diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,4233 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.8272438990762443,
+  "eval_steps": 500,
+  "global_step": 6000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0013787398317937406,
+      "grad_norm": 1.0734721422195435,
+      "learning_rate": 0.0002,
+      "loss": 2.8003,
+      "step": 10
+    },
+    {
+      "epoch": 0.002757479663587481,
+      "grad_norm": 0.19643057882785797,
+      "learning_rate": 0.0002,
+      "loss": 0.3727,
+      "step": 20
+    },
+    {
+      "epoch": 0.004136219495381221,
+      "grad_norm": 0.8137874007225037,
+      "learning_rate": 0.0002,
+      "loss": 0.3519,
+      "step": 30
+    },
+    {
+      "epoch": 0.005514959327174962,
+      "grad_norm": 0.4064357578754425,
+      "learning_rate": 0.0002,
+      "loss": 0.357,
+      "step": 40
+    },
+    {
+      "epoch": 0.0068936991589687024,
+      "grad_norm": 1.0673978328704834,
+      "learning_rate": 0.0002,
+      "loss": 0.3515,
+      "step": 50
+    },
+    {
+      "epoch": 0.008272438990762443,
+      "grad_norm": 2.2639453411102295,
+      "learning_rate": 0.0002,
+      "loss": 0.4068,
+      "step": 60
+    },
+    {
+      "epoch": 0.009651178822556184,
+      "grad_norm": 1.2247616052627563,
+      "learning_rate": 0.0002,
+      "loss": 0.5108,
+      "step": 70
+    },
+    {
+      "epoch": 0.011029918654349925,
+      "grad_norm": 0.07231716066598892,
+      "learning_rate": 0.0002,
+      "loss": 0.3644,
+      "step": 80
+    },
+    {
+      "epoch": 0.012408658486143665,
+      "grad_norm": 1.4329577684402466,
+      "learning_rate": 0.0002,
+      "loss": 0.3571,
+      "step": 90
+    },
+    {
+      "epoch": 0.013787398317937405,
+      "grad_norm": 0.8798255920410156,
+      "learning_rate": 0.0002,
+      "loss": 0.3521,
+      "step": 100
+    },
+    {
+      "epoch": 0.015166138149731145,
+      "grad_norm": 0.06499180942773819,
+      "learning_rate": 0.0002,
+      "loss": 0.3597,
+      "step": 110
+    },
+    {
+      "epoch": 0.016544877981524885,
+      "grad_norm": 0.284839928150177,
+      "learning_rate": 0.0002,
+      "loss": 0.4099,
+      "step": 120
+    },
+    {
+      "epoch": 0.017923617813318627,
+      "grad_norm": 0.12186373770236969,
+      "learning_rate": 0.0002,
+      "loss": 0.3572,
+      "step": 130
+    },
+    {
+      "epoch": 0.01930235764511237,
+      "grad_norm": 0.21747085452079773,
+      "learning_rate": 0.0002,
+      "loss": 0.3497,
+      "step": 140
+    },
+    {
+      "epoch": 0.020681097476906107,
+      "grad_norm": 0.22812429070472717,
+      "learning_rate": 0.0002,
+      "loss": 0.3778,
+      "step": 150
+    },
+    {
+      "epoch": 0.02205983730869985,
+      "grad_norm": 1.1110987663269043,
+      "learning_rate": 0.0002,
+      "loss": 0.3607,
+      "step": 160
+    },
+    {
+      "epoch": 0.023438577140493588,
+      "grad_norm": 0.8457236886024475,
+      "learning_rate": 0.0002,
+      "loss": 0.9183,
+      "step": 170
+    },
+    {
+      "epoch": 0.02481731697228733,
+      "grad_norm": 0.46427470445632935,
+      "learning_rate": 0.0002,
+      "loss": 0.3503,
+      "step": 180
+    },
+    {
+      "epoch": 0.02619605680408107,
+      "grad_norm": 2.4730632305145264,
+      "learning_rate": 0.0002,
+      "loss": 0.3824,
+      "step": 190
+    },
+    {
+      "epoch": 0.02757479663587481,
+      "grad_norm": 0.3589227497577667,
+      "learning_rate": 0.0002,
+      "loss": 1.0105,
+      "step": 200
+    },
+    {
+      "epoch": 0.02895353646766855,
+      "grad_norm": 3.5308260917663574,
+      "learning_rate": 0.0002,
+      "loss": 0.3605,
+      "step": 210
+    },
+    {
+      "epoch": 0.03033227629946229,
+      "grad_norm": 0.5237946510314941,
+      "learning_rate": 0.0002,
+      "loss": 0.3983,
+      "step": 220
+    },
+    {
+      "epoch": 0.031711016131256035,
+      "grad_norm": 0.5702632665634155,
+      "learning_rate": 0.0002,
+      "loss": 0.3521,
+      "step": 230
+    },
+    {
+      "epoch": 0.03308975596304977,
+      "grad_norm": 1.1318608522415161,
+      "learning_rate": 0.0002,
+      "loss": 0.3617,
+      "step": 240
+    },
+    {
+      "epoch": 0.03446849579484351,
+      "grad_norm": 0.536571741104126,
+      "learning_rate": 0.0002,
+      "loss": 0.3555,
+      "step": 250
+    },
+    {
+      "epoch": 0.035847235626637254,
+      "grad_norm": 0.12125778943300247,
+      "learning_rate": 0.0002,
+      "loss": 0.3733,
+      "step": 260
+    },
+    {
+      "epoch": 0.037225975458430996,
+      "grad_norm": 0.045536063611507416,
+      "learning_rate": 0.0002,
+      "loss": 0.3514,
+      "step": 270
+    },
+    {
+      "epoch": 0.03860471529022474,
+      "grad_norm": 0.31765612959861755,
+      "learning_rate": 0.0002,
+      "loss": 0.3536,
+      "step": 280
+    },
+    {
+      "epoch": 0.03998345512201847,
+      "grad_norm": 0.27900660037994385,
+      "learning_rate": 0.0002,
+      "loss": 0.3495,
+      "step": 290
+    },
+    {
+      "epoch": 0.041362194953812215,
+      "grad_norm": 0.15112566947937012,
+      "learning_rate": 0.0002,
+      "loss": 0.355,
+      "step": 300
+    },
+    {
+      "epoch": 0.04274093478560596,
+      "grad_norm": 0.2682303786277771,
+      "learning_rate": 0.0002,
+      "loss": 0.3609,
+      "step": 310
+    },
+    {
+      "epoch": 0.0441196746173997,
+      "grad_norm": 1.0106860399246216,
+      "learning_rate": 0.0002,
+      "loss": 0.3697,
+      "step": 320
+    },
+    {
+      "epoch": 0.04549841444919344,
+      "grad_norm": 1.0782426595687866,
+      "learning_rate": 0.0002,
+      "loss": 0.3594,
+      "step": 330
+    },
+    {
+      "epoch": 0.046877154280987175,
+      "grad_norm": 2.294581651687622,
+      "learning_rate": 0.0002,
+      "loss": 0.3676,
+      "step": 340
+    },
+    {
+      "epoch": 0.04825589411278092,
+      "grad_norm": 0.6223801970481873,
+      "learning_rate": 0.0002,
+      "loss": 0.3741,
+      "step": 350
+    },
+    {
+      "epoch": 0.04963463394457466,
+      "grad_norm": 0.2735952138900757,
+      "learning_rate": 0.0002,
+      "loss": 0.3628,
+      "step": 360
+    },
+    {
+      "epoch": 0.0510133737763684,
+      "grad_norm": 0.7569056153297424,
+      "learning_rate": 0.0002,
+      "loss": 1.1827,
+      "step": 370
+    },
+    {
+      "epoch": 0.05239211360816214,
+      "grad_norm": 0.6536706686019897,
+      "learning_rate": 0.0002,
+      "loss": 0.3543,
+      "step": 380
+    },
+    {
+      "epoch": 0.05377085343995588,
+      "grad_norm": 0.3573110103607178,
+      "learning_rate": 0.0002,
+      "loss": 0.3529,
+      "step": 390
+    },
+    {
+      "epoch": 0.05514959327174962,
+      "grad_norm": 0.8121228218078613,
+      "learning_rate": 0.0002,
+      "loss": 0.3566,
+      "step": 400
+    },
+    {
+      "epoch": 0.05652833310354336,
+      "grad_norm": 0.7444269061088562,
+      "learning_rate": 0.0002,
+      "loss": 0.7803,
+      "step": 410
+    },
+    {
+      "epoch": 0.0579070729353371,
+      "grad_norm": 0.5015038847923279,
+      "learning_rate": 0.0002,
+      "loss": 0.6134,
+      "step": 420
+    },
+    {
+      "epoch": 0.059285812767130845,
+      "grad_norm": 0.08748292177915573,
+      "learning_rate": 0.0002,
+      "loss": 0.3501,
+      "step": 430
+    },
+    {
+      "epoch": 0.06066455259892458,
+      "grad_norm": 0.3987080156803131,
+      "learning_rate": 0.0002,
+      "loss": 0.3484,
+      "step": 440
+    },
+    {
+      "epoch": 0.06204329243071832,
+      "grad_norm": 347.7005920410156,
+      "learning_rate": 0.0002,
+      "loss": 0.8413,
+      "step": 450
+    },
+    {
+      "epoch": 0.06342203226251207,
+      "grad_norm": 88.2750473022461,
+      "learning_rate": 0.0002,
+      "loss": 2.8013,
+      "step": 460
+    },
+    {
+      "epoch": 0.0648007720943058,
+      "grad_norm": 0.8716701865196228,
+      "learning_rate": 0.0002,
+      "loss": 0.8356,
+      "step": 470
+    },
+    {
+      "epoch": 0.06617951192609954,
+      "grad_norm": 0.8243119120597839,
+      "learning_rate": 0.0002,
+      "loss": 0.3616,
+      "step": 480
+    },
+    {
+      "epoch": 0.06755825175789329,
+      "grad_norm": 1.1744294166564941,
+      "learning_rate": 0.0002,
+      "loss": 0.3998,
+      "step": 490
+    },
+    {
+      "epoch": 0.06893699158968702,
+      "grad_norm": 0.03163053095340729,
+      "learning_rate": 0.0002,
+      "loss": 0.3549,
+      "step": 500
+    },
+    {
+      "epoch": 0.07031573142148077,
+      "grad_norm": 3.4403915405273438,
+      "learning_rate": 0.0002,
+      "loss": 0.3722,
+      "step": 510
+    },
+    {
+      "epoch": 0.07169447125327451,
+      "grad_norm": 1.0608879327774048,
+      "learning_rate": 0.0002,
+      "loss": 0.3701,
+      "step": 520
+    },
+    {
+      "epoch": 0.07307321108506824,
+      "grad_norm": 1.2809940576553345,
+      "learning_rate": 0.0002,
+      "loss": 0.3602,
+      "step": 530
+    },
+    {
+      "epoch": 0.07445195091686199,
+      "grad_norm": 0.40460118651390076,
+      "learning_rate": 0.0002,
+      "loss": 0.3532,
+      "step": 540
+    },
+    {
+      "epoch": 0.07583069074865573,
+      "grad_norm": 0.6290703415870667,
+      "learning_rate": 0.0002,
+      "loss": 0.3477,
+      "step": 550
+    },
+    {
+      "epoch": 0.07720943058044948,
+      "grad_norm": 0.2159261256456375,
+      "learning_rate": 0.0002,
+      "loss": 0.405,
+      "step": 560
+    },
+    {
+      "epoch": 0.07858817041224321,
+      "grad_norm": 0.37101301550865173,
+      "learning_rate": 0.0002,
+      "loss": 0.3752,
+      "step": 570
+    },
+    {
+      "epoch": 0.07996691024403695,
+      "grad_norm": 1.3007190227508545,
+      "learning_rate": 0.0002,
+      "loss": 0.3503,
+      "step": 580
+    },
+    {
+      "epoch": 0.0813456500758307,
+      "grad_norm": 0.4508918225765228,
+      "learning_rate": 0.0002,
+      "loss": 0.3531,
+      "step": 590
+    },
+    {
+      "epoch": 0.08272438990762443,
+      "grad_norm": 0.46898791193962097,
+      "learning_rate": 0.0002,
+      "loss": 0.347,
+      "step": 600
+    },
+    {
+      "epoch": 0.08410312973941818,
+      "grad_norm": 0.8449831604957581,
+      "learning_rate": 0.0002,
+      "loss": 0.3546,
+      "step": 610
+    },
+    {
+      "epoch": 0.08548186957121191,
+      "grad_norm": 0.7988163232803345,
+      "learning_rate": 0.0002,
+      "loss": 0.3505,
+      "step": 620
+    },
+    {
+      "epoch": 0.08686060940300565,
+      "grad_norm": 0.4426226317882538,
+      "learning_rate": 0.0002,
+      "loss": 0.3649,
+      "step": 630
+    },
+    {
+      "epoch": 0.0882393492347994,
+      "grad_norm": 0.2260913848876953,
+      "learning_rate": 0.0002,
+      "loss": 0.3499,
+      "step": 640
+    },
+    {
+      "epoch": 0.08961808906659313,
+      "grad_norm": 1.476747751235962,
+      "learning_rate": 0.0002,
+      "loss": 0.3546,
+      "step": 650
+    },
+    {
+      "epoch": 0.09099682889838688,
+      "grad_norm": 0.7640777230262756,
+      "learning_rate": 0.0002,
+      "loss": 0.3568,
+      "step": 660
+    },
+    {
+      "epoch": 0.09237556873018062,
+      "grad_norm": 0.8559088706970215,
+      "learning_rate": 0.0002,
+      "loss": 0.3507,
+      "step": 670
+    },
+    {
+      "epoch": 0.09375430856197435,
+      "grad_norm": 0.20833595097064972,
+      "learning_rate": 0.0002,
+      "loss": 0.3556,
+      "step": 680
+    },
+    {
+      "epoch": 0.0951330483937681,
+      "grad_norm": 1.1485021114349365,
+      "learning_rate": 0.0002,
+      "loss": 0.3516,
+      "step": 690
+    },
+    {
+      "epoch": 0.09651178822556183,
+      "grad_norm": 1.0206815004348755,
+      "learning_rate": 0.0002,
+      "loss": 0.3591,
+      "step": 700
+    },
+    {
+      "epoch": 0.09789052805735558,
+      "grad_norm": 0.9966775178909302,
+      "learning_rate": 0.0002,
+      "loss": 0.359,
+      "step": 710
+    },
+    {
+      "epoch": 0.09926926788914932,
+      "grad_norm": 0.8833585977554321,
+      "learning_rate": 0.0002,
+      "loss": 0.3546,
+      "step": 720
+    },
+    {
+      "epoch": 0.10064800772094305,
+      "grad_norm": 1.0842584371566772,
+      "learning_rate": 0.0002,
+      "loss": 0.3556,
+      "step": 730
+    },
+    {
+      "epoch": 0.1020267475527368,
+      "grad_norm": 0.3791511058807373,
+      "learning_rate": 0.0002,
+      "loss": 0.3571,
+      "step": 740
+    },
+    {
+      "epoch": 0.10340548738453054,
+      "grad_norm": 0.24666732549667358,
+      "learning_rate": 0.0002,
+      "loss": 0.3502,
+      "step": 750
+    },
+    {
+      "epoch": 0.10478422721632429,
+      "grad_norm": 0.21794968843460083,
+      "learning_rate": 0.0002,
+      "loss": 0.3483,
+      "step": 760
+    },
+    {
+      "epoch": 0.10616296704811802,
+      "grad_norm": 0.47017499804496765,
+      "learning_rate": 0.0002,
+      "loss": 0.3499,
+      "step": 770
+    },
+    {
+      "epoch": 0.10754170687991176,
+      "grad_norm": 0.2813131809234619,
+      "learning_rate": 0.0002,
+      "loss": 0.3482,
+      "step": 780
+    },
+    {
+      "epoch": 0.1089204467117055,
+      "grad_norm": 1.2175363302230835,
+      "learning_rate": 0.0002,
+      "loss": 0.3524,
+      "step": 790
+    },
+    {
+      "epoch": 0.11029918654349924,
+      "grad_norm": 0.2712210416793823,
+      "learning_rate": 0.0002,
+      "loss": 0.3526,
+      "step": 800
+    },
+    {
+      "epoch": 0.11167792637529299,
+      "grad_norm": 0.1428445726633072,
+      "learning_rate": 0.0002,
+      "loss": 0.3518,
+      "step": 810
+    },
+    {
+      "epoch": 0.11305666620708672,
+      "grad_norm": 0.23716595768928528,
+      "learning_rate": 0.0002,
+      "loss": 0.3501,
+      "step": 820
+    },
+    {
+      "epoch": 0.11443540603888046,
+      "grad_norm": 0.07993923872709274,
+      "learning_rate": 0.0002,
+      "loss": 0.349,
+      "step": 830
+    },
+    {
+      "epoch": 0.1158141458706742,
+      "grad_norm": 1.2958595752716064,
+      "learning_rate": 0.0002,
+      "loss": 0.352,
+      "step": 840
+    },
+    {
+      "epoch": 0.11719288570246794,
+      "grad_norm": 1.6257132291793823,
+      "learning_rate": 0.0002,
+      "loss": 0.3589,
+      "step": 850
+    },
+    {
+      "epoch": 0.11857162553426169,
+      "grad_norm": 0.20367591083049774,
+      "learning_rate": 0.0002,
+      "loss": 0.3602,
+      "step": 860
+    },
+    {
+      "epoch": 0.11995036536605543,
+      "grad_norm": 1.147210955619812,
+      "learning_rate": 0.0002,
+      "loss": 0.3619,
+      "step": 870
+    },
+    {
+      "epoch": 0.12132910519784916,
+      "grad_norm": 0.19706425070762634,
+      "learning_rate": 0.0002,
+      "loss": 0.3536,
+      "step": 880
+    },
+    {
+      "epoch": 0.12270784502964291,
+      "grad_norm": 0.17990930378437042,
+      "learning_rate": 0.0002,
+      "loss": 0.3501,
+      "step": 890
+    },
+    {
+      "epoch": 0.12408658486143664,
+      "grad_norm": 0.5770463943481445,
+      "learning_rate": 0.0002,
+      "loss": 0.3509,
+      "step": 900
+    },
+    {
+      "epoch": 0.12546532469323038,
+      "grad_norm": 0.24645955860614777,
+      "learning_rate": 0.0002,
+      "loss": 0.3526,
+      "step": 910
+    },
+    {
+      "epoch": 0.12684406452502414,
+      "grad_norm": 0.15745119750499725,
+      "learning_rate": 0.0002,
+      "loss": 0.3503,
+      "step": 920
+    },
+    {
+      "epoch": 0.12822280435681788,
+      "grad_norm": 0.054484590888023376,
+      "learning_rate": 0.0002,
+      "loss": 0.3508,
+      "step": 930
+    },
+    {
+      "epoch": 0.1296015441886116,
+      "grad_norm": 0.30564025044441223,
+      "learning_rate": 0.0002,
+      "loss": 0.3489,
+      "step": 940
+    },
+    {
+      "epoch": 0.13098028402040535,
+      "grad_norm": 0.3614678382873535,
+      "learning_rate": 0.0002,
+      "loss": 0.3477,
+      "step": 950
+    },
+    {
+      "epoch": 0.13235902385219908,
+      "grad_norm": 0.703029990196228,
+      "learning_rate": 0.0002,
+      "loss": 0.3552,
+      "step": 960
+    },
+    {
+      "epoch": 0.13373776368399284,
+      "grad_norm": 1.1954560279846191,
+      "learning_rate": 0.0002,
+      "loss": 0.3528,
+      "step": 970
+    },
+    {
+      "epoch": 0.13511650351578658,
+      "grad_norm": 0.8106504678726196,
+      "learning_rate": 0.0002,
+      "loss": 0.3586,
+      "step": 980
+    },
+    {
+      "epoch": 0.13649524334758031,
+      "grad_norm": 0.40758854150772095,
+      "learning_rate": 0.0002,
+      "loss": 0.3908,
+      "step": 990
+    },
+    {
+      "epoch": 0.13787398317937405,
+      "grad_norm": 0.613096296787262,
+      "learning_rate": 0.0002,
+      "loss": 0.3519,
+      "step": 1000
+    },
+    {
+      "epoch": 0.13925272301116778,
+      "grad_norm": 0.38185614347457886,
+      "learning_rate": 0.0002,
+      "loss": 0.3506,
+      "step": 1010
+    },
+    {
+      "epoch": 0.14063146284296155,
+      "grad_norm": 0.07220327854156494,
+      "learning_rate": 0.0002,
+      "loss": 0.3472,
+      "step": 1020
+    },
+    {
+      "epoch": 0.14201020267475528,
+      "grad_norm": 0.1451689898967743,
+      "learning_rate": 0.0002,
+      "loss": 0.3534,
+      "step": 1030
+    },
+    {
+      "epoch": 0.14338894250654902,
+      "grad_norm": 0.08052591234445572,
+      "learning_rate": 0.0002,
+      "loss": 0.3476,
+      "step": 1040
+    },
+    {
+      "epoch": 0.14476768233834275,
+      "grad_norm": 1.0108163356781006,
+      "learning_rate": 0.0002,
+      "loss": 0.3508,
+      "step": 1050
+    },
+    {
+      "epoch": 0.1461464221701365,
+      "grad_norm": 0.5895722508430481,
+      "learning_rate": 0.0002,
+      "loss": 0.3541,
+      "step": 1060
+    },
+    {
+      "epoch": 0.14752516200193025,
+      "grad_norm": 0.6988415718078613,
+      "learning_rate": 0.0002,
+      "loss": 0.3512,
+      "step": 1070
+    },
+    {
+      "epoch": 0.14890390183372398,
+      "grad_norm": 0.54078608751297,
+      "learning_rate": 0.0002,
+      "loss": 0.3479,
+      "step": 1080
+    },
+    {
+      "epoch": 0.15028264166551772,
+      "grad_norm": 0.19162333011627197,
+      "learning_rate": 0.0002,
+      "loss": 0.3518,
+      "step": 1090
+    },
+    {
+      "epoch": 0.15166138149731145,
+      "grad_norm": 0.36928215622901917,
+      "learning_rate": 0.0002,
+      "loss": 0.3505,
+      "step": 1100
+    },
+    {
+      "epoch": 0.1530401213291052,
+      "grad_norm": 0.572607696056366,
+      "learning_rate": 0.0002,
+      "loss": 0.355,
+      "step": 1110
+    },
+    {
+      "epoch": 0.15441886116089895,
+      "grad_norm": 0.20841191709041595,
+      "learning_rate": 0.0002,
+      "loss": 0.348,
+      "step": 1120
+    },
+    {
+      "epoch": 0.1557976009926927,
+      "grad_norm": 0.04682110995054245,
+      "learning_rate": 0.0002,
+      "loss": 0.3502,
+      "step": 1130
+    },
+    {
+      "epoch": 0.15717634082448642,
+      "grad_norm": 0.867899477481842,
+      "learning_rate": 0.0002,
+      "loss": 0.3476,
+      "step": 1140
+    },
+    {
+      "epoch": 0.15855508065628016,
+      "grad_norm": 0.2828502655029297,
+      "learning_rate": 0.0002,
+      "loss": 0.3525,
+      "step": 1150
+    },
+    {
+      "epoch": 0.1599338204880739,
+      "grad_norm": 0.44510889053344727,
+      "learning_rate": 0.0002,
+      "loss": 0.35,
+      "step": 1160
+    },
+    {
+      "epoch": 0.16131256031986765,
+      "grad_norm": 0.1896822154521942,
+      "learning_rate": 0.0002,
+      "loss": 0.3493,
+      "step": 1170
+    },
+    {
+      "epoch": 0.1626913001516614,
+      "grad_norm": 0.15781590342521667,
+      "learning_rate": 0.0002,
+      "loss": 0.3477,
+      "step": 1180
+    },
+    {
+      "epoch": 0.16407003998345512,
+      "grad_norm": 0.2315225899219513,
+      "learning_rate": 0.0002,
+      "loss": 0.3498,
+      "step": 1190
+    },
+    {
+      "epoch": 0.16544877981524886,
+      "grad_norm": 0.2198018729686737,
+      "learning_rate": 0.0002,
+      "loss": 0.3484,
+      "step": 1200
+    },
+    {
+      "epoch": 0.1668275196470426,
+      "grad_norm": 0.2039571851491928,
+      "learning_rate": 0.0002,
+      "loss": 0.348,
+      "step": 1210
+    },
+    {
+      "epoch": 0.16820625947883636,
+      "grad_norm": 0.009352603927254677,
+      "learning_rate": 0.0002,
+      "loss": 0.3481,
+      "step": 1220
+    },
+    {
+      "epoch": 0.1695849993106301,
+      "grad_norm": 0.2558707892894745,
+      "learning_rate": 0.0002,
+      "loss": 0.3475,
+      "step": 1230
+    },
+    {
+      "epoch": 0.17096373914242383,
+      "grad_norm": 0.07278712838888168,
+      "learning_rate": 0.0002,
+      "loss": 0.3471,
+      "step": 1240
+    },
+    {
+      "epoch": 0.17234247897421756,
+      "grad_norm": 0.4133436381816864,
+      "learning_rate": 0.0002,
+      "loss": 0.3453,
+      "step": 1250
+    },
+    {
+      "epoch": 0.1737212188060113,
+      "grad_norm": 0.16729828715324402,
+      "learning_rate": 0.0002,
+      "loss": 0.3527,
+      "step": 1260
+    },
+    {
+      "epoch": 0.17509995863780506,
+      "grad_norm": 0.33326980471611023,
+      "learning_rate": 0.0002,
+      "loss": 0.3463,
+      "step": 1270
+    },
+    {
+      "epoch": 0.1764786984695988,
+      "grad_norm": 0.7140666246414185,
+      "learning_rate": 0.0002,
+      "loss": 0.3627,
+      "step": 1280
+    },
+    {
+      "epoch": 0.17785743830139253,
+      "grad_norm": 0.17751634120941162,
+      "learning_rate": 0.0002,
+      "loss": 0.3506,
+      "step": 1290
+    },
+    {
+      "epoch": 0.17923617813318626,
+      "grad_norm": 0.44009125232696533,
+      "learning_rate": 0.0002,
+      "loss": 0.3516,
+      "step": 1300
+    },
+    {
+      "epoch": 0.18061491796498,
+      "grad_norm": 0.07371579110622406,
+      "learning_rate": 0.0002,
+      "loss": 0.3479,
+      "step": 1310
+    },
+    {
+      "epoch": 0.18199365779677376,
+      "grad_norm": 0.6804266571998596,
+      "learning_rate": 0.0002,
+      "loss": 0.3476,
+      "step": 1320
+    },
+    {
+      "epoch": 0.1833723976285675,
+      "grad_norm": 0.19634029269218445,
+      "learning_rate": 0.0002,
+      "loss": 0.354,
+      "step": 1330
+    },
+    {
+      "epoch": 0.18475113746036123,
+      "grad_norm": 0.34020882844924927,
+      "learning_rate": 0.0002,
+      "loss": 0.3481,
+      "step": 1340
+    },
+    {
+      "epoch": 0.18612987729215497,
+      "grad_norm": 0.38502731919288635,
+      "learning_rate": 0.0002,
+      "loss": 0.3501,
+      "step": 1350
+    },
+    {
+      "epoch": 0.1875086171239487,
+      "grad_norm": 0.0810522586107254,
+      "learning_rate": 0.0002,
+      "loss": 0.3473,
+      "step": 1360
+    },
+    {
+      "epoch": 0.18888735695574246,
+      "grad_norm": 0.4057389497756958,
+      "learning_rate": 0.0002,
+      "loss": 0.3489,
+      "step": 1370
+    },
+    {
+      "epoch": 0.1902660967875362,
+      "grad_norm": 0.17514599859714508,
+      "learning_rate": 0.0002,
+      "loss": 0.3472,
+      "step": 1380
+    },
+    {
+      "epoch": 0.19164483661932993,
+      "grad_norm": 0.10964088141918182,
+      "learning_rate": 0.0002,
+      "loss": 0.3479,
+      "step": 1390
+    },
+    {
+      "epoch": 0.19302357645112367,
+      "grad_norm": 0.20920871198177338,
+      "learning_rate": 0.0002,
+      "loss": 0.3488,
+      "step": 1400
+    },
+    {
+      "epoch": 0.1944023162829174,
+      "grad_norm": 1.149121880531311,
+      "learning_rate": 0.0002,
+      "loss": 0.3548,
+      "step": 1410
+    },
+    {
+      "epoch": 0.19578105611471117,
+      "grad_norm": 1.3394649028778076,
+      "learning_rate": 0.0002,
+      "loss": 0.3495,
+      "step": 1420
+    },
+    {
+      "epoch": 0.1971597959465049,
+      "grad_norm": 1.2763960361480713,
+      "learning_rate": 0.0002,
+      "loss": 0.3679,
+      "step": 1430
+    },
+    {
+      "epoch": 0.19853853577829864,
+      "grad_norm": 0.5421571731567383,
+      "learning_rate": 0.0002,
+      "loss": 0.3538,
+      "step": 1440
+    },
+    {
+      "epoch": 0.19991727561009237,
+      "grad_norm": 0.22273503243923187,
+      "learning_rate": 0.0002,
+      "loss": 0.3518,
+      "step": 1450
+    },
+    {
+      "epoch": 0.2012960154418861,
+      "grad_norm": 0.6335702538490295,
+      "learning_rate": 0.0002,
+      "loss": 0.3481,
+      "step": 1460
+    },
+    {
+      "epoch": 0.20267475527367987,
+      "grad_norm": 0.7090324759483337,
+      "learning_rate": 0.0002,
+      "loss": 0.3486,
+      "step": 1470
+    },
+    {
+      "epoch": 0.2040534951054736,
+      "grad_norm": 0.011333847418427467,
+      "learning_rate": 0.0002,
+      "loss": 0.3476,
+      "step": 1480
+    },
+    {
+      "epoch": 0.20543223493726734,
+      "grad_norm": 0.24088676273822784,
+      "learning_rate": 0.0002,
+      "loss": 0.3569,
+      "step": 1490
+    },
+    {
+      "epoch": 0.20681097476906107,
+      "grad_norm": 0.8654371500015259,
+      "learning_rate": 0.0002,
+      "loss": 0.3528,
+      "step": 1500
+    },
+    {
+      "epoch": 0.2081897146008548,
+      "grad_norm": 0.06135034188628197,
+      "learning_rate": 0.0002,
+      "loss": 0.3509,
+      "step": 1510
+    },
+    {
+      "epoch": 0.20956845443264857,
+      "grad_norm": 0.38141730427742004,
+      "learning_rate": 0.0002,
+      "loss": 0.3474,
+      "step": 1520
+    },
+    {
+      "epoch": 0.2109471942644423,
+      "grad_norm": 1.1622456312179565,
+      "learning_rate": 0.0002,
+      "loss": 0.353,
+      "step": 1530
+    },
+    {
+      "epoch": 0.21232593409623604,
+      "grad_norm": 0.5747712254524231,
+      "learning_rate": 0.0002,
+      "loss": 0.3513,
+      "step": 1540
+    },
+    {
+      "epoch": 0.21370467392802978,
+      "grad_norm": 0.09723293781280518,
+      "learning_rate": 0.0002,
+      "loss": 0.3482,
+      "step": 1550
+    },
+    {
+      "epoch": 0.2150834137598235,
+      "grad_norm": 0.18574804067611694,
+      "learning_rate": 0.0002,
+      "loss": 0.3512,
+      "step": 1560
+    },
+    {
+      "epoch": 0.21646215359161727,
+      "grad_norm": 0.33651217818260193,
+      "learning_rate": 0.0002,
+      "loss": 0.349,
+      "step": 1570
+    },
+    {
+      "epoch": 0.217840893423411,
+      "grad_norm": 0.07309216260910034,
+      "learning_rate": 0.0002,
+      "loss": 0.3493,
+      "step": 1580
+    },
+    {
+      "epoch": 0.21921963325520474,
+      "grad_norm": 0.19346486032009125,
+      "learning_rate": 0.0002,
+      "loss": 0.3478,
+      "step": 1590
+    },
+    {
+      "epoch": 0.22059837308699848,
+      "grad_norm": 0.3398933708667755,
+      "learning_rate": 0.0002,
+      "loss": 0.3496,
+      "step": 1600
+    },
+    {
+      "epoch": 0.2219771129187922,
+      "grad_norm": 0.34032130241394043,
+      "learning_rate": 0.0002,
+      "loss": 0.3488,
+      "step": 1610
+    },
+    {
+      "epoch": 0.22335585275058598,
+      "grad_norm": 0.901030421257019,
+      "learning_rate": 0.0002,
+      "loss": 0.3531,
+      "step": 1620
+    },
+    {
+      "epoch": 0.2247345925823797,
+      "grad_norm": 0.500088632106781,
+      "learning_rate": 0.0002,
+      "loss": 0.3516,
+      "step": 1630
+    },
+    {
+      "epoch": 0.22611333241417345,
+      "grad_norm": 0.3230324387550354,
+      "learning_rate": 0.0002,
+      "loss": 0.3546,
+      "step": 1640
+    },
+    {
+      "epoch": 0.22749207224596718,
+      "grad_norm": 1.2476601600646973,
+      "learning_rate": 0.0002,
+      "loss": 0.3512,
+      "step": 1650
+    },
+    {
+      "epoch": 0.22887081207776092,
+      "grad_norm": 0.23318485915660858,
+      "learning_rate": 0.0002,
+      "loss": 0.3456,
+      "step": 1660
+    },
+    {
+      "epoch": 0.23024955190955468,
+      "grad_norm": 0.472400963306427,
+      "learning_rate": 0.0002,
+      "loss": 0.3551,
+      "step": 1670
+    },
+    {
+      "epoch": 0.2316282917413484,
+      "grad_norm": 0.04836912825703621,
+      "learning_rate": 0.0002,
+      "loss": 0.3505,
+      "step": 1680
+    },
+    {
+      "epoch": 0.23300703157314215,
+      "grad_norm": 0.34590113162994385,
+      "learning_rate": 0.0002,
+      "loss": 0.3507,
+      "step": 1690
+    },
+    {
+      "epoch": 0.23438577140493588,
+      "grad_norm": 0.23341989517211914,
+      "learning_rate": 0.0002,
+      "loss": 0.3478,
+      "step": 1700
+    },
+    {
+      "epoch": 0.23576451123672962,
+      "grad_norm": 0.001271920627914369,
+      "learning_rate": 0.0002,
+      "loss": 0.3562,
+      "step": 1710
+    },
+    {
+      "epoch": 0.23714325106852338,
+      "grad_norm": 0.20549911260604858,
+      "learning_rate": 0.0002,
+      "loss": 0.3479,
+      "step": 1720
+    },
+    {
+      "epoch": 0.23852199090031712,
+      "grad_norm": 0.3825775384902954,
+      "learning_rate": 0.0002,
+      "loss": 0.3482,
+      "step": 1730
+    },
+    {
+      "epoch": 0.23990073073211085,
+      "grad_norm": 0.028804048895835876,
+      "learning_rate": 0.0002,
+      "loss": 0.3481,
+      "step": 1740
+    },
+    {
+      "epoch": 0.24127947056390459,
+      "grad_norm": 0.04462611302733421,
+      "learning_rate": 0.0002,
+      "loss": 0.3477,
+      "step": 1750
+    },
+    {
+      "epoch": 0.24265821039569832,
+      "grad_norm": 0.6634818315505981,
+      "learning_rate": 0.0002,
+      "loss": 0.3501,
+      "step": 1760
+    },
+    {
+      "epoch": 0.24403695022749208,
+      "grad_norm": 1.3807406425476074,
+      "learning_rate": 0.0002,
+      "loss": 0.3547,
+      "step": 1770
+    },
+    {
+      "epoch": 0.24541569005928582,
+      "grad_norm": 0.24347831308841705,
+      "learning_rate": 0.0002,
+      "loss": 0.3473,
+      "step": 1780
+    },
+    {
+      "epoch": 0.24679442989107955,
+      "grad_norm": 0.61258465051651,
+      "learning_rate": 0.0002,
+      "loss": 0.3492,
+      "step": 1790
+    },
+    {
+      "epoch": 0.2481731697228733,
+      "grad_norm": 0.011543272994458675,
+      "learning_rate": 0.0002,
+      "loss": 0.347,
+      "step": 1800
+    },
+    {
+      "epoch": 0.24955190955466702,
+      "grad_norm": 0.09996844828128815,
+      "learning_rate": 0.0002,
+      "loss": 0.3474,
+      "step": 1810
+    },
+    {
+      "epoch": 0.25093064938646076,
+      "grad_norm": 0.17044603824615479,
+      "learning_rate": 0.0002,
+      "loss": 0.3471,
+      "step": 1820
+    },
+    {
+      "epoch": 0.2523093892182545,
+      "grad_norm": 0.17940489947795868,
+      "learning_rate": 0.0002,
+      "loss": 0.3478,
+      "step": 1830
+    },
+    {
+      "epoch": 0.2536881290500483,
+      "grad_norm": 0.21834205090999603,
+      "learning_rate": 0.0002,
+      "loss": 0.3461,
+      "step": 1840
+    },
+    {
+      "epoch": 0.255066868881842,
+      "grad_norm": 0.2272634655237198,
+      "learning_rate": 0.0002,
+      "loss": 0.347,
+      "step": 1850
+    },
+    {
+      "epoch": 0.25644560871363575,
+      "grad_norm": 0.18734070658683777,
+      "learning_rate": 0.0002,
+      "loss": 0.3497,
+      "step": 1860
+    },
+    {
+      "epoch": 0.25782434854542946,
+      "grad_norm": 0.04078834876418114,
+      "learning_rate": 0.0002,
+      "loss": 0.3493,
+      "step": 1870
+    },
+    {
+      "epoch": 0.2592030883772232,
+      "grad_norm": 0.3463903069496155,
+      "learning_rate": 0.0002,
+      "loss": 0.3494,
+      "step": 1880
+    },
+    {
+      "epoch": 0.260581828209017,
+      "grad_norm": 0.3256634771823883,
+      "learning_rate": 0.0002,
+      "loss": 0.3488,
+      "step": 1890
+    },
+    {
+      "epoch": 0.2619605680408107,
+      "grad_norm": 0.04404434189200401,
+      "learning_rate": 0.0002,
+      "loss": 0.3509,
+      "step": 1900
+    },
+    {
+      "epoch": 0.26333930787260446,
+      "grad_norm": 0.20446011424064636,
+      "learning_rate": 0.0002,
+      "loss": 0.3526,
+      "step": 1910
+    },
+    {
+      "epoch": 0.26471804770439816,
+      "grad_norm": 0.06089179962873459,
+      "learning_rate": 0.0002,
+      "loss": 0.3491,
+      "step": 1920
+    },
+    {
+      "epoch": 0.2660967875361919,
+      "grad_norm": 0.32661405205726624,
+      "learning_rate": 0.0002,
+      "loss": 0.35,
+      "step": 1930
+    },
+    {
+      "epoch": 0.2674755273679857,
+      "grad_norm": 0.09823151677846909,
+      "learning_rate": 0.0002,
+      "loss": 0.3482,
+      "step": 1940
+    },
+    {
+      "epoch": 0.2688542671997794,
+      "grad_norm": 0.11397412419319153,
+      "learning_rate": 0.0002,
+      "loss": 0.347,
+      "step": 1950
+    },
+    {
+      "epoch": 0.27023300703157316,
+      "grad_norm": 0.2632172703742981,
+      "learning_rate": 0.0002,
+      "loss": 0.352,
+      "step": 1960
+    },
+    {
+      "epoch": 0.27161174686336687,
+      "grad_norm": 0.27215296030044556,
+      "learning_rate": 0.0002,
+      "loss": 0.3482,
+      "step": 1970
+    },
+    {
+      "epoch": 0.27299048669516063,
+      "grad_norm": 0.20016005635261536,
+      "learning_rate": 0.0002,
+      "loss": 0.3489,
+      "step": 1980
+    },
+    {
+      "epoch": 0.2743692265269544,
+      "grad_norm": 0.3071637749671936,
+      "learning_rate": 0.0002,
+      "loss": 0.354,
+      "step": 1990
+    },
+    {
+      "epoch": 0.2757479663587481,
+      "grad_norm": 1.0373337268829346,
+      "learning_rate": 0.0002,
+      "loss": 0.3481,
+      "step": 2000
+    },
+    {
+      "epoch": 0.27712670619054186,
+      "grad_norm": 0.49023887515068054,
+      "learning_rate": 0.0002,
+      "loss": 0.3501,
+      "step": 2010
+    },
+    {
+      "epoch": 0.27850544602233557,
+      "grad_norm": 0.7107551097869873,
+      "learning_rate": 0.0002,
+      "loss": 0.3515,
+      "step": 2020
+    },
+    {
+      "epoch": 0.27988418585412933,
+      "grad_norm": 1.5392403602600098,
+      "learning_rate": 0.0002,
+      "loss": 0.3468,
+      "step": 2030
+    },
+    {
+      "epoch": 0.2812629256859231,
+      "grad_norm": 2.7259292602539062,
+      "learning_rate": 0.0002,
+      "loss": 0.3585,
+      "step": 2040
+    },
+    {
+      "epoch": 0.2826416655177168,
+      "grad_norm": 2.783911943435669,
+      "learning_rate": 0.0002,
+      "loss": 0.3703,
+      "step": 2050
+    },
+    {
+      "epoch": 0.28402040534951056,
+      "grad_norm": 0.015472312457859516,
+      "learning_rate": 0.0002,
+      "loss": 0.3506,
+      "step": 2060
+    },
+    {
+      "epoch": 0.28539914518130427,
+      "grad_norm": 0.7833682298660278,
+      "learning_rate": 0.0002,
+      "loss": 0.3483,
+      "step": 2070
+    },
+    {
+      "epoch": 0.28677788501309803,
+      "grad_norm": 0.4828648269176483,
+      "learning_rate": 0.0002,
+      "loss": 0.3479,
+      "step": 2080
+    },
+    {
+      "epoch": 0.2881566248448918,
+      "grad_norm": 1.035452961921692,
+      "learning_rate": 0.0002,
+      "loss": 0.348,
+      "step": 2090
+    },
+    {
+      "epoch": 0.2895353646766855,
+      "grad_norm": 0.14552044868469238,
+      "learning_rate": 0.0002,
+      "loss": 0.3478,
+      "step": 2100
+    },
+    {
+      "epoch": 0.29091410450847927,
+      "grad_norm": 0.9040600061416626,
+      "learning_rate": 0.0002,
+      "loss": 0.3508,
+      "step": 2110
+    },
+    {
+      "epoch": 0.292292844340273,
+      "grad_norm": 0.36334726214408875,
+      "learning_rate": 0.0002,
+      "loss": 0.3472,
+      "step": 2120
+    },
+    {
+      "epoch": 0.29367158417206674,
+      "grad_norm": 0.6746371984481812,
+      "learning_rate": 0.0002,
+      "loss": 0.3546,
+      "step": 2130
+    },
+    {
+      "epoch": 0.2950503240038605,
+      "grad_norm": 0.49551185965538025,
+      "learning_rate": 0.0002,
+      "loss": 0.3464,
+      "step": 2140
+    },
+    {
+      "epoch": 0.2964290638356542,
+      "grad_norm": 0.04471205919981003,
+      "learning_rate": 0.0002,
+      "loss": 0.347,
+      "step": 2150
+    },
+    {
+      "epoch": 0.29780780366744797,
+      "grad_norm": 0.12050581723451614,
+      "learning_rate": 0.0002,
+      "loss": 0.3475,
+      "step": 2160
+    },
+    {
+      "epoch": 0.2991865434992417,
+      "grad_norm": 0.04496219381690025,
+      "learning_rate": 0.0002,
+      "loss": 0.346,
+      "step": 2170
+    },
+    {
+      "epoch": 0.30056528333103544,
+      "grad_norm": 0.11529727280139923,
+      "learning_rate": 0.0002,
+      "loss": 0.3475,
+      "step": 2180
+    },
+    {
+      "epoch": 0.3019440231628292,
+      "grad_norm": 0.12811416387557983,
+      "learning_rate": 0.0002,
+      "loss": 0.3466,
+      "step": 2190
+    },
+    {
+      "epoch": 0.3033227629946229,
+      "grad_norm": 0.0471930056810379,
+      "learning_rate": 0.0002,
+      "loss": 0.3468,
+      "step": 2200
+    },
+    {
+      "epoch": 0.30470150282641667,
+      "grad_norm": 2.6910486221313477,
+      "learning_rate": 0.0002,
+      "loss": 0.351,
+      "step": 2210
+    },
+    {
+      "epoch": 0.3060802426582104,
+      "grad_norm": 0.3302823305130005,
+      "learning_rate": 0.0002,
+      "loss": 0.3498,
+      "step": 2220
+    },
+    {
+      "epoch": 0.30745898249000414,
+      "grad_norm": 10.41483211517334,
+      "learning_rate": 0.0002,
+      "loss": 0.3694,
+      "step": 2230
+    },
+    {
+      "epoch": 0.3088377223217979,
+      "grad_norm": 12.485153198242188,
+      "learning_rate": 0.0002,
+      "loss": 0.3816,
+      "step": 2240
+    },
+    {
+      "epoch": 0.3102164621535916,
+      "grad_norm": 1.1303633451461792,
+      "learning_rate": 0.0002,
+      "loss": 0.3554,
+      "step": 2250
+    },
+    {
+      "epoch": 0.3115952019853854,
+      "grad_norm": 0.08931141346693039,
+      "learning_rate": 0.0002,
+      "loss": 0.3824,
+      "step": 2260
+    },
+    {
+      "epoch": 0.3129739418171791,
+      "grad_norm": 0.20106171071529388,
+      "learning_rate": 0.0002,
+      "loss": 0.3483,
+      "step": 2270
+    },
+    {
+      "epoch": 0.31435268164897284,
+      "grad_norm": 1.7839974164962769,
+      "learning_rate": 0.0002,
+      "loss": 0.354,
+      "step": 2280
+    },
+    {
+      "epoch": 0.3157314214807666,
+      "grad_norm": 0.6848243474960327,
+      "learning_rate": 0.0002,
+      "loss": 0.356,
+      "step": 2290
+    },
+    {
+      "epoch": 0.3171101613125603,
+      "grad_norm": 1.5771201848983765,
+      "learning_rate": 0.0002,
+      "loss": 0.35,
+      "step": 2300
+    },
+    {
+      "epoch": 0.3184889011443541,
+      "grad_norm": 1.2237290143966675,
+      "learning_rate": 0.0002,
+      "loss": 0.3521,
+      "step": 2310
+    },
+    {
+      "epoch": 0.3198676409761478,
+      "grad_norm": 0.04607740044593811,
+      "learning_rate": 0.0002,
+      "loss": 0.3455,
+      "step": 2320
+    },
+    {
+      "epoch": 0.32124638080794155,
+      "grad_norm": 0.0691566988825798,
+      "learning_rate": 0.0002,
+      "loss": 0.6835,
+      "step": 2330
+    },
+    {
+      "epoch": 0.3226251206397353,
+      "grad_norm": 0.17569975554943085,
+      "learning_rate": 0.0002,
+      "loss": 0.3471,
+      "step": 2340
+    },
+    {
+      "epoch": 0.324003860471529,
+      "grad_norm": 0.23300354182720184,
+      "learning_rate": 0.0002,
+      "loss": 0.3496,
+      "step": 2350
+    },
+    {
+      "epoch": 0.3253826003033228,
+      "grad_norm": 0.2859944999217987,
+      "learning_rate": 0.0002,
+      "loss": 0.3544,
+      "step": 2360
+    },
+    {
+      "epoch": 0.3267613401351165,
+      "grad_norm": 140.8813018798828,
+      "learning_rate": 0.0002,
+      "loss": 0.5267,
+      "step": 2370
+    },
+    {
+      "epoch": 0.32814007996691025,
+      "grad_norm": 0.4637663960456848,
+      "learning_rate": 0.0002,
+      "loss": 0.3538,
+      "step": 2380
+    },
+    {
+      "epoch": 0.329518819798704,
+      "grad_norm": 0.2639578580856323,
+      "learning_rate": 0.0002,
+      "loss": 0.3482,
+      "step": 2390
+    },
+    {
+      "epoch": 0.3308975596304977,
+      "grad_norm": 0.02536751516163349,
+      "learning_rate": 0.0002,
+      "loss": 0.3468,
+      "step": 2400
+    },
+    {
+      "epoch": 0.3322762994622915,
+      "grad_norm": 0.20604869723320007,
+      "learning_rate": 0.0002,
+      "loss": 0.3494,
+      "step": 2410
+    },
+    {
+      "epoch": 0.3336550392940852,
+      "grad_norm": 0.3395155668258667,
+      "learning_rate": 0.0002,
+      "loss": 0.3492,
+      "step": 2420
+    },
+    {
+      "epoch": 0.33503377912587895,
+      "grad_norm": 0.07880198955535889,
+      "learning_rate": 0.0002,
+      "loss": 0.3474,
+      "step": 2430
+    },
+    {
+      "epoch": 0.3364125189576727,
+      "grad_norm": 0.18549232184886932,
+      "learning_rate": 0.0002,
+      "loss": 0.3515,
+      "step": 2440
+    },
+    {
+      "epoch": 0.3377912587894664,
+      "grad_norm": 0.030663492158055305,
+      "learning_rate": 0.0002,
+      "loss": 0.3475,
+      "step": 2450
+    },
+    {
+      "epoch": 0.3391699986212602,
+      "grad_norm": 0.2116040587425232,
+      "learning_rate": 0.0002,
+      "loss": 0.3484,
+      "step": 2460
+    },
+    {
+      "epoch": 0.3405487384530539,
+      "grad_norm": 0.06269362568855286,
+      "learning_rate": 0.0002,
+      "loss": 0.3476,
+      "step": 2470
+    },
+    {
+      "epoch": 0.34192747828484765,
+      "grad_norm": 0.13339760899543762,
+      "learning_rate": 0.0002,
+      "loss": 0.3477,
+      "step": 2480
+    },
+    {
+      "epoch": 0.3433062181166414,
+      "grad_norm": 0.3202485740184784,
+      "learning_rate": 0.0002,
+      "loss": 0.3477,
+      "step": 2490
+    },
+    {
+      "epoch": 0.3446849579484351,
+      "grad_norm": 0.2552681565284729,
+      "learning_rate": 0.0002,
+      "loss": 0.3488,
+      "step": 2500
+    },
+    {
+      "epoch": 0.3460636977802289,
+      "grad_norm": 0.5942317247390747,
+      "learning_rate": 0.0002,
+      "loss": 0.3513,
+      "step": 2510
+    },
+    {
+      "epoch": 0.3474424376120226,
+      "grad_norm": 0.865836501121521,
+      "learning_rate": 0.0002,
+      "loss": 0.3517,
+      "step": 2520
+    },
+    {
+      "epoch": 0.34882117744381635,
+      "grad_norm": 0.47955092787742615,
+      "learning_rate": 0.0002,
+      "loss": 0.3508,
+      "step": 2530
+    },
+    {
+      "epoch": 0.3501999172756101,
+      "grad_norm": 0.07025259733200073,
+      "learning_rate": 0.0002,
+      "loss": 0.3476,
+      "step": 2540
+    },
+    {
+      "epoch": 0.3515786571074038,
+      "grad_norm": 0.08550610393285751,
+      "learning_rate": 0.0002,
+      "loss": 0.3483,
+      "step": 2550
+    },
+    {
+      "epoch": 0.3529573969391976,
+      "grad_norm": 0.07628582417964935,
+      "learning_rate": 0.0002,
+      "loss": 0.3503,
+      "step": 2560
+    },
+    {
+      "epoch": 0.3543361367709913,
+      "grad_norm": 0.09870871156454086,
+      "learning_rate": 0.0002,
+      "loss": 0.3506,
+      "step": 2570
+    },
+    {
+      "epoch": 0.35571487660278506,
+      "grad_norm": 0.03877127543091774,
+      "learning_rate": 0.0002,
+      "loss": 0.3477,
+      "step": 2580
+    },
+    {
+      "epoch": 0.3570936164345788,
+      "grad_norm": 0.01288803294301033,
+      "learning_rate": 0.0002,
+      "loss": 0.3472,
+      "step": 2590
+    },
+    {
+      "epoch": 0.3584723562663725,
+      "grad_norm": 0.2647448778152466,
+      "learning_rate": 0.0002,
+      "loss": 0.3499,
+      "step": 2600
+    },
+    {
+      "epoch": 0.3598510960981663,
+      "grad_norm": 0.5602620840072632,
+      "learning_rate": 0.0002,
+      "loss": 0.3478,
+      "step": 2610
+    },
+    {
+      "epoch": 0.36122983592996,
+      "grad_norm": 0.788970947265625,
+      "learning_rate": 0.0002,
+      "loss": 0.3513,
+      "step": 2620
+    },
+    {
+      "epoch": 0.36260857576175376,
+      "grad_norm": 0.6836252212524414,
+      "learning_rate": 0.0002,
+      "loss": 0.3502,
+      "step": 2630
+    },
+    {
+      "epoch": 0.3639873155935475,
+      "grad_norm": 0.14788776636123657,
+      "learning_rate": 0.0002,
+      "loss": 0.3654,
+      "step": 2640
+    },
+    {
+      "epoch": 0.36536605542534123,
+      "grad_norm": 0.7003443837165833,
+      "learning_rate": 0.0002,
+      "loss": 0.3526,
+      "step": 2650
+    },
+    {
+      "epoch": 0.366744795257135,
+      "grad_norm": 0.600309431552887,
+      "learning_rate": 0.0002,
+      "loss": 0.3475,
+      "step": 2660
+    },
+    {
+      "epoch": 0.3681235350889287,
+      "grad_norm": 0.10942361503839493,
+      "learning_rate": 0.0002,
+      "loss": 0.3467,
+      "step": 2670
+    },
+    {
+      "epoch": 0.36950227492072246,
+      "grad_norm": 0.2913811504840851,
+      "learning_rate": 0.0002,
+      "loss": 0.3473,
+      "step": 2680
+    },
+    {
+      "epoch": 0.3708810147525162,
+      "grad_norm": 0.11519110947847366,
+      "learning_rate": 0.0002,
+      "loss": 0.3492,
+      "step": 2690
+    },
+    {
+      "epoch": 0.37225975458430993,
+      "grad_norm": 0.04020654037594795,
+      "learning_rate": 0.0002,
+      "loss": 0.3465,
+      "step": 2700
+    },
+    {
+      "epoch": 0.3736384944161037,
+      "grad_norm": 0.3286079168319702,
+      "learning_rate": 0.0002,
+      "loss": 0.3481,
+      "step": 2710
+    },
+    {
+      "epoch": 0.3750172342478974,
+      "grad_norm": 0.09239518642425537,
+      "learning_rate": 0.0002,
+      "loss": 0.3478,
+      "step": 2720
+    },
+    {
+      "epoch": 0.37639597407969116,
+      "grad_norm": 0.18499009311199188,
+      "learning_rate": 0.0002,
+      "loss": 0.3474,
+      "step": 2730
+    },
+    {
+      "epoch": 0.3777747139114849,
+      "grad_norm": 0.031250640749931335,
+      "learning_rate": 0.0002,
+      "loss": 0.3478,
+      "step": 2740
+    },
+    {
+      "epoch": 0.37915345374327863,
+      "grad_norm": 0.013040668331086636,
+      "learning_rate": 0.0002,
+      "loss": 0.3463,
+      "step": 2750
+    },
+    {
+      "epoch": 0.3805321935750724,
+      "grad_norm": 0.11167818307876587,
+      "learning_rate": 0.0002,
+      "loss": 0.35,
+      "step": 2760
+    },
+    {
+      "epoch": 0.3819109334068661,
+      "grad_norm": 0.9923171401023865,
+      "learning_rate": 0.0002,
+      "loss": 0.3483,
+      "step": 2770
+    },
+    {
+      "epoch": 0.38328967323865987,
+      "grad_norm": 0.7532585263252258,
+      "learning_rate": 0.0002,
+      "loss": 0.3493,
+      "step": 2780
+    },
+    {
+      "epoch": 0.38466841307045363,
+      "grad_norm": 0.17266100645065308,
+      "learning_rate": 0.0002,
+      "loss": 0.3472,
+      "step": 2790
+    },
+    {
+      "epoch": 0.38604715290224734,
+      "grad_norm": 0.07107715308666229,
+      "learning_rate": 0.0002,
+      "loss": 0.3465,
+      "step": 2800
+    },
+    {
+      "epoch": 0.3874258927340411,
+      "grad_norm": 0.05912484973669052,
+      "learning_rate": 0.0002,
+      "loss": 0.3454,
+      "step": 2810
+    },
+    {
+      "epoch": 0.3888046325658348,
+      "grad_norm": 0.18973538279533386,
+      "learning_rate": 0.0002,
+      "loss": 0.3485,
+      "step": 2820
+    },
+    {
+      "epoch": 0.39018337239762857,
+      "grad_norm": 0.3309797942638397,
+      "learning_rate": 0.0002,
+      "loss": 0.3477,
+      "step": 2830
+    },
+    {
+      "epoch": 0.39156211222942233,
+      "grad_norm": 0.1095886304974556,
+      "learning_rate": 0.0002,
+      "loss": 0.3479,
+      "step": 2840
+    },
+    {
+      "epoch": 0.39294085206121604,
+      "grad_norm": 0.7054705619812012,
+      "learning_rate": 0.0002,
+      "loss": 0.3457,
+      "step": 2850
+    },
+    {
+      "epoch": 0.3943195918930098,
+      "grad_norm": 0.35234084725379944,
+      "learning_rate": 0.0002,
+      "loss": 0.3519,
+      "step": 2860
+    },
+    {
+      "epoch": 0.3956983317248035,
+      "grad_norm": 1.1833308935165405,
+      "learning_rate": 0.0002,
+      "loss": 0.3461,
+      "step": 2870
+    },
+    {
+      "epoch": 0.39707707155659727,
+      "grad_norm": 0.046266939491033554,
+      "learning_rate": 0.0002,
+      "loss": 0.3478,
+      "step": 2880
+    },
+    {
+      "epoch": 0.39845581138839103,
+      "grad_norm": 0.47695428133010864,
+      "learning_rate": 0.0002,
+      "loss": 0.3494,
+      "step": 2890
+    },
+    {
+      "epoch": 0.39983455122018474,
+      "grad_norm": 0.14398415386676788,
+      "learning_rate": 0.0002,
+      "loss": 0.347,
+      "step": 2900
+    },
+    {
+      "epoch": 0.4012132910519785,
+      "grad_norm": 0.0672549456357956,
+      "learning_rate": 0.0002,
+      "loss": 0.3446,
+      "step": 2910
+    },
+    {
+      "epoch": 0.4025920308837722,
+      "grad_norm": 0.7201761603355408,
+      "learning_rate": 0.0002,
+      "loss": 0.3536,
+      "step": 2920
+    },
+    {
+      "epoch": 0.403970770715566,
+      "grad_norm": 1.0536067485809326,
+      "learning_rate": 0.0002,
+      "loss": 0.3448,
+      "step": 2930
+    },
+    {
+      "epoch": 0.40534951054735974,
+      "grad_norm": 0.4652802050113678,
+      "learning_rate": 0.0002,
+      "loss": 0.3572,
+      "step": 2940
+    },
+    {
+      "epoch": 0.40672825037915344,
+      "grad_norm": 0.2714202404022217,
+      "learning_rate": 0.0002,
+      "loss": 0.3506,
+      "step": 2950
+    },
+    {
+      "epoch": 0.4081069902109472,
+      "grad_norm": 0.056970611214637756,
+      "learning_rate": 0.0002,
+      "loss": 0.3466,
+      "step": 2960
+    },
+    {
+      "epoch": 0.4094857300427409,
+      "grad_norm": 0.08109589666128159,
+      "learning_rate": 0.0002,
+      "loss": 0.3478,
+      "step": 2970
+    },
+    {
+      "epoch": 0.4108644698745347,
+      "grad_norm": 0.2987334728240967,
+      "learning_rate": 0.0002,
+      "loss": 0.3482,
+      "step": 2980
+    },
+    {
+      "epoch": 0.41224320970632844,
+      "grad_norm": 0.3519710898399353,
+      "learning_rate": 0.0002,
+      "loss": 0.351,
+      "step": 2990
+    },
+    {
+      "epoch": 0.41362194953812215,
+      "grad_norm": 0.20754824578762054,
+      "learning_rate": 0.0002,
+      "loss": 0.3455,
+      "step": 3000
+    },
+    {
+      "epoch": 0.4150006893699159,
+      "grad_norm": 0.00901414267718792,
+      "learning_rate": 0.0002,
+      "loss": 0.3497,
+      "step": 3010
+    },
+    {
+      "epoch": 0.4163794292017096,
+      "grad_norm": 0.5444782972335815,
+      "learning_rate": 0.0002,
+      "loss": 0.3698,
+      "step": 3020
+    },
+    {
+      "epoch": 0.4177581690335034,
+      "grad_norm": 1.026206374168396,
+      "learning_rate": 0.0002,
+      "loss": 0.3516,
+      "step": 3030
+    },
+    {
+      "epoch": 0.41913690886529714,
+      "grad_norm": 0.4680636525154114,
+      "learning_rate": 0.0002,
+      "loss": 0.3488,
+      "step": 3040
+    },
+    {
+      "epoch": 0.42051564869709085,
+      "grad_norm": 0.07180967926979065,
+      "learning_rate": 0.0002,
+      "loss": 0.3491,
+      "step": 3050
+    },
+    {
+      "epoch": 0.4218943885288846,
+      "grad_norm": 1.1045209169387817,
+      "learning_rate": 0.0002,
+      "loss": 0.3497,
+      "step": 3060
+    },
+    {
+      "epoch": 0.4232731283606783,
+      "grad_norm": 1.6731029748916626,
+      "learning_rate": 0.0002,
+      "loss": 0.3587,
+      "step": 3070
+    },
+    {
+      "epoch": 0.4246518681924721,
+      "grad_norm": 0.44295939803123474,
+      "learning_rate": 0.0002,
+      "loss": 0.348,
+      "step": 3080
+    },
+    {
+      "epoch": 0.42603060802426584,
+      "grad_norm": 0.011876541189849377,
+      "learning_rate": 0.0002,
+      "loss": 0.3471,
+      "step": 3090
+    },
+    {
+      "epoch": 0.42740934785605955,
+      "grad_norm": 0.5493383407592773,
+      "learning_rate": 0.0002,
+      "loss": 0.3474,
+      "step": 3100
+    },
+    {
+      "epoch": 0.4287880876878533,
+      "grad_norm": 0.013986635021865368,
+      "learning_rate": 0.0002,
+      "loss": 0.3467,
+      "step": 3110
+    },
+    {
+      "epoch": 0.430166827519647,
+      "grad_norm": 0.11136174947023392,
+      "learning_rate": 0.0002,
+      "loss": 0.3465,
+      "step": 3120
+    },
+    {
+      "epoch": 0.4315455673514408,
+      "grad_norm": 0.18628475069999695,
+      "learning_rate": 0.0002,
+      "loss": 0.347,
+      "step": 3130
+    },
+    {
+      "epoch": 0.43292430718323455,
+      "grad_norm": 0.07927969843149185,
+      "learning_rate": 0.0002,
+      "loss": 0.3477,
+      "step": 3140
+    },
+    {
+      "epoch": 0.43430304701502825,
+      "grad_norm": 0.32330620288848877,
+      "learning_rate": 0.0002,
+      "loss": 0.3477,
+      "step": 3150
+    },
+    {
+      "epoch": 0.435681786846822,
+      "grad_norm": 0.4383370578289032,
+      "learning_rate": 0.0002,
+      "loss": 0.3491,
+      "step": 3160
+    },
+    {
+      "epoch": 0.4370605266786157,
+      "grad_norm": 0.5952053070068359,
+      "learning_rate": 0.0002,
+      "loss": 0.3482,
+      "step": 3170
+    },
+    {
+      "epoch": 0.4384392665104095,
+      "grad_norm": 1.106527328491211,
+      "learning_rate": 0.0002,
+      "loss": 0.3569,
+      "step": 3180
+    },
+    {
+      "epoch": 0.43981800634220325,
+      "grad_norm": 0.6171185970306396,
+      "learning_rate": 0.0002,
+      "loss": 0.3474,
+      "step": 3190
+    },
+    {
+      "epoch": 0.44119674617399696,
+      "grad_norm": 0.16300754249095917,
+      "learning_rate": 0.0002,
+      "loss": 0.3474,
+      "step": 3200
+    },
+    {
+      "epoch": 0.4425754860057907,
+      "grad_norm": 0.03562415391206741,
+      "learning_rate": 0.0002,
+      "loss": 0.3474,
+      "step": 3210
+    },
+    {
+      "epoch": 0.4439542258375844,
+      "grad_norm": 0.30006295442581177,
+      "learning_rate": 0.0002,
+      "loss": 0.3475,
+      "step": 3220
+    },
+    {
+      "epoch": 0.4453329656693782,
+      "grad_norm": 0.4837903678417206,
+      "learning_rate": 0.0002,
+      "loss": 0.3618,
+      "step": 3230
+    },
+    {
+      "epoch": 0.44671170550117195,
+      "grad_norm": 0.4861944317817688,
+      "learning_rate": 0.0002,
+      "loss": 0.3536,
+      "step": 3240
+    },
+    {
+      "epoch": 0.44809044533296566,
+      "grad_norm": 1.048814296722412,
+      "learning_rate": 0.0002,
+      "loss": 0.3538,
+      "step": 3250
+    },
+    {
+      "epoch": 0.4494691851647594,
+      "grad_norm": 0.6691413521766663,
+      "learning_rate": 0.0002,
+      "loss": 0.3515,
+      "step": 3260
+    },
+    {
+      "epoch": 0.45084792499655313,
+      "grad_norm": 0.24432829022407532,
+      "learning_rate": 0.0002,
+      "loss": 0.3561,
+      "step": 3270
+    },
+    {
+      "epoch": 0.4522266648283469,
+      "grad_norm": 0.529619038105011,
+      "learning_rate": 0.0002,
+      "loss": 0.3488,
+      "step": 3280
+    },
+    {
+      "epoch": 0.45360540466014065,
+      "grad_norm": 0.10819733142852783,
+      "learning_rate": 0.0002,
+      "loss": 0.3464,
+      "step": 3290
+    },
+    {
+      "epoch": 0.45498414449193436,
+      "grad_norm": 0.3517175614833832,
+      "learning_rate": 0.0002,
+      "loss": 0.347,
+      "step": 3300
+    },
+    {
+      "epoch": 0.4563628843237281,
+      "grad_norm": 0.23464784026145935,
+      "learning_rate": 0.0002,
+      "loss": 0.3484,
+      "step": 3310
+    },
+    {
+      "epoch": 0.45774162415552183,
+      "grad_norm": 0.1873646080493927,
+      "learning_rate": 0.0002,
+      "loss": 0.3517,
+      "step": 3320
+    },
+    {
+      "epoch": 0.4591203639873156,
+      "grad_norm": 0.20844773948192596,
+      "learning_rate": 0.0002,
+      "loss": 0.352,
+      "step": 3330
+    },
+    {
+      "epoch": 0.46049910381910936,
+      "grad_norm": 0.45384037494659424,
+      "learning_rate": 0.0002,
+      "loss": 0.3492,
+      "step": 3340
+    },
+    {
+      "epoch": 0.46187784365090306,
+      "grad_norm": 0.32738834619522095,
+      "learning_rate": 0.0002,
+      "loss": 0.3509,
+      "step": 3350
+    },
+    {
+      "epoch": 0.4632565834826968,
+      "grad_norm": 1.4738928079605103,
+      "learning_rate": 0.0002,
+      "loss": 0.3483,
+      "step": 3360
+    },
+    {
+      "epoch": 0.46463532331449053,
+      "grad_norm": 0.5815828442573547,
+      "learning_rate": 0.0002,
+      "loss": 0.3509,
+      "step": 3370
+    },
+    {
+      "epoch": 0.4660140631462843,
+      "grad_norm": 0.6721034049987793,
+      "learning_rate": 0.0002,
+      "loss": 0.3467,
+      "step": 3380
+    },
+    {
+      "epoch": 0.46739280297807806,
+      "grad_norm": 0.8967618346214294,
+      "learning_rate": 0.0002,
+      "loss": 0.3548,
+      "step": 3390
+    },
+    {
+      "epoch": 0.46877154280987177,
+      "grad_norm": 2.0388050079345703,
+      "learning_rate": 0.0002,
+      "loss": 0.3612,
+      "step": 3400
+    },
+    {
+      "epoch": 0.47015028264166553,
+      "grad_norm": 0.2570609450340271,
+      "learning_rate": 0.0002,
+      "loss": 0.3489,
+      "step": 3410
+    },
+    {
+      "epoch": 0.47152902247345924,
+      "grad_norm": 0.6751313209533691,
+      "learning_rate": 0.0002,
+      "loss": 0.3486,
+      "step": 3420
+    },
+    {
+      "epoch": 0.472907762305253,
+      "grad_norm": 0.4083470106124878,
+      "learning_rate": 0.0002,
+      "loss": 0.3471,
+      "step": 3430
+    },
+    {
+      "epoch": 0.47428650213704676,
+      "grad_norm": 0.09502824395895004,
+      "learning_rate": 0.0002,
+      "loss": 0.3479,
+      "step": 3440
+    },
+    {
+      "epoch": 0.47566524196884047,
+      "grad_norm": 0.40624290704727173,
+      "learning_rate": 0.0002,
+      "loss": 0.3484,
+      "step": 3450
+    },
+    {
+      "epoch": 0.47704398180063423,
+      "grad_norm": 0.1163792535662651,
+      "learning_rate": 0.0002,
+      "loss": 0.3469,
+      "step": 3460
+    },
+    {
+      "epoch": 0.47842272163242794,
+      "grad_norm": 0.06055830419063568,
+      "learning_rate": 0.0002,
+      "loss": 0.3476,
+      "step": 3470
+    },
+    {
+      "epoch": 0.4798014614642217,
+      "grad_norm": 0.21893176436424255,
+      "learning_rate": 0.0002,
+      "loss": 0.3465,
+      "step": 3480
+    },
+    {
+      "epoch": 0.48118020129601546,
+      "grad_norm": 0.24376049637794495,
+      "learning_rate": 0.0002,
+      "loss": 0.3497,
+      "step": 3490
+    },
+    {
+      "epoch": 0.48255894112780917,
+      "grad_norm": 0.14530010521411896,
+      "learning_rate": 0.0002,
+      "loss": 0.3441,
+      "step": 3500
+    },
+    {
+      "epoch": 0.48393768095960293,
+      "grad_norm": 0.4784521460533142,
+      "learning_rate": 0.0002,
+      "loss": 0.3585,
+      "step": 3510
+    },
+    {
+      "epoch": 0.48531642079139664,
+      "grad_norm": 1.5072555541992188,
+      "learning_rate": 0.0002,
+      "loss": 0.352,
+      "step": 3520
+    },
+    {
+      "epoch": 0.4866951606231904,
+      "grad_norm": 1.2513431310653687,
+      "learning_rate": 0.0002,
+      "loss": 0.3495,
+      "step": 3530
+    },
+    {
+      "epoch": 0.48807390045498417,
+      "grad_norm": 1.5765403509140015,
+      "learning_rate": 0.0002,
+      "loss": 0.3575,
+      "step": 3540
+    },
+    {
+      "epoch": 0.4894526402867779,
+      "grad_norm": 2.110595703125,
+      "learning_rate": 0.0002,
+      "loss": 0.3464,
+      "step": 3550
+    },
+    {
+      "epoch": 0.49083138011857164,
+      "grad_norm": 3.3377187252044678,
+      "learning_rate": 0.0002,
+      "loss": 0.3518,
+      "step": 3560
+    },
+    {
+      "epoch": 0.49221011995036534,
+      "grad_norm": 1.1348721981048584,
+      "learning_rate": 0.0002,
+      "loss": 0.3461,
+      "step": 3570
+    },
+    {
+      "epoch": 0.4935888597821591,
+      "grad_norm": 1.115633249282837,
+      "learning_rate": 0.0002,
+      "loss": 0.3477,
+      "step": 3580
+    },
+    {
+      "epoch": 0.49496759961395287,
+      "grad_norm": 0.46544066071510315,
+      "learning_rate": 0.0002,
+      "loss": 0.3508,
+      "step": 3590
+    },
+    {
+      "epoch": 0.4963463394457466,
+      "grad_norm": 1.0968178510665894,
+      "learning_rate": 0.0002,
+      "loss": 0.3491,
+      "step": 3600
+    },
+    {
+      "epoch": 0.49772507927754034,
+      "grad_norm": 0.1590128093957901,
+      "learning_rate": 0.0002,
+      "loss": 0.3493,
+      "step": 3610
+    },
+    {
+      "epoch": 0.49910381910933405,
+      "grad_norm": 0.0618724524974823,
+      "learning_rate": 0.0002,
+      "loss": 0.3467,
+      "step": 3620
+    },
+    {
+      "epoch": 0.5004825589411278,
+      "grad_norm": 0.038633331656455994,
+      "learning_rate": 0.0002,
+      "loss": 0.347,
+      "step": 3630
+    },
+    {
+      "epoch": 0.5018612987729215,
+      "grad_norm": 0.08056730031967163,
+      "learning_rate": 0.0002,
+      "loss": 0.3469,
+      "step": 3640
+    },
+    {
+      "epoch": 0.5032400386047153,
+      "grad_norm": 0.07540078461170197,
+      "learning_rate": 0.0002,
+      "loss": 0.347,
+      "step": 3650
+    },
+    {
+      "epoch": 0.504618778436509,
+      "grad_norm": 0.0971425250172615,
+      "learning_rate": 0.0002,
+      "loss": 0.349,
+      "step": 3660
+    },
+    {
+      "epoch": 0.5059975182683027,
+      "grad_norm": 0.1605909764766693,
+      "learning_rate": 0.0002,
+      "loss": 0.3472,
+      "step": 3670
+    },
+    {
+      "epoch": 0.5073762581000966,
+      "grad_norm": 0.2239840030670166,
+      "learning_rate": 0.0002,
+      "loss": 0.3489,
+      "step": 3680
+    },
+    {
+      "epoch": 0.5087549979318903,
+      "grad_norm": 0.25974148511886597,
+      "learning_rate": 0.0002,
+      "loss": 0.3478,
+      "step": 3690
+    },
+    {
+      "epoch": 0.510133737763684,
+      "grad_norm": 0.7621527314186096,
+      "learning_rate": 0.0002,
+      "loss": 0.3501,
+      "step": 3700
+    },
+    {
+      "epoch": 0.5115124775954777,
+      "grad_norm": 0.6531065702438354,
+      "learning_rate": 0.0002,
+      "loss": 0.3473,
+      "step": 3710
+    },
+    {
+      "epoch": 0.5128912174272715,
+      "grad_norm": 0.0937981903553009,
+      "learning_rate": 0.0002,
+      "loss": 0.3494,
+      "step": 3720
+    },
+    {
+      "epoch": 0.5142699572590652,
+      "grad_norm": 0.353502094745636,
+      "learning_rate": 0.0002,
+      "loss": 0.3487,
+      "step": 3730
+    },
+    {
+      "epoch": 0.5156486970908589,
+      "grad_norm": 0.5151197910308838,
+      "learning_rate": 0.0002,
+      "loss": 0.3485,
+      "step": 3740
+    },
+    {
+      "epoch": 0.5170274369226527,
+      "grad_norm": 0.15747584402561188,
+      "learning_rate": 0.0002,
+      "loss": 0.3473,
+      "step": 3750
+    },
+    {
+      "epoch": 0.5184061767544464,
+      "grad_norm": 0.18814824521541595,
+      "learning_rate": 0.0002,
+      "loss": 0.3474,
+      "step": 3760
+    },
+    {
+      "epoch": 0.5197849165862402,
+      "grad_norm": 0.040331218391656876,
+      "learning_rate": 0.0002,
+      "loss": 0.3474,
+      "step": 3770
+    },
+    {
+      "epoch": 0.521163656418034,
+      "grad_norm": 0.21510355174541473,
+      "learning_rate": 0.0002,
+      "loss": 0.3471,
+      "step": 3780
+    },
+    {
+      "epoch": 0.5225423962498277,
+      "grad_norm": 0.12755675613880157,
+      "learning_rate": 0.0002,
+      "loss": 0.3477,
+      "step": 3790
+    },
+    {
+      "epoch": 0.5239211360816214,
+      "grad_norm": 0.5592456459999084,
+      "learning_rate": 0.0002,
+      "loss": 0.3509,
+      "step": 3800
+    },
+    {
+      "epoch": 0.5252998759134151,
+      "grad_norm": 0.30668896436691284,
+      "learning_rate": 0.0002,
+      "loss": 0.3475,
+      "step": 3810
+    },
+    {
+      "epoch": 0.5266786157452089,
+      "grad_norm": 0.12264347821474075,
+      "learning_rate": 0.0002,
+      "loss": 0.3691,
+      "step": 3820
+    },
+    {
+      "epoch": 0.5280573555770026,
+      "grad_norm": 0.21281813085079193,
+      "learning_rate": 0.0002,
+      "loss": 0.3493,
+      "step": 3830
+    },
+    {
+      "epoch": 0.5294360954087963,
+      "grad_norm": 0.547156572341919,
+      "learning_rate": 0.0002,
+      "loss": 0.3485,
+      "step": 3840
+    },
+    {
+      "epoch": 0.5308148352405901,
+      "grad_norm": 0.2829332947731018,
+      "learning_rate": 0.0002,
+      "loss": 0.3459,
+      "step": 3850
+    },
+    {
+      "epoch": 0.5321935750723839,
+      "grad_norm": 0.6659385561943054,
+      "learning_rate": 0.0002,
+      "loss": 0.3492,
+      "step": 3860
+    },
+    {
+      "epoch": 0.5335723149041776,
+      "grad_norm": 0.9354788064956665,
+      "learning_rate": 0.0002,
+      "loss": 0.3501,
+      "step": 3870
+    },
+    {
+      "epoch": 0.5349510547359714,
+      "grad_norm": 0.64557945728302,
+      "learning_rate": 0.0002,
+      "loss": 0.3544,
+      "step": 3880
+    },
+    {
+      "epoch": 0.5363297945677651,
+      "grad_norm": 0.07203350216150284,
+      "learning_rate": 0.0002,
+      "loss": 0.3485,
+      "step": 3890
+    },
+    {
+      "epoch": 0.5377085343995588,
+      "grad_norm": 0.8220388293266296,
+      "learning_rate": 0.0002,
+      "loss": 0.3472,
+      "step": 3900
+    },
+    {
+      "epoch": 0.5390872742313525,
+      "grad_norm": 0.2928883135318756,
+      "learning_rate": 0.0002,
+      "loss": 0.3549,
+      "step": 3910
+    },
+    {
+      "epoch": 0.5404660140631463,
+      "grad_norm": 0.14233669638633728,
+      "learning_rate": 0.0002,
+      "loss": 0.3465,
+      "step": 3920
+    },
+    {
+      "epoch": 0.54184475389494,
+      "grad_norm": 0.3891246020793915,
+      "learning_rate": 0.0002,
+      "loss": 0.3463,
+      "step": 3930
+    },
+    {
+      "epoch": 0.5432234937267337,
+      "grad_norm": 1.5544075965881348,
+      "learning_rate": 0.0002,
+      "loss": 0.357,
+      "step": 3940
+    },
+    {
+      "epoch": 0.5446022335585275,
+      "grad_norm": 0.28166115283966064,
+      "learning_rate": 0.0002,
+      "loss": 0.351,
+      "step": 3950
+    },
+    {
+      "epoch": 0.5459809733903213,
+      "grad_norm": 0.4519428014755249,
+      "learning_rate": 0.0002,
+      "loss": 0.3523,
+      "step": 3960
+    },
+    {
+      "epoch": 0.547359713222115,
+      "grad_norm": 0.05194510146975517,
+      "learning_rate": 0.0002,
+      "loss": 0.3518,
+      "step": 3970
+    },
+    {
+      "epoch": 0.5487384530539088,
+      "grad_norm": 0.5450552105903625,
+      "learning_rate": 0.0002,
+      "loss": 0.3492,
+      "step": 3980
+    },
+    {
+      "epoch": 0.5501171928857025,
+      "grad_norm": 0.24036559462547302,
+      "learning_rate": 0.0002,
+      "loss": 0.3473,
+      "step": 3990
+    },
+    {
+      "epoch": 0.5514959327174962,
+      "grad_norm": 0.15362729132175446,
+      "learning_rate": 0.0002,
+      "loss": 0.3471,
+      "step": 4000
+    },
+    {
+      "epoch": 0.5528746725492899,
+      "grad_norm": 0.38819682598114014,
+      "learning_rate": 0.0002,
+      "loss": 0.3515,
+      "step": 4010
+    },
+    {
+      "epoch": 0.5542534123810837,
+      "grad_norm": 1.4290597438812256,
+      "learning_rate": 0.0002,
+      "loss": 0.3507,
+      "step": 4020
+    },
+    {
+      "epoch": 0.5556321522128774,
+      "grad_norm": 0.3114960491657257,
+      "learning_rate": 0.0002,
+      "loss": 0.3524,
+      "step": 4030
+    },
+    {
+      "epoch": 0.5570108920446711,
+      "grad_norm": 1.4362361431121826,
+      "learning_rate": 0.0002,
+      "loss": 0.3508,
+      "step": 4040
+    },
+    {
+      "epoch": 0.558389631876465,
+      "grad_norm": 2.7773709297180176,
+      "learning_rate": 0.0002,
+      "loss": 0.3499,
+      "step": 4050
+    },
+    {
+      "epoch": 0.5597683717082587,
+      "grad_norm": 0.5773783922195435,
+      "learning_rate": 0.0002,
+      "loss": 0.3466,
+      "step": 4060
+    },
+    {
+      "epoch": 0.5611471115400524,
+      "grad_norm": 1.8094568252563477,
+      "learning_rate": 0.0002,
+      "loss": 0.3523,
+      "step": 4070
+    },
+    {
+      "epoch": 0.5625258513718462,
+      "grad_norm": 0.732036292552948,
+      "learning_rate": 0.0002,
+      "loss": 0.3491,
+      "step": 4080
+    },
+    {
+      "epoch": 0.5639045912036399,
+      "grad_norm": 0.48324140906333923,
+      "learning_rate": 0.0002,
+      "loss": 0.3484,
+      "step": 4090
+    },
+    {
+      "epoch": 0.5652833310354336,
+      "grad_norm": 0.42546725273132324,
+      "learning_rate": 0.0002,
+      "loss": 0.348,
+      "step": 4100
+    },
+    {
+      "epoch": 0.5666620708672273,
+      "grad_norm": 0.06215515360236168,
+      "learning_rate": 0.0002,
+      "loss": 0.3467,
+      "step": 4110
+    },
+    {
+      "epoch": 0.5680408106990211,
+      "grad_norm": 0.09043420851230621,
+      "learning_rate": 0.0002,
+      "loss": 0.3496,
+      "step": 4120
+    },
+    {
+      "epoch": 0.5694195505308148,
+      "grad_norm": 0.4240778088569641,
+      "learning_rate": 0.0002,
+      "loss": 0.3475,
+      "step": 4130
+    },
+    {
+      "epoch": 0.5707982903626085,
+      "grad_norm": 0.40331658720970154,
+      "learning_rate": 0.0002,
+      "loss": 0.3472,
+      "step": 4140
+    },
+    {
+      "epoch": 0.5721770301944024,
+      "grad_norm": 0.090058833360672,
+      "learning_rate": 0.0002,
+      "loss": 0.3482,
+      "step": 4150
+    },
+    {
+      "epoch": 0.5735557700261961,
+      "grad_norm": 0.1977396011352539,
+      "learning_rate": 0.0002,
+      "loss": 0.3467,
+      "step": 4160
+    },
+    {
+      "epoch": 0.5749345098579898,
+      "grad_norm": 0.3819331228733063,
+      "learning_rate": 0.0002,
+      "loss": 0.356,
+      "step": 4170
+    },
+    {
+      "epoch": 0.5763132496897836,
+      "grad_norm": 0.9690848588943481,
+      "learning_rate": 0.0002,
+      "loss": 0.3471,
+      "step": 4180
+    },
+    {
+      "epoch": 0.5776919895215773,
+      "grad_norm": 0.5054971575737,
+      "learning_rate": 0.0002,
+      "loss": 0.3471,
+      "step": 4190
+    },
+    {
+      "epoch": 0.579070729353371,
+      "grad_norm": 0.8218488097190857,
+      "learning_rate": 0.0002,
+      "loss": 0.3486,
+      "step": 4200
+    },
+    {
+      "epoch": 0.5804494691851647,
+      "grad_norm": 0.12472192943096161,
+      "learning_rate": 0.0002,
+      "loss": 0.3494,
+      "step": 4210
+    },
+    {
+      "epoch": 0.5818282090169585,
+      "grad_norm": 0.06102175638079643,
+      "learning_rate": 0.0002,
+      "loss": 0.3479,
+      "step": 4220
+    },
+    {
+      "epoch": 0.5832069488487522,
+      "grad_norm": 0.3293861746788025,
+      "learning_rate": 0.0002,
+      "loss": 0.3481,
+      "step": 4230
+    },
+    {
+      "epoch": 0.584585688680546,
+      "grad_norm": 0.8093200325965881,
+      "learning_rate": 0.0002,
+      "loss": 0.3468,
+      "step": 4240
+    },
+    {
+      "epoch": 0.5859644285123398,
+      "grad_norm": 0.24079467356204987,
+      "learning_rate": 0.0002,
+      "loss": 0.3494,
+      "step": 4250
+    },
+    {
+      "epoch": 0.5873431683441335,
+      "grad_norm": 0.7632653117179871,
+      "learning_rate": 0.0002,
+      "loss": 0.3475,
+      "step": 4260
+    },
+    {
+      "epoch": 0.5887219081759272,
+      "grad_norm": 0.0492396280169487,
+      "learning_rate": 0.0002,
+      "loss": 0.3495,
+      "step": 4270
+    },
+    {
+      "epoch": 0.590100648007721,
+      "grad_norm": 0.20296718180179596,
+      "learning_rate": 0.0002,
+      "loss": 0.3478,
+      "step": 4280
+    },
+    {
+      "epoch": 0.5914793878395147,
+      "grad_norm": 0.46310916543006897,
+      "learning_rate": 0.0002,
+      "loss": 0.3473,
+      "step": 4290
+    },
+    {
+      "epoch": 0.5928581276713084,
+      "grad_norm": 0.08670624345541,
+      "learning_rate": 0.0002,
+      "loss": 0.3474,
+      "step": 4300
+    },
+    {
+      "epoch": 0.5942368675031021,
+      "grad_norm": 0.09182599186897278,
+      "learning_rate": 0.0002,
+      "loss": 0.3474,
+      "step": 4310
+    },
+    {
+      "epoch": 0.5956156073348959,
+      "grad_norm": 0.07010811567306519,
+      "learning_rate": 0.0002,
+      "loss": 0.3498,
+      "step": 4320
+    },
+    {
+      "epoch": 0.5969943471666896,
+      "grad_norm": 0.3193844258785248,
+      "learning_rate": 0.0002,
+      "loss": 0.3483,
+      "step": 4330
+    },
+    {
+      "epoch": 0.5983730869984834,
+      "grad_norm": 0.022516721859574318,
+      "learning_rate": 0.0002,
+      "loss": 0.3474,
+      "step": 4340
+    },
+    {
+      "epoch": 0.5997518268302772,
+      "grad_norm": 0.35459697246551514,
+      "learning_rate": 0.0002,
+      "loss": 0.3486,
+      "step": 4350
+    },
+    {
+      "epoch": 0.6011305666620709,
+      "grad_norm": 0.5825755000114441,
+      "learning_rate": 0.0002,
+      "loss": 0.3494,
+      "step": 4360
+    },
+    {
+      "epoch": 0.6025093064938646,
+      "grad_norm": 0.06838054209947586,
+      "learning_rate": 0.0002,
+      "loss": 0.3494,
+      "step": 4370
+    },
+    {
+      "epoch": 0.6038880463256584,
+      "grad_norm": 0.08730319142341614,
+      "learning_rate": 0.0002,
+      "loss": 0.3501,
+      "step": 4380
+    },
+    {
+      "epoch": 0.6052667861574521,
+      "grad_norm": 0.6632748246192932,
+      "learning_rate": 0.0002,
+      "loss": 0.3483,
+      "step": 4390
+    },
+    {
+      "epoch": 0.6066455259892458,
+      "grad_norm": 0.06691185384988785,
+      "learning_rate": 0.0002,
+      "loss": 0.3477,
+      "step": 4400
+    },
+    {
+      "epoch": 0.6080242658210395,
+      "grad_norm": 0.24831068515777588,
+      "learning_rate": 0.0002,
+      "loss": 0.3474,
+      "step": 4410
+    },
+    {
+      "epoch": 0.6094030056528333,
+      "grad_norm": 0.14777064323425293,
+      "learning_rate": 0.0002,
+      "loss": 0.3473,
+      "step": 4420
+    },
+    {
+      "epoch": 0.610781745484627,
+      "grad_norm": 0.3845897912979126,
+      "learning_rate": 0.0002,
+      "loss": 0.3471,
+      "step": 4430
+    },
+    {
+      "epoch": 0.6121604853164208,
+      "grad_norm": 0.27106064558029175,
+      "learning_rate": 0.0002,
+      "loss": 0.3476,
+      "step": 4440
+    },
+    {
+      "epoch": 0.6135392251482146,
+      "grad_norm": 0.16815268993377686,
+      "learning_rate": 0.0002,
+      "loss": 0.3485,
+      "step": 4450
+    },
+    {
+      "epoch": 0.6149179649800083,
+      "grad_norm": 0.03021593764424324,
+      "learning_rate": 0.0002,
+      "loss": 0.3478,
+      "step": 4460
+    },
+    {
+      "epoch": 0.616296704811802,
+      "grad_norm": 0.12287360429763794,
+      "learning_rate": 0.0002,
+      "loss": 0.3466,
+      "step": 4470
+    },
+    {
+      "epoch": 0.6176754446435958,
+      "grad_norm": 0.8147618174552917,
+      "learning_rate": 0.0002,
+      "loss": 0.3506,
+      "step": 4480
+    },
+    {
+      "epoch": 0.6190541844753895,
+      "grad_norm": 0.35684388875961304,
+      "learning_rate": 0.0002,
+      "loss": 0.3506,
+      "step": 4490
+    },
+    {
+      "epoch": 0.6204329243071832,
+      "grad_norm": 0.7910260558128357,
+      "learning_rate": 0.0002,
+      "loss": 0.3488,
+      "step": 4500
+    },
+    {
+      "epoch": 0.6218116641389769,
+      "grad_norm": 0.046976786106824875,
+      "learning_rate": 0.0002,
+      "loss": 0.3458,
+      "step": 4510
+    },
+    {
+      "epoch": 0.6231904039707707,
+      "grad_norm": 0.20018818974494934,
+      "learning_rate": 0.0002,
+      "loss": 0.3474,
+      "step": 4520
+    },
+    {
+      "epoch": 0.6245691438025645,
+      "grad_norm": 0.12379112839698792,
+      "learning_rate": 0.0002,
+      "loss": 0.3479,
+      "step": 4530
+    },
+    {
+      "epoch": 0.6259478836343582,
+      "grad_norm": 0.6918463706970215,
+      "learning_rate": 0.0002,
+      "loss": 0.3461,
+      "step": 4540
+    },
+    {
+      "epoch": 0.627326623466152,
+      "grad_norm": 0.5008101463317871,
+      "learning_rate": 0.0002,
+      "loss": 0.3545,
+      "step": 4550
+    },
+    {
+      "epoch": 0.6287053632979457,
+      "grad_norm": 1.0159549713134766,
+      "learning_rate": 0.0002,
+      "loss": 0.3476,
+      "step": 4560
+    },
+    {
+      "epoch": 0.6300841031297394,
+      "grad_norm": 0.8494382500648499,
+      "learning_rate": 0.0002,
+      "loss": 0.3709,
+      "step": 4570
+    },
+    {
+      "epoch": 0.6314628429615332,
+      "grad_norm": 0.5443588495254517,
+      "learning_rate": 0.0002,
+      "loss": 0.3511,
+      "step": 4580
+    },
+    {
+      "epoch": 0.6328415827933269,
+      "grad_norm": 0.06652149558067322,
+      "learning_rate": 0.0002,
+      "loss": 0.3473,
+      "step": 4590
+    },
+    {
+      "epoch": 0.6342203226251206,
+      "grad_norm": 0.33754631876945496,
+      "learning_rate": 0.0002,
+      "loss": 0.3486,
+      "step": 4600
+    },
+    {
+      "epoch": 0.6355990624569143,
+      "grad_norm": 0.02498391829431057,
+      "learning_rate": 0.0002,
+      "loss": 0.3466,
+      "step": 4610
+    },
+    {
+      "epoch": 0.6369778022887082,
+      "grad_norm": 0.1519896537065506,
+      "learning_rate": 0.0002,
+      "loss": 0.3514,
+      "step": 4620
+    },
+    {
+      "epoch": 0.6383565421205019,
+      "grad_norm": 1.1180652379989624,
+      "learning_rate": 0.0002,
+      "loss": 0.3484,
+      "step": 4630
+    },
+    {
+      "epoch": 0.6397352819522956,
+      "grad_norm": 0.2721969783306122,
+      "learning_rate": 0.0002,
+      "loss": 0.3476,
+      "step": 4640
+    },
+    {
+      "epoch": 0.6411140217840894,
+      "grad_norm": 0.564969003200531,
+      "learning_rate": 0.0002,
+      "loss": 0.3498,
+      "step": 4650
+    },
+    {
+      "epoch": 0.6424927616158831,
+      "grad_norm": 0.15192922949790955,
+      "learning_rate": 0.0002,
+      "loss": 0.3495,
+      "step": 4660
+    },
+    {
+      "epoch": 0.6438715014476768,
+      "grad_norm": 0.12203380465507507,
+      "learning_rate": 0.0002,
+      "loss": 0.3463,
+      "step": 4670
+    },
+    {
+      "epoch": 0.6452502412794706,
+      "grad_norm": 0.428926557302475,
+      "learning_rate": 0.0002,
+      "loss": 0.3542,
+      "step": 4680
+    },
+    {
+      "epoch": 0.6466289811112643,
+      "grad_norm": 0.16655278205871582,
+      "learning_rate": 0.0002,
+      "loss": 0.3502,
+      "step": 4690
+    },
+    {
+      "epoch": 0.648007720943058,
+      "grad_norm": 0.783054530620575,
+      "learning_rate": 0.0002,
+      "loss": 0.3517,
+      "step": 4700
+    },
+    {
+      "epoch": 0.6493864607748517,
+      "grad_norm": 0.6207926273345947,
+      "learning_rate": 0.0002,
+      "loss": 0.3553,
+      "step": 4710
+    },
+    {
+      "epoch": 0.6507652006066456,
+      "grad_norm": 0.08468189090490341,
+      "learning_rate": 0.0002,
+      "loss": 0.3465,
+      "step": 4720
+    },
+    {
+      "epoch": 0.6521439404384393,
+      "grad_norm": 0.3856925368309021,
+      "learning_rate": 0.0002,
+      "loss": 0.3465,
+      "step": 4730
+    },
+    {
+      "epoch": 0.653522680270233,
+      "grad_norm": 0.07147127389907837,
+      "learning_rate": 0.0002,
+      "loss": 0.3461,
+      "step": 4740
+    },
+    {
+      "epoch": 0.6549014201020268,
+      "grad_norm": 0.6693950295448303,
+      "learning_rate": 0.0002,
+      "loss": 0.3472,
+      "step": 4750
+    },
+    {
+      "epoch": 0.6562801599338205,
+      "grad_norm": 0.22481206059455872,
+      "learning_rate": 0.0002,
+      "loss": 0.3488,
+      "step": 4760
+    },
+    {
+      "epoch": 0.6576588997656142,
+      "grad_norm": 0.1674708127975464,
+      "learning_rate": 0.0002,
+      "loss": 0.3494,
+      "step": 4770
+    },
+    {
+      "epoch": 0.659037639597408,
+      "grad_norm": 0.1923258900642395,
+      "learning_rate": 0.0002,
+      "loss": 0.3489,
+      "step": 4780
+    },
+    {
+      "epoch": 0.6604163794292017,
+      "grad_norm": 0.11305999755859375,
+      "learning_rate": 0.0002,
+      "loss": 0.3472,
+      "step": 4790
+    },
+    {
+      "epoch": 0.6617951192609954,
+      "grad_norm": 0.01815452240407467,
+      "learning_rate": 0.0002,
+      "loss": 0.3475,
+      "step": 4800
+    },
+    {
+      "epoch": 0.6631738590927891,
+      "grad_norm": 0.05602194741368294,
+      "learning_rate": 0.0002,
+      "loss": 0.3475,
+      "step": 4810
+    },
+    {
+      "epoch": 0.664552598924583,
+      "grad_norm": 0.1651487499475479,
+      "learning_rate": 0.0002,
+      "loss": 0.348,
+      "step": 4820
+    },
+    {
+      "epoch": 0.6659313387563767,
+      "grad_norm": 0.13739043474197388,
+      "learning_rate": 0.0002,
+      "loss": 0.3476,
+      "step": 4830
+    },
+    {
+      "epoch": 0.6673100785881704,
+      "grad_norm": 0.3567051887512207,
+      "learning_rate": 0.0002,
+      "loss": 0.3469,
+      "step": 4840
+    },
+    {
+      "epoch": 0.6686888184199642,
+      "grad_norm": 0.4653046131134033,
+      "learning_rate": 0.0002,
+      "loss": 0.3469,
+      "step": 4850
+    },
+    {
+      "epoch": 0.6700675582517579,
+      "grad_norm": 1.0645127296447754,
+      "learning_rate": 0.0002,
+      "loss": 0.3486,
+      "step": 4860
+    },
+    {
+      "epoch": 0.6714462980835516,
+      "grad_norm": 0.29499199986457825,
+      "learning_rate": 0.0002,
+      "loss": 0.35,
+      "step": 4870
+    },
+    {
+      "epoch": 0.6728250379153454,
+      "grad_norm": 0.03139637038111687,
+      "learning_rate": 0.0002,
+      "loss": 0.3476,
+      "step": 4880
+    },
+    {
+      "epoch": 0.6742037777471391,
+      "grad_norm": 0.9673721194267273,
+      "learning_rate": 0.0002,
+      "loss": 0.3495,
+      "step": 4890
+    },
+    {
+      "epoch": 0.6755825175789328,
+      "grad_norm": 0.410988450050354,
+      "learning_rate": 0.0002,
+      "loss": 0.3487,
+      "step": 4900
+    },
+    {
+      "epoch": 0.6769612574107265,
+      "grad_norm": 2.0737829208374023,
+      "learning_rate": 0.0002,
+      "loss": 0.3536,
+      "step": 4910
+    },
+    {
+      "epoch": 0.6783399972425204,
+      "grad_norm": 0.20992055535316467,
+      "learning_rate": 0.0002,
+      "loss": 0.3504,
+      "step": 4920
+    },
+    {
+      "epoch": 0.6797187370743141,
+      "grad_norm": 0.30803462862968445,
+      "learning_rate": 0.0002,
+      "loss": 0.3522,
+      "step": 4930
+    },
+    {
+      "epoch": 0.6810974769061078,
+      "grad_norm": 0.7711420655250549,
+      "learning_rate": 0.0002,
+      "loss": 0.3725,
+      "step": 4940
+    },
+    {
+      "epoch": 0.6824762167379016,
+      "grad_norm": 1.0478612184524536,
+      "learning_rate": 0.0002,
+      "loss": 0.3555,
+      "step": 4950
+    },
+    {
+      "epoch": 0.6838549565696953,
+      "grad_norm": 0.6949588656425476,
+      "learning_rate": 0.0002,
+      "loss": 0.3485,
+      "step": 4960
+    },
+    {
+      "epoch": 0.685233696401489,
+      "grad_norm": 0.13478349149227142,
+      "learning_rate": 0.0002,
+      "loss": 0.3497,
+      "step": 4970
+    },
+    {
+      "epoch": 0.6866124362332828,
+      "grad_norm": 0.0825844407081604,
+      "learning_rate": 0.0002,
+      "loss": 0.3475,
+      "step": 4980
+    },
+    {
+      "epoch": 0.6879911760650765,
+      "grad_norm": 0.013896561227738857,
+      "learning_rate": 0.0002,
+      "loss": 0.3472,
+      "step": 4990
+    },
+    {
+      "epoch": 0.6893699158968702,
+      "grad_norm": 0.10021249949932098,
+      "learning_rate": 0.0002,
+      "loss": 0.3472,
+      "step": 5000
+    },
+    {
+      "epoch": 0.690748655728664,
+      "grad_norm": 0.16724593937397003,
+      "learning_rate": 0.0002,
+      "loss": 0.3468,
+      "step": 5010
+    },
+    {
+      "epoch": 0.6921273955604578,
+      "grad_norm": 0.0654013380408287,
+      "learning_rate": 0.0002,
+      "loss": 0.3468,
+      "step": 5020
+    },
+    {
+      "epoch": 0.6935061353922515,
+      "grad_norm": 0.24155353009700775,
+      "learning_rate": 0.0002,
+      "loss": 0.3463,
+      "step": 5030
+    },
+    {
+      "epoch": 0.6948848752240452,
+      "grad_norm": 0.2128722220659256,
+      "learning_rate": 0.0002,
+      "loss": 0.3481,
+      "step": 5040
+    },
+    {
+      "epoch": 0.696263615055839,
+      "grad_norm": 0.26365959644317627,
+      "learning_rate": 0.0002,
+      "loss": 0.347,
+      "step": 5050
+    },
+    {
+      "epoch": 0.6976423548876327,
+      "grad_norm": 0.021076533943414688,
+      "learning_rate": 0.0002,
+      "loss": 0.3486,
+      "step": 5060
+    },
+    {
+      "epoch": 0.6990210947194264,
+      "grad_norm": 0.08810916543006897,
+      "learning_rate": 0.0002,
+      "loss": 0.3469,
+      "step": 5070
+    },
+    {
+      "epoch": 0.7003998345512202,
+      "grad_norm": 0.12958675622940063,
+      "learning_rate": 0.0002,
+      "loss": 0.3474,
+      "step": 5080
+    },
+    {
+      "epoch": 0.7017785743830139,
+      "grad_norm": 0.03748354688286781,
+      "learning_rate": 0.0002,
+      "loss": 0.3463,
+      "step": 5090
+    },
+    {
+      "epoch": 0.7031573142148076,
+      "grad_norm": 0.18368670344352722,
+      "learning_rate": 0.0002,
+      "loss": 0.3487,
+      "step": 5100
+    },
+    {
+      "epoch": 0.7045360540466014,
+      "grad_norm": 0.11506777256727219,
+      "learning_rate": 0.0002,
+      "loss": 0.3486,
+      "step": 5110
+    },
+    {
+      "epoch": 0.7059147938783952,
+      "grad_norm": 0.31118178367614746,
+      "learning_rate": 0.0002,
+      "loss": 0.3475,
+      "step": 5120
+    },
+    {
+      "epoch": 0.7072935337101889,
+      "grad_norm": 0.8338253498077393,
+      "learning_rate": 0.0002,
+      "loss": 0.3515,
+      "step": 5130
+    },
+    {
+      "epoch": 0.7086722735419826,
+      "grad_norm": 0.0813426598906517,
+      "learning_rate": 0.0002,
+      "loss": 0.3472,
+      "step": 5140
+    },
+    {
+      "epoch": 0.7100510133737764,
+      "grad_norm": 0.41932597756385803,
+      "learning_rate": 0.0002,
+      "loss": 0.3471,
+      "step": 5150
+    },
+    {
+      "epoch": 0.7114297532055701,
+      "grad_norm": 0.29162150621414185,
+      "learning_rate": 0.0002,
+      "loss": 0.3468,
+      "step": 5160
+    },
+    {
+      "epoch": 0.7128084930373638,
+      "grad_norm": 0.29316985607147217,
+      "learning_rate": 0.0002,
+      "loss": 0.3466,
+      "step": 5170
+    },
+    {
+      "epoch": 0.7141872328691576,
+      "grad_norm": 0.7843354344367981,
+      "learning_rate": 0.0002,
+      "loss": 0.3483,
+      "step": 5180
+    },
+    {
+      "epoch": 0.7155659727009513,
+      "grad_norm": 0.5645899176597595,
+      "learning_rate": 0.0002,
+      "loss": 0.3481,
+      "step": 5190
+    },
+    {
+      "epoch": 0.716944712532745,
+      "grad_norm": 0.05591394752264023,
+      "learning_rate": 0.0002,
+      "loss": 0.3455,
+      "step": 5200
+    },
+    {
+      "epoch": 0.7183234523645388,
+      "grad_norm": 0.23811687529087067,
+      "learning_rate": 0.0002,
+      "loss": 0.3507,
+      "step": 5210
+    },
+    {
+      "epoch": 0.7197021921963326,
+      "grad_norm": 1.6006462574005127,
+      "learning_rate": 0.0002,
+      "loss": 0.3537,
+      "step": 5220
+    },
+    {
+      "epoch": 0.7210809320281263,
+      "grad_norm": 0.9361100792884827,
+      "learning_rate": 0.0002,
+      "loss": 0.353,
+      "step": 5230
+    },
+    {
+      "epoch": 0.72245967185992,
+      "grad_norm": 1.1202062368392944,
+      "learning_rate": 0.0002,
+      "loss": 0.3558,
+      "step": 5240
+    },
+    {
+      "epoch": 0.7238384116917138,
+      "grad_norm": 0.058672916144132614,
+      "learning_rate": 0.0002,
+      "loss": 0.3493,
+      "step": 5250
+    },
+    {
+      "epoch": 0.7252171515235075,
+      "grad_norm": 0.6465821266174316,
+      "learning_rate": 0.0002,
+      "loss": 0.3455,
+      "step": 5260
+    },
+    {
+      "epoch": 0.7265958913553012,
+      "grad_norm": 2.3552298545837402,
+      "learning_rate": 0.0002,
+      "loss": 0.3504,
+      "step": 5270
+    },
+    {
+      "epoch": 0.727974631187095,
+      "grad_norm": 1.6489512920379639,
+      "learning_rate": 0.0002,
+      "loss": 0.3525,
+      "step": 5280
+    },
+    {
+      "epoch": 0.7293533710188888,
+      "grad_norm": 0.116278275847435,
+      "learning_rate": 0.0002,
+      "loss": 0.35,
+      "step": 5290
+    },
+    {
+      "epoch": 0.7307321108506825,
+      "grad_norm": 0.12393849343061447,
+      "learning_rate": 0.0002,
+      "loss": 0.347,
+      "step": 5300
+    },
+    {
+      "epoch": 0.7321108506824762,
+      "grad_norm": 0.04386143013834953,
+      "learning_rate": 0.0002,
+      "loss": 0.3459,
+      "step": 5310
+    },
+    {
+      "epoch": 0.73348959051427,
+      "grad_norm": 0.28474342823028564,
+      "learning_rate": 0.0002,
+      "loss": 0.3479,
+      "step": 5320
+    },
+    {
+      "epoch": 0.7348683303460637,
+      "grad_norm": 0.21442459523677826,
+      "learning_rate": 0.0002,
+      "loss": 0.3494,
+      "step": 5330
+    },
+    {
+      "epoch": 0.7362470701778574,
+      "grad_norm": 0.6859696507453918,
+      "learning_rate": 0.0002,
+      "loss": 0.3529,
+      "step": 5340
+    },
+    {
+      "epoch": 0.7376258100096512,
+      "grad_norm": 0.7702832818031311,
+      "learning_rate": 0.0002,
+      "loss": 0.3472,
+      "step": 5350
+    },
+    {
+      "epoch": 0.7390045498414449,
+      "grad_norm": 0.13723976910114288,
+      "learning_rate": 0.0002,
+      "loss": 0.3503,
+      "step": 5360
+    },
+    {
+      "epoch": 0.7403832896732386,
+      "grad_norm": 0.05630859360098839,
+      "learning_rate": 0.0002,
+      "loss": 0.4115,
+      "step": 5370
+    },
+    {
+      "epoch": 0.7417620295050324,
+      "grad_norm": 1219.2249755859375,
+      "learning_rate": 0.0002,
+      "loss": 0.5008,
+      "step": 5380
+    },
+    {
+      "epoch": 0.7431407693368262,
+      "grad_norm": 11.58250904083252,
+      "learning_rate": 0.0002,
+      "loss": 2.7799,
+      "step": 5390
+    },
+    {
+      "epoch": 0.7445195091686199,
+      "grad_norm": 0.5598617196083069,
+      "learning_rate": 0.0002,
+      "loss": 2.3202,
+      "step": 5400
+    },
+    {
+      "epoch": 0.7458982490004136,
+      "grad_norm": 8.439847946166992,
+      "learning_rate": 0.0002,
+      "loss": 14.4666,
+      "step": 5410
+    },
+    {
+      "epoch": 0.7472769888322074,
+      "grad_norm": 2.4195609092712402,
+      "learning_rate": 0.0002,
+      "loss": 0.3972,
+      "step": 5420
+    },
+    {
+      "epoch": 0.7486557286640011,
+      "grad_norm": 2.582343816757202,
+      "learning_rate": 0.0002,
+      "loss": 2.8621,
+      "step": 5430
+    },
+    {
+      "epoch": 0.7500344684957948,
+      "grad_norm": 1717.897216796875,
+      "learning_rate": 0.0002,
+      "loss": 1.1371,
+      "step": 5440
+    },
+    {
+      "epoch": 0.7514132083275886,
+      "grad_norm": 30.825328826904297,
+      "learning_rate": 0.0002,
+      "loss": 13.4539,
+      "step": 5450
+    },
+    {
+      "epoch": 0.7527919481593823,
+      "grad_norm": 5.435330390930176,
+      "learning_rate": 0.0002,
+      "loss": 2.7941,
+      "step": 5460
+    },
+    {
+      "epoch": 0.754170687991176,
+      "grad_norm": 1.5872547626495361,
+      "learning_rate": 0.0002,
+      "loss": 1.7573,
+      "step": 5470
+    },
+    {
+      "epoch": 0.7555494278229699,
+      "grad_norm": 6.4165496826171875,
+      "learning_rate": 0.0002,
+      "loss": 1.3329,
+      "step": 5480
+    },
+    {
+      "epoch": 0.7569281676547636,
+      "grad_norm": 4.3425703048706055,
+      "learning_rate": 0.0002,
+      "loss": 1.259,
+      "step": 5490
+    },
+    {
+      "epoch": 0.7583069074865573,
+      "grad_norm": 4.925943374633789,
+      "learning_rate": 0.0002,
+      "loss": 1.2657,
+      "step": 5500
+    },
+    {
+      "epoch": 0.759685647318351,
+      "grad_norm": 0.9678533673286438,
+      "learning_rate": 0.0002,
+      "loss": 1.1297,
+      "step": 5510
+    },
+    {
+      "epoch": 0.7610643871501448,
+      "grad_norm": 2.6134777069091797,
+      "learning_rate": 0.0002,
+      "loss": 1.0949,
+      "step": 5520
+    },
+    {
+      "epoch": 0.7624431269819385,
+      "grad_norm": 2.0054285526275635,
+      "learning_rate": 0.0002,
+      "loss": 1.0665,
+      "step": 5530
+    },
+    {
+      "epoch": 0.7638218668137322,
+      "grad_norm": 2.4693703651428223,
+      "learning_rate": 0.0002,
+      "loss": 1.1067,
+      "step": 5540
+    },
+    {
+      "epoch": 0.765200606645526,
+      "grad_norm": 2.8980472087860107,
+      "learning_rate": 0.0002,
+      "loss": 1.0521,
+      "step": 5550
+    },
+    {
+      "epoch": 0.7665793464773197,
+      "grad_norm": 1.2929408550262451,
+      "learning_rate": 0.0002,
+      "loss": 1.1144,
+      "step": 5560
+    },
+    {
+      "epoch": 0.7679580863091134,
+      "grad_norm": 2.822032928466797,
+      "learning_rate": 0.0002,
+      "loss": 0.6138,
+      "step": 5570
+    },
+    {
+      "epoch": 0.7693368261409073,
+      "grad_norm": 0.18209309875965118,
+      "learning_rate": 0.0002,
+      "loss": 0.7642,
+      "step": 5580
+    },
+    {
+      "epoch": 0.770715565972701,
+      "grad_norm": 0.6624636054039001,
+      "learning_rate": 0.0002,
+      "loss": 0.3621,
+      "step": 5590
+    },
+    {
+      "epoch": 0.7720943058044947,
+      "grad_norm": 1.829182744026184,
+      "learning_rate": 0.0002,
+      "loss": 0.362,
+      "step": 5600
+    },
+    {
+      "epoch": 0.7734730456362884,
+      "grad_norm": 6.315738201141357,
+      "learning_rate": 0.0002,
+      "loss": 4.5046,
+      "step": 5610
+    },
+    {
+      "epoch": 0.7748517854680822,
+      "grad_norm": 8.351112365722656,
+      "learning_rate": 0.0002,
+      "loss": 2.3984,
+      "step": 5620
+    },
+    {
+      "epoch": 0.7762305252998759,
+      "grad_norm": 5.36275577545166,
+      "learning_rate": 0.0002,
+      "loss": 2.0927,
+      "step": 5630
+    },
+    {
+      "epoch": 0.7776092651316696,
+      "grad_norm": 152.20603942871094,
+      "learning_rate": 0.0002,
+      "loss": 0.8589,
+      "step": 5640
+    },
+    {
+      "epoch": 0.7789880049634634,
+      "grad_norm": 31.267227172851562,
+      "learning_rate": 0.0002,
+      "loss": 0.5345,
+      "step": 5650
+    },
+    {
+      "epoch": 0.7803667447952571,
+      "grad_norm": 61105.32421875,
+      "learning_rate": 0.0002,
+      "loss": 0.9758,
+      "step": 5660
+    },
+    {
+      "epoch": 0.7817454846270508,
+      "grad_norm": 117.10787200927734,
+      "learning_rate": 0.0002,
+      "loss": 5.5246,
+      "step": 5670
+    },
+    {
+      "epoch": 0.7831242244588447,
+      "grad_norm": 110.45667266845703,
+      "learning_rate": 0.0002,
+      "loss": 1.4643,
+      "step": 5680
+    },
+    {
+      "epoch": 0.7845029642906384,
+      "grad_norm": 16.042375564575195,
+      "learning_rate": 0.0002,
+      "loss": 1.3579,
+      "step": 5690
+    },
+    {
+      "epoch": 0.7858817041224321,
+      "grad_norm": 2.7994253635406494,
+      "learning_rate": 0.0002,
+      "loss": 1.0792,
+      "step": 5700
+    },
+    {
+      "epoch": 0.7872604439542258,
+      "grad_norm": 2.488832712173462,
+      "learning_rate": 0.0002,
+      "loss": 1.1039,
+      "step": 5710
+    },
+    {
+      "epoch": 0.7886391837860196,
+      "grad_norm": 0.881956934928894,
+      "learning_rate": 0.0002,
+      "loss": 1.0412,
+      "step": 5720
+    },
+    {
+      "epoch": 0.7900179236178133,
+      "grad_norm": 0.8120241761207581,
+      "learning_rate": 0.0002,
+      "loss": 0.5623,
+      "step": 5730
+    },
+    {
+      "epoch": 0.791396663449607,
+      "grad_norm": 2.202730655670166,
+      "learning_rate": 0.0002,
+      "loss": 0.3895,
+      "step": 5740
+    },
+    {
+      "epoch": 0.7927754032814008,
+      "grad_norm": 1.8389441967010498,
+      "learning_rate": 0.0002,
+      "loss": 0.4279,
+      "step": 5750
+    },
+    {
+      "epoch": 0.7941541431131945,
+      "grad_norm": 9.026138305664062,
+      "learning_rate": 0.0002,
+      "loss": 1.1896,
+      "step": 5760
+    },
+    {
+      "epoch": 0.7955328829449883,
+      "grad_norm": 1.7441275119781494,
+      "learning_rate": 0.0002,
+      "loss": 0.4751,
+      "step": 5770
+    },
+    {
+      "epoch": 0.7969116227767821,
+      "grad_norm": 2.0509731769561768,
+      "learning_rate": 0.0002,
+      "loss": 0.4031,
+      "step": 5780
+    },
+    {
+      "epoch": 0.7982903626085758,
+      "grad_norm": 0.7825273871421814,
+      "learning_rate": 0.0002,
+      "loss": 0.3782,
+      "step": 5790
+    },
+    {
+      "epoch": 0.7996691024403695,
+      "grad_norm": 1.6086506843566895,
+      "learning_rate": 0.0002,
+      "loss": 0.3639,
+      "step": 5800
+    },
+    {
+      "epoch": 0.8010478422721632,
+      "grad_norm": 165.86508178710938,
+      "learning_rate": 0.0002,
+      "loss": 0.7169,
+      "step": 5810
+    },
+    {
+      "epoch": 0.802426582103957,
+      "grad_norm": 1.6798721551895142,
+      "learning_rate": 0.0002,
+      "loss": 0.6331,
+      "step": 5820
+    },
+    {
+      "epoch": 0.8038053219357507,
+      "grad_norm": 0.3252682089805603,
+      "learning_rate": 0.0002,
+      "loss": 0.3673,
+      "step": 5830
+    },
+    {
+      "epoch": 0.8051840617675444,
+      "grad_norm": 0.10813436657190323,
+      "learning_rate": 0.0002,
+      "loss": 0.3533,
+      "step": 5840
+    },
+    {
+      "epoch": 0.8065628015993382,
+      "grad_norm": 0.4101102948188782,
+      "learning_rate": 0.0002,
+      "loss": 0.3601,
+      "step": 5850
+    },
+    {
+      "epoch": 0.807941541431132,
+      "grad_norm": 4.0115742683410645,
+      "learning_rate": 0.0002,
+      "loss": 0.4159,
+      "step": 5860
+    },
+    {
+      "epoch": 0.8093202812629257,
+      "grad_norm": 0.7337489724159241,
+      "learning_rate": 0.0002,
+      "loss": 0.3701,
+      "step": 5870
+    },
+    {
+      "epoch": 0.8106990210947195,
+      "grad_norm": 0.19467441737651825,
+      "learning_rate": 0.0002,
+      "loss": 0.3516,
+      "step": 5880
+    },
+    {
+      "epoch": 0.8120777609265132,
+      "grad_norm": 1.1243358850479126,
+      "learning_rate": 0.0002,
+      "loss": 0.3537,
+      "step": 5890
+    },
+    {
+      "epoch": 0.8134565007583069,
+      "grad_norm": 0.646501898765564,
+      "learning_rate": 0.0002,
+      "loss": 0.3645,
+      "step": 5900
+    },
+    {
+      "epoch": 0.8148352405901006,
+      "grad_norm": 0.5523257851600647,
+      "learning_rate": 0.0002,
+      "loss": 0.3625,
+      "step": 5910
+    },
+    {
+      "epoch": 0.8162139804218944,
+      "grad_norm": 0.03894199803471565,
+      "learning_rate": 0.0002,
+      "loss": 0.4156,
+      "step": 5920
+    },
+    {
+      "epoch": 0.8175927202536881,
+      "grad_norm": 0.02224864810705185,
+      "learning_rate": 0.0002,
+      "loss": 0.3592,
+      "step": 5930
+    },
+    {
+      "epoch": 0.8189714600854818,
+      "grad_norm": 0.45560982823371887,
+      "learning_rate": 0.0002,
+      "loss": 0.3727,
+      "step": 5940
+    },
+    {
+      "epoch": 0.8203501999172756,
+      "grad_norm": 1.3106188774108887,
+      "learning_rate": 0.0002,
+      "loss": 0.3966,
+      "step": 5950
+    },
+    {
+      "epoch": 0.8217289397490694,
+      "grad_norm": 0.08093151450157166,
+      "learning_rate": 0.0002,
+      "loss": 0.3761,
+      "step": 5960
+    },
+    {
+      "epoch": 0.8231076795808631,
+      "grad_norm": 0.10217435657978058,
+      "learning_rate": 0.0002,
+      "loss": 0.4228,
+      "step": 5970
+    },
+    {
+      "epoch": 0.8244864194126569,
+      "grad_norm": 0.47661492228507996,
+      "learning_rate": 0.0002,
+      "loss": 0.3511,
+      "step": 5980
+    },
+    {
+      "epoch": 0.8258651592444506,
+      "grad_norm": 0.0656476691365242,
+      "learning_rate": 0.0002,
+      "loss": 0.3622,
+      "step": 5990
+    },
+    {
+      "epoch": 0.8272438990762443,
+      "grad_norm": 0.1670994758605957,
+      "learning_rate": 0.0002,
+      "loss": 0.3535,
+      "step": 6000
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 100000,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 14,
+  "save_steps": 1000,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.2276641541122163e+19,
+  "train_batch_size": 16,
+  "trial_name": null,
+  "trial_params": null
+}