Upload checkpoint-875 (shard 3, 0:506)

adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ad9957c6732464d4155353123efc916169f2a1d59b6df5402f3b29b7575ea956
+oid sha256:966b222a3b7f9bdc4e973aa5f6308cbfbda82657a2d60102bcab79bd8510174b
 size 154116312

optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d8f9c2caca1a25b43618b00293b9db43f57882645ddebe56f1f8f239d4a6bfb7
+oid sha256:a20efd67b670b3358df43dda9be51461cce1ffcb8b3a06b239b7b692fafd946e
 size 308668611

trainer_state.json

@@ -11,1752 +11,1752 @@
   "log_history": [
     {
       "epoch": 0.007905138339920948,
-      "grad_norm": 37.5876579284668,
+      "grad_norm": 29.443620681762695,
       "learning_rate": 0.0,
       "loss": 6.8345,
       "step": 1
     },
     {
       "epoch": 0.015810276679841896,
-      "grad_norm": 14.87977409362793,
+      "grad_norm": 14.875253677368164,
       "learning_rate": 2.5e-05,
       "loss": 6.6279,
       "step": 2
     },
     {
       "epoch": 0.023715415019762844,
-      "grad_norm": 16.453094482421875,
+      "grad_norm": 60.56179428100586,
       "learning_rate": 5e-05,
-      "loss": 6.7811,
+      "loss": 6.7804,
       "step": 3
     },
     {
       "epoch": 0.03162055335968379,
-      "grad_norm": 9.101632118225098,
+      "grad_norm": 8.8980712890625,
       "learning_rate": 7.500000000000001e-05,
-      "loss": 5.66,
+      "loss": 5.7093,
       "step": 4
     },
     {
       "epoch": 0.039525691699604744,
-      "grad_norm": 5.45641565322876,
+      "grad_norm": 11.861969947814941,
       "learning_rate": 0.0001,
-      "loss": 4.8428,
+      "loss": 4.9035,
       "step": 5
     },
     {
       "epoch": 0.04743083003952569,
-      "grad_norm": 4.674358367919922,
+      "grad_norm": 5.465580940246582,
       "learning_rate": 0.000125,
-      "loss": 4.0043,
+      "loss": 4.0769,
       "step": 6
     },
     {
       "epoch": 0.05533596837944664,
-      "grad_norm": 4.011600017547607,
+      "grad_norm": 3.910144329071045,
       "learning_rate": 0.00015000000000000001,
-      "loss": 4.102,
+      "loss": 4.1908,
       "step": 7
     },
     {
       "epoch": 0.06324110671936758,
-      "grad_norm": 2.9474167823791504,
+      "grad_norm": 2.9628067016601562,
       "learning_rate": 0.000175,
-      "loss": 3.0249,
+      "loss": 3.0805,
       "step": 8
     },
     {
       "epoch": 0.07114624505928854,
-      "grad_norm": 2.8972344398498535,
+      "grad_norm": 3.2732276916503906,
       "learning_rate": 0.0002,
-      "loss": 3.0272,
+      "loss": 3.063,
       "step": 9
     },
     {
       "epoch": 0.07905138339920949,
-      "grad_norm": 2.332426071166992,
+      "grad_norm": 2.3448641300201416,
       "learning_rate": 0.0001999915737775817,
-      "loss": 2.4029,
+      "loss": 2.4361,
       "step": 10
     },
     {
       "epoch": 0.08695652173913043,
-      "grad_norm": 2.9309306144714355,
+      "grad_norm": 3.226670265197754,
       "learning_rate": 0.00019996629653035126,
-      "loss": 2.1172,
+      "loss": 2.142,
       "step": 11
     },
     {
       "epoch": 0.09486166007905138,
-      "grad_norm": 2.907116174697876,
+      "grad_norm": 2.9033336639404297,
       "learning_rate": 0.00019992417251814282,
-      "loss": 1.8977,
+      "loss": 1.928,
       "step": 12
     },
     {
       "epoch": 0.10276679841897234,
-      "grad_norm": 2.2385010719299316,
+      "grad_norm": 2.5437309741973877,
       "learning_rate": 0.00019986520883988232,
-      "loss": 1.463,
+      "loss": 1.4877,
       "step": 13
     },
     {
       "epoch": 0.11067193675889328,
-      "grad_norm": 2.3392879962921143,
+      "grad_norm": 2.364232301712036,
       "learning_rate": 0.0001997894154323911,
-      "loss": 1.3303,
+      "loss": 1.3451,
       "step": 14
     },
     {
       "epoch": 0.11857707509881422,
-      "grad_norm": 2.8780765533447266,
+      "grad_norm": 2.6934092044830322,
       "learning_rate": 0.00019969680506871137,
-      "loss": 1.1526,
+      "loss": 1.1716,
       "step": 15
     },
     {
       "epoch": 0.12648221343873517,
-      "grad_norm": 2.6122989654541016,
+      "grad_norm": 2.518855333328247,
       "learning_rate": 0.0001995873933559535,
-      "loss": 0.9236,
+      "loss": 0.9187,
       "step": 16
     },
     {
       "epoch": 0.13438735177865613,
-      "grad_norm": 3.0627999305725098,
+      "grad_norm": 2.960550546646118,
       "learning_rate": 0.00019946119873266613,
-      "loss": 0.7549,
+      "loss": 0.7536,
       "step": 17
     },
     {
       "epoch": 0.1422924901185771,
-      "grad_norm": 2.3931822776794434,
+      "grad_norm": 2.3193135261535645,
       "learning_rate": 0.0001993182424657285,
-      "loss": 0.782,
+      "loss": 0.7687,
       "step": 18
     },
     {
       "epoch": 0.15019762845849802,
-      "grad_norm": 1.717180848121643,
+      "grad_norm": 1.7276065349578857,
       "learning_rate": 0.00019915854864676664,
-      "loss": 0.7529,
+      "loss": 0.7496,
       "step": 19
     },
     {
       "epoch": 0.15810276679841898,
-      "grad_norm": 2.4214677810668945,
+      "grad_norm": 1.7833150625228882,
       "learning_rate": 0.0001989821441880933,
-      "loss": 0.6544,
+      "loss": 0.6497,
       "step": 20
     },
     {
       "epoch": 0.16600790513833993,
-      "grad_norm": 1.7936805486679077,
+      "grad_norm": 1.5743191242218018,
       "learning_rate": 0.00019878905881817252,
-      "loss": 0.6009,
+      "loss": 0.5612,
       "step": 21
     },
     {
       "epoch": 0.17391304347826086,
-      "grad_norm": 3.7155938148498535,
+      "grad_norm": 1.4150418043136597,
       "learning_rate": 0.0001985793250766098,
-      "loss": 0.6768,
+      "loss": 0.6368,
       "step": 22
     },
     {
       "epoch": 0.18181818181818182,
-      "grad_norm": 1.343880534172058,
+      "grad_norm": 2.5285849571228027,
       "learning_rate": 0.00019835297830866826,
-      "loss": 0.5198,
+      "loss": 0.51,
       "step": 23
     },
     {
       "epoch": 0.18972332015810275,
-      "grad_norm": 1.1089560985565186,
+      "grad_norm": 1.3967912197113037,
       "learning_rate": 0.00019811005665931205,
-      "loss": 0.425,
+      "loss": 0.4095,
       "step": 24
     },
     {
       "epoch": 0.1976284584980237,
-      "grad_norm": 2.3267531394958496,
+      "grad_norm": 1.952337384223938,
       "learning_rate": 0.00019785060106677818,
-      "loss": 0.8532,
+      "loss": 0.8365,
       "step": 25
     },
     {
       "epoch": 0.20553359683794467,
-      "grad_norm": 1.536291241645813,
+      "grad_norm": 1.570603609085083,
       "learning_rate": 0.0001975746552556772,
-      "loss": 0.4212,
+      "loss": 0.3903,
       "step": 26
     },
     {
       "epoch": 0.2134387351778656,
-      "grad_norm": 1.1441535949707031,
+      "grad_norm": 1.3032807111740112,
       "learning_rate": 0.00019728226572962473,
-      "loss": 0.4482,
+      "loss": 0.4507,
       "step": 27
     },
     {
       "epoch": 0.22134387351778656,
-      "grad_norm": 1.1318137645721436,
+      "grad_norm": 1.129608392715454,
       "learning_rate": 0.0001969734817634044,
-      "loss": 0.6573,
+      "loss": 0.6392,
       "step": 28
     },
     {
       "epoch": 0.22924901185770752,
-      "grad_norm": 1.5648777484893799,
+      "grad_norm": 2.1530044078826904,
       "learning_rate": 0.0001966483553946637,
-      "loss": 0.6438,
+      "loss": 0.5927,
       "step": 29
     },
     {
       "epoch": 0.23715415019762845,
-      "grad_norm": 1.3132728338241577,
+      "grad_norm": 1.7631202936172485,
       "learning_rate": 0.00019630694141514464,
-      "loss": 0.3222,
+      "loss": 0.317,
       "step": 30
     },
     {
       "epoch": 0.2450592885375494,
-      "grad_norm": 1.2373923063278198,
+      "grad_norm": 3.5597596168518066,
       "learning_rate": 0.00019594929736144976,
-      "loss": 0.3504,
+      "loss": 0.3338,
       "step": 31
     },
     {
       "epoch": 0.25296442687747034,
-      "grad_norm": 1.2690372467041016,
+      "grad_norm": 1.4273171424865723,
       "learning_rate": 0.0001955754835053459,
-      "loss": 0.724,
+      "loss": 0.7187,
       "step": 32
     },
     {
       "epoch": 0.2608695652173913,
-      "grad_norm": 1.5825968980789185,
+      "grad_norm": 2.1714446544647217,
       "learning_rate": 0.00019518556284360696,
-      "loss": 0.4321,
+      "loss": 0.4304,
       "step": 33
     },
     {
       "epoch": 0.26877470355731226,
-      "grad_norm": 1.175012230873108,
+      "grad_norm": 1.3923094272613525,
       "learning_rate": 0.0001947796010873974,
-      "loss": 0.3895,
+      "loss": 0.3955,
       "step": 34
     },
     {
       "epoch": 0.2766798418972332,
-      "grad_norm": 1.0333178043365479,
+      "grad_norm": 1.089962124824524,
       "learning_rate": 0.0001943576666511982,
-      "loss": 0.5972,
+      "loss": 0.6046,
       "step": 35
     },
     {
       "epoch": 0.2845849802371542,
-      "grad_norm": 1.1397666931152344,
+      "grad_norm": 0.9553408622741699,
       "learning_rate": 0.0001939198306412775,
-      "loss": 0.6173,
+      "loss": 0.6164,
       "step": 36
     },
     {
       "epoch": 0.2924901185770751,
-      "grad_norm": 1.1599067449569702,
+      "grad_norm": 0.9228238463401794,
       "learning_rate": 0.0001934661668437073,
-      "loss": 0.3916,
+      "loss": 0.4104,
       "step": 37
     },
     {
       "epoch": 0.30039525691699603,
-      "grad_norm": 1.0097579956054688,
+      "grad_norm": 0.8698107004165649,
       "learning_rate": 0.0001929967517119289,
-      "loss": 0.6569,
+      "loss": 0.6589,
       "step": 38
     },
     {
       "epoch": 0.308300395256917,
-      "grad_norm": 1.6274114847183228,
+      "grad_norm": 0.7418029308319092,
       "learning_rate": 0.0001925116643538684,
-      "loss": 0.2624,
+      "loss": 0.2634,
       "step": 39
     },
     {
       "epoch": 0.31620553359683795,
-      "grad_norm": 1.0551433563232422,
+      "grad_norm": 1.038150668144226,
       "learning_rate": 0.0001920109865186052,
-      "loss": 0.6816,
+      "loss": 0.6815,
       "step": 40
     },
     {
       "epoch": 0.3241106719367589,
-      "grad_norm": 0.9606983661651611,
+      "grad_norm": 0.7771694660186768,
       "learning_rate": 0.00019149480258259533,
-      "loss": 0.4485,
+      "loss": 0.4593,
       "step": 41
     },
     {
       "epoch": 0.33201581027667987,
-      "grad_norm": 1.1261677742004395,
+      "grad_norm": 1.0811573266983032,
       "learning_rate": 0.00019096319953545185,
-      "loss": 0.3185,
+      "loss": 0.3129,
       "step": 42
     },
     {
       "epoch": 0.33992094861660077,
-      "grad_norm": 1.820938229560852,
+      "grad_norm": 1.2685577869415283,
       "learning_rate": 0.00019041626696528503,
-      "loss": 0.277,
+      "loss": 0.2788,
       "step": 43
     },
     {
       "epoch": 0.34782608695652173,
-      "grad_norm": 1.1593023538589478,
+      "grad_norm": 1.1799741983413696,
       "learning_rate": 0.00018985409704360456,
-      "loss": 0.6395,
+      "loss": 0.6224,
       "step": 44
     },
     {
       "epoch": 0.3557312252964427,
-      "grad_norm": 0.8903545141220093,
+      "grad_norm": 1.3152222633361816,
       "learning_rate": 0.0001892767845097864,
-      "loss": 0.4733,
+      "loss": 0.4609,
       "step": 45
     },
     {
       "epoch": 0.36363636363636365,
-      "grad_norm": 0.7684641480445862,
+      "grad_norm": 0.8674153089523315,
       "learning_rate": 0.00018868442665510678,
-      "loss": 0.3804,
+      "loss": 0.3661,
       "step": 46
     },
     {
       "epoch": 0.3715415019762846,
-      "grad_norm": 1.0813274383544922,
+      "grad_norm": 1.9650135040283203,
       "learning_rate": 0.00018807712330634642,
-      "loss": 0.428,
+      "loss": 0.4214,
       "step": 47
     },
     {
       "epoch": 0.3794466403162055,
-      "grad_norm": 0.8424346446990967,
+      "grad_norm": 0.8908179402351379,
       "learning_rate": 0.00018745497680896722,
-      "loss": 0.381,
+      "loss": 0.3789,
       "step": 48
     },
     {
       "epoch": 0.38735177865612647,
-      "grad_norm": 0.7333454489707947,
+      "grad_norm": 0.8767942786216736,
       "learning_rate": 0.0001868180920098644,
-      "loss": 0.4397,
+      "loss": 0.432,
       "step": 49
     },
     {
       "epoch": 0.3952569169960474,
-      "grad_norm": 1.145660161972046,
+      "grad_norm": 1.3174779415130615,
       "learning_rate": 0.0001861665762396974,
-      "loss": 0.813,
+      "loss": 0.8093,
       "step": 50
     },
     {
       "epoch": 0.4031620553359684,
-      "grad_norm": 1.0778045654296875,
+      "grad_norm": 1.12623929977417,
       "learning_rate": 0.00018550053929480202,
-      "loss": 0.3169,
+      "loss": 0.3248,
       "step": 51
     },
     {
       "epoch": 0.41106719367588934,
-      "grad_norm": 1.2810466289520264,
+      "grad_norm": 2.023007392883301,
       "learning_rate": 0.00018482009341868697,
-      "loss": 0.2757,
+      "loss": 0.2736,
       "step": 52
     },
     {
       "epoch": 0.4189723320158103,
-      "grad_norm": 1.754908800125122,
+      "grad_norm": 0.981887936592102,
       "learning_rate": 0.00018412535328311814,
-      "loss": 0.3722,
+      "loss": 0.3832,
       "step": 53
     },
     {
       "epoch": 0.4268774703557312,
-      "grad_norm": 1.2411037683486938,
+      "grad_norm": 2.4076719284057617,
       "learning_rate": 0.00018341643596879367,
-      "loss": 0.8517,
+      "loss": 0.857,
       "step": 54
     },
     {
       "epoch": 0.43478260869565216,
-      "grad_norm": 0.9490335583686829,
+      "grad_norm": 1.1959154605865479,
       "learning_rate": 0.0001826934609456129,
-      "loss": 0.4454,
+      "loss": 0.4592,
       "step": 55
     },
     {
       "epoch": 0.4426877470355731,
-      "grad_norm": 0.5774851441383362,
+      "grad_norm": 0.5865926742553711,
       "learning_rate": 0.00018195655005254273,
-      "loss": 0.2006,
+      "loss": 0.1814,
       "step": 56
     },
     {
       "epoch": 0.4505928853754941,
-      "grad_norm": 0.771795928478241,
+      "grad_norm": 0.8474725484848022,
       "learning_rate": 0.00018120582747708502,
-      "loss": 0.2744,
+      "loss": 0.2844,
       "step": 57
     },
     {
       "epoch": 0.45849802371541504,
-      "grad_norm": 0.7947430610656738,
+      "grad_norm": 1.0415821075439453,
       "learning_rate": 0.00018044141973434758,
-      "loss": 0.3595,
+      "loss": 0.3637,
       "step": 58
     },
     {
       "epoch": 0.466403162055336,
-      "grad_norm": 0.8362589478492737,
+      "grad_norm": 1.8770055770874023,
       "learning_rate": 0.0001796634556457236,
-      "loss": 0.2472,
+      "loss": 0.2462,
       "step": 59
     },
     {
       "epoch": 0.4743083003952569,
-      "grad_norm": 1.265906572341919,
+      "grad_norm": 0.9714164733886719,
       "learning_rate": 0.00017887206631718203,
-      "loss": 0.3196,
+      "loss": 0.3059,
       "step": 60
     },
     {
       "epoch": 0.48221343873517786,
-      "grad_norm": 0.9433225989341736,
+      "grad_norm": 0.9178167581558228,
       "learning_rate": 0.0001780673851171728,
-      "loss": 0.4213,
+      "loss": 0.4233,
       "step": 61
     },
     {
       "epoch": 0.4901185770750988,
-      "grad_norm": 1.061390995979309,
+      "grad_norm": 0.7472209930419922,
       "learning_rate": 0.00017724954765415137,
-      "loss": 0.2079,
+      "loss": 0.1977,
       "step": 62
     },
     {
       "epoch": 0.4980237154150198,
-      "grad_norm": 0.7185953259468079,
+      "grad_norm": 0.7857463955879211,
       "learning_rate": 0.00017641869175372493,
-      "loss": 0.2292,
+      "loss": 0.2279,
       "step": 63
     },
     {
       "epoch": 0.5059288537549407,
-      "grad_norm": 2.011293649673462,
+      "grad_norm": 2.13358211517334,
       "learning_rate": 0.00017557495743542585,
-      "loss": 0.4257,
+      "loss": 0.4481,
       "step": 64
     },
     {
       "epoch": 0.5138339920948617,
-      "grad_norm": 1.008697271347046,
+      "grad_norm": 0.9242135882377625,
       "learning_rate": 0.00017471848688911464,
-      "loss": 0.3885,
+      "loss": 0.3843,
       "step": 65
     },
     {
       "epoch": 0.5217391304347826,
-      "grad_norm": 0.6567645072937012,
+      "grad_norm": 0.7868145704269409,
       "learning_rate": 0.00017384942445101772,
-      "loss": 0.1438,
+      "loss": 0.1393,
       "step": 66
     },
     {
       "epoch": 0.5296442687747036,
-      "grad_norm": 0.7839428186416626,
+      "grad_norm": 0.6896469593048096,
       "learning_rate": 0.000172967916579403,
-      "loss": 0.3432,
+      "loss": 0.3294,
       "step": 67
     },
     {
       "epoch": 0.5375494071146245,
-      "grad_norm": 0.6802240610122681,
+      "grad_norm": 0.691632866859436,
       "learning_rate": 0.00017207411182989832,
-      "loss": 0.2519,
+      "loss": 0.2405,
       "step": 68
     },
     {
       "epoch": 0.5454545454545454,
-      "grad_norm": 0.7788658142089844,
+      "grad_norm": 0.6870527267456055,
       "learning_rate": 0.00017116816083045602,
-      "loss": 0.2402,
+      "loss": 0.2379,
       "step": 69
     },
     {
       "epoch": 0.5533596837944664,
-      "grad_norm": 1.4981237649917603,
+      "grad_norm": 0.6751531958580017,
       "learning_rate": 0.00017025021625596853,
-      "loss": 0.2386,
+      "loss": 0.2365,
       "step": 70
     },
     {
       "epoch": 0.5612648221343873,
-      "grad_norm": 1.2055702209472656,
+      "grad_norm": 1.061668872833252,
       "learning_rate": 0.0001693204328025389,
-      "loss": 0.5334,
+      "loss": 0.5389,
       "step": 71
     },
     {
       "epoch": 0.5691699604743083,
-      "grad_norm": 0.8050393462181091,
+      "grad_norm": 0.7164187431335449,
       "learning_rate": 0.0001683789671614107,
-      "loss": 0.3503,
+      "loss": 0.3441,
       "step": 72
     },
     {
       "epoch": 0.5770750988142292,
-      "grad_norm": 1.388447880744934,
+      "grad_norm": 1.1519713401794434,
       "learning_rate": 0.00016742597799256182,
-      "loss": 0.2713,
+      "loss": 0.2715,
       "step": 73
     },
     {
       "epoch": 0.5849802371541502,
-      "grad_norm": 1.34609854221344,
+      "grad_norm": 1.568219542503357,
       "learning_rate": 0.00016646162589796615,
-      "loss": 0.3561,
+      "loss": 0.3667,
       "step": 74
     },
     {
       "epoch": 0.5928853754940712,
-      "grad_norm": 0.6840114593505859,
+      "grad_norm": 0.7082597017288208,
       "learning_rate": 0.00016548607339452853,
-      "loss": 0.3319,
+      "loss": 0.3271,
       "step": 75
     },
     {
       "epoch": 0.6007905138339921,
-      "grad_norm": 0.8626895546913147,
+      "grad_norm": 0.7822222709655762,
       "learning_rate": 0.00016449948488669639,
-      "loss": 0.3875,
+      "loss": 0.395,
       "step": 76
     },
     {
       "epoch": 0.6086956521739131,
-      "grad_norm": 0.6978746652603149,
+      "grad_norm": 0.70323646068573,
       "learning_rate": 0.00016350202663875386,
-      "loss": 0.3392,
+      "loss": 0.3593,
       "step": 77
     },
     {
       "epoch": 0.616600790513834,
-      "grad_norm": 1.2156915664672852,
+      "grad_norm": 0.8593727350234985,
       "learning_rate": 0.00016249386674680184,
-      "loss": 0.4685,
+      "loss": 0.4751,
       "step": 78
     },
     {
       "epoch": 0.6245059288537549,
-      "grad_norm": 1.4607609510421753,
+      "grad_norm": 0.8436718583106995,
       "learning_rate": 0.0001614751751104301,
-      "loss": 0.2612,
+      "loss": 0.2788,
       "step": 79
     },
     {
       "epoch": 0.6324110671936759,
-      "grad_norm": 1.203145146369934,
+      "grad_norm": 1.0489970445632935,
       "learning_rate": 0.00016044612340408466,
-      "loss": 0.2867,
+      "loss": 0.2804,
       "step": 80
     },
     {
       "epoch": 0.6403162055335968,
-      "grad_norm": 0.5245752930641174,
+      "grad_norm": 0.4728718101978302,
       "learning_rate": 0.00015940688504813662,
-      "loss": 0.1539,
+      "loss": 0.1463,
       "step": 81
     },
     {
       "epoch": 0.6482213438735178,
-      "grad_norm": 0.959148108959198,
+      "grad_norm": 0.8874382972717285,
       "learning_rate": 0.00015835763517965673,
-      "loss": 0.5396,
+      "loss": 0.547,
       "step": 82
     },
     {
       "epoch": 0.6561264822134387,
-      "grad_norm": 1.129942536354065,
+      "grad_norm": 0.9559019804000854,
       "learning_rate": 0.00015729855062290022,
-      "loss": 0.2542,
+      "loss": 0.2586,
       "step": 83
     },
     {
       "epoch": 0.6640316205533597,
-      "grad_norm": 1.5355336666107178,
+      "grad_norm": 1.4548382759094238,
       "learning_rate": 0.0001562298098595078,
-      "loss": 0.41,
+      "loss": 0.4085,
       "step": 84
     },
     {
       "epoch": 0.6719367588932806,
-      "grad_norm": 1.1144013404846191,
+      "grad_norm": 1.6104789972305298,
       "learning_rate": 0.00015515159299842707,
-      "loss": 0.3745,
+      "loss": 0.3712,
       "step": 85
     },
     {
       "epoch": 0.6798418972332015,
-      "grad_norm": 0.7561334371566772,
+      "grad_norm": 0.7389092445373535,
       "learning_rate": 0.00015406408174555976,
-      "loss": 0.2736,
+      "loss": 0.2789,
       "step": 86
     },
     {
       "epoch": 0.6877470355731226,
-      "grad_norm": 0.8727364540100098,
+      "grad_norm": 0.6817464232444763,
       "learning_rate": 0.00015296745937313987,
-      "loss": 0.2456,
+      "loss": 0.2501,
       "step": 87
     },
     {
       "epoch": 0.6956521739130435,
-      "grad_norm": 0.9084362983703613,
+      "grad_norm": 0.9103575348854065,
       "learning_rate": 0.00015186191068884775,
-      "loss": 0.5981,
+      "loss": 0.5931,
       "step": 88
     },
     {
       "epoch": 0.7035573122529645,
-      "grad_norm": 0.9988125562667847,
+      "grad_norm": 0.9242210388183594,
       "learning_rate": 0.00015074762200466556,
-      "loss": 0.6487,
+      "loss": 0.6431,
       "step": 89
     },
     {
       "epoch": 0.7114624505928854,
-      "grad_norm": 0.9429134130477905,
+      "grad_norm": 0.9366337060928345,
       "learning_rate": 0.00014962478110547918,
-      "loss": 0.4248,
+      "loss": 0.4411,
       "step": 90
     },
     {
       "epoch": 0.7193675889328063,
-      "grad_norm": 0.7329612970352173,
+      "grad_norm": 0.531251072883606,
       "learning_rate": 0.00014849357721743168,
-      "loss": 0.1729,
+      "loss": 0.1745,
       "step": 91
     },
     {
       "epoch": 0.7272727272727273,
-      "grad_norm": 1.2303996086120605,
+      "grad_norm": 1.2808730602264404,
       "learning_rate": 0.0001473542009760343,
-      "loss": 0.4279,
+      "loss": 0.4046,
       "step": 92
     },
     {
       "epoch": 0.7351778656126482,
-      "grad_norm": 1.3390841484069824,
+      "grad_norm": 1.19551682472229,
       "learning_rate": 0.00014620684439403962,
-      "loss": 0.4451,
+      "loss": 0.4524,
       "step": 93
     },
     {
       "epoch": 0.7430830039525692,
-      "grad_norm": 0.8936579823493958,
+      "grad_norm": 0.8160001039505005,
       "learning_rate": 0.0001450517008290827,
-      "loss": 0.2279,
+      "loss": 0.2237,
       "step": 94
     },
     {
       "epoch": 0.7509881422924901,
-      "grad_norm": 0.6139077544212341,
+      "grad_norm": 0.5796943306922913,
       "learning_rate": 0.0001438889649510956,
-      "loss": 0.2379,
+      "loss": 0.2322,
       "step": 95
     },
     {
       "epoch": 0.758893280632411,
-      "grad_norm": 0.6658306121826172,
+      "grad_norm": 0.6521222591400146,
       "learning_rate": 0.00014271883270950073,
-      "loss": 0.2763,
+      "loss": 0.2682,
       "step": 96
     },
     {
       "epoch": 0.766798418972332,
-      "grad_norm": 1.5102399587631226,
+      "grad_norm": 1.001668095588684,
       "learning_rate": 0.00014154150130018866,
-      "loss": 0.3528,
+      "loss": 0.3352,
       "step": 97
     },
     {
       "epoch": 0.7747035573122529,
-      "grad_norm": 0.6491991877555847,
+      "grad_norm": 0.8112585544586182,
       "learning_rate": 0.00014035716913228568,
-      "loss": 0.1745,
+      "loss": 0.173,
       "step": 98
     },
     {
       "epoch": 0.782608695652174,
-      "grad_norm": 1.1244672536849976,
+      "grad_norm": 1.2103630304336548,
       "learning_rate": 0.00013916603579471705,
-      "loss": 1.0632,
+      "loss": 1.0565,
       "step": 99
     },
     {
       "epoch": 0.7905138339920948,
-      "grad_norm": 1.0054899454116821,
+      "grad_norm": 1.0214811563491821,
       "learning_rate": 0.0001379683020225714,
-      "loss": 0.3859,
+      "loss": 0.3749,
       "step": 100
     },
     {
       "epoch": 0.7984189723320159,
-      "grad_norm": 1.1216827630996704,
+      "grad_norm": 0.6681635975837708,
       "learning_rate": 0.000136764169663272,
-      "loss": 0.1602,
+      "loss": 0.1641,
       "step": 101
     },
     {
       "epoch": 0.8063241106719368,
-      "grad_norm": 1.0952600240707397,
+      "grad_norm": 0.9718304872512817,
       "learning_rate": 0.00013555384164256048,
-      "loss": 0.2047,
+      "loss": 0.1989,
       "step": 102
     },
     {
       "epoch": 0.8142292490118577,
-      "grad_norm": 0.6201446652412415,
+      "grad_norm": 0.5994592905044556,
       "learning_rate": 0.00013433752193029886,
-      "loss": 0.2862,
+      "loss": 0.2865,
       "step": 103
     },
     {
       "epoch": 0.8221343873517787,
-      "grad_norm": 0.7657504677772522,
+      "grad_norm": 0.7605760097503662,
       "learning_rate": 0.00013311541550609565,
-      "loss": 0.468,
+      "loss": 0.4609,
       "step": 104
     },
     {
       "epoch": 0.8300395256916996,
-      "grad_norm": 0.5368217825889587,
+      "grad_norm": 0.4863194227218628,
       "learning_rate": 0.00013188772832476188,
       "loss": 0.1356,
       "step": 105
     },
     {
       "epoch": 0.8379446640316206,
-      "grad_norm": 0.6289976835250854,
+      "grad_norm": 0.4657331705093384,
       "learning_rate": 0.00013065466728160252,
-      "loss": 0.1638,
+      "loss": 0.1501,
       "step": 106
     },
     {
       "epoch": 0.8458498023715415,
-      "grad_norm": 1.216394066810608,
+      "grad_norm": 0.6451147794723511,
       "learning_rate": 0.00012941644017754964,
-      "loss": 0.3181,
+      "loss": 0.3072,
       "step": 107
     },
     {
       "epoch": 0.8537549407114624,
-      "grad_norm": 0.6963629722595215,
+      "grad_norm": 0.7699165344238281,
       "learning_rate": 0.00012817325568414297,
-      "loss": 0.418,
+      "loss": 0.4145,
       "step": 108
     },
     {
       "epoch": 0.8616600790513834,
-      "grad_norm": 1.0949358940124512,
+      "grad_norm": 1.160872459411621,
       "learning_rate": 0.00012692532330836346,
-      "loss": 0.2907,
+      "loss": 0.2784,
       "step": 109
     },
     {
       "epoch": 0.8695652173913043,
-      "grad_norm": 0.7999882102012634,
+      "grad_norm": 0.9029427766799927,
       "learning_rate": 0.00012567285335732633,
-      "loss": 0.3369,
+      "loss": 0.3273,
       "step": 110
     },
     {
       "epoch": 0.8774703557312253,
-      "grad_norm": 0.7669695019721985,
+      "grad_norm": 0.7525882720947266,
       "learning_rate": 0.00012441605690283915,
-      "loss": 0.2744,
+      "loss": 0.2811,
       "step": 111
     },
     {
       "epoch": 0.8853754940711462,
-      "grad_norm": 0.8140245079994202,
+      "grad_norm": 0.7249767184257507,
       "learning_rate": 0.00012315514574583113,
-      "loss": 0.2297,
+      "loss": 0.2133,
       "step": 112
     },
     {
       "epoch": 0.8932806324110671,
-      "grad_norm": 1.2144274711608887,
+      "grad_norm": 2.000819683074951,
       "learning_rate": 0.0001218903323806595,
-      "loss": 0.4247,
+      "loss": 0.4112,
       "step": 113
     },
     {
       "epoch": 0.9011857707509882,
-      "grad_norm": 0.8045055866241455,
+      "grad_norm": 0.751440703868866,
       "learning_rate": 0.00012062182995929882,
-      "loss": 0.3203,
+      "loss": 0.3087,
       "step": 114
     },
     {
       "epoch": 0.9090909090909091,
-      "grad_norm": 4.573957443237305,
+      "grad_norm": 1.0243513584136963,
       "learning_rate": 0.00011934985225541998,
-      "loss": 0.4917,
+      "loss": 0.4851,
       "step": 115
     },
     {
       "epoch": 0.9169960474308301,
-      "grad_norm": 1.2436349391937256,
+      "grad_norm": 0.7862813472747803,
       "learning_rate": 0.0001180746136283638,
-      "loss": 0.3548,
+      "loss": 0.3612,
       "step": 116
     },
     {
       "epoch": 0.924901185770751,
-      "grad_norm": 1.3303828239440918,
+      "grad_norm": 1.0955418348312378,
       "learning_rate": 0.00011679632898701649,
-      "loss": 0.3161,
+      "loss": 0.3377,
       "step": 117
     },
     {
       "epoch": 0.932806324110672,
-      "grad_norm": 0.7288329005241394,
+      "grad_norm": 0.8309126496315002,
       "learning_rate": 0.00011551521375359206,
-      "loss": 0.2748,
+      "loss": 0.282,
       "step": 118
     },
     {
       "epoch": 0.9407114624505929,
-      "grad_norm": 0.94376140832901,
+      "grad_norm": 1.5201857089996338,
       "learning_rate": 0.00011423148382732853,
-      "loss": 0.2465,
+      "loss": 0.2581,
       "step": 119
     },
     {
       "epoch": 0.9486166007905138,
-      "grad_norm": 0.5832346081733704,
+      "grad_norm": 0.6001629829406738,
       "learning_rate": 0.00011294535554810354,
-      "loss": 0.1955,
+      "loss": 0.2002,
       "step": 120
     },
     {
       "epoch": 0.9565217391304348,
-      "grad_norm": 0.6223639845848083,
+      "grad_norm": 0.5865480899810791,
       "learning_rate": 0.00011165704565997593,
-      "loss": 0.1401,
+      "loss": 0.137,
       "step": 121
     },
     {
       "epoch": 0.9644268774703557,
-      "grad_norm": 0.8612600564956665,
+      "grad_norm": 0.8828125596046448,
       "learning_rate": 0.00011036677127465889,
-      "loss": 0.3112,
+      "loss": 0.3052,
       "step": 122
     },
     {
       "epoch": 0.9723320158102767,
-      "grad_norm": 1.0104589462280273,
+      "grad_norm": 0.6774571537971497,
       "learning_rate": 0.00010907474983493144,
-      "loss": 0.426,
+      "loss": 0.4175,
       "step": 123
     },
     {
       "epoch": 0.9802371541501976,
-      "grad_norm": 1.0083054304122925,
+      "grad_norm": 0.9369196891784668,
       "learning_rate": 0.00010778119907799398,
-      "loss": 0.4125,
+      "loss": 0.4036,
       "step": 124
     },
     {
       "epoch": 0.9881422924901185,
-      "grad_norm": 2.0528476238250732,
+      "grad_norm": 0.9793416261672974,
       "learning_rate": 0.0001064863369987743,
-      "loss": 0.2398,
+      "loss": 0.2515,
       "step": 125
     },
     {
       "epoch": 0.9960474308300395,
-      "grad_norm": 1.8364521265029907,
+      "grad_norm": 1.6688363552093506,
       "learning_rate": 0.00010519038181318999,
-      "loss": 0.2789,
+      "loss": 0.2686,
       "step": 126
     },
     {
       "epoch": 1.0,
-      "grad_norm": 1.1836934089660645,
+      "grad_norm": 1.589106559753418,
       "learning_rate": 0.00010389355192137377,
-      "loss": 0.3988,
+      "loss": 0.3921,
       "step": 127
     },
     {
       "epoch": 1.007905138339921,
-      "grad_norm": 0.7436236143112183,
+      "grad_norm": 0.6045113205909729,
       "learning_rate": 0.00010259606587086783,
-      "loss": 0.296,
+      "loss": 0.2638,
       "step": 128
     },
     {
       "epoch": 1.0158102766798418,
-      "grad_norm": 0.6589778661727905,
+      "grad_norm": 0.6287718415260315,
       "learning_rate": 0.0001012981423197931,
-      "loss": 0.4009,
+      "loss": 0.3818,
       "step": 129
     },
     {
       "epoch": 1.023715415019763,
-      "grad_norm": 0.5743900537490845,
+      "grad_norm": 0.6705343723297119,
       "learning_rate": 0.0001,
-      "loss": 0.1445,
+      "loss": 0.1436,
       "step": 130
     },
     {
       "epoch": 1.0316205533596838,
-      "grad_norm": 1.7947680950164795,
+      "grad_norm": 0.7970353960990906,
       "learning_rate": 9.870185768020693e-05,
-      "loss": 0.2082,
+      "loss": 0.1986,
       "step": 131
     },
     {
       "epoch": 1.0395256916996047,
-      "grad_norm": 0.9778440594673157,
+      "grad_norm": 1.4546546936035156,
       "learning_rate": 9.740393412913219e-05,
-      "loss": 0.3782,
+      "loss": 0.3551,
       "step": 132
     },
     {
       "epoch": 1.0474308300395256,
-      "grad_norm": 0.6362202167510986,
+      "grad_norm": 0.5461708307266235,
       "learning_rate": 9.610644807862625e-05,
-      "loss": 0.2059,
+      "loss": 0.2015,
       "step": 133
     },
     {
       "epoch": 1.0553359683794465,
-      "grad_norm": 0.7074435353279114,
+      "grad_norm": 1.8389110565185547,
       "learning_rate": 9.480961818681004e-05,
-      "loss": 0.2698,
+      "loss": 0.2769,
       "step": 134
     },
     {
       "epoch": 1.0632411067193677,
-      "grad_norm": 0.8866583704948425,
+      "grad_norm": 0.6100041270256042,
       "learning_rate": 9.35136630012257e-05,
-      "loss": 0.1588,
+      "loss": 0.1516,
       "step": 135
     },
     {
       "epoch": 1.0711462450592886,
-      "grad_norm": 0.5219342708587646,
+      "grad_norm": 0.5065872669219971,
       "learning_rate": 9.221880092200601e-05,
-      "loss": 0.1926,
+      "loss": 0.1945,
       "step": 136
     },
     {
       "epoch": 1.0790513833992095,
-      "grad_norm": 0.7011678218841553,
+      "grad_norm": 1.9398545026779175,
       "learning_rate": 9.092525016506858e-05,
-      "loss": 0.2249,
+      "loss": 0.2246,
       "step": 137
     },
     {
       "epoch": 1.0869565217391304,
-      "grad_norm": 0.5657678246498108,
+      "grad_norm": 0.6197378635406494,
       "learning_rate": 8.963322872534114e-05,
-      "loss": 0.1683,
+      "loss": 0.1768,
       "step": 138
     },
     {
       "epoch": 1.0948616600790513,
-      "grad_norm": 8.20474910736084,
+      "grad_norm": 1.256683349609375,
       "learning_rate": 8.83429543400241e-05,
-      "loss": 0.3678,
+      "loss": 0.3742,
       "step": 139
     },
     {
       "epoch": 1.1027667984189724,
-      "grad_norm": 0.7418918013572693,
+      "grad_norm": 0.8401638865470886,
       "learning_rate": 8.705464445189647e-05,
-      "loss": 0.1955,
+      "loss": 0.1974,
       "step": 140
     },
     {
       "epoch": 1.1106719367588933,
-      "grad_norm": 0.733155369758606,
+      "grad_norm": 0.7496947646141052,
       "learning_rate": 8.57685161726715e-05,
-      "loss": 0.3882,
+      "loss": 0.3825,
       "step": 141
     },
     {
       "epoch": 1.1185770750988142,
-      "grad_norm": 0.6813409924507141,
+      "grad_norm": 1.0039342641830444,
       "learning_rate": 8.448478624640797e-05,
-      "loss": 0.1533,
+      "loss": 0.1535,
       "step": 142
     },
     {
       "epoch": 1.1264822134387351,
-      "grad_norm": 0.6259405016899109,
+      "grad_norm": 0.5387095808982849,
       "learning_rate": 8.320367101298351e-05,
-      "loss": 0.2233,
+      "loss": 0.2106,
       "step": 143
     },
     {
       "epoch": 1.1343873517786562,
-      "grad_norm": 0.6203911304473877,
+      "grad_norm": 0.5525783896446228,
       "learning_rate": 8.192538637163621e-05,
-      "loss": 0.167,
+      "loss": 0.169,
       "step": 144
     },
     {
       "epoch": 1.1422924901185771,
-      "grad_norm": 0.6209132075309753,
+      "grad_norm": 0.6670098304748535,
       "learning_rate": 8.065014774458003e-05,
-      "loss": 0.2,
+      "loss": 0.1897,
       "step": 145
     },
     {
       "epoch": 1.150197628458498,
-      "grad_norm": 0.5935210585594177,
+      "grad_norm": 0.701342761516571,
       "learning_rate": 7.93781700407012e-05,
-      "loss": 0.176,
+      "loss": 0.1784,
       "step": 146
     },
     {
       "epoch": 1.158102766798419,
-      "grad_norm": 0.5734385251998901,
+      "grad_norm": 0.52852463722229,
       "learning_rate": 7.810966761934053e-05,
-      "loss": 0.1797,
+      "loss": 0.1712,
       "step": 147
     },
     {
       "epoch": 1.1660079051383399,
-      "grad_norm": 0.7462530136108398,
+      "grad_norm": 0.5911440849304199,
       "learning_rate": 7.684485425416888e-05,
-      "loss": 0.1302,
+      "loss": 0.1378,
       "step": 148
     },
     {
       "epoch": 1.1739130434782608,
-      "grad_norm": 0.5157163739204407,
+      "grad_norm": 0.7355263829231262,
       "learning_rate": 7.558394309716088e-05,
-      "loss": 0.109,
+      "loss": 0.1088,
       "step": 149
     },
     {
       "epoch": 1.1818181818181819,
-      "grad_norm": 0.5993911027908325,
+      "grad_norm": 0.5206398367881775,
       "learning_rate": 7.432714664267373e-05,
-      "loss": 0.1918,
+      "loss": 0.1724,
       "step": 150
     },
     {
       "epoch": 1.1897233201581028,
-      "grad_norm": 0.6078954339027405,
+      "grad_norm": 0.5356054902076721,
       "learning_rate": 7.307467669163655e-05,
-      "loss": 0.1903,
+      "loss": 0.1792,
       "step": 151
     },
     {
       "epoch": 1.1976284584980237,
-      "grad_norm": 0.7795295119285583,
+      "grad_norm": 0.5152002573013306,
       "learning_rate": 7.182674431585704e-05,
-      "loss": 0.2494,
+      "loss": 0.2179,
       "step": 152
     },
     {
       "epoch": 1.2055335968379446,
-      "grad_norm": 0.6875370144844055,
+      "grad_norm": 0.6617997288703918,
       "learning_rate": 7.058355982245037e-05,
-      "loss": 0.165,
+      "loss": 0.1654,
       "step": 153
     },
     {
       "epoch": 1.2134387351778657,
-      "grad_norm": 1.2836542129516602,
+      "grad_norm": 1.776577115058899,
       "learning_rate": 6.934533271839752e-05,
-      "loss": 0.2879,
+      "loss": 0.3008,
       "step": 154
     },
     {
       "epoch": 1.2213438735177866,
-      "grad_norm": 0.6149235963821411,
+      "grad_norm": 0.5596933364868164,
       "learning_rate": 6.811227167523815e-05,
-      "loss": 0.3759,
+      "loss": 0.3617,
       "step": 155
     },
     {
       "epoch": 1.2292490118577075,
-      "grad_norm": 0.6558827757835388,
+      "grad_norm": 0.5993364453315735,
       "learning_rate": 6.688458449390437e-05,
-      "loss": 0.2317,
+      "loss": 0.244,
       "step": 156
     },
     {
       "epoch": 1.2371541501976284,
-      "grad_norm": 0.5832497477531433,
+      "grad_norm": 0.9282941818237305,
       "learning_rate": 6.566247806970119e-05,
-      "loss": 0.243,
+      "loss": 0.2579,
       "step": 157
     },
     {
       "epoch": 1.2450592885375493,
-      "grad_norm": 0.7867661714553833,
+      "grad_norm": 0.677804172039032,
       "learning_rate": 6.444615835743955e-05,
-      "loss": 0.291,
+      "loss": 0.2758,
       "step": 158
     },
     {
       "epoch": 1.2529644268774702,
-      "grad_norm": 1.1524978876113892,
+      "grad_norm": 3.5637006759643555,
       "learning_rate": 6.323583033672799e-05,
-      "loss": 0.2394,
+      "loss": 0.2121,
       "step": 159
     },
     {
       "epoch": 1.2608695652173914,
-      "grad_norm": 1.0739554166793823,
+      "grad_norm": 0.7661057710647583,
       "learning_rate": 6.203169797742861e-05,
-      "loss": 0.3626,
+      "loss": 0.3532,
       "step": 160
     },
     {
       "epoch": 1.2687747035573123,
-      "grad_norm": 1.4459284543991089,
+      "grad_norm": 1.0855592489242554,
       "learning_rate": 6.083396420528298e-05,
-      "loss": 0.2245,
+      "loss": 0.1676,
       "step": 161
     },
     {
       "epoch": 1.2766798418972332,
-      "grad_norm": 1.0734719038009644,
+      "grad_norm": 1.2014670372009277,
       "learning_rate": 5.964283086771435e-05,
-      "loss": 0.4054,
+      "loss": 0.3823,
       "step": 162
     },
     {
       "epoch": 1.2845849802371543,
-      "grad_norm": 0.7080317139625549,
+      "grad_norm": 1.5054898262023926,
       "learning_rate": 5.845849869981137e-05,
-      "loss": 0.1684,
+      "loss": 0.1683,
       "step": 163
     },
     {
       "epoch": 1.2924901185770752,
-      "grad_norm": 0.9621386528015137,
+      "grad_norm": 1.071014404296875,
       "learning_rate": 5.728116729049928e-05,
-      "loss": 0.5877,
+      "loss": 0.6714,
       "step": 164
     },
     {
       "epoch": 1.300395256916996,
-      "grad_norm": 0.511576771736145,
+      "grad_norm": 0.34447920322418213,
       "learning_rate": 5.611103504890444e-05,
-      "loss": 0.0844,
+      "loss": 0.0817,
       "step": 165
     },
     {
       "epoch": 1.308300395256917,
-      "grad_norm": 0.8402789235115051,
+      "grad_norm": 0.7190909385681152,
       "learning_rate": 5.4948299170917325e-05,
-      "loss": 0.2352,
+      "loss": 0.2317,
       "step": 166
     },
     {
       "epoch": 1.316205533596838,
-      "grad_norm": 0.7360251545906067,
+      "grad_norm": 1.5183690786361694,
       "learning_rate": 5.379315560596038e-05,
-      "loss": 0.2368,
+      "loss": 0.2293,
       "step": 167
     },
     {
       "epoch": 1.3241106719367588,
-      "grad_norm": 0.8452273011207581,
+      "grad_norm": 1.6111899614334106,
       "learning_rate": 5.26457990239657e-05,
-      "loss": 0.2075,
+      "loss": 0.1938,
       "step": 168
     },
     {
       "epoch": 1.33201581027668,
-      "grad_norm": 1.056742787361145,
+      "grad_norm": 1.1006243228912354,
       "learning_rate": 5.1506422782568345e-05,
-      "loss": 0.2835,
+      "loss": 0.2764,
       "step": 169
     },
     {
       "epoch": 1.3399209486166008,
-      "grad_norm": 0.4637269079685211,
+      "grad_norm": 0.48925891518592834,
       "learning_rate": 5.0375218894520834e-05,
-      "loss": 0.0961,
+      "loss": 0.0881,
       "step": 170
     },
     {
       "epoch": 1.3478260869565217,
-      "grad_norm": 1.0665563344955444,
+      "grad_norm": 1.2127503156661987,
       "learning_rate": 4.9252377995334444e-05,
-      "loss": 0.1613,
+      "loss": 0.1581,
       "step": 171
     },
     {
       "epoch": 1.3557312252964426,
-      "grad_norm": 0.6855564713478088,
+      "grad_norm": 0.70034259557724,
       "learning_rate": 4.813808931115228e-05,
-      "loss": 0.1832,
+      "loss": 0.1765,
       "step": 172
     },
     {
       "epoch": 1.3636363636363638,
-      "grad_norm": 1.3846145868301392,
+      "grad_norm": 0.9129422307014465,
       "learning_rate": 4.703254062686017e-05,
-      "loss": 0.4217,
+      "loss": 0.4467,
       "step": 173
     },
     {
       "epoch": 1.3715415019762847,
-      "grad_norm": 0.9569350481033325,
+      "grad_norm": 1.0346431732177734,
       "learning_rate": 4.593591825444028e-05,
-      "loss": 0.3381,
+      "loss": 0.324,
       "step": 174
     },
     {
       "epoch": 1.3794466403162056,
-      "grad_norm": 2.7899603843688965,
+      "grad_norm": 1.1604621410369873,
       "learning_rate": 4.484840700157295e-05,
-      "loss": 0.2466,
+      "loss": 0.2579,
       "step": 175
     },
     {
       "epoch": 1.3873517786561265,
-      "grad_norm": 0.5576080679893494,
+      "grad_norm": 0.5738621354103088,
       "learning_rate": 4.377019014049223e-05,
-      "loss": 0.1441,
+      "loss": 0.1351,
       "step": 176
     },
     {
       "epoch": 1.3952569169960474,
-      "grad_norm": 0.9455075263977051,
+      "grad_norm": 0.8380118608474731,
       "learning_rate": 4.270144937709981e-05,
-      "loss": 0.1785,
+      "loss": 0.1808,
       "step": 177
     },
     {
       "epoch": 1.4031620553359683,
-      "grad_norm": 0.9868407845497131,
+      "grad_norm": 0.9386733770370483,
       "learning_rate": 4.164236482034327e-05,
-      "loss": 0.3179,
+      "loss": 0.2689,
       "step": 178
     },
     {
       "epoch": 1.4110671936758894,
-      "grad_norm": 0.39552775025367737,
+      "grad_norm": 0.451817125082016,
       "learning_rate": 4.059311495186338e-05,
-      "loss": 0.1057,
+      "loss": 0.1063,
       "step": 179
     },
     {
       "epoch": 1.4189723320158103,
-      "grad_norm": 1.4461482763290405,
+      "grad_norm": 0.9198683500289917,
       "learning_rate": 3.9553876595915375e-05,
-      "loss": 0.086,
+      "loss": 0.0927,
       "step": 180
     },
     {
       "epoch": 1.4268774703557312,
-      "grad_norm": 0.5234783887863159,
+      "grad_norm": 0.5517873167991638,
       "learning_rate": 3.852482488956992e-05,
-      "loss": 0.1233,
+      "loss": 0.1326,
       "step": 181
     },
     {
       "epoch": 1.434782608695652,
-      "grad_norm": 1.636184811592102,
+      "grad_norm": 2.099898338317871,
       "learning_rate": 3.750613325319817e-05,
-      "loss": 0.2031,
+      "loss": 0.2189,
       "step": 182
     },
     {
       "epoch": 1.4426877470355732,
-      "grad_norm": 0.6264592409133911,
+      "grad_norm": 0.5490242838859558,
       "learning_rate": 3.649797336124615e-05,
-      "loss": 0.1644,
+      "loss": 0.1408,
       "step": 183
     },
     {
       "epoch": 1.4505928853754941,
-      "grad_norm": 0.6155117750167847,
+      "grad_norm": 0.6548390984535217,
       "learning_rate": 3.550051511330361e-05,
-      "loss": 0.2239,
+      "loss": 0.2264,
       "step": 184
     },
     {
       "epoch": 1.458498023715415,
-      "grad_norm": 1.329161524772644,
+      "grad_norm": 1.2147459983825684,
       "learning_rate": 3.45139266054715e-05,
-      "loss": 0.2775,
+      "loss": 0.2715,
       "step": 185
     },
     {
       "epoch": 1.466403162055336,
-      "grad_norm": 0.8846868872642517,
+      "grad_norm": 0.9193414449691772,
       "learning_rate": 3.3538374102033866e-05,
-      "loss": 0.1838,
+      "loss": 0.1903,
       "step": 186
     },
     {
       "epoch": 1.4743083003952568,
-      "grad_norm": 0.929398238658905,
+      "grad_norm": 0.8377665281295776,
       "learning_rate": 3.257402200743821e-05,
-      "loss": 0.171,
+      "loss": 0.1726,
       "step": 187
     },
     {
       "epoch": 1.4822134387351777,
-      "grad_norm": 0.7083112597465515,
+      "grad_norm": 1.05824613571167,
       "learning_rate": 3.1621032838589305e-05,
-      "loss": 0.1866,
+      "loss": 0.1822,
       "step": 188
     },
     {
       "epoch": 1.4901185770750989,
-      "grad_norm": 1.1155579090118408,
+      "grad_norm": 0.8356701731681824,
       "learning_rate": 3.0679567197461134e-05,
-      "loss": 0.299,
+      "loss": 0.289,
       "step": 189
     },
     {
       "epoch": 1.4980237154150198,
-      "grad_norm": 0.7512155175209045,
+      "grad_norm": 1.1289795637130737,
       "learning_rate": 2.974978374403147e-05,
-      "loss": 0.2418,
+      "loss": 0.25,
       "step": 190
     },
     {
       "epoch": 1.5059288537549407,
-      "grad_norm": 0.7041618824005127,
+      "grad_norm": 0.7377334237098694,
       "learning_rate": 2.8831839169543996e-05,
-      "loss": 0.216,
+      "loss": 0.2193,
       "step": 191
     },
     {
       "epoch": 1.5138339920948618,
-      "grad_norm": 0.5851369500160217,
+      "grad_norm": 0.6371282935142517,
       "learning_rate": 2.7925888170101665e-05,
-      "loss": 0.0778,
+      "loss": 0.0678,
       "step": 192
     },
     {
       "epoch": 1.5217391304347827,
-      "grad_norm": 0.7017596364021301,
+      "grad_norm": 0.5883315801620483,
       "learning_rate": 2.7032083420597e-05,
-      "loss": 0.1147,
+      "loss": 0.0967,
       "step": 193
     },
     {
       "epoch": 1.5296442687747036,
-      "grad_norm": 0.8783608078956604,
+      "grad_norm": 0.9382833242416382,
       "learning_rate": 2.6150575548982292e-05,
-      "loss": 0.3006,
+      "loss": 0.297,
       "step": 194
     },
     {
       "epoch": 1.5375494071146245,
-      "grad_norm": 1.0361592769622803,
+      "grad_norm": 0.8458752036094666,
       "learning_rate": 2.528151311088537e-05,
-      "loss": 0.1789,
+      "loss": 0.1853,
       "step": 195
     },
     {
       "epoch": 1.5454545454545454,
-      "grad_norm": 0.7505379319190979,
+      "grad_norm": 0.5695004463195801,
       "learning_rate": 2.4425042564574184e-05,
-      "loss": 0.1592,
+      "loss": 0.1407,
       "step": 196
     },
     {
       "epoch": 1.5533596837944663,
-      "grad_norm": 0.8229113221168518,
+      "grad_norm": 0.46086326241493225,
       "learning_rate": 2.3581308246275103e-05,
-      "loss": 0.1244,
+      "loss": 0.1084,
       "step": 197
     },
     {
       "epoch": 1.5612648221343872,
-      "grad_norm": 1.0748467445373535,
+      "grad_norm": 0.9662849307060242,
       "learning_rate": 2.2750452345848682e-05,
-      "loss": 0.2833,
+      "loss": 0.2847,
       "step": 198
     },
     {
       "epoch": 1.5691699604743083,
-      "grad_norm": 0.7150699496269226,
+      "grad_norm": 0.799304723739624,
       "learning_rate": 2.1932614882827197e-05,
-      "loss": 0.1917,
+      "loss": 0.189,
       "step": 199
     },
     {
       "epoch": 1.5770750988142292,
-      "grad_norm": 0.5945517420768738,
+      "grad_norm": 0.6747804880142212,
       "learning_rate": 2.112793368281799e-05,
-      "loss": 0.193,
+      "loss": 0.1797,
       "step": 200
     },
     {
       "epoch": 1.5849802371541502,
-      "grad_norm": 0.5897226929664612,
+      "grad_norm": 0.5679678320884705,
       "learning_rate": 2.03365443542764e-05,
-      "loss": 0.1404,
+      "loss": 0.1284,
       "step": 201
     },
     {
       "epoch": 1.5928853754940713,
-      "grad_norm": 1.019797921180725,
+      "grad_norm": 0.669952929019928,
       "learning_rate": 1.9558580265652448e-05,
-      "loss": 0.2734,
+      "loss": 0.2349,
       "step": 202
     },
     {
       "epoch": 1.6007905138339922,
-      "grad_norm": 0.6552490592002869,
+      "grad_norm": 0.5827800035476685,
       "learning_rate": 1.879417252291502e-05,
-      "loss": 0.1588,
+      "loss": 0.1668,
       "step": 203
     },
     {
       "epoch": 1.608695652173913,
-      "grad_norm": 0.6750946044921875,
+      "grad_norm": 0.8047837018966675,
       "learning_rate": 1.804344994745727e-05,
-      "loss": 0.1418,
+      "loss": 0.1244,
       "step": 204
     },
     {
       "epoch": 1.616600790513834,
-      "grad_norm": 0.7633326053619385,
+      "grad_norm": 0.7034929990768433,
       "learning_rate": 1.730653905438714e-05,
-      "loss": 0.2295,
+      "loss": 0.2241,
       "step": 205
     },
     {
       "epoch": 1.6245059288537549,
-      "grad_norm": 1.0973371267318726,
+      "grad_norm": 0.9926624894142151,
       "learning_rate": 1.6583564031206357e-05,
-      "loss": 0.1814,
+      "loss": 0.1807,
       "step": 206
     },
     {
       "epoch": 1.6324110671936758,
-      "grad_norm": 0.7115572094917297,
+      "grad_norm": 0.7195786833763123,
       "learning_rate": 1.587464671688187e-05,
-      "loss": 0.2119,
+      "loss": 0.2133,
       "step": 207
     },
     {
       "epoch": 1.6403162055335967,
-      "grad_norm": 0.9038087725639343,
+      "grad_norm": 0.6933628916740417,
       "learning_rate": 1.5179906581313064e-05,
-      "loss": 0.2032,
+      "loss": 0.1928,
       "step": 208
     },
     {
       "epoch": 1.6482213438735178,
-      "grad_norm": 0.8214746117591858,
+      "grad_norm": 1.1264605522155762,
       "learning_rate": 1.4499460705197998e-05,
-      "loss": 0.3846,
+      "loss": 0.3693,
       "step": 209
     },
     {
       "epoch": 1.6561264822134387,
-      "grad_norm": 0.7156324982643127,
+      "grad_norm": 0.7283610701560974,
       "learning_rate": 1.3833423760302611e-05,
-      "loss": 0.1685,
+      "loss": 0.1677,
       "step": 210
     },
     {
       "epoch": 1.6640316205533598,
-      "grad_norm": 0.8988309502601624,
+      "grad_norm": 1.039823055267334,
       "learning_rate": 1.3181907990135622e-05,
-      "loss": 0.1499,
+      "loss": 0.1639,
       "step": 211
     },
     {
       "epoch": 1.6719367588932808,
-      "grad_norm": 0.6503781676292419,
+      "grad_norm": 0.933429479598999,
       "learning_rate": 1.2545023191032801e-05,
-      "loss": 0.1936,
+      "loss": 0.1798,
       "step": 212
     },
     {
       "epoch": 1.6798418972332017,
-      "grad_norm": 1.4845696687698364,
+      "grad_norm": 1.0857698917388916,
       "learning_rate": 1.1922876693653585e-05,
-      "loss": 0.1949,
+      "loss": 0.2107,
       "step": 213
     },
     {
       "epoch": 1.6877470355731226,
-      "grad_norm": 1.4460773468017578,
+      "grad_norm": 1.2482712268829346,
       "learning_rate": 1.131557334489326e-05,
-      "loss": 0.2512,
+      "loss": 0.2459,
       "step": 214
     },
     {
       "epoch": 1.6956521739130435,
-      "grad_norm": 2.0067126750946045,
+      "grad_norm": 0.8419239521026611,
       "learning_rate": 1.0723215490213634e-05,
-      "loss": 0.2433,
+      "loss": 0.2451,
       "step": 215
     },
     {
       "epoch": 1.7035573122529644,
-      "grad_norm": 0.6824133992195129,
+      "grad_norm": 0.8410497903823853,
       "learning_rate": 1.0145902956395447e-05,
-      "loss": 0.2181,
+      "loss": 0.2175,
       "step": 216
     },
     {
       "epoch": 1.7114624505928853,
-      "grad_norm": 1.23275887966156,
+      "grad_norm": 1.390134334564209,
       "learning_rate": 9.583733034714981e-06,
-      "loss": 0.4743,
+      "loss": 0.4648,
       "step": 217
     },
     {
       "epoch": 1.7193675889328062,
-      "grad_norm": 0.6075592041015625,
+      "grad_norm": 0.5143908262252808,
       "learning_rate": 9.036800464548157e-06,
-      "loss": 0.1184,
+      "loss": 0.112,
       "step": 218
     },
     {
       "epoch": 1.7272727272727273,
-      "grad_norm": 0.8228341937065125,
+      "grad_norm": 0.6791729927062988,
       "learning_rate": 8.505197417404687e-06,
-      "loss": 0.1745,
+      "loss": 0.1596,
       "step": 219
     },
     {
       "epoch": 1.7351778656126482,
-      "grad_norm": 0.7998052835464478,
+      "grad_norm": 0.794528603553772,
       "learning_rate": 7.989013481394814e-06,
-      "loss": 0.3358,
+      "loss": 0.3435,
       "step": 220
     },
     {
       "epoch": 1.7430830039525693,
-      "grad_norm": 0.9460540413856506,
+      "grad_norm": 0.7834582328796387,
       "learning_rate": 7.488335646131628e-06,
-      "loss": 0.3009,
+      "loss": 0.2931,
       "step": 221
     },
     {
       "epoch": 1.7509881422924902,
-      "grad_norm": 1.0127474069595337,
+      "grad_norm": 2.8932089805603027,
       "learning_rate": 7.003248288071118e-06,
-      "loss": 0.3223,
+      "loss": 0.3275,
       "step": 222
     },
     {
       "epoch": 1.7588932806324111,
-      "grad_norm": 1.1740081310272217,
+      "grad_norm": 0.5584789514541626,
       "learning_rate": 6.533833156292679e-06,
-      "loss": 0.1361,
+      "loss": 0.1267,
       "step": 223
     },
     {
       "epoch": 1.766798418972332,
-      "grad_norm": 1.0067321062088013,
+      "grad_norm": 0.9573125839233398,
       "learning_rate": 6.08016935872251e-06,
-      "loss": 0.2048,
+      "loss": 0.215,
       "step": 224
     },
     {
       "epoch": 1.774703557312253,
-      "grad_norm": 0.8940996527671814,
+      "grad_norm": 0.8806841969490051,
       "learning_rate": 5.6423333488018095e-06,
-      "loss": 0.3323,
+      "loss": 0.3208,
       "step": 225
     },
     {
       "epoch": 1.7826086956521738,
-      "grad_norm": 0.6231039762496948,
+      "grad_norm": 0.49562156200408936,
       "learning_rate": 5.22039891260262e-06,
-      "loss": 0.1062,
+      "loss": 0.093,
       "step": 226
     },
     {
       "epoch": 1.7905138339920947,
-      "grad_norm": 0.628740668296814,
+      "grad_norm": 0.5696113705635071,
       "learning_rate": 4.8144371563930476e-06,
-      "loss": 0.1944,
+      "loss": 0.1829,
       "step": 227
     },
     {
       "epoch": 1.7984189723320159,
-      "grad_norm": 0.766796350479126,
+      "grad_norm": 0.7890591621398926,
       "learning_rate": 4.424516494654118e-06,
-      "loss": 0.1936,
+      "loss": 0.1888,
       "step": 228
     },
     {
       "epoch": 1.8063241106719368,
-      "grad_norm": 1.4939095973968506,
+      "grad_norm": 1.111238718032837,
       "learning_rate": 4.050702638550275e-06,
-      "loss": 0.2269,
+      "loss": 0.2312,
       "step": 229
     },
     {
       "epoch": 1.8142292490118577,
-      "grad_norm": 1.01512610912323,
+      "grad_norm": 1.0172678232192993,
       "learning_rate": 3.693058584855369e-06,
-      "loss": 0.4442,
+      "loss": 0.4459,
       "step": 230
     },
     {
       "epoch": 1.8221343873517788,
-      "grad_norm": 0.6523633003234863,
+      "grad_norm": 0.6409620642662048,
       "learning_rate": 3.3516446053363015e-06,
-      "loss": 0.182,
+      "loss": 0.1838,
       "step": 231
     },
     {
       "epoch": 1.8300395256916997,
-      "grad_norm": 0.6220389008522034,
+      "grad_norm": 1.010557770729065,
       "learning_rate": 3.026518236595621e-06,
-      "loss": 0.1633,
+      "loss": 0.1648,
       "step": 232
     },
     {
       "epoch": 1.8379446640316206,
-      "grad_norm": 0.6830483078956604,
+      "grad_norm": 0.7010674476623535,
       "learning_rate": 2.717734270375272e-06,
-      "loss": 0.1698,
+      "loss": 0.1565,
       "step": 233
     },
     {
       "epoch": 1.8458498023715415,
-      "grad_norm": 2.016892194747925,
+      "grad_norm": 0.8774541616439819,
       "learning_rate": 2.4253447443228106e-06,
-      "loss": 0.2407,
+      "loss": 0.2482,
       "step": 234
     },
     {
       "epoch": 1.8537549407114624,
-      "grad_norm": 1.519871711730957,
+      "grad_norm": 1.4902609586715698,
       "learning_rate": 2.1493989332218468e-06,
-      "loss": 0.6052,
+      "loss": 0.6009,
       "step": 235
     },
     {
       "epoch": 1.8616600790513833,
-      "grad_norm": 0.8474094867706299,
+      "grad_norm": 0.8265174627304077,
       "learning_rate": 1.8899433406879608e-06,
-      "loss": 0.1317,
+      "loss": 0.1213,
       "step": 236
     },
     {
       "epoch": 1.8695652173913042,
-      "grad_norm": 1.0883585214614868,
+      "grad_norm": 2.709406852722168,
       "learning_rate": 1.6470216913317626e-06,
-      "loss": 0.205,
+      "loss": 0.1897,
       "step": 237
     },
     {
       "epoch": 1.8774703557312253,
-      "grad_norm": 0.5493878126144409,
+      "grad_norm": 0.6855682730674744,
       "learning_rate": 1.4206749233902084e-06,
-      "loss": 0.1517,
+      "loss": 0.1564,
       "step": 238
     },
     {
       "epoch": 1.8853754940711462,
-      "grad_norm": 0.5892478227615356,
+      "grad_norm": 0.5468905568122864,
       "learning_rate": 1.2109411818274852e-06,
-      "loss": 0.1239,
+      "loss": 0.1241,
       "step": 239
     },
     {
       "epoch": 1.8932806324110671,
-      "grad_norm": 0.6910550594329834,
+      "grad_norm": 0.5691235661506653,
       "learning_rate": 1.0178558119067315e-06,
-      "loss": 0.146,
+      "loss": 0.1394,
       "step": 240
     },
     {
       "epoch": 1.9011857707509883,
-      "grad_norm": 2.088679790496826,
+      "grad_norm": 0.8554819822311401,
       "learning_rate": 8.41451353233369e-07,
-      "loss": 0.2145,
+      "loss": 0.2113,
       "step": 241
     },
     {
       "epoch": 1.9090909090909092,
-      "grad_norm": 1.0188775062561035,
+      "grad_norm": 1.5547891855239868,
       "learning_rate": 6.817575342714988e-07,
-      "loss": 0.4285,
+      "loss": 0.4036,
       "step": 242
     },
     {
       "epoch": 1.91699604743083,
-      "grad_norm": 0.9871934056282043,
+      "grad_norm": 1.505486249923706,
       "learning_rate": 5.388012673338661e-07,
-      "loss": 0.248,
+      "loss": 0.2284,
       "step": 243
     },
     {
       "epoch": 1.924901185770751,
-      "grad_norm": 0.7955995798110962,
+      "grad_norm": 0.9467947483062744,
       "learning_rate": 4.126066440464982e-07,
-      "loss": 0.3334,
+      "loss": 0.3383,
       "step": 244
     },
     {
       "epoch": 1.9328063241106719,
-      "grad_norm": 1.2158209085464478,
+      "grad_norm": 1.066933274269104,
       "learning_rate": 3.0319493128866396e-07,
-      "loss": 0.3475,
+      "loss": 0.3404,
       "step": 245
     },
     {
       "epoch": 1.9407114624505928,
-      "grad_norm": 3.033383369445801,
+      "grad_norm": 1.2586652040481567,
       "learning_rate": 2.1058456760891798e-07,
-      "loss": 0.2237,
+      "loss": 0.2501,
       "step": 246
     },
     {
       "epoch": 1.9486166007905137,
-      "grad_norm": 0.9176952242851257,
+      "grad_norm": 1.629897117614746,
       "learning_rate": 1.3479116011769767e-07,
-      "loss": 0.2966,
+      "loss": 0.3085,
       "step": 247
     },
     {
       "epoch": 1.9565217391304348,
-      "grad_norm": 0.9858683943748474,
+      "grad_norm": 0.8343656063079834,
       "learning_rate": 7.582748185719358e-08,
-      "loss": 0.339,
+      "loss": 0.3449,
       "step": 248
     },
     {
       "epoch": 1.9644268774703557,
-      "grad_norm": 1.3175891637802124,
+      "grad_norm": 1.270969271659851,
       "learning_rate": 3.370346964876036e-08,
-      "loss": 0.4174,
+      "loss": 0.4198,
       "step": 249
     },
     {
       "epoch": 1.9723320158102768,
-      "grad_norm": 1.8579566478729248,
+      "grad_norm": 1.3988014459609985,
       "learning_rate": 8.426222418311814e-09,
-      "loss": 0.611,
+      "loss": 0.619,
       "step": 250
     }
   ],

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:870d6061a3cfd60bcc0dcabc7a4c223e02ed071b6cad5e90dd66a232cf240cc3
+oid sha256:e4e77794bbdd2eb0fb1754c2e6637d987d75bed42f3093d2f32cb758d7ef33e8
 size 6353