Training in progress, step 210, checkpoint

checkpoint-210/global_step210/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5439969fc361ff852c6adcadaa6f8422f75b1c4d9577abbdb47742237ea284d2
+oid sha256:a964100ea5fca29615c410816916dff4331b771b98631be51c9dfad420d8dc0e
 size 3402782599

checkpoint-210/global_step210/bf16_zero_pp_rank_1_mp_rank_00_optim_states.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ca94ebbc2b41927e345a475b57b07bbebde2352f49eb2e01389abc5481c17653
+oid sha256:cb1fbe340c6774fdd4be1723b46252a74d12ec659665f195b5e4a4cd2826776b
 size 3402782599

checkpoint-210/global_step210/bf16_zero_pp_rank_2_mp_rank_00_optim_states.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2d7eff2b58924e1815eb1d7264179128fcac4a7bf1d726837d9b98df23e7cf22
+oid sha256:a49654a6c631951b157efe0af80122e596176a6921398b0c50d0271805a21a9d
 size 3402782599

checkpoint-210/global_step210/bf16_zero_pp_rank_3_mp_rank_00_optim_states.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e4a1a49834444c03ae247951c1ab4b38824504a89a3570b7e37c5f7c89a2f5b6
+oid sha256:520d0e55e451cb020000f91f118b036577130ead5d17172e6f147114e111bb41
 size 3402782599

checkpoint-210/global_step210/bf16_zero_pp_rank_4_mp_rank_00_optim_states.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e501f13e779784e88bbeb31ebd7e9899c55fdd1372b2aa1398f76ab461785d48
+oid sha256:dfda00dacdaa5a7b5bed6d5fa097f4c0e66887e15984e97860ce82d6c9ed4d84
 size 3402782599

checkpoint-210/global_step210/bf16_zero_pp_rank_5_mp_rank_00_optim_states.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:14b55d23000f932481ca1ee03f5242821e2ec13c76e75f13e763169e7f807b65
+oid sha256:1894b2e2b0fa6ebad413225d533b0efe9d116f5b43c1bb2aaf23578e8891dfba
 size 3402782599

checkpoint-210/global_step210/bf16_zero_pp_rank_6_mp_rank_00_optim_states.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:73889a07bc11090a095476103ecf7b65f7ca60e712fdf3649960b4dc7d48ec94
+oid sha256:708026c96af37c8357c64d2cd5963f91211d30aa13a6dd11b6e7670bca29475e
 size 3402782599

checkpoint-210/global_step210/bf16_zero_pp_rank_7_mp_rank_00_optim_states.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dd35aefbd060783af9f0889e369c4cec1c55dc04e3d1bd0a1bac2749a6fd60aa
+oid sha256:cc9e6b9c7381f8a59b46a41e7e495aa8c9a157942f598dba67c955720d2d5f29
 size 3402782599

checkpoint-210/global_step210/zero_pp_rank_0_mp_rank_00_model_states.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:aa9cdf4f231f9565da95c795021c8bab2c84d38d67e870b70ca9ac3b5db3e406
+oid sha256:d267b5a5adc3b249e9b75831e76e86a47a12f768250b81d1e9298624098bc0ea
 size 150245

checkpoint-210/global_step210/zero_pp_rank_1_mp_rank_00_model_states.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:333aab5149bc5c5383397a9bda8848da74c34b5e8a6bf3ee6fc96f8c136f9d19
+oid sha256:c7a264a507c6632110323bf2dd4ff9cffcbd94327358c5e2d5f7875ab99ef672
 size 150245

checkpoint-210/global_step210/zero_pp_rank_2_mp_rank_00_model_states.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:844b838dfe4fc34d298a7c7f9fe9428cc5029cdcda33e5ffb033f0a1541d8f55
+oid sha256:7d0a2d8ea8538673663ea47f306617eaf01f6f590eaba5692481a2ac671d3e7e
 size 150245

checkpoint-210/global_step210/zero_pp_rank_3_mp_rank_00_model_states.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:86fc60090fd5cde3b58413c7bd09dcd59f654df46835f63114ba109d86f7ce5e
+oid sha256:b3865c1f5c7c8a93a5d8d152259de016603bd0a15f0037b70530f92e42069893
 size 150245

checkpoint-210/global_step210/zero_pp_rank_4_mp_rank_00_model_states.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:47345124e9554305b5412464ecde8f73bad2da742ffe28ebdc9ed3532902dca6
+oid sha256:f891eda0cc8c1e1ee4ee0eb197830188cc3eb4f16eba81352fc6fe72e906ddc6
 size 150245

checkpoint-210/global_step210/zero_pp_rank_5_mp_rank_00_model_states.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c2425a9e81eac9199883a06de96df5662dd47ef3f0030b90d61ce1a67e7fb473
+oid sha256:3e698ee6b9a1e09854f52fa65812a2669789b266475be3cc7a9699d6800b8fb0
 size 150245

checkpoint-210/global_step210/zero_pp_rank_6_mp_rank_00_model_states.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:55cd4356c201aed2ce55850198315a379593fcdb6f0aa90f5bb65a7754b86804
+oid sha256:992728c31e93e4e06f4e0cb6901e95ce4e7835d5cf950eb080b66ef4308865e4
 size 150245

checkpoint-210/global_step210/zero_pp_rank_7_mp_rank_00_model_states.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8771a82bdc6003651dd20dea06121ac1ba627c645120f8d31bc084a4ebbe61e8
+oid sha256:92ad76ac01a9040705a38518fa43cbe8b381235bc1f21f8c8c9e1fae661b7263
 size 150245

checkpoint-210/model-00001-of-00002.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:84c096c8d3a9b13352aff57b009e6ba81ddfbf6bd74ed35afb503bc99ecf8420
+oid sha256:d5ae6ffc7d1099523179e5c3458e9b03d17cd4ce6e8653aada4b018133540637
 size 4978354640

checkpoint-210/model-00002-of-00002.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a4d7f35c184d870bbd8122a0ef2eed4c1516bf3d0a51f7d6297d6ca4050eda16
+oid sha256:8d3d84c39fb4833d412123e6aa2b89388aa9e0f00b22ba52deedf5b586e8a41a
 size 4047172128

checkpoint-210/trainer_state.json

@@ -11,1496 +11,1496 @@
   "log_history": [
     {
       "epoch": 0.0023781212841854932,
-      "grad_norm": 32.382706174573094,
+      "grad_norm": 32.74397118558861,
       "learning_rate": 5.000000000000001e-07,
-      "loss": 2.9481,
+      "loss": 2.9478,
       "step": 1
     },
     {
       "epoch": 0.0023781212841854932,
-      "eval_loss": 3.3739373683929443,
-      "eval_runtime": 152.3599,
-      "eval_samples_per_second": 4.194,
-      "eval_steps_per_second": 0.525,
+      "eval_loss": 3.373392343521118,
+      "eval_runtime": 78.9756,
+      "eval_samples_per_second": 8.091,
+      "eval_steps_per_second": 1.013,
       "step": 1
     },
     {
       "epoch": 0.0047562425683709865,
-      "grad_norm": 36.523706488132554,
+      "grad_norm": 37.080911592721954,
       "learning_rate": 1.0000000000000002e-06,
-      "loss": 2.8267,
+      "loss": 2.8264,
       "step": 2
     },
     {
       "epoch": 0.007134363852556481,
-      "grad_norm": 30.823955989385283,
+      "grad_norm": 31.107267997702266,
       "learning_rate": 1.5e-06,
       "loss": 2.9936,
       "step": 3
     },
     {
       "epoch": 0.009512485136741973,
-      "grad_norm": 27.113313789497223,
+      "grad_norm": 27.574905774161167,
       "learning_rate": 2.0000000000000003e-06,
-      "loss": 2.8945,
+      "loss": 2.8944,
       "step": 4
     },
     {
       "epoch": 0.011890606420927468,
-      "grad_norm": 13.505949871248614,
+      "grad_norm": 13.643957484299273,
       "learning_rate": 2.5e-06,
-      "loss": 2.7689,
+      "loss": 2.7687,
       "step": 5
     },
     {
       "epoch": 0.014268727705112961,
-      "grad_norm": 15.0338641814596,
+      "grad_norm": 15.077028980653411,
       "learning_rate": 3e-06,
-      "loss": 2.6611,
+      "loss": 2.6623,
       "step": 6
     },
     {
       "epoch": 0.016646848989298454,
-      "grad_norm": 14.319329036066911,
+      "grad_norm": 14.569557474559408,
       "learning_rate": 3.5e-06,
-      "loss": 2.8985,
+      "loss": 2.9007,
       "step": 7
     },
     {
       "epoch": 0.019024970273483946,
-      "grad_norm": 14.23176344511718,
+      "grad_norm": 13.894984550517007,
       "learning_rate": 4.000000000000001e-06,
-      "loss": 2.834,
+      "loss": 2.8359,
       "step": 8
     },
     {
       "epoch": 0.02140309155766944,
-      "grad_norm": 13.977227656063016,
+      "grad_norm": 13.546442233583257,
       "learning_rate": 4.5e-06,
-      "loss": 2.7566,
+      "loss": 2.7583,
       "step": 9
     },
     {
       "epoch": 0.023781212841854936,
-      "grad_norm": 11.952136262918579,
+      "grad_norm": 11.166476039936938,
       "learning_rate": 5e-06,
-      "loss": 2.5198,
+      "loss": 2.518,
       "step": 10
     },
     {
       "epoch": 0.026159334126040427,
-      "grad_norm": 8.093250140015975,
+      "grad_norm": 7.787448179397784,
       "learning_rate": 5.500000000000001e-06,
-      "loss": 2.6486,
+      "loss": 2.6494,
       "step": 11
     },
     {
       "epoch": 0.028537455410225922,
-      "grad_norm": 4.914080620863233,
+      "grad_norm": 4.72349822440695,
       "learning_rate": 6e-06,
-      "loss": 2.4014,
+      "loss": 2.4022,
       "step": 12
     },
     {
       "epoch": 0.030915576694411414,
-      "grad_norm": 4.175963213321,
+      "grad_norm": 4.100722460414476,
       "learning_rate": 6.5000000000000004e-06,
-      "loss": 2.3925,
+      "loss": 2.3933,
       "step": 13
     },
     {
       "epoch": 0.03329369797859691,
-      "grad_norm": 4.098569808536842,
+      "grad_norm": 4.193151112965372,
       "learning_rate": 7e-06,
-      "loss": 2.5433,
+      "loss": 2.5468,
       "step": 14
     },
     {
       "epoch": 0.0356718192627824,
-      "grad_norm": 5.46644902034287,
+      "grad_norm": 5.502246954578136,
       "learning_rate": 7.500000000000001e-06,
-      "loss": 2.5083,
+      "loss": 2.5126,
       "step": 15
     },
     {
       "epoch": 0.03804994054696789,
-      "grad_norm": 5.787961659158823,
+      "grad_norm": 5.716937946349337,
       "learning_rate": 8.000000000000001e-06,
-      "loss": 2.6687,
+      "loss": 2.6761,
       "step": 16
     },
     {
       "epoch": 0.04042806183115339,
-      "grad_norm": 3.034638828158752,
+      "grad_norm": 3.008867017303434,
       "learning_rate": 8.5e-06,
-      "loss": 2.4193,
+      "loss": 2.4264,
       "step": 17
     },
     {
       "epoch": 0.04280618311533888,
-      "grad_norm": 3.6371355055713352,
+      "grad_norm": 3.6016120293217178,
       "learning_rate": 9e-06,
-      "loss": 2.3729,
+      "loss": 2.3836,
       "step": 18
     },
     {
       "epoch": 0.04518430439952437,
-      "grad_norm": 2.1215238627086714,
+      "grad_norm": 2.1431386543975908,
       "learning_rate": 9.5e-06,
-      "loss": 2.4799,
+      "loss": 2.4879,
       "step": 19
     },
     {
       "epoch": 0.04756242568370987,
-      "grad_norm": 3.987132196757572,
+      "grad_norm": 3.838539096237921,
       "learning_rate": 1e-05,
-      "loss": 2.3981,
+      "loss": 2.4116,
       "step": 20
     },
     {
       "epoch": 0.04994054696789536,
-      "grad_norm": 1.8643899752237216,
+      "grad_norm": 1.9119200890173822,
       "learning_rate": 1.0500000000000001e-05,
-      "loss": 2.6625,
+      "loss": 2.6716,
       "step": 21
     },
     {
       "epoch": 0.052318668252080855,
-      "grad_norm": 1.9762840454371524,
+      "grad_norm": 2.0853737807318904,
       "learning_rate": 1.1000000000000001e-05,
-      "loss": 2.4424,
+      "loss": 2.4523,
       "step": 22
     },
     {
       "epoch": 0.054696789536266346,
-      "grad_norm": 1.6430481169991258,
+      "grad_norm": 1.6109539551135528,
       "learning_rate": 1.15e-05,
-      "loss": 2.3021,
+      "loss": 2.3086,
       "step": 23
     },
     {
       "epoch": 0.057074910820451845,
-      "grad_norm": 2.539281280765958,
+      "grad_norm": 2.575933824126331,
       "learning_rate": 1.2e-05,
-      "loss": 2.5471,
+      "loss": 2.5576,
       "step": 24
     },
     {
       "epoch": 0.059453032104637336,
-      "grad_norm": 1.457528590578011,
+      "grad_norm": 1.4827660467902501,
       "learning_rate": 1.25e-05,
-      "loss": 2.2526,
+      "loss": 2.264,
       "step": 25
     },
     {
       "epoch": 0.06183115338882283,
-      "grad_norm": 2.1155531281068884,
+      "grad_norm": 2.1022593224903128,
       "learning_rate": 1.3000000000000001e-05,
-      "loss": 2.8667,
+      "loss": 2.8781,
       "step": 26
     },
     {
       "epoch": 0.06420927467300833,
-      "grad_norm": 1.73342938568033,
+      "grad_norm": 2.756765764204405,
       "learning_rate": 1.3500000000000001e-05,
-      "loss": 2.4915,
+      "loss": 2.5071,
       "step": 27
     },
     {
       "epoch": 0.06658739595719382,
-      "grad_norm": 1.7317610327219517,
+      "grad_norm": 1.6249455774563617,
       "learning_rate": 1.4e-05,
-      "loss": 2.498,
+      "loss": 2.5085,
       "step": 28
     },
     {
       "epoch": 0.06896551724137931,
-      "grad_norm": 1.8363122567034704,
+      "grad_norm": 1.8653671914008603,
       "learning_rate": 1.45e-05,
-      "loss": 2.4243,
+      "loss": 2.4368,
       "step": 29
     },
     {
       "epoch": 0.0713436385255648,
-      "grad_norm": 1.6309580804999364,
+      "grad_norm": 1.867632087572305,
       "learning_rate": 1.5000000000000002e-05,
-      "loss": 2.1101,
+      "loss": 2.1204,
       "step": 30
     },
     {
       "epoch": 0.07372175980975029,
-      "grad_norm": 1.334654998985815,
+      "grad_norm": 1.2367329590346625,
       "learning_rate": 1.55e-05,
-      "loss": 2.1046,
+      "loss": 2.1118,
       "step": 31
     },
     {
       "epoch": 0.07609988109393578,
-      "grad_norm": 1.2376006409087055,
+      "grad_norm": 1.5969207000997636,
       "learning_rate": 1.6000000000000003e-05,
-      "loss": 2.4625,
+      "loss": 2.4747,
       "step": 32
     },
     {
       "epoch": 0.07847800237812129,
-      "grad_norm": 1.8515124675338408,
+      "grad_norm": 1.5075133034220278,
       "learning_rate": 1.65e-05,
-      "loss": 2.3255,
+      "loss": 2.3349,
       "step": 33
     },
     {
       "epoch": 0.08085612366230678,
-      "grad_norm": 2.4630792631286575,
+      "grad_norm": 1.4803692479229955,
       "learning_rate": 1.7e-05,
-      "loss": 2.3863,
+      "loss": 2.3915,
       "step": 34
     },
     {
       "epoch": 0.08323424494649227,
-      "grad_norm": 1.4437945302508726,
+      "grad_norm": 2.692949771759104,
       "learning_rate": 1.7500000000000002e-05,
-      "loss": 2.6266,
+      "loss": 2.6397,
       "step": 35
     },
     {
       "epoch": 0.08561236623067776,
-      "grad_norm": 1.6987676732458694,
+      "grad_norm": 1.3636512770329847,
       "learning_rate": 1.8e-05,
-      "loss": 2.2871,
+      "loss": 2.2944,
       "step": 36
     },
     {
       "epoch": 0.08799048751486326,
-      "grad_norm": 1.8155955386416776,
+      "grad_norm": 1.8310086696195464,
       "learning_rate": 1.8500000000000002e-05,
-      "loss": 2.2534,
+      "loss": 2.2614,
       "step": 37
     },
     {
       "epoch": 0.09036860879904875,
-      "grad_norm": 1.1727394458786922,
+      "grad_norm": 1.7765027708264853,
       "learning_rate": 1.9e-05,
-      "loss": 2.3458,
+      "loss": 2.3579,
       "step": 38
     },
     {
       "epoch": 0.09274673008323424,
-      "grad_norm": 1.4758138327709573,
+      "grad_norm": 1.4484769960901491,
       "learning_rate": 1.95e-05,
-      "loss": 2.4411,
+      "loss": 2.4548,
       "step": 39
     },
     {
       "epoch": 0.09512485136741974,
-      "grad_norm": 2.683368502920404,
+      "grad_norm": 3.1520205275209414,
       "learning_rate": 2e-05,
-      "loss": 2.5046,
+      "loss": 2.5208,
       "step": 40
     },
     {
       "epoch": 0.09750297265160524,
-      "grad_norm": 1.6044285552596205,
+      "grad_norm": 1.5897739849482102,
       "learning_rate": 1.9999981652287733e-05,
-      "loss": 2.413,
+      "loss": 2.4216,
       "step": 41
     },
     {
       "epoch": 0.09988109393579073,
-      "grad_norm": 1.2802177698200177,
+      "grad_norm": 2.4520591326987495,
       "learning_rate": 1.999992660921826e-05,
-      "loss": 2.2159,
+      "loss": 2.2326,
       "step": 42
     },
     {
       "epoch": 0.10225921521997622,
-      "grad_norm": 1.8164033444979393,
+      "grad_norm": 1.4817926319377914,
       "learning_rate": 1.999983487099356e-05,
-      "loss": 2.5287,
+      "loss": 2.541,
       "step": 43
     },
     {
       "epoch": 0.10463733650416171,
-      "grad_norm": 1.8420609096797744,
+      "grad_norm": 2.4022755616863956,
       "learning_rate": 1.999970643795027e-05,
-      "loss": 2.3484,
+      "loss": 2.3645,
       "step": 44
     },
     {
       "epoch": 0.1070154577883472,
-      "grad_norm": 1.4183887604989038,
+      "grad_norm": 2.476073007712477,
       "learning_rate": 1.9999541310559686e-05,
-      "loss": 2.4876,
+      "loss": 2.5051,
       "step": 45
     },
     {
       "epoch": 0.10939357907253269,
-      "grad_norm": 1.419110897512839,
+      "grad_norm": 3.428213096316913,
       "learning_rate": 1.9999339489427746e-05,
-      "loss": 2.3251,
+      "loss": 2.3605,
       "step": 46
     },
     {
       "epoch": 0.1117717003567182,
-      "grad_norm": 1.6548987738436802,
+      "grad_norm": 2.464783346708793,
       "learning_rate": 1.9999100975295046e-05,
-      "loss": 2.358,
+      "loss": 2.3785,
       "step": 47
     },
     {
       "epoch": 0.11414982164090369,
-      "grad_norm": 1.7822053646902,
+      "grad_norm": 2.0686444585541754,
       "learning_rate": 1.999882576903682e-05,
-      "loss": 2.5853,
+      "loss": 2.6035,
       "step": 48
     },
     {
       "epoch": 0.11652794292508918,
-      "grad_norm": 1.1885208188962406,
+      "grad_norm": 2.30832669621963,
       "learning_rate": 1.9998513871662945e-05,
-      "loss": 2.3794,
+      "loss": 2.3982,
       "step": 49
     },
     {
       "epoch": 0.11890606420927467,
-      "grad_norm": 1.7227839639402496,
+      "grad_norm": 1.7443884357427357,
       "learning_rate": 1.9998165284317944e-05,
-      "loss": 2.1933,
+      "loss": 2.2344,
       "step": 50
     },
     {
       "epoch": 0.12128418549346016,
-      "grad_norm": 1.3654158039316031,
+      "grad_norm": 1.862498116501275,
       "learning_rate": 1.999778000828098e-05,
-      "loss": 2.0518,
+      "loss": 2.0631,
       "step": 51
     },
     {
       "epoch": 0.12366230677764566,
-      "grad_norm": 1.3281085763249942,
+      "grad_norm": 1.3054707043181313,
       "learning_rate": 1.9997358044965833e-05,
-      "loss": 2.4352,
+      "loss": 2.4434,
       "step": 52
     },
     {
       "epoch": 0.12604042806183116,
-      "grad_norm": 1.1185472935459022,
+      "grad_norm": 2.6645481874919583,
       "learning_rate": 1.9996899395920915e-05,
-      "loss": 2.4415,
+      "loss": 2.4647,
       "step": 53
     },
     {
       "epoch": 0.12841854934601665,
-      "grad_norm": 1.343325378951958,
+      "grad_norm": 1.6884769054479885,
       "learning_rate": 1.999640406282926e-05,
-      "loss": 2.2661,
+      "loss": 2.2864,
       "step": 54
     },
     {
       "epoch": 0.13079667063020214,
-      "grad_norm": 1.371349840462257,
+      "grad_norm": 2.281038989939936,
       "learning_rate": 1.9995872047508516e-05,
-      "loss": 2.2813,
+      "loss": 2.2999,
       "step": 55
     },
     {
       "epoch": 0.13317479191438764,
-      "grad_norm": 1.2560078627168356,
+      "grad_norm": 1.8457967068582515,
       "learning_rate": 1.9995303351910934e-05,
-      "loss": 2.494,
+      "loss": 2.5117,
       "step": 56
     },
     {
       "epoch": 0.13555291319857313,
-      "grad_norm": 1.3801141775649166,
+      "grad_norm": 1.593901860141829,
       "learning_rate": 1.9994697978123363e-05,
-      "loss": 2.1525,
+      "loss": 2.1696,
       "step": 57
     },
     {
       "epoch": 0.13793103448275862,
-      "grad_norm": 2.8895107871445167,
+      "grad_norm": 2.168613470386471,
       "learning_rate": 1.9994055928367256e-05,
-      "loss": 2.6727,
+      "loss": 2.6852,
       "step": 58
     },
     {
       "epoch": 0.1403091557669441,
-      "grad_norm": 1.231028014153545,
+      "grad_norm": 1.5398971278043392,
       "learning_rate": 1.999337720499863e-05,
-      "loss": 2.3854,
+      "loss": 2.396,
       "step": 59
     },
     {
       "epoch": 0.1426872770511296,
-      "grad_norm": 1.5588938179669447,
+      "grad_norm": 1.7618365641290346,
       "learning_rate": 1.99926618105081e-05,
-      "loss": 2.033,
+      "loss": 2.047,
       "step": 60
     },
     {
       "epoch": 0.1450653983353151,
-      "grad_norm": 1.1690094119529322,
+      "grad_norm": 1.3935889380673343,
       "learning_rate": 1.9991909747520835e-05,
-      "loss": 2.1,
+      "loss": 2.1071,
       "step": 61
     },
     {
       "epoch": 0.14744351961950058,
-      "grad_norm": 1.5407871708756533,
+      "grad_norm": 1.3063537354837544,
       "learning_rate": 1.999112101879656e-05,
-      "loss": 2.3949,
+      "loss": 2.3992,
       "step": 62
     },
     {
       "epoch": 0.14982164090368608,
-      "grad_norm": 1.0249789290696483,
+      "grad_norm": 1.025531260937785,
       "learning_rate": 1.9990295627229544e-05,
-      "loss": 2.463,
+      "loss": 2.4764,
       "step": 63
     },
     {
       "epoch": 0.15219976218787157,
-      "grad_norm": 1.5221618005694364,
+      "grad_norm": 1.0662434476421614,
       "learning_rate": 1.99894335758486e-05,
-      "loss": 2.0575,
+      "loss": 2.0608,
       "step": 64
     },
     {
       "epoch": 0.1545778834720571,
-      "grad_norm": 1.0758079202744963,
+      "grad_norm": 1.335312659171346,
       "learning_rate": 1.9988534867817065e-05,
-      "loss": 2.0227,
+      "loss": 2.0345,
       "step": 65
     },
     {
       "epoch": 0.15695600475624258,
-      "grad_norm": 1.3430324490071632,
+      "grad_norm": 1.7866360477276542,
       "learning_rate": 1.9987599506432785e-05,
-      "loss": 2.467,
+      "loss": 2.4781,
       "step": 66
     },
     {
       "epoch": 0.15933412604042807,
-      "grad_norm": 1.0714364507553935,
+      "grad_norm": 1.3661388854405736,
       "learning_rate": 1.9986627495128105e-05,
-      "loss": 2.3812,
+      "loss": 2.3895,
       "step": 67
     },
     {
       "epoch": 0.16171224732461356,
-      "grad_norm": 1.296279438065536,
+      "grad_norm": 1.69413799763372,
       "learning_rate": 1.9985618837469864e-05,
-      "loss": 2.4045,
+      "loss": 2.415,
       "step": 68
     },
     {
       "epoch": 0.16409036860879905,
-      "grad_norm": 1.1423020273292732,
+      "grad_norm": 1.6458921632697698,
       "learning_rate": 1.998457353715938e-05,
-      "loss": 2.386,
+      "loss": 2.4017,
       "step": 69
     },
     {
       "epoch": 0.16646848989298454,
-      "grad_norm": 1.5165435155836928,
+      "grad_norm": 1.705150979387567,
       "learning_rate": 1.998349159803241e-05,
-      "loss": 2.3063,
+      "loss": 2.317,
       "step": 70
     },
     {
       "epoch": 0.16884661117717004,
-      "grad_norm": 2.758568140385349,
+      "grad_norm": 1.4786622132550975,
       "learning_rate": 1.9982373024059195e-05,
-      "loss": 2.4007,
+      "loss": 2.4046,
       "step": 71
     },
     {
       "epoch": 0.17122473246135553,
-      "grad_norm": 1.268753870102164,
+      "grad_norm": 1.1257378401253821,
       "learning_rate": 1.998121781934438e-05,
-      "loss": 2.1938,
+      "loss": 2.196,
       "step": 72
     },
     {
       "epoch": 0.17360285374554102,
-      "grad_norm": 1.1116353498200209,
+      "grad_norm": 1.2450243917414825,
       "learning_rate": 1.9980025988127037e-05,
-      "loss": 2.1202,
+      "loss": 2.1235,
       "step": 73
     },
     {
       "epoch": 0.1759809750297265,
-      "grad_norm": 1.266229091733924,
+      "grad_norm": 1.2574977788035384,
       "learning_rate": 1.9978797534780646e-05,
-      "loss": 2.4397,
+      "loss": 2.4466,
       "step": 74
     },
     {
       "epoch": 0.178359096313912,
-      "grad_norm": 1.2442419764935988,
+      "grad_norm": 1.390309850165232,
       "learning_rate": 1.9977532463813064e-05,
-      "loss": 2.4345,
+      "loss": 2.4469,
       "step": 75
     },
     {
       "epoch": 0.1807372175980975,
-      "grad_norm": 1.0991664177678122,
+      "grad_norm": 1.2146735833909619,
       "learning_rate": 1.9976230779866527e-05,
-      "loss": 2.3604,
+      "loss": 2.3705,
       "step": 76
     },
     {
       "epoch": 0.18311533888228299,
-      "grad_norm": 1.4648399555465317,
+      "grad_norm": 1.26356031715395,
       "learning_rate": 1.9974892487717613e-05,
-      "loss": 2.4796,
+      "loss": 2.4926,
       "step": 77
     },
     {
       "epoch": 0.18549346016646848,
-      "grad_norm": 1.4075214014031838,
+      "grad_norm": 1.3934309027656608,
       "learning_rate": 1.997351759227725e-05,
-      "loss": 2.4604,
+      "loss": 2.47,
       "step": 78
     },
     {
       "epoch": 0.187871581450654,
-      "grad_norm": 1.1658962187610111,
+      "grad_norm": 1.129998342751621,
       "learning_rate": 1.9972106098590665e-05,
-      "loss": 2.3639,
+      "loss": 2.3718,
       "step": 79
     },
     {
       "epoch": 0.1902497027348395,
-      "grad_norm": 1.0920203834125308,
+      "grad_norm": 1.3957720808228478,
       "learning_rate": 1.9970658011837404e-05,
-      "loss": 2.1962,
+      "loss": 2.2057,
       "step": 80
     },
     {
       "epoch": 0.19262782401902498,
-      "grad_norm": 1.0547465850317588,
+      "grad_norm": 1.5835508346410572,
       "learning_rate": 1.9969173337331283e-05,
-      "loss": 2.2381,
+      "loss": 2.2551,
       "step": 81
     },
     {
       "epoch": 0.19500594530321047,
-      "grad_norm": 0.9700896578179515,
+      "grad_norm": 1.0726363834452401,
       "learning_rate": 1.996765208052037e-05,
-      "loss": 1.9818,
+      "loss": 1.9962,
       "step": 82
     },
     {
       "epoch": 0.19738406658739596,
-      "grad_norm": 1.3721631541122385,
+      "grad_norm": 1.5380248440103288,
       "learning_rate": 1.9966094246986983e-05,
-      "loss": 2.1842,
+      "loss": 2.1986,
       "step": 83
     },
     {
       "epoch": 0.19976218787158145,
-      "grad_norm": 1.1909292893353944,
+      "grad_norm": 1.2953387209833067,
       "learning_rate": 1.9964499842447665e-05,
-      "loss": 2.5704,
+      "loss": 2.5842,
       "step": 84
     },
     {
       "epoch": 0.20214030915576695,
-      "grad_norm": 1.0845849173658786,
+      "grad_norm": 1.0712033116668103,
       "learning_rate": 1.9962868872753144e-05,
-      "loss": 2.1158,
+      "loss": 2.1298,
       "step": 85
     },
     {
       "epoch": 0.20451843043995244,
-      "grad_norm": 1.1512213319968665,
+      "grad_norm": 1.095761902776689,
       "learning_rate": 1.996120134388834e-05,
-      "loss": 2.3564,
+      "loss": 2.3641,
       "step": 86
     },
     {
       "epoch": 0.20689655172413793,
-      "grad_norm": 1.1653706380489515,
+      "grad_norm": 1.186679631328553,
       "learning_rate": 1.995949726197231e-05,
-      "loss": 2.4664,
+      "loss": 2.4801,
       "step": 87
     },
     {
       "epoch": 0.20927467300832342,
-      "grad_norm": 1.2375083589347724,
+      "grad_norm": 1.1887498108170933,
       "learning_rate": 1.9957756633258264e-05,
-      "loss": 2.2763,
+      "loss": 2.2866,
       "step": 88
     },
     {
       "epoch": 0.2116527942925089,
-      "grad_norm": 1.180995352120129,
+      "grad_norm": 1.0909023774872124,
       "learning_rate": 1.9955979464133515e-05,
-      "loss": 2.28,
+      "loss": 2.2916,
       "step": 89
     },
     {
       "epoch": 0.2140309155766944,
-      "grad_norm": 1.0060015116786936,
+      "grad_norm": 1.1077175360558418,
       "learning_rate": 1.995416576111945e-05,
-      "loss": 2.0939,
+      "loss": 2.1077,
       "step": 90
     },
     {
       "epoch": 0.2164090368608799,
-      "grad_norm": 1.123230557994905,
+      "grad_norm": 1.1526064687436712,
       "learning_rate": 1.9952315530871537e-05,
-      "loss": 2.2575,
+      "loss": 2.2723,
       "step": 91
     },
     {
       "epoch": 0.21878715814506539,
-      "grad_norm": 1.033250223812201,
+      "grad_norm": 1.1818210038912647,
       "learning_rate": 1.9950428780179274e-05,
-      "loss": 2.2192,
+      "loss": 2.2338,
       "step": 92
     },
     {
       "epoch": 0.2211652794292509,
-      "grad_norm": 1.01496288741104,
+      "grad_norm": 1.164942154271255,
       "learning_rate": 1.994850551596617e-05,
-      "loss": 2.3692,
+      "loss": 2.3817,
       "step": 93
     },
     {
       "epoch": 0.2235434007134364,
-      "grad_norm": 1.2509512883238079,
+      "grad_norm": 1.339398993177121,
       "learning_rate": 1.9946545745289727e-05,
-      "loss": 2.5349,
+      "loss": 2.5508,
       "step": 94
     },
     {
       "epoch": 0.2259215219976219,
-      "grad_norm": 1.1631889518067213,
+      "grad_norm": 1.3267763052855093,
       "learning_rate": 1.9944549475341404e-05,
-      "loss": 2.2335,
+      "loss": 2.247,
       "step": 95
     },
     {
       "epoch": 0.22829964328180738,
-      "grad_norm": 1.0176487393302203,
+      "grad_norm": 1.1953250811556597,
       "learning_rate": 1.99425167134466e-05,
-      "loss": 2.325,
+      "loss": 2.3373,
       "step": 96
     },
     {
       "epoch": 0.23067776456599287,
-      "grad_norm": 1.0766159235170416,
+      "grad_norm": 1.4321452409301854,
       "learning_rate": 1.9940447467064624e-05,
-      "loss": 2.4656,
+      "loss": 2.4776,
       "step": 97
     },
     {
       "epoch": 0.23305588585017836,
-      "grad_norm": 1.1250639831138038,
+      "grad_norm": 1.0224444212683161,
       "learning_rate": 1.9938341743788658e-05,
-      "loss": 2.1741,
+      "loss": 2.1837,
       "step": 98
     },
     {
       "epoch": 0.23543400713436385,
-      "grad_norm": 1.1024674310720775,
+      "grad_norm": 1.0977574950238398,
       "learning_rate": 1.9936199551345744e-05,
-      "loss": 2.1336,
+      "loss": 2.1478,
       "step": 99
     },
     {
       "epoch": 0.23781212841854935,
-      "grad_norm": 1.2527734635640946,
+      "grad_norm": 1.0660069054078747,
       "learning_rate": 1.9934020897596752e-05,
-      "loss": 2.2741,
+      "loss": 2.2816,
       "step": 100
     },
     {
       "epoch": 0.24019024970273484,
-      "grad_norm": 0.9699919156060421,
+      "grad_norm": 1.0312993672336248,
       "learning_rate": 1.9931805790536342e-05,
-      "loss": 2.2369,
+      "loss": 2.2468,
       "step": 101
     },
     {
       "epoch": 0.24256837098692033,
-      "grad_norm": 1.1834325873202396,
+      "grad_norm": 1.1278898252066067,
       "learning_rate": 1.9929554238292944e-05,
-      "loss": 2.0419,
+      "loss": 2.0526,
       "step": 102
     },
     {
       "epoch": 0.24494649227110582,
-      "grad_norm": 1.1078024399344104,
+      "grad_norm": 1.288343002944789,
       "learning_rate": 1.992726624912872e-05,
-      "loss": 2.4991,
+      "loss": 2.512,
       "step": 103
     },
     {
       "epoch": 0.2473246135552913,
-      "grad_norm": 1.1465140647519878,
+      "grad_norm": 1.1840020386119305,
       "learning_rate": 1.992494183143955e-05,
-      "loss": 2.6218,
+      "loss": 2.6356,
       "step": 104
     },
     {
       "epoch": 0.2497027348394768,
-      "grad_norm": 1.143333452765713,
+      "grad_norm": 1.1014678408276726,
       "learning_rate": 1.9922580993754985e-05,
-      "loss": 2.3428,
+      "loss": 2.3521,
       "step": 105
     },
     {
       "epoch": 0.2497027348394768,
-      "eval_loss": 2.499577760696411,
-      "eval_runtime": 151.1827,
-      "eval_samples_per_second": 4.227,
-      "eval_steps_per_second": 0.529,
+      "eval_loss": 2.5365779399871826,
+      "eval_runtime": 66.0796,
+      "eval_samples_per_second": 9.67,
+      "eval_steps_per_second": 1.211,
       "step": 105
     },
     {
       "epoch": 0.2520808561236623,
-      "grad_norm": 1.0672145334944887,
+      "grad_norm": 0.9957844757920508,
       "learning_rate": 1.9920183744738208e-05,
-      "loss": 2.3485,
+      "loss": 2.355,
       "step": 106
     },
     {
       "epoch": 0.2544589774078478,
-      "grad_norm": 1.0044582600402383,
+      "grad_norm": 1.0070598447313825,
       "learning_rate": 1.9917750093186036e-05,
-      "loss": 2.1678,
+      "loss": 2.1747,
       "step": 107
     },
     {
       "epoch": 0.2568370986920333,
-      "grad_norm": 1.1029447003396373,
+      "grad_norm": 1.2463453868295562,
       "learning_rate": 1.9915280048028853e-05,
-      "loss": 2.3967,
+      "loss": 2.4131,
       "step": 108
     },
     {
       "epoch": 0.25921521997621877,
-      "grad_norm": 1.260079077740416,
+      "grad_norm": 1.158558292534161,
       "learning_rate": 1.9912773618330595e-05,
-      "loss": 2.4385,
+      "loss": 2.4527,
       "step": 109
     },
     {
       "epoch": 0.2615933412604043,
-      "grad_norm": 1.0846663572847435,
+      "grad_norm": 1.1875643459332377,
       "learning_rate": 1.9910230813288713e-05,
-      "loss": 2.1431,
+      "loss": 2.1523,
       "step": 110
     },
     {
       "epoch": 0.26397146254458975,
-      "grad_norm": 0.991674367348856,
+      "grad_norm": 0.892269173897758,
       "learning_rate": 1.9907651642234138e-05,
-      "loss": 1.9523,
+      "loss": 1.9606,
       "step": 111
     },
     {
       "epoch": 0.26634958382877527,
-      "grad_norm": 1.105394178981242,
+      "grad_norm": 1.181952902180908,
       "learning_rate": 1.9905036114631247e-05,
-      "loss": 2.3063,
+      "loss": 2.3201,
       "step": 112
     },
     {
       "epoch": 0.26872770511296074,
-      "grad_norm": 0.9979450972746343,
+      "grad_norm": 0.9689153704257877,
       "learning_rate": 1.990238424007783e-05,
-      "loss": 2.2218,
+      "loss": 2.2329,
       "step": 113
     },
     {
       "epoch": 0.27110582639714625,
-      "grad_norm": 1.2796796522419402,
+      "grad_norm": 1.3665918769424286,
       "learning_rate": 1.989969602830505e-05,
-      "loss": 2.2271,
+      "loss": 2.2387,
       "step": 114
     },
     {
       "epoch": 0.2734839476813318,
-      "grad_norm": 1.1720433923322253,
+      "grad_norm": 1.0478434719151144,
       "learning_rate": 1.9896971489177417e-05,
-      "loss": 2.2736,
+      "loss": 2.2798,
       "step": 115
     },
     {
       "epoch": 0.27586206896551724,
-      "grad_norm": 1.0744949490065667,
+      "grad_norm": 1.5752154316391798,
       "learning_rate": 1.9894210632692745e-05,
-      "loss": 2.1864,
+      "loss": 2.201,
       "step": 116
     },
     {
       "epoch": 0.27824019024970276,
-      "grad_norm": 1.9078078540875143,
+      "grad_norm": 1.0264277011384757,
       "learning_rate": 1.9891413468982112e-05,
-      "loss": 2.2678,
+      "loss": 2.2756,
       "step": 117
     },
     {
       "epoch": 0.2806183115338882,
-      "grad_norm": 1.0427118571864202,
+      "grad_norm": 1.1750703393359614,
       "learning_rate": 1.988858000830983e-05,
-      "loss": 2.1757,
+      "loss": 2.1907,
       "step": 118
     },
     {
       "epoch": 0.28299643281807374,
-      "grad_norm": 1.8597307152991636,
+      "grad_norm": 0.9456957190962577,
       "learning_rate": 1.9885710261073402e-05,
-      "loss": 2.2913,
+      "loss": 2.2993,
       "step": 119
     },
     {
       "epoch": 0.2853745541022592,
-      "grad_norm": 1.0401590654407316,
+      "grad_norm": 1.37591692336223,
       "learning_rate": 1.9882804237803487e-05,
-      "loss": 2.0629,
+      "loss": 2.0751,
       "step": 120
     },
     {
       "epoch": 0.2877526753864447,
-      "grad_norm": 1.5079024059776849,
+      "grad_norm": 0.991181263305241,
       "learning_rate": 1.9879861949163863e-05,
-      "loss": 2.1827,
+      "loss": 2.1946,
       "step": 121
     },
     {
       "epoch": 0.2901307966706302,
-      "grad_norm": 1.1499157717530648,
+      "grad_norm": 1.2826616603092615,
       "learning_rate": 1.9876883405951378e-05,
-      "loss": 2.2959,
+      "loss": 2.3084,
       "step": 122
     },
     {
       "epoch": 0.2925089179548157,
-      "grad_norm": 1.030399166377001,
+      "grad_norm": 1.3162982027829009,
       "learning_rate": 1.987386861909593e-05,
-      "loss": 2.279,
+      "loss": 2.294,
       "step": 123
     },
     {
       "epoch": 0.29488703923900117,
-      "grad_norm": 1.062349469745721,
+      "grad_norm": 1.086311999313279,
       "learning_rate": 1.98708175996604e-05,
-      "loss": 2.2865,
+      "loss": 2.3025,
       "step": 124
     },
     {
       "epoch": 0.2972651605231867,
-      "grad_norm": 1.0853089306877393,
+      "grad_norm": 1.10683170372015,
       "learning_rate": 1.986773035884064e-05,
-      "loss": 2.332,
+      "loss": 2.3447,
       "step": 125
     },
     {
       "epoch": 0.29964328180737215,
-      "grad_norm": 1.0066223498076698,
+      "grad_norm": 1.090568761480393,
       "learning_rate": 1.9864606907965407e-05,
-      "loss": 2.3971,
+      "loss": 2.4104,
       "step": 126
     },
     {
       "epoch": 0.30202140309155767,
-      "grad_norm": 1.101756427229776,
+      "grad_norm": 1.4024759238343605,
       "learning_rate": 1.986144725849634e-05,
-      "loss": 2.2862,
+      "loss": 2.298,
       "step": 127
     },
     {
       "epoch": 0.30439952437574314,
-      "grad_norm": 1.1264625682995106,
+      "grad_norm": 0.9324914520062791,
       "learning_rate": 1.9858251422027903e-05,
-      "loss": 2.0974,
+      "loss": 2.1123,
       "step": 128
     },
     {
       "epoch": 0.30677764565992865,
-      "grad_norm": 1.0012622487091893,
+      "grad_norm": 1.3818136151492852,
       "learning_rate": 1.9855019410287355e-05,
-      "loss": 2.2648,
+      "loss": 2.2786,
       "step": 129
     },
     {
       "epoch": 0.3091557669441142,
-      "grad_norm": 1.0030908712085922,
+      "grad_norm": 0.9879756737720099,
       "learning_rate": 1.98517512351347e-05,
-      "loss": 2.2626,
+      "loss": 2.2735,
       "step": 130
     },
     {
       "epoch": 0.31153388822829964,
-      "grad_norm": 1.1376149636856583,
+      "grad_norm": 1.4107106057474024,
       "learning_rate": 1.9848446908562647e-05,
-      "loss": 2.2315,
+      "loss": 2.2421,
       "step": 131
     },
     {
       "epoch": 0.31391200951248516,
-      "grad_norm": 0.9616294646184662,
+      "grad_norm": 0.978862094447652,
       "learning_rate": 1.9845106442696563e-05,
-      "loss": 2.4033,
+      "loss": 2.4152,
       "step": 132
     },
     {
       "epoch": 0.3162901307966706,
-      "grad_norm": 1.3402719458865533,
+      "grad_norm": 1.3714074038447606,
       "learning_rate": 1.9841729849794427e-05,
-      "loss": 2.4429,
+      "loss": 2.4567,
       "step": 133
     },
     {
       "epoch": 0.31866825208085614,
-      "grad_norm": 1.3575140959134089,
+      "grad_norm": 1.030641093673837,
       "learning_rate": 1.983831714224679e-05,
-      "loss": 2.291,
+      "loss": 2.3015,
       "step": 134
     },
     {
       "epoch": 0.3210463733650416,
-      "grad_norm": 0.9521047622766085,
+      "grad_norm": 1.1744699755999302,
       "learning_rate": 1.9834868332576727e-05,
-      "loss": 2.2754,
+      "loss": 2.2878,
       "step": 135
     },
     {
       "epoch": 0.3234244946492271,
-      "grad_norm": 1.0875381930222732,
+      "grad_norm": 0.9733999816490441,
       "learning_rate": 1.9831383433439798e-05,
-      "loss": 2.1469,
+      "loss": 2.1571,
       "step": 136
     },
     {
       "epoch": 0.3258026159334126,
-      "grad_norm": 1.0257090212605473,
+      "grad_norm": 1.0470367999253474,
       "learning_rate": 1.982786245762398e-05,
-      "loss": 2.0848,
+      "loss": 2.0943,
       "step": 137
     },
     {
       "epoch": 0.3281807372175981,
-      "grad_norm": 1.0372156134974286,
+      "grad_norm": 1.0748276455064096,
       "learning_rate": 1.9824305418049645e-05,
-      "loss": 2.4043,
+      "loss": 2.4156,
       "step": 138
     },
     {
       "epoch": 0.33055885850178357,
-      "grad_norm": 0.9832029689246473,
+      "grad_norm": 1.0220509349947084,
       "learning_rate": 1.9820712327769503e-05,
-      "loss": 2.177,
+      "loss": 2.1898,
       "step": 139
     },
     {
       "epoch": 0.3329369797859691,
-      "grad_norm": 1.0472107045649877,
+      "grad_norm": 0.9811166423920332,
       "learning_rate": 1.9817083199968552e-05,
-      "loss": 2.3309,
+      "loss": 2.3449,
       "step": 140
     },
     {
       "epoch": 0.33531510107015455,
-      "grad_norm": 1.0277879008926316,
+      "grad_norm": 1.0664757695722766,
       "learning_rate": 1.9813418047964025e-05,
-      "loss": 2.1389,
+      "loss": 2.1514,
       "step": 141
     },
     {
       "epoch": 0.3376932223543401,
-      "grad_norm": 1.15382327979194,
+      "grad_norm": 1.1228830278366924,
       "learning_rate": 1.9809716885205363e-05,
-      "loss": 2.3254,
+      "loss": 2.3371,
       "step": 142
     },
     {
       "epoch": 0.3400713436385256,
-      "grad_norm": 0.9933827586398313,
+      "grad_norm": 1.0703957613617774,
       "learning_rate": 1.980597972527413e-05,
-      "loss": 2.2454,
+      "loss": 2.2577,
       "step": 143
     },
     {
       "epoch": 0.34244946492271106,
-      "grad_norm": 0.9637942573486198,
+      "grad_norm": 0.9971842999532138,
       "learning_rate": 1.9802206581883992e-05,
-      "loss": 2.1945,
+      "loss": 2.2048,
       "step": 144
     },
     {
       "epoch": 0.3448275862068966,
-      "grad_norm": 0.918837452187297,
+      "grad_norm": 0.9969712850303254,
       "learning_rate": 1.979839746888067e-05,
-      "loss": 2.1599,
+      "loss": 2.1725,
       "step": 145
     },
     {
       "epoch": 0.34720570749108204,
-      "grad_norm": 0.962272148586432,
+      "grad_norm": 0.9782490093980141,
       "learning_rate": 1.979455240024186e-05,
-      "loss": 2.1504,
+      "loss": 2.1598,
       "step": 146
     },
     {
       "epoch": 0.34958382877526756,
-      "grad_norm": 1.0902104459187203,
+      "grad_norm": 1.1595035293528873,
       "learning_rate": 1.97906713900772e-05,
-      "loss": 2.1671,
+      "loss": 2.1812,
       "step": 147
     },
     {
       "epoch": 0.351961950059453,
-      "grad_norm": 0.9528336584077055,
+      "grad_norm": 1.0488323565717943,
       "learning_rate": 1.9786754452628226e-05,
-      "loss": 2.113,
+      "loss": 2.126,
       "step": 148
     },
     {
       "epoch": 0.35434007134363854,
-      "grad_norm": 0.9216300894356183,
+      "grad_norm": 1.0236205683546673,
       "learning_rate": 1.9782801602268306e-05,
-      "loss": 1.9271,
+      "loss": 1.9399,
       "step": 149
     },
     {
       "epoch": 0.356718192627824,
-      "grad_norm": 0.9148117236258577,
+      "grad_norm": 0.983049547537296,
       "learning_rate": 1.9778812853502592e-05,
-      "loss": 2.0221,
+      "loss": 2.0336,
       "step": 150
     },
     {
       "epoch": 0.3590963139120095,
-      "grad_norm": 0.9737793407143137,
+      "grad_norm": 1.0856474713800959,
       "learning_rate": 1.9774788220967968e-05,
-      "loss": 2.2007,
+      "loss": 2.2103,
       "step": 151
     },
     {
       "epoch": 0.361474435196195,
-      "grad_norm": 0.8979472123958888,
+      "grad_norm": 1.098143269144179,
       "learning_rate": 1.9770727719432994e-05,
-      "loss": 2.1295,
+      "loss": 2.1425,
       "step": 152
     },
     {
       "epoch": 0.3638525564803805,
-      "grad_norm": 1.0391719811646754,
+      "grad_norm": 1.1908904777112574,
       "learning_rate": 1.9766631363797852e-05,
-      "loss": 2.2383,
+      "loss": 2.2516,
       "step": 153
     },
     {
       "epoch": 0.36623067776456597,
-      "grad_norm": 1.0898893562871452,
+      "grad_norm": 1.1823343263781934,
       "learning_rate": 1.9762499169094288e-05,
-      "loss": 2.085,
+      "loss": 2.0991,
       "step": 154
     },
     {
       "epoch": 0.3686087990487515,
-      "grad_norm": 2.9046586074281686,
+      "grad_norm": 1.1543274307271654,
       "learning_rate": 1.9758331150485576e-05,
-      "loss": 2.2903,
+      "loss": 2.2917,
       "step": 155
     },
     {
       "epoch": 0.37098692033293695,
-      "grad_norm": 1.0438135264691892,
+      "grad_norm": 1.1828452156246019,
       "learning_rate": 1.9754127323266426e-05,
-      "loss": 2.349,
+      "loss": 2.3577,
       "step": 156
     },
     {
       "epoch": 0.3733650416171225,
-      "grad_norm": 1.2082790526216014,
+      "grad_norm": 1.2458434785978698,
       "learning_rate": 1.9749887702862972e-05,
-      "loss": 2.2182,
+      "loss": 2.2291,
       "step": 157
     },
     {
       "epoch": 0.375743162901308,
-      "grad_norm": 1.2415740026323197,
+      "grad_norm": 1.0632348458757013,
       "learning_rate": 1.9745612304832672e-05,
-      "loss": 2.4834,
+      "loss": 2.495,
       "step": 158
     },
     {
       "epoch": 0.37812128418549346,
-      "grad_norm": 0.9901876124346225,
+      "grad_norm": 1.2413557275846534,
       "learning_rate": 1.9741301144864284e-05,
-      "loss": 2.2873,
+      "loss": 2.3006,
       "step": 159
     },
     {
       "epoch": 0.380499405469679,
-      "grad_norm": 1.1185971951047096,
+      "grad_norm": 1.068837985332943,
       "learning_rate": 1.9736954238777793e-05,
-      "loss": 2.2114,
+      "loss": 2.2228,
       "step": 160
     },
     {
       "epoch": 0.38287752675386444,
-      "grad_norm": 1.0186645068648283,
+      "grad_norm": 1.181973772137545,
       "learning_rate": 1.9732571602524353e-05,
-      "loss": 2.3323,
+      "loss": 2.3419,
       "step": 161
     },
     {
       "epoch": 0.38525564803804996,
-      "grad_norm": 0.9856339888297305,
+      "grad_norm": 0.9361759344356807,
       "learning_rate": 1.972815325218624e-05,
-      "loss": 2.2638,
+      "loss": 2.2727,
       "step": 162
     },
     {
       "epoch": 0.3876337693222354,
-      "grad_norm": 1.287711819624049,
+      "grad_norm": 1.2300672941710984,
       "learning_rate": 1.9723699203976768e-05,
-      "loss": 2.3897,
+      "loss": 2.3947,
       "step": 163
     },
     {
       "epoch": 0.39001189060642094,
-      "grad_norm": 0.9474533284935532,
+      "grad_norm": 0.9647921025871186,
       "learning_rate": 1.9719209474240263e-05,
-      "loss": 1.8287,
+      "loss": 1.8388,
       "step": 164
     },
     {
       "epoch": 0.3923900118906064,
-      "grad_norm": 1.0505224096035144,
+      "grad_norm": 1.1390311715526416,
       "learning_rate": 1.971468407945198e-05,
-      "loss": 2.3906,
+      "loss": 2.4054,
       "step": 165
     },
     {
       "epoch": 0.3947681331747919,
-      "grad_norm": 0.9322039774829307,
+      "grad_norm": 0.9830051867519547,
       "learning_rate": 1.9710123036218044e-05,
-      "loss": 2.0246,
+      "loss": 2.0355,
       "step": 166
     },
     {
       "epoch": 0.3971462544589774,
-      "grad_norm": 1.1428006052468438,
+      "grad_norm": 1.1244517585073737,
       "learning_rate": 1.97055263612754e-05,
-      "loss": 2.0085,
+      "loss": 2.0188,
       "step": 167
     },
     {
       "epoch": 0.3995243757431629,
-      "grad_norm": 0.9233456322532203,
+      "grad_norm": 1.0256020852263494,
       "learning_rate": 1.9700894071491736e-05,
-      "loss": 2.0657,
+      "loss": 2.0774,
       "step": 168
     },
     {
       "epoch": 0.40190249702734837,
-      "grad_norm": 1.1387607148614496,
+      "grad_norm": 1.011023720252716,
       "learning_rate": 1.9696226183865436e-05,
-      "loss": 2.2507,
+      "loss": 2.2592,
       "step": 169
     },
     {
       "epoch": 0.4042806183115339,
-      "grad_norm": 1.0240739510681864,
+      "grad_norm": 1.046975898884085,
       "learning_rate": 1.969152271552552e-05,
-      "loss": 2.1685,
+      "loss": 2.1791,
       "step": 170
     },
     {
       "epoch": 0.40665873959571935,
-      "grad_norm": 1.2665975670284688,
+      "grad_norm": 1.1800984480399852,
       "learning_rate": 1.9686783683731557e-05,
-      "loss": 2.3869,
+      "loss": 2.3941,
       "step": 171
     },
     {
       "epoch": 0.4090368608799049,
-      "grad_norm": 1.0148421037850517,
+      "grad_norm": 1.2459882622321672,
       "learning_rate": 1.9682009105873633e-05,
-      "loss": 2.1379,
+      "loss": 2.1522,
       "step": 172
     },
     {
       "epoch": 0.4114149821640904,
-      "grad_norm": 1.0117482642225601,
+      "grad_norm": 1.0732133381850257,
       "learning_rate": 1.9677198999472257e-05,
-      "loss": 2.1104,
+      "loss": 2.1233,
       "step": 173
     },
     {
       "epoch": 0.41379310344827586,
-      "grad_norm": 0.9720066057353862,
+      "grad_norm": 1.2405484917580802,
       "learning_rate": 1.967235338217832e-05,
-      "loss": 2.2884,
+      "loss": 2.3016,
       "step": 174
     },
     {
       "epoch": 0.4161712247324614,
-      "grad_norm": 1.2836956527083296,
+      "grad_norm": 1.0759940201219593,
       "learning_rate": 1.9667472271773026e-05,
-      "loss": 2.281,
+      "loss": 2.2947,
       "step": 175
     },
     {
       "epoch": 0.41854934601664684,
-      "grad_norm": 0.9738075739171279,
+      "grad_norm": 1.2008734320661734,
       "learning_rate": 1.9662555686167808e-05,
-      "loss": 2.2039,
+      "loss": 2.2155,
       "step": 176
     },
     {
       "epoch": 0.42092746730083236,
-      "grad_norm": 1.0382703606377657,
+      "grad_norm": 0.9303619935178572,
       "learning_rate": 1.965760364340429e-05,
-      "loss": 2.1142,
+      "loss": 2.1234,
       "step": 177
     },
     {
       "epoch": 0.4233055885850178,
-      "grad_norm": 0.9318167386351257,
+      "grad_norm": 1.3884826767438652,
       "learning_rate": 1.9652616161654204e-05,
-      "loss": 2.2409,
+      "loss": 2.2539,
       "step": 178
     },
     {
       "epoch": 0.42568370986920334,
-      "grad_norm": 1.3023888510009893,
+      "grad_norm": 0.9947187673832885,
       "learning_rate": 1.9647593259219328e-05,
-      "loss": 2.1972,
+      "loss": 2.2052,
       "step": 179
     },
     {
       "epoch": 0.4280618311533888,
-      "grad_norm": 2.3590946551757583,
+      "grad_norm": 1.4655922792083054,
       "learning_rate": 1.964253495453141e-05,
-      "loss": 2.1922,
+      "loss": 2.1552,
       "step": 180
     },
     {
       "epoch": 0.4304399524375743,
-      "grad_norm": 1.277353180787431,
+      "grad_norm": 1.1481294188693778,
       "learning_rate": 1.963744126615212e-05,
-      "loss": 2.384,
+      "loss": 2.3942,
       "step": 181
     },
     {
       "epoch": 0.4328180737217598,
-      "grad_norm": 0.9912373962254282,
+      "grad_norm": 1.239760521409481,
       "learning_rate": 1.9632312212772956e-05,
-      "loss": 2.3005,
+      "loss": 2.3091,
       "step": 182
     },
     {
       "epoch": 0.4351961950059453,
-      "grad_norm": 1.3088391002508755,
+      "grad_norm": 1.0524654460411744,
       "learning_rate": 1.9627147813215207e-05,
-      "loss": 2.2924,
+      "loss": 2.302,
       "step": 183
     },
     {
       "epoch": 0.43757431629013077,
-      "grad_norm": 1.0523417119399674,
+      "grad_norm": 1.0231645108607732,
       "learning_rate": 1.9621948086429847e-05,
-      "loss": 2.2255,
+      "loss": 2.2334,
       "step": 184
     },
     {
       "epoch": 0.4399524375743163,
-      "grad_norm": 1.386504728655603,
+      "grad_norm": 1.0600582051447691,
       "learning_rate": 1.9616713051497496e-05,
-      "loss": 2.1963,
+      "loss": 2.2044,
       "step": 185
     },
     {
       "epoch": 0.4423305588585018,
-      "grad_norm": 0.9623064139518317,
+      "grad_norm": 1.0861978175484295,
       "learning_rate": 1.9611442727628344e-05,
-      "loss": 2.3175,
+      "loss": 2.3267,
       "step": 186
     },
     {
       "epoch": 0.4447086801426873,
-      "grad_norm": 1.7130236244329065,
+      "grad_norm": 1.0122924353396487,
       "learning_rate": 1.960613713416206e-05,
-      "loss": 2.2246,
+      "loss": 2.2327,
       "step": 187
     },
     {
       "epoch": 0.4470868014268728,
-      "grad_norm": 1.2183030146426241,
+      "grad_norm": 1.1275635495135592,
       "learning_rate": 1.9600796290567747e-05,
-      "loss": 2.2345,
+      "loss": 2.2474,
       "step": 188
     },
     {
       "epoch": 0.44946492271105826,
-      "grad_norm": 2.1564007093614497,
+      "grad_norm": 1.0778906611663819,
       "learning_rate": 1.9595420216443864e-05,
-      "loss": 2.269,
+      "loss": 2.2777,
       "step": 189
     },
     {
       "epoch": 0.4518430439952438,
-      "grad_norm": 1.1685306329495788,
+      "grad_norm": 1.0593499669893551,
       "learning_rate": 1.9590008931518133e-05,
-      "loss": 2.4817,
+      "loss": 2.4937,
       "step": 190
     },
     {
       "epoch": 0.45422116527942924,
-      "grad_norm": 1.9314760188762214,
+      "grad_norm": 1.0887914371115388,
       "learning_rate": 1.9584562455647494e-05,
-      "loss": 2.2482,
+      "loss": 2.2577,
       "step": 191
     },
     {
       "epoch": 0.45659928656361476,
-      "grad_norm": 1.2361103482091276,
+      "grad_norm": 1.0280779311785984,
       "learning_rate": 1.9579080808818035e-05,
-      "loss": 2.2229,
+      "loss": 2.2352,
       "step": 192
     },
     {
       "epoch": 0.4589774078478002,
-      "grad_norm": 1.7230964883799738,
+      "grad_norm": 1.1201705856067985,
       "learning_rate": 1.9573564011144873e-05,
-      "loss": 2.142,
+      "loss": 2.1482,
       "step": 193
     },
     {
       "epoch": 0.46135552913198574,
-      "grad_norm": 1.2344882256405865,
+      "grad_norm": 1.0039435227655624,
       "learning_rate": 1.9568012082872148e-05,
-      "loss": 2.0984,
+      "loss": 2.1069,
       "step": 194
     },
     {
       "epoch": 0.4637336504161712,
-      "grad_norm": 1.563594408096457,
+      "grad_norm": 1.0523831000821406,
       "learning_rate": 1.9562425044372884e-05,
-      "loss": 1.9245,
+      "loss": 1.9268,
       "step": 195
     },
     {
       "epoch": 0.4661117717003567,
-      "grad_norm": 1.942680386259649,
+      "grad_norm": 1.0635880350342213,
       "learning_rate": 1.9556802916148963e-05,
-      "loss": 2.2633,
+      "loss": 2.2722,
       "step": 196
     },
     {
       "epoch": 0.4684898929845422,
-      "grad_norm": 1.3339191105911707,
+      "grad_norm": 2.4351848601787287,
       "learning_rate": 1.955114571883102e-05,
-      "loss": 2.1356,
+      "loss": 2.1402,
       "step": 197
     },
     {
       "epoch": 0.4708680142687277,
-      "grad_norm": 2.204421390949867,
+      "grad_norm": 1.2199308274597462,
       "learning_rate": 1.9545453473178384e-05,
-      "loss": 2.2506,
+      "loss": 2.2599,
       "step": 198
     },
     {
       "epoch": 0.47324613555291317,
-      "grad_norm": 1.106826809772669,
+      "grad_norm": 0.9936114796299212,
       "learning_rate": 1.9539726200078987e-05,
-      "loss": 2.0526,
+      "loss": 2.0662,
       "step": 199
     },
     {
       "epoch": 0.4756242568370987,
-      "grad_norm": 2.7022504927488136,
+      "grad_norm": 1.0692703333507547,
       "learning_rate": 1.9533963920549307e-05,
-      "loss": 2.3654,
+      "loss": 2.3739,
       "step": 200
     },
     {
       "epoch": 0.4780023781212842,
-      "grad_norm": 2.20401718688023,
+      "grad_norm": 1.0406002686664542,
       "learning_rate": 1.9528166655734267e-05,
-      "loss": 2.3523,
+      "loss": 2.3611,
       "step": 201
     },
     {
       "epoch": 0.4803804994054697,
-      "grad_norm": 1.7098249314197853,
+      "grad_norm": 1.9375905536343168,
       "learning_rate": 1.9522334426907185e-05,
-      "loss": 2.0789,
+      "loss": 2.0971,
       "step": 202
     },
     {
       "epoch": 0.4827586206896552,
-      "grad_norm": 2.893610442346491,
+      "grad_norm": 1.024548704059581,
       "learning_rate": 1.951646725546966e-05,
-      "loss": 2.2456,
+      "loss": 2.2498,
       "step": 203
     },
     {
       "epoch": 0.48513674197384066,
-      "grad_norm": 2.2821472751414937,
+      "grad_norm": 1.0033895284405978,
       "learning_rate": 1.9510565162951538e-05,
-      "loss": 2.3351,
+      "loss": 2.299,
       "step": 204
     },
     {
       "epoch": 0.4875148632580262,
-      "grad_norm": 1.300863641854685,
+      "grad_norm": 1.541631519071697,
       "learning_rate": 1.950462817101079e-05,
-      "loss": 2.3617,
+      "loss": 2.4076,
       "step": 205
     },
     {
       "epoch": 0.48989298454221164,
-      "grad_norm": 1.3899023575579752,
+      "grad_norm": 0.9499702987331401,
       "learning_rate": 1.9498656301433466e-05,
-      "loss": 2.0628,
+      "loss": 2.0754,
       "step": 206
     },
     {
       "epoch": 0.49227110582639716,
-      "grad_norm": 1.018395556873715,
+      "grad_norm": 1.099383371761328,
       "learning_rate": 1.9492649576133594e-05,
-      "loss": 2.2362,
+      "loss": 2.2514,
       "step": 207
     },
     {
       "epoch": 0.4946492271105826,
-      "grad_norm": 1.4600451736607643,
+      "grad_norm": 0.9296431838496088,
       "learning_rate": 1.94866080171531e-05,
-      "loss": 2.2194,
+      "loss": 2.2308,
       "step": 208
     },
     {
       "epoch": 0.49702734839476814,
-      "grad_norm": 4.244045726741938,
+      "grad_norm": 4.140796209905845,
       "learning_rate": 1.9480531646661753e-05,
-      "loss": 2.434,
+      "loss": 2.4388,
       "step": 209
     },
     {
       "epoch": 0.4994054696789536,
-      "grad_norm": 1.2125371460079588,
+      "grad_norm": 1.011142238194789,
       "learning_rate": 1.9474420486957045e-05,
-      "loss": 2.228,
+      "loss": 2.2414,
       "step": 210
     },
     {
       "epoch": 0.4994054696789536,
-      "eval_loss": 2.4567697048187256,
-      "eval_runtime": 151.5129,
-      "eval_samples_per_second": 4.217,
-      "eval_steps_per_second": 0.528,
+      "eval_loss": 2.49302339553833,
+      "eval_runtime": 65.8636,
+      "eval_samples_per_second": 9.702,
+      "eval_steps_per_second": 1.215,
       "step": 210
     }
   ],
@@ -1521,7 +1521,7 @@
       "attributes": {}
     }
   },
-  "total_flos": 1.3027053762772992e+17,
+  "total_flos": 1.3027116576669696e+17,
   "train_batch_size": 1,
   "trial_name": null,
   "trial_params": null

checkpoint-210/training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:14afcc01255b192ff7972e2f101184c05665fd6b6b6ac8aba6cded13c143de1c
+oid sha256:2c4afa27d1a40ecd661f9724785c2e34f68177e49c01784ff719ada5cf02a780
 size 8504