Training in progress, step 10000, checkpoint

checkpoint-10000/model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ad96b9c160abbb286c9901f6668f5c89fb035af44adc8dec20a236ee5a73aac9
+oid sha256:e702a31647d627c560e91594f22bce35b32778e7635ec5412b5f6343c5638b9b
 size 503128704

checkpoint-10000/optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bb52d7abd4dcfc37ce37351b954e2a755221104864201eeae51a11baf4ab3d6b
+oid sha256:33f4dcba3de2ff71c9068859b580dbcc6faea60dcf354e243c5932e6227e92f9
 size 1006351290

checkpoint-10000/rng_state.pth

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3c8f1b0ec2ac4559b9a7317e5177712fc8549e54b1b53dfcf14de603fbc071d0
+oid sha256:793dbdfe1dd819d2a2df7a014dacf916da5d2986ca0f4bc141c769a4b74f4323
 size 14244

checkpoint-10000/scheduler.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:90c311f0069114271a5dba5d72dcffc82e7a33081dd5f150aa8c9b2278c1c497
+oid sha256:baaddc0dc177f093cb7927eae851400a36fde6f45129bcaf2df443bdde5269f4
 size 1064

checkpoint-10000/trainer_state.json

@@ -1,7 +1,7 @@
 {
-  "best_metric": 3.7523183822631836,
-  "best_model_checkpoint": "/scratch/cl5625/exceptions/models/100M_8397/checkpoint-10000",
-  "epoch": 1.0763104079216446,
+  "best_metric": 3.7442734241485596,
+  "best_model_checkpoint": "/scratch/cl5625/exceptions/models/100M_low_2000_8397/checkpoint-10000",
+  "epoch": 1.0781671159029649,
   "eval_steps": 1000,
   "global_step": 10000,
   "is_hyper_param_search": false,
@@ -9,1498 +9,1498 @@
   "is_world_process_zero": true,
   "log_history": [
     {
-      "epoch": 0.005381552039608223,
-      "grad_norm": 1.4122834205627441,
-      "learning_rate": 0.00028799999999999995,
-      "loss": 8.5357,
+      "epoch": 0.005390835579514825,
+      "grad_norm": 1.5813435316085815,
+      "learning_rate": 0.0003,
+      "loss": 8.6318,
       "step": 50
     },
     {
-      "epoch": 0.010763104079216447,
-      "grad_norm": 3.4872183799743652,
-      "learning_rate": 0.000588,
-      "loss": 6.9207,
+      "epoch": 0.01078167115902965,
+      "grad_norm": 1.0082982778549194,
+      "learning_rate": 0.0006,
+      "loss": 6.8697,
       "step": 100
     },
     {
-      "epoch": 0.01614465611882467,
-      "grad_norm": 2.1926398277282715,
-      "learning_rate": 0.000599689688611141,
-      "loss": 6.4649,
+      "epoch": 0.016172506738544475,
+      "grad_norm": 2.0049550533294678,
+      "learning_rate": 0.0005996762007555315,
+      "loss": 6.4573,
       "step": 150
     },
     {
-      "epoch": 0.021526208158432893,
-      "grad_norm": 1.9389663934707642,
-      "learning_rate": 0.0005993664475810796,
-      "loss": 6.2572,
+      "epoch": 0.0215633423180593,
+      "grad_norm": 2.044867515563965,
+      "learning_rate": 0.000599352401511063,
+      "loss": 6.239,
       "step": 200
     },
     {
-      "epoch": 0.026907760198041114,
-      "grad_norm": 1.5312936305999756,
-      "learning_rate": 0.0005990432065510182,
-      "loss": 6.1012,
+      "epoch": 0.026954177897574125,
+      "grad_norm": 1.7265796661376953,
+      "learning_rate": 0.0005990286022665946,
+      "loss": 6.0844,
       "step": 250
     },
     {
-      "epoch": 0.03228931223764934,
-      "grad_norm": 1.838247299194336,
-      "learning_rate": 0.0005987199655209567,
-      "loss": 5.9633,
+      "epoch": 0.03234501347708895,
+      "grad_norm": 3.774665594100952,
+      "learning_rate": 0.0005987048030221263,
+      "loss": 5.9817,
       "step": 300
     },
     {
-      "epoch": 0.03767086427725756,
-      "grad_norm": 1.8276017904281616,
-      "learning_rate": 0.0005983967244908953,
-      "loss": 5.8677,
+      "epoch": 0.03773584905660377,
+      "grad_norm": 1.1093494892120361,
+      "learning_rate": 0.0005983810037776578,
+      "loss": 5.8816,
       "step": 350
     },
     {
-      "epoch": 0.04305241631686579,
-      "grad_norm": 1.031121015548706,
-      "learning_rate": 0.0005980734834608338,
-      "loss": 5.7804,
+      "epoch": 0.0431266846361186,
+      "grad_norm": 1.401320457458496,
+      "learning_rate": 0.0005980572045331894,
+      "loss": 5.8369,
       "step": 400
     },
     {
-      "epoch": 0.048433968356474004,
-      "grad_norm": 2.2222535610198975,
-      "learning_rate": 0.0005977502424307725,
-      "loss": 5.6924,
+      "epoch": 0.04851752021563342,
+      "grad_norm": 1.4539614915847778,
+      "learning_rate": 0.0005977334052887209,
+      "loss": 5.756,
       "step": 450
     },
     {
-      "epoch": 0.05381552039608223,
-      "grad_norm": 1.698441982269287,
-      "learning_rate": 0.0005974270014007111,
-      "loss": 5.654,
+      "epoch": 0.05390835579514825,
+      "grad_norm": 1.237868309020996,
+      "learning_rate": 0.0005974096060442526,
+      "loss": 5.6613,
       "step": 500
     },
     {
-      "epoch": 0.05919707243569045,
-      "grad_norm": 1.586506962776184,
-      "learning_rate": 0.0005971037603706497,
-      "loss": 5.5769,
+      "epoch": 0.05929919137466307,
+      "grad_norm": 1.7472739219665527,
+      "learning_rate": 0.0005970858067997841,
+      "loss": 5.5746,
       "step": 550
     },
     {
-      "epoch": 0.06457862447529868,
-      "grad_norm": 1.3522851467132568,
-      "learning_rate": 0.0005967805193405882,
-      "loss": 5.4942,
+      "epoch": 0.0646900269541779,
+      "grad_norm": 1.1309614181518555,
+      "learning_rate": 0.0005967620075553157,
+      "loss": 5.5044,
       "step": 600
     },
     {
-      "epoch": 0.0699601765149069,
-      "grad_norm": 1.594616413116455,
-      "learning_rate": 0.0005964572783105269,
-      "loss": 5.4497,
+      "epoch": 0.07008086253369272,
+      "grad_norm": 1.2399049997329712,
+      "learning_rate": 0.0005964382083108472,
+      "loss": 5.4345,
       "step": 650
     },
     {
-      "epoch": 0.07534172855451512,
-      "grad_norm": 1.3532969951629639,
-      "learning_rate": 0.0005961340372804654,
-      "loss": 5.3653,
+      "epoch": 0.07547169811320754,
+      "grad_norm": 1.7309972047805786,
+      "learning_rate": 0.0005961144090663788,
+      "loss": 5.386,
       "step": 700
     },
     {
-      "epoch": 0.08072328059412334,
-      "grad_norm": 1.399992823600769,
-      "learning_rate": 0.000595810796250404,
-      "loss": 5.3081,
+      "epoch": 0.08086253369272237,
+      "grad_norm": 1.1939959526062012,
+      "learning_rate": 0.0005957906098219104,
+      "loss": 5.3088,
       "step": 750
     },
     {
-      "epoch": 0.08610483263373157,
-      "grad_norm": 1.2744522094726562,
-      "learning_rate": 0.0005954875552203426,
-      "loss": 5.2635,
+      "epoch": 0.0862533692722372,
+      "grad_norm": 1.6205377578735352,
+      "learning_rate": 0.0005954668105774419,
+      "loss": 5.2622,
       "step": 800
     },
     {
-      "epoch": 0.09148638467333979,
-      "grad_norm": 1.311295747756958,
-      "learning_rate": 0.0005951643141902811,
-      "loss": 5.2182,
+      "epoch": 0.09164420485175202,
+      "grad_norm": 1.3945573568344116,
+      "learning_rate": 0.0005951430113329735,
+      "loss": 5.2251,
       "step": 850
     },
     {
-      "epoch": 0.09686793671294801,
-      "grad_norm": 1.1952992677688599,
-      "learning_rate": 0.0005948410731602198,
-      "loss": 5.1703,
+      "epoch": 0.09703504043126684,
+      "grad_norm": 1.3046308755874634,
+      "learning_rate": 0.0005948192120885051,
+      "loss": 5.1654,
       "step": 900
     },
     {
-      "epoch": 0.10224948875255624,
-      "grad_norm": 1.2511310577392578,
-      "learning_rate": 0.0005945178321301583,
-      "loss": 5.132,
+      "epoch": 0.10242587601078167,
+      "grad_norm": 1.2131174802780151,
+      "learning_rate": 0.0005944954128440366,
+      "loss": 5.1448,
       "step": 950
     },
     {
-      "epoch": 0.10763104079216446,
-      "grad_norm": 1.1788593530654907,
-      "learning_rate": 0.000594194591100097,
-      "loss": 5.0646,
+      "epoch": 0.1078167115902965,
+      "grad_norm": 0.9675126075744629,
+      "learning_rate": 0.0005941716135995682,
+      "loss": 5.1035,
       "step": 1000
     },
     {
-      "epoch": 0.10763104079216446,
-      "eval_accuracy": 0.22730227591944518,
-      "eval_loss": 5.024896144866943,
-      "eval_runtime": 187.644,
-      "eval_samples_per_second": 95.985,
-      "eval_steps_per_second": 6.001,
+      "epoch": 0.1078167115902965,
+      "eval_accuracy": 0.22749035406825205,
+      "eval_loss": 5.018069267272949,
+      "eval_runtime": 184.6816,
+      "eval_samples_per_second": 97.525,
+      "eval_steps_per_second": 6.097,
       "step": 1000
     },
     {
-      "epoch": 0.11301259283177269,
-      "grad_norm": 1.5058064460754395,
-      "learning_rate": 0.0005938713500700355,
-      "loss": 5.0372,
+      "epoch": 0.11320754716981132,
+      "grad_norm": 1.0149263143539429,
+      "learning_rate": 0.0005938478143550997,
+      "loss": 5.0486,
       "step": 1050
     },
     {
-      "epoch": 0.1183941448713809,
-      "grad_norm": 1.2659214735031128,
-      "learning_rate": 0.000593548109039974,
-      "loss": 5.0221,
+      "epoch": 0.11859838274932614,
+      "grad_norm": 1.3123433589935303,
+      "learning_rate": 0.0005935240151106314,
+      "loss": 5.0207,
       "step": 1100
     },
     {
-      "epoch": 0.12377569691098914,
-      "grad_norm": 1.4247196912765503,
-      "learning_rate": 0.0005932248680099127,
-      "loss": 4.9852,
+      "epoch": 0.12398921832884097,
+      "grad_norm": 0.9050843715667725,
+      "learning_rate": 0.0005932002158661629,
+      "loss": 4.9928,
       "step": 1150
     },
     {
-      "epoch": 0.12915724895059735,
-      "grad_norm": 1.7715990543365479,
-      "learning_rate": 0.0005929016269798512,
-      "loss": 4.943,
+      "epoch": 0.1293800539083558,
+      "grad_norm": 1.274807095527649,
+      "learning_rate": 0.0005928764166216945,
+      "loss": 4.9535,
       "step": 1200
     },
     {
-      "epoch": 0.13453880099020557,
-      "grad_norm": 1.0012747049331665,
-      "learning_rate": 0.0005925783859497898,
-      "loss": 4.9166,
+      "epoch": 0.1347708894878706,
+      "grad_norm": 1.00777006149292,
+      "learning_rate": 0.000592552617377226,
+      "loss": 4.9137,
       "step": 1250
     },
     {
-      "epoch": 0.1399203530298138,
-      "grad_norm": 1.266779899597168,
-      "learning_rate": 0.0005922551449197284,
-      "loss": 4.8965,
+      "epoch": 0.14016172506738545,
+      "grad_norm": 0.8887598514556885,
+      "learning_rate": 0.0005922288181327577,
+      "loss": 4.8818,
       "step": 1300
     },
     {
-      "epoch": 0.14530190506942203,
-      "grad_norm": 0.9170055389404297,
-      "learning_rate": 0.0005919319038896671,
-      "loss": 4.8706,
+      "epoch": 0.14555256064690028,
+      "grad_norm": 0.9829823970794678,
+      "learning_rate": 0.0005919050188882893,
+      "loss": 4.8648,
       "step": 1350
     },
     {
-      "epoch": 0.15068345710903025,
-      "grad_norm": 0.9823539853096008,
-      "learning_rate": 0.0005916086628596056,
-      "loss": 4.8504,
+      "epoch": 0.1509433962264151,
+      "grad_norm": 0.9974054098129272,
+      "learning_rate": 0.0005915812196438207,
+      "loss": 4.8498,
       "step": 1400
     },
     {
-      "epoch": 0.15606500914863847,
-      "grad_norm": 1.2595840692520142,
-      "learning_rate": 0.0005912854218295442,
-      "loss": 4.813,
+      "epoch": 0.15633423180592992,
+      "grad_norm": 1.0419261455535889,
+      "learning_rate": 0.0005912574203993524,
+      "loss": 4.8005,
       "step": 1450
     },
     {
-      "epoch": 0.16144656118824668,
-      "grad_norm": 0.8510214686393738,
-      "learning_rate": 0.0005909621807994827,
-      "loss": 4.7813,
+      "epoch": 0.16172506738544473,
+      "grad_norm": 1.1737334728240967,
+      "learning_rate": 0.0005909336211548839,
+      "loss": 4.8138,
       "step": 1500
     },
     {
-      "epoch": 0.1668281132278549,
-      "grad_norm": 1.1120244264602661,
-      "learning_rate": 0.0005906389397694213,
-      "loss": 4.7637,
+      "epoch": 0.16711590296495957,
+      "grad_norm": 1.0961012840270996,
+      "learning_rate": 0.0005906098219104155,
+      "loss": 4.7659,
       "step": 1550
     },
     {
-      "epoch": 0.17220966526746315,
-      "grad_norm": 1.0976771116256714,
-      "learning_rate": 0.00059031569873936,
-      "loss": 4.7563,
+      "epoch": 0.1725067385444744,
+      "grad_norm": 1.1323438882827759,
+      "learning_rate": 0.000590286022665947,
+      "loss": 4.7663,
       "step": 1600
     },
     {
-      "epoch": 0.17759121730707136,
-      "grad_norm": 0.9833604693412781,
-      "learning_rate": 0.0005899924577092985,
-      "loss": 4.7421,
+      "epoch": 0.1778975741239892,
+      "grad_norm": 0.8736843466758728,
+      "learning_rate": 0.0005899622234214787,
+      "loss": 4.7157,
       "step": 1650
     },
     {
-      "epoch": 0.18297276934667958,
-      "grad_norm": 0.8846125602722168,
-      "learning_rate": 0.0005896692166792371,
-      "loss": 4.7031,
+      "epoch": 0.18328840970350405,
+      "grad_norm": 0.7891756892204285,
+      "learning_rate": 0.0005896384241770102,
+      "loss": 4.7099,
       "step": 1700
     },
     {
-      "epoch": 0.1883543213862878,
-      "grad_norm": 1.0292339324951172,
-      "learning_rate": 0.0005893459756491757,
-      "loss": 4.6573,
+      "epoch": 0.18867924528301888,
+      "grad_norm": 1.0397777557373047,
+      "learning_rate": 0.0005893146249325418,
+      "loss": 4.6729,
       "step": 1750
     },
     {
-      "epoch": 0.19373587342589602,
-      "grad_norm": 1.0036226511001587,
-      "learning_rate": 0.0005890227346191143,
-      "loss": 4.6505,
+      "epoch": 0.1940700808625337,
+      "grad_norm": 1.151475429534912,
+      "learning_rate": 0.0005889908256880733,
+      "loss": 4.6804,
       "step": 1800
     },
     {
-      "epoch": 0.19911742546550426,
-      "grad_norm": 0.9965416789054871,
-      "learning_rate": 0.0005886994935890529,
-      "loss": 4.6383,
+      "epoch": 0.19946091644204852,
+      "grad_norm": 1.5584900379180908,
+      "learning_rate": 0.0005886670264436049,
+      "loss": 4.6534,
       "step": 1850
     },
     {
-      "epoch": 0.20449897750511248,
-      "grad_norm": 1.1083658933639526,
-      "learning_rate": 0.0005883762525589915,
-      "loss": 4.6478,
+      "epoch": 0.20485175202156333,
+      "grad_norm": 0.7592549920082092,
+      "learning_rate": 0.0005883432271991365,
+      "loss": 4.6368,
       "step": 1900
     },
     {
-      "epoch": 0.2098805295447207,
-      "grad_norm": 0.93230140209198,
-      "learning_rate": 0.00058805301152893,
-      "loss": 4.5879,
+      "epoch": 0.21024258760107817,
+      "grad_norm": 1.132956862449646,
+      "learning_rate": 0.0005880194279546681,
+      "loss": 4.6073,
       "step": 1950
     },
     {
-      "epoch": 0.2152620815843289,
-      "grad_norm": 1.1331114768981934,
-      "learning_rate": 0.0005877297704988686,
-      "loss": 4.5685,
+      "epoch": 0.215633423180593,
+      "grad_norm": 0.904139518737793,
+      "learning_rate": 0.0005876956287101996,
+      "loss": 4.5922,
       "step": 2000
     },
     {
-      "epoch": 0.2152620815843289,
-      "eval_accuracy": 0.271470652692022,
-      "eval_loss": 4.505092620849609,
-      "eval_runtime": 185.4709,
-      "eval_samples_per_second": 97.11,
-      "eval_steps_per_second": 6.071,
+      "epoch": 0.215633423180593,
+      "eval_accuracy": 0.26994136329603313,
+      "eval_loss": 4.514337062835693,
+      "eval_runtime": 183.9245,
+      "eval_samples_per_second": 97.926,
+      "eval_steps_per_second": 6.122,
       "step": 2000
     },
     {
-      "epoch": 0.22064363362393713,
-      "grad_norm": 0.9773377776145935,
-      "learning_rate": 0.0005874065294688072,
-      "loss": 4.5528,
+      "epoch": 0.2210242587601078,
+      "grad_norm": 0.8630588054656982,
+      "learning_rate": 0.0005873718294657312,
+      "loss": 4.5736,
       "step": 2050
     },
     {
-      "epoch": 0.22602518566354537,
-      "grad_norm": 0.7613441348075867,
-      "learning_rate": 0.0005870832884387457,
-      "loss": 4.5416,
+      "epoch": 0.22641509433962265,
+      "grad_norm": 0.751631498336792,
+      "learning_rate": 0.0005870480302212628,
+      "loss": 4.5577,
       "step": 2100
     },
     {
-      "epoch": 0.2314067377031536,
-      "grad_norm": 0.9072310924530029,
-      "learning_rate": 0.0005867600474086844,
-      "loss": 4.5192,
+      "epoch": 0.23180592991913745,
+      "grad_norm": 0.9096190333366394,
+      "learning_rate": 0.0005867242309767943,
+      "loss": 4.5327,
       "step": 2150
     },
     {
-      "epoch": 0.2367882897427618,
-      "grad_norm": 0.9957188963890076,
-      "learning_rate": 0.0005864368063786229,
-      "loss": 4.5067,
+      "epoch": 0.2371967654986523,
+      "grad_norm": 0.9275792241096497,
+      "learning_rate": 0.0005864004317323259,
+      "loss": 4.513,
       "step": 2200
     },
     {
-      "epoch": 0.24216984178237003,
-      "grad_norm": 1.090625286102295,
-      "learning_rate": 0.0005861135653485616,
-      "loss": 4.4835,
+      "epoch": 0.24258760107816713,
+      "grad_norm": 0.8501319885253906,
+      "learning_rate": 0.0005860766324878575,
+      "loss": 4.4957,
       "step": 2250
     },
     {
-      "epoch": 0.24755139382197827,
-      "grad_norm": 1.0402551889419556,
-      "learning_rate": 0.0005857903243185001,
-      "loss": 4.4773,
+      "epoch": 0.24797843665768193,
+      "grad_norm": 0.8335550427436829,
+      "learning_rate": 0.000585752833243389,
+      "loss": 4.505,
       "step": 2300
     },
     {
-      "epoch": 0.2529329458615865,
-      "grad_norm": 0.9853291511535645,
-      "learning_rate": 0.0005854670832884386,
-      "loss": 4.4576,
+      "epoch": 0.25336927223719674,
+      "grad_norm": 0.7996338605880737,
+      "learning_rate": 0.0005854290339989206,
+      "loss": 4.4554,
       "step": 2350
     },
     {
-      "epoch": 0.2583144979011947,
-      "grad_norm": 0.777766227722168,
-      "learning_rate": 0.0005851438422583773,
-      "loss": 4.4518,
+      "epoch": 0.2587601078167116,
+      "grad_norm": 1.0249171257019043,
+      "learning_rate": 0.0005851052347544521,
+      "loss": 4.4629,
       "step": 2400
     },
     {
-      "epoch": 0.2636960499408029,
-      "grad_norm": 1.0316593647003174,
-      "learning_rate": 0.0005848206012283159,
-      "loss": 4.43,
+      "epoch": 0.2641509433962264,
+      "grad_norm": 0.9536342620849609,
+      "learning_rate": 0.0005847814355099838,
+      "loss": 4.4412,
       "step": 2450
     },
     {
-      "epoch": 0.26907760198041114,
-      "grad_norm": 0.7335191369056702,
-      "learning_rate": 0.0005844973601982545,
-      "loss": 4.4154,
+      "epoch": 0.2695417789757412,
+      "grad_norm": 1.2684301137924194,
+      "learning_rate": 0.0005844576362655154,
+      "loss": 4.4274,
       "step": 2500
     },
     {
-      "epoch": 0.27445915402001936,
-      "grad_norm": 0.9138801097869873,
-      "learning_rate": 0.000584174119168193,
-      "loss": 4.396,
+      "epoch": 0.2749326145552561,
+      "grad_norm": 0.8013315796852112,
+      "learning_rate": 0.0005841338370210469,
+      "loss": 4.4099,
       "step": 2550
     },
     {
-      "epoch": 0.2798407060596276,
-      "grad_norm": 0.6920284628868103,
-      "learning_rate": 0.0005838508781381316,
-      "loss": 4.4028,
+      "epoch": 0.2803234501347709,
+      "grad_norm": 0.8502827882766724,
+      "learning_rate": 0.0005838100377765785,
+      "loss": 4.3802,
       "step": 2600
     },
     {
-      "epoch": 0.2852222580992358,
-      "grad_norm": 0.9748484492301941,
-      "learning_rate": 0.0005835276371080702,
-      "loss": 4.367,
+      "epoch": 0.2857142857142857,
+      "grad_norm": 0.8461282849311829,
+      "learning_rate": 0.0005834862385321101,
+      "loss": 4.3947,
       "step": 2650
     },
     {
-      "epoch": 0.29060381013884407,
-      "grad_norm": 0.7634653449058533,
-      "learning_rate": 0.0005832043960780088,
-      "loss": 4.3595,
+      "epoch": 0.29110512129380056,
+      "grad_norm": 1.0242016315460205,
+      "learning_rate": 0.0005831624392876417,
+      "loss": 4.3976,
       "step": 2700
     },
     {
-      "epoch": 0.2959853621784523,
-      "grad_norm": 0.6365559101104736,
-      "learning_rate": 0.0005828811550479474,
-      "loss": 4.3588,
+      "epoch": 0.29649595687331537,
+      "grad_norm": 0.8631992936134338,
+      "learning_rate": 0.0005828386400431731,
+      "loss": 4.3521,
       "step": 2750
     },
     {
-      "epoch": 0.3013669142180605,
-      "grad_norm": 0.632849395275116,
-      "learning_rate": 0.0005825579140178859,
-      "loss": 4.3405,
+      "epoch": 0.3018867924528302,
+      "grad_norm": 0.7447966933250427,
+      "learning_rate": 0.0005825148407987048,
+      "loss": 4.3323,
       "step": 2800
     },
     {
-      "epoch": 0.3067484662576687,
-      "grad_norm": 0.8527909517288208,
-      "learning_rate": 0.0005822346729878246,
-      "loss": 4.3207,
+      "epoch": 0.30727762803234504,
+      "grad_norm": 0.7234294414520264,
+      "learning_rate": 0.0005821910415542363,
+      "loss": 4.3364,
       "step": 2850
     },
     {
-      "epoch": 0.31213001829727693,
-      "grad_norm": 0.948337197303772,
-      "learning_rate": 0.0005819114319577631,
-      "loss": 4.3312,
+      "epoch": 0.31266846361185985,
+      "grad_norm": 0.8805587887763977,
+      "learning_rate": 0.0005818672423097679,
+      "loss": 4.3307,
       "step": 2900
     },
     {
-      "epoch": 0.31751157033688515,
-      "grad_norm": 0.8351525664329529,
-      "learning_rate": 0.0005815881909277017,
-      "loss": 4.3143,
+      "epoch": 0.31805929919137466,
+      "grad_norm": 0.8215044140815735,
+      "learning_rate": 0.0005815434430652994,
+      "loss": 4.3141,
       "step": 2950
     },
     {
-      "epoch": 0.32289312237649337,
-      "grad_norm": 0.8625895977020264,
-      "learning_rate": 0.0005812649498976403,
-      "loss": 4.3002,
+      "epoch": 0.32345013477088946,
+      "grad_norm": 0.7359305620193481,
+      "learning_rate": 0.0005812196438208311,
+      "loss": 4.311,
       "step": 3000
     },
     {
-      "epoch": 0.32289312237649337,
-      "eval_accuracy": 0.29852044103076386,
-      "eval_loss": 4.236849308013916,
-      "eval_runtime": 184.7432,
-      "eval_samples_per_second": 97.492,
-      "eval_steps_per_second": 6.095,
+      "epoch": 0.32345013477088946,
+      "eval_accuracy": 0.298171882566655,
+      "eval_loss": 4.2451090812683105,
+      "eval_runtime": 183.7967,
+      "eval_samples_per_second": 97.994,
+      "eval_steps_per_second": 6.126,
       "step": 3000
     },
     {
-      "epoch": 0.3282746744161016,
-      "grad_norm": 0.7663789391517639,
-      "learning_rate": 0.0005809417088675789,
-      "loss": 4.2959,
+      "epoch": 0.3288409703504043,
+      "grad_norm": 0.770511269569397,
+      "learning_rate": 0.0005808958445763626,
+      "loss": 4.3027,
       "step": 3050
     },
     {
-      "epoch": 0.3336562264557098,
-      "grad_norm": 0.7020168304443359,
-      "learning_rate": 0.0005806184678375175,
-      "loss": 4.2839,
+      "epoch": 0.33423180592991913,
+      "grad_norm": 0.677880048751831,
+      "learning_rate": 0.0005805720453318942,
+      "loss": 4.3083,
       "step": 3100
     },
     {
-      "epoch": 0.3390377784953181,
-      "grad_norm": 0.7872812151908875,
-      "learning_rate": 0.000580295226807456,
-      "loss": 4.2729,
+      "epoch": 0.33962264150943394,
+      "grad_norm": 0.6858367919921875,
+      "learning_rate": 0.0005802482460874257,
+      "loss": 4.2657,
       "step": 3150
     },
     {
-      "epoch": 0.3444193305349263,
-      "grad_norm": 0.8375337719917297,
-      "learning_rate": 0.0005799719857773946,
-      "loss": 4.281,
+      "epoch": 0.3450134770889488,
+      "grad_norm": 0.6485952138900757,
+      "learning_rate": 0.0005799244468429573,
+      "loss": 4.261,
       "step": 3200
     },
     {
-      "epoch": 0.3498008825745345,
-      "grad_norm": 0.7012325525283813,
-      "learning_rate": 0.0005796487447473331,
-      "loss": 4.26,
+      "epoch": 0.3504043126684636,
+      "grad_norm": 0.706744372844696,
+      "learning_rate": 0.0005796006475984889,
+      "loss": 4.2609,
       "step": 3250
     },
     {
-      "epoch": 0.35518243461414273,
-      "grad_norm": 0.7597373723983765,
-      "learning_rate": 0.0005793255037172718,
-      "loss": 4.2512,
+      "epoch": 0.3557951482479784,
+      "grad_norm": 0.7495291233062744,
+      "learning_rate": 0.0005792768483540205,
+      "loss": 4.2458,
       "step": 3300
     },
     {
-      "epoch": 0.36056398665375095,
-      "grad_norm": 0.7354339361190796,
-      "learning_rate": 0.0005790022626872104,
-      "loss": 4.2585,
+      "epoch": 0.3611859838274933,
+      "grad_norm": 0.6684705018997192,
+      "learning_rate": 0.000578953049109552,
+      "loss": 4.2625,
       "step": 3350
     },
     {
-      "epoch": 0.36594553869335916,
-      "grad_norm": 0.9357606768608093,
-      "learning_rate": 0.000578679021657149,
-      "loss": 4.2464,
+      "epoch": 0.3665768194070081,
+      "grad_norm": 0.6015046834945679,
+      "learning_rate": 0.0005786292498650836,
+      "loss": 4.2344,
       "step": 3400
     },
     {
-      "epoch": 0.3713270907329674,
-      "grad_norm": 0.6800475716590881,
-      "learning_rate": 0.0005783557806270875,
-      "loss": 4.2159,
+      "epoch": 0.3719676549865229,
+      "grad_norm": 0.6746869087219238,
+      "learning_rate": 0.0005783054506206152,
+      "loss": 4.2115,
       "step": 3450
     },
     {
-      "epoch": 0.3767086427725756,
-      "grad_norm": 0.8798818588256836,
-      "learning_rate": 0.0005780325395970262,
-      "loss": 4.2192,
+      "epoch": 0.37735849056603776,
+      "grad_norm": 0.6800693273544312,
+      "learning_rate": 0.0005779816513761467,
+      "loss": 4.2238,
       "step": 3500
     },
     {
-      "epoch": 0.3820901948121838,
-      "grad_norm": 0.6990943551063538,
-      "learning_rate": 0.0005777092985669647,
-      "loss": 4.2043,
+      "epoch": 0.38274932614555257,
+      "grad_norm": 0.8763014674186707,
+      "learning_rate": 0.0005776578521316782,
+      "loss": 4.2176,
       "step": 3550
     },
     {
-      "epoch": 0.38747174685179203,
-      "grad_norm": 0.732219398021698,
-      "learning_rate": 0.0005773860575369033,
-      "loss": 4.2124,
+      "epoch": 0.3881401617250674,
+      "grad_norm": 0.6071361899375916,
+      "learning_rate": 0.0005773340528872099,
+      "loss": 4.2172,
       "step": 3600
     },
     {
-      "epoch": 0.3928532988914003,
-      "grad_norm": 0.768298327922821,
-      "learning_rate": 0.0005770628165068419,
-      "loss": 4.194,
+      "epoch": 0.3935309973045822,
+      "grad_norm": 0.8424677848815918,
+      "learning_rate": 0.0005770102536427414,
+      "loss": 4.1909,
       "step": 3650
     },
     {
-      "epoch": 0.3982348509310085,
-      "grad_norm": 0.7568483948707581,
-      "learning_rate": 0.0005767395754767804,
-      "loss": 4.1734,
+      "epoch": 0.39892183288409705,
+      "grad_norm": 0.7024722099304199,
+      "learning_rate": 0.000576686454398273,
+      "loss": 4.193,
       "step": 3700
     },
     {
-      "epoch": 0.40361640297061674,
-      "grad_norm": 0.7396944165229797,
-      "learning_rate": 0.0005764163344467191,
-      "loss": 4.1768,
+      "epoch": 0.40431266846361186,
+      "grad_norm": 0.6678298711776733,
+      "learning_rate": 0.0005763626551538045,
+      "loss": 4.1954,
       "step": 3750
     },
     {
-      "epoch": 0.40899795501022496,
-      "grad_norm": 0.7737883925437927,
-      "learning_rate": 0.0005760930934166576,
-      "loss": 4.1866,
+      "epoch": 0.40970350404312667,
+      "grad_norm": 0.591792643070221,
+      "learning_rate": 0.0005760388559093362,
+      "loss": 4.1897,
       "step": 3800
     },
     {
-      "epoch": 0.4143795070498332,
-      "grad_norm": 0.6525033116340637,
-      "learning_rate": 0.0005757698523865963,
-      "loss": 4.1749,
+      "epoch": 0.41509433962264153,
+      "grad_norm": 0.7370854020118713,
+      "learning_rate": 0.0005757150566648678,
+      "loss": 4.1703,
       "step": 3850
     },
     {
-      "epoch": 0.4197610590894414,
-      "grad_norm": 0.7739155888557434,
-      "learning_rate": 0.0005754466113565348,
-      "loss": 4.18,
+      "epoch": 0.42048517520215634,
+      "grad_norm": 0.8055949807167053,
+      "learning_rate": 0.0005753912574203993,
+      "loss": 4.1666,
       "step": 3900
     },
     {
-      "epoch": 0.4251426111290496,
-      "grad_norm": 0.7185007333755493,
-      "learning_rate": 0.0005751233703264733,
-      "loss": 4.1561,
+      "epoch": 0.42587601078167114,
+      "grad_norm": 0.6214135885238647,
+      "learning_rate": 0.0005750674581759309,
+      "loss": 4.1586,
       "step": 3950
     },
     {
-      "epoch": 0.4305241631686578,
-      "grad_norm": 0.6594641804695129,
-      "learning_rate": 0.000574800129296412,
-      "loss": 4.1579,
+      "epoch": 0.431266846361186,
+      "grad_norm": 0.7452210783958435,
+      "learning_rate": 0.0005747436589314624,
+      "loss": 4.159,
       "step": 4000
     },
     {
-      "epoch": 0.4305241631686578,
-      "eval_accuracy": 0.3129942008694187,
-      "eval_loss": 4.0840744972229,
-      "eval_runtime": 184.6939,
-      "eval_samples_per_second": 97.518,
-      "eval_steps_per_second": 6.097,
+      "epoch": 0.431266846361186,
+      "eval_accuracy": 0.31298409615085193,
+      "eval_loss": 4.08732271194458,
+      "eval_runtime": 183.8347,
+      "eval_samples_per_second": 97.974,
+      "eval_steps_per_second": 6.125,
       "step": 4000
     },
     {
-      "epoch": 0.43590571520826604,
-      "grad_norm": 0.6553388833999634,
-      "learning_rate": 0.0005744768882663505,
-      "loss": 4.1519,
+      "epoch": 0.4366576819407008,
+      "grad_norm": 0.5454065203666687,
+      "learning_rate": 0.0005744198596869941,
+      "loss": 4.1509,
       "step": 4050
     },
     {
-      "epoch": 0.44128726724787426,
-      "grad_norm": 0.7919243574142456,
-      "learning_rate": 0.0005741536472362891,
-      "loss": 4.165,
+      "epoch": 0.4420485175202156,
+      "grad_norm": 0.6480352282524109,
+      "learning_rate": 0.0005740960604425255,
+      "loss": 4.1515,
       "step": 4100
     },
     {
-      "epoch": 0.44666881928748253,
-      "grad_norm": 0.5930292010307312,
-      "learning_rate": 0.0005738304062062277,
-      "loss": 4.1667,
+      "epoch": 0.4474393530997305,
+      "grad_norm": 0.7491263151168823,
+      "learning_rate": 0.0005737722611980572,
+      "loss": 4.1543,
       "step": 4150
     },
     {
-      "epoch": 0.45205037132709075,
-      "grad_norm": 0.7211865782737732,
-      "learning_rate": 0.0005735071651761664,
-      "loss": 4.1264,
+      "epoch": 0.4528301886792453,
+      "grad_norm": 0.5414714217185974,
+      "learning_rate": 0.0005734484619535887,
+      "loss": 4.1397,
       "step": 4200
     },
     {
-      "epoch": 0.45743192336669897,
-      "grad_norm": 0.7048370242118835,
-      "learning_rate": 0.0005731839241461049,
-      "loss": 4.1397,
+      "epoch": 0.4582210242587601,
+      "grad_norm": 0.735364556312561,
+      "learning_rate": 0.0005731246627091203,
+      "loss": 4.1231,
       "step": 4250
     },
     {
-      "epoch": 0.4628134754063072,
-      "grad_norm": 0.7572526931762695,
-      "learning_rate": 0.0005728606831160435,
-      "loss": 4.1205,
+      "epoch": 0.4636118598382749,
+      "grad_norm": 0.7785477042198181,
+      "learning_rate": 0.0005728008634646518,
+      "loss": 4.1242,
       "step": 4300
     },
     {
-      "epoch": 0.4681950274459154,
-      "grad_norm": 0.7019510865211487,
-      "learning_rate": 0.000572537442085982,
-      "loss": 4.1274,
+      "epoch": 0.46900269541778977,
+      "grad_norm": 0.6108700633049011,
+      "learning_rate": 0.0005724770642201835,
+      "loss": 4.118,
       "step": 4350
     },
     {
-      "epoch": 0.4735765794855236,
-      "grad_norm": 0.693013072013855,
-      "learning_rate": 0.0005722142010559206,
-      "loss": 4.1165,
+      "epoch": 0.4743935309973046,
+      "grad_norm": 0.6415695548057556,
+      "learning_rate": 0.000572153264975715,
+      "loss": 4.1178,
       "step": 4400
     },
     {
-      "epoch": 0.47895813152513184,
-      "grad_norm": 0.6860193014144897,
-      "learning_rate": 0.0005718909600258593,
-      "loss": 4.1139,
+      "epoch": 0.4797843665768194,
+      "grad_norm": 0.8770371675491333,
+      "learning_rate": 0.0005718294657312466,
+      "loss": 4.1028,
       "step": 4450
     },
     {
-      "epoch": 0.48433968356474005,
-      "grad_norm": 0.6195184588432312,
-      "learning_rate": 0.0005715677189957978,
-      "loss": 4.1266,
+      "epoch": 0.48517520215633425,
+      "grad_norm": 0.7115625143051147,
+      "learning_rate": 0.0005715056664867781,
+      "loss": 4.1109,
       "step": 4500
     },
     {
-      "epoch": 0.48972123560434827,
-      "grad_norm": 0.7034777402877808,
-      "learning_rate": 0.0005712444779657364,
-      "loss": 4.0937,
+      "epoch": 0.49056603773584906,
+      "grad_norm": 0.660191535949707,
+      "learning_rate": 0.0005711818672423097,
+      "loss": 4.1011,
       "step": 4550
     },
     {
-      "epoch": 0.49510278764395654,
-      "grad_norm": 0.6804658770561218,
-      "learning_rate": 0.000570921236935675,
-      "loss": 4.1195,
+      "epoch": 0.49595687331536387,
+      "grad_norm": 0.6192442178726196,
+      "learning_rate": 0.0005708580679978413,
+      "loss": 4.0857,
       "step": 4600
     },
     {
-      "epoch": 0.5004843396835648,
-      "grad_norm": 0.7629151940345764,
-      "learning_rate": 0.0005705979959056136,
-      "loss": 4.0871,
+      "epoch": 0.5013477088948787,
+      "grad_norm": 0.6783645153045654,
+      "learning_rate": 0.0005705342687533729,
+      "loss": 4.078,
       "step": 4650
     },
     {
-      "epoch": 0.505865891723173,
-      "grad_norm": 0.5974324345588684,
-      "learning_rate": 0.0005702747548755522,
-      "loss": 4.1022,
+      "epoch": 0.5067385444743935,
+      "grad_norm": 0.6928367018699646,
+      "learning_rate": 0.0005702104695089044,
+      "loss": 4.0956,
       "step": 4700
     },
     {
-      "epoch": 0.5112474437627812,
-      "grad_norm": 0.5412039160728455,
-      "learning_rate": 0.0005699515138454908,
-      "loss": 4.0876,
+      "epoch": 0.5121293800539084,
+      "grad_norm": 0.6265180110931396,
+      "learning_rate": 0.000569886670264436,
+      "loss": 4.0822,
       "step": 4750
     },
     {
-      "epoch": 0.5166289958023894,
-      "grad_norm": 0.7911316752433777,
-      "learning_rate": 0.0005696282728154293,
-      "loss": 4.0688,
+      "epoch": 0.5175202156334232,
+      "grad_norm": 0.7510149478912354,
+      "learning_rate": 0.0005695628710199675,
+      "loss": 4.0755,
       "step": 4800
     },
     {
-      "epoch": 0.5220105478419976,
-      "grad_norm": 0.68361496925354,
-      "learning_rate": 0.0005693050317853679,
-      "loss": 4.0913,
+      "epoch": 0.522911051212938,
+      "grad_norm": 0.7354061603546143,
+      "learning_rate": 0.0005692390717754991,
+      "loss": 4.0655,
       "step": 4850
     },
     {
-      "epoch": 0.5273920998816058,
-      "grad_norm": 0.577229380607605,
-      "learning_rate": 0.0005689817907553065,
-      "loss": 4.0676,
+      "epoch": 0.5283018867924528,
+      "grad_norm": 0.6540889739990234,
+      "learning_rate": 0.0005689152725310306,
+      "loss": 4.055,
       "step": 4900
     },
     {
-      "epoch": 0.5327736519212141,
-      "grad_norm": 0.7196305394172668,
-      "learning_rate": 0.000568658549725245,
-      "loss": 4.0467,
+      "epoch": 0.5336927223719676,
+      "grad_norm": 0.674555778503418,
+      "learning_rate": 0.0005685914732865623,
+      "loss": 4.0696,
       "step": 4950
     },
     {
-      "epoch": 0.5381552039608223,
-      "grad_norm": 0.5695436596870422,
-      "learning_rate": 0.0005683353086951837,
-      "loss": 4.062,
+      "epoch": 0.5390835579514824,
+      "grad_norm": 0.4996623992919922,
+      "learning_rate": 0.0005682676740420939,
+      "loss": 4.0472,
       "step": 5000
     },
     {
-      "epoch": 0.5381552039608223,
-      "eval_accuracy": 0.3209510691498487,
-      "eval_loss": 3.9919753074645996,
-      "eval_runtime": 184.7142,
-      "eval_samples_per_second": 97.507,
-      "eval_steps_per_second": 6.096,
+      "epoch": 0.5390835579514824,
+      "eval_accuracy": 0.3215982057496718,
+      "eval_loss": 3.9894087314605713,
+      "eval_runtime": 183.8499,
+      "eval_samples_per_second": 97.966,
+      "eval_steps_per_second": 6.125,
       "step": 5000
     },
     {
-      "epoch": 0.5435367560004305,
-      "grad_norm": 0.5603345632553101,
-      "learning_rate": 0.0005680120676651222,
-      "loss": 4.0619,
+      "epoch": 0.5444743935309974,
+      "grad_norm": 0.7127335071563721,
+      "learning_rate": 0.0005679438747976254,
+      "loss": 4.0568,
       "step": 5050
     },
     {
-      "epoch": 0.5489183080400387,
-      "grad_norm": 0.6007739305496216,
-      "learning_rate": 0.0005676888266350609,
-      "loss": 4.039,
+      "epoch": 0.5498652291105122,
+      "grad_norm": 0.6820212602615356,
+      "learning_rate": 0.000567620075553157,
+      "loss": 4.0516,
       "step": 5100
     },
     {
-      "epoch": 0.5542998600796469,
-      "grad_norm": 0.7048094272613525,
-      "learning_rate": 0.0005673655856049994,
-      "loss": 4.0414,
+      "epoch": 0.555256064690027,
+      "grad_norm": 0.7278046011924744,
+      "learning_rate": 0.0005672962763086886,
+      "loss": 4.0498,
       "step": 5150
     },
     {
-      "epoch": 0.5596814121192552,
-      "grad_norm": 0.6947103142738342,
-      "learning_rate": 0.0005670423445749379,
-      "loss": 4.0415,
+      "epoch": 0.5606469002695418,
+      "grad_norm": 0.624863862991333,
+      "learning_rate": 0.0005669724770642202,
+      "loss": 4.0671,
       "step": 5200
     },
     {
-      "epoch": 0.5650629641588634,
-      "grad_norm": 0.6355268359184265,
-      "learning_rate": 0.0005667191035448766,
-      "loss": 4.0565,
+      "epoch": 0.5660377358490566,
+      "grad_norm": 0.6197010278701782,
+      "learning_rate": 0.0005666486778197517,
+      "loss": 4.041,
       "step": 5250
     },
     {
-      "epoch": 0.5704445161984716,
-      "grad_norm": 0.6978614926338196,
-      "learning_rate": 0.0005663958625148152,
-      "loss": 4.0299,
+      "epoch": 0.5714285714285714,
+      "grad_norm": 0.7419002056121826,
+      "learning_rate": 0.0005663248785752833,
+      "loss": 4.0326,
       "step": 5300
     },
     {
-      "epoch": 0.5758260682380799,
-      "grad_norm": 0.5592193007469177,
-      "learning_rate": 0.0005660726214847538,
-      "loss": 4.0185,
+      "epoch": 0.5768194070080862,
+      "grad_norm": 0.6294003129005432,
+      "learning_rate": 0.0005660010793308148,
+      "loss": 4.0263,
       "step": 5350
     },
     {
-      "epoch": 0.5812076202776881,
-      "grad_norm": 0.6000945568084717,
-      "learning_rate": 0.0005657493804546923,
-      "loss": 4.0321,
+      "epoch": 0.5822102425876011,
+      "grad_norm": 0.5749978423118591,
+      "learning_rate": 0.0005656772800863465,
+      "loss": 4.0359,
       "step": 5400
     },
     {
-      "epoch": 0.5865891723172963,
-      "grad_norm": 0.6557536721229553,
-      "learning_rate": 0.0005654261394246309,
-      "loss": 4.0062,
+      "epoch": 0.5876010781671159,
+      "grad_norm": 0.5415307879447937,
+      "learning_rate": 0.0005653534808418779,
+      "loss": 4.0382,
       "step": 5450
     },
     {
-      "epoch": 0.5919707243569046,
-      "grad_norm": 0.6313118934631348,
-      "learning_rate": 0.0005651028983945695,
-      "loss": 4.0171,
+      "epoch": 0.5929919137466307,
+      "grad_norm": 0.6727957725524902,
+      "learning_rate": 0.0005650296815974096,
+      "loss": 4.0181,
       "step": 5500
     },
     {
-      "epoch": 0.5973522763965128,
-      "grad_norm": 0.6380894184112549,
-      "learning_rate": 0.000564779657364508,
-      "loss": 4.0139,
+      "epoch": 0.5983827493261455,
+      "grad_norm": 0.6418060660362244,
+      "learning_rate": 0.0005647058823529411,
+      "loss": 4.0069,
       "step": 5550
     },
     {
-      "epoch": 0.602733828436121,
-      "grad_norm": 0.5402942299842834,
-      "learning_rate": 0.0005644564163344467,
-      "loss": 4.0111,
+      "epoch": 0.6037735849056604,
+      "grad_norm": 0.7099882960319519,
+      "learning_rate": 0.0005643820831084727,
+      "loss": 4.0226,
       "step": 5600
     },
     {
-      "epoch": 0.6081153804757292,
-      "grad_norm": 0.5897926688194275,
-      "learning_rate": 0.0005641331753043852,
-      "loss": 4.0321,
+      "epoch": 0.6091644204851752,
+      "grad_norm": 0.6498609185218811,
+      "learning_rate": 0.0005640582838640042,
+      "loss": 4.0046,
       "step": 5650
     },
     {
-      "epoch": 0.6134969325153374,
-      "grad_norm": 0.6751918792724609,
-      "learning_rate": 0.0005638099342743239,
-      "loss": 4.014,
+      "epoch": 0.6145552560646901,
+      "grad_norm": 0.5713142156600952,
+      "learning_rate": 0.0005637344846195358,
+      "loss": 4.0043,
       "step": 5700
     },
     {
-      "epoch": 0.6188784845549457,
-      "grad_norm": 0.5889696478843689,
-      "learning_rate": 0.0005634866932442624,
-      "loss": 3.9981,
+      "epoch": 0.6199460916442049,
+      "grad_norm": 0.6617730259895325,
+      "learning_rate": 0.0005634106853750674,
+      "loss": 3.979,
       "step": 5750
     },
     {
-      "epoch": 0.6242600365945539,
-      "grad_norm": 0.6100760698318481,
-      "learning_rate": 0.000563163452214201,
-      "loss": 4.0152,
+      "epoch": 0.6253369272237197,
+      "grad_norm": 0.6460627317428589,
+      "learning_rate": 0.000563086886130599,
+      "loss": 3.9846,
       "step": 5800
     },
     {
-      "epoch": 0.6296415886341621,
-      "grad_norm": 0.6364511847496033,
-      "learning_rate": 0.0005628402111841396,
-      "loss": 4.0011,
+      "epoch": 0.6307277628032345,
+      "grad_norm": 0.6782198548316956,
+      "learning_rate": 0.0005627630868861305,
+      "loss": 3.9947,
       "step": 5850
     },
     {
-      "epoch": 0.6350231406737703,
-      "grad_norm": 0.6010422706604004,
-      "learning_rate": 0.0005625169701540782,
-      "loss": 3.9959,
+      "epoch": 0.6361185983827493,
+      "grad_norm": 0.5800774097442627,
+      "learning_rate": 0.0005624392876416621,
+      "loss": 3.9922,
       "step": 5900
     },
     {
-      "epoch": 0.6404046927133785,
-      "grad_norm": 0.5707629919052124,
-      "learning_rate": 0.0005621937291240168,
-      "loss": 3.9918,
+      "epoch": 0.6415094339622641,
+      "grad_norm": 0.676412045955658,
+      "learning_rate": 0.0005621154883971937,
+      "loss": 3.9791,
       "step": 5950
     },
     {
-      "epoch": 0.6457862447529867,
-      "grad_norm": 0.6960643529891968,
-      "learning_rate": 0.0005618704880939553,
-      "loss": 3.9778,
+      "epoch": 0.6469002695417789,
+      "grad_norm": 0.583685040473938,
+      "learning_rate": 0.0005617916891527253,
+      "loss": 3.9931,
       "step": 6000
     },
     {
-      "epoch": 0.6457862447529867,
-      "eval_accuracy": 0.32842921280656817,
-      "eval_loss": 3.9220640659332275,
-      "eval_runtime": 184.5812,
-      "eval_samples_per_second": 97.578,
-      "eval_steps_per_second": 6.1,
+      "epoch": 0.6469002695417789,
+      "eval_accuracy": 0.32843508006251015,
+      "eval_loss": 3.9174036979675293,
+      "eval_runtime": 183.9133,
+      "eval_samples_per_second": 97.932,
+      "eval_steps_per_second": 6.122,
       "step": 6000
     },
     {
-      "epoch": 0.651167796792595,
-      "grad_norm": 0.5975825190544128,
-      "learning_rate": 0.0005615472470638939,
-      "loss": 3.9885,
+      "epoch": 0.6522911051212938,
+      "grad_norm": 0.5780828595161438,
+      "learning_rate": 0.0005614678899082568,
+      "loss": 3.9663,
       "step": 6050
     },
     {
-      "epoch": 0.6565493488322032,
-      "grad_norm": 0.5811153650283813,
-      "learning_rate": 0.0005612240060338325,
-      "loss": 3.9683,
+      "epoch": 0.6576819407008087,
+      "grad_norm": 0.5652017593383789,
+      "learning_rate": 0.0005611440906637884,
+      "loss": 3.9971,
       "step": 6100
     },
     {
-      "epoch": 0.6619309008718114,
-      "grad_norm": 0.6349126696586609,
-      "learning_rate": 0.0005609007650037711,
-      "loss": 3.978,
+      "epoch": 0.6630727762803235,
+      "grad_norm": 0.8341758847236633,
+      "learning_rate": 0.00056082029141932,
+      "loss": 3.9793,
       "step": 6150
     },
     {
-      "epoch": 0.6673124529114196,
-      "grad_norm": 0.5223124623298645,
-      "learning_rate": 0.0005605775239737097,
-      "loss": 3.9877,
+      "epoch": 0.6684636118598383,
+      "grad_norm": 0.578648567199707,
+      "learning_rate": 0.0005604964921748515,
+      "loss": 3.9499,
       "step": 6200
     },
     {
-      "epoch": 0.6726940049510278,
-      "grad_norm": 0.5649115443229675,
-      "learning_rate": 0.0005602542829436483,
-      "loss": 3.9654,
+      "epoch": 0.6738544474393531,
+      "grad_norm": 0.59043949842453,
+      "learning_rate": 0.000560172692930383,
+      "loss": 3.9715,
       "step": 6250
     },
     {
-      "epoch": 0.6780755569906362,
-      "grad_norm": 0.5698593258857727,
-      "learning_rate": 0.0005599310419135868,
-      "loss": 3.9748,
+      "epoch": 0.6792452830188679,
+      "grad_norm": 0.6629766225814819,
+      "learning_rate": 0.0005598488936859147,
+      "loss": 3.9749,
       "step": 6300
     },
     {
-      "epoch": 0.6834571090302444,
-      "grad_norm": 0.5707403421401978,
-      "learning_rate": 0.0005596078008835255,
-      "loss": 3.9573,
+      "epoch": 0.6846361185983828,
+      "grad_norm": 0.5541216135025024,
+      "learning_rate": 0.0005595250944414463,
+      "loss": 3.9608,
       "step": 6350
     },
     {
-      "epoch": 0.6888386610698526,
-      "grad_norm": 0.6776087284088135,
-      "learning_rate": 0.000559284559853464,
-      "loss": 3.9625,
+      "epoch": 0.6900269541778976,
+      "grad_norm": 0.6767361164093018,
+      "learning_rate": 0.0005592012951969778,
+      "loss": 3.9694,
       "step": 6400
     },
     {
-      "epoch": 0.6942202131094608,
-      "grad_norm": 0.5434262752532959,
-      "learning_rate": 0.0005589613188234026,
-      "loss": 3.9497,
+      "epoch": 0.6954177897574124,
+      "grad_norm": 0.6981496810913086,
+      "learning_rate": 0.0005588774959525094,
+      "loss": 3.9655,
       "step": 6450
     },
     {
-      "epoch": 0.699601765149069,
-      "grad_norm": 0.621336817741394,
-      "learning_rate": 0.0005586380777933412,
-      "loss": 3.9608,
+      "epoch": 0.7008086253369272,
+      "grad_norm": 0.5768890380859375,
+      "learning_rate": 0.000558553696708041,
+      "loss": 3.96,
       "step": 6500
     },
     {
-      "epoch": 0.7049833171886772,
-      "grad_norm": 0.6960099339485168,
-      "learning_rate": 0.0005583148367632797,
-      "loss": 3.9539,
+      "epoch": 0.706199460916442,
+      "grad_norm": 0.6919478178024292,
+      "learning_rate": 0.0005582298974635726,
+      "loss": 3.958,
       "step": 6550
     },
     {
-      "epoch": 0.7103648692282855,
-      "grad_norm": 0.6768355965614319,
-      "learning_rate": 0.0005579915957332184,
-      "loss": 3.9431,
+      "epoch": 0.7115902964959568,
+      "grad_norm": 0.5978683829307556,
+      "learning_rate": 0.0005579060982191041,
+      "loss": 3.9587,
       "step": 6600
     },
     {
-      "epoch": 0.7157464212678937,
-      "grad_norm": 0.7968887686729431,
-      "learning_rate": 0.0005576683547031569,
-      "loss": 3.963,
+      "epoch": 0.7169811320754716,
+      "grad_norm": 0.6285462379455566,
+      "learning_rate": 0.0005575822989746357,
+      "loss": 3.9703,
       "step": 6650
     },
     {
-      "epoch": 0.7211279733075019,
-      "grad_norm": 0.6040535569190979,
-      "learning_rate": 0.0005573451136730956,
-      "loss": 3.9357,
+      "epoch": 0.7223719676549866,
+      "grad_norm": 0.7304993867874146,
+      "learning_rate": 0.0005572584997301672,
+      "loss": 3.9348,
       "step": 6700
     },
     {
-      "epoch": 0.7265095253471101,
-      "grad_norm": 0.6244126558303833,
-      "learning_rate": 0.0005570218726430341,
-      "loss": 3.9461,
+      "epoch": 0.7277628032345014,
+      "grad_norm": 0.5624628663063049,
+      "learning_rate": 0.0005569347004856989,
+      "loss": 3.9602,
       "step": 6750
     },
     {
-      "epoch": 0.7318910773867183,
-      "grad_norm": 0.6094521284103394,
-      "learning_rate": 0.0005566986316129728,
-      "loss": 3.9556,
+      "epoch": 0.7331536388140162,
+      "grad_norm": 0.6064887642860413,
+      "learning_rate": 0.0005566109012412303,
+      "loss": 3.9331,
       "step": 6800
     },
     {
-      "epoch": 0.7372726294263265,
-      "grad_norm": 0.6165975332260132,
-      "learning_rate": 0.0005563753905829113,
-      "loss": 3.9451,
+      "epoch": 0.738544474393531,
+      "grad_norm": 0.5876811146736145,
+      "learning_rate": 0.000556287101996762,
+      "loss": 3.9323,
       "step": 6850
     },
     {
-      "epoch": 0.7426541814659348,
-      "grad_norm": 0.6186493635177612,
-      "learning_rate": 0.0005560521495528498,
-      "loss": 3.9314,
+      "epoch": 0.7439353099730458,
+      "grad_norm": 0.5871797204017639,
+      "learning_rate": 0.0005559633027522935,
+      "loss": 3.9329,
       "step": 6900
     },
     {
-      "epoch": 0.748035733505543,
-      "grad_norm": 0.60491943359375,
-      "learning_rate": 0.0005557289085227884,
-      "loss": 3.9439,
+      "epoch": 0.7493261455525606,
+      "grad_norm": 0.6020776629447937,
+      "learning_rate": 0.0005556395035078251,
+      "loss": 3.9362,
       "step": 6950
     },
     {
-      "epoch": 0.7534172855451512,
-      "grad_norm": 0.5608195662498474,
-      "learning_rate": 0.000555405667492727,
-      "loss": 3.9329,
+      "epoch": 0.7547169811320755,
+      "grad_norm": 0.6208156943321228,
+      "learning_rate": 0.0005553157042633566,
+      "loss": 3.9191,
       "step": 7000
     },
     {
-      "epoch": 0.7534172855451512,
-      "eval_accuracy": 0.3331331222787205,
-      "eval_loss": 3.8616700172424316,
-      "eval_runtime": 184.9857,
-      "eval_samples_per_second": 97.364,
-      "eval_steps_per_second": 6.087,
+      "epoch": 0.7547169811320755,
+      "eval_accuracy": 0.33344278300899177,
+      "eval_loss": 3.8618791103363037,
+      "eval_runtime": 183.7498,
+      "eval_samples_per_second": 98.019,
+      "eval_steps_per_second": 6.128,
       "step": 7000
     },
     {
-      "epoch": 0.7587988375847594,
-      "grad_norm": 0.6202792525291443,
-      "learning_rate": 0.0005550824264626657,
-      "loss": 3.9319,
+      "epoch": 0.7601078167115903,
+      "grad_norm": 0.6798801422119141,
+      "learning_rate": 0.0005549919050188882,
+      "loss": 3.9331,
       "step": 7050
     },
     {
-      "epoch": 0.7641803896243676,
-      "grad_norm": 0.5904057025909424,
-      "learning_rate": 0.0005547591854326042,
-      "loss": 3.9246,
+      "epoch": 0.7654986522911051,
+      "grad_norm": 1.4628221988677979,
+      "learning_rate": 0.0005546681057744198,
+      "loss": 3.906,
       "step": 7100
     },
     {
-      "epoch": 0.7695619416639758,
-      "grad_norm": 0.5333248972892761,
-      "learning_rate": 0.0005544359444025428,
-      "loss": 3.9287,
+      "epoch": 0.77088948787062,
+      "grad_norm": 0.6019647121429443,
+      "learning_rate": 0.0005543443065299514,
+      "loss": 3.9351,
       "step": 7150
     },
     {
-      "epoch": 0.7749434937035841,
-      "grad_norm": 0.6606253981590271,
-      "learning_rate": 0.0005541127033724813,
-      "loss": 3.9263,
+      "epoch": 0.7762803234501348,
+      "grad_norm": 0.5555400252342224,
+      "learning_rate": 0.000554020507285483,
+      "loss": 3.9058,
       "step": 7200
     },
     {
-      "epoch": 0.7803250457431924,
-      "grad_norm": 0.6061251759529114,
-      "learning_rate": 0.0005537894623424199,
-      "loss": 3.9207,
+      "epoch": 0.7816711590296496,
+      "grad_norm": 0.6362798810005188,
+      "learning_rate": 0.0005536967080410145,
+      "loss": 3.8967,
       "step": 7250
     },
     {
-      "epoch": 0.7857065977828006,
-      "grad_norm": 0.5618706345558167,
-      "learning_rate": 0.0005534662213123586,
-      "loss": 3.9275,
+      "epoch": 0.7870619946091644,
+      "grad_norm": 0.5722786784172058,
+      "learning_rate": 0.0005533729087965462,
+      "loss": 3.9111,
       "step": 7300
     },
     {
-      "epoch": 0.7910881498224088,
-      "grad_norm": 0.5616447329521179,
-      "learning_rate": 0.0005531429802822971,
-      "loss": 3.9249,
+      "epoch": 0.7924528301886793,
+      "grad_norm": 0.580388069152832,
+      "learning_rate": 0.0005530491095520777,
+      "loss": 3.9099,
       "step": 7350
     },
     {
-      "epoch": 0.796469701862017,
-      "grad_norm": 0.6080626845359802,
-      "learning_rate": 0.0005528197392522357,
-      "loss": 3.9069,
+      "epoch": 0.7978436657681941,
+      "grad_norm": 0.5334943532943726,
+      "learning_rate": 0.0005527253103076093,
+      "loss": 3.9072,
       "step": 7400
     },
     {
-      "epoch": 0.8018512539016253,
-      "grad_norm": 0.6487952470779419,
-      "learning_rate": 0.0005524964982221743,
-      "loss": 3.9027,
+      "epoch": 0.8032345013477089,
+      "grad_norm": 0.6169307827949524,
+      "learning_rate": 0.0005524015110631408,
+      "loss": 3.8832,
       "step": 7450
     },
     {
-      "epoch": 0.8072328059412335,
-      "grad_norm": 0.5572729706764221,
-      "learning_rate": 0.0005521732571921129,
-      "loss": 3.8851,
+      "epoch": 0.8086253369272237,
+      "grad_norm": 0.6007163524627686,
+      "learning_rate": 0.0005520777118186724,
+      "loss": 3.9024,
       "step": 7500
     },
     {
-      "epoch": 0.8126143579808417,
-      "grad_norm": 0.6633656620979309,
-      "learning_rate": 0.0005518500161620514,
-      "loss": 3.9135,
+      "epoch": 0.8140161725067385,
+      "grad_norm": 0.5735055804252625,
+      "learning_rate": 0.0005517539125742039,
+      "loss": 3.8979,
       "step": 7550
     },
     {
-      "epoch": 0.8179959100204499,
-      "grad_norm": 0.5475841164588928,
-      "learning_rate": 0.00055152677513199,
-      "loss": 3.9074,
+      "epoch": 0.8194070080862533,
+      "grad_norm": 0.5585691928863525,
+      "learning_rate": 0.0005514365893146249,
+      "loss": 3.8902,
       "step": 7600
     },
     {
-      "epoch": 0.8233774620600581,
-      "grad_norm": 0.5891292691230774,
-      "learning_rate": 0.0005512035341019286,
-      "loss": 3.9187,
+      "epoch": 0.8247978436657682,
+      "grad_norm": 0.5613856315612793,
+      "learning_rate": 0.0005511127900701564,
+      "loss": 3.8952,
       "step": 7650
     },
     {
-      "epoch": 0.8287590140996663,
-      "grad_norm": 0.6392335891723633,
-      "learning_rate": 0.0005508802930718672,
-      "loss": 3.8935,
+      "epoch": 0.8301886792452831,
+      "grad_norm": 0.5988004803657532,
+      "learning_rate": 0.000550788990825688,
+      "loss": 3.8894,
       "step": 7700
     },
     {
-      "epoch": 0.8341405661392746,
-      "grad_norm": 0.6276848316192627,
-      "learning_rate": 0.0005505570520418058,
-      "loss": 3.8951,
+      "epoch": 0.8355795148247979,
+      "grad_norm": 0.5915306210517883,
+      "learning_rate": 0.0005504651915812196,
+      "loss": 3.8922,
       "step": 7750
     },
     {
-      "epoch": 0.8395221181788828,
-      "grad_norm": 0.5669248700141907,
-      "learning_rate": 0.0005502338110117443,
-      "loss": 3.9057,
+      "epoch": 0.8409703504043127,
+      "grad_norm": 0.7122664451599121,
+      "learning_rate": 0.0005501413923367512,
+      "loss": 3.8766,
       "step": 7800
     },
     {
-      "epoch": 0.844903670218491,
-      "grad_norm": 0.666827380657196,
-      "learning_rate": 0.000549910569981683,
-      "loss": 3.8974,
+      "epoch": 0.8463611859838275,
+      "grad_norm": 0.5410688519477844,
+      "learning_rate": 0.0005498175930922827,
+      "loss": 3.8794,
       "step": 7850
     },
     {
-      "epoch": 0.8502852222580992,
-      "grad_norm": 0.6152955293655396,
-      "learning_rate": 0.0005495873289516215,
-      "loss": 3.9008,
+      "epoch": 0.8517520215633423,
+      "grad_norm": 0.5719017386436462,
+      "learning_rate": 0.0005494937938478143,
+      "loss": 3.8727,
       "step": 7900
     },
     {
-      "epoch": 0.8556667742977074,
-      "grad_norm": 0.5386107563972473,
-      "learning_rate": 0.0005492640879215602,
-      "loss": 3.8832,
+      "epoch": 0.8571428571428571,
+      "grad_norm": 0.570475697517395,
+      "learning_rate": 0.0005491699946033459,
+      "loss": 3.9063,
       "step": 7950
     },
     {
-      "epoch": 0.8610483263373157,
-      "grad_norm": 0.553949236869812,
-      "learning_rate": 0.0005489408468914987,
-      "loss": 3.8799,
+      "epoch": 0.862533692722372,
+      "grad_norm": 0.5836682915687561,
+      "learning_rate": 0.0005488461953588775,
+      "loss": 3.8542,
       "step": 8000
     },
     {
-      "epoch": 0.8610483263373157,
-      "eval_accuracy": 0.3376146192895166,
-      "eval_loss": 3.817253589630127,
-      "eval_runtime": 184.9597,
-      "eval_samples_per_second": 97.378,
-      "eval_steps_per_second": 6.088,
+      "epoch": 0.862533692722372,
+      "eval_accuracy": 0.33794155582895036,
+      "eval_loss": 3.816067695617676,
+      "eval_runtime": 183.5803,
+      "eval_samples_per_second": 98.11,
+      "eval_steps_per_second": 6.134,
       "step": 8000
     },
     {
-      "epoch": 0.8664298783769239,
-      "grad_norm": 0.5128387212753296,
-      "learning_rate": 0.0005486176058614372,
-      "loss": 3.8821,
+      "epoch": 0.8679245283018868,
+      "grad_norm": 0.6498636603355408,
+      "learning_rate": 0.000548522396114409,
+      "loss": 3.895,
       "step": 8050
     },
     {
-      "epoch": 0.8718114304165321,
-      "grad_norm": 0.6132450699806213,
-      "learning_rate": 0.0005482943648313759,
-      "loss": 3.8667,
+      "epoch": 0.8733153638814016,
+      "grad_norm": 0.6250914931297302,
+      "learning_rate": 0.0005481985968699406,
+      "loss": 3.8735,
       "step": 8100
     },
     {
-      "epoch": 0.8771929824561403,
-      "grad_norm": 0.6007569432258606,
-      "learning_rate": 0.0005479711238013145,
-      "loss": 3.8853,
+      "epoch": 0.8787061994609164,
+      "grad_norm": 0.5511185526847839,
+      "learning_rate": 0.0005478747976254721,
+      "loss": 3.8758,
       "step": 8150
     },
     {
-      "epoch": 0.8825745344957485,
-      "grad_norm": 0.599909782409668,
-      "learning_rate": 0.0005476478827712531,
-      "loss": 3.871,
+      "epoch": 0.8840970350404312,
+      "grad_norm": 0.6440311074256897,
+      "learning_rate": 0.0005475509983810037,
+      "loss": 3.869,
       "step": 8200
     },
     {
-      "epoch": 0.8879560865353568,
-      "grad_norm": 0.6086229681968689,
-      "learning_rate": 0.0005473246417411916,
-      "loss": 3.8779,
+      "epoch": 0.889487870619946,
+      "grad_norm": 0.5398238301277161,
+      "learning_rate": 0.0005472271991365352,
+      "loss": 3.866,
       "step": 8250
     },
     {
-      "epoch": 0.8933376385749651,
-      "grad_norm": 0.6416434645652771,
-      "learning_rate": 0.0005470014007111302,
-      "loss": 3.8791,
+      "epoch": 0.894878706199461,
+      "grad_norm": 0.5822187066078186,
+      "learning_rate": 0.0005469033998920669,
+      "loss": 3.8864,
       "step": 8300
     },
     {
-      "epoch": 0.8987191906145733,
-      "grad_norm": 0.549599289894104,
-      "learning_rate": 0.0005466781596810688,
-      "loss": 3.8874,
+      "epoch": 0.9002695417789758,
+      "grad_norm": 0.5405433773994446,
+      "learning_rate": 0.0005465796006475984,
+      "loss": 3.857,
       "step": 8350
     },
     {
-      "epoch": 0.9041007426541815,
-      "grad_norm": 0.5874208211898804,
-      "learning_rate": 0.0005463549186510073,
-      "loss": 3.8738,
+      "epoch": 0.9056603773584906,
+      "grad_norm": 0.5735152959823608,
+      "learning_rate": 0.00054625580140313,
+      "loss": 3.8628,
       "step": 8400
     },
     {
-      "epoch": 0.9094822946937897,
-      "grad_norm": 0.5840559005737305,
-      "learning_rate": 0.000546031677620946,
-      "loss": 3.864,
+      "epoch": 0.9110512129380054,
+      "grad_norm": 0.5623797178268433,
+      "learning_rate": 0.0005459320021586615,
+      "loss": 3.858,
       "step": 8450
     },
     {
-      "epoch": 0.9148638467333979,
-      "grad_norm": 0.5824464559555054,
-      "learning_rate": 0.0005457084365908845,
-      "loss": 3.8568,
+      "epoch": 0.9164420485175202,
+      "grad_norm": 0.5893810391426086,
+      "learning_rate": 0.0005456082029141932,
+      "loss": 3.8638,
       "step": 8500
     },
     {
-      "epoch": 0.9202453987730062,
-      "grad_norm": 0.6089414358139038,
-      "learning_rate": 0.0005453851955608232,
-      "loss": 3.8317,
+      "epoch": 0.921832884097035,
+      "grad_norm": 0.6306450366973877,
+      "learning_rate": 0.0005452844036697248,
+      "loss": 3.8565,
       "step": 8550
     },
     {
-      "epoch": 0.9256269508126144,
-      "grad_norm": 0.5441460609436035,
-      "learning_rate": 0.0005450619545307617,
-      "loss": 3.8802,
+      "epoch": 0.9272237196765498,
+      "grad_norm": 0.5499542951583862,
+      "learning_rate": 0.0005449606044252563,
+      "loss": 3.8639,
       "step": 8600
     },
     {
-      "epoch": 0.9310085028522226,
-      "grad_norm": 0.5227726697921753,
-      "learning_rate": 0.0005447387135007003,
-      "loss": 3.8594,
+      "epoch": 0.9326145552560647,
+      "grad_norm": 0.563671886920929,
+      "learning_rate": 0.0005446368051807879,
+      "loss": 3.844,
       "step": 8650
     },
     {
-      "epoch": 0.9363900548918308,
-      "grad_norm": 0.5715673565864563,
-      "learning_rate": 0.0005444154724706389,
-      "loss": 3.8604,
+      "epoch": 0.9380053908355795,
+      "grad_norm": 0.5789870619773865,
+      "learning_rate": 0.0005443130059363194,
+      "loss": 3.8492,
       "step": 8700
     },
     {
-      "epoch": 0.941771606931439,
-      "grad_norm": 0.5205848813056946,
-      "learning_rate": 0.0005440922314405775,
-      "loss": 3.8593,
+      "epoch": 0.9433962264150944,
+      "grad_norm": 0.5480877757072449,
+      "learning_rate": 0.0005439892066918511,
+      "loss": 3.8458,
       "step": 8750
     },
     {
-      "epoch": 0.9471531589710472,
-      "grad_norm": 0.6547507047653198,
-      "learning_rate": 0.0005437689904105161,
-      "loss": 3.8436,
+      "epoch": 0.9487870619946092,
+      "grad_norm": 0.5352542996406555,
+      "learning_rate": 0.0005436654074473825,
+      "loss": 3.8487,
       "step": 8800
     },
     {
-      "epoch": 0.9525347110106555,
-      "grad_norm": 0.61313396692276,
-      "learning_rate": 0.0005434457493804546,
-      "loss": 3.848,
+      "epoch": 0.954177897574124,
+      "grad_norm": 0.5493354797363281,
+      "learning_rate": 0.0005433416082029142,
+      "loss": 3.8454,
       "step": 8850
     },
     {
-      "epoch": 0.9579162630502637,
-      "grad_norm": 0.644686758518219,
-      "learning_rate": 0.0005431225083503932,
-      "loss": 3.844,
+      "epoch": 0.9595687331536388,
+      "grad_norm": 0.5753820538520813,
+      "learning_rate": 0.0005430178089584457,
+      "loss": 3.8372,
       "step": 8900
     },
     {
-      "epoch": 0.9632978150898719,
-      "grad_norm": 0.5130710005760193,
-      "learning_rate": 0.0005427992673203318,
-      "loss": 3.8736,
+      "epoch": 0.9649595687331537,
+      "grad_norm": 0.5328227877616882,
+      "learning_rate": 0.0005426940097139773,
+      "loss": 3.8276,
       "step": 8950
     },
     {
-      "epoch": 0.9686793671294801,
-      "grad_norm": 0.5622977018356323,
-      "learning_rate": 0.0005424760262902704,
-      "loss": 3.8502,
+      "epoch": 0.9703504043126685,
+      "grad_norm": 0.5667588114738464,
+      "learning_rate": 0.0005423702104695088,
+      "loss": 3.8531,
       "step": 9000
     },
     {
-      "epoch": 0.9686793671294801,
-      "eval_accuracy": 0.3408010825304519,
-      "eval_loss": 3.7829928398132324,
-      "eval_runtime": 184.3628,
-      "eval_samples_per_second": 97.693,
-      "eval_steps_per_second": 6.108,
+      "epoch": 0.9703504043126685,
+      "eval_accuracy": 0.34165379039401994,
+      "eval_loss": 3.7784414291381836,
+      "eval_runtime": 184.2031,
+      "eval_samples_per_second": 97.778,
+      "eval_steps_per_second": 6.113,
       "step": 9000
     },
     {
-      "epoch": 0.9740609191690883,
-      "grad_norm": 0.5300456285476685,
-      "learning_rate": 0.000542152785260209,
-      "loss": 3.8365,
+      "epoch": 0.9757412398921833,
+      "grad_norm": 0.6010326743125916,
+      "learning_rate": 0.0005420464112250404,
+      "loss": 3.8678,
       "step": 9050
     },
     {
-      "epoch": 0.9794424712086965,
-      "grad_norm": 0.5774227976799011,
-      "learning_rate": 0.0005418295442301476,
-      "loss": 3.8359,
+      "epoch": 0.9811320754716981,
+      "grad_norm": 0.5293521285057068,
+      "learning_rate": 0.000541722611980572,
+      "loss": 3.8241,
       "step": 9100
     },
     {
-      "epoch": 0.9848240232483048,
-      "grad_norm": 0.5969952344894409,
-      "learning_rate": 0.0005415063032000861,
-      "loss": 3.8564,
+      "epoch": 0.9865229110512129,
+      "grad_norm": 0.5834252238273621,
+      "learning_rate": 0.0005413988127361036,
+      "loss": 3.8378,
       "step": 9150
     },
     {
-      "epoch": 0.9902055752879131,
-      "grad_norm": 0.6162004470825195,
-      "learning_rate": 0.0005411830621700248,
-      "loss": 3.8361,
+      "epoch": 0.9919137466307277,
+      "grad_norm": 0.5734151005744934,
+      "learning_rate": 0.0005410750134916351,
+      "loss": 3.8295,
       "step": 9200
     },
     {
-      "epoch": 0.9955871273275213,
-      "grad_norm": 0.5422845482826233,
-      "learning_rate": 0.0005408598211399633,
-      "loss": 3.8151,
+      "epoch": 0.9973045822102425,
+      "grad_norm": 0.5477326512336731,
+      "learning_rate": 0.0005407512142471667,
+      "loss": 3.8451,
       "step": 9250
     },
     {
-      "epoch": 1.0009686793671295,
-      "grad_norm": 0.5742968916893005,
-      "learning_rate": 0.0005405365801099019,
-      "loss": 3.8236,
+      "epoch": 1.0026954177897573,
+      "grad_norm": 0.5776128172874451,
+      "learning_rate": 0.0005404274150026983,
+      "loss": 3.7938,
       "step": 9300
     },
     {
-      "epoch": 1.0063502314067376,
-      "grad_norm": 0.5724886655807495,
-      "learning_rate": 0.0005402133390798405,
-      "loss": 3.7718,
+      "epoch": 1.0080862533692723,
+      "grad_norm": 0.5947125554084778,
+      "learning_rate": 0.0005401036157582299,
+      "loss": 3.7605,
       "step": 9350
     },
     {
-      "epoch": 1.011731783446346,
-      "grad_norm": 0.6036879420280457,
-      "learning_rate": 0.000539890098049779,
-      "loss": 3.7636,
+      "epoch": 1.013477088948787,
+      "grad_norm": 0.6069636940956116,
+      "learning_rate": 0.0005397798165137614,
+      "loss": 3.7525,
       "step": 9400
     },
     {
-      "epoch": 1.017113335485954,
-      "grad_norm": 0.5474029779434204,
-      "learning_rate": 0.0005395668570197177,
-      "loss": 3.7549,
+      "epoch": 1.0188679245283019,
+      "grad_norm": 0.5255810618400574,
+      "learning_rate": 0.000539456017269293,
+      "loss": 3.7525,
       "step": 9450
     },
     {
-      "epoch": 1.0224948875255624,
-      "grad_norm": 0.5884007811546326,
-      "learning_rate": 0.0005392436159896562,
-      "loss": 3.7614,
+      "epoch": 1.0242587601078168,
+      "grad_norm": 0.546853244304657,
+      "learning_rate": 0.0005391322180248245,
+      "loss": 3.7576,
       "step": 9500
     },
     {
-      "epoch": 1.0278764395651705,
-      "grad_norm": 0.5328190326690674,
-      "learning_rate": 0.0005389203749595948,
-      "loss": 3.7569,
+      "epoch": 1.0296495956873315,
+      "grad_norm": 0.5615027546882629,
+      "learning_rate": 0.0005388084187803561,
+      "loss": 3.7679,
       "step": 9550
     },
     {
-      "epoch": 1.0332579916047788,
-      "grad_norm": 0.5494911670684814,
-      "learning_rate": 0.0005385971339295334,
-      "loss": 3.7681,
+      "epoch": 1.0350404312668464,
+      "grad_norm": 0.579430341720581,
+      "learning_rate": 0.0005384846195358876,
+      "loss": 3.7661,
       "step": 9600
     },
     {
-      "epoch": 1.0386395436443872,
-      "grad_norm": 0.6018995046615601,
-      "learning_rate": 0.000538273892899472,
-      "loss": 3.7921,
+      "epoch": 1.0404312668463611,
+      "grad_norm": 0.6099395751953125,
+      "learning_rate": 0.0005381608202914193,
+      "loss": 3.7707,
       "step": 9650
     },
     {
-      "epoch": 1.0440210956839953,
-      "grad_norm": 0.6208644509315491,
-      "learning_rate": 0.0005379506518694106,
-      "loss": 3.7763,
+      "epoch": 1.045822102425876,
+      "grad_norm": 0.5670583248138428,
+      "learning_rate": 0.0005378370210469509,
+      "loss": 3.7544,
       "step": 9700
     },
     {
-      "epoch": 1.0494026477236036,
-      "grad_norm": 0.6352260112762451,
-      "learning_rate": 0.0005376274108393491,
-      "loss": 3.7876,
+      "epoch": 1.0512129380053907,
+      "grad_norm": 0.5511836409568787,
+      "learning_rate": 0.0005375132218024824,
+      "loss": 3.7525,
       "step": 9750
     },
     {
-      "epoch": 1.0547841997632117,
-      "grad_norm": 0.565929114818573,
-      "learning_rate": 0.0005373041698092877,
-      "loss": 3.7864,
+      "epoch": 1.0566037735849056,
+      "grad_norm": 0.5288443565368652,
+      "learning_rate": 0.000537189422558014,
+      "loss": 3.7805,
       "step": 9800
     },
     {
-      "epoch": 1.06016575180282,
-      "grad_norm": 0.5776049494743347,
-      "learning_rate": 0.0005369809287792263,
-      "loss": 3.7535,
+      "epoch": 1.0619946091644206,
+      "grad_norm": 0.5281191468238831,
+      "learning_rate": 0.0005368656233135455,
+      "loss": 3.7818,
       "step": 9850
     },
     {
-      "epoch": 1.0655473038424281,
-      "grad_norm": 0.5360944867134094,
-      "learning_rate": 0.000536657687749165,
-      "loss": 3.7739,
+      "epoch": 1.0673854447439353,
+      "grad_norm": 0.5530230402946472,
+      "learning_rate": 0.0005365483000539665,
+      "loss": 3.7735,
       "step": 9900
     },
     {
-      "epoch": 1.0709288558820365,
-      "grad_norm": 0.5402177572250366,
-      "learning_rate": 0.0005363344467191035,
-      "loss": 3.7569,
+      "epoch": 1.0727762803234502,
+      "grad_norm": 0.6312490701675415,
+      "learning_rate": 0.0005362245008094981,
+      "loss": 3.7825,
       "step": 9950
     },
     {
-      "epoch": 1.0763104079216446,
-      "grad_norm": 0.5348372459411621,
-      "learning_rate": 0.000536011205689042,
-      "loss": 3.7597,
+      "epoch": 1.0781671159029649,
+      "grad_norm": 0.5409021377563477,
+      "learning_rate": 0.0005359007015650297,
+      "loss": 3.7613,
       "step": 10000
     },
     {
-      "epoch": 1.0763104079216446,
-      "eval_accuracy": 0.3438087031180445,
-      "eval_loss": 3.7523183822631836,
-      "eval_runtime": 185.331,
-      "eval_samples_per_second": 97.183,
-      "eval_steps_per_second": 6.076,
+      "epoch": 1.0781671159029649,
+      "eval_accuracy": 0.3451733833889944,
+      "eval_loss": 3.7442734241485596,
+      "eval_runtime": 183.7504,
+      "eval_samples_per_second": 98.019,
+      "eval_steps_per_second": 6.128,
       "step": 10000
     }
   ],
   "logging_steps": 50,
-  "max_steps": 92910,
+  "max_steps": 92750,
   "num_input_tokens_seen": 0,
   "num_train_epochs": 10,
   "save_steps": 10000,
@@ -1516,7 +1516,7 @@
       "attributes": {}
     }
   },
-  "total_flos": 8.3609792151552e+16,
+  "total_flos": 8.36069179392e+16,
   "train_batch_size": 32,
   "trial_name": null,
   "trial_params": null

checkpoint-10000/training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:097b2f4878fa458807831f34b125c9126dbb86ed90dd7b8414ba94f45393ff0f
+oid sha256:d82950a3fe07264a227b14657436d7461495ae286247a2020c333c91d4e11e02
 size 5304