Training in progress, step 4000, checkpoint

checkpoint-4000/config.json

@@ -84,7 +84,7 @@
   "speech_decoder_prenet_layers": 2,
   "speech_decoder_prenet_units": 256,
   "tie_word_embeddings": true,
-  "transformers_version": "5.19.0.dev0",
+  "transformers_version": "5.19.0",
   "use_cache": false,
   "use_guided_attention_loss": false,
   "vocab_size": 81

checkpoint-4000/generation_config.json

@@ -6,6 +6,6 @@
   "output_attentions": false,
   "output_hidden_states": false,
   "pad_token_id": 1,
-  "transformers_version": "5.19.0.dev0",
+  "transformers_version": "5.19.0",
   "use_cache": true
 }

checkpoint-4000/model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:27e459c5f7f0464df3af0838433916b887be053576245bdfcc1559c3f736bddf
+oid sha256:8cd71b8fd3a858f984ac9e75ffeb4c20d781c2e8045914e208851e3ca632233f
 size 577789320

checkpoint-4000/optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:82c2df09c6168958bddf9c64bc8f813d8a5034492dd548c0ff26d10d83ad42a6
+oid sha256:5f3aad8a01fbc68cf8bb62cf70700abfcea0a6e122150cd141ed4548539f493d
 size 1155772634

checkpoint-4000/rng_state.pth

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:70bb84e7cae61858d146cfc480d242d55e181d6e7bf4abeb65fdc8568b633c0b
+oid sha256:62598e980391bb7746347f70cb114fe3982477c73538995ae1fd42f784063e5c
 size 14645

checkpoint-4000/scheduler.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:51e2bdb3387aab664d6aa962ae96b60dd7d64c2d53dca5db8681fbfd246c6aca
+oid sha256:0f966efa7e63a187b503624edc0c09d8db72e8713aa4445f871e5274bdaf562f
 size 1465

checkpoint-4000/trainer_state.json

@@ -1,8 +1,8 @@
 {
   "best_global_step": 4000,
-  "best_metric": 0.9094438552856445,
+  "best_metric": 0.9033195376396179,
   "best_model_checkpoint": "/kaggle/working/tts_processed/checkpoint-4000",
-  "epoch": 19.80247678018576,
+  "epoch": 19.801980198019802,
   "eval_steps": 500,
   "global_step": 4000,
   "is_hyper_param_search": false,
@@ -10,1187 +10,1187 @@
   "is_world_process_zero": true,
   "log_history": [
     {
-      "epoch": 0.1238390092879257,
-      "grad_norm": 829.4996337890625,
+      "epoch": 0.12376237623762376,
+      "grad_norm": 363.43377685546875,
       "learning_rate": 4.8e-08,
-      "loss": 21.73699951171875,
+      "loss": 10.89256103515625,
       "step": 25
     },
     {
-      "epoch": 0.2476780185758514,
-      "grad_norm": 701.2330932617188,
+      "epoch": 0.24752475247524752,
+      "grad_norm": 336.6938171386719,
       "learning_rate": 9.8e-08,
-      "loss": 21.00220947265625,
+      "loss": 10.47541259765625,
       "step": 50
     },
     {
-      "epoch": 0.3715170278637771,
-      "grad_norm": 631.0809326171875,
+      "epoch": 0.3712871287128713,
+      "grad_norm": 255.0070343017578,
       "learning_rate": 1.4799999999999998e-07,
-      "loss": 19.772623291015623,
+      "loss": 10.08507568359375,
       "step": 75
     },
     {
-      "epoch": 0.4953560371517028,
-      "grad_norm": 556.35791015625,
+      "epoch": 0.49504950495049505,
+      "grad_norm": 247.54541015625,
       "learning_rate": 1.98e-07,
-      "loss": 18.46118896484375,
+      "loss": 9.282888793945313,
       "step": 100
     },
     {
-      "epoch": 0.6191950464396285,
-      "grad_norm": 536.2819213867188,
+      "epoch": 0.6188118811881188,
+      "grad_norm": 124.208740234375,
       "learning_rate": 2.48e-07,
-      "loss": 17.51901611328125,
+      "loss": 8.717689208984375,
       "step": 125
     },
     {
-      "epoch": 0.7430340557275542,
-      "grad_norm": 459.290283203125,
+      "epoch": 0.7425742574257426,
+      "grad_norm": 194.5498809814453,
       "learning_rate": 2.98e-07,
-      "loss": 16.43676513671875,
+      "loss": 8.212156372070313,
       "step": 150
     },
     {
-      "epoch": 0.8668730650154799,
-      "grad_norm": 230.56405639648438,
+      "epoch": 0.8663366336633663,
+      "grad_norm": 143.2777557373047,
       "learning_rate": 3.4799999999999994e-07,
-      "loss": 15.75008056640625,
+      "loss": 7.8293896484375,
       "step": 175
     },
     {
-      "epoch": 0.9907120743034056,
-      "grad_norm": 233.90635681152344,
+      "epoch": 0.9900990099009901,
+      "grad_norm": 55.50076675415039,
       "learning_rate": 3.98e-07,
-      "loss": 14.56845947265625,
+      "loss": 7.3809716796875,
       "step": 200
     },
     {
-      "epoch": 1.1139318885448917,
-      "grad_norm": 107.61080169677734,
+      "epoch": 1.113861386138614,
+      "grad_norm": 69.43047332763672,
       "learning_rate": 4.48e-07,
-      "loss": 13.99685546875,
+      "loss": 6.937587890625,
       "step": 225
     },
     {
-      "epoch": 1.2377708978328172,
-      "grad_norm": 115.48291015625,
+      "epoch": 1.2376237623762376,
+      "grad_norm": 49.69321060180664,
       "learning_rate": 4.979999999999999e-07,
-      "loss": 13.390751953125,
+      "loss": 6.8110888671875,
       "step": 250
     },
     {
-      "epoch": 1.361609907120743,
-      "grad_norm": 150.88845825195312,
+      "epoch": 1.3613861386138613,
+      "grad_norm": 26.72427749633789,
       "learning_rate": 5.48e-07,
-      "loss": 12.885958251953125,
+      "loss": 6.522671508789062,
       "step": 275
     },
     {
-      "epoch": 1.4854489164086688,
-      "grad_norm": 102.25688934326172,
+      "epoch": 1.4851485148514851,
+      "grad_norm": 51.36345672607422,
       "learning_rate": 5.979999999999999e-07,
-      "loss": 12.281358642578125,
+      "loss": 6.398966064453125,
       "step": 300
     },
     {
-      "epoch": 1.6092879256965944,
-      "grad_norm": 91.96704864501953,
+      "epoch": 1.608910891089109,
+      "grad_norm": 38.85439682006836,
       "learning_rate": 6.48e-07,
-      "loss": 12.51485107421875,
+      "loss": 6.198662109375,
       "step": 325
     },
     {
-      "epoch": 1.73312693498452,
-      "grad_norm": 89.19316864013672,
+      "epoch": 1.7326732673267327,
+      "grad_norm": 27.508665084838867,
       "learning_rate": 6.979999999999999e-07,
-      "loss": 11.758416748046875,
+      "loss": 6.075740356445312,
       "step": 350
     },
     {
-      "epoch": 1.8569659442724458,
-      "grad_norm": 43.06995391845703,
+      "epoch": 1.8564356435643563,
+      "grad_norm": 36.8642463684082,
       "learning_rate": 7.48e-07,
-      "loss": 11.7668603515625,
+      "loss": 6.0834722900390625,
       "step": 375
     },
     {
-      "epoch": 1.9808049535603716,
-      "grad_norm": 37.787078857421875,
+      "epoch": 1.9801980198019802,
+      "grad_norm": 27.07192039489746,
       "learning_rate": 7.98e-07,
-      "loss": 11.696016845703125,
+      "loss": 5.9166015625,
       "step": 400
     },
     {
-      "epoch": 2.1040247678018575,
-      "grad_norm": 52.63007354736328,
+      "epoch": 2.103960396039604,
+      "grad_norm": 22.465463638305664,
       "learning_rate": 8.48e-07,
-      "loss": 11.60255859375,
+      "loss": 5.795056762695313,
       "step": 425
     },
     {
-      "epoch": 2.2278637770897833,
-      "grad_norm": 37.73246383666992,
+      "epoch": 2.227722772277228,
+      "grad_norm": 34.86723709106445,
       "learning_rate": 8.98e-07,
-      "loss": 11.121099853515625,
+      "loss": 5.741965942382812,
       "step": 450
     },
     {
-      "epoch": 2.351702786377709,
-      "grad_norm": 50.654396057128906,
+      "epoch": 2.3514851485148514,
+      "grad_norm": 22.37932777404785,
       "learning_rate": 9.479999999999999e-07,
-      "loss": 10.981490478515624,
+      "loss": 5.716455688476563,
       "step": 475
     },
     {
-      "epoch": 2.4755417956656345,
-      "grad_norm": 34.40877151489258,
+      "epoch": 2.4752475247524752,
+      "grad_norm": 29.54213523864746,
       "learning_rate": 9.98e-07,
-      "loss": 10.89916015625,
+      "loss": 5.599214477539062,
       "step": 500
     },
     {
-      "epoch": 2.4755417956656345,
-      "eval_loss": 1.2403377294540405,
-      "eval_runtime": 78.2214,
-      "eval_samples_per_second": 20.659,
-      "eval_steps_per_second": 5.165,
+      "epoch": 2.4752475247524752,
+      "eval_loss": 1.2534884214401245,
+      "eval_runtime": 82.4141,
+      "eval_samples_per_second": 19.608,
+      "eval_steps_per_second": 4.902,
       "step": 500
     },
     {
-      "epoch": 2.5993808049535603,
-      "grad_norm": 32.53867721557617,
-      "learning_rate": 9.986285714285714e-07,
-      "loss": 10.70981201171875,
+      "epoch": 2.599009900990099,
+      "grad_norm": 21.54619026184082,
+      "learning_rate": 9.999953592789052e-07,
+      "loss": 5.526549072265625,
       "step": 525
     },
     {
-      "epoch": 2.723219814241486,
-      "grad_norm": 37.28243637084961,
-      "learning_rate": 9.972e-07,
-      "loss": 10.68586669921875,
+      "epoch": 2.7227722772277225,
+      "grad_norm": 31.9982967376709,
+      "learning_rate": 9.999806557001091e-07,
+      "loss": 5.38673828125,
       "step": 550
     },
     {
-      "epoch": 2.847058823529412,
-      "grad_norm": 33.349178314208984,
-      "learning_rate": 9.957714285714285e-07,
-      "loss": 10.5280859375,
+      "epoch": 2.8465346534653464,
+      "grad_norm": 19.33993148803711,
+      "learning_rate": 9.999558815029513e-07,
+      "loss": 5.291504516601562,
       "step": 575
     },
     {
-      "epoch": 2.9708978328173377,
-      "grad_norm": 40.658817291259766,
-      "learning_rate": 9.94342857142857e-07,
-      "loss": 10.160640258789062,
+      "epoch": 2.9702970297029703,
+      "grad_norm": 18.838518142700195,
+      "learning_rate": 9.999210371864342e-07,
+      "loss": 5.345247192382812,
       "step": 600
     },
     {
-      "epoch": 3.0941176470588236,
-      "grad_norm": 35.320701599121094,
-      "learning_rate": 9.929142857142856e-07,
-      "loss": 10.215652465820312,
+      "epoch": 3.094059405940594,
+      "grad_norm": 19.941740036010742,
+      "learning_rate": 9.998761234523925e-07,
+      "loss": 5.081980285644531,
       "step": 625
     },
     {
-      "epoch": 3.2179566563467494,
-      "grad_norm": 31.319137573242188,
-      "learning_rate": 9.914857142857142e-07,
-      "loss": 10.158941650390625,
+      "epoch": 3.217821782178218,
+      "grad_norm": 24.84575080871582,
+      "learning_rate": 9.998211412054793e-07,
+      "loss": 5.025005493164063,
       "step": 650
     },
     {
-      "epoch": 3.341795665634675,
-      "grad_norm": 28.438013076782227,
-      "learning_rate": 9.900571428571427e-07,
-      "loss": 9.933291625976562,
+      "epoch": 3.3415841584158414,
+      "grad_norm": 24.532045364379883,
+      "learning_rate": 9.99756091553148e-07,
+      "loss": 5.004798889160156,
       "step": 675
     },
     {
-      "epoch": 3.4656346749226006,
-      "grad_norm": 41.65836715698242,
-      "learning_rate": 9.886285714285713e-07,
-      "loss": 9.679817504882813,
+      "epoch": 3.4653465346534653,
+      "grad_norm": 21.20513916015625,
+      "learning_rate": 9.996809758056298e-07,
+      "loss": 4.870877685546875,
       "step": 700
     },
     {
-      "epoch": 3.5894736842105264,
-      "grad_norm": 49.5134162902832,
-      "learning_rate": 9.871999999999998e-07,
-      "loss": 9.66293212890625,
+      "epoch": 3.589108910891089,
+      "grad_norm": 16.589248657226562,
+      "learning_rate": 9.995957954759072e-07,
+      "loss": 4.95923095703125,
       "step": 725
     },
     {
-      "epoch": 3.713312693498452,
-      "grad_norm": 37.431331634521484,
-      "learning_rate": 9.857714285714284e-07,
-      "loss": 9.642633056640625,
+      "epoch": 3.7128712871287126,
+      "grad_norm": 26.863059997558594,
+      "learning_rate": 9.995005522796837e-07,
+      "loss": 4.81072265625,
       "step": 750
     },
     {
-      "epoch": 3.837151702786378,
-      "grad_norm": 39.974544525146484,
-      "learning_rate": 9.843428571428572e-07,
-      "loss": 9.296731567382812,
+      "epoch": 3.8366336633663365,
+      "grad_norm": 14.973981857299805,
+      "learning_rate": 9.993952481353493e-07,
+      "loss": 4.8048876953125,
       "step": 775
     },
     {
-      "epoch": 3.9609907120743033,
-      "grad_norm": 39.033599853515625,
-      "learning_rate": 9.829142857142857e-07,
-      "loss": 9.472877807617188,
+      "epoch": 3.9603960396039604,
+      "grad_norm": 18.018110275268555,
+      "learning_rate": 9.992798851639418e-07,
+      "loss": 4.770014343261718,
       "step": 800
     },
     {
-      "epoch": 4.08421052631579,
-      "grad_norm": 29.367778778076172,
-      "learning_rate": 9.814857142857143e-07,
-      "loss": 9.286629028320313,
+      "epoch": 4.084158415841584,
+      "grad_norm": 15.32919979095459,
+      "learning_rate": 9.991544656891043e-07,
+      "loss": 4.663901672363282,
       "step": 825
     },
     {
-      "epoch": 4.208049535603715,
-      "grad_norm": 30.283451080322266,
-      "learning_rate": 9.800571428571428e-07,
-      "loss": 9.21810546875,
+      "epoch": 4.207920792079208,
+      "grad_norm": 14.925799369812012,
+      "learning_rate": 9.990189922370373e-07,
+      "loss": 4.614352111816406,
       "step": 850
     },
     {
-      "epoch": 4.33188854489164,
-      "grad_norm": 43.98243713378906,
-      "learning_rate": 9.786285714285714e-07,
-      "loss": 9.4026904296875,
+      "epoch": 4.3316831683168315,
+      "grad_norm": 14.5320405960083,
+      "learning_rate": 9.988734675364496e-07,
+      "loss": 4.607881774902344,
       "step": 875
     },
     {
-      "epoch": 4.455727554179567,
-      "grad_norm": 35.94754409790039,
-      "learning_rate": 9.772e-07,
-      "loss": 9.243385009765625,
+      "epoch": 4.455445544554456,
+      "grad_norm": 17.06329345703125,
+      "learning_rate": 9.987178945185019e-07,
+      "loss": 4.582254028320312,
       "step": 900
     },
     {
-      "epoch": 4.579566563467492,
-      "grad_norm": 97.34254455566406,
-      "learning_rate": 9.757714285714285e-07,
-      "loss": 9.081926879882813,
+      "epoch": 4.579207920792079,
+      "grad_norm": 16.860803604125977,
+      "learning_rate": 9.985522763167486e-07,
+      "loss": 4.6025848388671875,
       "step": 925
     },
     {
-      "epoch": 4.703405572755418,
-      "grad_norm": 35.22529602050781,
-      "learning_rate": 9.74342857142857e-07,
-      "loss": 9.177518310546875,
+      "epoch": 4.702970297029703,
+      "grad_norm": 12.743483543395996,
+      "learning_rate": 9.983766162670739e-07,
+      "loss": 4.603981628417968,
       "step": 950
     },
     {
-      "epoch": 4.827244582043344,
-      "grad_norm": 31.761302947998047,
-      "learning_rate": 9.729142857142856e-07,
-      "loss": 9.266145629882812,
+      "epoch": 4.826732673267327,
+      "grad_norm": 11.648216247558594,
+      "learning_rate": 9.981909179076255e-07,
+      "loss": 4.5333544921875,
       "step": 975
     },
     {
-      "epoch": 4.951083591331269,
-      "grad_norm": 34.29395294189453,
-      "learning_rate": 9.714857142857142e-07,
-      "loss": 9.146566162109375,
+      "epoch": 4.9504950495049505,
+      "grad_norm": 12.38879108428955,
+      "learning_rate": 9.979951849787429e-07,
+      "loss": 4.538429565429688,
       "step": 1000
     },
     {
-      "epoch": 4.951083591331269,
-      "eval_loss": 1.01394784450531,
-      "eval_runtime": 78.015,
-      "eval_samples_per_second": 20.714,
-      "eval_steps_per_second": 5.178,
+      "epoch": 4.9504950495049505,
+      "eval_loss": 1.0112695693969727,
+      "eval_runtime": 82.132,
+      "eval_samples_per_second": 19.676,
+      "eval_steps_per_second": 4.919,
       "step": 1000
     },
     {
-      "epoch": 5.074303405572755,
-      "grad_norm": 28.847089767456055,
-      "learning_rate": 9.700571428571427e-07,
-      "loss": 9.07594970703125,
+      "epoch": 5.074257425742574,
+      "grad_norm": 39.850196838378906,
+      "learning_rate": 9.977894214228815e-07,
+      "loss": 4.55607666015625,
       "step": 1025
     },
     {
-      "epoch": 5.198142414860681,
-      "grad_norm": 31.509410858154297,
-      "learning_rate": 9.686285714285713e-07,
-      "loss": 9.165030517578124,
+      "epoch": 5.198019801980198,
+      "grad_norm": 13.423649787902832,
+      "learning_rate": 9.975736313845345e-07,
+      "loss": 4.43293212890625,
       "step": 1050
     },
     {
-      "epoch": 5.321981424148607,
-      "grad_norm": 28.566282272338867,
-      "learning_rate": 9.671999999999998e-07,
-      "loss": 8.919168701171875,
+      "epoch": 5.321782178217822,
+      "grad_norm": 15.34720516204834,
+      "learning_rate": 9.97347819210148e-07,
+      "loss": 4.4848849487304685,
       "step": 1075
     },
     {
-      "epoch": 5.445820433436532,
-      "grad_norm": 26.087688446044922,
-      "learning_rate": 9.657714285714286e-07,
-      "loss": 8.877406005859376,
+      "epoch": 5.445544554455446,
+      "grad_norm": 10.895915031433105,
+      "learning_rate": 9.971119894480344e-07,
+      "loss": 4.418946228027344,
       "step": 1100
     },
     {
-      "epoch": 5.569659442724459,
-      "grad_norm": 27.453046798706055,
-      "learning_rate": 9.643428571428571e-07,
-      "loss": 8.87409912109375,
+      "epoch": 5.569306930693069,
+      "grad_norm": 13.77745532989502,
+      "learning_rate": 9.96866146848281e-07,
+      "loss": 4.675699768066406,
       "step": 1125
     },
     {
-      "epoch": 5.693498452012384,
-      "grad_norm": 26.87615203857422,
-      "learning_rate": 9.629142857142857e-07,
-      "loss": 8.98948486328125,
+      "epoch": 5.693069306930693,
+      "grad_norm": 21.53974723815918,
+      "learning_rate": 9.96610296362653e-07,
+      "loss": 4.483356628417969,
       "step": 1150
     },
     {
-      "epoch": 5.817337461300309,
-      "grad_norm": 28.69729232788086,
-      "learning_rate": 9.614857142857143e-07,
-      "loss": 8.854622192382813,
+      "epoch": 5.816831683168317,
+      "grad_norm": 14.47573184967041,
+      "learning_rate": 9.963444431444948e-07,
+      "loss": 4.555756530761719,
       "step": 1175
     },
     {
-      "epoch": 5.9411764705882355,
-      "grad_norm": 31.937828063964844,
-      "learning_rate": 9.600571428571428e-07,
-      "loss": 9.01401123046875,
+      "epoch": 5.9405940594059405,
+      "grad_norm": 11.701032638549805,
+      "learning_rate": 9.960685925486265e-07,
+      "loss": 4.419072265625,
       "step": 1200
     },
     {
-      "epoch": 6.064396284829721,
-      "grad_norm": 42.471954345703125,
-      "learning_rate": 9.586285714285714e-07,
-      "loss": 8.731229858398438,
+      "epoch": 6.064356435643564,
+      "grad_norm": 17.197729110717773,
+      "learning_rate": 9.957827501312352e-07,
+      "loss": 4.375239562988281,
       "step": 1225
     },
     {
-      "epoch": 6.188235294117647,
-      "grad_norm": 23.436538696289062,
-      "learning_rate": 9.572e-07,
-      "loss": 8.83049560546875,
+      "epoch": 6.188118811881188,
+      "grad_norm": 12.936083793640137,
+      "learning_rate": 9.954869216497633e-07,
+      "loss": 4.3699853515625,
       "step": 1250
     },
     {
-      "epoch": 6.312074303405573,
-      "grad_norm": 36.96808624267578,
-      "learning_rate": 9.557714285714285e-07,
-      "loss": 8.846947631835938,
+      "epoch": 6.311881188118812,
+      "grad_norm": 15.114112854003906,
+      "learning_rate": 9.95181113062793e-07,
+      "loss": 4.41375732421875,
       "step": 1275
     },
     {
-      "epoch": 6.435913312693499,
-      "grad_norm": 34.232627868652344,
-      "learning_rate": 9.54342857142857e-07,
-      "loss": 8.8575048828125,
+      "epoch": 6.435643564356436,
+      "grad_norm": 17.72418785095215,
+      "learning_rate": 9.948653305299256e-07,
+      "loss": 4.430754089355469,
       "step": 1300
     },
     {
-      "epoch": 6.559752321981424,
-      "grad_norm": 25.819372177124023,
-      "learning_rate": 9.529142857142857e-07,
-      "loss": 8.967839965820312,
+      "epoch": 6.5594059405940595,
+      "grad_norm": 18.342178344726562,
+      "learning_rate": 9.945395804116578e-07,
+      "loss": 4.4040380859375,
       "step": 1325
     },
     {
-      "epoch": 6.68359133126935,
-      "grad_norm": 30.442367553710938,
-      "learning_rate": 9.514857142857142e-07,
-      "loss": 8.79695556640625,
+      "epoch": 6.683168316831683,
+      "grad_norm": 12.649271965026855,
+      "learning_rate": 9.94203869269254e-07,
+      "loss": 4.3536474609375,
       "step": 1350
     },
     {
-      "epoch": 6.807430340557276,
-      "grad_norm": 25.49448585510254,
-      "learning_rate": 9.500571428571428e-07,
-      "loss": 8.629199829101562,
+      "epoch": 6.806930693069307,
+      "grad_norm": 32.92485809326172,
+      "learning_rate": 9.938582038646131e-07,
+      "loss": 4.349139709472656,
       "step": 1375
     },
     {
-      "epoch": 6.931269349845201,
-      "grad_norm": 50.014862060546875,
-      "learning_rate": 9.486285714285714e-07,
-      "loss": 8.7688330078125,
+      "epoch": 6.930693069306931,
+      "grad_norm": 11.99206829071045,
+      "learning_rate": 9.935025911601334e-07,
+      "loss": 4.317200317382812,
       "step": 1400
     },
     {
-      "epoch": 7.054489164086688,
-      "grad_norm": 29.534305572509766,
-      "learning_rate": 9.472e-07,
-      "loss": 8.635714721679687,
+      "epoch": 7.054455445544554,
+      "grad_norm": 15.798395156860352,
+      "learning_rate": 9.931370383185716e-07,
+      "loss": 4.390576782226563,
       "step": 1425
     },
     {
-      "epoch": 7.178328173374613,
-      "grad_norm": 24.548370361328125,
-      "learning_rate": 9.457714285714286e-07,
-      "loss": 8.751875,
+      "epoch": 7.178217821782178,
+      "grad_norm": 12.485000610351562,
+      "learning_rate": 9.92761552702899e-07,
+      "loss": 4.415244750976562,
       "step": 1450
     },
     {
-      "epoch": 7.302167182662538,
-      "grad_norm": 27.02566146850586,
-      "learning_rate": 9.443428571428571e-07,
-      "loss": 8.664575805664063,
+      "epoch": 7.301980198019802,
+      "grad_norm": 14.873079299926758,
+      "learning_rate": 9.923761418761526e-07,
+      "loss": 4.3524868774414065,
       "step": 1475
     },
     {
-      "epoch": 7.4260061919504645,
-      "grad_norm": 26.448801040649414,
-      "learning_rate": 9.429142857142857e-07,
-      "loss": 8.631251831054687,
+      "epoch": 7.425742574257426,
+      "grad_norm": 13.601946830749512,
+      "learning_rate": 9.919808136012837e-07,
+      "loss": 4.294336547851563,
       "step": 1500
     },
     {
-      "epoch": 7.4260061919504645,
-      "eval_loss": 0.9785507321357727,
-      "eval_runtime": 78.788,
-      "eval_samples_per_second": 20.511,
-      "eval_steps_per_second": 5.128,
+      "epoch": 7.425742574257426,
+      "eval_loss": 0.9704130291938782,
+      "eval_runtime": 86.4979,
+      "eval_samples_per_second": 18.683,
+      "eval_steps_per_second": 4.671,
       "step": 1500
     },
     {
-      "epoch": 7.54984520123839,
-      "grad_norm": 26.66581916809082,
-      "learning_rate": 9.414857142857142e-07,
-      "loss": 8.60765869140625,
+      "epoch": 7.5495049504950495,
+      "grad_norm": 15.53011417388916,
+      "learning_rate": 9.915755758410008e-07,
+      "loss": 4.254639587402344,
       "step": 1525
     },
     {
-      "epoch": 7.673684210526316,
-      "grad_norm": 24.334070205688477,
-      "learning_rate": 9.400571428571428e-07,
-      "loss": 8.694551391601562,
+      "epoch": 7.673267326732673,
+      "grad_norm": 11.795248985290527,
+      "learning_rate": 9.91160436757609e-07,
+      "loss": 4.307816467285156,
       "step": 1550
     },
     {
-      "epoch": 7.7975232198142415,
-      "grad_norm": 25.746564865112305,
-      "learning_rate": 9.386285714285714e-07,
-      "loss": 8.605336303710937,
+      "epoch": 7.797029702970297,
+      "grad_norm": 15.00704574584961,
+      "learning_rate": 9.907354047128464e-07,
+      "loss": 4.349948120117188,
       "step": 1575
     },
     {
-      "epoch": 7.921362229102167,
-      "grad_norm": 26.221471786499023,
-      "learning_rate": 9.372e-07,
-      "loss": 8.670101928710938,
+      "epoch": 7.920792079207921,
+      "grad_norm": 13.17168140411377,
+      "learning_rate": 9.903004882677155e-07,
+      "loss": 4.2257516479492185,
       "step": 1600
     },
     {
-      "epoch": 8.044582043343654,
-      "grad_norm": 40.2961311340332,
-      "learning_rate": 9.357714285714286e-07,
-      "loss": 8.697719116210937,
+      "epoch": 8.044554455445544,
+      "grad_norm": 13.770243644714355,
+      "learning_rate": 9.898556961823103e-07,
+      "loss": 4.310787048339844,
       "step": 1625
     },
     {
-      "epoch": 8.16842105263158,
-      "grad_norm": 29.41912078857422,
-      "learning_rate": 9.343428571428571e-07,
-      "loss": 8.6234228515625,
+      "epoch": 8.168316831683168,
+      "grad_norm": 13.734089851379395,
+      "learning_rate": 9.894010374156398e-07,
+      "loss": 4.148223266601563,
       "step": 1650
     },
     {
-      "epoch": 8.292260061919505,
-      "grad_norm": 23.59851837158203,
-      "learning_rate": 9.329142857142857e-07,
-      "loss": 8.555197143554688,
+      "epoch": 8.292079207920793,
+      "grad_norm": 15.90095329284668,
+      "learning_rate": 9.889365211254485e-07,
+      "loss": 4.3510107421875,
       "step": 1675
     },
     {
-      "epoch": 8.41609907120743,
-      "grad_norm": 23.335365295410156,
-      "learning_rate": 9.314857142857142e-07,
-      "loss": 8.533560180664063,
+      "epoch": 8.415841584158416,
+      "grad_norm": 12.133323669433594,
+      "learning_rate": 9.884621566680306e-07,
+      "loss": 4.337423400878906,
       "step": 1700
     },
     {
-      "epoch": 8.539938080495356,
-      "grad_norm": 19.63068389892578,
-      "learning_rate": 9.300571428571428e-07,
-      "loss": 8.534976196289062,
+      "epoch": 8.53960396039604,
+      "grad_norm": 50.35639190673828,
+      "learning_rate": 9.87977953598043e-07,
+      "loss": 4.3044869995117185,
       "step": 1725
     },
     {
-      "epoch": 8.66377708978328,
-      "grad_norm": 26.558069229125977,
-      "learning_rate": 9.286285714285714e-07,
-      "loss": 8.529254150390624,
+      "epoch": 8.663366336633663,
+      "grad_norm": 12.457736015319824,
+      "learning_rate": 9.874839216683112e-07,
+      "loss": 4.3085205078125,
       "step": 1750
     },
     {
-      "epoch": 8.787616099071208,
-      "grad_norm": 28.553796768188477,
-      "learning_rate": 9.272e-07,
-      "loss": 8.4541943359375,
+      "epoch": 8.787128712871286,
+      "grad_norm": 14.483731269836426,
+      "learning_rate": 9.869800708296346e-07,
+      "loss": 4.1831143188476565,
       "step": 1775
     },
     {
-      "epoch": 8.911455108359133,
-      "grad_norm": 26.726491928100586,
-      "learning_rate": 9.257714285714285e-07,
-      "loss": 8.505891723632812,
+      "epoch": 8.910891089108912,
+      "grad_norm": 11.174741744995117,
+      "learning_rate": 9.864664112305846e-07,
+      "loss": 4.162017822265625,
       "step": 1800
     },
     {
-      "epoch": 9.034674922600619,
-      "grad_norm": 33.665992736816406,
-      "learning_rate": 9.243428571428571e-07,
-      "loss": 8.420493774414062,
+      "epoch": 9.034653465346535,
+      "grad_norm": 10.312405586242676,
+      "learning_rate": 9.859429532173012e-07,
+      "loss": 4.320256958007812,
       "step": 1825
     },
     {
-      "epoch": 9.158513931888544,
-      "grad_norm": 26.62388801574707,
-      "learning_rate": 9.229142857142857e-07,
-      "loss": 8.4884375,
+      "epoch": 9.158415841584159,
+      "grad_norm": 13.814125061035156,
+      "learning_rate": 9.85409707333284e-07,
+      "loss": 4.314511413574219,
       "step": 1850
     },
     {
-      "epoch": 9.282352941176471,
-      "grad_norm": 21.369853973388672,
-      "learning_rate": 9.214857142857142e-07,
-      "loss": 8.413807983398437,
+      "epoch": 9.282178217821782,
+      "grad_norm": 11.276887893676758,
+      "learning_rate": 9.848666843191804e-07,
+      "loss": 4.327219543457031,
       "step": 1875
     },
     {
-      "epoch": 9.406191950464397,
-      "grad_norm": 23.232192993164062,
-      "learning_rate": 9.200571428571429e-07,
-      "loss": 8.473324584960938,
+      "epoch": 9.405940594059405,
+      "grad_norm": 13.48475170135498,
+      "learning_rate": 9.84313895112568e-07,
+      "loss": 4.2580517578125,
       "step": 1900
     },
     {
-      "epoch": 9.530030959752322,
-      "grad_norm": 48.790531158447266,
-      "learning_rate": 9.186285714285714e-07,
-      "loss": 8.452188720703125,
+      "epoch": 9.52970297029703,
+      "grad_norm": 13.70500659942627,
+      "learning_rate": 9.837513508477365e-07,
+      "loss": 4.269812316894531,
       "step": 1925
     },
     {
-      "epoch": 9.653869969040247,
-      "grad_norm": 23.798295974731445,
-      "learning_rate": 9.172e-07,
-      "loss": 8.495001831054687,
+      "epoch": 9.653465346534654,
+      "grad_norm": 12.701525688171387,
+      "learning_rate": 9.831790628554612e-07,
+      "loss": 4.257364196777344,
       "step": 1950
     },
     {
-      "epoch": 9.777708978328173,
-      "grad_norm": 33.404537200927734,
-      "learning_rate": 9.157714285714285e-07,
-      "loss": 8.58928955078125,
+      "epoch": 9.777227722772277,
+      "grad_norm": 11.086860656738281,
+      "learning_rate": 9.82597042662776e-07,
+      "loss": 4.120067443847656,
       "step": 1975
     },
     {
-      "epoch": 9.9015479876161,
-      "grad_norm": 30.602357864379883,
-      "learning_rate": 9.143428571428571e-07,
-      "loss": 8.496845703125,
+      "epoch": 9.900990099009901,
+      "grad_norm": 14.811179161071777,
+      "learning_rate": 9.820053019927408e-07,
+      "loss": 4.13207275390625,
       "step": 2000
     },
     {
-      "epoch": 9.9015479876161,
-      "eval_loss": 0.9522278904914856,
-      "eval_runtime": 78.3957,
-      "eval_samples_per_second": 20.613,
-      "eval_steps_per_second": 5.153,
+      "epoch": 9.900990099009901,
+      "eval_loss": 0.9480290412902832,
+      "eval_runtime": 81.9519,
+      "eval_samples_per_second": 19.719,
+      "eval_steps_per_second": 4.93,
       "step": 2000
     },
     {
-      "epoch": 10.024767801857585,
-      "grad_norm": 28.26131248474121,
-      "learning_rate": 9.129142857142856e-07,
-      "loss": 8.348336181640626,
+      "epoch": 10.024752475247524,
+      "grad_norm": 13.266257286071777,
+      "learning_rate": 9.814038527642059e-07,
+      "loss": 4.281947021484375,
       "step": 2025
     },
     {
-      "epoch": 10.14860681114551,
-      "grad_norm": 27.35468864440918,
-      "learning_rate": 9.114857142857142e-07,
-      "loss": 8.510921020507812,
+      "epoch": 10.148514851485148,
+      "grad_norm": 13.618546485900879,
+      "learning_rate": 9.80792707091571e-07,
+      "loss": 4.166450500488281,
       "step": 2050
     },
     {
-      "epoch": 10.272445820433436,
-      "grad_norm": 21.89419174194336,
-      "learning_rate": 9.100571428571429e-07,
-      "loss": 8.4816162109375,
+      "epoch": 10.272277227722773,
+      "grad_norm": 12.208344459533691,
+      "learning_rate": 9.801718772845422e-07,
+      "loss": 4.151670227050781,
       "step": 2075
     },
     {
-      "epoch": 10.396284829721361,
-      "grad_norm": 24.475194931030273,
-      "learning_rate": 9.086285714285714e-07,
-      "loss": 8.524483032226563,
+      "epoch": 10.396039603960396,
+      "grad_norm": 12.220300674438477,
+      "learning_rate": 9.795413758478835e-07,
+      "loss": 4.250581359863281,
       "step": 2100
     },
     {
-      "epoch": 10.520123839009289,
-      "grad_norm": 28.319612503051758,
-      "learning_rate": 9.072e-07,
-      "loss": 8.434509887695313,
+      "epoch": 10.51980198019802,
+      "grad_norm": 10.937910079956055,
+      "learning_rate": 9.789012154811646e-07,
+      "loss": 4.219160461425782,
       "step": 2125
     },
     {
-      "epoch": 10.643962848297214,
-      "grad_norm": 22.886144638061523,
-      "learning_rate": 9.057714285714285e-07,
-      "loss": 8.604608764648438,
+      "epoch": 10.643564356435643,
+      "grad_norm": 12.94577407836914,
+      "learning_rate": 9.78251409078506e-07,
+      "loss": 4.256142272949218,
       "step": 2150
     },
     {
-      "epoch": 10.76780185758514,
-      "grad_norm": 31.52608871459961,
-      "learning_rate": 9.043428571428571e-07,
-      "loss": 8.302666015625,
+      "epoch": 10.767326732673267,
+      "grad_norm": 12.101743698120117,
+      "learning_rate": 9.77591969728319e-07,
+      "loss": 4.145942077636719,
       "step": 2175
     },
     {
-      "epoch": 10.891640866873065,
-      "grad_norm": 23.449596405029297,
-      "learning_rate": 9.029142857142856e-07,
-      "loss": 8.249719848632813,
+      "epoch": 10.891089108910892,
+      "grad_norm": 15.079996109008789,
+      "learning_rate": 9.769229107130408e-07,
+      "loss": 4.2266848754882815,
       "step": 2200
     },
     {
-      "epoch": 11.014860681114552,
-      "grad_norm": 26.614627838134766,
-      "learning_rate": 9.014857142857143e-07,
-      "loss": 8.312252807617188,
+      "epoch": 11.014851485148515,
+      "grad_norm": 13.785998344421387,
+      "learning_rate": 9.762442455088694e-07,
+      "loss": 4.173150634765625,
       "step": 2225
     },
     {
-      "epoch": 11.138699690402477,
-      "grad_norm": 25.212867736816406,
-      "learning_rate": 9.000571428571429e-07,
-      "loss": 8.25859130859375,
+      "epoch": 11.138613861386139,
+      "grad_norm": 33.18384552001953,
+      "learning_rate": 9.7555598778549e-07,
+      "loss": 4.193008728027344,
       "step": 2250
     },
     {
-      "epoch": 11.262538699690403,
-      "grad_norm": 27.4830265045166,
-      "learning_rate": 8.986285714285714e-07,
-      "loss": 8.3023095703125,
+      "epoch": 11.262376237623762,
+      "grad_norm": 17.419424057006836,
+      "learning_rate": 9.748581514058005e-07,
+      "loss": 4.254207153320312,
       "step": 2275
     },
     {
-      "epoch": 11.386377708978328,
-      "grad_norm": 23.062416076660156,
-      "learning_rate": 8.972e-07,
-      "loss": 8.250548095703126,
+      "epoch": 11.386138613861386,
+      "grad_norm": 11.025071144104004,
+      "learning_rate": 9.741507504256326e-07,
+      "loss": 4.130965270996094,
       "step": 2300
     },
     {
-      "epoch": 11.510216718266253,
-      "grad_norm": 26.304353713989258,
-      "learning_rate": 8.957714285714285e-07,
-      "loss": 8.221282958984375,
+      "epoch": 11.509900990099009,
+      "grad_norm": 11.500884056091309,
+      "learning_rate": 9.734337990934681e-07,
+      "loss": 4.12019775390625,
       "step": 2325
     },
     {
-      "epoch": 11.63405572755418,
-      "grad_norm": 30.5177001953125,
-      "learning_rate": 8.943428571428571e-07,
-      "loss": 8.381609497070313,
+      "epoch": 11.633663366336634,
+      "grad_norm": 13.27956771850586,
+      "learning_rate": 9.727073118501523e-07,
+      "loss": 4.216612854003906,
       "step": 2350
     },
     {
-      "epoch": 11.757894736842106,
-      "grad_norm": 23.5266056060791,
-      "learning_rate": 8.929142857142856e-07,
-      "loss": 8.32532958984375,
+      "epoch": 11.757425742574258,
+      "grad_norm": 16.143634796142578,
+      "learning_rate": 9.719713033286028e-07,
+      "loss": 4.163922424316406,
       "step": 2375
     },
     {
-      "epoch": 11.881733746130031,
-      "grad_norm": 24.466337203979492,
-      "learning_rate": 8.914857142857143e-07,
-      "loss": 8.319666137695313,
+      "epoch": 11.881188118811881,
+      "grad_norm": 14.004642486572266,
+      "learning_rate": 9.712257883535152e-07,
+      "loss": 4.1760952758789065,
       "step": 2400
     },
     {
-      "epoch": 12.004953560371517,
-      "grad_norm": 28.647558212280273,
-      "learning_rate": 8.900571428571428e-07,
-      "loss": 8.305534057617187,
+      "epoch": 12.004950495049505,
+      "grad_norm": 11.351595878601074,
+      "learning_rate": 9.704707819410634e-07,
+      "loss": 4.229373168945313,
       "step": 2425
     },
     {
-      "epoch": 12.128792569659442,
-      "grad_norm": 25.798559188842773,
-      "learning_rate": 8.886285714285714e-07,
-      "loss": 8.325819091796875,
+      "epoch": 12.128712871287128,
+      "grad_norm": 9.855143547058105,
+      "learning_rate": 9.697062992985993e-07,
+      "loss": 4.096666259765625,
       "step": 2450
     },
     {
-      "epoch": 12.25263157894737,
-      "grad_norm": 23.050085067749023,
-      "learning_rate": 8.872e-07,
-      "loss": 8.418189086914062,
+      "epoch": 12.252475247524753,
+      "grad_norm": 11.369580268859863,
+      "learning_rate": 9.689323558243444e-07,
+      "loss": 4.218254089355469,
       "step": 2475
     },
     {
-      "epoch": 12.376470588235295,
-      "grad_norm": 45.27325439453125,
-      "learning_rate": 8.857714285714285e-07,
-      "loss": 8.232376098632812,
+      "epoch": 12.376237623762377,
+      "grad_norm": 9.929821014404297,
+      "learning_rate": 9.681489671070802e-07,
+      "loss": 4.054186401367187,
       "step": 2500
     },
     {
-      "epoch": 12.376470588235295,
-      "eval_loss": 0.9387099742889404,
-      "eval_runtime": 78.8561,
-      "eval_samples_per_second": 20.493,
-      "eval_steps_per_second": 5.123,
+      "epoch": 12.376237623762377,
+      "eval_loss": 0.9306696653366089,
+      "eval_runtime": 82.6001,
+      "eval_samples_per_second": 19.564,
+      "eval_steps_per_second": 4.891,
       "step": 2500
     },
     {
-      "epoch": 12.50030959752322,
-      "grad_norm": 30.00120735168457,
-      "learning_rate": 8.843428571428571e-07,
-      "loss": 8.216619262695312,
+      "epoch": 12.5,
+      "grad_norm": 23.94037628173828,
+      "learning_rate": 9.673561489258352e-07,
+      "loss": 4.108751220703125,
       "step": 2525
     },
     {
-      "epoch": 12.624148606811145,
-      "grad_norm": 20.330636978149414,
-      "learning_rate": 8.829142857142857e-07,
-      "loss": 8.305621948242187,
+      "epoch": 12.623762376237623,
+      "grad_norm": 10.571653366088867,
+      "learning_rate": 9.665539172495656e-07,
+      "loss": 4.077666320800781,
       "step": 2550
     },
     {
-      "epoch": 12.74798761609907,
-      "grad_norm": 23.445192337036133,
-      "learning_rate": 8.814857142857143e-07,
-      "loss": 8.283675537109374,
+      "epoch": 12.747524752475247,
+      "grad_norm": 12.341602325439453,
+      "learning_rate": 9.657422882368347e-07,
+      "loss": 4.187429504394531,
       "step": 2575
     },
     {
-      "epoch": 12.871826625386998,
-      "grad_norm": 19.532381057739258,
-      "learning_rate": 8.800571428571428e-07,
-      "loss": 8.197587890625,
+      "epoch": 12.871287128712872,
+      "grad_norm": 9.840188026428223,
+      "learning_rate": 9.64921278235487e-07,
+      "loss": 4.135216369628906,
       "step": 2600
     },
     {
-      "epoch": 12.995665634674923,
-      "grad_norm": 24.901432037353516,
-      "learning_rate": 8.786285714285714e-07,
-      "loss": 8.2815869140625,
+      "epoch": 12.995049504950495,
+      "grad_norm": 11.124446868896484,
+      "learning_rate": 9.640909037823194e-07,
+      "loss": 4.087601318359375,
       "step": 2625
     },
     {
-      "epoch": 13.118885448916409,
-      "grad_norm": 23.698482513427734,
-      "learning_rate": 8.771999999999999e-07,
-      "loss": 8.184017333984375,
+      "epoch": 13.118811881188119,
+      "grad_norm": 11.180383682250977,
+      "learning_rate": 9.63251181602747e-07,
+      "loss": 4.0938818359375,
       "step": 2650
     },
     {
-      "epoch": 13.242724458204334,
-      "grad_norm": 22.537649154663086,
-      "learning_rate": 8.757714285714285e-07,
-      "loss": 8.187122802734375,
+      "epoch": 13.242574257425742,
+      "grad_norm": 12.539691925048828,
+      "learning_rate": 9.624021286104674e-07,
+      "loss": 4.085841369628906,
       "step": 2675
     },
     {
-      "epoch": 13.36656346749226,
-      "grad_norm": 23.588315963745117,
-      "learning_rate": 8.74342857142857e-07,
-      "loss": 8.233043212890625,
+      "epoch": 13.366336633663366,
+      "grad_norm": 11.089624404907227,
+      "learning_rate": 9.615437619071198e-07,
+      "loss": 4.057616577148438,
       "step": 2700
     },
     {
-      "epoch": 13.490402476780186,
-      "grad_norm": 31.038589477539062,
-      "learning_rate": 8.729142857142857e-07,
-      "loss": 8.288897094726563,
+      "epoch": 13.490099009900991,
+      "grad_norm": 10.118108749389648,
+      "learning_rate": 9.6067609878194e-07,
+      "loss": 4.123738403320313,
       "step": 2725
     },
     {
-      "epoch": 13.614241486068112,
-      "grad_norm": 27.92134666442871,
-      "learning_rate": 8.714857142857143e-07,
-      "loss": 8.241436767578126,
+      "epoch": 13.613861386138614,
+      "grad_norm": 13.25112247467041,
+      "learning_rate": 9.597991567114124e-07,
+      "loss": 4.035686645507813,
       "step": 2750
     },
     {
-      "epoch": 13.738080495356037,
-      "grad_norm": 22.192489624023438,
-      "learning_rate": 8.700571428571428e-07,
-      "loss": 8.224349975585938,
+      "epoch": 13.737623762376238,
+      "grad_norm": 10.743168830871582,
+      "learning_rate": 9.589129533589185e-07,
+      "loss": 4.089259643554687,
       "step": 2775
     },
     {
-      "epoch": 13.861919504643962,
-      "grad_norm": 29.820369720458984,
-      "learning_rate": 8.686285714285714e-07,
-      "loss": 8.162620849609375,
+      "epoch": 13.861386138613861,
+      "grad_norm": 13.900023460388184,
+      "learning_rate": 9.580175065743798e-07,
+      "loss": 4.086239929199219,
       "step": 2800
     },
     {
-      "epoch": 13.985758513931888,
-      "grad_norm": 28.781845092773438,
-      "learning_rate": 8.671999999999999e-07,
-      "loss": 8.23837890625,
+      "epoch": 13.985148514851485,
+      "grad_norm": 14.214149475097656,
+      "learning_rate": 9.571128343939005e-07,
+      "loss": 4.133528747558594,
       "step": 2825
     },
     {
-      "epoch": 14.108978328173375,
-      "grad_norm": 21.96494483947754,
-      "learning_rate": 8.657714285714285e-07,
-      "loss": 8.158751831054687,
+      "epoch": 14.108910891089108,
+      "grad_norm": 13.158830642700195,
+      "learning_rate": 9.561989550394016e-07,
+      "loss": 4.137518005371094,
       "step": 2850
     },
     {
-      "epoch": 14.2328173374613,
-      "grad_norm": 22.800256729125977,
-      "learning_rate": 8.643428571428571e-07,
-      "loss": 8.228921508789062,
+      "epoch": 14.232673267326733,
+      "grad_norm": 13.256293296813965,
+      "learning_rate": 9.552758869182557e-07,
+      "loss": 4.049294128417968,
       "step": 2875
     },
     {
-      "epoch": 14.356656346749226,
-      "grad_norm": 23.881921768188477,
-      "learning_rate": 8.629142857142857e-07,
-      "loss": 8.183645629882813,
+      "epoch": 14.356435643564357,
+      "grad_norm": 10.833374977111816,
+      "learning_rate": 9.543436486229161e-07,
+      "loss": 4.0441677856445315,
       "step": 2900
     },
     {
-      "epoch": 14.480495356037151,
-      "grad_norm": 22.399667739868164,
-      "learning_rate": 8.614857142857143e-07,
-      "loss": 8.157160034179688,
+      "epoch": 14.48019801980198,
+      "grad_norm": 12.035900115966797,
+      "learning_rate": 9.534022589305411e-07,
+      "loss": 4.099019775390625,
       "step": 2925
     },
     {
-      "epoch": 14.604334365325077,
-      "grad_norm": 22.374099731445312,
-      "learning_rate": 8.600571428571428e-07,
-      "loss": 8.129403076171876,
+      "epoch": 14.603960396039604,
+      "grad_norm": 11.711991310119629,
+      "learning_rate": 9.524517368026168e-07,
+      "loss": 4.064933166503907,
       "step": 2950
     },
     {
-      "epoch": 14.728173374613004,
-      "grad_norm": 22.495939254760742,
-      "learning_rate": 8.586285714285714e-07,
-      "loss": 8.19878662109375,
+      "epoch": 14.727722772277227,
+      "grad_norm": 20.470388412475586,
+      "learning_rate": 9.514921013845754e-07,
+      "loss": 4.110359802246093,
       "step": 2975
     },
     {
-      "epoch": 14.852012383900929,
-      "grad_norm": 26.343931198120117,
-      "learning_rate": 8.571999999999999e-07,
-      "loss": 8.154603881835937,
+      "epoch": 14.851485148514852,
+      "grad_norm": 12.352797508239746,
+      "learning_rate": 9.505233720054085e-07,
+      "loss": 4.122251892089844,
       "step": 3000
     },
     {
-      "epoch": 14.852012383900929,
-      "eval_loss": 0.9244788289070129,
-      "eval_runtime": 78.6746,
-      "eval_samples_per_second": 20.54,
-      "eval_steps_per_second": 5.135,
+      "epoch": 14.851485148514852,
+      "eval_loss": 0.920803427696228,
+      "eval_runtime": 86.0706,
+      "eval_samples_per_second": 18.775,
+      "eval_steps_per_second": 4.694,
       "step": 3000
     },
     {
-      "epoch": 14.975851393188854,
-      "grad_norm": 20.63393783569336,
-      "learning_rate": 8.557714285714285e-07,
-      "loss": 8.259378051757812,
+      "epoch": 14.975247524752476,
+      "grad_norm": 12.874964714050293,
+      "learning_rate": 9.495455681772791e-07,
+      "loss": 4.0896240234375,
       "step": 3025
     },
     {
-      "epoch": 15.09907120743034,
-      "grad_norm": 28.88397789001465,
-      "learning_rate": 8.543428571428571e-07,
-      "loss": 8.186115112304687,
+      "epoch": 15.099009900990099,
+      "grad_norm": 12.206281661987305,
+      "learning_rate": 9.485587095951275e-07,
+      "loss": 4.052577819824219,
       "step": 3050
     },
     {
-      "epoch": 15.222910216718267,
-      "grad_norm": 25.326343536376953,
-      "learning_rate": 8.529142857142857e-07,
-      "loss": 8.17370849609375,
+      "epoch": 15.222772277227723,
+      "grad_norm": 11.373126983642578,
+      "learning_rate": 9.475628161362747e-07,
+      "loss": 4.107190551757813,
       "step": 3075
     },
     {
-      "epoch": 15.346749226006192,
-      "grad_norm": 22.252532958984375,
-      "learning_rate": 8.514857142857142e-07,
-      "loss": 7.99333984375,
+      "epoch": 15.346534653465346,
+      "grad_norm": 10.882997512817383,
+      "learning_rate": 9.465579078600232e-07,
+      "loss": 4.058759155273438,
       "step": 3100
     },
     {
-      "epoch": 15.470588235294118,
-      "grad_norm": 22.498172760009766,
-      "learning_rate": 8.500571428571428e-07,
-      "loss": 8.131607666015626,
+      "epoch": 15.47029702970297,
+      "grad_norm": 14.810848236083984,
+      "learning_rate": 9.455440050072511e-07,
+      "loss": 4.046248168945312,
       "step": 3125
     },
     {
-      "epoch": 15.594427244582043,
-      "grad_norm": 19.141489028930664,
-      "learning_rate": 8.486285714285714e-07,
-      "loss": 8.157739868164063,
+      "epoch": 15.594059405940595,
+      "grad_norm": 13.74386215209961,
+      "learning_rate": 9.44521128000006e-07,
+      "loss": 4.134948120117188,
       "step": 3150
     },
     {
-      "epoch": 15.718266253869968,
-      "grad_norm": 24.114906311035156,
-      "learning_rate": 8.471999999999999e-07,
-      "loss": 8.153767700195312,
+      "epoch": 15.717821782178218,
+      "grad_norm": 14.098150253295898,
+      "learning_rate": 9.434892974410932e-07,
+      "loss": 3.9966055297851564,
       "step": 3175
     },
     {
-      "epoch": 15.842105263157894,
-      "grad_norm": 24.1041259765625,
-      "learning_rate": 8.457714285714286e-07,
-      "loss": 8.211907348632813,
+      "epoch": 15.841584158415841,
+      "grad_norm": 10.933466911315918,
+      "learning_rate": 9.424485341136599e-07,
+      "loss": 4.099876708984375,
       "step": 3200
     },
     {
-      "epoch": 15.965944272445821,
-      "grad_norm": 26.138748168945312,
-      "learning_rate": 8.443428571428571e-07,
-      "loss": 8.123822631835937,
+      "epoch": 15.965346534653465,
+      "grad_norm": 9.829030990600586,
+      "learning_rate": 9.41398858980778e-07,
+      "loss": 3.9940283203125,
       "step": 3225
     },
     {
-      "epoch": 16.08916408668731,
-      "grad_norm": 19.59873390197754,
-      "learning_rate": 8.429142857142857e-07,
-      "loss": 8.09514404296875,
+      "epoch": 16.08910891089109,
+      "grad_norm": 10.526725769042969,
+      "learning_rate": 9.403402931850207e-07,
+      "loss": 4.016290588378906,
       "step": 3250
     },
     {
-      "epoch": 16.213003095975232,
-      "grad_norm": 23.773271560668945,
-      "learning_rate": 8.414857142857142e-07,
-      "loss": 8.156911010742187,
+      "epoch": 16.212871287128714,
+      "grad_norm": 9.416884422302246,
+      "learning_rate": 9.392728580480371e-07,
+      "loss": 4.113368835449219,
       "step": 3275
     },
     {
-      "epoch": 16.33684210526316,
-      "grad_norm": 25.273815155029297,
-      "learning_rate": 8.400571428571428e-07,
-      "loss": 8.161531982421875,
+      "epoch": 16.336633663366335,
+      "grad_norm": 11.515923500061035,
+      "learning_rate": 9.381965750701228e-07,
+      "loss": 4.077259216308594,
       "step": 3300
     },
     {
-      "epoch": 16.460681114551083,
-      "grad_norm": 22.487651824951172,
-      "learning_rate": 8.386285714285713e-07,
-      "loss": 8.00319091796875,
+      "epoch": 16.46039603960396,
+      "grad_norm": 12.539885520935059,
+      "learning_rate": 9.37111465929787e-07,
+      "loss": 3.9970358276367186,
       "step": 3325
     },
     {
-      "epoch": 16.58452012383901,
-      "grad_norm": 28.419109344482422,
-      "learning_rate": 8.372e-07,
-      "loss": 8.031408081054687,
+      "epoch": 16.584158415841586,
+      "grad_norm": 11.099287986755371,
+      "learning_rate": 9.360175524833151e-07,
+      "loss": 4.0322119140625,
       "step": 3350
     },
     {
-      "epoch": 16.708359133126933,
-      "grad_norm": 21.66465187072754,
-      "learning_rate": 8.357714285714286e-07,
-      "loss": 8.159840087890625,
+      "epoch": 16.707920792079207,
+      "grad_norm": 12.52721118927002,
+      "learning_rate": 9.349148567643294e-07,
+      "loss": 4.074447937011719,
       "step": 3375
     },
     {
-      "epoch": 16.83219814241486,
-      "grad_norm": 23.416608810424805,
-      "learning_rate": 8.343428571428571e-07,
-      "loss": 8.243095092773437,
+      "epoch": 16.831683168316832,
+      "grad_norm": 9.58317756652832,
+      "learning_rate": 9.338034009833445e-07,
+      "loss": 4.086669921875,
       "step": 3400
     },
     {
-      "epoch": 16.956037151702787,
-      "grad_norm": 24.853750228881836,
-      "learning_rate": 8.329142857142857e-07,
-      "loss": 8.11635986328125,
+      "epoch": 16.955445544554454,
+      "grad_norm": 11.126392364501953,
+      "learning_rate": 9.326832075273207e-07,
+      "loss": 4.079499206542969,
       "step": 3425
     },
     {
-      "epoch": 17.07925696594427,
-      "grad_norm": 24.341514587402344,
-      "learning_rate": 8.314857142857142e-07,
-      "loss": 7.91652587890625,
+      "epoch": 17.07920792079208,
+      "grad_norm": 11.195257186889648,
+      "learning_rate": 9.315542989592124e-07,
+      "loss": 4.001406555175781,
       "step": 3450
     },
     {
-      "epoch": 17.2030959752322,
-      "grad_norm": 25.844379425048828,
-      "learning_rate": 8.300571428571428e-07,
-      "loss": 8.211973876953126,
+      "epoch": 17.202970297029704,
+      "grad_norm": 11.249403953552246,
+      "learning_rate": 9.304166980175141e-07,
+      "loss": 4.042621765136719,
       "step": 3475
     },
     {
-      "epoch": 17.326934984520125,
-      "grad_norm": 23.98019790649414,
-      "learning_rate": 8.286285714285713e-07,
-      "loss": 8.040772705078124,
+      "epoch": 17.326732673267326,
+      "grad_norm": 22.27759552001953,
+      "learning_rate": 9.292704276158021e-07,
+      "loss": 4.025510559082031,
       "step": 3500
     },
     {
-      "epoch": 17.326934984520125,
-      "eval_loss": 0.9177140593528748,
-      "eval_runtime": 79.4513,
-      "eval_samples_per_second": 20.34,
-      "eval_steps_per_second": 5.085,
+      "epoch": 17.326732673267326,
+      "eval_loss": 0.9127047657966614,
+      "eval_runtime": 81.0553,
+      "eval_samples_per_second": 19.937,
+      "eval_steps_per_second": 4.984,
       "step": 3500
     },
     {
-      "epoch": 17.45077399380805,
-      "grad_norm": 27.95872688293457,
-      "learning_rate": 8.272e-07,
-      "loss": 7.994537963867187,
+      "epoch": 17.45049504950495,
+      "grad_norm": 10.390947341918945,
+      "learning_rate": 9.281155108422731e-07,
+      "loss": 4.0325146484375,
       "step": 3525
     },
     {
-      "epoch": 17.574613003095976,
-      "grad_norm": 23.433879852294922,
-      "learning_rate": 8.257714285714286e-07,
-      "loss": 8.102257080078125,
+      "epoch": 17.574257425742573,
+      "grad_norm": 11.612289428710938,
+      "learning_rate": 9.269519709592794e-07,
+      "loss": 4.141344909667969,
       "step": 3550
     },
     {
-      "epoch": 17.6984520123839,
-      "grad_norm": 25.75996208190918,
-      "learning_rate": 8.243428571428571e-07,
-      "loss": 8.0616259765625,
+      "epoch": 17.698019801980198,
+      "grad_norm": 15.810525894165039,
+      "learning_rate": 9.257798314028599e-07,
+      "loss": 4.035428771972656,
       "step": 3575
     },
     {
-      "epoch": 17.822291021671827,
-      "grad_norm": 20.315242767333984,
-      "learning_rate": 8.229142857142857e-07,
-      "loss": 8.003837280273437,
+      "epoch": 17.821782178217823,
+      "grad_norm": 17.175111770629883,
+      "learning_rate": 9.245991157822681e-07,
+      "loss": 4.041367492675781,
       "step": 3600
     },
     {
-      "epoch": 17.946130030959754,
-      "grad_norm": 34.34598922729492,
-      "learning_rate": 8.214857142857142e-07,
-      "loss": 8.088245239257812,
+      "epoch": 17.945544554455445,
+      "grad_norm": 10.926711082458496,
+      "learning_rate": 9.234098478794969e-07,
+      "loss": 4.121484069824219,
       "step": 3625
     },
     {
-      "epoch": 18.069349845201238,
-      "grad_norm": 20.82579231262207,
-      "learning_rate": 8.200571428571428e-07,
-      "loss": 8.047254638671875,
+      "epoch": 18.06930693069307,
+      "grad_norm": 11.38853931427002,
+      "learning_rate": 9.222120516487997e-07,
+      "loss": 3.9692178344726563,
       "step": 3650
     },
     {
-      "epoch": 18.193188854489165,
-      "grad_norm": 29.7625732421875,
-      "learning_rate": 8.186285714285714e-07,
-      "loss": 8.1514794921875,
+      "epoch": 18.193069306930692,
+      "grad_norm": 11.953936576843262,
+      "learning_rate": 9.210057512162072e-07,
+      "loss": 4.020705261230469,
       "step": 3675
     },
     {
-      "epoch": 18.31702786377709,
-      "grad_norm": 24.978836059570312,
-      "learning_rate": 8.172e-07,
-      "loss": 7.944334106445313,
+      "epoch": 18.316831683168317,
+      "grad_norm": 13.49695110321045,
+      "learning_rate": 9.197909708790419e-07,
+      "loss": 4.005714721679688,
       "step": 3700
     },
     {
-      "epoch": 18.440866873065016,
-      "grad_norm": 28.676979064941406,
-      "learning_rate": 8.157714285714285e-07,
-      "loss": 8.004066772460938,
+      "epoch": 18.440594059405942,
+      "grad_norm": 12.650299072265625,
+      "learning_rate": 9.185677351054289e-07,
+      "loss": 4.006802062988282,
       "step": 3725
     },
     {
-      "epoch": 18.564705882352943,
-      "grad_norm": 22.64858627319336,
-      "learning_rate": 8.143428571428571e-07,
-      "loss": 8.025421752929688,
+      "epoch": 18.564356435643564,
+      "grad_norm": 14.999792098999023,
+      "learning_rate": 9.173360685338026e-07,
+      "loss": 3.983961486816406,
       "step": 3750
     },
     {
-      "epoch": 18.688544891640866,
-      "grad_norm": 24.17665672302246,
-      "learning_rate": 8.129142857142857e-07,
-      "loss": 8.024356079101562,
+      "epoch": 18.68811881188119,
+      "grad_norm": 8.753050804138184,
+      "learning_rate": 9.160959959724109e-07,
+      "loss": 4.121242980957032,
       "step": 3775
     },
     {
-      "epoch": 18.812383900928793,
-      "grad_norm": 36.6374397277832,
-      "learning_rate": 8.114857142857142e-07,
-      "loss": 8.029654541015624,
+      "epoch": 18.81188118811881,
+      "grad_norm": 11.767355918884277,
+      "learning_rate": 9.148475423988149e-07,
+      "loss": 3.988392333984375,
       "step": 3800
     },
     {
-      "epoch": 18.936222910216717,
-      "grad_norm": 27.306865692138672,
-      "learning_rate": 8.100571428571428e-07,
-      "loss": 8.086602172851563,
+      "epoch": 18.935643564356436,
+      "grad_norm": 13.041923522949219,
+      "learning_rate": 9.135907329593869e-07,
+      "loss": 4.037374877929688,
       "step": 3825
     },
     {
-      "epoch": 19.059442724458204,
-      "grad_norm": 21.691247940063477,
-      "learning_rate": 8.086285714285714e-07,
-      "loss": 7.949559326171875,
+      "epoch": 19.059405940594058,
+      "grad_norm": 17.08255386352539,
+      "learning_rate": 9.123255929688023e-07,
+      "loss": 4.015922241210937,
       "step": 3850
     },
     {
-      "epoch": 19.18328173374613,
-      "grad_norm": 21.185199737548828,
-      "learning_rate": 8.072e-07,
-      "loss": 7.867513427734375,
+      "epoch": 19.183168316831683,
+      "grad_norm": 10.86015796661377,
+      "learning_rate": 9.110521479095312e-07,
+      "loss": 4.031682739257812,
       "step": 3875
     },
     {
-      "epoch": 19.307120743034055,
-      "grad_norm": 20.94526481628418,
-      "learning_rate": 8.057714285714285e-07,
-      "loss": 7.994710693359375,
+      "epoch": 19.306930693069308,
+      "grad_norm": 14.082168579101562,
+      "learning_rate": 9.097704234313243e-07,
+      "loss": 3.973893737792969,
       "step": 3900
     },
     {
-      "epoch": 19.430959752321982,
-      "grad_norm": 20.232423782348633,
-      "learning_rate": 8.043428571428571e-07,
-      "loss": 8.108115234375,
+      "epoch": 19.43069306930693,
+      "grad_norm": 12.722586631774902,
+      "learning_rate": 9.084804453506966e-07,
+      "loss": 4.083378295898438,
       "step": 3925
     },
     {
-      "epoch": 19.554798761609906,
-      "grad_norm": 22.80320930480957,
-      "learning_rate": 8.029142857142856e-07,
-      "loss": 8.067723388671874,
+      "epoch": 19.554455445544555,
+      "grad_norm": 12.841690063476562,
+      "learning_rate": 9.071822396504072e-07,
+      "loss": 4.002433166503907,
       "step": 3950
     },
     {
-      "epoch": 19.678637770897833,
-      "grad_norm": 25.35050392150879,
-      "learning_rate": 8.014857142857142e-07,
-      "loss": 7.935881958007813,
+      "epoch": 19.678217821782177,
+      "grad_norm": 11.057418823242188,
+      "learning_rate": 9.058758324789359e-07,
+      "loss": 4.0087548828125,
       "step": 3975
     },
     {
-      "epoch": 19.80247678018576,
-      "grad_norm": 30.46210479736328,
-      "learning_rate": 8.000571428571429e-07,
-      "loss": 8.018241577148437,
+      "epoch": 19.801980198019802,
+      "grad_norm": 10.621847152709961,
+      "learning_rate": 9.045612501499573e-07,
+      "loss": 3.928231201171875,
       "step": 4000
     },
     {
-      "epoch": 19.80247678018576,
-      "eval_loss": 0.9094438552856445,
-      "eval_runtime": 80.6469,
-      "eval_samples_per_second": 20.038,
-      "eval_steps_per_second": 5.009,
+      "epoch": 19.801980198019802,
+      "eval_loss": 0.9033195376396179,
+      "eval_runtime": 84.5771,
+      "eval_samples_per_second": 19.107,
+      "eval_steps_per_second": 4.777,
       "step": 4000
     }
   ],
@@ -1220,8 +1220,8 @@
       "attributes": {}
     }
   },
-  "total_flos": 1.5554271788757504e+16,
-  "train_batch_size": 4,
+  "total_flos": 1.7164828967143968e+16,
+  "train_batch_size": 8,
   "trial_name": null,
   "trial_params": null
 }

checkpoint-4000/training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bc3ae1f3188fc78f14b1e24c35f0f902ff13c8475d4503f269d542bb33a09e53
+oid sha256:5d34645042a62af83117180e7d8c3a51dc32127e71ed6c053e7766610460e0b1
 size 5393