Training in progress, step 500, checkpoint

last-checkpoint/model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:79f312030784e0c4c1ac477ec8be9baccaf31bb453a5f1f0ba47bae3a11d6736
+oid sha256:d3eec04603345e3344b56349bbf3c8f2ab8c2870fb856bc03b802e4a9b84023a
 size 791869518

last-checkpoint/optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b9f0413ff5d7932d1e3cf4853d666b874dcafb75ec0e3ed0aec08553a6ec61db
+oid sha256:e596f723566f73180179176318fe50c96148d6e69a806bcd5b46bd82b7f1cd4e
 size 2375752250

last-checkpoint/scheduler.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:68d8c53b42f31b6c7313464394859eb2afcdd967a1a813faffd1e0be60216485
+oid sha256:4b2a49e94ec437b7eac846837816e3fa765a1213db3ca835c9332ec400042339
 size 1000

last-checkpoint/trainer_state.json

@@ -1,986 +1,720 @@
 {
   "best_metric": null,
   "best_model_checkpoint": null,
-  "epoch": 0.999909559555033,
+  "epoch": 0.7235235597359139,
   "eval_steps": 500,
-  "global_step": 691,
+  "global_step": 500,
   "is_hyper_param_search": false,
   "is_local_process_zero": true,
   "is_world_process_zero": true,
   "log_history": [
     {
       "epoch": 0.007235235597359139,
-      "grad_norm": 152.25,
+      "grad_norm": 240.625,
       "learning_rate": 7.142857142857143e-07,
-      "loss": 124.6481,
+      "loss": 119.8148,
       "step": 5
     },
     {
       "epoch": 0.014470471194718278,
-      "grad_norm": 150.625,
+      "grad_norm": 138.625,
       "learning_rate": 1.4285714285714286e-06,
-      "loss": 119.964,
+      "loss": 104.262,
       "step": 10
     },
     {
       "epoch": 0.021705706792077416,
-      "grad_norm": 150.375,
+      "grad_norm": 121.875,
       "learning_rate": 2.1428571428571427e-06,
-      "loss": 118.5022,
+      "loss": 93.6603,
       "step": 15
     },
     {
       "epoch": 0.028940942389436557,
-      "grad_norm": 168.125,
+      "grad_norm": 120.625,
       "learning_rate": 2.8571428571428573e-06,
-      "loss": 116.7498,
+      "loss": 90.7159,
       "step": 20
     },
     {
       "epoch": 0.0361761779867957,
-      "grad_norm": 143.375,
+      "grad_norm": 106.5625,
       "learning_rate": 3.5714285714285718e-06,
-      "loss": 114.4503,
+      "loss": 86.0796,
       "step": 25
     },
     {
       "epoch": 0.04341141358415483,
-      "grad_norm": 135.125,
+      "grad_norm": 103.875,
       "learning_rate": 4.2857142857142855e-06,
-      "loss": 114.6372,
+      "loss": 86.2538,
       "step": 30
     },
     {
       "epoch": 0.05064664918151397,
-      "grad_norm": 149.875,
+      "grad_norm": 108.375,
       "learning_rate": 5e-06,
-      "loss": 111.9492,
+      "loss": 84.5117,
       "step": 35
     },
     {
       "epoch": 0.05788188477887311,
-      "grad_norm": 141.75,
+      "grad_norm": 101.0,
       "learning_rate": 5.7142857142857145e-06,
-      "loss": 112.9367,
+      "loss": 83.5148,
       "step": 40
     },
     {
       "epoch": 0.06511712037623225,
-      "grad_norm": 159.0,
+      "grad_norm": 109.875,
       "learning_rate": 6.4285714285714295e-06,
-      "loss": 113.2503,
+      "loss": 82.3289,
       "step": 45
     },
     {
       "epoch": 0.0723523559735914,
-      "grad_norm": 147.0,
+      "grad_norm": 102.5625,
       "learning_rate": 7.1428571428571436e-06,
-      "loss": 110.1401,
+      "loss": 81.4988,
       "step": 50
     },
     {
       "epoch": 0.07958759157095052,
-      "grad_norm": 161.125,
+      "grad_norm": 115.375,
       "learning_rate": 7.857142857142858e-06,
-      "loss": 110.8394,
+      "loss": 82.3074,
       "step": 55
     },
     {
       "epoch": 0.08682282716830966,
-      "grad_norm": 158.125,
+      "grad_norm": 115.625,
       "learning_rate": 8.571428571428571e-06,
-      "loss": 113.1161,
+      "loss": 81.3627,
       "step": 60
     },
     {
       "epoch": 0.0940580627656688,
-      "grad_norm": 159.0,
+      "grad_norm": 103.0625,
       "learning_rate": 9.285714285714288e-06,
-      "loss": 111.898,
+      "loss": 83.3712,
       "step": 65
     },
     {
       "epoch": 0.10129329836302794,
-      "grad_norm": 143.0,
+      "grad_norm": 118.625,
       "learning_rate": 1e-05,
-      "loss": 112.3912,
+      "loss": 81.1766,
       "step": 70
     },
     {
       "epoch": 0.10852853396038709,
-      "grad_norm": 159.5,
+      "grad_norm": 120.75,
       "learning_rate": 9.919484702093398e-06,
-      "loss": 111.8873,
+      "loss": 80.6615,
       "step": 75
     },
     {
       "epoch": 0.11576376955774623,
-      "grad_norm": 162.875,
+      "grad_norm": 108.5,
       "learning_rate": 9.838969404186796e-06,
-      "loss": 111.136,
+      "loss": 78.3733,
       "step": 80
     },
     {
       "epoch": 0.12299900515510537,
-      "grad_norm": 159.0,
+      "grad_norm": 102.375,
       "learning_rate": 9.758454106280194e-06,
-      "loss": 113.4827,
+      "loss": 80.2047,
       "step": 85
     },
     {
       "epoch": 0.1302342407524645,
-      "grad_norm": 155.625,
+      "grad_norm": 106.875,
       "learning_rate": 9.677938808373591e-06,
-      "loss": 111.4002,
+      "loss": 79.1781,
       "step": 90
     },
     {
       "epoch": 0.13746947634982365,
-      "grad_norm": 167.75,
+      "grad_norm": 105.6875,
       "learning_rate": 9.59742351046699e-06,
-      "loss": 111.9107,
+      "loss": 79.691,
       "step": 95
     },
     {
       "epoch": 0.1447047119471828,
-      "grad_norm": 153.375,
+      "grad_norm": 116.125,
       "learning_rate": 9.516908212560388e-06,
-      "loss": 113.0265,
+      "loss": 80.0193,
       "step": 100
     },
     {
       "epoch": 0.15193994754454193,
-      "grad_norm": 166.0,
+      "grad_norm": 115.0,
       "learning_rate": 9.436392914653784e-06,
-      "loss": 110.3215,
+      "loss": 80.1022,
       "step": 105
     },
     {
       "epoch": 0.15917518314190104,
-      "grad_norm": 141.0,
+      "grad_norm": 117.875,
       "learning_rate": 9.355877616747183e-06,
-      "loss": 110.7277,
+      "loss": 78.4374,
       "step": 110
     },
     {
       "epoch": 0.16641041873926019,
-      "grad_norm": 147.5,
+      "grad_norm": 103.3125,
       "learning_rate": 9.275362318840581e-06,
-      "loss": 112.1578,
+      "loss": 78.8168,
       "step": 115
     },
     {
       "epoch": 0.17364565433661933,
-      "grad_norm": 155.125,
+      "grad_norm": 102.125,
       "learning_rate": 9.194847020933978e-06,
-      "loss": 111.7271,
+      "loss": 77.7842,
       "step": 120
     },
     {
       "epoch": 0.18088088993397847,
-      "grad_norm": 149.25,
+      "grad_norm": 105.375,
       "learning_rate": 9.114331723027376e-06,
-      "loss": 109.4189,
+      "loss": 77.5425,
       "step": 125
     },
     {
       "epoch": 0.1881161255313376,
-      "grad_norm": 159.125,
+      "grad_norm": 105.5,
       "learning_rate": 9.033816425120775e-06,
-      "loss": 110.2367,
+      "loss": 76.4404,
       "step": 130
     },
     {
       "epoch": 0.19535136112869675,
-      "grad_norm": 146.75,
+      "grad_norm": 105.125,
       "learning_rate": 8.953301127214171e-06,
-      "loss": 110.2772,
+      "loss": 76.6364,
       "step": 135
     },
     {
       "epoch": 0.2025865967260559,
-      "grad_norm": 143.375,
+      "grad_norm": 124.375,
       "learning_rate": 8.87278582930757e-06,
-      "loss": 111.3193,
+      "loss": 76.9181,
       "step": 140
     },
     {
       "epoch": 0.20982183232341503,
-      "grad_norm": 153.375,
+      "grad_norm": 103.0625,
       "learning_rate": 8.792270531400966e-06,
-      "loss": 108.8334,
+      "loss": 77.4936,
       "step": 145
     },
     {
       "epoch": 0.21705706792077417,
-      "grad_norm": 165.375,
+      "grad_norm": 104.875,
       "learning_rate": 8.711755233494365e-06,
-      "loss": 108.6805,
+      "loss": 76.7636,
       "step": 150
     },
     {
       "epoch": 0.2242923035181333,
-      "grad_norm": 152.125,
+      "grad_norm": 113.8125,
       "learning_rate": 8.631239935587761e-06,
-      "loss": 109.1355,
+      "loss": 76.7539,
       "step": 155
     },
     {
       "epoch": 0.23152753911549245,
-      "grad_norm": 157.375,
+      "grad_norm": 111.875,
       "learning_rate": 8.55072463768116e-06,
-      "loss": 108.0569,
+      "loss": 75.9734,
       "step": 160
     },
     {
       "epoch": 0.2387627747128516,
-      "grad_norm": 146.625,
+      "grad_norm": 98.75,
       "learning_rate": 8.470209339774558e-06,
-      "loss": 109.2913,
+      "loss": 74.6409,
       "step": 165
     },
     {
       "epoch": 0.24599801031021074,
-      "grad_norm": 158.375,
+      "grad_norm": 107.5625,
       "learning_rate": 8.389694041867955e-06,
-      "loss": 107.8965,
+      "loss": 76.4473,
       "step": 170
     },
     {
       "epoch": 0.25323324590756985,
-      "grad_norm": 153.25,
+      "grad_norm": 109.5,
       "learning_rate": 8.309178743961353e-06,
-      "loss": 107.1813,
+      "loss": 75.0732,
       "step": 175
     },
     {
       "epoch": 0.260468481504929,
-      "grad_norm": 145.25,
+      "grad_norm": 114.5,
       "learning_rate": 8.228663446054752e-06,
-      "loss": 107.5406,
+      "loss": 74.5601,
       "step": 180
     },
     {
       "epoch": 0.26770371710228813,
-      "grad_norm": 142.125,
+      "grad_norm": 101.9375,
       "learning_rate": 8.148148148148148e-06,
-      "loss": 107.5163,
+      "loss": 77.3542,
       "step": 185
     },
     {
       "epoch": 0.2749389526996473,
-      "grad_norm": 151.5,
+      "grad_norm": 107.75,
       "learning_rate": 8.067632850241547e-06,
-      "loss": 107.8385,
+      "loss": 74.8798,
       "step": 190
     },
     {
       "epoch": 0.2821741882970064,
-      "grad_norm": 143.125,
+      "grad_norm": 107.75,
       "learning_rate": 7.987117552334945e-06,
-      "loss": 106.562,
+      "loss": 75.2804,
       "step": 195
     },
     {
       "epoch": 0.2894094238943656,
-      "grad_norm": 147.75,
+      "grad_norm": 97.3125,
       "learning_rate": 7.906602254428342e-06,
-      "loss": 104.995,
+      "loss": 74.8395,
       "step": 200
     },
     {
       "epoch": 0.2966446594917247,
-      "grad_norm": 151.375,
+      "grad_norm": 96.9375,
       "learning_rate": 7.82608695652174e-06,
-      "loss": 106.541,
+      "loss": 72.9746,
       "step": 205
     },
     {
       "epoch": 0.30387989508908386,
-      "grad_norm": 153.25,
+      "grad_norm": 102.25,
       "learning_rate": 7.745571658615137e-06,
-      "loss": 106.8991,
+      "loss": 74.5221,
       "step": 210
     },
     {
       "epoch": 0.311115130686443,
-      "grad_norm": 160.25,
+      "grad_norm": 100.9375,
       "learning_rate": 7.665056360708535e-06,
-      "loss": 104.9197,
+      "loss": 72.7894,
       "step": 215
     },
     {
       "epoch": 0.3183503662838021,
-      "grad_norm": 152.875,
+      "grad_norm": 102.625,
       "learning_rate": 7.584541062801934e-06,
-      "loss": 105.1669,
+      "loss": 73.3365,
       "step": 220
     },
     {
       "epoch": 0.32558560188116126,
-      "grad_norm": 150.875,
+      "grad_norm": 102.125,
       "learning_rate": 7.504025764895331e-06,
-      "loss": 103.8021,
+      "loss": 73.3862,
       "step": 225
     },
     {
       "epoch": 0.33282083747852037,
-      "grad_norm": 135.125,
+      "grad_norm": 100.25,
       "learning_rate": 7.423510466988728e-06,
-      "loss": 104.6592,
+      "loss": 71.9684,
       "step": 230
     },
     {
       "epoch": 0.34005607307587954,
-      "grad_norm": 145.5,
+      "grad_norm": 102.5625,
       "learning_rate": 7.342995169082127e-06,
-      "loss": 103.5859,
+      "loss": 72.8994,
       "step": 235
     },
     {
       "epoch": 0.34729130867323865,
-      "grad_norm": 148.875,
+      "grad_norm": 105.1875,
       "learning_rate": 7.262479871175524e-06,
-      "loss": 104.8537,
+      "loss": 71.1332,
       "step": 240
     },
     {
       "epoch": 0.3545265442705978,
-      "grad_norm": 149.75,
+      "grad_norm": 100.125,
       "learning_rate": 7.181964573268921e-06,
-      "loss": 104.8267,
+      "loss": 70.3095,
       "step": 245
     },
     {
       "epoch": 0.36176177986795693,
-      "grad_norm": 162.625,
+      "grad_norm": 99.375,
       "learning_rate": 7.10144927536232e-06,
-      "loss": 104.704,
+      "loss": 70.9021,
       "step": 250
     },
     {
       "epoch": 0.3689970154653161,
-      "grad_norm": 143.375,
+      "grad_norm": 110.25,
       "learning_rate": 7.020933977455717e-06,
-      "loss": 105.0362,
+      "loss": 72.08,
       "step": 255
     },
     {
       "epoch": 0.3762322510626752,
-      "grad_norm": 148.375,
+      "grad_norm": 95.0625,
       "learning_rate": 6.940418679549115e-06,
-      "loss": 102.2345,
+      "loss": 70.8334,
       "step": 260
     },
     {
       "epoch": 0.3834674866600344,
-      "grad_norm": 152.375,
+      "grad_norm": 101.3125,
       "learning_rate": 6.859903381642513e-06,
-      "loss": 105.5086,
+      "loss": 70.9255,
       "step": 265
     },
     {
       "epoch": 0.3907027222573935,
-      "grad_norm": 140.125,
+      "grad_norm": 102.8125,
       "learning_rate": 6.779388083735911e-06,
-      "loss": 103.5476,
+      "loss": 72.2681,
       "step": 270
     },
     {
       "epoch": 0.39793795785475267,
-      "grad_norm": 148.375,
+      "grad_norm": 101.25,
       "learning_rate": 6.698872785829308e-06,
-      "loss": 103.0274,
+      "loss": 70.6592,
       "step": 275
     },
     {
       "epoch": 0.4051731934521118,
-      "grad_norm": 151.0,
+      "grad_norm": 104.0,
       "learning_rate": 6.6183574879227065e-06,
-      "loss": 103.2487,
+      "loss": 70.5197,
       "step": 280
     },
     {
       "epoch": 0.41240842904947095,
-      "grad_norm": 143.5,
+      "grad_norm": 103.5,
       "learning_rate": 6.537842190016104e-06,
-      "loss": 102.4844,
+      "loss": 70.3267,
       "step": 285
     },
     {
       "epoch": 0.41964366464683006,
-      "grad_norm": 139.625,
+      "grad_norm": 95.6875,
       "learning_rate": 6.457326892109501e-06,
-      "loss": 102.6165,
+      "loss": 69.8351,
       "step": 290
     },
     {
       "epoch": 0.4268789002441892,
-      "grad_norm": 135.0,
+      "grad_norm": 95.9375,
       "learning_rate": 6.376811594202898e-06,
-      "loss": 101.8949,
+      "loss": 69.768,
       "step": 295
     },
     {
       "epoch": 0.43411413584154834,
-      "grad_norm": 159.125,
+      "grad_norm": 99.9375,
       "learning_rate": 6.296296296296297e-06,
-      "loss": 101.1282,
+      "loss": 68.9199,
       "step": 300
     },
     {
       "epoch": 0.44134937143890746,
-      "grad_norm": 146.125,
+      "grad_norm": 103.8125,
       "learning_rate": 6.215780998389694e-06,
-      "loss": 102.0691,
+      "loss": 70.4392,
       "step": 305
     },
     {
       "epoch": 0.4485846070362666,
-      "grad_norm": 145.625,
+      "grad_norm": 94.6875,
       "learning_rate": 6.135265700483092e-06,
-      "loss": 101.359,
+      "loss": 69.9065,
       "step": 310
     },
     {
       "epoch": 0.45581984263362574,
-      "grad_norm": 148.5,
+      "grad_norm": 110.9375,
       "learning_rate": 6.05475040257649e-06,
-      "loss": 101.5264,
+      "loss": 68.7817,
       "step": 315
     },
     {
       "epoch": 0.4630550782309849,
-      "grad_norm": 139.375,
+      "grad_norm": 95.25,
       "learning_rate": 5.974235104669888e-06,
-      "loss": 99.743,
+      "loss": 69.2291,
       "step": 320
     },
     {
       "epoch": 0.470290313828344,
-      "grad_norm": 145.0,
+      "grad_norm": 94.5,
       "learning_rate": 5.893719806763285e-06,
-      "loss": 103.0054,
+      "loss": 69.6696,
       "step": 325
     },
     {
       "epoch": 0.4775255494257032,
-      "grad_norm": 155.875,
+      "grad_norm": 100.9375,
       "learning_rate": 5.8132045088566835e-06,
-      "loss": 99.6074,
+      "loss": 68.0409,
       "step": 330
     },
     {
       "epoch": 0.4847607850230623,
-      "grad_norm": 157.25,
+      "grad_norm": 95.5,
       "learning_rate": 5.732689210950081e-06,
-      "loss": 100.957,
+      "loss": 68.8582,
       "step": 335
     },
     {
       "epoch": 0.49199602062042147,
-      "grad_norm": 145.125,
+      "grad_norm": 94.6875,
       "learning_rate": 5.652173913043479e-06,
-      "loss": 100.6761,
+      "loss": 68.1767,
       "step": 340
     },
     {
       "epoch": 0.4992312562177806,
-      "grad_norm": 144.5,
+      "grad_norm": 96.0625,
       "learning_rate": 5.571658615136877e-06,
-      "loss": 99.3499,
+      "loss": 68.3804,
       "step": 345
     },
     {
       "epoch": 0.5064664918151397,
-      "grad_norm": 165.25,
+      "grad_norm": 97.125,
       "learning_rate": 5.4911433172302745e-06,
-      "loss": 101.0515,
+      "loss": 66.2784,
       "step": 350
     },
     {
       "epoch": 0.5137017274124989,
-      "grad_norm": 140.625,
+      "grad_norm": 93.75,
       "learning_rate": 5.410628019323671e-06,
-      "loss": 100.2075,
+      "loss": 66.8873,
       "step": 355
     },
     {
       "epoch": 0.520936963009858,
-      "grad_norm": 148.625,
+      "grad_norm": 96.0,
       "learning_rate": 5.3301127214170704e-06,
-      "loss": 101.0406,
+      "loss": 67.6164,
       "step": 360
     },
     {
       "epoch": 0.5281721986072172,
-      "grad_norm": 137.625,
+      "grad_norm": 106.625,
       "learning_rate": 5.249597423510467e-06,
-      "loss": 100.4312,
+      "loss": 68.5509,
       "step": 365
     },
     {
       "epoch": 0.5354074342045763,
-      "grad_norm": 143.25,
+      "grad_norm": 102.125,
       "learning_rate": 5.169082125603865e-06,
-      "loss": 98.7602,
+      "loss": 65.9846,
       "step": 370
     },
     {
       "epoch": 0.5426426698019354,
-      "grad_norm": 145.625,
+      "grad_norm": 99.125,
       "learning_rate": 5.088566827697263e-06,
-      "loss": 99.5661,
+      "loss": 67.4898,
       "step": 375
     },
     {
       "epoch": 0.5498779053992946,
-      "grad_norm": 130.875,
+      "grad_norm": 98.25,
       "learning_rate": 5.0080515297906606e-06,
-      "loss": 98.5447,
+      "loss": 68.5553,
       "step": 380
     },
     {
       "epoch": 0.5571131409966537,
-      "grad_norm": 149.25,
+      "grad_norm": 99.5,
       "learning_rate": 4.927536231884059e-06,
-      "loss": 98.0122,
+      "loss": 66.9819,
       "step": 385
     },
     {
       "epoch": 0.5643483765940128,
-      "grad_norm": 139.625,
+      "grad_norm": 102.375,
       "learning_rate": 4.847020933977456e-06,
-      "loss": 99.2832,
+      "loss": 65.8663,
       "step": 390
     },
     {
       "epoch": 0.571583612191372,
-      "grad_norm": 151.25,
+      "grad_norm": 97.5,
       "learning_rate": 4.766505636070854e-06,
-      "loss": 98.1141,
+      "loss": 67.4699,
       "step": 395
     },
     {
       "epoch": 0.5788188477887312,
-      "grad_norm": 153.75,
+      "grad_norm": 104.4375,
       "learning_rate": 4.6859903381642516e-06,
-      "loss": 96.6736,
+      "loss": 66.4532,
       "step": 400
     },
     {
       "epoch": 0.5860540833860902,
-      "grad_norm": 153.375,
+      "grad_norm": 100.9375,
       "learning_rate": 4.605475040257649e-06,
-      "loss": 98.9147,
+      "loss": 66.7904,
       "step": 405
     },
     {
       "epoch": 0.5932893189834494,
-      "grad_norm": 135.0,
+      "grad_norm": 96.875,
       "learning_rate": 4.5249597423510475e-06,
-      "loss": 97.6523,
+      "loss": 66.644,
       "step": 410
     },
     {
       "epoch": 0.6005245545808086,
-      "grad_norm": 140.375,
+      "grad_norm": 95.8125,
       "learning_rate": 4.444444444444444e-06,
-      "loss": 96.3539,
+      "loss": 64.846,
       "step": 415
     },
     {
       "epoch": 0.6077597901781677,
-      "grad_norm": 143.125,
+      "grad_norm": 96.625,
       "learning_rate": 4.3639291465378425e-06,
-      "loss": 99.4855,
+      "loss": 65.7853,
       "step": 420
     },
     {
       "epoch": 0.6149950257755268,
-      "grad_norm": 150.5,
+      "grad_norm": 93.0625,
       "learning_rate": 4.28341384863124e-06,
-      "loss": 98.0349,
+      "loss": 64.4889,
       "step": 425
     },
     {
       "epoch": 0.622230261372886,
-      "grad_norm": 135.5,
+      "grad_norm": 106.5,
       "learning_rate": 4.202898550724638e-06,
-      "loss": 97.4008,
+      "loss": 65.0533,
       "step": 430
     },
     {
       "epoch": 0.6294654969702451,
-      "grad_norm": 139.25,
+      "grad_norm": 92.5,
       "learning_rate": 4.122383252818036e-06,
-      "loss": 96.3096,
+      "loss": 65.3124,
       "step": 435
     },
     {
       "epoch": 0.6367007325676042,
-      "grad_norm": 138.0,
+      "grad_norm": 94.875,
       "learning_rate": 4.0418679549114335e-06,
-      "loss": 95.9571,
+      "loss": 65.9648,
       "step": 440
     },
     {
       "epoch": 0.6439359681649633,
-      "grad_norm": 148.125,
+      "grad_norm": 98.125,
       "learning_rate": 3.961352657004831e-06,
-      "loss": 95.6152,
+      "loss": 64.7386,
       "step": 445
     },
     {
       "epoch": 0.6511712037623225,
-      "grad_norm": 147.25,
+      "grad_norm": 92.6875,
       "learning_rate": 3.880837359098229e-06,
-      "loss": 95.9353,
+      "loss": 64.4612,
       "step": 450
     },
     {
       "epoch": 0.6584064393596817,
-      "grad_norm": 140.25,
+      "grad_norm": 103.5,
       "learning_rate": 3.800322061191627e-06,
-      "loss": 97.5711,
+      "loss": 64.202,
       "step": 455
     },
     {
       "epoch": 0.6656416749570407,
-      "grad_norm": 161.25,
+      "grad_norm": 95.6875,
       "learning_rate": 3.7198067632850245e-06,
-      "loss": 94.4532,
+      "loss": 63.8991,
       "step": 460
     },
     {
       "epoch": 0.6728769105543999,
-      "grad_norm": 138.125,
+      "grad_norm": 95.625,
       "learning_rate": 3.6392914653784224e-06,
-      "loss": 95.4708,
+      "loss": 63.4502,
       "step": 465
     },
     {
       "epoch": 0.6801121461517591,
-      "grad_norm": 147.25,
+      "grad_norm": 92.4375,
       "learning_rate": 3.5587761674718204e-06,
-      "loss": 93.6638,
+      "loss": 63.3427,
       "step": 470
     },
     {
       "epoch": 0.6873473817491182,
-      "grad_norm": 139.5,
+      "grad_norm": 93.9375,
       "learning_rate": 3.4782608695652175e-06,
-      "loss": 95.4091,
+      "loss": 64.2563,
       "step": 475
     },
     {
       "epoch": 0.6945826173464773,
-      "grad_norm": 132.25,
+      "grad_norm": 97.25,
       "learning_rate": 3.3977455716586155e-06,
-      "loss": 93.6919,
+      "loss": 63.9529,
       "step": 480
     },
     {
       "epoch": 0.7018178529438365,
-      "grad_norm": 144.875,
+      "grad_norm": 94.375,
       "learning_rate": 3.317230273752013e-06,
-      "loss": 96.2361,
+      "loss": 63.9575,
       "step": 485
     },
     {
       "epoch": 0.7090530885411956,
-      "grad_norm": 133.625,
+      "grad_norm": 94.75,
       "learning_rate": 3.236714975845411e-06,
-      "loss": 95.2073,
+      "loss": 64.8717,
       "step": 490
     },
     {
       "epoch": 0.7162883241385548,
-      "grad_norm": 136.625,
+      "grad_norm": 97.0625,
       "learning_rate": 3.156199677938809e-06,
-      "loss": 93.5076,
+      "loss": 63.5308,
       "step": 495
     },
     {
       "epoch": 0.7235235597359139,
-      "grad_norm": 137.5,
+      "grad_norm": 95.25,
       "learning_rate": 3.075684380032206e-06,
-      "loss": 94.2644,
+      "loss": 63.161,
       "step": 500
     },
     {
       "epoch": 0.7235235597359139,
-      "eval_loss": NaN,
-      "eval_runtime": 64.9086,
-      "eval_samples_per_second": 573.792,
-      "eval_steps_per_second": 35.866,
+      "eval_loss": 1.7313103675842285,
+      "eval_runtime": 63.4514,
+      "eval_samples_per_second": 586.969,
+      "eval_steps_per_second": 36.69,
       "step": 500
-    },
-    {
-      "epoch": 0.730758795333273,
-      "grad_norm": 154.0,
-      "learning_rate": 2.995169082125604e-06,
-      "loss": 92.6857,
-      "step": 505
-    },
-    {
-      "epoch": 0.7379940309306322,
-      "grad_norm": 148.625,
-      "learning_rate": 2.914653784219002e-06,
-      "loss": 94.1288,
-      "step": 510
-    },
-    {
-      "epoch": 0.7452292665279913,
-      "grad_norm": 141.625,
-      "learning_rate": 2.8341384863123995e-06,
-      "loss": 93.0597,
-      "step": 515
-    },
-    {
-      "epoch": 0.7524645021253504,
-      "grad_norm": 146.375,
-      "learning_rate": 2.7536231884057974e-06,
-      "loss": 94.0807,
-      "step": 520
-    },
-    {
-      "epoch": 0.7596997377227096,
-      "grad_norm": 141.875,
-      "learning_rate": 2.6731078904991954e-06,
-      "loss": 95.1949,
-      "step": 525
-    },
-    {
-      "epoch": 0.7669349733200688,
-      "grad_norm": 142.25,
-      "learning_rate": 2.5925925925925925e-06,
-      "loss": 94.323,
-      "step": 530
-    },
-    {
-      "epoch": 0.7741702089174278,
-      "grad_norm": 137.125,
-      "learning_rate": 2.5120772946859904e-06,
-      "loss": 92.8978,
-      "step": 535
-    },
-    {
-      "epoch": 0.781405444514787,
-      "grad_norm": 155.0,
-      "learning_rate": 2.4315619967793884e-06,
-      "loss": 92.656,
-      "step": 540
-    },
-    {
-      "epoch": 0.7886406801121462,
-      "grad_norm": 136.875,
-      "learning_rate": 2.351046698872786e-06,
-      "loss": 94.4373,
-      "step": 545
-    },
-    {
-      "epoch": 0.7958759157095053,
-      "grad_norm": 137.375,
-      "learning_rate": 2.270531400966184e-06,
-      "loss": 93.7375,
-      "step": 550
-    },
-    {
-      "epoch": 0.8031111513068644,
-      "grad_norm": 141.125,
-      "learning_rate": 2.1900161030595814e-06,
-      "loss": 93.2835,
-      "step": 555
-    },
-    {
-      "epoch": 0.8103463869042236,
-      "grad_norm": 136.125,
-      "learning_rate": 2.109500805152979e-06,
-      "loss": 92.7297,
-      "step": 560
-    },
-    {
-      "epoch": 0.8175816225015827,
-      "grad_norm": 132.0,
-      "learning_rate": 2.028985507246377e-06,
-      "loss": 93.2055,
-      "step": 565
-    },
-    {
-      "epoch": 0.8248168580989419,
-      "grad_norm": 142.0,
-      "learning_rate": 1.948470209339775e-06,
-      "loss": 92.6126,
-      "step": 570
-    },
-    {
-      "epoch": 0.832052093696301,
-      "grad_norm": 128.375,
-      "learning_rate": 1.8679549114331724e-06,
-      "loss": 91.9927,
-      "step": 575
-    },
-    {
-      "epoch": 0.8392873292936601,
-      "grad_norm": 142.625,
-      "learning_rate": 1.7874396135265702e-06,
-      "loss": 92.6083,
-      "step": 580
-    },
-    {
-      "epoch": 0.8465225648910193,
-      "grad_norm": 142.5,
-      "learning_rate": 1.7069243156199681e-06,
-      "loss": 93.0809,
-      "step": 585
-    },
-    {
-      "epoch": 0.8537578004883783,
-      "grad_norm": 131.625,
-      "learning_rate": 1.6264090177133656e-06,
-      "loss": 90.4466,
-      "step": 590
-    },
-    {
-      "epoch": 0.8609930360857375,
-      "grad_norm": 153.0,
-      "learning_rate": 1.5458937198067634e-06,
-      "loss": 90.9484,
-      "step": 595
-    },
-    {
-      "epoch": 0.8682282716830967,
-      "grad_norm": 141.75,
-      "learning_rate": 1.4653784219001613e-06,
-      "loss": 92.5591,
-      "step": 600
-    },
-    {
-      "epoch": 0.8754635072804559,
-      "grad_norm": 149.25,
-      "learning_rate": 1.3848631239935589e-06,
-      "loss": 93.0069,
-      "step": 605
-    },
-    {
-      "epoch": 0.8826987428778149,
-      "grad_norm": 131.625,
-      "learning_rate": 1.3043478260869566e-06,
-      "loss": 92.2291,
-      "step": 610
-    },
-    {
-      "epoch": 0.8899339784751741,
-      "grad_norm": 129.875,
-      "learning_rate": 1.2238325281803544e-06,
-      "loss": 91.3298,
-      "step": 615
-    },
-    {
-      "epoch": 0.8971692140725332,
-      "grad_norm": 139.25,
-      "learning_rate": 1.1433172302737521e-06,
-      "loss": 92.7737,
-      "step": 620
-    },
-    {
-      "epoch": 0.9044044496698924,
-      "grad_norm": 134.375,
-      "learning_rate": 1.0628019323671499e-06,
-      "loss": 93.0011,
-      "step": 625
-    },
-    {
-      "epoch": 0.9116396852672515,
-      "grad_norm": 131.625,
-      "learning_rate": 9.822866344605476e-07,
-      "loss": 92.2996,
-      "step": 630
-    },
-    {
-      "epoch": 0.9188749208646106,
-      "grad_norm": 136.75,
-      "learning_rate": 9.017713365539453e-07,
-      "loss": 92.7781,
-      "step": 635
-    },
-    {
-      "epoch": 0.9261101564619698,
-      "grad_norm": 139.75,
-      "learning_rate": 8.212560386473431e-07,
-      "loss": 91.0231,
-      "step": 640
-    },
-    {
-      "epoch": 0.933345392059329,
-      "grad_norm": 136.875,
-      "learning_rate": 7.407407407407407e-07,
-      "loss": 91.6645,
-      "step": 645
-    },
-    {
-      "epoch": 0.940580627656688,
-      "grad_norm": 144.875,
-      "learning_rate": 6.602254428341386e-07,
-      "loss": 90.7734,
-      "step": 650
-    },
-    {
-      "epoch": 0.9478158632540472,
-      "grad_norm": 142.75,
-      "learning_rate": 5.797101449275363e-07,
-      "loss": 91.3177,
-      "step": 655
-    },
-    {
-      "epoch": 0.9550510988514064,
-      "grad_norm": 148.75,
-      "learning_rate": 4.991948470209341e-07,
-      "loss": 90.913,
-      "step": 660
-    },
-    {
-      "epoch": 0.9622863344487654,
-      "grad_norm": 128.125,
-      "learning_rate": 4.1867954911433176e-07,
-      "loss": 90.6037,
-      "step": 665
-    },
-    {
-      "epoch": 0.9695215700461246,
-      "grad_norm": 132.25,
-      "learning_rate": 3.3816425120772945e-07,
-      "loss": 91.3471,
-      "step": 670
-    },
-    {
-      "epoch": 0.9767568056434838,
-      "grad_norm": 126.4375,
-      "learning_rate": 2.5764895330112725e-07,
-      "loss": 91.4777,
-      "step": 675
-    },
-    {
-      "epoch": 0.9839920412408429,
-      "grad_norm": 139.0,
-      "learning_rate": 1.7713365539452497e-07,
-      "loss": 92.0473,
-      "step": 680
-    },
-    {
-      "epoch": 0.991227276838202,
-      "grad_norm": 134.875,
-      "learning_rate": 9.661835748792271e-08,
-      "loss": 90.7339,
-      "step": 685
-    },
-    {
-      "epoch": 0.9984625124355612,
-      "grad_norm": 139.375,
-      "learning_rate": 1.6103059581320453e-08,
-      "loss": 92.0409,
-      "step": 690
     }
   ],
   "logging_steps": 5,
@@ -995,12 +729,12 @@
         "should_evaluate": false,
         "should_log": false,
         "should_save": true,
-        "should_training_stop": true
+        "should_training_stop": false
       },
       "attributes": {}
     }
   },
-  "total_flos": 2.993659372963365e+18,
+  "total_flos": 2.1661789963943936e+18,
   "train_batch_size": 4,
   "trial_name": null,
   "trial_params": null

last-checkpoint/training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c88d63b229521e393dd15e94a0dce5631f524c246b8dceda3a877ef2c9acdd7e
+oid sha256:f9ac479c0792c767bddcc93c08dda3764b8a2580cb4ea96fcc5be3a954af60ee
 size 5752