Training in progress, epoch 0, checkpoint

last-checkpoint/config.json

@@ -16,6 +16,6 @@
   "norm_eps": 1e-05,
   "swiglu_multiple_of": 256,
   "torch_dtype": "float32",
-  "transformers_version": "4.46.0",
+  "transformers_version": "4.46.3",
   "vocab_size": 37
 }

last-checkpoint/model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:26d8f5a8a6c2b2142e1a897b44d8d10c4b31cd9bfb6af9a2f209cb1c664219db
+oid sha256:d1c25afc108a9327d3cdd7c728c1d34b04467bb2920b765b6464fe6ec5d70e99
 size 2682482800

last-checkpoint/optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d5e3926040695ffd36b68b04adaa45e63e11c719919bb0b1c4e2bd27ee1e14c0
+oid sha256:aa971a7fbc6a7cfdfa3d9f7f643c82fc86bb204d25a82c4b7273e6c9b3fc0ad7
 size 5365108834

last-checkpoint/rng_state.pth

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dd261f24a3b802018daa7344b83f247d318502ad53b463783d522e7fc68f088e
-size 14244
+oid sha256:8e6a34fab0a1d10aa24e64c94ccfc5f36fca4a5f6412c70d7125c97f6c837c89
+size 15006

last-checkpoint/scheduler.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2b32f7bc8829feee7543ba545ec8f492f6a76931d31cd30ffef13d25c9c59f4e
+oid sha256:d53991ab07b79829d2e4729a2df1ec31fceb46e178383f1b9683278175822105
 size 1064

last-checkpoint/trainer_state.json

@@ -1,576 +1,91 @@
 {
   "best_metric": null,
   "best_model_checkpoint": null,
-  "epoch": 0.30688378694593094,
-  "eval_steps": 500,
-  "global_step": 4000,
+  "epoch": 0.9998851234922458,
+  "eval_steps": 50,
+  "global_step": 272,
   "is_hyper_param_search": false,
   "is_local_process_zero": true,
   "is_world_process_zero": true,
   "log_history": [
     {
-      "epoch": 0.0038360473368241363,
-      "grad_norm": 42.40585708618164,
-      "learning_rate": 9.961638790854688e-05,
-      "loss": 76.6444,
+      "epoch": 0.18380241240666284,
+      "grad_norm": 88.13125610351562,
+      "learning_rate": 8.161764705882353e-05,
+      "loss": 1.4061,
       "step": 50
     },
     {
-      "epoch": 0.0076720946736482725,
-      "grad_norm": 47.012596130371094,
-      "learning_rate": 9.923277581709376e-05,
-      "loss": 73.8558,
-      "step": 100
-    },
-    {
-      "epoch": 0.011508142010472408,
-      "grad_norm": 37.02272033691406,
-      "learning_rate": 9.884916372564063e-05,
-      "loss": 72.8816,
-      "step": 150
-    },
-    {
-      "epoch": 0.015344189347296545,
-      "grad_norm": 47.1681022644043,
-      "learning_rate": 9.846555163418752e-05,
-      "loss": 72.3309,
-      "step": 200
-    },
-    {
-      "epoch": 0.019180236684120684,
-      "grad_norm": 61.863441467285156,
-      "learning_rate": 9.80819395427344e-05,
-      "loss": 71.639,
-      "step": 250
-    },
-    {
-      "epoch": 0.023016284020944817,
-      "grad_norm": 65.1348648071289,
-      "learning_rate": 9.769832745128127e-05,
-      "loss": 71.186,
-      "step": 300
-    },
-    {
-      "epoch": 0.026852331357768953,
-      "grad_norm": 54.1230583190918,
-      "learning_rate": 9.731471535982815e-05,
-      "loss": 70.5527,
-      "step": 350
-    },
-    {
-      "epoch": 0.03068837869459309,
-      "grad_norm": 24.713777542114258,
-      "learning_rate": 9.693110326837502e-05,
-      "loss": 70.4304,
-      "step": 400
-    },
-    {
-      "epoch": 0.03452442603141723,
-      "grad_norm": 33.9341926574707,
-      "learning_rate": 9.65474911769219e-05,
-      "loss": 69.9943,
-      "step": 450
-    },
-    {
-      "epoch": 0.03836047336824137,
-      "grad_norm": 45.115230560302734,
-      "learning_rate": 9.616387908546877e-05,
-      "loss": 69.615,
-      "step": 500
-    },
-    {
-      "epoch": 0.0421965207050655,
-      "grad_norm": 32.29951858520508,
-      "learning_rate": 9.578026699401565e-05,
-      "loss": 69.3197,
-      "step": 550
-    },
-    {
-      "epoch": 0.046032568041889634,
-      "grad_norm": 47.040401458740234,
-      "learning_rate": 9.539665490256254e-05,
-      "loss": 69.4096,
-      "step": 600
-    },
-    {
-      "epoch": 0.049868615378713774,
-      "grad_norm": 15.688956260681152,
-      "learning_rate": 9.501304281110941e-05,
-      "loss": 68.8061,
-      "step": 650
-    },
-    {
-      "epoch": 0.05370466271553791,
-      "grad_norm": 47.72418212890625,
-      "learning_rate": 9.462943071965629e-05,
-      "loss": 68.8162,
-      "step": 700
-    },
-    {
-      "epoch": 0.05754071005236205,
-      "grad_norm": 60.38280487060547,
-      "learning_rate": 9.424581862820317e-05,
-      "loss": 68.3997,
-      "step": 750
-    },
-    {
-      "epoch": 0.06137675738918618,
-      "grad_norm": 18.39830207824707,
-      "learning_rate": 9.386220653675004e-05,
-      "loss": 68.264,
-      "step": 800
-    },
-    {
-      "epoch": 0.06521280472601032,
-      "grad_norm": 55.10166931152344,
-      "learning_rate": 9.347859444529692e-05,
-      "loss": 67.9374,
-      "step": 850
-    },
-    {
-      "epoch": 0.06904885206283445,
-      "grad_norm": 28.2165584564209,
-      "learning_rate": 9.309498235384379e-05,
-      "loss": 67.6585,
-      "step": 900
-    },
-    {
-      "epoch": 0.07288489939965859,
-      "grad_norm": 43.54179763793945,
-      "learning_rate": 9.271137026239067e-05,
-      "loss": 67.4189,
-      "step": 950
-    },
-    {
-      "epoch": 0.07672094673648273,
-      "grad_norm": 25.794546127319336,
-      "learning_rate": 9.232775817093756e-05,
-      "loss": 67.0964,
-      "step": 1000
-    },
-    {
-      "epoch": 0.08055699407330687,
-      "grad_norm": 24.354610443115234,
-      "learning_rate": 9.194414607948443e-05,
-      "loss": 66.8448,
-      "step": 1050
-    },
-    {
-      "epoch": 0.084393041410131,
-      "grad_norm": 32.821502685546875,
-      "learning_rate": 9.156053398803131e-05,
-      "loss": 66.7666,
-      "step": 1100
-    },
-    {
-      "epoch": 0.08822908874695513,
-      "grad_norm": 28.09437370300293,
-      "learning_rate": 9.117692189657818e-05,
-      "loss": 66.4893,
-      "step": 1150
-    },
-    {
-      "epoch": 0.09206513608377927,
-      "grad_norm": 48.216896057128906,
-      "learning_rate": 9.079330980512506e-05,
-      "loss": 66.352,
-      "step": 1200
-    },
-    {
-      "epoch": 0.09590118342060341,
-      "grad_norm": 30.080474853515625,
-      "learning_rate": 9.040969771367193e-05,
-      "loss": 66.2678,
-      "step": 1250
-    },
-    {
-      "epoch": 0.09973723075742755,
-      "grad_norm": 31.63680648803711,
-      "learning_rate": 9.002608562221881e-05,
-      "loss": 66.0015,
-      "step": 1300
-    },
-    {
-      "epoch": 0.10357327809425168,
-      "grad_norm": 64.34364318847656,
-      "learning_rate": 8.964247353076569e-05,
-      "loss": 65.6839,
-      "step": 1350
-    },
-    {
-      "epoch": 0.10740932543107581,
-      "grad_norm": 43.31911087036133,
-      "learning_rate": 8.925886143931257e-05,
-      "loss": 65.4607,
-      "step": 1400
-    },
-    {
-      "epoch": 0.11124537276789996,
-      "grad_norm": 43.40916442871094,
-      "learning_rate": 8.887524934785945e-05,
-      "loss": 65.2276,
-      "step": 1450
-    },
-    {
-      "epoch": 0.1150814201047241,
-      "grad_norm": 27.121234893798828,
-      "learning_rate": 8.849163725640633e-05,
-      "loss": 65.2446,
-      "step": 1500
-    },
-    {
-      "epoch": 0.11891746744154823,
-      "grad_norm": 14.789810180664062,
-      "learning_rate": 8.81080251649532e-05,
-      "loss": 64.8275,
-      "step": 1550
-    },
-    {
-      "epoch": 0.12275351477837236,
-      "grad_norm": 44.35984802246094,
-      "learning_rate": 8.772441307350008e-05,
-      "loss": 64.7482,
-      "step": 1600
-    },
-    {
-      "epoch": 0.1265895621151965,
-      "grad_norm": 16.479351043701172,
-      "learning_rate": 8.734080098204695e-05,
-      "loss": 64.3666,
-      "step": 1650
-    },
-    {
-      "epoch": 0.13042560945202064,
-      "grad_norm": 27.057039260864258,
-      "learning_rate": 8.695718889059383e-05,
-      "loss": 64.3815,
-      "step": 1700
-    },
-    {
-      "epoch": 0.13426165678884477,
-      "grad_norm": 47.9273681640625,
-      "learning_rate": 8.65735767991407e-05,
-      "loss": 64.2401,
-      "step": 1750
-    },
-    {
-      "epoch": 0.1380977041256689,
-      "grad_norm": 29.331714630126953,
-      "learning_rate": 8.618996470768759e-05,
-      "loss": 63.8462,
-      "step": 1800
-    },
-    {
-      "epoch": 0.14193375146249304,
-      "grad_norm": 57.99611282348633,
-      "learning_rate": 8.580635261623447e-05,
-      "loss": 63.6392,
-      "step": 1850
-    },
-    {
-      "epoch": 0.14576979879931717,
-      "grad_norm": 54.53227233886719,
-      "learning_rate": 8.542274052478134e-05,
-      "loss": 63.5283,
-      "step": 1900
-    },
-    {
-      "epoch": 0.1496058461361413,
-      "grad_norm": 30.265884399414062,
-      "learning_rate": 8.503912843332822e-05,
-      "loss": 63.378,
-      "step": 1950
-    },
-    {
-      "epoch": 0.15344189347296547,
-      "grad_norm": 19.72339630126953,
-      "learning_rate": 8.46555163418751e-05,
-      "loss": 63.123,
-      "step": 2000
-    },
-    {
-      "epoch": 0.1572779408097896,
-      "grad_norm": 31.457361221313477,
-      "learning_rate": 8.427190425042197e-05,
-      "loss": 63.1952,
-      "step": 2050
-    },
-    {
-      "epoch": 0.16111398814661373,
-      "grad_norm": 19.572940826416016,
-      "learning_rate": 8.388829215896885e-05,
-      "loss": 62.8529,
-      "step": 2100
-    },
-    {
-      "epoch": 0.16495003548343787,
-      "grad_norm": 34.743125915527344,
-      "learning_rate": 8.350468006751573e-05,
-      "loss": 62.6967,
-      "step": 2150
-    },
-    {
-      "epoch": 0.168786082820262,
-      "grad_norm": 27.31527328491211,
-      "learning_rate": 8.312106797606261e-05,
-      "loss": 62.6359,
-      "step": 2200
-    },
-    {
-      "epoch": 0.17262213015708613,
-      "grad_norm": 35.15655517578125,
-      "learning_rate": 8.273745588460949e-05,
-      "loss": 62.3291,
-      "step": 2250
-    },
-    {
-      "epoch": 0.17645817749391027,
-      "grad_norm": 38.50543212890625,
-      "learning_rate": 8.235384379315636e-05,
-      "loss": 62.4786,
-      "step": 2300
-    },
-    {
-      "epoch": 0.1802942248307344,
-      "grad_norm": 33.36336135864258,
-      "learning_rate": 8.197023170170324e-05,
-      "loss": 62.0836,
-      "step": 2350
-    },
-    {
-      "epoch": 0.18413027216755853,
-      "grad_norm": 50.12577438354492,
-      "learning_rate": 8.158661961025011e-05,
-      "loss": 61.8858,
-      "step": 2400
-    },
-    {
-      "epoch": 0.1879663195043827,
-      "grad_norm": 62.16424560546875,
-      "learning_rate": 8.120300751879699e-05,
-      "loss": 61.8654,
-      "step": 2450
-    },
-    {
-      "epoch": 0.19180236684120683,
-      "grad_norm": 47.132232666015625,
-      "learning_rate": 8.081939542734386e-05,
-      "loss": 61.6991,
-      "step": 2500
-    },
-    {
-      "epoch": 0.19563841417803096,
-      "grad_norm": 35.88410186767578,
-      "learning_rate": 8.043578333589075e-05,
-      "loss": 61.6338,
-      "step": 2550
-    },
-    {
-      "epoch": 0.1994744615148551,
-      "grad_norm": 35.40159225463867,
-      "learning_rate": 8.005217124443763e-05,
-      "loss": 61.4948,
-      "step": 2600
-    },
-    {
-      "epoch": 0.20331050885167923,
-      "grad_norm": 26.17939567565918,
-      "learning_rate": 7.96685591529845e-05,
-      "loss": 61.1859,
-      "step": 2650
-    },
-    {
-      "epoch": 0.20714655618850336,
-      "grad_norm": 40.978424072265625,
-      "learning_rate": 7.928494706153138e-05,
-      "loss": 61.0628,
-      "step": 2700
-    },
-    {
-      "epoch": 0.2109826035253275,
-      "grad_norm": 48.16961669921875,
-      "learning_rate": 7.890133497007825e-05,
-      "loss": 61.0898,
-      "step": 2750
-    },
-    {
-      "epoch": 0.21481865086215163,
-      "grad_norm": 38.556034088134766,
-      "learning_rate": 7.851772287862513e-05,
-      "loss": 60.5673,
-      "step": 2800
-    },
-    {
-      "epoch": 0.2186546981989758,
-      "grad_norm": 47.85882568359375,
-      "learning_rate": 7.8134110787172e-05,
-      "loss": 60.3441,
-      "step": 2850
-    },
-    {
-      "epoch": 0.22249074553579992,
-      "grad_norm": 37.2432975769043,
-      "learning_rate": 7.775049869571888e-05,
-      "loss": 60.3992,
-      "step": 2900
-    },
-    {
-      "epoch": 0.22632679287262406,
-      "grad_norm": 22.40682029724121,
-      "learning_rate": 7.736688660426577e-05,
-      "loss": 60.2252,
-      "step": 2950
-    },
-    {
-      "epoch": 0.2301628402094482,
-      "grad_norm": 63.746803283691406,
-      "learning_rate": 7.698327451281265e-05,
-      "loss": 60.2993,
-      "step": 3000
-    },
-    {
-      "epoch": 0.23399888754627232,
-      "grad_norm": 27.40711212158203,
-      "learning_rate": 7.659966242135954e-05,
-      "loss": 60.103,
-      "step": 3050
-    },
-    {
-      "epoch": 0.23783493488309645,
-      "grad_norm": 26.063871383666992,
-      "learning_rate": 7.621605032990641e-05,
-      "loss": 60.1669,
-      "step": 3100
-    },
-    {
-      "epoch": 0.2416709822199206,
-      "grad_norm": 55.009674072265625,
-      "learning_rate": 7.583243823845329e-05,
-      "loss": 59.8542,
-      "step": 3150
-    },
-    {
-      "epoch": 0.24550702955674472,
-      "grad_norm": 54.10274124145508,
-      "learning_rate": 7.544882614700016e-05,
-      "loss": 59.586,
-      "step": 3200
-    },
-    {
-      "epoch": 0.24934307689356885,
-      "grad_norm": 59.13345718383789,
-      "learning_rate": 7.506521405554704e-05,
-      "loss": 59.2741,
-      "step": 3250
-    },
-    {
-      "epoch": 0.253179124230393,
-      "grad_norm": 26.184877395629883,
-      "learning_rate": 7.468160196409391e-05,
-      "loss": 59.3165,
-      "step": 3300
-    },
-    {
-      "epoch": 0.25701517156721715,
-      "grad_norm": 29.710121154785156,
-      "learning_rate": 7.429798987264079e-05,
-      "loss": 59.6604,
-      "step": 3350
-    },
-    {
-      "epoch": 0.2608512189040413,
-      "grad_norm": 67.96011352539062,
-      "learning_rate": 7.391437778118768e-05,
-      "loss": 59.4448,
-      "step": 3400
-    },
-    {
-      "epoch": 0.2646872662408654,
-      "grad_norm": 77.67021942138672,
-      "learning_rate": 7.353076568973455e-05,
-      "loss": 59.168,
-      "step": 3450
-    },
-    {
-      "epoch": 0.26852331357768955,
-      "grad_norm": 14.086892127990723,
-      "learning_rate": 7.314715359828143e-05,
-      "loss": 58.981,
-      "step": 3500
-    },
-    {
-      "epoch": 0.2723593609145137,
-      "grad_norm": 51.17266082763672,
-      "learning_rate": 7.27635415068283e-05,
-      "loss": 58.9879,
-      "step": 3550
-    },
-    {
-      "epoch": 0.2761954082513378,
-      "grad_norm": 48.200164794921875,
-      "learning_rate": 7.237992941537518e-05,
-      "loss": 58.5352,
-      "step": 3600
+      "epoch": 0.18380241240666284,
+      "eval_loss": 1.3573001623153687,
+      "eval_runtime": 115.6752,
+      "eval_samples_per_second": 41.72,
+      "eval_steps_per_second": 20.86,
+      "step": 50
     },
     {
-      "epoch": 0.28003145558816195,
-      "grad_norm": 37.883766174316406,
-      "learning_rate": 7.199631732392206e-05,
-      "loss": 58.6675,
-      "step": 3650
+      "epoch": 0.3676048248133257,
+      "grad_norm": 81.8663330078125,
+      "learning_rate": 6.323529411764705e-05,
+      "loss": 1.3151,
+      "step": 100
     },
     {
-      "epoch": 0.2838675029249861,
-      "grad_norm": 45.00859832763672,
-      "learning_rate": 7.161270523246893e-05,
-      "loss": 58.7792,
-      "step": 3700
+      "epoch": 0.3676048248133257,
+      "eval_loss": 1.3327406644821167,
+      "eval_runtime": 115.7469,
+      "eval_samples_per_second": 41.694,
+      "eval_steps_per_second": 20.847,
+      "step": 100
     },
     {
-      "epoch": 0.2877035502618102,
-      "grad_norm": 26.977285385131836,
-      "learning_rate": 7.12290931410158e-05,
-      "loss": 58.4862,
-      "step": 3750
+      "epoch": 0.5514072372199885,
+      "grad_norm": 75.3175048828125,
+      "learning_rate": 4.485294117647059e-05,
+      "loss": 1.2816,
+      "step": 150
     },
     {
-      "epoch": 0.29153959759863435,
-      "grad_norm": 38.32195281982422,
-      "learning_rate": 7.08454810495627e-05,
-      "loss": 58.3661,
-      "step": 3800
+      "epoch": 0.5514072372199885,
+      "eval_loss": 1.312986969947815,
+      "eval_runtime": 115.6659,
+      "eval_samples_per_second": 41.724,
+      "eval_steps_per_second": 20.862,
+      "step": 150
     },
     {
-      "epoch": 0.2953756449354585,
-      "grad_norm": 63.87961196899414,
-      "learning_rate": 7.046186895810957e-05,
-      "loss": 58.25,
-      "step": 3850
+      "epoch": 0.7352096496266514,
+      "grad_norm": 71.32404327392578,
+      "learning_rate": 2.647058823529412e-05,
+      "loss": 1.2492,
+      "step": 200
     },
     {
-      "epoch": 0.2992116922722826,
-      "grad_norm": 35.58587646484375,
-      "learning_rate": 7.007825686665645e-05,
-      "loss": 58.3304,
-      "step": 3900
+      "epoch": 0.7352096496266514,
+      "eval_loss": 1.3017219305038452,
+      "eval_runtime": 115.6996,
+      "eval_samples_per_second": 41.711,
+      "eval_steps_per_second": 20.856,
+      "step": 200
     },
     {
-      "epoch": 0.3030477396091068,
-      "grad_norm": 60.112632751464844,
-      "learning_rate": 6.969464477520332e-05,
-      "loss": 58.1747,
-      "step": 3950
+      "epoch": 0.9190120620333142,
+      "grad_norm": 63.50423812866211,
+      "learning_rate": 8.088235294117648e-06,
+      "loss": 1.2329,
+      "step": 250
     },
     {
-      "epoch": 0.30688378694593094,
-      "grad_norm": 41.148353576660156,
-      "learning_rate": 6.93110326837502e-05,
-      "loss": 57.8954,
-      "step": 4000
+      "epoch": 0.9190120620333142,
+      "eval_loss": 1.2930370569229126,
+      "eval_runtime": 115.8112,
+      "eval_samples_per_second": 41.671,
+      "eval_steps_per_second": 20.836,
+      "step": 250
     }
   ],
   "logging_steps": 50,
-  "max_steps": 13034,
+  "max_steps": 272,
   "num_input_tokens_seen": 0,
   "num_train_epochs": 1,
   "save_steps": 500,
@@ -581,12 +96,12 @@
         "should_evaluate": false,
         "should_log": false,
         "should_save": true,
-        "should_training_stop": false
+        "should_training_stop": true
       },
       "attributes": {}
     }
   },
-  "total_flos": 6.855218432384696e+18,
+  "total_flos": 7.013472912644506e+16,
   "train_batch_size": 2,
   "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:eb6051eee29c31ca0092abe3bc9f1d55068e299690664ab4b9484f44a7a3cebf
-size 5240
+oid sha256:75d9aec9ea7211c3e87a441ee9e1a32bad36657c98d667f4b0503d440eacb58e
+size 5304