809515e-07,
+      "loss": 0.7336,
+      "step": 6180
+    },
+    {
+      "epoch": 7.4873035066505444,
+      "grad_norm": 6.577304840087891,
+      "learning_rate": 3.1527591662595725e-07,
+      "loss": 0.553,
+      "step": 6190
+    },
+    {
+      "epoch": 7.499395405078597,
+      "grad_norm": 4.713655948638916,
+      "learning_rate": 3.0069703414319804e-07,
+      "loss": 0.7218,
+      "step": 6200
+    },
+    {
+      "epoch": 7.511487303506651,
+      "grad_norm": 4.648543834686279,
+      "learning_rate": 2.8645984262442024e-07,
+      "loss": 0.6861,
+      "step": 6210
+    },
+    {
+      "epoch": 7.511487303506651,
+      "eval_loss": 1.1386197805404663,
+      "eval_runtime": 62.2247,
+      "eval_samples_per_second": 3.053,
+      "eval_steps_per_second": 1.527,
+      "step": 6210
+    },
+    {
+      "epoch": 7.523579201934703,
+      "grad_norm": 4.095178127288818,
+      "learning_rate": 2.725646730186221e-07,
+      "loss": 0.5349,
+      "step": 6220
+    },
+    {
+      "epoch": 7.535671100362757,
+      "grad_norm": 7.081197738647461,
+      "learning_rate": 2.5901184832438473e-07,
+      "loss": 0.5566,
+      "step": 6230
+    },
+    {
+      "epoch": 7.547762998790811,
+      "grad_norm": 4.9830451011657715,
+      "learning_rate": 2.458016835823368e-07,
+      "loss": 0.7235,
+      "step": 6240
+    },
+    {
+      "epoch": 7.559854897218863,
+      "grad_norm": 3.604832649230957,
+      "learning_rate": 2.329344858678567e-07,
+      "loss": 0.6745,
+      "step": 6250
+    },
+    {
+      "epoch": 7.571946795646917,
+      "grad_norm": 4.3431010246276855,
+      "learning_rate": 2.2041055428390855e-07,
+      "loss": 0.681,
+      "step": 6260
+    },
+    {
+      "epoch": 7.5840386940749696,
+      "grad_norm": 4.137078285217285,
+      "learning_rate": 2.0823017995411585e-07,
+      "loss": 0.5737,
+      "step": 6270
+    },
+    {
+      "epoch": 7.596130592503023,
+      "grad_norm": 4.049239635467529,
+      "learning_rate": 1.9639364601597376e-07,
+      "loss": 0.608,
+      "step": 6280
+    },
+    {
+      "epoch": 7.608222490931076,
+      "grad_norm": 4.289693355560303,
+      "learning_rate": 1.849012276142742e-07,
+      "loss": 0.7035,
+      "step": 6290
+    },
+    {
+      "epoch": 7.620314389359129,
+      "grad_norm": 5.548840045928955,
+      "learning_rate": 1.737531918947144e-07,
+      "loss": 0.6729,
+      "step": 6300
+    },
+    {
+      "epoch": 7.632406287787182,
+      "grad_norm": 3.503091812133789,
+      "learning_rate": 1.6294979799768516e-07,
+      "loss": 0.7248,
+      "step": 6310
+    },
+    {
+      "epoch": 7.644498186215236,
+      "grad_norm": 4.313474655151367,
+      "learning_rate": 1.524912970522374e-07,
+      "loss": 0.6973,
+      "step": 6320
+    },
+    {
+      "epoch": 7.656590084643289,
+      "grad_norm": 3.1831257343292236,
+      "learning_rate": 1.4237793217026008e-07,
+      "loss": 0.6437,
+      "step": 6330
+    },
+    {
+      "epoch": 7.668681983071342,
+      "grad_norm": 5.381216526031494,
+      "learning_rate": 1.326099384408197e-07,
+      "loss": 0.7226,
+      "step": 6340
+    },
+    {
+      "epoch": 7.6807738814993956,
+      "grad_norm": 5.752452373504639,
+      "learning_rate": 1.2318754292469315e-07,
+      "loss": 0.6116,
+      "step": 6350
+    },
+    {
+      "epoch": 7.692865779927448,
+      "grad_norm": 3.416691780090332,
+      "learning_rate": 1.1411096464909843e-07,
+      "loss": 0.6333,
+      "step": 6360
+    },
+    {
+      "epoch": 7.704957678355502,
+      "grad_norm": 3.4745635986328125,
+      "learning_rate": 1.053804146025955e-07,
+      "loss": 0.5811,
+      "step": 6370
+    },
+    {
+      "epoch": 7.717049576783555,
+      "grad_norm": 3.8336737155914307,
+      "learning_rate": 9.699609573018686e-08,
+      "loss": 0.6614,
+      "step": 6380
+    },
+    {
+      "epoch": 7.729141475211608,
+      "grad_norm": 3.6881539821624756,
+      "learning_rate": 8.895820292859624e-08,
+      "loss": 0.6613,
+      "step": 6390
+    },
+    {
+      "epoch": 7.741233373639662,
+      "grad_norm": 4.259592056274414,
+      "learning_rate": 8.126692304174065e-08,
+      "loss": 0.6911,
+      "step": 6400
+    },
+    {
+      "epoch": 7.753325272067714,
+      "grad_norm": 3.5432159900665283,
+      "learning_rate": 7.39224348563855e-08,
+      "loss": 0.6985,
+      "step": 6410
+    },
+    {
+      "epoch": 7.765417170495768,
+      "grad_norm": 5.046801567077637,
+      "learning_rate": 6.69249090979912e-08,
+      "loss": 0.7313,
+      "step": 6420
+    },
+    {
+      "epoch": 7.777509068923821,
+      "grad_norm": 4.582681655883789,
+      "learning_rate": 6.027450842673976e-08,
+      "loss": 0.6667,
+      "step": 6430
+    },
+    {
+      "epoch": 7.789600967351874,
+      "grad_norm": 4.538397312164307,
+      "learning_rate": 5.39713874337594e-08,
+      "loss": 0.8269,
+      "step": 6440
+    },
+    {
+      "epoch": 7.801692865779928,
+      "grad_norm": 3.4865572452545166,
+      "learning_rate": 4.801569263752747e-08,
+      "loss": 0.6478,
+      "step": 6450
+    },
+    {
+      "epoch": 7.8137847642079805,
+      "grad_norm": 3.949676036834717,
+      "learning_rate": 4.2407562480464846e-08,
+      "loss": 0.7094,
+      "step": 6460
+    },
+    {
+      "epoch": 7.825876662636034,
+      "grad_norm": 4.50856876373291,
+      "learning_rate": 3.714712732572012e-08,
+      "loss": 0.6551,
+      "step": 6470
+    },
+    {
+      "epoch": 7.837968561064087,
+      "grad_norm": 3.412080764770508,
+      "learning_rate": 3.223450945413542e-08,
+      "loss": 0.6534,
+      "step": 6480
+    },
+    {
+      "epoch": 7.85006045949214,
+      "grad_norm": 3.6913983821868896,
+      "learning_rate": 2.7669823061406972e-08,
+      "loss": 0.6306,
+      "step": 6490
+    },
+    {
+      "epoch": 7.862152357920193,
+      "grad_norm": 5.678858757019043,
+      "learning_rate": 2.345317425542892e-08,
+      "loss": 0.5884,
+      "step": 6500
+    },
+    {
+      "epoch": 7.874244256348247,
+      "grad_norm": 3.923584461212158,
+      "learning_rate": 1.9584661053830278e-08,
+      "loss": 0.6475,
+      "step": 6510
+    },
+    {
+      "epoch": 7.8863361547763,
+      "grad_norm": 5.143980503082275,
+      "learning_rate": 1.606437338169009e-08,
+      "loss": 0.6475,
+      "step": 6520
+    },
+    {
+      "epoch": 7.898428053204353,
+      "grad_norm": 3.7609965801239014,
+      "learning_rate": 1.2892393069454111e-08,
+      "loss": 0.5987,
+      "step": 6530
+    },
+    {
+      "epoch": 7.9105199516324065,
+      "grad_norm": 5.001131057739258,
+      "learning_rate": 1.006879385102466e-08,
+      "loss": 0.6181,
+      "step": 6540
+    },
+    {
+      "epoch": 7.922611850060459,
+      "grad_norm": 5.9533257484436035,
+      "learning_rate": 7.59364136205698e-09,
+      "loss": 0.6446,
+      "step": 6550
+    },
+    {
+      "epoch": 7.934703748488513,
+      "grad_norm": 4.852860450744629,
+      "learning_rate": 5.466993138423804e-09,
+      "loss": 0.5722,
+      "step": 6560
+    },
+    {
+      "epoch": 7.946795646916566,
+      "grad_norm": 3.6675336360931396,
+      "learning_rate": 3.6888986148814196e-09,
+      "loss": 0.668,
+      "step": 6570
+    },
+    {
+      "epoch": 7.958887545344619,
+      "grad_norm": 3.777907133102417,
+      "learning_rate": 2.259399123923922e-09,
+      "loss": 0.5827,
+      "step": 6580
+    },
+    {
+      "epoch": 7.970979443772673,
+      "grad_norm": 3.8638296127319336,
+      "learning_rate": 1.1785278948189815e-09,
+      "loss": 0.6919,
+      "step": 6590
+    },
+    {
+      "epoch": 7.983071342200725,
+      "grad_norm": 3.947110652923584,
+      "learning_rate": 4.4631005283179983e-10,
+      "loss": 0.5722,
+      "step": 6600
+    },
+    {
+      "epoch": 7.995163240628779,
+      "grad_norm": 4.3253655433654785,
+      "learning_rate": 6.276261865223453e-11,
+      "loss": 0.5308,
+      "step": 6610
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 6616,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 8,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 5.3489398056380006e+17,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}