diff --git "a/trainer_state.json" "b/trainer_state.json"
--- "a/trainer_state.json"
+++ "b/trainer_state.json"
@@ -2,7 +2,7 @@
   "best_metric": null,
   "best_model_checkpoint": null,
   "epoch": 1.0,
-  "eval_steps": 200,
+  "eval_steps": 100,
   "global_step": 1563,
   "is_hyper_param_search": false,
   "is_local_process_zero": true,
@@ -10,7 +10,7 @@
   "log_history": [
     {
       "epoch": 0.0006397952655150352,
-      "grad_norm": 412.2480368303884,
+      "grad_norm": 412.2005907726637,
       "learning_rate": 3.1847133757961784e-09,
       "logits/generated": -2.53759503364563,
       "logits/real": -2.290781259536743,
@@ -25,2471 +25,2599 @@
     },
     {
       "epoch": 0.006397952655150352,
-      "grad_norm": 431.96820478477065,
+      "grad_norm": 434.41016443323946,
       "learning_rate": 3.184713375796178e-08,
-      "logits/generated": -2.5309884548187256,
-      "logits/real": -2.523167371749878,
-      "logps/generated": -88.78756713867188,
-      "logps/real": -236.1966094970703,
-      "loss": 1.1221,
-      "rewards/accuracies": 0.5555555820465088,
-      "rewards/generated": -0.010531743057072163,
-      "rewards/margins": 0.02550431713461876,
-      "rewards/real": 0.014972572214901447,
+      "logits/generated": -2.5310893058776855,
+      "logits/real": -2.523176670074463,
+      "logps/generated": -88.80826568603516,
+      "logps/real": -236.27447509765625,
+      "loss": 1.1251,
+      "rewards/accuracies": 0.5277777910232544,
+      "rewards/generated": -0.01260171364992857,
+      "rewards/margins": 0.01978916861116886,
+      "rewards/real": 0.007187454029917717,
       "step": 10
     },
     {
       "epoch": 0.012795905310300703,
-      "grad_norm": 299.862381030409,
+      "grad_norm": 297.9763241749388,
       "learning_rate": 6.369426751592356e-08,
-      "logits/generated": -2.578387498855591,
-      "logits/real": -2.5629818439483643,
-      "logps/generated": -90.43721771240234,
-      "logps/real": -232.0134735107422,
-      "loss": 0.9789,
+      "logits/generated": -2.5787622928619385,
+      "logits/real": -2.5631511211395264,
+      "logps/generated": -90.50495147705078,
+      "logps/real": -232.0811767578125,
+      "loss": 0.9777,
       "rewards/accuracies": 0.9624999761581421,
-      "rewards/generated": -0.10627230256795883,
-      "rewards/margins": 0.31521138548851013,
-      "rewards/real": 0.2089390754699707,
+      "rewards/generated": -0.11304464191198349,
+      "rewards/margins": 0.3152126371860504,
+      "rewards/real": 0.20216801762580872,
       "step": 20
     },
     {
       "epoch": 0.019193857965451054,
-      "grad_norm": 128.80078242264278,
+      "grad_norm": 127.27796238923372,
       "learning_rate": 9.554140127388536e-08,
-      "logits/generated": -2.639655590057373,
-      "logits/real": -2.623234748840332,
-      "logps/generated": -81.7702865600586,
-      "logps/real": -192.8977813720703,
-      "loss": 0.732,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -0.4113238453865051,
-      "rewards/margins": 1.0968698263168335,
-      "rewards/real": 0.6855460405349731,
+      "logits/generated": -2.6398799419403076,
+      "logits/real": -2.6236019134521484,
+      "logps/generated": -81.77882385253906,
+      "logps/real": -192.82749938964844,
+      "loss": 0.7306,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/generated": -0.4121777415275574,
+      "rewards/margins": 1.104752540588379,
+      "rewards/real": 0.6925747990608215,
       "step": 30
     },
     {
       "epoch": 0.025591810620601407,
-      "grad_norm": 53.90372491881725,
+      "grad_norm": 54.110633436002345,
       "learning_rate": 1.2738853503184713e-07,
-      "logits/generated": -2.754666566848755,
-      "logits/real": -2.771085262298584,
-      "logps/generated": -93.06287384033203,
-      "logps/real": -208.916015625,
-      "loss": 0.551,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -0.8528655171394348,
-      "rewards/margins": 2.697784423828125,
-      "rewards/real": 1.844918966293335,
+      "logits/generated": -2.7545366287231445,
+      "logits/real": -2.770876407623291,
+      "logps/generated": -93.11697387695312,
+      "logps/real": -208.94491577148438,
+      "loss": 0.5505,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -0.8582771420478821,
+      "rewards/margins": 2.700303792953491,
+      "rewards/real": 1.842026710510254,
       "step": 40
     },
     {
       "epoch": 0.03198976327575176,
-      "grad_norm": 36.32945064394617,
+      "grad_norm": 36.311814112732144,
       "learning_rate": 1.592356687898089e-07,
-      "logits/generated": -2.875619649887085,
-      "logits/real": -2.8736255168914795,
-      "logps/generated": -102.63410949707031,
-      "logps/real": -185.98129272460938,
-      "loss": 0.4387,
+      "logits/generated": -2.8756887912750244,
+      "logits/real": -2.8734962940216064,
+      "logps/generated": -102.60001373291016,
+      "logps/real": -186.04449462890625,
+      "loss": 0.4389,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/generated": -1.5957601070404053,
-      "rewards/margins": 4.849114894866943,
-      "rewards/real": 3.253354549407959,
+      "rewards/generated": -1.592349886894226,
+      "rewards/margins": 4.839385509490967,
+      "rewards/real": 3.247035503387451,
       "step": 50
     },
     {
       "epoch": 0.03838771593090211,
-      "grad_norm": 22.26335616683524,
+      "grad_norm": 22.292502262551583,
       "learning_rate": 1.9108280254777072e-07,
-      "logits/generated": -2.9478697776794434,
-      "logits/real": -3.007082223892212,
-      "logps/generated": -110.52901458740234,
-      "logps/real": -171.8012237548828,
-      "loss": 0.3865,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -1.786839485168457,
-      "rewards/margins": 6.540804862976074,
-      "rewards/real": 4.753964900970459,
+      "logits/generated": -2.9486448764801025,
+      "logits/real": -3.0080959796905518,
+      "logps/generated": -110.56791687011719,
+      "logps/real": -171.86045837402344,
+      "loss": 0.3867,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -1.7907304763793945,
+      "rewards/margins": 6.538771152496338,
+      "rewards/real": 4.748041152954102,
       "step": 60
     },
     {
       "epoch": 0.044785668586052464,
-      "grad_norm": 18.72255550731988,
+      "grad_norm": 20.28796261535442,
       "learning_rate": 2.2292993630573247e-07,
-      "logits/generated": -2.9968416690826416,
-      "logits/real": -3.001485824584961,
-      "logps/generated": -107.23085021972656,
-      "logps/real": -146.6729736328125,
-      "loss": 0.3283,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -1.93621826171875,
-      "rewards/margins": 8.847390174865723,
-      "rewards/real": 6.911171913146973,
+      "logits/generated": -3.008568048477173,
+      "logits/real": -3.0137972831726074,
+      "logps/generated": -104.3636245727539,
+      "logps/real": -148.05299377441406,
+      "loss": 0.3308,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -1.649496078491211,
+      "rewards/margins": 8.422666549682617,
+      "rewards/real": 6.773171424865723,
       "step": 70
     },
     {
       "epoch": 0.05118362124120281,
-      "grad_norm": 9.385567995761765,
+      "grad_norm": 6.868749133343201,
       "learning_rate": 2.5477707006369425e-07,
-      "logits/generated": -2.985713481903076,
-      "logits/real": -3.0017921924591064,
-      "logps/generated": -111.2695083618164,
-      "logps/real": -128.3047332763672,
-      "loss": 0.2744,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -2.9947776794433594,
-      "rewards/margins": 10.760424613952637,
-      "rewards/real": 7.765646934509277,
+      "logits/generated": -2.988903284072876,
+      "logits/real": -3.000037908554077,
+      "logps/generated": -109.73497009277344,
+      "logps/real": -128.55386352539062,
+      "loss": 0.2753,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -2.8413243293762207,
+      "rewards/margins": 10.582056999206543,
+      "rewards/real": 7.7407331466674805,
       "step": 80
     },
     {
       "epoch": 0.05758157389635317,
-      "grad_norm": 4.574434721884733,
+      "grad_norm": 4.629693260914371,
       "learning_rate": 2.86624203821656e-07,
-      "logits/generated": -2.93389630317688,
-      "logits/real": -2.9760384559631348,
-      "logps/generated": -124.31734466552734,
-      "logps/real": -127.97233581542969,
-      "loss": 0.269,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -3.8817825317382812,
-      "rewards/margins": 12.403055191040039,
-      "rewards/real": 8.521272659301758,
+      "logits/generated": -2.9299612045288086,
+      "logits/real": -2.967792272567749,
+      "logps/generated": -122.9028091430664,
+      "logps/real": -127.64404296875,
+      "loss": 0.2688,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -3.740330934524536,
+      "rewards/margins": 12.294435501098633,
+      "rewards/real": 8.55410385131836,
       "step": 90
     },
     {
       "epoch": 0.06397952655150352,
-      "grad_norm": 9.260926951901052,
+      "grad_norm": 7.713169024269941,
       "learning_rate": 3.184713375796178e-07,
-      "logits/generated": -2.948780059814453,
-      "logits/real": -2.924206495285034,
-      "logps/generated": -117.82643127441406,
-      "logps/real": -140.2669219970703,
-      "loss": 0.2787,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -3.5896174907684326,
-      "rewards/margins": 13.11004638671875,
-      "rewards/real": 9.520429611206055,
+      "logits/generated": -2.945857524871826,
+      "logits/real": -2.9198944568634033,
+      "logps/generated": -117.79829406738281,
+      "logps/real": -140.08273315429688,
+      "loss": 0.278,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -3.586803436279297,
+      "rewards/margins": 13.125653266906738,
+      "rewards/real": 9.538848876953125,
+      "step": 100
+    },
+    {
+      "epoch": 0.06397952655150352,
+      "eval_logits/generated": -2.905454158782959,
+      "eval_logits/real": -2.8962948322296143,
+      "eval_logps/generated": -112.96752166748047,
+      "eval_logps/real": -126.31481170654297,
+      "eval_loss": 0.27034586668014526,
+      "eval_rewards/accuracies": 0.9921875,
+      "eval_rewards/generated": -3.425051212310791,
+      "eval_rewards/margins": 12.061681747436523,
+      "eval_rewards/real": 8.636629104614258,
+      "eval_runtime": 57.1662,
+      "eval_samples_per_second": 8.746,
+      "eval_steps_per_second": 0.28,
       "step": 100
     },
     {
       "epoch": 0.07037747920665387,
-      "grad_norm": 5.655236437132318,
+      "grad_norm": 6.7112612127648505,
       "learning_rate": 3.5031847133757957e-07,
-      "logits/generated": -2.9382705688476562,
-      "logits/real": -2.895289421081543,
-      "logps/generated": -111.1765365600586,
-      "logps/real": -122.95256042480469,
-      "loss": 0.2607,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -3.181370258331299,
-      "rewards/margins": 12.501322746276855,
-      "rewards/real": 9.319952011108398,
+      "logits/generated": -2.9418439865112305,
+      "logits/real": -2.896251678466797,
+      "logps/generated": -111.11647033691406,
+      "logps/real": -122.94500732421875,
+      "loss": 0.2608,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -3.175363302230835,
+      "rewards/margins": 12.496070861816406,
+      "rewards/real": 9.320706367492676,
       "step": 110
     },
     {
       "epoch": 0.07677543186180422,
-      "grad_norm": 10.474178085979151,
+      "grad_norm": 26.396184251399095,
       "learning_rate": 3.8216560509554143e-07,
-      "logits/generated": -2.9170148372650146,
-      "logits/real": -2.9494102001190186,
-      "logps/generated": -120.6019515991211,
-      "logps/real": -120.84870910644531,
-      "loss": 0.2583,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -3.854489803314209,
-      "rewards/margins": 13.146855354309082,
-      "rewards/real": 9.292365074157715,
+      "logits/generated": -2.932110548019409,
+      "logits/real": -2.9671242237091064,
+      "logps/generated": -120.4710693359375,
+      "logps/real": -121.0118408203125,
+      "loss": 0.2575,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -3.8414013385772705,
+      "rewards/margins": 13.117452621459961,
+      "rewards/real": 9.27605152130127,
       "step": 120
     },
     {
       "epoch": 0.08317338451695458,
-      "grad_norm": 3.425127673225565,
+      "grad_norm": 3.4397935654373515,
       "learning_rate": 4.140127388535032e-07,
-      "logits/generated": -2.962573528289795,
-      "logits/real": -2.9198384284973145,
-      "logps/generated": -119.30030822753906,
-      "logps/real": -114.26997375488281,
-      "loss": 0.2475,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -4.209897041320801,
-      "rewards/margins": 13.073972702026367,
-      "rewards/real": 8.864076614379883,
+      "logits/generated": -2.922434091567993,
+      "logits/real": -2.8773040771484375,
+      "logps/generated": -118.9048080444336,
+      "logps/real": -114.76216125488281,
+      "loss": 0.2439,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -4.17034912109375,
+      "rewards/margins": 12.985206604003906,
+      "rewards/real": 8.814857482910156,
       "step": 130
     },
     {
       "epoch": 0.08957133717210493,
-      "grad_norm": 2.8766539348394065,
+      "grad_norm": 2.869584316013309,
       "learning_rate": 4.4585987261146494e-07,
-      "logits/generated": -2.902782440185547,
-      "logits/real": -2.8874409198760986,
-      "logps/generated": -138.11538696289062,
-      "logps/real": -128.03103637695312,
-      "loss": 0.2566,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.331695556640625,
-      "rewards/margins": 15.117105484008789,
-      "rewards/real": 9.785411834716797,
+      "logits/generated": -2.8454012870788574,
+      "logits/real": -2.831343173980713,
+      "logps/generated": -137.0986328125,
+      "logps/real": -128.74740600585938,
+      "loss": 0.2553,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.2300214767456055,
+      "rewards/margins": 14.943796157836914,
+      "rewards/real": 9.713775634765625,
       "step": 140
     },
     {
       "epoch": 0.09596928982725528,
-      "grad_norm": 2.928535512875351,
+      "grad_norm": 3.3106537474929794,
       "learning_rate": 4.777070063694267e-07,
-      "logits/generated": -2.9390759468078613,
-      "logits/real": -2.9156928062438965,
-      "logps/generated": -128.9432373046875,
-      "logps/real": -112.00594329833984,
-      "loss": 0.2422,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -4.531386375427246,
-      "rewards/margins": 13.208353042602539,
-      "rewards/real": 8.676966667175293,
+      "logits/generated": -2.8886618614196777,
+      "logits/real": -2.8667774200439453,
+      "logps/generated": -129.49241638183594,
+      "logps/real": -112.80342864990234,
+      "loss": 0.2406,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -4.586302757263184,
+      "rewards/margins": 13.183520317077637,
+      "rewards/real": 8.597217559814453,
       "step": 150
     },
     {
       "epoch": 0.10236724248240563,
-      "grad_norm": 4.646412917226807,
+      "grad_norm": 6.648711627699365,
       "learning_rate": 4.989331436699858e-07,
-      "logits/generated": -2.907299041748047,
-      "logits/real": -2.8829636573791504,
-      "logps/generated": -147.1040802001953,
-      "logps/real": -130.88232421875,
-      "loss": 0.2552,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.514917373657227,
-      "rewards/margins": 16.280481338500977,
-      "rewards/real": 10.76556396484375,
+      "logits/generated": -2.8701493740081787,
+      "logits/real": -2.834799289703369,
+      "logps/generated": -145.00540161132812,
+      "logps/real": -130.97731018066406,
+      "loss": 0.254,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.305049419403076,
+      "rewards/margins": 16.061115264892578,
+      "rewards/real": 10.75606632232666,
       "step": 160
     },
     {
       "epoch": 0.10876519513755598,
-      "grad_norm": 2.67790411455109,
+      "grad_norm": 2.896327038817593,
       "learning_rate": 4.953769559032717e-07,
-      "logits/generated": -2.8610851764678955,
-      "logits/real": -2.931917667388916,
-      "logps/generated": -134.75003051757812,
-      "logps/real": -93.88365173339844,
-      "loss": 0.2215,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -4.963212013244629,
-      "rewards/margins": 13.223382949829102,
-      "rewards/real": 8.260170936584473,
+      "logits/generated": -2.818129777908325,
+      "logits/real": -2.861079454421997,
+      "logps/generated": -133.56591796875,
+      "logps/real": -94.10267639160156,
+      "loss": 0.2217,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -4.84480094909668,
+      "rewards/margins": 13.083070755004883,
+      "rewards/real": 8.23826789855957,
       "step": 170
     },
     {
       "epoch": 0.11516314779270634,
-      "grad_norm": 3.0194042444553317,
+      "grad_norm": 3.3492243318176094,
       "learning_rate": 4.918207681365576e-07,
-      "logits/generated": -2.8236758708953857,
-      "logits/real": -2.895191192626953,
-      "logps/generated": -128.0216522216797,
-      "logps/real": -105.7012939453125,
-      "loss": 0.2328,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -4.834359645843506,
-      "rewards/margins": 14.000758171081543,
-      "rewards/real": 9.166399002075195,
+      "logits/generated": -2.8010871410369873,
+      "logits/real": -2.8660852909088135,
+      "logps/generated": -127.16666412353516,
+      "logps/real": -106.13145446777344,
+      "loss": 0.2314,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -4.748859882354736,
+      "rewards/margins": 13.87224292755127,
+      "rewards/real": 9.123383522033691,
       "step": 180
     },
     {
       "epoch": 0.12156110044785669,
-      "grad_norm": 7.731581783616942,
+      "grad_norm": 9.892939505336098,
       "learning_rate": 4.882645803698435e-07,
-      "logits/generated": -2.880075693130493,
-      "logits/real": -2.863041400909424,
-      "logps/generated": -129.73480224609375,
-      "logps/real": -121.33485412597656,
-      "loss": 0.2407,
+      "logits/generated": -2.896153211593628,
+      "logits/real": -2.870706796646118,
+      "logps/generated": -129.23936462402344,
+      "logps/real": -121.87593841552734,
+      "loss": 0.2402,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/generated": -4.882136344909668,
-      "rewards/margins": 15.376531600952148,
-      "rewards/real": 10.49439525604248,
+      "rewards/generated": -4.832592964172363,
+      "rewards/margins": 15.272882461547852,
+      "rewards/real": 10.440288543701172,
       "step": 190
     },
     {
       "epoch": 0.12795905310300704,
-      "grad_norm": 6.783378094196307,
+      "grad_norm": 12.363845887191474,
       "learning_rate": 4.847083926031294e-07,
-      "logits/generated": -2.8695290088653564,
-      "logits/real": -2.863180637359619,
-      "logps/generated": -143.60906982421875,
-      "logps/real": -90.62736511230469,
-      "loss": 0.2262,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.7865214347839355,
-      "rewards/margins": 13.516904830932617,
-      "rewards/real": 7.73038387298584,
+      "logits/generated": -2.863784074783325,
+      "logits/real": -2.8600029945373535,
+      "logps/generated": -142.89822387695312,
+      "logps/real": -91.83329010009766,
+      "loss": 0.2283,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.715437889099121,
+      "rewards/margins": 13.325228691101074,
+      "rewards/real": 7.609792232513428,
       "step": 200
     },
     {
       "epoch": 0.12795905310300704,
-      "eval_logits/generated": -2.833332061767578,
-      "eval_logits/real": -2.8265442848205566,
-      "eval_logps/generated": -126.3611831665039,
-      "eval_logps/real": -117.04948425292969,
-      "eval_loss": 0.2435438483953476,
+      "eval_logits/generated": -2.8307945728302,
+      "eval_logits/real": -2.819204330444336,
+      "eval_logps/generated": -124.98797607421875,
+      "eval_logps/real": -116.981689453125,
+      "eval_loss": 0.24378155171871185,
       "eval_rewards/accuracies": 0.9921875,
-      "eval_rewards/generated": -4.7644171714782715,
-      "eval_rewards/margins": 14.327579498291016,
-      "eval_rewards/real": 9.563161849975586,
-      "eval_runtime": 57.2388,
-      "eval_samples_per_second": 8.735,
-      "eval_steps_per_second": 0.28,
+      "eval_rewards/generated": -4.627096176147461,
+      "eval_rewards/margins": 14.197038650512695,
+      "eval_rewards/real": 9.569941520690918,
+      "eval_runtime": 68.1859,
+      "eval_samples_per_second": 7.333,
+      "eval_steps_per_second": 0.235,
       "step": 200
     },
     {
       "epoch": 0.1343570057581574,
-      "grad_norm": 4.630294431619136,
+      "grad_norm": 11.343173710749236,
       "learning_rate": 4.811522048364154e-07,
-      "logits/generated": -2.848015546798706,
-      "logits/real": -2.878896474838257,
-      "logps/generated": -134.64739990234375,
-      "logps/real": -130.23609924316406,
-      "loss": 0.2535,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.333014965057373,
-      "rewards/margins": 16.449054718017578,
-      "rewards/real": 11.11603832244873,
+      "logits/generated": -2.834120273590088,
+      "logits/real": -2.8591551780700684,
+      "logps/generated": -133.72886657714844,
+      "logps/real": -130.7188262939453,
+      "loss": 0.2534,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.2411603927612305,
+      "rewards/margins": 16.30892562866211,
+      "rewards/real": 11.067764282226562,
       "step": 210
     },
     {
       "epoch": 0.14075495841330773,
-      "grad_norm": 3.37570780842826,
+      "grad_norm": 3.4857481537447996,
       "learning_rate": 4.775960170697012e-07,
-      "logits/generated": -2.9035279750823975,
-      "logits/real": -2.8899435997009277,
-      "logps/generated": -159.3094024658203,
-      "logps/real": -124.04632568359375,
-      "loss": 0.243,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -6.095156669616699,
-      "rewards/margins": 16.67739486694336,
-      "rewards/real": 10.58223819732666,
+      "logits/generated": -2.8847694396972656,
+      "logits/real": -2.8639719486236572,
+      "logps/generated": -159.26048278808594,
+      "logps/real": -124.52852630615234,
+      "loss": 0.2438,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -6.090262413024902,
+      "rewards/margins": 16.624279022216797,
+      "rewards/real": 10.534018516540527,
       "step": 220
     },
     {
       "epoch": 0.1471529110684581,
-      "grad_norm": 2.451067761798054,
+      "grad_norm": 2.5439877470663137,
       "learning_rate": 4.7403982930298717e-07,
-      "logits/generated": -2.87707257270813,
-      "logits/real": -2.901421546936035,
-      "logps/generated": -120.0963134765625,
-      "logps/real": -123.16734313964844,
-      "loss": 0.2359,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -4.565489768981934,
-      "rewards/margins": 15.016607284545898,
-      "rewards/real": 10.451117515563965,
+      "logits/generated": -2.856072187423706,
+      "logits/real": -2.8673479557037354,
+      "logps/generated": -120.0430679321289,
+      "logps/real": -123.7723617553711,
+      "loss": 0.2368,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -4.560164451599121,
+      "rewards/margins": 14.950778007507324,
+      "rewards/real": 10.390615463256836,
       "step": 230
     },
     {
       "epoch": 0.15355086372360843,
-      "grad_norm": 2.3458265316121145,
+      "grad_norm": 2.3771095937929765,
       "learning_rate": 4.7048364153627306e-07,
-      "logits/generated": -2.845377206802368,
-      "logits/real": -2.9469656944274902,
-      "logps/generated": -130.70401000976562,
-      "logps/real": -102.62498474121094,
-      "loss": 0.2292,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -4.917192459106445,
-      "rewards/margins": 13.963025093078613,
-      "rewards/real": 9.045831680297852,
+      "logits/generated": -2.8317062854766846,
+      "logits/real": -2.8830018043518066,
+      "logps/generated": -129.07260131835938,
+      "logps/real": -103.09846496582031,
+      "loss": 0.2315,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/generated": -4.754051685333252,
+      "rewards/margins": 13.752535820007324,
+      "rewards/real": 8.99848461151123,
       "step": 240
     },
     {
       "epoch": 0.1599488163787588,
-      "grad_norm": 2.2252235869873425,
+      "grad_norm": 2.2413244849118565,
       "learning_rate": 4.66927453769559e-07,
-      "logits/generated": -2.814479351043701,
-      "logits/real": -2.79661226272583,
-      "logps/generated": -129.10992431640625,
-      "logps/real": -110.62039947509766,
-      "loss": 0.2418,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -4.840092182159424,
-      "rewards/margins": 14.007853507995605,
-      "rewards/real": 9.16776180267334,
+      "logits/generated": -2.803177833557129,
+      "logits/real": -2.7696690559387207,
+      "logps/generated": -127.9206314086914,
+      "logps/real": -110.67021179199219,
+      "loss": 0.2428,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/generated": -4.72116231918335,
+      "rewards/margins": 13.883941650390625,
+      "rewards/real": 9.1627779006958,
       "step": 250
     },
     {
       "epoch": 0.16634676903390916,
-      "grad_norm": 3.2120162920530775,
+      "grad_norm": 3.7576364979309087,
       "learning_rate": 4.633712660028449e-07,
-      "logits/generated": -2.7691586017608643,
-      "logits/real": -2.803241729736328,
-      "logps/generated": -130.2401885986328,
-      "logps/real": -103.78974914550781,
-      "loss": 0.2207,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -4.814304828643799,
-      "rewards/margins": 14.570037841796875,
-      "rewards/real": 9.755732536315918,
+      "logits/generated": -2.786233425140381,
+      "logits/real": -2.7940025329589844,
+      "logps/generated": -128.66128540039062,
+      "logps/real": -103.73394775390625,
+      "loss": 0.221,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -4.656413555145264,
+      "rewards/margins": 14.41772747039795,
+      "rewards/real": 9.761313438415527,
       "step": 260
     },
     {
       "epoch": 0.1727447216890595,
-      "grad_norm": 2.265151763557376,
+      "grad_norm": 3.0970314717177425,
       "learning_rate": 4.5981507823613085e-07,
-      "logits/generated": -2.7566680908203125,
-      "logits/real": -2.747866153717041,
-      "logps/generated": -152.24049377441406,
-      "logps/real": -107.55720520019531,
-      "loss": 0.2249,
+      "logits/generated": -2.7715699672698975,
+      "logits/real": -2.736011028289795,
+      "logps/generated": -151.93618774414062,
+      "logps/real": -107.65568542480469,
+      "loss": 0.2243,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/generated": -6.19102144241333,
-      "rewards/margins": 15.924577713012695,
-      "rewards/real": 9.733556747436523,
+      "rewards/generated": -6.160590648651123,
+      "rewards/margins": 15.884298324584961,
+      "rewards/real": 9.72370719909668,
       "step": 270
     },
     {
       "epoch": 0.17914267434420986,
-      "grad_norm": 3.5602666832695844,
+      "grad_norm": 3.1841304688976515,
       "learning_rate": 4.562588904694168e-07,
-      "logits/generated": -2.7723679542541504,
-      "logits/real": -2.785216808319092,
-      "logps/generated": -128.1282958984375,
-      "logps/real": -111.1799545288086,
-      "loss": 0.2202,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -4.806519508361816,
-      "rewards/margins": 15.082513809204102,
-      "rewards/real": 10.275992393493652,
+      "logits/generated": -2.7978196144104004,
+      "logits/real": -2.7954959869384766,
+      "logps/generated": -127.94291687011719,
+      "logps/real": -111.31756591796875,
+      "loss": 0.2204,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -4.7879815101623535,
+      "rewards/margins": 15.050213813781738,
+      "rewards/real": 10.262231826782227,
       "step": 280
     },
     {
       "epoch": 0.1855406269993602,
-      "grad_norm": 3.11519890787563,
+      "grad_norm": 3.0783563347015477,
       "learning_rate": 4.5270270270270264e-07,
-      "logits/generated": -2.757876396179199,
-      "logits/real": -2.795422077178955,
-      "logps/generated": -128.11813354492188,
-      "logps/real": -102.88358306884766,
-      "loss": 0.233,
+      "logits/generated": -2.7986302375793457,
+      "logits/real": -2.799823045730591,
+      "logps/generated": -127.54069519042969,
+      "logps/real": -103.19920349121094,
+      "loss": 0.2406,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/generated": -4.833106994628906,
-      "rewards/margins": 14.29919147491455,
-      "rewards/real": 9.466082572937012,
+      "rewards/generated": -4.77536153793335,
+      "rewards/margins": 14.209879875183105,
+      "rewards/real": 9.43451976776123,
       "step": 290
     },
     {
       "epoch": 0.19193857965451055,
-      "grad_norm": 3.487329720349146,
+      "grad_norm": 3.5430526371071456,
       "learning_rate": 4.491465149359886e-07,
-      "logits/generated": -2.8109488487243652,
-      "logits/real": -2.7747206687927246,
-      "logps/generated": -129.84274291992188,
-      "logps/real": -106.18394470214844,
-      "loss": 0.2286,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -4.749394416809082,
-      "rewards/margins": 14.446981430053711,
-      "rewards/real": 9.697588920593262,
+      "logits/generated": -2.85805082321167,
+      "logits/real": -2.804370164871216,
+      "logps/generated": -129.9695281982422,
+      "logps/real": -106.55607604980469,
+      "loss": 0.2284,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -4.7620744705200195,
+      "rewards/margins": 14.422449111938477,
+      "rewards/real": 9.660375595092773,
+      "step": 300
+    },
+    {
+      "epoch": 0.19193857965451055,
+      "eval_logits/generated": -2.8395633697509766,
+      "eval_logits/real": -2.8204095363616943,
+      "eval_logps/generated": -129.49810791015625,
+      "eval_logps/real": -114.83212280273438,
+      "eval_loss": 0.2383781373500824,
+      "eval_rewards/accuracies": 0.9921875,
+      "eval_rewards/generated": -5.078109264373779,
+      "eval_rewards/margins": 14.863008499145508,
+      "eval_rewards/real": 9.784897804260254,
+      "eval_runtime": 65.3189,
+      "eval_samples_per_second": 7.655,
+      "eval_steps_per_second": 0.245,
       "step": 300
     },
     {
       "epoch": 0.19833653230966092,
-      "grad_norm": 4.236344907363098,
+      "grad_norm": 4.338072435391337,
       "learning_rate": 4.4559032716927454e-07,
-      "logits/generated": -2.7962634563446045,
-      "logits/real": -2.86857533454895,
-      "logps/generated": -137.8242645263672,
-      "logps/real": -126.29652404785156,
-      "loss": 0.2454,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.464461803436279,
-      "rewards/margins": 16.495359420776367,
-      "rewards/real": 11.030899047851562,
+      "logits/generated": -2.8614089488983154,
+      "logits/real": -2.921618938446045,
+      "logps/generated": -138.57562255859375,
+      "logps/real": -126.46247863769531,
+      "loss": 0.246,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.5395989418029785,
+      "rewards/margins": 16.55390167236328,
+      "rewards/real": 11.014303207397461,
       "step": 310
     },
     {
       "epoch": 0.20473448496481125,
-      "grad_norm": 2.0886040022985206,
+      "grad_norm": 2.2368921036154736,
       "learning_rate": 4.420341394025605e-07,
-      "logits/generated": -2.780796766281128,
-      "logits/real": -2.830658197402954,
-      "logps/generated": -128.07977294921875,
-      "logps/real": -102.74151611328125,
-      "loss": 0.2217,
+      "logits/generated": -2.850996732711792,
+      "logits/real": -2.8796534538269043,
+      "logps/generated": -130.45950317382812,
+      "logps/real": -102.80158996582031,
+      "loss": 0.2227,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/generated": -4.719244956970215,
-      "rewards/margins": 13.948654174804688,
-      "rewards/real": 9.229410171508789,
+      "rewards/generated": -4.957219123840332,
+      "rewards/margins": 14.180620193481445,
+      "rewards/real": 9.22340202331543,
       "step": 320
     },
     {
       "epoch": 0.21113243761996162,
-      "grad_norm": 2.7732179589224737,
+      "grad_norm": 2.5901712732984046,
       "learning_rate": 4.384779516358463e-07,
-      "logits/generated": -2.771653175354004,
-      "logits/real": -2.719320058822632,
-      "logps/generated": -128.0609130859375,
-      "logps/real": -110.12910461425781,
-      "loss": 0.2298,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -4.775327682495117,
-      "rewards/margins": 14.936683654785156,
-      "rewards/real": 10.161355972290039,
+      "logits/generated": -2.8232226371765137,
+      "logits/real": -2.7724738121032715,
+      "logps/generated": -129.60939025878906,
+      "logps/real": -110.3594741821289,
+      "loss": 0.2302,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -4.930176734924316,
+      "rewards/margins": 15.068493843078613,
+      "rewards/real": 10.138318061828613,
       "step": 330
     },
     {
       "epoch": 0.21753039027511195,
-      "grad_norm": 2.4432792911747327,
+      "grad_norm": 2.635446881296208,
       "learning_rate": 4.3492176386913227e-07,
-      "logits/generated": -2.7587437629699707,
-      "logits/real": -2.802103042602539,
-      "logps/generated": -152.65858459472656,
-      "logps/real": -112.84379577636719,
-      "loss": 0.2312,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -6.072720527648926,
-      "rewards/margins": 16.01877212524414,
-      "rewards/real": 9.946051597595215,
+      "logits/generated": -2.8154995441436768,
+      "logits/real": -2.841954469680786,
+      "logps/generated": -153.96469116210938,
+      "logps/real": -112.9310531616211,
+      "loss": 0.231,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -6.203330993652344,
+      "rewards/margins": 16.14065933227539,
+      "rewards/real": 9.937326431274414,
       "step": 340
     },
     {
       "epoch": 0.22392834293026231,
-      "grad_norm": 4.160225941183556,
+      "grad_norm": 4.3992604924575405,
       "learning_rate": 4.313655761024182e-07,
-      "logits/generated": -2.7678041458129883,
-      "logits/real": -2.7790327072143555,
-      "logps/generated": -120.19953918457031,
-      "logps/real": -113.22270202636719,
-      "loss": 0.2164,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -4.654033660888672,
-      "rewards/margins": 14.861625671386719,
-      "rewards/real": 10.207592010498047,
+      "logits/generated": -2.8233864307403564,
+      "logits/real": -2.828899383544922,
+      "logps/generated": -120.79425048828125,
+      "logps/real": -113.58453369140625,
+      "loss": 0.2168,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -4.713505744934082,
+      "rewards/margins": 14.884913444519043,
+      "rewards/real": 10.171408653259277,
       "step": 350
     },
     {
       "epoch": 0.23032629558541268,
-      "grad_norm": 2.2403711981241106,
+      "grad_norm": 11.744067946216896,
       "learning_rate": 4.278093883357041e-07,
-      "logits/generated": -2.836555004119873,
-      "logits/real": -2.6989388465881348,
-      "logps/generated": -133.00775146484375,
-      "logps/real": -108.97412109375,
-      "loss": 0.2254,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/generated": -4.826836109161377,
-      "rewards/margins": 14.98741340637207,
-      "rewards/real": 10.160576820373535,
+      "logits/generated": -2.8917436599731445,
+      "logits/real": -2.757673978805542,
+      "logps/generated": -134.52706909179688,
+      "logps/real": -109.41776275634766,
+      "loss": 0.2238,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/generated": -4.9787678718566895,
+      "rewards/margins": 15.094983100891113,
+      "rewards/real": 10.116214752197266,
       "step": 360
     },
     {
       "epoch": 0.236724248240563,
-      "grad_norm": 3.08941181412325,
+      "grad_norm": 2.5704831812387376,
       "learning_rate": 4.2425320056899e-07,
-      "logits/generated": -2.7016754150390625,
-      "logits/real": -2.7019612789154053,
-      "logps/generated": -125.4333724975586,
-      "logps/real": -107.02604675292969,
-      "loss": 0.2184,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -4.93725061416626,
-      "rewards/margins": 14.800321578979492,
-      "rewards/real": 9.86307144165039,
+      "logits/generated": -2.7886219024658203,
+      "logits/real": -2.7841835021972656,
+      "logps/generated": -126.40400695800781,
+      "logps/real": -107.36631774902344,
+      "loss": 0.2183,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.034313678741455,
+      "rewards/margins": 14.863359451293945,
+      "rewards/real": 9.829044342041016,
       "step": 370
     },
     {
       "epoch": 0.24312220089571338,
-      "grad_norm": 8.811415551257076,
+      "grad_norm": 8.580079528543488,
       "learning_rate": 4.2069701280227595e-07,
-      "logits/generated": -2.7188172340393066,
-      "logits/real": -2.777225971221924,
-      "logps/generated": -131.77468872070312,
-      "logps/real": -97.81922912597656,
-      "loss": 0.2132,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -4.891592979431152,
-      "rewards/margins": 14.796887397766113,
-      "rewards/real": 9.905295372009277,
+      "logits/generated": -2.773625612258911,
+      "logits/real": -2.8157546520233154,
+      "logps/generated": -130.76388549804688,
+      "logps/real": -97.5969467163086,
+      "loss": 0.2136,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -4.7905097007751465,
+      "rewards/margins": 14.718034744262695,
+      "rewards/real": 9.927522659301758,
       "step": 380
     },
     {
       "epoch": 0.2495201535508637,
-      "grad_norm": 2.1656082848995992,
+      "grad_norm": 2.1703781751821363,
       "learning_rate": 4.1714082503556185e-07,
-      "logits/generated": -2.81782865524292,
-      "logits/real": -2.733973264694214,
-      "logps/generated": -125.90080261230469,
-      "logps/real": -108.9613037109375,
-      "loss": 0.208,
+      "logits/generated": -2.8539223670959473,
+      "logits/real": -2.776035785675049,
+      "logps/generated": -125.4119873046875,
+      "logps/real": -109.11286926269531,
+      "loss": 0.2083,
       "rewards/accuracies": 1.0,
-      "rewards/generated": -4.474228858947754,
-      "rewards/margins": 14.704130172729492,
-      "rewards/real": 10.229901313781738,
+      "rewards/generated": -4.4253458976745605,
+      "rewards/margins": 14.640092849731445,
+      "rewards/real": 10.214746475219727,
       "step": 390
     },
     {
       "epoch": 0.2559181062060141,
-      "grad_norm": 2.1913448139958307,
+      "grad_norm": 3.7601859793500116,
       "learning_rate": 4.135846372688478e-07,
-      "logits/generated": -2.741250514984131,
-      "logits/real": -2.797882080078125,
-      "logps/generated": -140.27984619140625,
-      "logps/real": -113.54410552978516,
-      "loss": 0.2141,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.6347832679748535,
-      "rewards/margins": 16.515918731689453,
-      "rewards/real": 10.881135940551758,
+      "logits/generated": -2.7994043827056885,
+      "logits/real": -2.8369317054748535,
+      "logps/generated": -141.95204162597656,
+      "logps/real": -113.73207092285156,
+      "loss": 0.2154,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.802001953125,
+      "rewards/margins": 16.664339065551758,
+      "rewards/real": 10.862336158752441,
       "step": 400
     },
     {
       "epoch": 0.2559181062060141,
-      "eval_logits/generated": -2.7752091884613037,
-      "eval_logits/real": -2.76131010055542,
-      "eval_logps/generated": -128.1855010986328,
-      "eval_logps/real": -113.70217895507812,
-      "eval_loss": 0.23566634953022003,
+      "eval_logits/generated": -2.830336093902588,
+      "eval_logits/real": -2.8085427284240723,
+      "eval_logps/generated": -127.63105773925781,
+      "eval_logps/real": -113.71005249023438,
+      "eval_loss": 0.2360752820968628,
       "eval_rewards/accuracies": 0.9921875,
-      "eval_rewards/generated": -4.946849822998047,
-      "eval_rewards/margins": 14.844742774963379,
-      "eval_rewards/real": 9.897892951965332,
-      "eval_runtime": 70.5415,
-      "eval_samples_per_second": 7.088,
-      "eval_steps_per_second": 0.227,
+      "eval_rewards/generated": -4.891404628753662,
+      "eval_rewards/margins": 14.788509368896484,
+      "eval_rewards/real": 9.89710521697998,
+      "eval_runtime": 65.951,
+      "eval_samples_per_second": 7.581,
+      "eval_steps_per_second": 0.243,
       "step": 400
     },
     {
       "epoch": 0.26231605886116444,
-      "grad_norm": 3.5382231043936563,
+      "grad_norm": 6.383162655219248,
       "learning_rate": 4.100284495021337e-07,
-      "logits/generated": -2.7606406211853027,
-      "logits/real": -2.8516087532043457,
-      "logps/generated": -150.7864990234375,
-      "logps/real": -107.80538177490234,
-      "loss": 0.222,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -6.30859899520874,
-      "rewards/margins": 16.771678924560547,
-      "rewards/real": 10.463083267211914,
+      "logits/generated": -2.8110575675964355,
+      "logits/real": -2.867323875427246,
+      "logps/generated": -149.95602416992188,
+      "logps/real": -107.71382904052734,
+      "loss": 0.2229,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -6.225547790527344,
+      "rewards/margins": 16.697784423828125,
+      "rewards/real": 10.472236633300781,
       "step": 410
     },
     {
       "epoch": 0.2687140115163148,
-      "grad_norm": 2.180636822570479,
+      "grad_norm": 5.623968249132037,
       "learning_rate": 4.064722617354196e-07,
-      "logits/generated": -2.777144193649292,
-      "logits/real": -2.7791004180908203,
-      "logps/generated": -141.62130737304688,
-      "logps/real": -111.6278305053711,
-      "loss": 0.2179,
+      "logits/generated": -2.823819160461426,
+      "logits/real": -2.796905755996704,
+      "logps/generated": -142.87222290039062,
+      "logps/real": -111.68009185791016,
+      "loss": 0.2184,
       "rewards/accuracies": 1.0,
-      "rewards/generated": -5.845305442810059,
-      "rewards/margins": 16.107940673828125,
-      "rewards/real": 10.262636184692383,
+      "rewards/generated": -5.970399856567383,
+      "rewards/margins": 16.22780990600586,
+      "rewards/real": 10.257412910461426,
       "step": 420
     },
     {
       "epoch": 0.2751119641714651,
-      "grad_norm": 7.826230059811001,
+      "grad_norm": 9.012135226128816,
       "learning_rate": 4.0291607396870553e-07,
-      "logits/generated": -2.7701034545898438,
-      "logits/real": -2.835702657699585,
-      "logps/generated": -134.1055450439453,
-      "logps/real": -115.15937805175781,
-      "loss": 0.2277,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.376823425292969,
-      "rewards/margins": 16.15007972717285,
-      "rewards/real": 10.773252487182617,
+      "logits/generated": -2.818744659423828,
+      "logits/real": -2.854924201965332,
+      "logps/generated": -134.91293334960938,
+      "logps/real": -115.4572982788086,
+      "loss": 0.2283,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.4575629234313965,
+      "rewards/margins": 16.20102310180664,
+      "rewards/real": 10.743459701538086,
       "step": 430
     },
     {
       "epoch": 0.28150991682661547,
-      "grad_norm": 4.574173821690705,
+      "grad_norm": 6.106542046916075,
       "learning_rate": 3.993598862019915e-07,
-      "logits/generated": -2.7749392986297607,
-      "logits/real": -2.82127046585083,
-      "logps/generated": -133.71063232421875,
-      "logps/real": -103.6177978515625,
-      "loss": 0.2214,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.1843953132629395,
-      "rewards/margins": 14.904150009155273,
-      "rewards/real": 9.719755172729492,
+      "logits/generated": -2.816333293914795,
+      "logits/real": -2.8392767906188965,
+      "logps/generated": -134.35879516601562,
+      "logps/real": -103.79875183105469,
+      "loss": 0.2198,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.249211311340332,
+      "rewards/margins": 14.950872421264648,
+      "rewards/real": 9.70166015625,
       "step": 440
     },
     {
       "epoch": 0.28790786948176583,
-      "grad_norm": 2.774800684163752,
+      "grad_norm": 3.069855045067045,
       "learning_rate": 3.9580369843527737e-07,
-      "logits/generated": -2.744696617126465,
-      "logits/real": -2.7654058933258057,
-      "logps/generated": -157.8831024169922,
-      "logps/real": -103.58988952636719,
-      "loss": 0.2138,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -6.751193046569824,
-      "rewards/margins": 17.019277572631836,
-      "rewards/real": 10.268084526062012,
+      "logits/generated": -2.80080509185791,
+      "logits/real": -2.7943968772888184,
+      "logps/generated": -158.69754028320312,
+      "logps/real": -104.19126892089844,
+      "loss": 0.2134,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -6.832635402679443,
+      "rewards/margins": 17.04058265686035,
+      "rewards/real": 10.207947731018066,
       "step": 450
     },
     {
       "epoch": 0.2943058221369162,
-      "grad_norm": 5.160429480281925,
+      "grad_norm": 4.764120113518142,
       "learning_rate": 3.9224751066856327e-07,
-      "logits/generated": -2.7339484691619873,
-      "logits/real": -2.685930013656616,
-      "logps/generated": -135.38714599609375,
-      "logps/real": -114.84906005859375,
-      "loss": 0.2204,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.209853649139404,
-      "rewards/margins": 15.920175552368164,
-      "rewards/real": 10.710321426391602,
+      "logits/generated": -2.7790608406066895,
+      "logits/real": -2.733055591583252,
+      "logps/generated": -136.2861785888672,
+      "logps/real": -114.9732437133789,
+      "loss": 0.2209,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.299757480621338,
+      "rewards/margins": 15.997659683227539,
+      "rewards/real": 10.697901725769043,
       "step": 460
     },
     {
       "epoch": 0.30070377479206656,
-      "grad_norm": 4.706717543122211,
+      "grad_norm": 6.02409783080102,
       "learning_rate": 3.886913229018492e-07,
-      "logits/generated": -2.7286899089813232,
-      "logits/real": -2.782683849334717,
-      "logps/generated": -146.5428924560547,
-      "logps/real": -119.84122467041016,
-      "loss": 0.2176,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.839329719543457,
-      "rewards/margins": 16.770212173461914,
-      "rewards/real": 10.930883407592773,
+      "logits/generated": -2.7785205841064453,
+      "logits/real": -2.806605815887451,
+      "logps/generated": -146.60862731933594,
+      "logps/real": -120.13336181640625,
+      "loss": 0.2187,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.845905303955078,
+      "rewards/margins": 16.747573852539062,
+      "rewards/real": 10.9016695022583,
       "step": 470
     },
     {
       "epoch": 0.30710172744721687,
-      "grad_norm": 2.3058413282374715,
+      "grad_norm": 2.4688287560199376,
       "learning_rate": 3.851351351351351e-07,
-      "logits/generated": -2.7684266567230225,
-      "logits/real": -2.802767276763916,
-      "logps/generated": -138.41696166992188,
-      "logps/real": -113.2447738647461,
-      "loss": 0.2321,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.6439032554626465,
-      "rewards/margins": 15.897249221801758,
-      "rewards/real": 10.253345489501953,
+      "logits/generated": -2.8146071434020996,
+      "logits/real": -2.835864543914795,
+      "logps/generated": -139.945068359375,
+      "logps/real": -113.28076171875,
+      "loss": 0.2331,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/generated": -5.7967119216918945,
+      "rewards/margins": 16.046459197998047,
+      "rewards/real": 10.249748229980469,
       "step": 480
     },
     {
       "epoch": 0.31349968010236723,
-      "grad_norm": 3.3424369618748706,
+      "grad_norm": 4.977535722385729,
       "learning_rate": 3.8157894736842105e-07,
-      "logits/generated": -2.766950845718384,
-      "logits/real": -2.794663906097412,
-      "logps/generated": -123.19486236572266,
-      "logps/real": -98.98766326904297,
-      "loss": 0.2199,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -4.739457130432129,
-      "rewards/margins": 14.55401611328125,
-      "rewards/real": 9.814558029174805,
+      "logits/generated": -2.804661273956299,
+      "logits/real": -2.8211963176727295,
+      "logps/generated": -123.10560607910156,
+      "logps/real": -99.4258041381836,
+      "loss": 0.2192,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -4.730529308319092,
+      "rewards/margins": 14.501276016235352,
+      "rewards/real": 9.770744323730469,
       "step": 490
     },
     {
       "epoch": 0.3198976327575176,
-      "grad_norm": 2.169410849214599,
+      "grad_norm": 12.370683444840274,
       "learning_rate": 3.7802275960170695e-07,
-      "logits/generated": -2.715301513671875,
-      "logits/real": -2.8804736137390137,
-      "logps/generated": -140.2711181640625,
-      "logps/real": -100.17481994628906,
-      "loss": 0.2348,
+      "logits/generated": -2.769005060195923,
+      "logits/real": -2.884453058242798,
+      "logps/generated": -140.912353515625,
+      "logps/real": -99.95684814453125,
+      "loss": 0.2368,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/generated": -5.455787658691406,
-      "rewards/margins": 15.202703475952148,
-      "rewards/real": 9.746915817260742,
+      "rewards/generated": -5.519913196563721,
+      "rewards/margins": 15.28862476348877,
+      "rewards/real": 9.768712043762207,
+      "step": 500
+    },
+    {
+      "epoch": 0.3198976327575176,
+      "eval_logits/generated": -2.8227732181549072,
+      "eval_logits/real": -2.808349132537842,
+      "eval_logps/generated": -129.20452880859375,
+      "eval_logps/real": -112.91948699951172,
+      "eval_loss": 0.2350633293390274,
+      "eval_rewards/accuracies": 0.9921875,
+      "eval_rewards/generated": -5.048752784729004,
+      "eval_rewards/margins": 15.024911880493164,
+      "eval_rewards/real": 9.976161003112793,
+      "eval_runtime": 67.7839,
+      "eval_samples_per_second": 7.376,
+      "eval_steps_per_second": 0.236,
       "step": 500
     },
     {
       "epoch": 0.32629558541266795,
-      "grad_norm": 2.04238252540844,
+      "grad_norm": 2.080419019105328,
       "learning_rate": 3.7446657183499284e-07,
-      "logits/generated": -2.7904624938964844,
-      "logits/real": -2.7905654907226562,
-      "logps/generated": -145.4995574951172,
-      "logps/real": -121.39151763916016,
-      "loss": 0.2262,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.898383140563965,
-      "rewards/margins": 17.675220489501953,
-      "rewards/real": 11.776837348937988,
+      "logits/generated": -2.835850238800049,
+      "logits/real": -2.8236746788024902,
+      "logps/generated": -147.96896362304688,
+      "logps/real": -121.75404357910156,
+      "loss": 0.2259,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -6.145321846008301,
+      "rewards/margins": 17.885906219482422,
+      "rewards/real": 11.740584373474121,
       "step": 510
     },
     {
       "epoch": 0.3326935380678183,
-      "grad_norm": 2.4330841754647823,
+      "grad_norm": 2.8736781905864945,
       "learning_rate": 3.709103840682788e-07,
-      "logits/generated": -2.728691577911377,
-      "logits/real": -2.8865561485290527,
-      "logps/generated": -156.2779998779297,
-      "logps/real": -108.7221450805664,
-      "loss": 0.2081,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -6.528653144836426,
-      "rewards/margins": 16.937610626220703,
-      "rewards/real": 10.408956527709961,
+      "logits/generated": -2.7863011360168457,
+      "logits/real": -2.889273166656494,
+      "logps/generated": -157.24008178710938,
+      "logps/real": -108.51228332519531,
+      "loss": 0.207,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -6.624861240386963,
+      "rewards/margins": 17.054805755615234,
+      "rewards/real": 10.42994213104248,
       "step": 520
     },
     {
       "epoch": 0.3390914907229686,
-      "grad_norm": 3.3742934939891174,
+      "grad_norm": 3.0886072446221653,
       "learning_rate": 3.6735419630156474e-07,
-      "logits/generated": -2.819417953491211,
-      "logits/real": -2.816087245941162,
-      "logps/generated": -153.2522430419922,
-      "logps/real": -133.01565551757812,
-      "loss": 0.2314,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -6.505625247955322,
-      "rewards/margins": 18.493694305419922,
-      "rewards/real": 11.988069534301758,
+      "logits/generated": -2.866192579269409,
+      "logits/real": -2.8452916145324707,
+      "logps/generated": -153.5027313232422,
+      "logps/real": -133.2013702392578,
+      "loss": 0.2307,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -6.530673027038574,
+      "rewards/margins": 18.50016975402832,
+      "rewards/real": 11.969497680664062,
       "step": 530
     },
     {
       "epoch": 0.345489443378119,
-      "grad_norm": 2.5651527495210242,
+      "grad_norm": 2.6106684247503744,
       "learning_rate": 3.637980085348506e-07,
-      "logits/generated": -2.7839221954345703,
-      "logits/real": -2.799818515777588,
-      "logps/generated": -126.00923156738281,
-      "logps/real": -103.7330093383789,
-      "loss": 0.2194,
+      "logits/generated": -2.820413589477539,
+      "logits/real": -2.815932273864746,
+      "logps/generated": -127.53736877441406,
+      "logps/real": -103.6552963256836,
+      "loss": 0.2185,
       "rewards/accuracies": 1.0,
-      "rewards/generated": -4.9756646156311035,
-      "rewards/margins": 15.352749824523926,
-      "rewards/real": 10.377084732055664,
+      "rewards/generated": -5.128478527069092,
+      "rewards/margins": 15.513333320617676,
+      "rewards/real": 10.384856224060059,
       "step": 540
     },
     {
       "epoch": 0.35188739603326935,
-      "grad_norm": 2.096110953513035,
+      "grad_norm": 2.061456575800624,
       "learning_rate": 3.602418207681365e-07,
-      "logits/generated": -2.7390735149383545,
-      "logits/real": -2.818681240081787,
-      "logps/generated": -123.09765625,
-      "logps/real": -97.98001861572266,
-      "loss": 0.211,
+      "logits/generated": -2.786287546157837,
+      "logits/real": -2.829831123352051,
+      "logps/generated": -122.35221862792969,
+      "logps/real": -97.89688110351562,
+      "loss": 0.2113,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/generated": -4.679843902587891,
-      "rewards/margins": 14.732965469360352,
-      "rewards/real": 10.053121566772461,
+      "rewards/generated": -4.605299949645996,
+      "rewards/margins": 14.66673469543457,
+      "rewards/real": 10.061434745788574,
       "step": 550
     },
     {
       "epoch": 0.3582853486884197,
-      "grad_norm": 2.3805224123873767,
+      "grad_norm": 3.671533429639384,
       "learning_rate": 3.5668563300142247e-07,
-      "logits/generated": -2.74855899810791,
-      "logits/real": -2.8764424324035645,
-      "logps/generated": -138.00978088378906,
-      "logps/real": -104.4416275024414,
-      "loss": 0.2092,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.503247261047363,
-      "rewards/margins": 15.430496215820312,
-      "rewards/real": 9.927248001098633,
+      "logits/generated": -2.804115056991577,
+      "logits/real": -2.8819785118103027,
+      "logps/generated": -141.40806579589844,
+      "logps/real": -104.39363098144531,
+      "loss": 0.2093,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.843076229095459,
+      "rewards/margins": 15.775125503540039,
+      "rewards/real": 9.932049751281738,
       "step": 560
     },
     {
       "epoch": 0.3646833013435701,
-      "grad_norm": 2.9036557884699996,
+      "grad_norm": 2.694307081619078,
       "learning_rate": 3.5312944523470837e-07,
-      "logits/generated": -2.8142824172973633,
-      "logits/real": -2.8175225257873535,
-      "logps/generated": -138.83673095703125,
-      "logps/real": -105.87809753417969,
-      "loss": 0.2188,
+      "logits/generated": -2.8499112129211426,
+      "logits/real": -2.8300094604492188,
+      "logps/generated": -139.9486541748047,
+      "logps/real": -105.6432876586914,
+      "loss": 0.2179,
       "rewards/accuracies": 1.0,
-      "rewards/generated": -5.465325355529785,
-      "rewards/margins": 15.622749328613281,
-      "rewards/real": 10.157423973083496,
+      "rewards/generated": -5.576516628265381,
+      "rewards/margins": 15.757423400878906,
+      "rewards/real": 10.18090534210205,
       "step": 570
     },
     {
       "epoch": 0.3710812539987204,
-      "grad_norm": 2.1880847814800948,
+      "grad_norm": 2.238346737670723,
       "learning_rate": 3.495732574679943e-07,
-      "logits/generated": -2.8173575401306152,
-      "logits/real": -2.8300251960754395,
-      "logps/generated": -149.59030151367188,
-      "logps/real": -106.2435302734375,
-      "loss": 0.2234,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.7458953857421875,
-      "rewards/margins": 16.33285140991211,
-      "rewards/real": 10.586955070495605,
+      "logits/generated": -2.8515191078186035,
+      "logits/real": -2.8456454277038574,
+      "logps/generated": -148.60128784179688,
+      "logps/real": -106.5953369140625,
+      "loss": 0.2233,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.646993160247803,
+      "rewards/margins": 16.198768615722656,
+      "rewards/real": 10.551774978637695,
       "step": 580
     },
     {
       "epoch": 0.37747920665387075,
-      "grad_norm": 2.192659377942886,
+      "grad_norm": 2.0280848997133836,
       "learning_rate": 3.460170697012802e-07,
-      "logits/generated": -2.777217388153076,
-      "logits/real": -2.863020181655884,
-      "logps/generated": -129.68231201171875,
-      "logps/real": -108.95207214355469,
-      "loss": 0.2224,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -4.835152626037598,
-      "rewards/margins": 15.670806884765625,
-      "rewards/real": 10.835652351379395,
+      "logits/generated": -2.815340042114258,
+      "logits/real": -2.8666415214538574,
+      "logps/generated": -131.1103515625,
+      "logps/real": -109.18595886230469,
+      "loss": 0.2228,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -4.977957725524902,
+      "rewards/margins": 15.790220260620117,
+      "rewards/real": 10.812261581420898,
       "step": 590
     },
     {
       "epoch": 0.3838771593090211,
-      "grad_norm": 12.154950352869962,
+      "grad_norm": 8.22767535864096,
       "learning_rate": 3.424608819345661e-07,
-      "logits/generated": -2.823272228240967,
-      "logits/real": -2.7826805114746094,
-      "logps/generated": -117.91121673583984,
-      "logps/real": -111.25467681884766,
-      "loss": 0.2089,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -4.435954570770264,
-      "rewards/margins": 14.716769218444824,
-      "rewards/real": 10.280814170837402,
+      "logits/generated": -2.8348073959350586,
+      "logits/real": -2.797548294067383,
+      "logps/generated": -121.09053039550781,
+      "logps/real": -111.13722229003906,
+      "loss": 0.2065,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -4.753885269165039,
+      "rewards/margins": 15.046445846557617,
+      "rewards/real": 10.292560577392578,
       "step": 600
     },
     {
       "epoch": 0.3838771593090211,
-      "eval_logits/generated": -2.7913894653320312,
-      "eval_logits/real": -2.798403024673462,
-      "eval_logps/generated": -127.67301940917969,
-      "eval_logps/real": -112.43653106689453,
-      "eval_loss": 0.23413488268852234,
+      "eval_logits/generated": -2.8203775882720947,
+      "eval_logits/real": -2.808619976043701,
+      "eval_logps/generated": -128.3267059326172,
+      "eval_logps/real": -112.25542449951172,
+      "eval_loss": 0.23463907837867737,
       "eval_rewards/accuracies": 0.9921875,
-      "eval_rewards/generated": -4.895601272583008,
-      "eval_rewards/margins": 14.920058250427246,
-      "eval_rewards/real": 10.024457931518555,
-      "eval_runtime": 70.7535,
-      "eval_samples_per_second": 7.067,
-      "eval_steps_per_second": 0.226,
+      "eval_rewards/generated": -4.960969924926758,
+      "eval_rewards/margins": 15.003538131713867,
+      "eval_rewards/real": 10.04256820678711,
+      "eval_runtime": 66.9288,
+      "eval_samples_per_second": 7.471,
+      "eval_steps_per_second": 0.239,
       "step": 600
     },
     {
       "epoch": 0.3902751119641715,
-      "grad_norm": 2.0157880028714548,
+      "grad_norm": 2.0392053720088903,
       "learning_rate": 3.3890469416785205e-07,
-      "logits/generated": -2.7878739833831787,
-      "logits/real": -2.886143445968628,
-      "logps/generated": -132.29232788085938,
-      "logps/real": -107.32408142089844,
-      "loss": 0.2185,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.14271354675293,
-      "rewards/margins": 15.367979049682617,
-      "rewards/real": 10.225264549255371,
+      "logits/generated": -2.824596881866455,
+      "logits/real": -2.876988649368286,
+      "logps/generated": -133.09503173828125,
+      "logps/real": -107.67557525634766,
+      "loss": 0.2191,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.2229838371276855,
+      "rewards/margins": 15.41309928894043,
+      "rewards/real": 10.190114974975586,
       "step": 610
     },
     {
       "epoch": 0.39667306461932184,
-      "grad_norm": 3.4761506983717623,
+      "grad_norm": 4.866540394486641,
       "learning_rate": 3.35348506401138e-07,
-      "logits/generated": -2.800938129425049,
-      "logits/real": -2.9011588096618652,
-      "logps/generated": -152.92681884765625,
-      "logps/real": -90.30223083496094,
-      "loss": 0.2051,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -6.535437107086182,
-      "rewards/margins": 15.690652847290039,
-      "rewards/real": 9.155214309692383,
+      "logits/generated": -2.826312303543091,
+      "logits/real": -2.8815627098083496,
+      "logps/generated": -154.94284057617188,
+      "logps/real": -90.3093032836914,
+      "loss": 0.2062,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -6.737037658691406,
+      "rewards/margins": 15.891546249389648,
+      "rewards/real": 9.154508590698242,
       "step": 620
     },
     {
       "epoch": 0.40307101727447214,
-      "grad_norm": 2.066415328730325,
+      "grad_norm": 2.029869456907254,
       "learning_rate": 3.3179231863442384e-07,
-      "logits/generated": -2.7363858222961426,
-      "logits/real": -2.9205002784729004,
-      "logps/generated": -139.44544982910156,
-      "logps/real": -93.95768737792969,
-      "loss": 0.2054,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/generated": -5.3508195877075195,
-      "rewards/margins": 14.589759826660156,
-      "rewards/real": 9.23893928527832,
+      "logits/generated": -2.7833971977233887,
+      "logits/real": -2.901395797729492,
+      "logps/generated": -139.93069458007812,
+      "logps/real": -93.89498138427734,
+      "loss": 0.2045,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.399343490600586,
+      "rewards/margins": 14.644552230834961,
+      "rewards/real": 9.245210647583008,
       "step": 630
     },
     {
       "epoch": 0.4094689699296225,
-      "grad_norm": 2.720280779674223,
+      "grad_norm": 2.746032157210293,
       "learning_rate": 3.282361308677098e-07,
-      "logits/generated": -2.717585325241089,
-      "logits/real": -2.8081445693969727,
-      "logps/generated": -139.2246551513672,
-      "logps/real": -95.35455322265625,
-      "loss": 0.2105,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.649844169616699,
-      "rewards/margins": 15.598958015441895,
-      "rewards/real": 9.949111938476562,
+      "logits/generated": -2.760668992996216,
+      "logits/real": -2.8092033863067627,
+      "logps/generated": -143.58538818359375,
+      "logps/real": -96.08203887939453,
+      "loss": 0.2123,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -6.085916996002197,
+      "rewards/margins": 15.962282180786133,
+      "rewards/real": 9.876363754272461,
       "step": 640
     },
     {
       "epoch": 0.41586692258477287,
-      "grad_norm": 2.4467415453795534,
+      "grad_norm": 2.5187366822563866,
       "learning_rate": 3.2467994310099573e-07,
-      "logits/generated": -2.746037244796753,
-      "logits/real": -2.7760324478149414,
-      "logps/generated": -151.04168701171875,
-      "logps/real": -117.99039459228516,
-      "loss": 0.228,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -6.320367336273193,
-      "rewards/margins": 17.100696563720703,
-      "rewards/real": 10.780328750610352,
+      "logits/generated": -2.7855262756347656,
+      "logits/real": -2.7854223251342773,
+      "logps/generated": -151.7242889404297,
+      "logps/real": -117.95533752441406,
+      "loss": 0.2275,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -6.388627052307129,
+      "rewards/margins": 17.172462463378906,
+      "rewards/real": 10.783835411071777,
       "step": 650
     },
     {
       "epoch": 0.42226487523992323,
-      "grad_norm": 2.5369736094086752,
+      "grad_norm": 2.793059409553602,
       "learning_rate": 3.211237553342817e-07,
-      "logits/generated": -2.7743470668792725,
-      "logits/real": -2.9016330242156982,
-      "logps/generated": -169.43215942382812,
-      "logps/real": -110.16139221191406,
-      "loss": 0.2213,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -7.090481758117676,
-      "rewards/margins": 17.460203170776367,
-      "rewards/real": 10.369722366333008,
+      "logits/generated": -2.807608127593994,
+      "logits/real": -2.866541862487793,
+      "logps/generated": -167.69371032714844,
+      "logps/real": -109.91790771484375,
+      "loss": 0.2221,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -6.916634559631348,
+      "rewards/margins": 17.31070327758789,
+      "rewards/real": 10.394070625305176,
       "step": 660
     },
     {
       "epoch": 0.4286628278950736,
-      "grad_norm": 3.940886446614688,
+      "grad_norm": 3.829620174016922,
       "learning_rate": 3.175675675675675e-07,
-      "logits/generated": -2.7578988075256348,
-      "logits/real": -2.7422776222229004,
-      "logps/generated": -142.6200408935547,
-      "logps/real": -109.90318298339844,
-      "loss": 0.2226,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.739054203033447,
-      "rewards/margins": 16.534833908081055,
-      "rewards/real": 10.79577922821045,
+      "logits/generated": -2.78635835647583,
+      "logits/real": -2.7660999298095703,
+      "logps/generated": -142.71031188964844,
+      "logps/real": -110.40629577636719,
+      "loss": 0.2219,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.748080253601074,
+      "rewards/margins": 16.493549346923828,
+      "rewards/real": 10.745469093322754,
       "step": 670
     },
     {
       "epoch": 0.4350607805502239,
-      "grad_norm": 1.9885549537383231,
+      "grad_norm": 2.131853509135701,
       "learning_rate": 3.1401137980085347e-07,
-      "logits/generated": -2.753906011581421,
-      "logits/real": -2.8121540546417236,
-      "logps/generated": -142.36412048339844,
-      "logps/real": -95.9502182006836,
-      "loss": 0.2135,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -6.0002851486206055,
-      "rewards/margins": 15.877609252929688,
-      "rewards/real": 9.877325057983398,
+      "logits/generated": -2.782644033432007,
+      "logits/real": -2.806847095489502,
+      "logps/generated": -140.91482543945312,
+      "logps/real": -95.88102722167969,
+      "loss": 0.2134,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.855355262756348,
+      "rewards/margins": 15.739598274230957,
+      "rewards/real": 9.884242057800293,
       "step": 680
     },
     {
       "epoch": 0.44145873320537427,
-      "grad_norm": 2.360303864176274,
+      "grad_norm": 2.3182724658645895,
       "learning_rate": 3.104551920341394e-07,
-      "logits/generated": -2.7917466163635254,
-      "logits/real": -2.8809592723846436,
-      "logps/generated": -128.33108520507812,
-      "logps/real": -109.88875579833984,
-      "loss": 0.2091,
+      "logits/generated": -2.805422067642212,
+      "logits/real": -2.8506126403808594,
+      "logps/generated": -130.77435302734375,
+      "logps/real": -109.96136474609375,
+      "loss": 0.2102,
       "rewards/accuracies": 1.0,
-      "rewards/generated": -5.061340808868408,
-      "rewards/margins": 15.903111457824707,
-      "rewards/real": 10.841771125793457,
+      "rewards/generated": -5.305667877197266,
+      "rewards/margins": 16.140178680419922,
+      "rewards/real": 10.834508895874023,
       "step": 690
     },
     {
       "epoch": 0.44785668586052463,
-      "grad_norm": 2.2232483398700884,
+      "grad_norm": 2.2151091580518405,
       "learning_rate": 3.068990042674253e-07,
-      "logits/generated": -2.7540788650512695,
-      "logits/real": -2.8440656661987305,
-      "logps/generated": -137.61160278320312,
-      "logps/real": -110.4582290649414,
-      "loss": 0.2235,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.629530906677246,
-      "rewards/margins": 16.784801483154297,
-      "rewards/real": 11.155269622802734,
+      "logits/generated": -2.7822375297546387,
+      "logits/real": -2.835947036743164,
+      "logps/generated": -137.30563354492188,
+      "logps/real": -110.43853759765625,
+      "loss": 0.2244,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.598933219909668,
+      "rewards/margins": 16.75617218017578,
+      "rewards/real": 11.15723705291748,
+      "step": 700
+    },
+    {
+      "epoch": 0.44785668586052463,
+      "eval_logits/generated": -2.8203015327453613,
+      "eval_logits/real": -2.8076000213623047,
+      "eval_logps/generated": -130.0162353515625,
+      "eval_logps/real": -112.26402282714844,
+      "eval_loss": 0.23173104226589203,
+      "eval_rewards/accuracies": 1.0,
+      "eval_rewards/generated": -5.129922866821289,
+      "eval_rewards/margins": 15.171630859375,
+      "eval_rewards/real": 10.041707992553711,
+      "eval_runtime": 66.1369,
+      "eval_samples_per_second": 7.56,
+      "eval_steps_per_second": 0.242,
       "step": 700
     },
     {
       "epoch": 0.454254638515675,
-      "grad_norm": 2.420484904502962,
+      "grad_norm": 3.131717649421431,
       "learning_rate": 3.033428165007112e-07,
-      "logits/generated": -2.7791481018066406,
-      "logits/real": -2.823207378387451,
-      "logps/generated": -130.23995971679688,
-      "logps/real": -104.56422424316406,
-      "loss": 0.2045,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.046719551086426,
-      "rewards/margins": 15.610448837280273,
-      "rewards/real": 10.563729286193848,
+      "logits/generated": -2.8134236335754395,
+      "logits/real": -2.829641819000244,
+      "logps/generated": -133.2042236328125,
+      "logps/real": -105.28382873535156,
+      "loss": 0.2037,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.343146800994873,
+      "rewards/margins": 15.834917068481445,
+      "rewards/real": 10.491768836975098,
       "step": 710
     },
     {
       "epoch": 0.46065259117082535,
-      "grad_norm": 1.8814105476658205,
+      "grad_norm": 2.039165566664337,
       "learning_rate": 2.9978662873399715e-07,
-      "logits/generated": -2.754664897918701,
-      "logits/real": -2.868396043777466,
-      "logps/generated": -138.9653778076172,
-      "logps/real": -104.86256408691406,
+      "logits/generated": -2.7993435859680176,
+      "logits/real": -2.8707499504089355,
+      "logps/generated": -142.11581420898438,
+      "logps/real": -105.18668365478516,
       "loss": 0.2139,
       "rewards/accuracies": 1.0,
-      "rewards/generated": -5.563291549682617,
-      "rewards/margins": 16.40620994567871,
-      "rewards/real": 10.842918395996094,
+      "rewards/generated": -5.878334045410156,
+      "rewards/margins": 16.688838958740234,
+      "rewards/real": 10.810506820678711,
       "step": 720
     },
     {
       "epoch": 0.46705054382597566,
-      "grad_norm": 1.9123445490996223,
+      "grad_norm": 2.108008595873001,
       "learning_rate": 2.9623044096728305e-07,
-      "logits/generated": -2.7599892616271973,
-      "logits/real": -2.836871862411499,
-      "logps/generated": -143.76820373535156,
-      "logps/real": -101.16252136230469,
-      "loss": 0.2134,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.618950843811035,
-      "rewards/margins": 16.395917892456055,
-      "rewards/real": 10.776965141296387,
+      "logits/generated": -2.806464672088623,
+      "logits/real": -2.8528928756713867,
+      "logps/generated": -146.18896484375,
+      "logps/real": -101.5972900390625,
+      "loss": 0.213,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.861027717590332,
+      "rewards/margins": 16.594514846801758,
+      "rewards/real": 10.733488082885742,
       "step": 730
     },
     {
       "epoch": 0.473448496481126,
-      "grad_norm": 1.9685905302445135,
+      "grad_norm": 1.975375872847753,
       "learning_rate": 2.92674253200569e-07,
-      "logits/generated": -2.7747621536254883,
-      "logits/real": -2.7902274131774902,
-      "logps/generated": -118.72608947753906,
-      "logps/real": -107.44352722167969,
-      "loss": 0.2035,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -4.259701728820801,
-      "rewards/margins": 15.15577220916748,
-      "rewards/real": 10.89607048034668,
+      "logits/generated": -2.8044772148132324,
+      "logits/real": -2.8095054626464844,
+      "logps/generated": -118.6663818359375,
+      "logps/real": -107.431884765625,
+      "loss": 0.2036,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -4.2537312507629395,
+      "rewards/margins": 15.150965690612793,
+      "rewards/real": 10.897233963012695,
       "step": 740
     },
     {
       "epoch": 0.4798464491362764,
-      "grad_norm": 2.185389542016207,
+      "grad_norm": 2.1076532976513525,
       "learning_rate": 2.8911806543385494e-07,
-      "logits/generated": -2.694744348526001,
-      "logits/real": -2.7872262001037598,
-      "logps/generated": -117.7671890258789,
-      "logps/real": -92.30455017089844,
-      "loss": 0.2175,
+      "logits/generated": -2.7521071434020996,
+      "logits/real": -2.8067069053649902,
+      "logps/generated": -120.24348449707031,
+      "logps/real": -92.26268005371094,
+      "loss": 0.2203,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/generated": -4.346129894256592,
-      "rewards/margins": 13.698951721191406,
-      "rewards/real": 9.352822303771973,
+      "rewards/generated": -4.593760013580322,
+      "rewards/margins": 13.950767517089844,
+      "rewards/real": 9.357008934020996,
       "step": 750
     },
     {
       "epoch": 0.48624440179142675,
-      "grad_norm": 2.0382414250965732,
+      "grad_norm": 1.99233545437077,
       "learning_rate": 2.855618776671408e-07,
-      "logits/generated": -2.7974352836608887,
-      "logits/real": -2.8290600776672363,
-      "logps/generated": -132.22039794921875,
-      "logps/real": -104.40059661865234,
-      "loss": 0.2157,
+      "logits/generated": -2.8306665420532227,
+      "logits/real": -2.832305431365967,
+      "logps/generated": -133.2040557861328,
+      "logps/real": -105.5262222290039,
+      "loss": 0.2153,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/generated": -5.511082649230957,
-      "rewards/margins": 15.783149719238281,
-      "rewards/real": 10.272066116333008,
+      "rewards/generated": -5.609448432922363,
+      "rewards/margins": 15.768956184387207,
+      "rewards/real": 10.159505844116211,
       "step": 760
     },
     {
       "epoch": 0.4926423544465771,
-      "grad_norm": 2.1388688577679016,
+      "grad_norm": 5.594213807868116,
       "learning_rate": 2.8200568990042673e-07,
-      "logits/generated": -2.7909514904022217,
-      "logits/real": -2.808551073074341,
-      "logps/generated": -151.1422119140625,
-      "logps/real": -116.09953308105469,
-      "loss": 0.2271,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -6.148834228515625,
-      "rewards/margins": 17.4340877532959,
-      "rewards/real": 11.28525447845459,
+      "logits/generated": -2.832336902618408,
+      "logits/real": -2.828530788421631,
+      "logps/generated": -152.615478515625,
+      "logps/real": -116.57222747802734,
+      "loss": 0.2255,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -6.296160697937012,
+      "rewards/margins": 17.534143447875977,
+      "rewards/real": 11.237985610961914,
       "step": 770
     },
     {
       "epoch": 0.4990403071017274,
-      "grad_norm": 2.2656673226538717,
+      "grad_norm": 2.234187106468982,
       "learning_rate": 2.784495021337127e-07,
-      "logits/generated": -2.7903616428375244,
-      "logits/real": -2.8122215270996094,
-      "logps/generated": -138.20358276367188,
-      "logps/real": -104.3212661743164,
-      "loss": 0.2234,
+      "logits/generated": -2.811738967895508,
+      "logits/real": -2.813356876373291,
+      "logps/generated": -138.08277893066406,
+      "logps/real": -104.5354232788086,
+      "loss": 0.2246,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/generated": -5.353945732116699,
-      "rewards/margins": 16.125553131103516,
-      "rewards/real": 10.771608352661133,
+      "rewards/generated": -5.341864585876465,
+      "rewards/margins": 16.092058181762695,
+      "rewards/real": 10.750192642211914,
       "step": 780
     },
     {
       "epoch": 0.5054382597568778,
-      "grad_norm": 2.1885026864804624,
+      "grad_norm": 2.1986792668849917,
       "learning_rate": 2.7489331436699857e-07,
-      "logits/generated": -2.8039913177490234,
-      "logits/real": -2.7904744148254395,
-      "logps/generated": -129.17857360839844,
-      "logps/real": -115.23606872558594,
-      "loss": 0.2302,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.161337852478027,
-      "rewards/margins": 16.69219398498535,
-      "rewards/real": 11.530855178833008,
+      "logits/generated": -2.819045305252075,
+      "logits/real": -2.796511173248291,
+      "logps/generated": -129.47288513183594,
+      "logps/real": -115.36392974853516,
+      "loss": 0.2306,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.190769195556641,
+      "rewards/margins": 16.708837509155273,
+      "rewards/real": 11.518068313598633,
       "step": 790
     },
     {
       "epoch": 0.5118362124120281,
-      "grad_norm": 2.869890153550306,
+      "grad_norm": 18.006741390903635,
       "learning_rate": 2.7133712660028446e-07,
-      "logits/generated": -2.773207426071167,
-      "logits/real": -2.7846627235412598,
-      "logps/generated": -119.42828369140625,
-      "logps/real": -115.4027099609375,
-      "loss": 0.2148,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -4.340595722198486,
-      "rewards/margins": 15.677358627319336,
-      "rewards/real": 11.336763381958008,
+      "logits/generated": -2.7980234622955322,
+      "logits/real": -2.7978062629699707,
+      "logps/generated": -119.58663177490234,
+      "logps/real": -115.9960708618164,
+      "loss": 0.2161,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/generated": -4.35643196105957,
+      "rewards/margins": 15.633859634399414,
+      "rewards/real": 11.277426719665527,
       "step": 800
     },
     {
       "epoch": 0.5118362124120281,
-      "eval_logits/generated": -2.8194947242736816,
-      "eval_logits/real": -2.823777437210083,
-      "eval_logps/generated": -129.62098693847656,
-      "eval_logps/real": -112.27095031738281,
-      "eval_loss": 0.23086096346378326,
-      "eval_rewards/accuracies": 0.9921875,
-      "eval_rewards/generated": -5.09039831161499,
-      "eval_rewards/margins": 15.131412506103516,
-      "eval_rewards/real": 10.041014671325684,
-      "eval_runtime": 69.2574,
-      "eval_samples_per_second": 7.219,
-      "eval_steps_per_second": 0.231,
+      "eval_logits/generated": -2.843674659729004,
+      "eval_logits/real": -2.833674430847168,
+      "eval_logps/generated": -129.2824249267578,
+      "eval_logps/real": -111.94398498535156,
+      "eval_loss": 0.22970730066299438,
+      "eval_rewards/accuracies": 1.0,
+      "eval_rewards/generated": -5.056541919708252,
+      "eval_rewards/margins": 15.130252838134766,
+      "eval_rewards/real": 10.073712348937988,
+      "eval_runtime": 70.0028,
+      "eval_samples_per_second": 7.143,
+      "eval_steps_per_second": 0.229,
       "step": 800
     },
     {
       "epoch": 0.5182341650671785,
-      "grad_norm": 2.356636174793149,
+      "grad_norm": 2.4618069592133605,
       "learning_rate": 2.677809388335704e-07,
-      "logits/generated": -2.8063623905181885,
-      "logits/real": -2.863706111907959,
-      "logps/generated": -147.79519653320312,
-      "logps/real": -103.06642150878906,
-      "loss": 0.2038,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.835212707519531,
-      "rewards/margins": 16.04536247253418,
-      "rewards/real": 10.210151672363281,
+      "logits/generated": -2.8439269065856934,
+      "logits/real": -2.866291046142578,
+      "logps/generated": -151.0102081298828,
+      "logps/real": -103.1385269165039,
+      "loss": 0.2025,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -6.156715393066406,
+      "rewards/margins": 16.35965919494629,
+      "rewards/real": 10.20294189453125,
       "step": 810
     },
     {
       "epoch": 0.5246321177223289,
-      "grad_norm": 2.4379730238495885,
+      "grad_norm": 4.833177875201332,
       "learning_rate": 2.642247510668563e-07,
-      "logits/generated": -2.8116323947906494,
-      "logits/real": -2.8085341453552246,
-      "logps/generated": -135.0247039794922,
-      "logps/real": -88.69074249267578,
-      "loss": 0.2247,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.124009132385254,
-      "rewards/margins": 14.631864547729492,
-      "rewards/real": 9.507856369018555,
+      "logits/generated": -2.843188762664795,
+      "logits/real": -2.822864532470703,
+      "logps/generated": -136.66146850585938,
+      "logps/real": -89.46752166748047,
+      "loss": 0.2261,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.287683010101318,
+      "rewards/margins": 14.717862129211426,
+      "rewards/real": 9.430177688598633,
       "step": 820
     },
     {
       "epoch": 0.5310300703774792,
-      "grad_norm": 2.4726280945236083,
+      "grad_norm": 2.469263347083119,
       "learning_rate": 2.6066856330014225e-07,
-      "logits/generated": -2.7714428901672363,
-      "logits/real": -2.796095132827759,
-      "logps/generated": -138.78175354003906,
-      "logps/real": -103.1611099243164,
-      "loss": 0.2042,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.705328941345215,
-      "rewards/margins": 16.654027938842773,
-      "rewards/real": 10.948700904846191,
+      "logits/generated": -2.810912609100342,
+      "logits/real": -2.8138461112976074,
+      "logps/generated": -139.83778381347656,
+      "logps/real": -103.55058288574219,
+      "loss": 0.2036,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.8109307289123535,
+      "rewards/margins": 16.720684051513672,
+      "rewards/real": 10.909751892089844,
       "step": 830
     },
     {
       "epoch": 0.5374280230326296,
-      "grad_norm": 2.0930312705666965,
+      "grad_norm": 2.086295169247383,
       "learning_rate": 2.5711237553342815e-07,
-      "logits/generated": -2.796297311782837,
-      "logits/real": -2.761157512664795,
-      "logps/generated": -132.76963806152344,
-      "logps/real": -101.53390502929688,
-      "loss": 0.2154,
+      "logits/generated": -2.8145735263824463,
+      "logits/real": -2.7811641693115234,
+      "logps/generated": -134.34349060058594,
+      "logps/real": -102.11433410644531,
+      "loss": 0.2159,
       "rewards/accuracies": 1.0,
-      "rewards/generated": -5.048023223876953,
-      "rewards/margins": 16.279930114746094,
-      "rewards/real": 11.231904983520508,
+      "rewards/generated": -5.205408096313477,
+      "rewards/margins": 16.3792724609375,
+      "rewards/real": 11.173863410949707,
       "step": 840
     },
     {
       "epoch": 0.5438259756877799,
-      "grad_norm": 1.995146822930404,
+      "grad_norm": 1.9811986098946064,
       "learning_rate": 2.5355618776671404e-07,
-      "logits/generated": -2.74373722076416,
-      "logits/real": -2.8401942253112793,
-      "logps/generated": -134.88999938964844,
-      "logps/real": -108.44051361083984,
+      "logits/generated": -2.776414394378662,
+      "logits/real": -2.8270602226257324,
+      "logps/generated": -137.35968017578125,
+      "logps/real": -108.5652847290039,
       "loss": 0.2085,
       "rewards/accuracies": 1.0,
-      "rewards/generated": -5.090542793273926,
-      "rewards/margins": 16.1528377532959,
-      "rewards/real": 11.06229305267334,
+      "rewards/generated": -5.337510585784912,
+      "rewards/margins": 16.387327194213867,
+      "rewards/real": 11.04981803894043,
       "step": 850
     },
     {
       "epoch": 0.5502239283429302,
-      "grad_norm": 2.383171664852739,
+      "grad_norm": 2.3515626470248385,
       "learning_rate": 2.5e-07,
-      "logits/generated": -2.734670639038086,
-      "logits/real": -2.836339235305786,
-      "logps/generated": -138.05905151367188,
-      "logps/real": -96.95460510253906,
-      "loss": 0.2,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.488851070404053,
-      "rewards/margins": 15.819448471069336,
-      "rewards/real": 10.330598831176758,
+      "logits/generated": -2.775594472885132,
+      "logits/real": -2.8108696937561035,
+      "logps/generated": -138.37057495117188,
+      "logps/real": -96.95921325683594,
+      "loss": 0.2011,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.520004749298096,
+      "rewards/margins": 15.850143432617188,
+      "rewards/real": 10.330138206481934,
       "step": 860
     },
     {
       "epoch": 0.5566218809980806,
-      "grad_norm": 2.20520607110751,
+      "grad_norm": 8.080615991325786,
       "learning_rate": 2.4644381223328594e-07,
-      "logits/generated": -2.7709481716156006,
-      "logits/real": -2.7831263542175293,
-      "logps/generated": -137.1595916748047,
-      "logps/real": -114.11800384521484,
-      "loss": 0.2189,
+      "logits/generated": -2.811049222946167,
+      "logits/real": -2.7949414253234863,
+      "logps/generated": -140.1456756591797,
+      "logps/real": -114.25732421875,
+      "loss": 0.2188,
       "rewards/accuracies": 1.0,
-      "rewards/generated": -5.659204959869385,
-      "rewards/margins": 17.742206573486328,
-      "rewards/real": 12.083002090454102,
+      "rewards/generated": -5.957813262939453,
+      "rewards/margins": 18.026884078979492,
+      "rewards/real": 12.069070816040039,
       "step": 870
     },
     {
       "epoch": 0.5630198336532309,
-      "grad_norm": 2.6353918937836847,
+      "grad_norm": 3.352802751671241,
       "learning_rate": 2.4288762446657183e-07,
-      "logits/generated": -2.708103656768799,
-      "logits/real": -2.895563840866089,
-      "logps/generated": -141.7598876953125,
-      "logps/real": -99.23255920410156,
-      "loss": 0.2069,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.729107856750488,
-      "rewards/margins": 15.888574600219727,
-      "rewards/real": 10.159466743469238,
+      "logits/generated": -2.769892930984497,
+      "logits/real": -2.8765125274658203,
+      "logps/generated": -143.98390197753906,
+      "logps/real": -99.53087615966797,
+      "loss": 0.2068,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.951510906219482,
+      "rewards/margins": 16.081144332885742,
+      "rewards/real": 10.129633903503418,
       "step": 880
     },
     {
       "epoch": 0.5694177863083814,
-      "grad_norm": 2.159987018302931,
+      "grad_norm": 2.0932681040703334,
       "learning_rate": 2.393314366998578e-07,
-      "logits/generated": -2.8029890060424805,
-      "logits/real": -2.899193048477173,
-      "logps/generated": -142.26734924316406,
-      "logps/real": -89.63228607177734,
-      "loss": 0.2086,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.678102493286133,
-      "rewards/margins": 15.533955574035645,
-      "rewards/real": 9.855853080749512,
+      "logits/generated": -2.8301260471343994,
+      "logits/real": -2.87371826171875,
+      "logps/generated": -140.91387939453125,
+      "logps/real": -90.23018646240234,
+      "loss": 0.2098,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.542755603790283,
+      "rewards/margins": 15.338818550109863,
+      "rewards/real": 9.796061515808105,
       "step": 890
     },
     {
       "epoch": 0.5758157389635317,
-      "grad_norm": 2.829485679186667,
+      "grad_norm": 2.70893172288666,
       "learning_rate": 2.3577524893314365e-07,
-      "logits/generated": -2.8933000564575195,
-      "logits/real": -2.8898868560791016,
-      "logps/generated": -140.3587646484375,
-      "logps/real": -119.44731140136719,
-      "loss": 0.2126,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.4967241287231445,
-      "rewards/margins": 17.632753372192383,
-      "rewards/real": 12.136028289794922,
+      "logits/generated": -2.893622398376465,
+      "logits/real": -2.871110439300537,
+      "logps/generated": -140.7932586669922,
+      "logps/real": -119.5684814453125,
+      "loss": 0.2127,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.540173530578613,
+      "rewards/margins": 17.664085388183594,
+      "rewards/real": 12.123909950256348,
+      "step": 900
+    },
+    {
+      "epoch": 0.5758157389635317,
+      "eval_logits/generated": -2.8250679969787598,
+      "eval_logits/real": -2.815018653869629,
+      "eval_logps/generated": -129.62173461914062,
+      "eval_logps/real": -111.76828002929688,
+      "eval_loss": 0.2302427589893341,
+      "eval_rewards/accuracies": 1.0,
+      "eval_rewards/generated": -5.090472221374512,
+      "eval_rewards/margins": 15.181754112243652,
+      "eval_rewards/real": 10.091282844543457,
+      "eval_runtime": 46.5119,
+      "eval_samples_per_second": 10.75,
+      "eval_steps_per_second": 0.344,
       "step": 900
     },
     {
       "epoch": 0.582213691618682,
-      "grad_norm": 2.076889704737332,
+      "grad_norm": 2.049164473274404,
       "learning_rate": 2.322190611664296e-07,
-      "logits/generated": -2.783379077911377,
-      "logits/real": -2.837111473083496,
-      "logps/generated": -159.20602416992188,
-      "logps/real": -101.35649108886719,
-      "loss": 0.2152,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -6.505612850189209,
-      "rewards/margins": 16.421142578125,
-      "rewards/real": 9.91553020477295,
+      "logits/generated": -2.8185391426086426,
+      "logits/real": -2.8172860145568848,
+      "logps/generated": -161.87673950195312,
+      "logps/real": -101.66253662109375,
+      "loss": 0.2155,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -6.772686004638672,
+      "rewards/margins": 16.657611846923828,
+      "rewards/real": 9.884923934936523,
       "step": 910
     },
     {
       "epoch": 0.5886116442738324,
-      "grad_norm": 2.9561242755966917,
+      "grad_norm": 3.2668564053336144,
       "learning_rate": 2.2866287339971549e-07,
-      "logits/generated": -2.7626872062683105,
-      "logits/real": -2.775630235671997,
-      "logps/generated": -142.0345916748047,
-      "logps/real": -105.5779037475586,
-      "loss": 0.2106,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.930171012878418,
-      "rewards/margins": 16.78856086730957,
-      "rewards/real": 10.858389854431152,
+      "logits/generated": -2.7991254329681396,
+      "logits/real": -2.792893886566162,
+      "logps/generated": -141.7483673095703,
+      "logps/real": -105.80220031738281,
+      "loss": 0.21,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.901549339294434,
+      "rewards/margins": 16.73750877380371,
+      "rewards/real": 10.83596134185791,
       "step": 920
     },
     {
       "epoch": 0.5950095969289827,
-      "grad_norm": 5.343288980733804,
+      "grad_norm": 6.9616976220540305,
       "learning_rate": 2.251066856330014e-07,
-      "logits/generated": -2.863882064819336,
-      "logits/real": -2.7736942768096924,
-      "logps/generated": -146.3633575439453,
-      "logps/real": -116.1565170288086,
-      "loss": 0.2372,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -6.034173965454102,
-      "rewards/margins": 17.2874755859375,
-      "rewards/real": 11.253301620483398,
+      "logits/generated": -2.8697328567504883,
+      "logits/real": -2.7932534217834473,
+      "logps/generated": -146.28128051757812,
+      "logps/real": -116.77775573730469,
+      "loss": 0.2377,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -6.02596378326416,
+      "rewards/margins": 17.217140197753906,
+      "rewards/real": 11.191176414489746,
       "step": 930
     },
     {
       "epoch": 0.6014075495841331,
-      "grad_norm": 2.1700141539005418,
+      "grad_norm": 2.9973051859959794,
       "learning_rate": 2.2155049786628733e-07,
-      "logits/generated": -2.7782211303710938,
-      "logits/real": -2.784595012664795,
-      "logps/generated": -126.44002532958984,
-      "logps/real": -98.55274200439453,
-      "loss": 0.2121,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -4.89090633392334,
-      "rewards/margins": 15.318072319030762,
-      "rewards/real": 10.427166938781738,
+      "logits/generated": -2.8155462741851807,
+      "logits/real": -2.782264232635498,
+      "logps/generated": -131.8110809326172,
+      "logps/real": -98.78057098388672,
+      "loss": 0.212,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.428011894226074,
+      "rewards/margins": 15.83239459991455,
+      "rewards/real": 10.404382705688477,
       "step": 940
     },
     {
       "epoch": 0.6078055022392834,
-      "grad_norm": 2.086474199547169,
+      "grad_norm": 2.1077180741024435,
       "learning_rate": 2.1799431009957325e-07,
-      "logits/generated": -2.7844066619873047,
-      "logits/real": -2.87266206741333,
-      "logps/generated": -154.29930114746094,
-      "logps/real": -118.48374938964844,
-      "loss": 0.2242,
+      "logits/generated": -2.824289083480835,
+      "logits/real": -2.8804006576538086,
+      "logps/generated": -157.98342895507812,
+      "logps/real": -118.81966400146484,
+      "loss": 0.2247,
       "rewards/accuracies": 1.0,
-      "rewards/generated": -5.924124717712402,
-      "rewards/margins": 17.47787857055664,
-      "rewards/real": 11.553754806518555,
+      "rewards/generated": -6.292536735534668,
+      "rewards/margins": 17.812700271606445,
+      "rewards/real": 11.520162582397461,
       "step": 950
     },
     {
       "epoch": 0.6142034548944337,
-      "grad_norm": 5.893270475169604,
+      "grad_norm": 8.878441364444791,
       "learning_rate": 2.1443812233285914e-07,
-      "logits/generated": -2.7291111946105957,
-      "logits/real": -2.7862040996551514,
-      "logps/generated": -128.32461547851562,
-      "logps/real": -88.90164947509766,
-      "loss": 0.2049,
+      "logits/generated": -2.77471923828125,
+      "logits/real": -2.7926316261291504,
+      "logps/generated": -131.18397521972656,
+      "logps/real": -89.53666687011719,
+      "loss": 0.2104,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/generated": -5.07694149017334,
-      "rewards/margins": 14.397714614868164,
-      "rewards/real": 9.320773124694824,
+      "rewards/generated": -5.362875938415527,
+      "rewards/margins": 14.620147705078125,
+      "rewards/real": 9.257271766662598,
       "step": 960
     },
     {
       "epoch": 0.6206014075495841,
-      "grad_norm": 3.4745504874973854,
+      "grad_norm": 8.82863549180699,
       "learning_rate": 2.108819345661451e-07,
-      "logits/generated": -2.7670676708221436,
-      "logits/real": -2.906858205795288,
-      "logps/generated": -142.1813507080078,
-      "logps/real": -99.36964416503906,
-      "loss": 0.2152,
+      "logits/generated": -2.8006834983825684,
+      "logits/real": -2.877106189727783,
+      "logps/generated": -143.05490112304688,
+      "logps/real": -99.6307144165039,
+      "loss": 0.2142,
       "rewards/accuracies": 1.0,
-      "rewards/generated": -5.543498992919922,
-      "rewards/margins": 15.908276557922363,
-      "rewards/real": 10.364778518676758,
+      "rewards/generated": -5.630855560302734,
+      "rewards/margins": 15.969528198242188,
+      "rewards/real": 10.33867073059082,
       "step": 970
     },
     {
       "epoch": 0.6269993602047345,
-      "grad_norm": 1.9536154527758394,
+      "grad_norm": 1.926193237909525,
       "learning_rate": 2.0732574679943098e-07,
-      "logits/generated": -2.7182888984680176,
-      "logits/real": -2.795628070831299,
-      "logps/generated": -123.13966369628906,
-      "logps/real": -92.99507141113281,
-      "loss": 0.2048,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -4.866164207458496,
-      "rewards/margins": 14.955586433410645,
-      "rewards/real": 10.089422225952148,
+      "logits/generated": -2.7720258235931396,
+      "logits/real": -2.8031325340270996,
+      "logps/generated": -125.90155029296875,
+      "logps/real": -93.4397201538086,
+      "loss": 0.2034,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.14235258102417,
+      "rewards/margins": 15.187309265136719,
+      "rewards/real": 10.044957160949707,
       "step": 980
     },
     {
       "epoch": 0.6333973128598849,
-      "grad_norm": 2.3702688907183425,
+      "grad_norm": 2.3719036195798657,
       "learning_rate": 2.0376955903271693e-07,
-      "logits/generated": -2.792490243911743,
-      "logits/real": -2.814042568206787,
-      "logps/generated": -122.5984878540039,
-      "logps/real": -119.72834777832031,
-      "loss": 0.2154,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -4.933380603790283,
-      "rewards/margins": 16.830997467041016,
-      "rewards/real": 11.897615432739258,
+      "logits/generated": -2.826439380645752,
+      "logits/real": -2.830770492553711,
+      "logps/generated": -124.8687973022461,
+      "logps/real": -120.2898178100586,
+      "loss": 0.2166,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.160411357879639,
+      "rewards/margins": 17.001880645751953,
+      "rewards/real": 11.841466903686523,
       "step": 990
     },
     {
       "epoch": 0.6397952655150352,
-      "grad_norm": 2.710438187714284,
+      "grad_norm": 2.7028662367061553,
       "learning_rate": 2.0021337126600283e-07,
-      "logits/generated": -2.7410390377044678,
-      "logits/real": -2.817781925201416,
-      "logps/generated": -161.62539672851562,
-      "logps/real": -91.64434051513672,
-      "loss": 0.1994,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -6.895718574523926,
-      "rewards/margins": 17.067646026611328,
-      "rewards/real": 10.171927452087402,
+      "logits/generated": -2.785853862762451,
+      "logits/real": -2.812053680419922,
+      "logps/generated": -159.1813507080078,
+      "logps/real": -92.27577209472656,
+      "loss": 0.2017,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -6.651316165924072,
+      "rewards/margins": 16.760101318359375,
+      "rewards/real": 10.108785629272461,
       "step": 1000
     },
     {
       "epoch": 0.6397952655150352,
-      "eval_logits/generated": -2.74424147605896,
-      "eval_logits/real": -2.7461187839508057,
-      "eval_logps/generated": -130.59327697753906,
-      "eval_logps/real": -111.54972839355469,
-      "eval_loss": 0.23031392693519592,
-      "eval_rewards/accuracies": 0.9921875,
-      "eval_rewards/generated": -5.187626838684082,
-      "eval_rewards/margins": 15.300764083862305,
-      "eval_rewards/real": 10.113137245178223,
-      "eval_runtime": 69.5849,
-      "eval_samples_per_second": 7.185,
-      "eval_steps_per_second": 0.23,
+      "eval_logits/generated": -2.795530319213867,
+      "eval_logits/real": -2.7831103801727295,
+      "eval_logps/generated": -131.34414672851562,
+      "eval_logps/real": -111.43624877929688,
+      "eval_loss": 0.22981801629066467,
+      "eval_rewards/accuracies": 1.0,
+      "eval_rewards/generated": -5.262712478637695,
+      "eval_rewards/margins": 15.387198448181152,
+      "eval_rewards/real": 10.124486923217773,
+      "eval_runtime": 45.5245,
+      "eval_samples_per_second": 10.983,
+      "eval_steps_per_second": 0.351,
       "step": 1000
     },
     {
       "epoch": 0.6461932181701855,
-      "grad_norm": 2.2354079100751845,
+      "grad_norm": 2.1272152784462115,
       "learning_rate": 1.9665718349928875e-07,
-      "logits/generated": -2.7386083602905273,
-      "logits/real": -2.7921862602233887,
-      "logps/generated": -145.14804077148438,
-      "logps/real": -100.52366638183594,
-      "loss": 0.1945,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -6.059700012207031,
-      "rewards/margins": 16.424558639526367,
-      "rewards/real": 10.364858627319336,
+      "logits/generated": -2.78995418548584,
+      "logits/real": -2.792102098464966,
+      "logps/generated": -147.998291015625,
+      "logps/real": -101.14006805419922,
+      "loss": 0.1966,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -6.344725131988525,
+      "rewards/margins": 16.647945404052734,
+      "rewards/real": 10.303218841552734,
       "step": 1010
     },
     {
       "epoch": 0.6525911708253359,
-      "grad_norm": 2.1844813037214124,
+      "grad_norm": 2.161034970911341,
       "learning_rate": 1.931009957325747e-07,
-      "logits/generated": -2.7453203201293945,
-      "logits/real": -2.812004327774048,
-      "logps/generated": -147.8553466796875,
-      "logps/real": -100.11280822753906,
-      "loss": 0.2096,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.994841575622559,
-      "rewards/margins": 15.893636703491211,
-      "rewards/real": 9.898796081542969,
+      "logits/generated": -2.8014883995056152,
+      "logits/real": -2.820871591567993,
+      "logps/generated": -150.88238525390625,
+      "logps/real": -100.12489318847656,
+      "loss": 0.2099,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -6.297544479370117,
+      "rewards/margins": 16.195133209228516,
+      "rewards/real": 9.897587776184082,
       "step": 1020
     },
     {
       "epoch": 0.6589891234804862,
-      "grad_norm": 2.046398564222724,
+      "grad_norm": 1.9857012011776554,
       "learning_rate": 1.895448079658606e-07,
-      "logits/generated": -2.7719972133636475,
-      "logits/real": -2.837329387664795,
-      "logps/generated": -141.85464477539062,
-      "logps/real": -95.72025299072266,
-      "loss": 0.1927,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.702435493469238,
-      "rewards/margins": 15.670855522155762,
-      "rewards/real": 9.968420028686523,
+      "logits/generated": -2.8102505207061768,
+      "logits/real": -2.831676483154297,
+      "logps/generated": -144.0465545654297,
+      "logps/real": -96.27753448486328,
+      "loss": 0.1915,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.921627044677734,
+      "rewards/margins": 15.834318161010742,
+      "rewards/real": 9.912692070007324,
       "step": 1030
     },
     {
       "epoch": 0.6653870761356366,
-      "grad_norm": 2.2879001250497297,
+      "grad_norm": 2.666915560470661,
       "learning_rate": 1.859886201991465e-07,
-      "logits/generated": -2.708482265472412,
-      "logits/real": -2.903043270111084,
-      "logps/generated": -139.8241729736328,
-      "logps/real": -103.22698974609375,
-      "loss": 0.2013,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.350104331970215,
-      "rewards/margins": 16.140804290771484,
-      "rewards/real": 10.790700912475586,
+      "logits/generated": -2.775191307067871,
+      "logits/real": -2.8828163146972656,
+      "logps/generated": -141.82388305664062,
+      "logps/real": -103.5179214477539,
+      "loss": 0.2014,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.550075531005859,
+      "rewards/margins": 16.311683654785156,
+      "rewards/real": 10.76160717010498,
       "step": 1040
     },
     {
       "epoch": 0.6717850287907869,
-      "grad_norm": 4.628698290056213,
+      "grad_norm": 4.176748037838257,
       "learning_rate": 1.8243243243243243e-07,
-      "logits/generated": -2.748556137084961,
-      "logits/real": -2.8339285850524902,
-      "logps/generated": -148.6436004638672,
-      "logps/real": -114.01179504394531,
-      "loss": 0.2241,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.570124626159668,
-      "rewards/margins": 17.3154354095459,
-      "rewards/real": 11.745311737060547,
+      "logits/generated": -2.793463706970215,
+      "logits/real": -2.8246817588806152,
+      "logps/generated": -152.6351776123047,
+      "logps/real": -114.48309326171875,
+      "loss": 0.2242,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.969282627105713,
+      "rewards/margins": 17.667463302612305,
+      "rewards/real": 11.69818115234375,
       "step": 1050
     },
     {
       "epoch": 0.6781829814459372,
-      "grad_norm": 3.6247485539345417,
+      "grad_norm": 1.9721004025025954,
       "learning_rate": 1.7887624466571835e-07,
-      "logits/generated": -2.8411951065063477,
-      "logits/real": -2.7839694023132324,
-      "logps/generated": -123.47579193115234,
-      "logps/real": -112.1775131225586,
-      "loss": 0.2218,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -4.763791561126709,
-      "rewards/margins": 16.458126068115234,
-      "rewards/real": 11.694334030151367,
+      "logits/generated": -2.8584046363830566,
+      "logits/real": -2.7998831272125244,
+      "logps/generated": -128.0937042236328,
+      "logps/real": -112.5428466796875,
+      "loss": 0.2208,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.225579261779785,
+      "rewards/margins": 16.883380889892578,
+      "rewards/real": 11.657800674438477,
       "step": 1060
     },
     {
       "epoch": 0.6845809341010877,
-      "grad_norm": 2.2522203971012487,
+      "grad_norm": 2.6204362567382846,
       "learning_rate": 1.7532005689900424e-07,
-      "logits/generated": -2.7332136631011963,
-      "logits/real": -2.804201602935791,
-      "logps/generated": -142.97860717773438,
-      "logps/real": -85.4603271484375,
-      "loss": 0.1947,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.868603706359863,
-      "rewards/margins": 15.239709854125977,
-      "rewards/real": 9.37110424041748,
+      "logits/generated": -2.7668404579162598,
+      "logits/real": -2.789151668548584,
+      "logps/generated": -146.34744262695312,
+      "logps/real": -85.44964599609375,
+      "loss": 0.1944,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -6.205488681793213,
+      "rewards/margins": 15.577661514282227,
+      "rewards/real": 9.372171401977539,
       "step": 1070
     },
     {
       "epoch": 0.690978886756238,
-      "grad_norm": 2.2382250321484225,
+      "grad_norm": 2.2693636503855203,
       "learning_rate": 1.717638691322902e-07,
-      "logits/generated": -2.7342700958251953,
-      "logits/real": -2.8574981689453125,
-      "logps/generated": -151.73580932617188,
-      "logps/real": -103.33231353759766,
-      "loss": 0.2102,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -6.019370079040527,
-      "rewards/margins": 16.717998504638672,
-      "rewards/real": 10.698628425598145,
+      "logits/generated": -2.779353380203247,
+      "logits/real": -2.822978973388672,
+      "logps/generated": -153.47012329101562,
+      "logps/real": -103.8845443725586,
+      "loss": 0.2082,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -6.192803382873535,
+      "rewards/margins": 16.83620834350586,
+      "rewards/real": 10.64340591430664,
       "step": 1080
     },
     {
       "epoch": 0.6973768394113884,
-      "grad_norm": 4.541077368968895,
+      "grad_norm": 4.263244900035963,
       "learning_rate": 1.6820768136557609e-07,
-      "logits/generated": -2.6836509704589844,
-      "logits/real": -2.7782435417175293,
-      "logps/generated": -140.19631958007812,
-      "logps/real": -103.10017395019531,
-      "loss": 0.2103,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.432912349700928,
-      "rewards/margins": 16.41040802001953,
-      "rewards/real": 10.977496147155762,
+      "logits/generated": -2.7456464767456055,
+      "logits/real": -2.7892534732818604,
+      "logps/generated": -143.53396606445312,
+      "logps/real": -103.48819732666016,
+      "loss": 0.212,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.766679286956787,
+      "rewards/margins": 16.705373764038086,
+      "rewards/real": 10.938694953918457,
       "step": 1090
     },
     {
       "epoch": 0.7037747920665387,
-      "grad_norm": 2.134995135769877,
+      "grad_norm": 2.1937590854299387,
       "learning_rate": 1.64651493598862e-07,
-      "logits/generated": -2.700814962387085,
-      "logits/real": -2.7614195346832275,
-      "logps/generated": -139.90142822265625,
-      "logps/real": -101.66877746582031,
-      "loss": 0.2131,
+      "logits/generated": -2.769009828567505,
+      "logits/real": -2.7877819538116455,
+      "logps/generated": -144.3419647216797,
+      "logps/real": -102.21932220458984,
+      "loss": 0.2152,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/generated": -5.645236015319824,
-      "rewards/margins": 16.219711303710938,
-      "rewards/real": 10.57447624206543,
+      "rewards/generated": -6.089290142059326,
+      "rewards/margins": 16.60870933532715,
+      "rewards/real": 10.519418716430664,
+      "step": 1100
+    },
+    {
+      "epoch": 0.7037747920665387,
+      "eval_logits/generated": -2.7789735794067383,
+      "eval_logits/real": -2.760922431945801,
+      "eval_logps/generated": -132.22039794921875,
+      "eval_logps/real": -111.79248046875,
+      "eval_loss": 0.229682058095932,
+      "eval_rewards/accuracies": 1.0,
+      "eval_rewards/generated": -5.350338935852051,
+      "eval_rewards/margins": 15.439202308654785,
+      "eval_rewards/real": 10.088862419128418,
+      "eval_runtime": 46.5634,
+      "eval_samples_per_second": 10.738,
+      "eval_steps_per_second": 0.344,
       "step": 1100
     },
     {
       "epoch": 0.710172744721689,
-      "grad_norm": 2.1362568113809277,
+      "grad_norm": 2.0480356732786817,
       "learning_rate": 1.6109530583214793e-07,
-      "logits/generated": -2.7405338287353516,
-      "logits/real": -2.801772117614746,
-      "logps/generated": -135.68980407714844,
-      "logps/real": -115.05977630615234,
-      "loss": 0.2147,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.29919958114624,
-      "rewards/margins": 17.55141258239746,
-      "rewards/real": 12.252215385437012,
+      "logits/generated": -2.782161235809326,
+      "logits/real": -2.815020799636841,
+      "logps/generated": -137.57997131347656,
+      "logps/real": -115.2011489868164,
+      "loss": 0.2159,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.488217353820801,
+      "rewards/margins": 17.726293563842773,
+      "rewards/real": 12.238077163696289,
       "step": 1110
     },
     {
       "epoch": 0.7165706973768394,
-      "grad_norm": 2.6687487627386965,
+      "grad_norm": 2.6686832423114897,
       "learning_rate": 1.5753911806543385e-07,
-      "logits/generated": -2.655336380004883,
-      "logits/real": -2.716346502304077,
-      "logps/generated": -138.2339630126953,
-      "logps/real": -106.83113098144531,
-      "loss": 0.2236,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.799440860748291,
-      "rewards/margins": 16.477214813232422,
-      "rewards/real": 10.677773475646973,
+      "logits/generated": -2.73688006401062,
+      "logits/real": -2.7528836727142334,
+      "logps/generated": -137.2068634033203,
+      "logps/real": -106.5394515991211,
+      "loss": 0.2232,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.696730613708496,
+      "rewards/margins": 16.403671264648438,
+      "rewards/real": 10.706939697265625,
       "step": 1120
     },
     {
       "epoch": 0.7229686500319897,
-      "grad_norm": 1.9796488466759263,
+      "grad_norm": 4.518229737327693,
       "learning_rate": 1.5398293029871974e-07,
-      "logits/generated": -2.7700233459472656,
-      "logits/real": -2.719053268432617,
-      "logps/generated": -138.2145233154297,
-      "logps/real": -101.98434448242188,
-      "loss": 0.1982,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.574553966522217,
-      "rewards/margins": 15.94153118133545,
-      "rewards/real": 10.366975784301758,
+      "logits/generated": -2.8164429664611816,
+      "logits/real": -2.7557027339935303,
+      "logps/generated": -141.74234008789062,
+      "logps/real": -102.67829895019531,
+      "loss": 0.1976,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.927333831787109,
+      "rewards/margins": 16.224916458129883,
+      "rewards/real": 10.297581672668457,
       "step": 1130
     },
     {
       "epoch": 0.7293666026871402,
-      "grad_norm": 1.97652173855478,
+      "grad_norm": 1.977409807564797,
       "learning_rate": 1.504267425320057e-07,
-      "logits/generated": -2.749962091445923,
-      "logits/real": -2.7098100185394287,
-      "logps/generated": -128.74168395996094,
-      "logps/real": -103.87724304199219,
-      "loss": 0.203,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.09400749206543,
-      "rewards/margins": 16.894054412841797,
-      "rewards/real": 11.800048828125,
+      "logits/generated": -2.7976534366607666,
+      "logits/real": -2.763936758041382,
+      "logps/generated": -131.56748962402344,
+      "logps/real": -103.92170715332031,
+      "loss": 0.2012,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.376586437225342,
+      "rewards/margins": 17.172191619873047,
+      "rewards/real": 11.795602798461914,
       "step": 1140
     },
     {
       "epoch": 0.7357645553422905,
-      "grad_norm": 1.9276505695865507,
+      "grad_norm": 1.8922199269618114,
       "learning_rate": 1.4687055476529158e-07,
-      "logits/generated": -2.7056586742401123,
-      "logits/real": -2.84588885307312,
-      "logps/generated": -132.11521911621094,
-      "logps/real": -103.04747009277344,
-      "loss": 0.2,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.32986307144165,
-      "rewards/margins": 16.5743408203125,
-      "rewards/real": 11.244477272033691,
+      "logits/generated": -2.7850334644317627,
+      "logits/real": -2.8464672565460205,
+      "logps/generated": -136.84791564941406,
+      "logps/real": -103.54283142089844,
+      "loss": 0.2027,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.8031325340271,
+      "rewards/margins": 16.99807357788086,
+      "rewards/real": 11.194940567016602,
       "step": 1150
     },
     {
       "epoch": 0.7421625079974408,
-      "grad_norm": 2.4003697136809587,
+      "grad_norm": 2.6213217056885787,
       "learning_rate": 1.4331436699857753e-07,
-      "logits/generated": -2.7689154148101807,
-      "logits/real": -2.7181622982025146,
-      "logps/generated": -128.201171875,
-      "logps/real": -111.31385803222656,
-      "loss": 0.2257,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.305639266967773,
-      "rewards/margins": 16.98516273498535,
-      "rewards/real": 11.679524421691895,
+      "logits/generated": -2.8033339977264404,
+      "logits/real": -2.7528252601623535,
+      "logps/generated": -129.36412048339844,
+      "logps/real": -111.74896240234375,
+      "loss": 0.2287,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.421934127807617,
+      "rewards/margins": 17.057947158813477,
+      "rewards/real": 11.63601303100586,
       "step": 1160
     },
     {
       "epoch": 0.7485604606525912,
-      "grad_norm": 2.1662792343990898,
+      "grad_norm": 2.0727377414015598,
       "learning_rate": 1.3975817923186345e-07,
-      "logits/generated": -2.7471747398376465,
-      "logits/real": -2.745327949523926,
-      "logps/generated": -131.89920043945312,
-      "logps/real": -101.08728790283203,
-      "loss": 0.2126,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/generated": -5.3999738693237305,
-      "rewards/margins": 16.271684646606445,
-      "rewards/real": 10.871710777282715,
+      "logits/generated": -2.792191505432129,
+      "logits/real": -2.7740938663482666,
+      "logps/generated": -132.3335418701172,
+      "logps/real": -101.607666015625,
+      "loss": 0.2133,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.443409442901611,
+      "rewards/margins": 16.263084411621094,
+      "rewards/real": 10.819673538208008,
       "step": 1170
     },
     {
       "epoch": 0.7549584133077415,
-      "grad_norm": 1.9618990548812567,
+      "grad_norm": 2.0740248268939316,
       "learning_rate": 1.3620199146514935e-07,
-      "logits/generated": -2.73884654045105,
-      "logits/real": -2.792114734649658,
-      "logps/generated": -139.20669555664062,
-      "logps/real": -111.1391830444336,
-      "loss": 0.2157,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.806328296661377,
-      "rewards/margins": 17.363449096679688,
-      "rewards/real": 11.557123184204102,
+      "logits/generated": -2.8059496879577637,
+      "logits/real": -2.8090994358062744,
+      "logps/generated": -141.11962890625,
+      "logps/real": -111.667236328125,
+      "loss": 0.2159,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.99761962890625,
+      "rewards/margins": 17.501935958862305,
+      "rewards/real": 11.504318237304688,
       "step": 1180
     },
     {
       "epoch": 0.7613563659628919,
-      "grad_norm": 1.8565236814707453,
+      "grad_norm": 2.0778738450899397,
       "learning_rate": 1.326458036984353e-07,
-      "logits/generated": -2.708221673965454,
-      "logits/real": -2.8019871711730957,
-      "logps/generated": -141.60244750976562,
-      "logps/real": -103.365966796875,
-      "loss": 0.2066,
+      "logits/generated": -2.788054943084717,
+      "logits/real": -2.809278964996338,
+      "logps/generated": -144.7250213623047,
+      "logps/real": -104.4089126586914,
+      "loss": 0.2087,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/generated": -5.380032539367676,
-      "rewards/margins": 15.911615371704102,
-      "rewards/real": 10.53158187866211,
+      "rewards/generated": -5.692290306091309,
+      "rewards/margins": 16.119579315185547,
+      "rewards/real": 10.427286148071289,
       "step": 1190
     },
     {
       "epoch": 0.7677543186180422,
-      "grad_norm": 5.832771744417194,
+      "grad_norm": 9.932766239342575,
       "learning_rate": 1.290896159317212e-07,
-      "logits/generated": -2.7190277576446533,
-      "logits/real": -2.782357692718506,
-      "logps/generated": -133.46572875976562,
-      "logps/real": -102.03229522705078,
-      "loss": 0.2075,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.16973876953125,
-      "rewards/margins": 16.539508819580078,
-      "rewards/real": 11.369768142700195,
+      "logits/generated": -2.7904648780822754,
+      "logits/real": -2.7976596355438232,
+      "logps/generated": -137.5325469970703,
+      "logps/real": -101.97086334228516,
+      "loss": 0.2074,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.576420783996582,
+      "rewards/margins": 16.95233154296875,
+      "rewards/real": 11.375910758972168,
       "step": 1200
     },
     {
       "epoch": 0.7677543186180422,
-      "eval_logits/generated": -2.7359671592712402,
-      "eval_logits/real": -2.7372217178344727,
-      "eval_logps/generated": -128.395751953125,
-      "eval_logps/real": -111.5260009765625,
-      "eval_loss": 0.23038950562477112,
-      "eval_rewards/accuracies": 0.9921875,
-      "eval_rewards/generated": -4.9678730964660645,
-      "eval_rewards/margins": 15.083383560180664,
-      "eval_rewards/real": 10.115509986877441,
-      "eval_runtime": 69.835,
-      "eval_samples_per_second": 7.16,
-      "eval_steps_per_second": 0.229,
+      "eval_logits/generated": -2.7918548583984375,
+      "eval_logits/real": -2.7734375,
+      "eval_logps/generated": -131.92091369628906,
+      "eval_logps/real": -111.53852844238281,
+      "eval_loss": 0.22978967428207397,
+      "eval_rewards/accuracies": 1.0,
+      "eval_rewards/generated": -5.320389747619629,
+      "eval_rewards/margins": 15.434647560119629,
+      "eval_rewards/real": 10.114256858825684,
+      "eval_runtime": 47.5249,
+      "eval_samples_per_second": 10.521,
+      "eval_steps_per_second": 0.337,
       "step": 1200
     },
     {
       "epoch": 0.7741522712731925,
-      "grad_norm": 2.3644426695343914,
+      "grad_norm": 2.1890944986757046,
       "learning_rate": 1.255334281650071e-07,
-      "logits/generated": -2.702448606491089,
-      "logits/real": -2.849297046661377,
-      "logps/generated": -133.25411987304688,
-      "logps/real": -94.89964294433594,
-      "loss": 0.2192,
+      "logits/generated": -2.774868965148926,
+      "logits/real": -2.8254475593566895,
+      "logps/generated": -135.10997009277344,
+      "logps/real": -95.09330749511719,
+      "loss": 0.2194,
       "rewards/accuracies": 1.0,
-      "rewards/generated": -5.127745151519775,
-      "rewards/margins": 15.297076225280762,
-      "rewards/real": 10.169332504272461,
+      "rewards/generated": -5.313329219818115,
+      "rewards/margins": 15.463293075561523,
+      "rewards/real": 10.149965286254883,
       "step": 1210
     },
     {
       "epoch": 0.780550223928343,
-      "grad_norm": 4.2920815806287695,
+      "grad_norm": 3.3483681397588274,
       "learning_rate": 1.2197724039829303e-07,
-      "logits/generated": -2.8339786529541016,
-      "logits/real": -2.757995128631592,
-      "logps/generated": -148.0692138671875,
-      "logps/real": -121.7199478149414,
-      "loss": 0.2159,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -6.0756330490112305,
-      "rewards/margins": 18.304834365844727,
-      "rewards/real": 12.229202270507812,
+      "logits/generated": -2.8592276573181152,
+      "logits/real": -2.791822910308838,
+      "logps/generated": -150.6198272705078,
+      "logps/real": -122.3782730102539,
+      "loss": 0.2143,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -6.330694198608398,
+      "rewards/margins": 18.494060516357422,
+      "rewards/real": 12.16336727142334,
       "step": 1220
     },
     {
       "epoch": 0.7869481765834933,
-      "grad_norm": 2.624742549146953,
+      "grad_norm": 2.4905358123030705,
       "learning_rate": 1.1842105263157894e-07,
-      "logits/generated": -2.7471704483032227,
-      "logits/real": -2.8416290283203125,
-      "logps/generated": -147.3047637939453,
-      "logps/real": -111.00262451171875,
-      "loss": 0.2119,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.8865203857421875,
-      "rewards/margins": 17.451772689819336,
-      "rewards/real": 11.565254211425781,
+      "logits/generated": -2.7992920875549316,
+      "logits/real": -2.826453447341919,
+      "logps/generated": -150.37808227539062,
+      "logps/real": -111.29316711425781,
+      "loss": 0.2125,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -6.193851947784424,
+      "rewards/margins": 17.730052947998047,
+      "rewards/real": 11.536202430725098,
       "step": 1230
     },
     {
       "epoch": 0.7933461292386437,
-      "grad_norm": 2.1307989909566256,
+      "grad_norm": 2.14516868618303,
       "learning_rate": 1.1486486486486487e-07,
-      "logits/generated": -2.731872797012329,
-      "logits/real": -2.8878135681152344,
-      "logps/generated": -172.3035430908203,
-      "logps/real": -110.19361877441406,
-      "loss": 0.2202,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -7.301609039306641,
-      "rewards/margins": 18.403263092041016,
-      "rewards/real": 11.101654052734375,
+      "logits/generated": -2.787862777709961,
+      "logits/real": -2.851105213165283,
+      "logps/generated": -173.12652587890625,
+      "logps/real": -110.62249755859375,
+      "loss": 0.2214,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -7.38390588760376,
+      "rewards/margins": 18.442672729492188,
+      "rewards/real": 11.058765411376953,
       "step": 1240
     },
     {
       "epoch": 0.799744081893794,
-      "grad_norm": 2.18810616018797,
+      "grad_norm": 2.5675940201211067,
       "learning_rate": 1.1130867709815078e-07,
-      "logits/generated": -2.635159969329834,
-      "logits/real": -2.8312065601348877,
-      "logps/generated": -154.04550170898438,
-      "logps/real": -104.9063949584961,
-      "loss": 0.195,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -6.582805633544922,
-      "rewards/margins": 17.949687957763672,
-      "rewards/real": 11.366880416870117,
+      "logits/generated": -2.7247021198272705,
+      "logits/real": -2.826866388320923,
+      "logps/generated": -156.9348907470703,
+      "logps/real": -104.7955093383789,
+      "loss": 0.1951,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -6.871743202209473,
+      "rewards/margins": 18.249713897705078,
+      "rewards/real": 11.377972602844238,
       "step": 1250
     },
     {
       "epoch": 0.8061420345489443,
-      "grad_norm": 2.3243034253667,
+      "grad_norm": 2.3220199048245886,
       "learning_rate": 1.077524893314367e-07,
-      "logits/generated": -2.7570877075195312,
-      "logits/real": -2.676670789718628,
-      "logps/generated": -132.24571228027344,
-      "logps/real": -110.49552154541016,
-      "loss": 0.2048,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.188769340515137,
-      "rewards/margins": 16.681943893432617,
-      "rewards/real": 11.49317455291748,
+      "logits/generated": -2.79024600982666,
+      "logits/real": -2.7265076637268066,
+      "logps/generated": -136.0201416015625,
+      "logps/real": -111.22340393066406,
+      "loss": 0.2053,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.566214084625244,
+      "rewards/margins": 16.98659896850586,
+      "rewards/real": 11.420385360717773,
       "step": 1260
     },
     {
       "epoch": 0.8125399872040947,
-      "grad_norm": 2.22786920352928,
+      "grad_norm": 2.1814610270763954,
       "learning_rate": 1.0419630156472262e-07,
-      "logits/generated": -2.7040412425994873,
-      "logits/real": -2.7337164878845215,
-      "logps/generated": -129.7920379638672,
-      "logps/real": -102.24781799316406,
-      "loss": 0.2098,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -4.8722944259643555,
-      "rewards/margins": 16.266225814819336,
-      "rewards/real": 11.393930435180664,
+      "logits/generated": -2.769841194152832,
+      "logits/real": -2.773723602294922,
+      "logps/generated": -133.33799743652344,
+      "logps/real": -102.33349609375,
+      "loss": 0.2082,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.22689151763916,
+      "rewards/margins": 16.612255096435547,
+      "rewards/real": 11.385361671447754,
       "step": 1270
     },
     {
       "epoch": 0.818937939859245,
-      "grad_norm": 2.000001121079601,
+      "grad_norm": 1.9182196152080584,
       "learning_rate": 1.0064011379800854e-07,
-      "logits/generated": -2.724949359893799,
-      "logits/real": -2.7215514183044434,
-      "logps/generated": -146.55990600585938,
-      "logps/real": -114.5284423828125,
-      "loss": 0.2211,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.995248794555664,
-      "rewards/margins": 17.621685028076172,
-      "rewards/real": 11.626437187194824,
+      "logits/generated": -2.7872636318206787,
+      "logits/real": -2.765195369720459,
+      "logps/generated": -148.7725830078125,
+      "logps/real": -115.18281555175781,
+      "loss": 0.2226,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -6.216516971588135,
+      "rewards/margins": 17.777515411376953,
+      "rewards/real": 11.560999870300293,
       "step": 1280
     },
     {
       "epoch": 0.8253358925143954,
-      "grad_norm": 2.220536708583519,
+      "grad_norm": 2.1969885281530446,
       "learning_rate": 9.708392603129445e-08,
-      "logits/generated": -2.6592986583709717,
-      "logits/real": -2.766054391860962,
-      "logps/generated": -141.36831665039062,
-      "logps/real": -96.46757507324219,
-      "loss": 0.2112,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.908102989196777,
-      "rewards/margins": 16.13393783569336,
-      "rewards/real": 10.225834846496582,
+      "logits/generated": -2.7381138801574707,
+      "logits/real": -2.7735724449157715,
+      "logps/generated": -145.60491943359375,
+      "logps/real": -96.8055191040039,
+      "loss": 0.2097,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -6.331765174865723,
+      "rewards/margins": 16.523807525634766,
+      "rewards/real": 10.192041397094727,
       "step": 1290
     },
     {
       "epoch": 0.8317338451695457,
-      "grad_norm": 1.8458396373240298,
+      "grad_norm": 1.9560467278934162,
       "learning_rate": 9.352773826458037e-08,
-      "logits/generated": -2.7686662673950195,
-      "logits/real": -2.6808652877807617,
-      "logps/generated": -147.9801788330078,
-      "logps/real": -109.27278900146484,
-      "loss": 0.2098,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -6.193856716156006,
-      "rewards/margins": 17.859333038330078,
-      "rewards/real": 11.66547679901123,
+      "logits/generated": -2.804027795791626,
+      "logits/real": -2.725379467010498,
+      "logps/generated": -148.90652465820312,
+      "logps/real": -109.75422668457031,
+      "loss": 0.2107,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -6.286490440368652,
+      "rewards/margins": 17.903825759887695,
+      "rewards/real": 11.617334365844727,
+      "step": 1300
+    },
+    {
+      "epoch": 0.8317338451695457,
+      "eval_logits/generated": -2.773404836654663,
+      "eval_logits/real": -2.7523679733276367,
+      "eval_logps/generated": -131.85391235351562,
+      "eval_logps/real": -111.3324203491211,
+      "eval_loss": 0.22871987521648407,
+      "eval_rewards/accuracies": 1.0,
+      "eval_rewards/generated": -5.313689231872559,
+      "eval_rewards/margins": 15.448558807373047,
+      "eval_rewards/real": 10.134868621826172,
+      "eval_runtime": 48.0387,
+      "eval_samples_per_second": 10.408,
+      "eval_steps_per_second": 0.333,
       "step": 1300
     },
     {
       "epoch": 0.838131797824696,
-      "grad_norm": 2.2735605764303304,
+      "grad_norm": 2.898650764405823,
       "learning_rate": 8.997155049786629e-08,
-      "logits/generated": -2.791935443878174,
-      "logits/real": -2.7119717597961426,
-      "logps/generated": -136.2453155517578,
-      "logps/real": -101.1696548461914,
-      "loss": 0.2048,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.072253227233887,
-      "rewards/margins": 15.385726928710938,
-      "rewards/real": 10.313471794128418,
+      "logits/generated": -2.8157131671905518,
+      "logits/real": -2.7465660572052,
+      "logps/generated": -139.5260772705078,
+      "logps/real": -101.49971008300781,
+      "loss": 0.2059,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.40032958984375,
+      "rewards/margins": 15.680795669555664,
+      "rewards/real": 10.280465126037598,
       "step": 1310
     },
     {
       "epoch": 0.8445297504798465,
-      "grad_norm": 7.320467978107467,
+      "grad_norm": 3.225916689696325,
       "learning_rate": 8.64153627311522e-08,
-      "logits/generated": -2.716169834136963,
-      "logits/real": -2.7271907329559326,
-      "logps/generated": -142.0037384033203,
-      "logps/real": -96.90984344482422,
-      "loss": 0.2039,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.719296455383301,
-      "rewards/margins": 16.5465087890625,
-      "rewards/real": 10.827211380004883,
+      "logits/generated": -2.7690861225128174,
+      "logits/real": -2.7490289211273193,
+      "logps/generated": -146.69351196289062,
+      "logps/real": -98.00682830810547,
+      "loss": 0.2022,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -6.188272953033447,
+      "rewards/margins": 16.905786514282227,
+      "rewards/real": 10.717513084411621,
       "step": 1320
     },
     {
       "epoch": 0.8509277031349968,
-      "grad_norm": 2.1090442333698927,
+      "grad_norm": 2.0920801958873403,
       "learning_rate": 8.285917496443812e-08,
-      "logits/generated": -2.7019193172454834,
-      "logits/real": -2.7879014015197754,
-      "logps/generated": -131.5638885498047,
-      "logps/real": -103.13008880615234,
-      "loss": 0.2036,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.254150390625,
-      "rewards/margins": 15.576313972473145,
-      "rewards/real": 10.322164535522461,
+      "logits/generated": -2.7538485527038574,
+      "logits/real": -2.7750000953674316,
+      "logps/generated": -135.82803344726562,
+      "logps/real": -102.59940338134766,
+      "loss": 0.2043,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.680565357208252,
+      "rewards/margins": 16.055797576904297,
+      "rewards/real": 10.375231742858887,
       "step": 1330
     },
     {
       "epoch": 0.8573256557901472,
-      "grad_norm": 1.9764553854622655,
+      "grad_norm": 1.9642535936545011,
       "learning_rate": 7.930298719772404e-08,
-      "logits/generated": -2.7741990089416504,
-      "logits/real": -2.786623954772949,
-      "logps/generated": -128.60775756835938,
-      "logps/real": -95.55816650390625,
-      "loss": 0.2071,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -4.881311893463135,
-      "rewards/margins": 15.2656888961792,
-      "rewards/real": 10.384378433227539,
+      "logits/generated": -2.8019230365753174,
+      "logits/real": -2.7879862785339355,
+      "logps/generated": -130.47720336914062,
+      "logps/real": -95.87910461425781,
+      "loss": 0.2064,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.068253517150879,
+      "rewards/margins": 15.420539855957031,
+      "rewards/real": 10.35228443145752,
       "step": 1340
     },
     {
       "epoch": 0.8637236084452975,
-      "grad_norm": 2.4449760593366765,
+      "grad_norm": 2.460552007438197,
       "learning_rate": 7.574679943100994e-08,
-      "logits/generated": -2.7581117153167725,
-      "logits/real": -2.8711161613464355,
-      "logps/generated": -159.32579040527344,
-      "logps/real": -127.8974838256836,
-      "loss": 0.2075,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -6.493968963623047,
-      "rewards/margins": 19.491052627563477,
-      "rewards/real": 12.997082710266113,
+      "logits/generated": -2.8074569702148438,
+      "logits/real": -2.8504409790039062,
+      "logps/generated": -162.3019561767578,
+      "logps/real": -128.05322265625,
+      "loss": 0.208,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -6.791585445404053,
+      "rewards/margins": 19.773096084594727,
+      "rewards/real": 12.981508255004883,
       "step": 1350
     },
     {
       "epoch": 0.8701215611004478,
-      "grad_norm": 1.8244220811357807,
+      "grad_norm": 2.0596823227897985,
       "learning_rate": 7.219061166429587e-08,
-      "logits/generated": -2.740255832672119,
-      "logits/real": -2.8236019611358643,
-      "logps/generated": -140.54568481445312,
-      "logps/real": -108.64134216308594,
-      "loss": 0.1994,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.692780494689941,
-      "rewards/margins": 16.945453643798828,
-      "rewards/real": 11.25267219543457,
+      "logits/generated": -2.7889676094055176,
+      "logits/real": -2.8090572357177734,
+      "logps/generated": -142.484130859375,
+      "logps/real": -109.0689926147461,
+      "loss": 0.1997,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.886624813079834,
+      "rewards/margins": 17.09653091430664,
+      "rewards/real": 11.209908485412598,
       "step": 1360
     },
     {
       "epoch": 0.8765195137555982,
-      "grad_norm": 2.9020057472557528,
+      "grad_norm": 2.627724188734489,
       "learning_rate": 6.863442389758179e-08,
-      "logits/generated": -2.748126745223999,
-      "logits/real": -2.682790517807007,
-      "logps/generated": -129.32846069335938,
-      "logps/real": -92.95919036865234,
-      "loss": 0.2034,
+      "logits/generated": -2.7995376586914062,
+      "logits/real": -2.732335329055786,
+      "logps/generated": -133.498291015625,
+      "logps/real": -93.63945007324219,
+      "loss": 0.203,
       "rewards/accuracies": 1.0,
-      "rewards/generated": -5.053739547729492,
-      "rewards/margins": 15.391085624694824,
-      "rewards/real": 10.337347030639648,
+      "rewards/generated": -5.470720291137695,
+      "rewards/margins": 15.74004077911377,
+      "rewards/real": 10.269319534301758,
       "step": 1370
     },
     {
       "epoch": 0.8829174664107485,
-      "grad_norm": 2.342066981273815,
+      "grad_norm": 2.0576376488108665,
       "learning_rate": 6.507823613086771e-08,
-      "logits/generated": -2.734999179840088,
-      "logits/real": -2.705516815185547,
-      "logps/generated": -126.626708984375,
-      "logps/real": -108.821044921875,
-      "loss": 0.2083,
+      "logits/generated": -2.786233901977539,
+      "logits/real": -2.752906084060669,
+      "logps/generated": -130.2691650390625,
+      "logps/real": -109.58293151855469,
+      "loss": 0.2036,
       "rewards/accuracies": 1.0,
-      "rewards/generated": -5.249380111694336,
-      "rewards/margins": 17.1163387298584,
-      "rewards/real": 11.866960525512695,
+      "rewards/generated": -5.61362361907959,
+      "rewards/margins": 17.404396057128906,
+      "rewards/real": 11.790771484375,
       "step": 1380
     },
     {
       "epoch": 0.889315419065899,
-      "grad_norm": 2.2461978230592665,
+      "grad_norm": 2.1180329972387457,
       "learning_rate": 6.152204836415363e-08,
-      "logits/generated": -2.715066432952881,
-      "logits/real": -2.795833110809326,
-      "logps/generated": -136.6764373779297,
-      "logps/real": -112.36259460449219,
-      "loss": 0.219,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.541347026824951,
-      "rewards/margins": 17.154422760009766,
-      "rewards/real": 11.613073348999023,
+      "logits/generated": -2.7736868858337402,
+      "logits/real": -2.8066494464874268,
+      "logps/generated": -141.78265380859375,
+      "logps/real": -112.9043197631836,
+      "loss": 0.2192,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -6.051970958709717,
+      "rewards/margins": 17.610872268676758,
+      "rewards/real": 11.558903694152832,
       "step": 1390
     },
     {
       "epoch": 0.8957133717210493,
-      "grad_norm": 1.9643094015838993,
+      "grad_norm": 2.1797083625640905,
       "learning_rate": 5.796586059743954e-08,
-      "logits/generated": -2.7037057876586914,
-      "logits/real": -2.892324447631836,
-      "logps/generated": -144.58990478515625,
-      "logps/real": -93.6194839477539,
-      "loss": 0.1961,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.912665843963623,
-      "rewards/margins": 16.394804000854492,
-      "rewards/real": 10.482137680053711,
+      "logits/generated": -2.768073320388794,
+      "logits/real": -2.839402675628662,
+      "logps/generated": -148.06509399414062,
+      "logps/real": -94.10154724121094,
+      "loss": 0.1947,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -6.2601847648620605,
+      "rewards/margins": 16.694116592407227,
+      "rewards/real": 10.433931350708008,
       "step": 1400
     },
     {
       "epoch": 0.8957133717210493,
-      "eval_logits/generated": -2.731992244720459,
-      "eval_logits/real": -2.7331748008728027,
-      "eval_logps/generated": -128.72311401367188,
-      "eval_logps/real": -111.41729736328125,
-      "eval_loss": 0.22951382398605347,
-      "eval_rewards/accuracies": 0.9921875,
-      "eval_rewards/generated": -5.000611305236816,
-      "eval_rewards/margins": 15.126992225646973,
-      "eval_rewards/real": 10.12637996673584,
-      "eval_runtime": 68.2753,
-      "eval_samples_per_second": 7.323,
-      "eval_steps_per_second": 0.234,
+      "eval_logits/generated": -2.7803165912628174,
+      "eval_logits/real": -2.761274814605713,
+      "eval_logps/generated": -131.96864318847656,
+      "eval_logps/real": -111.41598510742188,
+      "eval_loss": 0.22879938781261444,
+      "eval_rewards/accuracies": 1.0,
+      "eval_rewards/generated": -5.325164318084717,
+      "eval_rewards/margins": 15.451674461364746,
+      "eval_rewards/real": 10.12651252746582,
+      "eval_runtime": 46.421,
+      "eval_samples_per_second": 10.771,
+      "eval_steps_per_second": 0.345,
       "step": 1400
     },
     {
       "epoch": 0.9021113243761996,
-      "grad_norm": 1.9508599183671798,
+      "grad_norm": 4.0872187897445755,
       "learning_rate": 5.4409672830725456e-08,
-      "logits/generated": -2.771930694580078,
-      "logits/real": -2.9047772884368896,
-      "logps/generated": -157.674072265625,
-      "logps/real": -95.12882232666016,
-      "loss": 0.2091,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -6.542889595031738,
-      "rewards/margins": 16.641141891479492,
-      "rewards/real": 10.098252296447754,
+      "logits/generated": -2.819594621658325,
+      "logits/real": -2.8617167472839355,
+      "logps/generated": -158.43478393554688,
+      "logps/real": -95.21733093261719,
+      "loss": 0.2097,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -6.618961334228516,
+      "rewards/margins": 16.708362579345703,
+      "rewards/real": 10.089402198791504,
       "step": 1410
     },
     {
       "epoch": 0.90850927703135,
-      "grad_norm": 8.200787208344204,
+      "grad_norm": 3.6943177243425716,
       "learning_rate": 5.0853485064011376e-08,
-      "logits/generated": -2.743286609649658,
-      "logits/real": -2.764103412628174,
-      "logps/generated": -124.41139221191406,
-      "logps/real": -113.6160659790039,
-      "loss": 0.2089,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -4.869805812835693,
-      "rewards/margins": 16.73537826538086,
-      "rewards/real": 11.865571975708008,
+      "logits/generated": -2.779648542404175,
+      "logits/real": -2.775038003921509,
+      "logps/generated": -129.05410766601562,
+      "logps/real": -114.03041076660156,
+      "loss": 0.2084,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.334076881408691,
+      "rewards/margins": 17.158214569091797,
+      "rewards/real": 11.824136734008789,
       "step": 1420
     },
     {
       "epoch": 0.9149072296865003,
-      "grad_norm": 4.459379575662199,
+      "grad_norm": 4.8280629189298505,
       "learning_rate": 4.72972972972973e-08,
-      "logits/generated": -2.746967077255249,
-      "logits/real": -2.726330280303955,
-      "logps/generated": -127.21881103515625,
-      "logps/real": -105.5604248046875,
-      "loss": 0.2142,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -4.854151248931885,
-      "rewards/margins": 16.145641326904297,
-      "rewards/real": 11.291491508483887,
+      "logits/generated": -2.7740349769592285,
+      "logits/real": -2.7425434589385986,
+      "logps/generated": -130.97879028320312,
+      "logps/real": -106.00202941894531,
+      "loss": 0.2151,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.230149269104004,
+      "rewards/margins": 16.477481842041016,
+      "rewards/real": 11.247332572937012,
       "step": 1430
     },
     {
       "epoch": 0.9213051823416507,
-      "grad_norm": 2.74872128438021,
+      "grad_norm": 2.148815074874242,
       "learning_rate": 4.374110953058322e-08,
-      "logits/generated": -2.7054824829101562,
-      "logits/real": -2.7788712978363037,
-      "logps/generated": -141.21192932128906,
-      "logps/real": -107.06086730957031,
-      "loss": 0.2159,
+      "logits/generated": -2.763887882232666,
+      "logits/real": -2.7833125591278076,
+      "logps/generated": -145.046630859375,
+      "logps/real": -107.34722900390625,
+      "loss": 0.2189,
       "rewards/accuracies": 1.0,
-      "rewards/generated": -5.830003261566162,
-      "rewards/margins": 17.112892150878906,
-      "rewards/real": 11.282888412475586,
+      "rewards/generated": -6.21347188949585,
+      "rewards/margins": 17.467723846435547,
+      "rewards/real": 11.254251480102539,
       "step": 1440
     },
     {
       "epoch": 0.927703134996801,
-      "grad_norm": 2.4246159341403537,
+      "grad_norm": 2.541830296879894,
       "learning_rate": 4.018492176386913e-08,
-      "logits/generated": -2.6539907455444336,
-      "logits/real": -2.773902177810669,
-      "logps/generated": -130.4512939453125,
-      "logps/real": -101.6545181274414,
-      "loss": 0.2101,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -4.960906028747559,
-      "rewards/margins": 15.910059928894043,
-      "rewards/real": 10.949153900146484,
+      "logits/generated": -2.7273221015930176,
+      "logits/real": -2.784517765045166,
+      "logps/generated": -135.05178833007812,
+      "logps/real": -102.04019927978516,
+      "loss": 0.2091,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.420957088470459,
+      "rewards/margins": 16.331541061401367,
+      "rewards/real": 10.910585403442383,
       "step": 1450
     },
     {
       "epoch": 0.9341010876519513,
-      "grad_norm": 1.8673176828484437,
+      "grad_norm": 1.826090506408138,
       "learning_rate": 3.6628733997155046e-08,
-      "logits/generated": -2.7114219665527344,
-      "logits/real": -2.8677616119384766,
-      "logps/generated": -130.7792510986328,
-      "logps/real": -88.69828796386719,
-      "loss": 0.1885,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.1922407150268555,
-      "rewards/margins": 14.514605522155762,
-      "rewards/real": 9.32236385345459,
+      "logits/generated": -2.773800849914551,
+      "logits/real": -2.8277792930603027,
+      "logps/generated": -135.4071807861328,
+      "logps/real": -88.8923110961914,
+      "loss": 0.1887,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.65503454208374,
+      "rewards/margins": 14.957998275756836,
+      "rewards/real": 9.302962303161621,
       "step": 1460
     },
     {
       "epoch": 0.9404990403071017,
-      "grad_norm": 2.167805338937868,
+      "grad_norm": 2.208244651604884,
       "learning_rate": 3.3072546230440967e-08,
-      "logits/generated": -2.709365129470825,
-      "logits/real": -2.7279558181762695,
-      "logps/generated": -129.8989715576172,
-      "logps/real": -96.73854064941406,
-      "loss": 0.1966,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.115813255310059,
-      "rewards/margins": 15.31916332244873,
-      "rewards/real": 10.203348159790039,
+      "logits/generated": -2.7581300735473633,
+      "logits/real": -2.734788417816162,
+      "logps/generated": -133.66445922851562,
+      "logps/real": -97.28591918945312,
+      "loss": 0.1969,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.492359638214111,
+      "rewards/margins": 15.640971183776855,
+      "rewards/real": 10.148611068725586,
       "step": 1470
     },
     {
       "epoch": 0.946896992962252,
-      "grad_norm": 2.343424293796162,
+      "grad_norm": 2.3306023563595972,
       "learning_rate": 2.9516358463726884e-08,
-      "logits/generated": -2.6958348751068115,
-      "logits/real": -2.7627506256103516,
-      "logps/generated": -132.286376953125,
-      "logps/real": -85.90814971923828,
-      "loss": 0.2066,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.348180294036865,
-      "rewards/margins": 14.917572021484375,
-      "rewards/real": 9.569391250610352,
+      "logits/generated": -2.747053384780884,
+      "logits/real": -2.7607858180999756,
+      "logps/generated": -133.93447875976562,
+      "logps/real": -86.20393371582031,
+      "loss": 0.2093,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.512990474700928,
+      "rewards/margins": 15.052803039550781,
+      "rewards/real": 9.539812088012695,
       "step": 1480
     },
     {
       "epoch": 0.9532949456174025,
-      "grad_norm": 2.443526667812465,
+      "grad_norm": 2.075410140017091,
       "learning_rate": 2.59601706970128e-08,
-      "logits/generated": -2.731065273284912,
-      "logits/real": -2.722964286804199,
-      "logps/generated": -144.53329467773438,
-      "logps/real": -111.11639404296875,
-      "loss": 0.2125,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.713242530822754,
-      "rewards/margins": 17.423105239868164,
-      "rewards/real": 11.70986270904541,
+      "logits/generated": -2.777174472808838,
+      "logits/real": -2.7545228004455566,
+      "logps/generated": -146.89410400390625,
+      "logps/real": -111.48350524902344,
+      "loss": 0.2124,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.9493231773376465,
+      "rewards/margins": 17.622472763061523,
+      "rewards/real": 11.673151016235352,
       "step": 1490
     },
     {
       "epoch": 0.9596928982725528,
-      "grad_norm": 2.5802622540203726,
+      "grad_norm": 2.5791143344257694,
       "learning_rate": 2.240398293029872e-08,
-      "logits/generated": -2.7666499614715576,
-      "logits/real": -2.7990810871124268,
-      "logps/generated": -151.97225952148438,
-      "logps/real": -102.18070983886719,
-      "loss": 0.2062,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -6.25309419631958,
-      "rewards/margins": 17.280641555786133,
-      "rewards/real": 11.027546882629395,
+      "logits/generated": -2.800865888595581,
+      "logits/real": -2.7849934101104736,
+      "logps/generated": -155.4287872314453,
+      "logps/real": -102.43949127197266,
+      "loss": 0.2056,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -6.598745822906494,
+      "rewards/margins": 17.60041618347168,
+      "rewards/real": 11.001669883728027,
+      "step": 1500
+    },
+    {
+      "epoch": 0.9596928982725528,
+      "eval_logits/generated": -2.7693557739257812,
+      "eval_logits/real": -2.749887466430664,
+      "eval_logps/generated": -131.87547302246094,
+      "eval_logps/real": -111.33661651611328,
+      "eval_loss": 0.2284335047006607,
+      "eval_rewards/accuracies": 1.0,
+      "eval_rewards/generated": -5.315846920013428,
+      "eval_rewards/margins": 15.450297355651855,
+      "eval_rewards/real": 10.134449005126953,
+      "eval_runtime": 46.3769,
+      "eval_samples_per_second": 10.781,
+      "eval_steps_per_second": 0.345,
       "step": 1500
     },
     {
       "epoch": 0.9660908509277031,
-      "grad_norm": 4.900620511000788,
+      "grad_norm": 3.786311955695454,
       "learning_rate": 1.8847795163584636e-08,
-      "logits/generated": -2.7653520107269287,
-      "logits/real": -2.737842082977295,
-      "logps/generated": -132.54611206054688,
-      "logps/real": -105.3329086303711,
-      "loss": 0.2044,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.18030309677124,
-      "rewards/margins": 16.7701358795166,
-      "rewards/real": 11.58983325958252,
+      "logits/generated": -2.7993459701538086,
+      "logits/real": -2.766533374786377,
+      "logps/generated": -136.34921264648438,
+      "logps/real": -106.26002502441406,
+      "loss": 0.2036,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.560615539550781,
+      "rewards/margins": 17.0577392578125,
+      "rewards/real": 11.497122764587402,
       "step": 1510
     },
     {
       "epoch": 0.9724888035828535,
-      "grad_norm": 2.2165972838393935,
+      "grad_norm": 2.1316442858968334,
       "learning_rate": 1.5291607396870554e-08,
-      "logits/generated": -2.709165334701538,
-      "logits/real": -2.7248358726501465,
-      "logps/generated": -131.63613891601562,
-      "logps/real": -111.87278747558594,
-      "loss": 0.2149,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.262539863586426,
-      "rewards/margins": 17.233844757080078,
-      "rewards/real": 11.971303939819336,
+      "logits/generated": -2.7540335655212402,
+      "logits/real": -2.753798246383667,
+      "logps/generated": -136.44451904296875,
+      "logps/real": -112.07403564453125,
+      "loss": 0.2154,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.74337911605835,
+      "rewards/margins": 17.69455909729004,
+      "rewards/real": 11.951177597045898,
       "step": 1520
     },
     {
       "epoch": 0.9788867562380038,
-      "grad_norm": 4.970086199198337,
+      "grad_norm": 4.30706646214191,
       "learning_rate": 1.1735419630156473e-08,
-      "logits/generated": -2.7099928855895996,
-      "logits/real": -2.7440731525421143,
-      "logps/generated": -127.85652160644531,
-      "logps/real": -98.89935302734375,
-      "loss": 0.1979,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -4.844382286071777,
-      "rewards/margins": 15.250204086303711,
-      "rewards/real": 10.4058198928833,
+      "logits/generated": -2.7646923065185547,
+      "logits/real": -2.760829210281372,
+      "logps/generated": -132.25070190429688,
+      "logps/real": -99.32478332519531,
+      "loss": 0.1991,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.283801078796387,
+      "rewards/margins": 15.647079467773438,
+      "rewards/real": 10.363277435302734,
       "step": 1530
     },
     {
       "epoch": 0.9852847088931542,
-      "grad_norm": 2.031754041409757,
+      "grad_norm": 2.3548816010878273,
       "learning_rate": 8.179231863442388e-09,
-      "logits/generated": -2.6747050285339355,
-      "logits/real": -2.6660659313201904,
-      "logps/generated": -139.36756896972656,
-      "logps/real": -103.53269958496094,
+      "logits/generated": -2.7391297817230225,
+      "logits/real": -2.7132420539855957,
+      "logps/generated": -142.3213653564453,
+      "logps/real": -104.17442321777344,
       "loss": 0.2106,
       "rewards/accuracies": 1.0,
-      "rewards/generated": -5.578815460205078,
-      "rewards/margins": 16.89048194885254,
-      "rewards/real": 11.311666488647461,
+      "rewards/generated": -5.874195575714111,
+      "rewards/margins": 17.12169075012207,
+      "rewards/real": 11.247493743896484,
       "step": 1540
     },
     {
       "epoch": 0.9916826615483045,
-      "grad_norm": 2.2877158724102564,
+      "grad_norm": 2.217442538024829,
       "learning_rate": 4.623044096728307e-09,
-      "logits/generated": -2.660313606262207,
-      "logits/real": -2.855464220046997,
-      "logps/generated": -145.4078826904297,
-      "logps/real": -106.03314208984375,
-      "loss": 0.2083,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -5.917939186096191,
-      "rewards/margins": 17.319114685058594,
-      "rewards/real": 11.40117359161377,
+      "logits/generated": -2.7288365364074707,
+      "logits/real": -2.814866542816162,
+      "logps/generated": -146.8229217529297,
+      "logps/real": -105.86663818359375,
+      "loss": 0.2079,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -6.05944299697876,
+      "rewards/margins": 17.47726821899414,
+      "rewards/real": 11.417825698852539,
       "step": 1550
     },
     {
       "epoch": 0.9980806142034548,
-      "grad_norm": 2.427910151920255,
+      "grad_norm": 2.150021894295723,
       "learning_rate": 1.0668563300142248e-09,
-      "logits/generated": -2.697697401046753,
-      "logits/real": -2.7648138999938965,
-      "logps/generated": -132.71572875976562,
-      "logps/real": -100.33382415771484,
-      "loss": 0.2061,
-      "rewards/accuracies": 1.0,
-      "rewards/generated": -4.841822624206543,
-      "rewards/margins": 15.7789306640625,
-      "rewards/real": 10.937108039855957,
+      "logits/generated": -2.7558183670043945,
+      "logits/real": -2.773977756500244,
+      "logps/generated": -135.76776123046875,
+      "logps/real": -100.64725494384766,
+      "loss": 0.2071,
+      "rewards/accuracies": 1.0,
+      "rewards/generated": -5.147028923034668,
+      "rewards/margins": 16.052793502807617,
+      "rewards/real": 10.905765533447266,
       "step": 1560
     },
     {
       "epoch": 1.0,
       "step": 1563,
       "total_flos": 0.0,
-      "train_loss": 0.2369839982275618,
-      "train_runtime": 14627.5004,
-      "train_samples_per_second": 3.418,
-      "train_steps_per_second": 0.107
+      "train_loss": 0.2371997828675781,
+      "train_runtime": 22626.3768,
+      "train_samples_per_second": 2.21,
+      "train_steps_per_second": 0.069
     }
   ],
   "logging_steps": 10,
   "max_steps": 1563,
   "num_input_tokens_seen": 0,
   "num_train_epochs": 1,
-  "save_steps": 500,
+  "save_steps": 50,
   "stateful_callbacks": {
     "TrainerControl": {
       "args": {