Upload folder using huggingface_hub

.gitattributes

@@ -33,3 +33,34 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+checkpoint-1172/optimizer_0/.metadata filter=lfs diff=lfs merge=lfs -text
+checkpoint-1172/optimizer_0/__0_0.distcp filter=lfs diff=lfs merge=lfs -text
+checkpoint-1172/optimizer_0/__1_0.distcp filter=lfs diff=lfs merge=lfs -text
+checkpoint-1172/optimizer_0/__2_0.distcp filter=lfs diff=lfs merge=lfs -text
+checkpoint-1172/optimizer_0/__3_0.distcp filter=lfs diff=lfs merge=lfs -text
+checkpoint-1172/pytorch_model_fsdp_0/.metadata filter=lfs diff=lfs merge=lfs -text
+checkpoint-1172/pytorch_model_fsdp_0/__0_0.distcp filter=lfs diff=lfs merge=lfs -text
+checkpoint-1172/pytorch_model_fsdp_0/__1_0.distcp filter=lfs diff=lfs merge=lfs -text
+checkpoint-1172/pytorch_model_fsdp_0/__2_0.distcp filter=lfs diff=lfs merge=lfs -text
+checkpoint-1172/pytorch_model_fsdp_0/__3_0.distcp filter=lfs diff=lfs merge=lfs -text
+checkpoint-2344/optimizer_0/.metadata filter=lfs diff=lfs merge=lfs -text
+checkpoint-2344/optimizer_0/__0_0.distcp filter=lfs diff=lfs merge=lfs -text
+checkpoint-2344/optimizer_0/__1_0.distcp filter=lfs diff=lfs merge=lfs -text
+checkpoint-2344/optimizer_0/__2_0.distcp filter=lfs diff=lfs merge=lfs -text
+checkpoint-2344/optimizer_0/__3_0.distcp filter=lfs diff=lfs merge=lfs -text
+checkpoint-2344/pytorch_model_fsdp_0/.metadata filter=lfs diff=lfs merge=lfs -text
+checkpoint-2344/pytorch_model_fsdp_0/__0_0.distcp filter=lfs diff=lfs merge=lfs -text
+checkpoint-2344/pytorch_model_fsdp_0/__1_0.distcp filter=lfs diff=lfs merge=lfs -text
+checkpoint-2344/pytorch_model_fsdp_0/__2_0.distcp filter=lfs diff=lfs merge=lfs -text
+checkpoint-2344/pytorch_model_fsdp_0/__3_0.distcp filter=lfs diff=lfs merge=lfs -text
+checkpoint-3513/optimizer_0/.metadata filter=lfs diff=lfs merge=lfs -text
+checkpoint-3513/optimizer_0/__0_0.distcp filter=lfs diff=lfs merge=lfs -text
+checkpoint-3513/optimizer_0/__1_0.distcp filter=lfs diff=lfs merge=lfs -text
+checkpoint-3513/optimizer_0/__2_0.distcp filter=lfs diff=lfs merge=lfs -text
+checkpoint-3513/optimizer_0/__3_0.distcp filter=lfs diff=lfs merge=lfs -text
+checkpoint-3513/pytorch_model_fsdp_0/.metadata filter=lfs diff=lfs merge=lfs -text
+checkpoint-3513/pytorch_model_fsdp_0/__0_0.distcp filter=lfs diff=lfs merge=lfs -text
+checkpoint-3513/pytorch_model_fsdp_0/__1_0.distcp filter=lfs diff=lfs merge=lfs -text
+checkpoint-3513/pytorch_model_fsdp_0/__2_0.distcp filter=lfs diff=lfs merge=lfs -text
+checkpoint-3513/pytorch_model_fsdp_0/__3_0.distcp filter=lfs diff=lfs merge=lfs -text
+tokenizer.json filter=lfs diff=lfs merge=lfs -text

checkpoint-1172/optimizer_0/.metadata

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:aaa48a3b43fa9f1b0a3a1b436f9d0de00d48740f75f7948f7499f12ba4718613
+size 342948

checkpoint-1172/optimizer_0/__0_0.distcp

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7d63f23b28902612af523ecde9a75b25469eb671c5db3f2438c70ef19fe8ef19
+size 2472090861

checkpoint-1172/optimizer_0/__1_0.distcp

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3ed693dd619945fafd13b1bf503e69d4d7607c6eccd9fa4b41e67a17a3dc9a18
+size 2472174123

checkpoint-1172/optimizer_0/__2_0.distcp

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:980dfa0ab9828c5a0ae79ae3b3cb850ff43bba52dd8bc9e1acf9c28d2a356a90
+size 2472182518

checkpoint-1172/optimizer_0/__3_0.distcp

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:61f8d13cefae2f7d8e43096599b731b018a1855adf4d44ed3fc8a048aeb85876
+size 2472179190

checkpoint-1172/pytorch_model_fsdp_0/.metadata

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:429d26665d6ce130813c7255fbf001ff00a15a0edea8cad2455305fcde6b9c0d
+size 129152

checkpoint-1172/pytorch_model_fsdp_0/__0_0.distcp

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8106dbe9810025c25ece98e9ce8bc2e48180092d3519229c7c8084106f16573c
+size 1498714507

checkpoint-1172/pytorch_model_fsdp_0/__1_0.distcp

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7479ae57d47bed41ba5ddb904bf17c1b23c7a1ddd39d0bcd058c671c37f62f4c
+size 1498714507

checkpoint-1172/pytorch_model_fsdp_0/__2_0.distcp

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:201f586f24a22475062808ef104aee1146892d3e1b121547b32e295536b86876
+size 1498714507

checkpoint-1172/pytorch_model_fsdp_0/__3_0.distcp

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fde1d7d49a72f2f2f89daa9f0539d0cdc0780c4b7a388c0ce2503d909d5ac9f3
+size 1498714507

checkpoint-1172/rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4188048c882e5b595f5e2cce07e8c73ea45afa11bc95f5627f2d6191ccaa5e40
+size 15429

checkpoint-1172/rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0eb1f7b1acfd920c8cdef076eee48ccfec7d3ed4d8c5d83c2592fbbb4a4f9b38
+size 15429

checkpoint-1172/rng_state_2.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5664fd5079c69dc8edcc429ded884f07d45c771f27a5a65fe1ad751348f9e1de
+size 15429

checkpoint-1172/rng_state_3.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0a3f17c4e252e5b7e99f3573c3bca992070f9ef436dd578ee9e88c333a94cef0
+size 15429

checkpoint-1172/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:27cb5271954fb800869f57c870c518e11db7b57d104b7bbfcf287f92b05c7805
+size 1465

checkpoint-1172/trainer_state.json

@@ -0,0 +1,867 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.0,
+  "eval_steps": 500,
+  "global_step": 1172,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0,
+      "eval_loss": 2.1641297340393066,
+      "eval_runtime": 1.9384,
+      "eval_samples_per_second": 10.318,
+      "eval_steps_per_second": 1.032,
+      "step": 0
+    },
+    {
+      "epoch": 0.0008535154166222128,
+      "grad_norm": 4.246492385864258,
+      "learning_rate": 2.840909090909091e-08,
+      "loss": 1.0032,
+      "step": 1
+    },
+    {
+      "epoch": 0.008535154166222128,
+      "grad_norm": 4.319986820220947,
+      "learning_rate": 2.840909090909091e-07,
+      "loss": 0.983,
+      "step": 10
+    },
+    {
+      "epoch": 0.017070308332444255,
+      "grad_norm": 3.9372925758361816,
+      "learning_rate": 5.681818181818182e-07,
+      "loss": 0.9573,
+      "step": 20
+    },
+    {
+      "epoch": 0.02560546249866638,
+      "grad_norm": 3.070388078689575,
+      "learning_rate": 8.522727272727273e-07,
+      "loss": 0.9306,
+      "step": 30
+    },
+    {
+      "epoch": 0.03414061666488851,
+      "grad_norm": 1.9363492727279663,
+      "learning_rate": 1.1363636363636364e-06,
+      "loss": 0.8481,
+      "step": 40
+    },
+    {
+      "epoch": 0.04267577083111064,
+      "grad_norm": 1.282257318496704,
+      "learning_rate": 1.4204545454545458e-06,
+      "loss": 0.7776,
+      "step": 50
+    },
+    {
+      "epoch": 0.05121092499733276,
+      "grad_norm": 1.0056549310684204,
+      "learning_rate": 1.7045454545454546e-06,
+      "loss": 0.7324,
+      "step": 60
+    },
+    {
+      "epoch": 0.05974607916355489,
+      "grad_norm": 0.8602527379989624,
+      "learning_rate": 1.9886363636363638e-06,
+      "loss": 0.6952,
+      "step": 70
+    },
+    {
+      "epoch": 0.06828123332977702,
+      "grad_norm": 0.822441577911377,
+      "learning_rate": 2.2727272727272728e-06,
+      "loss": 0.6724,
+      "step": 80
+    },
+    {
+      "epoch": 0.07681638749599914,
+      "grad_norm": 0.7585397958755493,
+      "learning_rate": 2.556818181818182e-06,
+      "loss": 0.6635,
+      "step": 90
+    },
+    {
+      "epoch": 0.08535154166222128,
+      "grad_norm": 0.7706250548362732,
+      "learning_rate": 2.8409090909090916e-06,
+      "loss": 0.6502,
+      "step": 100
+    },
+    {
+      "epoch": 0.0938866958284434,
+      "grad_norm": 0.8293640613555908,
+      "learning_rate": 3.125e-06,
+      "loss": 0.6334,
+      "step": 110
+    },
+    {
+      "epoch": 0.10242184999466553,
+      "grad_norm": 0.781350314617157,
+      "learning_rate": 3.409090909090909e-06,
+      "loss": 0.6268,
+      "step": 120
+    },
+    {
+      "epoch": 0.11095700416088766,
+      "grad_norm": 0.7660292387008667,
+      "learning_rate": 3.6931818181818186e-06,
+      "loss": 0.6137,
+      "step": 130
+    },
+    {
+      "epoch": 0.11949215832710978,
+      "grad_norm": 0.8327275514602661,
+      "learning_rate": 3.9772727272727275e-06,
+      "loss": 0.6159,
+      "step": 140
+    },
+    {
+      "epoch": 0.1280273124933319,
+      "grad_norm": 0.8584268093109131,
+      "learning_rate": 4.2613636363636365e-06,
+      "loss": 0.5995,
+      "step": 150
+    },
+    {
+      "epoch": 0.13656246665955404,
+      "grad_norm": 0.7809672951698303,
+      "learning_rate": 4.5454545454545455e-06,
+      "loss": 0.6043,
+      "step": 160
+    },
+    {
+      "epoch": 0.14509762082577615,
+      "grad_norm": 0.8208410739898682,
+      "learning_rate": 4.829545454545455e-06,
+      "loss": 0.6015,
+      "step": 170
+    },
+    {
+      "epoch": 0.1536327749919983,
+      "grad_norm": 0.828789472579956,
+      "learning_rate": 5.113636363636364e-06,
+      "loss": 0.5871,
+      "step": 180
+    },
+    {
+      "epoch": 0.16216792915822043,
+      "grad_norm": 0.7799698114395142,
+      "learning_rate": 5.397727272727273e-06,
+      "loss": 0.5946,
+      "step": 190
+    },
+    {
+      "epoch": 0.17070308332444256,
+      "grad_norm": 0.7569297552108765,
+      "learning_rate": 5.681818181818183e-06,
+      "loss": 0.5868,
+      "step": 200
+    },
+    {
+      "epoch": 0.17923823749066467,
+      "grad_norm": 0.7858031392097473,
+      "learning_rate": 5.965909090909091e-06,
+      "loss": 0.5732,
+      "step": 210
+    },
+    {
+      "epoch": 0.1877733916568868,
+      "grad_norm": 0.7812625765800476,
+      "learning_rate": 6.25e-06,
+      "loss": 0.5745,
+      "step": 220
+    },
+    {
+      "epoch": 0.19630854582310894,
+      "grad_norm": 0.7922148704528809,
+      "learning_rate": 6.53409090909091e-06,
+      "loss": 0.5742,
+      "step": 230
+    },
+    {
+      "epoch": 0.20484369998933105,
+      "grad_norm": 0.8420217633247375,
+      "learning_rate": 6.818181818181818e-06,
+      "loss": 0.5593,
+      "step": 240
+    },
+    {
+      "epoch": 0.2133788541555532,
+      "grad_norm": 0.8149869441986084,
+      "learning_rate": 7.102272727272727e-06,
+      "loss": 0.5694,
+      "step": 250
+    },
+    {
+      "epoch": 0.22191400832177532,
+      "grad_norm": 0.8200048208236694,
+      "learning_rate": 7.386363636363637e-06,
+      "loss": 0.5609,
+      "step": 260
+    },
+    {
+      "epoch": 0.23044916248799743,
+      "grad_norm": 0.7718726396560669,
+      "learning_rate": 7.670454545454547e-06,
+      "loss": 0.5596,
+      "step": 270
+    },
+    {
+      "epoch": 0.23898431665421957,
+      "grad_norm": 0.8079963326454163,
+      "learning_rate": 7.954545454545455e-06,
+      "loss": 0.5515,
+      "step": 280
+    },
+    {
+      "epoch": 0.2475194708204417,
+      "grad_norm": 0.8398913741111755,
+      "learning_rate": 8.238636363636365e-06,
+      "loss": 0.5615,
+      "step": 290
+    },
+    {
+      "epoch": 0.2560546249866638,
+      "grad_norm": 0.8522515892982483,
+      "learning_rate": 8.522727272727273e-06,
+      "loss": 0.5405,
+      "step": 300
+    },
+    {
+      "epoch": 0.26458977915288595,
+      "grad_norm": 0.965617299079895,
+      "learning_rate": 8.806818181818183e-06,
+      "loss": 0.5438,
+      "step": 310
+    },
+    {
+      "epoch": 0.2731249333191081,
+      "grad_norm": 0.834233283996582,
+      "learning_rate": 9.090909090909091e-06,
+      "loss": 0.5342,
+      "step": 320
+    },
+    {
+      "epoch": 0.2816600874853302,
+      "grad_norm": 0.8867103457450867,
+      "learning_rate": 9.375000000000001e-06,
+      "loss": 0.5272,
+      "step": 330
+    },
+    {
+      "epoch": 0.2901952416515523,
+      "grad_norm": 0.767537534236908,
+      "learning_rate": 9.65909090909091e-06,
+      "loss": 0.5319,
+      "step": 340
+    },
+    {
+      "epoch": 0.29873039581777444,
+      "grad_norm": 0.8806461095809937,
+      "learning_rate": 9.943181818181819e-06,
+      "loss": 0.5322,
+      "step": 350
+    },
+    {
+      "epoch": 0.3072655499839966,
+      "grad_norm": 0.7467280030250549,
+      "learning_rate": 9.999841959480543e-06,
+      "loss": 0.5267,
+      "step": 360
+    },
+    {
+      "epoch": 0.3158007041502187,
+      "grad_norm": 0.7319239974021912,
+      "learning_rate": 9.999199936992976e-06,
+      "loss": 0.5221,
+      "step": 370
+    },
+    {
+      "epoch": 0.32433585831644085,
+      "grad_norm": 0.7590662837028503,
+      "learning_rate": 9.998064118371841e-06,
+      "loss": 0.525,
+      "step": 380
+    },
+    {
+      "epoch": 0.332871012482663,
+      "grad_norm": 0.7705526351928711,
+      "learning_rate": 9.996434615807662e-06,
+      "loss": 0.5254,
+      "step": 390
+    },
+    {
+      "epoch": 0.3414061666488851,
+      "grad_norm": 0.8061032295227051,
+      "learning_rate": 9.994311590254603e-06,
+      "loss": 0.5221,
+      "step": 400
+    },
+    {
+      "epoch": 0.3499413208151072,
+      "grad_norm": 0.7729495167732239,
+      "learning_rate": 9.991695251414584e-06,
+      "loss": 0.53,
+      "step": 410
+    },
+    {
+      "epoch": 0.35847647498132934,
+      "grad_norm": 0.768975019454956,
+      "learning_rate": 9.988585857716555e-06,
+      "loss": 0.5099,
+      "step": 420
+    },
+    {
+      "epoch": 0.3670116291475515,
+      "grad_norm": 0.7381827235221863,
+      "learning_rate": 9.984983716290981e-06,
+      "loss": 0.5182,
+      "step": 430
+    },
+    {
+      "epoch": 0.3755467833137736,
+      "grad_norm": 0.8239282965660095,
+      "learning_rate": 9.980889182939495e-06,
+      "loss": 0.52,
+      "step": 440
+    },
+    {
+      "epoch": 0.38408193747999575,
+      "grad_norm": 0.8047374486923218,
+      "learning_rate": 9.976302662099758e-06,
+      "loss": 0.5091,
+      "step": 450
+    },
+    {
+      "epoch": 0.3926170916462179,
+      "grad_norm": 0.7568238377571106,
+      "learning_rate": 9.971224606805514e-06,
+      "loss": 0.5171,
+      "step": 460
+    },
+    {
+      "epoch": 0.40115224581243997,
+      "grad_norm": 0.7178462147712708,
+      "learning_rate": 9.965655518641837e-06,
+      "loss": 0.5119,
+      "step": 470
+    },
+    {
+      "epoch": 0.4096873999786621,
+      "grad_norm": 0.8041744232177734,
+      "learning_rate": 9.959595947695588e-06,
+      "loss": 0.506,
+      "step": 480
+    },
+    {
+      "epoch": 0.41822255414488424,
+      "grad_norm": 0.7083892226219177,
+      "learning_rate": 9.953046492501082e-06,
+      "loss": 0.5066,
+      "step": 490
+    },
+    {
+      "epoch": 0.4267577083111064,
+      "grad_norm": 0.7124646902084351,
+      "learning_rate": 9.946007799980963e-06,
+      "loss": 0.5098,
+      "step": 500
+    },
+    {
+      "epoch": 0.4352928624773285,
+      "grad_norm": 0.7167072892189026,
+      "learning_rate": 9.938480565382312e-06,
+      "loss": 0.513,
+      "step": 510
+    },
+    {
+      "epoch": 0.44382801664355065,
+      "grad_norm": 0.7211044430732727,
+      "learning_rate": 9.930465532207967e-06,
+      "loss": 0.5094,
+      "step": 520
+    },
+    {
+      "epoch": 0.45236317080977273,
+      "grad_norm": 0.7868006229400635,
+      "learning_rate": 9.921963492143083e-06,
+      "loss": 0.5131,
+      "step": 530
+    },
+    {
+      "epoch": 0.46089832497599487,
+      "grad_norm": 0.7130579948425293,
+      "learning_rate": 9.912975284976942e-06,
+      "loss": 0.506,
+      "step": 540
+    },
+    {
+      "epoch": 0.469433479142217,
+      "grad_norm": 0.7939994931221008,
+      "learning_rate": 9.903501798519994e-06,
+      "loss": 0.5025,
+      "step": 550
+    },
+    {
+      "epoch": 0.47796863330843914,
+      "grad_norm": 0.7138738036155701,
+      "learning_rate": 9.893543968516166e-06,
+      "loss": 0.5037,
+      "step": 560
+    },
+    {
+      "epoch": 0.4865037874746613,
+      "grad_norm": 0.716701328754425,
+      "learning_rate": 9.883102778550434e-06,
+      "loss": 0.5034,
+      "step": 570
+    },
+    {
+      "epoch": 0.4950389416408834,
+      "grad_norm": 0.7295582294464111,
+      "learning_rate": 9.872179259951672e-06,
+      "loss": 0.4949,
+      "step": 580
+    },
+    {
+      "epoch": 0.5035740958071055,
+      "grad_norm": 0.7438467144966125,
+      "learning_rate": 9.860774491690777e-06,
+      "loss": 0.4969,
+      "step": 590
+    },
+    {
+      "epoch": 0.5121092499733276,
+      "grad_norm": 0.696615993976593,
+      "learning_rate": 9.848889600274101e-06,
+      "loss": 0.4926,
+      "step": 600
+    },
+    {
+      "epoch": 0.5206444041395498,
+      "grad_norm": 0.6948277354240417,
+      "learning_rate": 9.836525759632171e-06,
+      "loss": 0.5034,
+      "step": 610
+    },
+    {
+      "epoch": 0.5291795583057719,
+      "grad_norm": 0.8091464042663574,
+      "learning_rate": 9.823684191003744e-06,
+      "loss": 0.5015,
+      "step": 620
+    },
+    {
+      "epoch": 0.537714712471994,
+      "grad_norm": 0.7501440048217773,
+      "learning_rate": 9.810366162815171e-06,
+      "loss": 0.5024,
+      "step": 630
+    },
+    {
+      "epoch": 0.5462498666382162,
+      "grad_norm": 0.7048625946044922,
+      "learning_rate": 9.796572990555112e-06,
+      "loss": 0.4955,
+      "step": 640
+    },
+    {
+      "epoch": 0.5547850208044383,
+      "grad_norm": 0.7141663432121277,
+      "learning_rate": 9.782306036644596e-06,
+      "loss": 0.5045,
+      "step": 650
+    },
+    {
+      "epoch": 0.5633201749706604,
+      "grad_norm": 0.7329632639884949,
+      "learning_rate": 9.767566710302453e-06,
+      "loss": 0.5038,
+      "step": 660
+    },
+    {
+      "epoch": 0.5718553291368825,
+      "grad_norm": 0.699938952922821,
+      "learning_rate": 9.752356467406114e-06,
+      "loss": 0.5066,
+      "step": 670
+    },
+    {
+      "epoch": 0.5803904833031046,
+      "grad_norm": 0.6912060976028442,
+      "learning_rate": 9.736676810347807e-06,
+      "loss": 0.4918,
+      "step": 680
+    },
+    {
+      "epoch": 0.5889256374693268,
+      "grad_norm": 0.6718906164169312,
+      "learning_rate": 9.720529287886159e-06,
+      "loss": 0.4932,
+      "step": 690
+    },
+    {
+      "epoch": 0.5974607916355489,
+      "grad_norm": 0.6830474734306335,
+      "learning_rate": 9.703915494993215e-06,
+      "loss": 0.489,
+      "step": 700
+    },
+    {
+      "epoch": 0.6059959458017711,
+      "grad_norm": 0.6920751333236694,
+      "learning_rate": 9.686837072696902e-06,
+      "loss": 0.4976,
+      "step": 710
+    },
+    {
+      "epoch": 0.6145310999679932,
+      "grad_norm": 0.695093035697937,
+      "learning_rate": 9.669295707918928e-06,
+      "loss": 0.4877,
+      "step": 720
+    },
+    {
+      "epoch": 0.6230662541342153,
+      "grad_norm": 0.683686375617981,
+      "learning_rate": 9.651293133308159e-06,
+      "loss": 0.4939,
+      "step": 730
+    },
+    {
+      "epoch": 0.6316014083004374,
+      "grad_norm": 0.7766408920288086,
+      "learning_rate": 9.632831127069479e-06,
+      "loss": 0.4993,
+      "step": 740
+    },
+    {
+      "epoch": 0.6401365624666595,
+      "grad_norm": 0.7032186985015869,
+      "learning_rate": 9.61391151278814e-06,
+      "loss": 0.4896,
+      "step": 750
+    },
+    {
+      "epoch": 0.6486717166328817,
+      "grad_norm": 0.7620243430137634,
+      "learning_rate": 9.594536159249654e-06,
+      "loss": 0.4926,
+      "step": 760
+    },
+    {
+      "epoch": 0.6572068707991038,
+      "grad_norm": 0.6803762316703796,
+      "learning_rate": 9.57470698025518e-06,
+      "loss": 0.4902,
+      "step": 770
+    },
+    {
+      "epoch": 0.665742024965326,
+      "grad_norm": 0.6878206133842468,
+      "learning_rate": 9.55442593443251e-06,
+      "loss": 0.4841,
+      "step": 780
+    },
+    {
+      "epoch": 0.674277179131548,
+      "grad_norm": 0.7530574202537537,
+      "learning_rate": 9.533695025042591e-06,
+      "loss": 0.4839,
+      "step": 790
+    },
+    {
+      "epoch": 0.6828123332977702,
+      "grad_norm": 0.7192838191986084,
+      "learning_rate": 9.512516299781665e-06,
+      "loss": 0.4841,
+      "step": 800
+    },
+    {
+      "epoch": 0.6913474874639923,
+      "grad_norm": 0.7086024880409241,
+      "learning_rate": 9.490891850578993e-06,
+      "loss": 0.4869,
+      "step": 810
+    },
+    {
+      "epoch": 0.6998826416302144,
+      "grad_norm": 0.6931254267692566,
+      "learning_rate": 9.468823813390236e-06,
+      "loss": 0.4834,
+      "step": 820
+    },
+    {
+      "epoch": 0.7084177957964366,
+      "grad_norm": 0.6703010201454163,
+      "learning_rate": 9.446314367986476e-06,
+      "loss": 0.4904,
+      "step": 830
+    },
+    {
+      "epoch": 0.7169529499626587,
+      "grad_norm": 0.721376359462738,
+      "learning_rate": 9.423365737738898e-06,
+      "loss": 0.4896,
+      "step": 840
+    },
+    {
+      "epoch": 0.7254881041288809,
+      "grad_norm": 0.7349815964698792,
+      "learning_rate": 9.399980189399191e-06,
+      "loss": 0.4938,
+      "step": 850
+    },
+    {
+      "epoch": 0.734023258295103,
+      "grad_norm": 0.7041159272193909,
+      "learning_rate": 9.376160032875638e-06,
+      "loss": 0.4889,
+      "step": 860
+    },
+    {
+      "epoch": 0.742558412461325,
+      "grad_norm": 0.7001343965530396,
+      "learning_rate": 9.351907621004963e-06,
+      "loss": 0.4908,
+      "step": 870
+    },
+    {
+      "epoch": 0.7510935666275472,
+      "grad_norm": 0.7233178019523621,
+      "learning_rate": 9.327225349319922e-06,
+      "loss": 0.4856,
+      "step": 880
+    },
+    {
+      "epoch": 0.7596287207937693,
+      "grad_norm": 0.6933467984199524,
+      "learning_rate": 9.30211565581269e-06,
+      "loss": 0.4817,
+      "step": 890
+    },
+    {
+      "epoch": 0.7681638749599915,
+      "grad_norm": 0.6627389192581177,
+      "learning_rate": 9.276581020694048e-06,
+      "loss": 0.4804,
+      "step": 900
+    },
+    {
+      "epoch": 0.7766990291262136,
+      "grad_norm": 0.6689618229866028,
+      "learning_rate": 9.250623966148392e-06,
+      "loss": 0.4854,
+      "step": 910
+    },
+    {
+      "epoch": 0.7852341832924358,
+      "grad_norm": 0.7105383276939392,
+      "learning_rate": 9.224247056084617e-06,
+      "loss": 0.4831,
+      "step": 920
+    },
+    {
+      "epoch": 0.7937693374586579,
+      "grad_norm": 0.7106824517250061,
+      "learning_rate": 9.197452895882857e-06,
+      "loss": 0.4802,
+      "step": 930
+    },
+    {
+      "epoch": 0.8023044916248799,
+      "grad_norm": 0.6922078132629395,
+      "learning_rate": 9.170244132137143e-06,
+      "loss": 0.4854,
+      "step": 940
+    },
+    {
+      "epoch": 0.8108396457911021,
+      "grad_norm": 0.7484738230705261,
+      "learning_rate": 9.142623452393984e-06,
+      "loss": 0.4708,
+      "step": 950
+    },
+    {
+      "epoch": 0.8193747999573242,
+      "grad_norm": 0.7449445128440857,
+      "learning_rate": 9.114593584886902e-06,
+      "loss": 0.4786,
+      "step": 960
+    },
+    {
+      "epoch": 0.8279099541235464,
+      "grad_norm": 0.6751102209091187,
+      "learning_rate": 9.086157298266957e-06,
+      "loss": 0.4783,
+      "step": 970
+    },
+    {
+      "epoch": 0.8364451082897685,
+      "grad_norm": 0.7621346712112427,
+      "learning_rate": 9.057317401329271e-06,
+      "loss": 0.4868,
+      "step": 980
+    },
+    {
+      "epoch": 0.8449802624559906,
+      "grad_norm": 0.7197836637496948,
+      "learning_rate": 9.028076742735583e-06,
+      "loss": 0.479,
+      "step": 990
+    },
+    {
+      "epoch": 0.8535154166222128,
+      "grad_norm": 0.7285887598991394,
+      "learning_rate": 8.998438210732882e-06,
+      "loss": 0.4851,
+      "step": 1000
+    },
+    {
+      "epoch": 0.8620505707884348,
+      "grad_norm": 0.7014701962471008,
+      "learning_rate": 8.968404732868112e-06,
+      "loss": 0.4703,
+      "step": 1010
+    },
+    {
+      "epoch": 0.870585724954657,
+      "grad_norm": 0.6849453449249268,
+      "learning_rate": 8.937979275699012e-06,
+      "loss": 0.4729,
+      "step": 1020
+    },
+    {
+      "epoch": 0.8791208791208791,
+      "grad_norm": 0.7209568023681641,
+      "learning_rate": 8.907164844501084e-06,
+      "loss": 0.4777,
+      "step": 1030
+    },
+    {
+      "epoch": 0.8876560332871013,
+      "grad_norm": 0.6904247403144836,
+      "learning_rate": 8.875964482970756e-06,
+      "loss": 0.4731,
+      "step": 1040
+    },
+    {
+      "epoch": 0.8961911874533234,
+      "grad_norm": 0.6657469272613525,
+      "learning_rate": 8.844381272924733e-06,
+      "loss": 0.4773,
+      "step": 1050
+    },
+    {
+      "epoch": 0.9047263416195455,
+      "grad_norm": 0.6649881601333618,
+      "learning_rate": 8.812418333995597e-06,
+      "loss": 0.4659,
+      "step": 1060
+    },
+    {
+      "epoch": 0.9132614957857677,
+      "grad_norm": 0.6824942231178284,
+      "learning_rate": 8.780078823323667e-06,
+      "loss": 0.4736,
+      "step": 1070
+    },
+    {
+      "epoch": 0.9217966499519897,
+      "grad_norm": 0.6798464059829712,
+      "learning_rate": 8.74736593524514e-06,
+      "loss": 0.4695,
+      "step": 1080
+    },
+    {
+      "epoch": 0.9303318041182119,
+      "grad_norm": 0.6828529238700867,
+      "learning_rate": 8.71428290097658e-06,
+      "loss": 0.4678,
+      "step": 1090
+    },
+    {
+      "epoch": 0.938866958284434,
+      "grad_norm": 0.6548129916191101,
+      "learning_rate": 8.680832988295757e-06,
+      "loss": 0.4748,
+      "step": 1100
+    },
+    {
+      "epoch": 0.9474021124506561,
+      "grad_norm": 0.7184150815010071,
+      "learning_rate": 8.647019501218862e-06,
+      "loss": 0.4851,
+      "step": 1110
+    },
+    {
+      "epoch": 0.9559372666168783,
+      "grad_norm": 0.6707058548927307,
+      "learning_rate": 8.612845779674166e-06,
+      "loss": 0.4621,
+      "step": 1120
+    },
+    {
+      "epoch": 0.9644724207831004,
+      "grad_norm": 0.6430131793022156,
+      "learning_rate": 8.578315199172105e-06,
+      "loss": 0.4733,
+      "step": 1130
+    },
+    {
+      "epoch": 0.9730075749493226,
+      "grad_norm": 0.652946949005127,
+      "learning_rate": 8.543431170471872e-06,
+      "loss": 0.469,
+      "step": 1140
+    },
+    {
+      "epoch": 0.9815427291155446,
+      "grad_norm": 0.678126871585846,
+      "learning_rate": 8.508197139244515e-06,
+      "loss": 0.4645,
+      "step": 1150
+    },
+    {
+      "epoch": 0.9900778832817668,
+      "grad_norm": 0.6810508966445923,
+      "learning_rate": 8.472616585732597e-06,
+      "loss": 0.4722,
+      "step": 1160
+    },
+    {
+      "epoch": 0.9986130374479889,
+      "grad_norm": 0.6804456114768982,
+      "learning_rate": 8.436693024406427e-06,
+      "loss": 0.4796,
+      "step": 1170
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 3513,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 3,
+  "save_steps": 500.0,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.729895927368909e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}

checkpoint-2344/optimizer_0/.metadata

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:de8fe0e298c50a180f372bb8baa7b6bbc8853af1a69ca9296a6e416515a95a83
+size 342948

checkpoint-2344/optimizer_0/__0_0.distcp

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a61391225763d87ad1bb0ed7098ec0a63afedd99757c7271469db6f70ff3bd87
+size 2472090861

checkpoint-2344/optimizer_0/__1_0.distcp

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6aa81b0b9948117913ea7ae8b6414b18370d9bd8be462198a4581d7b3832be48
+size 2472174123

checkpoint-2344/optimizer_0/__2_0.distcp

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:835dca2ffff9b9a6533c97b9abaf08a58a4a30f275bdc8b1715682b9c0f07f41
+size 2472182518

checkpoint-2344/optimizer_0/__3_0.distcp

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d93dbbc311bc2aec76d7fb6627cc84b84cb806a352445309b8dd75fb94feb444
+size 2472179190

checkpoint-2344/pytorch_model_fsdp_0/.metadata

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1becd24b4d9ba4e39e1ca5a165fa89c6f308a4f52bcc5fc6c80868e344c0d0d9
+size 129152

checkpoint-2344/pytorch_model_fsdp_0/__0_0.distcp

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7d82e43cdc5bc33e9193e583359f6c76211ac44502c70a4047f9537d29ce0167
+size 1498714507

checkpoint-2344/pytorch_model_fsdp_0/__1_0.distcp

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:45662ceec43e95a7eabc5cb376ad1dc392a5e7a3fb71330695a25289a9a6dc17
+size 1498714507

checkpoint-2344/pytorch_model_fsdp_0/__2_0.distcp

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e7105517998f94a9a141da96174d9822c1bf15f18d3f68249236f1e894ce57c7
+size 1498714507

checkpoint-2344/pytorch_model_fsdp_0/__3_0.distcp

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:598e38766de9736d2d7f0b08ce94c5df7a71b14c84bf9307117623d7506ed029
+size 1498714507

checkpoint-2344/rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9eee49a2ad561e75c5f0f3e8056308fc27088e19e19620813602767a759d8e5a
+size 15429

checkpoint-2344/rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e34c5da13614815d740bb91790974fffad60cd4e1e61540d824668f639ceb8c4
+size 15429

checkpoint-2344/rng_state_2.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:84cf36eb77e76f4d60fcc92f3755828769fe8aab6a2ca29f9573827e93d0c9f5
+size 15429

checkpoint-2344/rng_state_3.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e5001f16e2075ef4f6a1c0dbc75d9121e1906a36d5ad35279fed02e36a6c7d13
+size 15429

checkpoint-2344/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9b14df43ab6e329ad8793a3e7456044dba5f3aec1d8556442e9967ec4fbed219
+size 1465

checkpoint-2344/trainer_state.json

@@ -0,0 +1,1686 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 2.0,
+  "eval_steps": 500,
+  "global_step": 2344,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0,
+      "eval_loss": 2.1641297340393066,
+      "eval_runtime": 1.9384,
+      "eval_samples_per_second": 10.318,
+      "eval_steps_per_second": 1.032,
+      "step": 0
+    },
+    {
+      "epoch": 0.0008535154166222128,
+      "grad_norm": 4.246492385864258,
+      "learning_rate": 2.840909090909091e-08,
+      "loss": 1.0032,
+      "step": 1
+    },
+    {
+      "epoch": 0.008535154166222128,
+      "grad_norm": 4.319986820220947,
+      "learning_rate": 2.840909090909091e-07,
+      "loss": 0.983,
+      "step": 10
+    },
+    {
+      "epoch": 0.017070308332444255,
+      "grad_norm": 3.9372925758361816,
+      "learning_rate": 5.681818181818182e-07,
+      "loss": 0.9573,
+      "step": 20
+    },
+    {
+      "epoch": 0.02560546249866638,
+      "grad_norm": 3.070388078689575,
+      "learning_rate": 8.522727272727273e-07,
+      "loss": 0.9306,
+      "step": 30
+    },
+    {
+      "epoch": 0.03414061666488851,
+      "grad_norm": 1.9363492727279663,
+      "learning_rate": 1.1363636363636364e-06,
+      "loss": 0.8481,
+      "step": 40
+    },
+    {
+      "epoch": 0.04267577083111064,
+      "grad_norm": 1.282257318496704,
+      "learning_rate": 1.4204545454545458e-06,
+      "loss": 0.7776,
+      "step": 50
+    },
+    {
+      "epoch": 0.05121092499733276,
+      "grad_norm": 1.0056549310684204,
+      "learning_rate": 1.7045454545454546e-06,
+      "loss": 0.7324,
+      "step": 60
+    },
+    {
+      "epoch": 0.05974607916355489,
+      "grad_norm": 0.8602527379989624,
+      "learning_rate": 1.9886363636363638e-06,
+      "loss": 0.6952,
+      "step": 70
+    },
+    {
+      "epoch": 0.06828123332977702,
+      "grad_norm": 0.822441577911377,
+      "learning_rate": 2.2727272727272728e-06,
+      "loss": 0.6724,
+      "step": 80
+    },
+    {
+      "epoch": 0.07681638749599914,
+      "grad_norm": 0.7585397958755493,
+      "learning_rate": 2.556818181818182e-06,
+      "loss": 0.6635,
+      "step": 90
+    },
+    {
+      "epoch": 0.08535154166222128,
+      "grad_norm": 0.7706250548362732,
+      "learning_rate": 2.8409090909090916e-06,
+      "loss": 0.6502,
+      "step": 100
+    },
+    {
+      "epoch": 0.0938866958284434,
+      "grad_norm": 0.8293640613555908,
+      "learning_rate": 3.125e-06,
+      "loss": 0.6334,
+      "step": 110
+    },
+    {
+      "epoch": 0.10242184999466553,
+      "grad_norm": 0.781350314617157,
+      "learning_rate": 3.409090909090909e-06,
+      "loss": 0.6268,
+      "step": 120
+    },
+    {
+      "epoch": 0.11095700416088766,
+      "grad_norm": 0.7660292387008667,
+      "learning_rate": 3.6931818181818186e-06,
+      "loss": 0.6137,
+      "step": 130
+    },
+    {
+      "epoch": 0.11949215832710978,
+      "grad_norm": 0.8327275514602661,
+      "learning_rate": 3.9772727272727275e-06,
+      "loss": 0.6159,
+      "step": 140
+    },
+    {
+      "epoch": 0.1280273124933319,
+      "grad_norm": 0.8584268093109131,
+      "learning_rate": 4.2613636363636365e-06,
+      "loss": 0.5995,
+      "step": 150
+    },
+    {
+      "epoch": 0.13656246665955404,
+      "grad_norm": 0.7809672951698303,
+      "learning_rate": 4.5454545454545455e-06,
+      "loss": 0.6043,
+      "step": 160
+    },
+    {
+      "epoch": 0.14509762082577615,
+      "grad_norm": 0.8208410739898682,
+      "learning_rate": 4.829545454545455e-06,
+      "loss": 0.6015,
+      "step": 170
+    },
+    {
+      "epoch": 0.1536327749919983,
+      "grad_norm": 0.828789472579956,
+      "learning_rate": 5.113636363636364e-06,
+      "loss": 0.5871,
+      "step": 180
+    },
+    {
+      "epoch": 0.16216792915822043,
+      "grad_norm": 0.7799698114395142,
+      "learning_rate": 5.397727272727273e-06,
+      "loss": 0.5946,
+      "step": 190
+    },
+    {
+      "epoch": 0.17070308332444256,
+      "grad_norm": 0.7569297552108765,
+      "learning_rate": 5.681818181818183e-06,
+      "loss": 0.5868,
+      "step": 200
+    },
+    {
+      "epoch": 0.17923823749066467,
+      "grad_norm": 0.7858031392097473,
+      "learning_rate": 5.965909090909091e-06,
+      "loss": 0.5732,
+      "step": 210
+    },
+    {
+      "epoch": 0.1877733916568868,
+      "grad_norm": 0.7812625765800476,
+      "learning_rate": 6.25e-06,
+      "loss": 0.5745,
+      "step": 220
+    },
+    {
+      "epoch": 0.19630854582310894,
+      "grad_norm": 0.7922148704528809,
+      "learning_rate": 6.53409090909091e-06,
+      "loss": 0.5742,
+      "step": 230
+    },
+    {
+      "epoch": 0.20484369998933105,
+      "grad_norm": 0.8420217633247375,
+      "learning_rate": 6.818181818181818e-06,
+      "loss": 0.5593,
+      "step": 240
+    },
+    {
+      "epoch": 0.2133788541555532,
+      "grad_norm": 0.8149869441986084,
+      "learning_rate": 7.102272727272727e-06,
+      "loss": 0.5694,
+      "step": 250
+    },
+    {
+      "epoch": 0.22191400832177532,
+      "grad_norm": 0.8200048208236694,
+      "learning_rate": 7.386363636363637e-06,
+      "loss": 0.5609,
+      "step": 260
+    },
+    {
+      "epoch": 0.23044916248799743,
+      "grad_norm": 0.7718726396560669,
+      "learning_rate": 7.670454545454547e-06,
+      "loss": 0.5596,
+      "step": 270
+    },
+    {
+      "epoch": 0.23898431665421957,
+      "grad_norm": 0.8079963326454163,
+      "learning_rate": 7.954545454545455e-06,
+      "loss": 0.5515,
+      "step": 280
+    },
+    {
+      "epoch": 0.2475194708204417,
+      "grad_norm": 0.8398913741111755,
+      "learning_rate": 8.238636363636365e-06,
+      "loss": 0.5615,
+      "step": 290
+    },
+    {
+      "epoch": 0.2560546249866638,
+      "grad_norm": 0.8522515892982483,
+      "learning_rate": 8.522727272727273e-06,
+      "loss": 0.5405,
+      "step": 300
+    },
+    {
+      "epoch": 0.26458977915288595,
+      "grad_norm": 0.965617299079895,
+      "learning_rate": 8.806818181818183e-06,
+      "loss": 0.5438,
+      "step": 310
+    },
+    {
+      "epoch": 0.2731249333191081,
+      "grad_norm": 0.834233283996582,
+      "learning_rate": 9.090909090909091e-06,
+      "loss": 0.5342,
+      "step": 320
+    },
+    {
+      "epoch": 0.2816600874853302,
+      "grad_norm": 0.8867103457450867,
+      "learning_rate": 9.375000000000001e-06,
+      "loss": 0.5272,
+      "step": 330
+    },
+    {
+      "epoch": 0.2901952416515523,
+      "grad_norm": 0.767537534236908,
+      "learning_rate": 9.65909090909091e-06,
+      "loss": 0.5319,
+      "step": 340
+    },
+    {
+      "epoch": 0.29873039581777444,
+      "grad_norm": 0.8806461095809937,
+      "learning_rate": 9.943181818181819e-06,
+      "loss": 0.5322,
+      "step": 350
+    },
+    {
+      "epoch": 0.3072655499839966,
+      "grad_norm": 0.7467280030250549,
+      "learning_rate": 9.999841959480543e-06,
+      "loss": 0.5267,
+      "step": 360
+    },
+    {
+      "epoch": 0.3158007041502187,
+      "grad_norm": 0.7319239974021912,
+      "learning_rate": 9.999199936992976e-06,
+      "loss": 0.5221,
+      "step": 370
+    },
+    {
+      "epoch": 0.32433585831644085,
+      "grad_norm": 0.7590662837028503,
+      "learning_rate": 9.998064118371841e-06,
+      "loss": 0.525,
+      "step": 380
+    },
+    {
+      "epoch": 0.332871012482663,
+      "grad_norm": 0.7705526351928711,
+      "learning_rate": 9.996434615807662e-06,
+      "loss": 0.5254,
+      "step": 390
+    },
+    {
+      "epoch": 0.3414061666488851,
+      "grad_norm": 0.8061032295227051,
+      "learning_rate": 9.994311590254603e-06,
+      "loss": 0.5221,
+      "step": 400
+    },
+    {
+      "epoch": 0.3499413208151072,
+      "grad_norm": 0.7729495167732239,
+      "learning_rate": 9.991695251414584e-06,
+      "loss": 0.53,
+      "step": 410
+    },
+    {
+      "epoch": 0.35847647498132934,
+      "grad_norm": 0.768975019454956,
+      "learning_rate": 9.988585857716555e-06,
+      "loss": 0.5099,
+      "step": 420
+    },
+    {
+      "epoch": 0.3670116291475515,
+      "grad_norm": 0.7381827235221863,
+      "learning_rate": 9.984983716290981e-06,
+      "loss": 0.5182,
+      "step": 430
+    },
+    {
+      "epoch": 0.3755467833137736,
+      "grad_norm": 0.8239282965660095,
+      "learning_rate": 9.980889182939495e-06,
+      "loss": 0.52,
+      "step": 440
+    },
+    {
+      "epoch": 0.38408193747999575,
+      "grad_norm": 0.8047374486923218,
+      "learning_rate": 9.976302662099758e-06,
+      "loss": 0.5091,
+      "step": 450
+    },
+    {
+      "epoch": 0.3926170916462179,
+      "grad_norm": 0.7568238377571106,
+      "learning_rate": 9.971224606805514e-06,
+      "loss": 0.5171,
+      "step": 460
+    },
+    {
+      "epoch": 0.40115224581243997,
+      "grad_norm": 0.7178462147712708,
+      "learning_rate": 9.965655518641837e-06,
+      "loss": 0.5119,
+      "step": 470
+    },
+    {
+      "epoch": 0.4096873999786621,
+      "grad_norm": 0.8041744232177734,
+      "learning_rate": 9.959595947695588e-06,
+      "loss": 0.506,
+      "step": 480
+    },
+    {
+      "epoch": 0.41822255414488424,
+      "grad_norm": 0.7083892226219177,
+      "learning_rate": 9.953046492501082e-06,
+      "loss": 0.5066,
+      "step": 490
+    },
+    {
+      "epoch": 0.4267577083111064,
+      "grad_norm": 0.7124646902084351,
+      "learning_rate": 9.946007799980963e-06,
+      "loss": 0.5098,
+      "step": 500
+    },
+    {
+      "epoch": 0.4352928624773285,
+      "grad_norm": 0.7167072892189026,
+      "learning_rate": 9.938480565382312e-06,
+      "loss": 0.513,
+      "step": 510
+    },
+    {
+      "epoch": 0.44382801664355065,
+      "grad_norm": 0.7211044430732727,
+      "learning_rate": 9.930465532207967e-06,
+      "loss": 0.5094,
+      "step": 520
+    },
+    {
+      "epoch": 0.45236317080977273,
+      "grad_norm": 0.7868006229400635,
+      "learning_rate": 9.921963492143083e-06,
+      "loss": 0.5131,
+      "step": 530
+    },
+    {
+      "epoch": 0.46089832497599487,
+      "grad_norm": 0.7130579948425293,
+      "learning_rate": 9.912975284976942e-06,
+      "loss": 0.506,
+      "step": 540
+    },
+    {
+      "epoch": 0.469433479142217,
+      "grad_norm": 0.7939994931221008,
+      "learning_rate": 9.903501798519994e-06,
+      "loss": 0.5025,
+      "step": 550
+    },
+    {
+      "epoch": 0.47796863330843914,
+      "grad_norm": 0.7138738036155701,
+      "learning_rate": 9.893543968516166e-06,
+      "loss": 0.5037,
+      "step": 560
+    },
+    {
+      "epoch": 0.4865037874746613,
+      "grad_norm": 0.716701328754425,
+      "learning_rate": 9.883102778550434e-06,
+      "loss": 0.5034,
+      "step": 570
+    },
+    {
+      "epoch": 0.4950389416408834,
+      "grad_norm": 0.7295582294464111,
+      "learning_rate": 9.872179259951672e-06,
+      "loss": 0.4949,
+      "step": 580
+    },
+    {
+      "epoch": 0.5035740958071055,
+      "grad_norm": 0.7438467144966125,
+      "learning_rate": 9.860774491690777e-06,
+      "loss": 0.4969,
+      "step": 590
+    },
+    {
+      "epoch": 0.5121092499733276,
+      "grad_norm": 0.696615993976593,
+      "learning_rate": 9.848889600274101e-06,
+      "loss": 0.4926,
+      "step": 600
+    },
+    {
+      "epoch": 0.5206444041395498,
+      "grad_norm": 0.6948277354240417,
+      "learning_rate": 9.836525759632171e-06,
+      "loss": 0.5034,
+      "step": 610
+    },
+    {
+      "epoch": 0.5291795583057719,
+      "grad_norm": 0.8091464042663574,
+      "learning_rate": 9.823684191003744e-06,
+      "loss": 0.5015,
+      "step": 620
+    },
+    {
+      "epoch": 0.537714712471994,
+      "grad_norm": 0.7501440048217773,
+      "learning_rate": 9.810366162815171e-06,
+      "loss": 0.5024,
+      "step": 630
+    },
+    {
+      "epoch": 0.5462498666382162,
+      "grad_norm": 0.7048625946044922,
+      "learning_rate": 9.796572990555112e-06,
+      "loss": 0.4955,
+      "step": 640
+    },
+    {
+      "epoch": 0.5547850208044383,
+      "grad_norm": 0.7141663432121277,
+      "learning_rate": 9.782306036644596e-06,
+      "loss": 0.5045,
+      "step": 650
+    },
+    {
+      "epoch": 0.5633201749706604,
+      "grad_norm": 0.7329632639884949,
+      "learning_rate": 9.767566710302453e-06,
+      "loss": 0.5038,
+      "step": 660
+    },
+    {
+      "epoch": 0.5718553291368825,
+      "grad_norm": 0.699938952922821,
+      "learning_rate": 9.752356467406114e-06,
+      "loss": 0.5066,
+      "step": 670
+    },
+    {
+      "epoch": 0.5803904833031046,
+      "grad_norm": 0.6912060976028442,
+      "learning_rate": 9.736676810347807e-06,
+      "loss": 0.4918,
+      "step": 680
+    },
+    {
+      "epoch": 0.5889256374693268,
+      "grad_norm": 0.6718906164169312,
+      "learning_rate": 9.720529287886159e-06,
+      "loss": 0.4932,
+      "step": 690
+    },
+    {
+      "epoch": 0.5974607916355489,
+      "grad_norm": 0.6830474734306335,
+      "learning_rate": 9.703915494993215e-06,
+      "loss": 0.489,
+      "step": 700
+    },
+    {
+      "epoch": 0.6059959458017711,
+      "grad_norm": 0.6920751333236694,
+      "learning_rate": 9.686837072696902e-06,
+      "loss": 0.4976,
+      "step": 710
+    },
+    {
+      "epoch": 0.6145310999679932,
+      "grad_norm": 0.695093035697937,
+      "learning_rate": 9.669295707918928e-06,
+      "loss": 0.4877,
+      "step": 720
+    },
+    {
+      "epoch": 0.6230662541342153,
+      "grad_norm": 0.683686375617981,
+      "learning_rate": 9.651293133308159e-06,
+      "loss": 0.4939,
+      "step": 730
+    },
+    {
+      "epoch": 0.6316014083004374,
+      "grad_norm": 0.7766408920288086,
+      "learning_rate": 9.632831127069479e-06,
+      "loss": 0.4993,
+      "step": 740
+    },
+    {
+      "epoch": 0.6401365624666595,
+      "grad_norm": 0.7032186985015869,
+      "learning_rate": 9.61391151278814e-06,
+      "loss": 0.4896,
+      "step": 750
+    },
+    {
+      "epoch": 0.6486717166328817,
+      "grad_norm": 0.7620243430137634,
+      "learning_rate": 9.594536159249654e-06,
+      "loss": 0.4926,
+      "step": 760
+    },
+    {
+      "epoch": 0.6572068707991038,
+      "grad_norm": 0.6803762316703796,
+      "learning_rate": 9.57470698025518e-06,
+      "loss": 0.4902,
+      "step": 770
+    },
+    {
+      "epoch": 0.665742024965326,
+      "grad_norm": 0.6878206133842468,
+      "learning_rate": 9.55442593443251e-06,
+      "loss": 0.4841,
+      "step": 780
+    },
+    {
+      "epoch": 0.674277179131548,
+      "grad_norm": 0.7530574202537537,
+      "learning_rate": 9.533695025042591e-06,
+      "loss": 0.4839,
+      "step": 790
+    },
+    {
+      "epoch": 0.6828123332977702,
+      "grad_norm": 0.7192838191986084,
+      "learning_rate": 9.512516299781665e-06,
+      "loss": 0.4841,
+      "step": 800
+    },
+    {
+      "epoch": 0.6913474874639923,
+      "grad_norm": 0.7086024880409241,
+      "learning_rate": 9.490891850578993e-06,
+      "loss": 0.4869,
+      "step": 810
+    },
+    {
+      "epoch": 0.6998826416302144,
+      "grad_norm": 0.6931254267692566,
+      "learning_rate": 9.468823813390236e-06,
+      "loss": 0.4834,
+      "step": 820
+    },
+    {
+      "epoch": 0.7084177957964366,
+      "grad_norm": 0.6703010201454163,
+      "learning_rate": 9.446314367986476e-06,
+      "loss": 0.4904,
+      "step": 830
+    },
+    {
+      "epoch": 0.7169529499626587,
+      "grad_norm": 0.721376359462738,
+      "learning_rate": 9.423365737738898e-06,
+      "loss": 0.4896,
+      "step": 840
+    },
+    {
+      "epoch": 0.7254881041288809,
+      "grad_norm": 0.7349815964698792,
+      "learning_rate": 9.399980189399191e-06,
+      "loss": 0.4938,
+      "step": 850
+    },
+    {
+      "epoch": 0.734023258295103,
+      "grad_norm": 0.7041159272193909,
+      "learning_rate": 9.376160032875638e-06,
+      "loss": 0.4889,
+      "step": 860
+    },
+    {
+      "epoch": 0.742558412461325,
+      "grad_norm": 0.7001343965530396,
+      "learning_rate": 9.351907621004963e-06,
+      "loss": 0.4908,
+      "step": 870
+    },
+    {
+      "epoch": 0.7510935666275472,
+      "grad_norm": 0.7233178019523621,
+      "learning_rate": 9.327225349319922e-06,
+      "loss": 0.4856,
+      "step": 880
+    },
+    {
+      "epoch": 0.7596287207937693,
+      "grad_norm": 0.6933467984199524,
+      "learning_rate": 9.30211565581269e-06,
+      "loss": 0.4817,
+      "step": 890
+    },
+    {
+      "epoch": 0.7681638749599915,
+      "grad_norm": 0.6627389192581177,
+      "learning_rate": 9.276581020694048e-06,
+      "loss": 0.4804,
+      "step": 900
+    },
+    {
+      "epoch": 0.7766990291262136,
+      "grad_norm": 0.6689618229866028,
+      "learning_rate": 9.250623966148392e-06,
+      "loss": 0.4854,
+      "step": 910
+    },
+    {
+      "epoch": 0.7852341832924358,
+      "grad_norm": 0.7105383276939392,
+      "learning_rate": 9.224247056084617e-06,
+      "loss": 0.4831,
+      "step": 920
+    },
+    {
+      "epoch": 0.7937693374586579,
+      "grad_norm": 0.7106824517250061,
+      "learning_rate": 9.197452895882857e-06,
+      "loss": 0.4802,
+      "step": 930
+    },
+    {
+      "epoch": 0.8023044916248799,
+      "grad_norm": 0.6922078132629395,
+      "learning_rate": 9.170244132137143e-06,
+      "loss": 0.4854,
+      "step": 940
+    },
+    {
+      "epoch": 0.8108396457911021,
+      "grad_norm": 0.7484738230705261,
+      "learning_rate": 9.142623452393984e-06,
+      "loss": 0.4708,
+      "step": 950
+    },
+    {
+      "epoch": 0.8193747999573242,
+      "grad_norm": 0.7449445128440857,
+      "learning_rate": 9.114593584886902e-06,
+      "loss": 0.4786,
+      "step": 960
+    },
+    {
+      "epoch": 0.8279099541235464,
+      "grad_norm": 0.6751102209091187,
+      "learning_rate": 9.086157298266957e-06,
+      "loss": 0.4783,
+      "step": 970
+    },
+    {
+      "epoch": 0.8364451082897685,
+      "grad_norm": 0.7621346712112427,
+      "learning_rate": 9.057317401329271e-06,
+      "loss": 0.4868,
+      "step": 980
+    },
+    {
+      "epoch": 0.8449802624559906,
+      "grad_norm": 0.7197836637496948,
+      "learning_rate": 9.028076742735583e-06,
+      "loss": 0.479,
+      "step": 990
+    },
+    {
+      "epoch": 0.8535154166222128,
+      "grad_norm": 0.7285887598991394,
+      "learning_rate": 8.998438210732882e-06,
+      "loss": 0.4851,
+      "step": 1000
+    },
+    {
+      "epoch": 0.8620505707884348,
+      "grad_norm": 0.7014701962471008,
+      "learning_rate": 8.968404732868112e-06,
+      "loss": 0.4703,
+      "step": 1010
+    },
+    {
+      "epoch": 0.870585724954657,
+      "grad_norm": 0.6849453449249268,
+      "learning_rate": 8.937979275699012e-06,
+      "loss": 0.4729,
+      "step": 1020
+    },
+    {
+      "epoch": 0.8791208791208791,
+      "grad_norm": 0.7209568023681641,
+      "learning_rate": 8.907164844501084e-06,
+      "loss": 0.4777,
+      "step": 1030
+    },
+    {
+      "epoch": 0.8876560332871013,
+      "grad_norm": 0.6904247403144836,
+      "learning_rate": 8.875964482970756e-06,
+      "loss": 0.4731,
+      "step": 1040
+    },
+    {
+      "epoch": 0.8961911874533234,
+      "grad_norm": 0.6657469272613525,
+      "learning_rate": 8.844381272924733e-06,
+      "loss": 0.4773,
+      "step": 1050
+    },
+    {
+      "epoch": 0.9047263416195455,
+      "grad_norm": 0.6649881601333618,
+      "learning_rate": 8.812418333995597e-06,
+      "loss": 0.4659,
+      "step": 1060
+    },
+    {
+      "epoch": 0.9132614957857677,
+      "grad_norm": 0.6824942231178284,
+      "learning_rate": 8.780078823323667e-06,
+      "loss": 0.4736,
+      "step": 1070
+    },
+    {
+      "epoch": 0.9217966499519897,
+      "grad_norm": 0.6798464059829712,
+      "learning_rate": 8.74736593524514e-06,
+      "loss": 0.4695,
+      "step": 1080
+    },
+    {
+      "epoch": 0.9303318041182119,
+      "grad_norm": 0.6828529238700867,
+      "learning_rate": 8.71428290097658e-06,
+      "loss": 0.4678,
+      "step": 1090
+    },
+    {
+      "epoch": 0.938866958284434,
+      "grad_norm": 0.6548129916191101,
+      "learning_rate": 8.680832988295757e-06,
+      "loss": 0.4748,
+      "step": 1100
+    },
+    {
+      "epoch": 0.9474021124506561,
+      "grad_norm": 0.7184150815010071,
+      "learning_rate": 8.647019501218862e-06,
+      "loss": 0.4851,
+      "step": 1110
+    },
+    {
+      "epoch": 0.9559372666168783,
+      "grad_norm": 0.6707058548927307,
+      "learning_rate": 8.612845779674166e-06,
+      "loss": 0.4621,
+      "step": 1120
+    },
+    {
+      "epoch": 0.9644724207831004,
+      "grad_norm": 0.6430131793022156,
+      "learning_rate": 8.578315199172105e-06,
+      "loss": 0.4733,
+      "step": 1130
+    },
+    {
+      "epoch": 0.9730075749493226,
+      "grad_norm": 0.652946949005127,
+      "learning_rate": 8.543431170471872e-06,
+      "loss": 0.469,
+      "step": 1140
+    },
+    {
+      "epoch": 0.9815427291155446,
+      "grad_norm": 0.678126871585846,
+      "learning_rate": 8.508197139244515e-06,
+      "loss": 0.4645,
+      "step": 1150
+    },
+    {
+      "epoch": 0.9900778832817668,
+      "grad_norm": 0.6810508966445923,
+      "learning_rate": 8.472616585732597e-06,
+      "loss": 0.4722,
+      "step": 1160
+    },
+    {
+      "epoch": 0.9986130374479889,
+      "grad_norm": 0.6804456114768982,
+      "learning_rate": 8.436693024406427e-06,
+      "loss": 0.4796,
+      "step": 1170
+    },
+    {
+      "epoch": 1.0068281233329777,
+      "grad_norm": 0.7665020227432251,
+      "learning_rate": 8.400430003616924e-06,
+      "loss": 0.4501,
+      "step": 1180
+    },
+    {
+      "epoch": 1.0153632774991999,
+      "grad_norm": 0.7121028304100037,
+      "learning_rate": 8.363831105245125e-06,
+      "loss": 0.4436,
+      "step": 1190
+    },
+    {
+      "epoch": 1.023898431665422,
+      "grad_norm": 0.7494065761566162,
+      "learning_rate": 8.326899944348389e-06,
+      "loss": 0.4459,
+      "step": 1200
+    },
+    {
+      "epoch": 1.032433585831644,
+      "grad_norm": 0.6524202227592468,
+      "learning_rate": 8.289640168803308e-06,
+      "loss": 0.4445,
+      "step": 1210
+    },
+    {
+      "epoch": 1.0409687399978662,
+      "grad_norm": 0.6446959972381592,
+      "learning_rate": 8.25205545894541e-06,
+      "loss": 0.4406,
+      "step": 1220
+    },
+    {
+      "epoch": 1.0495038941640884,
+      "grad_norm": 0.6765570044517517,
+      "learning_rate": 8.214149527205606e-06,
+      "loss": 0.4538,
+      "step": 1230
+    },
+    {
+      "epoch": 1.0580390483303104,
+      "grad_norm": 0.6525077223777771,
+      "learning_rate": 8.17592611774352e-06,
+      "loss": 0.4492,
+      "step": 1240
+    },
+    {
+      "epoch": 1.0665742024965326,
+      "grad_norm": 0.6981220245361328,
+      "learning_rate": 8.13738900607764e-06,
+      "loss": 0.4428,
+      "step": 1250
+    },
+    {
+      "epoch": 1.0751093566627548,
+      "grad_norm": 0.7378373742103577,
+      "learning_rate": 8.098541998712405e-06,
+      "loss": 0.442,
+      "step": 1260
+    },
+    {
+      "epoch": 1.083644510828977,
+      "grad_norm": 0.6806818246841431,
+      "learning_rate": 8.059388932762207e-06,
+      "loss": 0.4428,
+      "step": 1270
+    },
+    {
+      "epoch": 1.092179664995199,
+      "grad_norm": 0.704845130443573,
+      "learning_rate": 8.019933675572389e-06,
+      "loss": 0.4402,
+      "step": 1280
+    },
+    {
+      "epoch": 1.100714819161421,
+      "grad_norm": 0.8347475528717041,
+      "learning_rate": 7.980180124337239e-06,
+      "loss": 0.448,
+      "step": 1290
+    },
+    {
+      "epoch": 1.1092499733276433,
+      "grad_norm": 0.7030121684074402,
+      "learning_rate": 7.940132205715053e-06,
+      "loss": 0.4462,
+      "step": 1300
+    },
+    {
+      "epoch": 1.1177851274938653,
+      "grad_norm": 0.6647533178329468,
+      "learning_rate": 7.89979387544027e-06,
+      "loss": 0.4481,
+      "step": 1310
+    },
+    {
+      "epoch": 1.1263202816600875,
+      "grad_norm": 0.724305272102356,
+      "learning_rate": 7.859169117932752e-06,
+      "loss": 0.4426,
+      "step": 1320
+    },
+    {
+      "epoch": 1.1348554358263097,
+      "grad_norm": 0.6487372517585754,
+      "learning_rate": 7.818261945904226e-06,
+      "loss": 0.4442,
+      "step": 1330
+    },
+    {
+      "epoch": 1.1433905899925318,
+      "grad_norm": 0.7011922597885132,
+      "learning_rate": 7.777076399961912e-06,
+      "loss": 0.4486,
+      "step": 1340
+    },
+    {
+      "epoch": 1.1519257441587538,
+      "grad_norm": 0.6805099844932556,
+      "learning_rate": 7.735616548209425e-06,
+      "loss": 0.4469,
+      "step": 1350
+    },
+    {
+      "epoch": 1.160460898324976,
+      "grad_norm": 0.683447003364563,
+      "learning_rate": 7.693886485844955e-06,
+      "loss": 0.4553,
+      "step": 1360
+    },
+    {
+      "epoch": 1.1689960524911982,
+      "grad_norm": 0.8037645816802979,
+      "learning_rate": 7.651890334756736e-06,
+      "loss": 0.4387,
+      "step": 1370
+    },
+    {
+      "epoch": 1.1775312066574202,
+      "grad_norm": 0.668209433555603,
+      "learning_rate": 7.609632243115941e-06,
+      "loss": 0.4373,
+      "step": 1380
+    },
+    {
+      "epoch": 1.1860663608236424,
+      "grad_norm": 0.7479778528213501,
+      "learning_rate": 7.567116384966918e-06,
+      "loss": 0.4525,
+      "step": 1390
+    },
+    {
+      "epoch": 1.1946015149898646,
+      "grad_norm": 0.743675172328949,
+      "learning_rate": 7.524346959814918e-06,
+      "loss": 0.4458,
+      "step": 1400
+    },
+    {
+      "epoch": 1.2031366691560867,
+      "grad_norm": 0.6677897572517395,
+      "learning_rate": 7.481328192211278e-06,
+      "loss": 0.4521,
+      "step": 1410
+    },
+    {
+      "epoch": 1.2116718233223087,
+      "grad_norm": 0.655086100101471,
+      "learning_rate": 7.4380643313361385e-06,
+      "loss": 0.4439,
+      "step": 1420
+    },
+    {
+      "epoch": 1.220206977488531,
+      "grad_norm": 0.7043260335922241,
+      "learning_rate": 7.394559650578746e-06,
+      "loss": 0.4387,
+      "step": 1430
+    },
+    {
+      "epoch": 1.228742131654753,
+      "grad_norm": 0.6985997557640076,
+      "learning_rate": 7.3508184471153274e-06,
+      "loss": 0.4396,
+      "step": 1440
+    },
+    {
+      "epoch": 1.237277285820975,
+      "grad_norm": 0.6879217028617859,
+      "learning_rate": 7.306845041484656e-06,
+      "loss": 0.4438,
+      "step": 1450
+    },
+    {
+      "epoch": 1.2458124399871973,
+      "grad_norm": 0.6632692217826843,
+      "learning_rate": 7.262643777161279e-06,
+      "loss": 0.4421,
+      "step": 1460
+    },
+    {
+      "epoch": 1.2543475941534195,
+      "grad_norm": 0.6956544518470764,
+      "learning_rate": 7.218219020126491e-06,
+      "loss": 0.45,
+      "step": 1470
+    },
+    {
+      "epoch": 1.2628827483196416,
+      "grad_norm": 0.6400607824325562,
+      "learning_rate": 7.1735751584370894e-06,
+      "loss": 0.4372,
+      "step": 1480
+    },
+    {
+      "epoch": 1.2714179024858636,
+      "grad_norm": 0.735914945602417,
+      "learning_rate": 7.128716601791938e-06,
+      "loss": 0.4459,
+      "step": 1490
+    },
+    {
+      "epoch": 1.2799530566520858,
+      "grad_norm": 0.7256124019622803,
+      "learning_rate": 7.083647781096404e-06,
+      "loss": 0.4437,
+      "step": 1500
+    },
+    {
+      "epoch": 1.2884882108183078,
+      "grad_norm": 0.6894016861915588,
+      "learning_rate": 7.038373148024693e-06,
+      "loss": 0.4468,
+      "step": 1510
+    },
+    {
+      "epoch": 1.29702336498453,
+      "grad_norm": 0.676726222038269,
+      "learning_rate": 6.992897174580133e-06,
+      "loss": 0.4409,
+      "step": 1520
+    },
+    {
+      "epoch": 1.3055585191507522,
+      "grad_norm": 0.7112148404121399,
+      "learning_rate": 6.947224352653454e-06,
+      "loss": 0.4387,
+      "step": 1530
+    },
+    {
+      "epoch": 1.3140936733169744,
+      "grad_norm": 0.6504147052764893,
+      "learning_rate": 6.901359193579104e-06,
+      "loss": 0.4357,
+      "step": 1540
+    },
+    {
+      "epoch": 1.3226288274831965,
+      "grad_norm": 0.6736079454421997,
+      "learning_rate": 6.855306227689634e-06,
+      "loss": 0.4403,
+      "step": 1550
+    },
+    {
+      "epoch": 1.3311639816494185,
+      "grad_norm": 0.6631302237510681,
+      "learning_rate": 6.809070003868221e-06,
+      "loss": 0.4368,
+      "step": 1560
+    },
+    {
+      "epoch": 1.3396991358156407,
+      "grad_norm": 0.6592603921890259,
+      "learning_rate": 6.762655089099353e-06,
+      "loss": 0.446,
+      "step": 1570
+    },
+    {
+      "epoch": 1.3482342899818627,
+      "grad_norm": 0.6875331997871399,
+      "learning_rate": 6.716066068017722e-06,
+      "loss": 0.4459,
+      "step": 1580
+    },
+    {
+      "epoch": 1.3567694441480849,
+      "grad_norm": 0.8026809096336365,
+      "learning_rate": 6.669307542455369e-06,
+      "loss": 0.4363,
+      "step": 1590
+    },
+    {
+      "epoch": 1.365304598314307,
+      "grad_norm": 0.8665246367454529,
+      "learning_rate": 6.6223841309871535e-06,
+      "loss": 0.4415,
+      "step": 1600
+    },
+    {
+      "epoch": 1.3738397524805293,
+      "grad_norm": 0.7898995280265808,
+      "learning_rate": 6.575300468474546e-06,
+      "loss": 0.4411,
+      "step": 1610
+    },
+    {
+      "epoch": 1.3823749066467514,
+      "grad_norm": 0.6701722145080566,
+      "learning_rate": 6.52806120560781e-06,
+      "loss": 0.441,
+      "step": 1620
+    },
+    {
+      "epoch": 1.3909100608129734,
+      "grad_norm": 0.6245713233947754,
+      "learning_rate": 6.480671008446651e-06,
+      "loss": 0.4357,
+      "step": 1630
+    },
+    {
+      "epoch": 1.3994452149791956,
+      "grad_norm": 0.6475566029548645,
+      "learning_rate": 6.433134557959303e-06,
+      "loss": 0.4487,
+      "step": 1640
+    },
+    {
+      "epoch": 1.4079803691454176,
+      "grad_norm": 0.7050219178199768,
+      "learning_rate": 6.385456549560187e-06,
+      "loss": 0.4488,
+      "step": 1650
+    },
+    {
+      "epoch": 1.4165155233116398,
+      "grad_norm": 0.6392747759819031,
+      "learning_rate": 6.337641692646106e-06,
+      "loss": 0.4453,
+      "step": 1660
+    },
+    {
+      "epoch": 1.425050677477862,
+      "grad_norm": 0.6791578531265259,
+      "learning_rate": 6.2896947101310756e-06,
+      "loss": 0.4406,
+      "step": 1670
+    },
+    {
+      "epoch": 1.4335858316440842,
+      "grad_norm": 0.6261289119720459,
+      "learning_rate": 6.24162033797983e-06,
+      "loss": 0.4364,
+      "step": 1680
+    },
+    {
+      "epoch": 1.4421209858103061,
+      "grad_norm": 0.6886487603187561,
+      "learning_rate": 6.193423324740018e-06,
+      "loss": 0.44,
+      "step": 1690
+    },
+    {
+      "epoch": 1.4506561399765283,
+      "grad_norm": 0.7670247554779053,
+      "learning_rate": 6.1451084310731545e-06,
+      "loss": 0.4348,
+      "step": 1700
+    },
+    {
+      "epoch": 1.4591912941427505,
+      "grad_norm": 0.6851808428764343,
+      "learning_rate": 6.096680429284412e-06,
+      "loss": 0.4358,
+      "step": 1710
+    },
+    {
+      "epoch": 1.4677264483089725,
+      "grad_norm": 0.660381555557251,
+      "learning_rate": 6.0481441028512154e-06,
+      "loss": 0.4419,
+      "step": 1720
+    },
+    {
+      "epoch": 1.4762616024751947,
+      "grad_norm": 0.6627777218818665,
+      "learning_rate": 5.9995042459507555e-06,
+      "loss": 0.4443,
+      "step": 1730
+    },
+    {
+      "epoch": 1.4847967566414169,
+      "grad_norm": 0.7264851331710815,
+      "learning_rate": 5.9507656629864554e-06,
+      "loss": 0.4419,
+      "step": 1740
+    },
+    {
+      "epoch": 1.493331910807639,
+      "grad_norm": 0.6353123784065247,
+      "learning_rate": 5.901933168113405e-06,
+      "loss": 0.4325,
+      "step": 1750
+    },
+    {
+      "epoch": 1.5018670649738612,
+      "grad_norm": 0.7154319882392883,
+      "learning_rate": 5.853011584762846e-06,
+      "loss": 0.4413,
+      "step": 1760
+    },
+    {
+      "epoch": 1.5104022191400832,
+      "grad_norm": 0.678537905216217,
+      "learning_rate": 5.804005745165735e-06,
+      "loss": 0.4418,
+      "step": 1770
+    },
+    {
+      "epoch": 1.5189373733063052,
+      "grad_norm": 0.6648243069648743,
+      "learning_rate": 5.754920489875447e-06,
+      "loss": 0.438,
+      "step": 1780
+    },
+    {
+      "epoch": 1.5274725274725274,
+      "grad_norm": 0.7042447924613953,
+      "learning_rate": 5.705760667289644e-06,
+      "loss": 0.441,
+      "step": 1790
+    },
+    {
+      "epoch": 1.5360076816387496,
+      "grad_norm": 0.6617943048477173,
+      "learning_rate": 5.65653113317137e-06,
+      "loss": 0.4375,
+      "step": 1800
+    },
+    {
+      "epoch": 1.5445428358049718,
+      "grad_norm": 0.6619049310684204,
+      "learning_rate": 5.607236750169431e-06,
+      "loss": 0.4395,
+      "step": 1810
+    },
+    {
+      "epoch": 1.553077989971194,
+      "grad_norm": 0.6774610280990601,
+      "learning_rate": 5.557882387338086e-06,
+      "loss": 0.4417,
+      "step": 1820
+    },
+    {
+      "epoch": 1.5616131441374161,
+      "grad_norm": 0.6360573768615723,
+      "learning_rate": 5.508472919656101e-06,
+      "loss": 0.4346,
+      "step": 1830
+    },
+    {
+      "epoch": 1.5701482983036381,
+      "grad_norm": 0.6723465919494629,
+      "learning_rate": 5.459013227545224e-06,
+      "loss": 0.4387,
+      "step": 1840
+    },
+    {
+      "epoch": 1.57868345246986,
+      "grad_norm": 0.6785566806793213,
+      "learning_rate": 5.4095081963881225e-06,
+      "loss": 0.4364,
+      "step": 1850
+    },
+    {
+      "epoch": 1.5872186066360823,
+      "grad_norm": 0.6658756136894226,
+      "learning_rate": 5.359962716045836e-06,
+      "loss": 0.4302,
+      "step": 1860
+    },
+    {
+      "epoch": 1.5957537608023045,
+      "grad_norm": 0.6586979031562805,
+      "learning_rate": 5.310381680374767e-06,
+      "loss": 0.4386,
+      "step": 1870
+    },
+    {
+      "epoch": 1.6042889149685267,
+      "grad_norm": 0.7203266620635986,
+      "learning_rate": 5.260769986743296e-06,
+      "loss": 0.4377,
+      "step": 1880
+    },
+    {
+      "epoch": 1.6128240691347489,
+      "grad_norm": 0.7455547451972961,
+      "learning_rate": 5.2111325355480506e-06,
+      "loss": 0.4428,
+      "step": 1890
+    },
+    {
+      "epoch": 1.6213592233009708,
+      "grad_norm": 0.6452232599258423,
+      "learning_rate": 5.161474229729858e-06,
+      "loss": 0.4336,
+      "step": 1900
+    },
+    {
+      "epoch": 1.629894377467193,
+      "grad_norm": 0.6487184166908264,
+      "learning_rate": 5.111799974289462e-06,
+      "loss": 0.4426,
+      "step": 1910
+    },
+    {
+      "epoch": 1.638429531633415,
+      "grad_norm": 0.6176945567131042,
+      "learning_rate": 5.062114675803032e-06,
+      "loss": 0.4356,
+      "step": 1920
+    },
+    {
+      "epoch": 1.6469646857996372,
+      "grad_norm": 0.6743866205215454,
+      "learning_rate": 5.012423241937515e-06,
+      "loss": 0.4308,
+      "step": 1930
+    },
+    {
+      "epoch": 1.6554998399658594,
+      "grad_norm": 0.6723053455352783,
+      "learning_rate": 4.962730580965879e-06,
+      "loss": 0.4344,
+      "step": 1940
+    },
+    {
+      "epoch": 1.6640349941320816,
+      "grad_norm": 0.7293416261672974,
+      "learning_rate": 4.913041601282298e-06,
+      "loss": 0.4229,
+      "step": 1950
+    },
+    {
+      "epoch": 1.6725701482983037,
+      "grad_norm": 0.6705866456031799,
+      "learning_rate": 4.863361210917332e-06,
+      "loss": 0.4412,
+      "step": 1960
+    },
+    {
+      "epoch": 1.6811053024645257,
+      "grad_norm": 0.676863968372345,
+      "learning_rate": 4.813694317053122e-06,
+      "loss": 0.4423,
+      "step": 1970
+    },
+    {
+      "epoch": 1.689640456630748,
+      "grad_norm": 0.659705400466919,
+      "learning_rate": 4.764045825538702e-06,
+      "loss": 0.4372,
+      "step": 1980
+    },
+    {
+      "epoch": 1.6981756107969699,
+      "grad_norm": 0.6498283743858337,
+      "learning_rate": 4.714420640405406e-06,
+      "loss": 0.4344,
+      "step": 1990
+    },
+    {
+      "epoch": 1.706710764963192,
+      "grad_norm": 0.6507246494293213,
+      "learning_rate": 4.664823663382482e-06,
+      "loss": 0.4274,
+      "step": 2000
+    },
+    {
+      "epoch": 1.7152459191294143,
+      "grad_norm": 0.705443263053894,
+      "learning_rate": 4.61525979341292e-06,
+      "loss": 0.4365,
+      "step": 2010
+    },
+    {
+      "epoch": 1.7237810732956365,
+      "grad_norm": 0.6291791200637817,
+      "learning_rate": 4.565733926169561e-06,
+      "loss": 0.4354,
+      "step": 2020
+    },
+    {
+      "epoch": 1.7323162274618586,
+      "grad_norm": 0.6487744450569153,
+      "learning_rate": 4.516250953571527e-06,
+      "loss": 0.439,
+      "step": 2030
+    },
+    {
+      "epoch": 1.7408513816280806,
+      "grad_norm": 0.6847457885742188,
+      "learning_rate": 4.466815763301014e-06,
+      "loss": 0.4245,
+      "step": 2040
+    },
+    {
+      "epoch": 1.7493865357943028,
+      "grad_norm": 0.6660692095756531,
+      "learning_rate": 4.417433238320524e-06,
+      "loss": 0.4345,
+      "step": 2050
+    },
+    {
+      "epoch": 1.7579216899605248,
+      "grad_norm": 0.6606684327125549,
+      "learning_rate": 4.36810825639054e-06,
+      "loss": 0.4442,
+      "step": 2060
+    },
+    {
+      "epoch": 1.766456844126747,
+      "grad_norm": 0.7084365487098694,
+      "learning_rate": 4.318845689587727e-06,
+      "loss": 0.4348,
+      "step": 2070
+    },
+    {
+      "epoch": 1.7749919982929692,
+      "grad_norm": 0.6337569952011108,
+      "learning_rate": 4.269650403823698e-06,
+      "loss": 0.4384,
+      "step": 2080
+    },
+    {
+      "epoch": 1.7835271524591914,
+      "grad_norm": 0.6552258133888245,
+      "learning_rate": 4.220527258364373e-06,
+      "loss": 0.4376,
+      "step": 2090
+    },
+    {
+      "epoch": 1.7920623066254135,
+      "grad_norm": 0.6303353309631348,
+      "learning_rate": 4.171481105350021e-06,
+      "loss": 0.4373,
+      "step": 2100
+    },
+    {
+      "epoch": 1.8005974607916355,
+      "grad_norm": 0.650391697883606,
+      "learning_rate": 4.122516789315969e-06,
+      "loss": 0.4437,
+      "step": 2110
+    },
+    {
+      "epoch": 1.8091326149578577,
+      "grad_norm": 0.6542319059371948,
+      "learning_rate": 4.073639146714099e-06,
+      "loss": 0.4262,
+      "step": 2120
+    },
+    {
+      "epoch": 1.8176677691240797,
+      "grad_norm": 0.651104211807251,
+      "learning_rate": 4.024853005435121e-06,
+      "loss": 0.4363,
+      "step": 2130
+    },
+    {
+      "epoch": 1.8262029232903019,
+      "grad_norm": 0.6116424798965454,
+      "learning_rate": 3.976163184331702e-06,
+      "loss": 0.4291,
+      "step": 2140
+    },
+    {
+      "epoch": 1.834738077456524,
+      "grad_norm": 0.8228034377098083,
+      "learning_rate": 3.927574492742473e-06,
+      "loss": 0.4306,
+      "step": 2150
+    },
+    {
+      "epoch": 1.8432732316227463,
+      "grad_norm": 0.6177427768707275,
+      "learning_rate": 3.879091730017003e-06,
+      "loss": 0.4274,
+      "step": 2160
+    },
+    {
+      "epoch": 1.8518083857889684,
+      "grad_norm": 0.6379538178443909,
+      "learning_rate": 3.830719685041728e-06,
+      "loss": 0.4344,
+      "step": 2170
+    },
+    {
+      "epoch": 1.8603435399551904,
+      "grad_norm": 0.6659247279167175,
+      "learning_rate": 3.782463135766938e-06,
+      "loss": 0.4318,
+      "step": 2180
+    },
+    {
+      "epoch": 1.8688786941214126,
+      "grad_norm": 0.6846089959144592,
+      "learning_rate": 3.7343268487348273e-06,
+      "loss": 0.4354,
+      "step": 2190
+    },
+    {
+      "epoch": 1.8774138482876346,
+      "grad_norm": 0.6371617913246155,
+      "learning_rate": 3.6863155786086886e-06,
+      "loss": 0.4302,
+      "step": 2200
+    },
+    {
+      "epoch": 1.8859490024538568,
+      "grad_norm": 0.6230723857879639,
+      "learning_rate": 3.6384340677032616e-06,
+      "loss": 0.4367,
+      "step": 2210
+    },
+    {
+      "epoch": 1.894484156620079,
+      "grad_norm": 0.6282981634140015,
+      "learning_rate": 3.5906870455163175e-06,
+      "loss": 0.4284,
+      "step": 2220
+    },
+    {
+      "epoch": 1.9030193107863012,
+      "grad_norm": 0.6560613512992859,
+      "learning_rate": 3.5430792282614978e-06,
+      "loss": 0.4324,
+      "step": 2230
+    },
+    {
+      "epoch": 1.9115544649525233,
+      "grad_norm": 0.6765207052230835,
+      "learning_rate": 3.4956153184024765e-06,
+      "loss": 0.4261,
+      "step": 2240
+    },
+    {
+      "epoch": 1.9200896191187453,
+      "grad_norm": 0.6762336492538452,
+      "learning_rate": 3.448300004188463e-06,
+      "loss": 0.4304,
+      "step": 2250
+    },
+    {
+      "epoch": 1.9286247732849675,
+      "grad_norm": 0.6394000053405762,
+      "learning_rate": 3.401137959191133e-06,
+      "loss": 0.4294,
+      "step": 2260
+    },
+    {
+      "epoch": 1.9371599274511895,
+      "grad_norm": 0.6332142949104309,
+      "learning_rate": 3.354133841842989e-06,
+      "loss": 0.4308,
+      "step": 2270
+    },
+    {
+      "epoch": 1.9456950816174117,
+      "grad_norm": 0.6541900038719177,
+      "learning_rate": 3.3072922949772235e-06,
+      "loss": 0.4343,
+      "step": 2280
+    },
+    {
+      "epoch": 1.9542302357836339,
+      "grad_norm": 0.652773380279541,
+      "learning_rate": 3.2606179453691255e-06,
+      "loss": 0.4311,
+      "step": 2290
+    },
+    {
+      "epoch": 1.962765389949856,
+      "grad_norm": 0.6701607704162598,
+      "learning_rate": 3.21411540327907e-06,
+      "loss": 0.4244,
+      "step": 2300
+    },
+    {
+      "epoch": 1.9713005441160782,
+      "grad_norm": 0.6590484976768494,
+      "learning_rate": 3.16778926199714e-06,
+      "loss": 0.4366,
+      "step": 2310
+    },
+    {
+      "epoch": 1.9798356982823002,
+      "grad_norm": 0.6363989114761353,
+      "learning_rate": 3.1216440973894184e-06,
+      "loss": 0.4267,
+      "step": 2320
+    },
+    {
+      "epoch": 1.9883708524485222,
+      "grad_norm": 0.6702283024787903,
+      "learning_rate": 3.075684467446014e-06,
+      "loss": 0.4305,
+      "step": 2330
+    },
+    {
+      "epoch": 1.9969060066147444,
+      "grad_norm": 0.6762630939483643,
+      "learning_rate": 3.029914911830846e-06,
+      "loss": 0.4339,
+      "step": 2340
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 3513,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 3,
+  "save_steps": 500.0,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 7.459688578893742e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}

checkpoint-3513/optimizer_0/.metadata

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:64ebca3ef3e81d560c5bdfc1a11e5060851c49ac7ed1fa50b5accc082a3cadf0
+size 342948

checkpoint-3513/optimizer_0/__0_0.distcp

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3d408253066e5d77c637d6dbea1612a78f42cefd5d4a2e56986a71cbecc20e73
+size 2472090861

checkpoint-3513/optimizer_0/__1_0.distcp

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8836e176c8a18a3a48d3d4e894795566efef8990b7b6efb9ee17513d229e0d1f
+size 2472174123

checkpoint-3513/optimizer_0/__2_0.distcp

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:be8853db9ac1f2210862de0c39b535f5298aa98726b2fcd39c691c2303aca2d9
+size 2472182518

checkpoint-3513/optimizer_0/__3_0.distcp

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4f828323f9db2d6ab4896c70f109894b76feafd4ef69faa78b609f258f141641
+size 2472179190

checkpoint-3513/pytorch_model_fsdp_0/.metadata

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2588cfb169e2dfe11ac38292e717db6f116d47852573a7b95bfef96b2d735122
+size 129152

checkpoint-3513/pytorch_model_fsdp_0/__0_0.distcp

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fe8ee71d9922f8f640c19cdcd40ca0418ce1eef0b9b4b37210ba0115023ce65d
+size 1498714507

checkpoint-3513/pytorch_model_fsdp_0/__1_0.distcp

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6ada85394c14a70f6c6cda8be12576502bb093ef1ab5cd9b6702521a46b613bd
+size 1498714507

checkpoint-3513/pytorch_model_fsdp_0/__2_0.distcp

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a74c5bdac352841029679327983cceefa92f910ccf1507abaf98505a9abd9e13
+size 1498714507

checkpoint-3513/pytorch_model_fsdp_0/__3_0.distcp

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5ced202d13c8626e264265e16084fb886e998f0c3e668467f687972f2f9187af
+size 1498714507

checkpoint-3513/rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b80829a78a755ba0cf5e985f9212f164229ce25f7c02fa877b4cdebf230922b5
+size 15429

checkpoint-3513/rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0c6f4074cb5241eff8959c1f45b18c23c893119f6c03fa2da27359dbe7dba66c
+size 15429

checkpoint-3513/rng_state_2.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f1e84026d823a1c04897f710cd119a65b654c08ad443e97c4486fc7a26f6e64b
+size 15429

checkpoint-3513/rng_state_3.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d9b60a41c989678f542cc75416e61e7a43f81d6221d259490134253d1f38a920
+size 15429

checkpoint-3513/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a3d65a72a5d03c0d7fbab19feab674b0f14ec8bddc1521fd78f3f902b2296fd1
+size 1465

checkpoint-3513/trainer_state.json

@@ -0,0 +1,2505 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 2.997759522031367,
+  "eval_steps": 500,
+  "global_step": 3513,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0,
+      "eval_loss": 2.1641297340393066,
+      "eval_runtime": 1.9384,
+      "eval_samples_per_second": 10.318,
+      "eval_steps_per_second": 1.032,
+      "step": 0
+    },
+    {
+      "epoch": 0.0008535154166222128,
+      "grad_norm": 4.246492385864258,
+      "learning_rate": 2.840909090909091e-08,
+      "loss": 1.0032,
+      "step": 1
+    },
+    {
+      "epoch": 0.008535154166222128,
+      "grad_norm": 4.319986820220947,
+      "learning_rate": 2.840909090909091e-07,
+      "loss": 0.983,
+      "step": 10
+    },
+    {
+      "epoch": 0.017070308332444255,
+      "grad_norm": 3.9372925758361816,
+      "learning_rate": 5.681818181818182e-07,
+      "loss": 0.9573,
+      "step": 20
+    },
+    {
+      "epoch": 0.02560546249866638,
+      "grad_norm": 3.070388078689575,
+      "learning_rate": 8.522727272727273e-07,
+      "loss": 0.9306,
+      "step": 30
+    },
+    {
+      "epoch": 0.03414061666488851,
+      "grad_norm": 1.9363492727279663,
+      "learning_rate": 1.1363636363636364e-06,
+      "loss": 0.8481,
+      "step": 40
+    },
+    {
+      "epoch": 0.04267577083111064,
+      "grad_norm": 1.282257318496704,
+      "learning_rate": 1.4204545454545458e-06,
+      "loss": 0.7776,
+      "step": 50
+    },
+    {
+      "epoch": 0.05121092499733276,
+      "grad_norm": 1.0056549310684204,
+      "learning_rate": 1.7045454545454546e-06,
+      "loss": 0.7324,
+      "step": 60
+    },
+    {
+      "epoch": 0.05974607916355489,
+      "grad_norm": 0.8602527379989624,
+      "learning_rate": 1.9886363636363638e-06,
+      "loss": 0.6952,
+      "step": 70
+    },
+    {
+      "epoch": 0.06828123332977702,
+      "grad_norm": 0.822441577911377,
+      "learning_rate": 2.2727272727272728e-06,
+      "loss": 0.6724,
+      "step": 80
+    },
+    {
+      "epoch": 0.07681638749599914,
+      "grad_norm": 0.7585397958755493,
+      "learning_rate": 2.556818181818182e-06,
+      "loss": 0.6635,
+      "step": 90
+    },
+    {
+      "epoch": 0.08535154166222128,
+      "grad_norm": 0.7706250548362732,
+      "learning_rate": 2.8409090909090916e-06,
+      "loss": 0.6502,
+      "step": 100
+    },
+    {
+      "epoch": 0.0938866958284434,
+      "grad_norm": 0.8293640613555908,
+      "learning_rate": 3.125e-06,
+      "loss": 0.6334,
+      "step": 110
+    },
+    {
+      "epoch": 0.10242184999466553,
+      "grad_norm": 0.781350314617157,
+      "learning_rate": 3.409090909090909e-06,
+      "loss": 0.6268,
+      "step": 120
+    },
+    {
+      "epoch": 0.11095700416088766,
+      "grad_norm": 0.7660292387008667,
+      "learning_rate": 3.6931818181818186e-06,
+      "loss": 0.6137,
+      "step": 130
+    },
+    {
+      "epoch": 0.11949215832710978,
+      "grad_norm": 0.8327275514602661,
+      "learning_rate": 3.9772727272727275e-06,
+      "loss": 0.6159,
+      "step": 140
+    },
+    {
+      "epoch": 0.1280273124933319,
+      "grad_norm": 0.8584268093109131,
+      "learning_rate": 4.2613636363636365e-06,
+      "loss": 0.5995,
+      "step": 150
+    },
+    {
+      "epoch": 0.13656246665955404,
+      "grad_norm": 0.7809672951698303,
+      "learning_rate": 4.5454545454545455e-06,
+      "loss": 0.6043,
+      "step": 160
+    },
+    {
+      "epoch": 0.14509762082577615,
+      "grad_norm": 0.8208410739898682,
+      "learning_rate": 4.829545454545455e-06,
+      "loss": 0.6015,
+      "step": 170
+    },
+    {
+      "epoch": 0.1536327749919983,
+      "grad_norm": 0.828789472579956,
+      "learning_rate": 5.113636363636364e-06,
+      "loss": 0.5871,
+      "step": 180
+    },
+    {
+      "epoch": 0.16216792915822043,
+      "grad_norm": 0.7799698114395142,
+      "learning_rate": 5.397727272727273e-06,
+      "loss": 0.5946,
+      "step": 190
+    },
+    {
+      "epoch": 0.17070308332444256,
+      "grad_norm": 0.7569297552108765,
+      "learning_rate": 5.681818181818183e-06,
+      "loss": 0.5868,
+      "step": 200
+    },
+    {
+      "epoch": 0.17923823749066467,
+      "grad_norm": 0.7858031392097473,
+      "learning_rate": 5.965909090909091e-06,
+      "loss": 0.5732,
+      "step": 210
+    },
+    {
+      "epoch": 0.1877733916568868,
+      "grad_norm": 0.7812625765800476,
+      "learning_rate": 6.25e-06,
+      "loss": 0.5745,
+      "step": 220
+    },
+    {
+      "epoch": 0.19630854582310894,
+      "grad_norm": 0.7922148704528809,
+      "learning_rate": 6.53409090909091e-06,
+      "loss": 0.5742,
+      "step": 230
+    },
+    {
+      "epoch": 0.20484369998933105,
+      "grad_norm": 0.8420217633247375,
+      "learning_rate": 6.818181818181818e-06,
+      "loss": 0.5593,
+      "step": 240
+    },
+    {
+      "epoch": 0.2133788541555532,
+      "grad_norm": 0.8149869441986084,
+      "learning_rate": 7.102272727272727e-06,
+      "loss": 0.5694,
+      "step": 250
+    },
+    {
+      "epoch": 0.22191400832177532,
+      "grad_norm": 0.8200048208236694,
+      "learning_rate": 7.386363636363637e-06,
+      "loss": 0.5609,
+      "step": 260
+    },
+    {
+      "epoch": 0.23044916248799743,
+      "grad_norm": 0.7718726396560669,
+      "learning_rate": 7.670454545454547e-06,
+      "loss": 0.5596,
+      "step": 270
+    },
+    {
+      "epoch": 0.23898431665421957,
+      "grad_norm": 0.8079963326454163,
+      "learning_rate": 7.954545454545455e-06,
+      "loss": 0.5515,
+      "step": 280
+    },
+    {
+      "epoch": 0.2475194708204417,
+      "grad_norm": 0.8398913741111755,
+      "learning_rate": 8.238636363636365e-06,
+      "loss": 0.5615,
+      "step": 290
+    },
+    {
+      "epoch": 0.2560546249866638,
+      "grad_norm": 0.8522515892982483,
+      "learning_rate": 8.522727272727273e-06,
+      "loss": 0.5405,
+      "step": 300
+    },
+    {
+      "epoch": 0.26458977915288595,
+      "grad_norm": 0.965617299079895,
+      "learning_rate": 8.806818181818183e-06,
+      "loss": 0.5438,
+      "step": 310
+    },
+    {
+      "epoch": 0.2731249333191081,
+      "grad_norm": 0.834233283996582,
+      "learning_rate": 9.090909090909091e-06,
+      "loss": 0.5342,
+      "step": 320
+    },
+    {
+      "epoch": 0.2816600874853302,
+      "grad_norm": 0.8867103457450867,
+      "learning_rate": 9.375000000000001e-06,
+      "loss": 0.5272,
+      "step": 330
+    },
+    {
+      "epoch": 0.2901952416515523,
+      "grad_norm": 0.767537534236908,
+      "learning_rate": 9.65909090909091e-06,
+      "loss": 0.5319,
+      "step": 340
+    },
+    {
+      "epoch": 0.29873039581777444,
+      "grad_norm": 0.8806461095809937,
+      "learning_rate": 9.943181818181819e-06,
+      "loss": 0.5322,
+      "step": 350
+    },
+    {
+      "epoch": 0.3072655499839966,
+      "grad_norm": 0.7467280030250549,
+      "learning_rate": 9.999841959480543e-06,
+      "loss": 0.5267,
+      "step": 360
+    },
+    {
+      "epoch": 0.3158007041502187,
+      "grad_norm": 0.7319239974021912,
+      "learning_rate": 9.999199936992976e-06,
+      "loss": 0.5221,
+      "step": 370
+    },
+    {
+      "epoch": 0.32433585831644085,
+      "grad_norm": 0.7590662837028503,
+      "learning_rate": 9.998064118371841e-06,
+      "loss": 0.525,
+      "step": 380
+    },
+    {
+      "epoch": 0.332871012482663,
+      "grad_norm": 0.7705526351928711,
+      "learning_rate": 9.996434615807662e-06,
+      "loss": 0.5254,
+      "step": 390
+    },
+    {
+      "epoch": 0.3414061666488851,
+      "grad_norm": 0.8061032295227051,
+      "learning_rate": 9.994311590254603e-06,
+      "loss": 0.5221,
+      "step": 400
+    },
+    {
+      "epoch": 0.3499413208151072,
+      "grad_norm": 0.7729495167732239,
+      "learning_rate": 9.991695251414584e-06,
+      "loss": 0.53,
+      "step": 410
+    },
+    {
+      "epoch": 0.35847647498132934,
+      "grad_norm": 0.768975019454956,
+      "learning_rate": 9.988585857716555e-06,
+      "loss": 0.5099,
+      "step": 420
+    },
+    {
+      "epoch": 0.3670116291475515,
+      "grad_norm": 0.7381827235221863,
+      "learning_rate": 9.984983716290981e-06,
+      "loss": 0.5182,
+      "step": 430
+    },
+    {
+      "epoch": 0.3755467833137736,
+      "grad_norm": 0.8239282965660095,
+      "learning_rate": 9.980889182939495e-06,
+      "loss": 0.52,
+      "step": 440
+    },
+    {
+      "epoch": 0.38408193747999575,
+      "grad_norm": 0.8047374486923218,
+      "learning_rate": 9.976302662099758e-06,
+      "loss": 0.5091,
+      "step": 450
+    },
+    {
+      "epoch": 0.3926170916462179,
+      "grad_norm": 0.7568238377571106,
+      "learning_rate": 9.971224606805514e-06,
+      "loss": 0.5171,
+      "step": 460
+    },
+    {
+      "epoch": 0.40115224581243997,
+      "grad_norm": 0.7178462147712708,
+      "learning_rate": 9.965655518641837e-06,
+      "loss": 0.5119,
+      "step": 470
+    },
+    {
+      "epoch": 0.4096873999786621,
+      "grad_norm": 0.8041744232177734,
+      "learning_rate": 9.959595947695588e-06,
+      "loss": 0.506,
+      "step": 480
+    },
+    {
+      "epoch": 0.41822255414488424,
+      "grad_norm": 0.7083892226219177,
+      "learning_rate": 9.953046492501082e-06,
+      "loss": 0.5066,
+      "step": 490
+    },
+    {
+      "epoch": 0.4267577083111064,
+      "grad_norm": 0.7124646902084351,
+      "learning_rate": 9.946007799980963e-06,
+      "loss": 0.5098,
+      "step": 500
+    },
+    {
+      "epoch": 0.4352928624773285,
+      "grad_norm": 0.7167072892189026,
+      "learning_rate": 9.938480565382312e-06,
+      "loss": 0.513,
+      "step": 510
+    },
+    {
+      "epoch": 0.44382801664355065,
+      "grad_norm": 0.7211044430732727,
+      "learning_rate": 9.930465532207967e-06,
+      "loss": 0.5094,
+      "step": 520
+    },
+    {
+      "epoch": 0.45236317080977273,
+      "grad_norm": 0.7868006229400635,
+      "learning_rate": 9.921963492143083e-06,
+      "loss": 0.5131,
+      "step": 530
+    },
+    {
+      "epoch": 0.46089832497599487,
+      "grad_norm": 0.7130579948425293,
+      "learning_rate": 9.912975284976942e-06,
+      "loss": 0.506,
+      "step": 540
+    },
+    {
+      "epoch": 0.469433479142217,
+      "grad_norm": 0.7939994931221008,
+      "learning_rate": 9.903501798519994e-06,
+      "loss": 0.5025,
+      "step": 550
+    },
+    {
+      "epoch": 0.47796863330843914,
+      "grad_norm": 0.7138738036155701,
+      "learning_rate": 9.893543968516166e-06,
+      "loss": 0.5037,
+      "step": 560
+    },
+    {
+      "epoch": 0.4865037874746613,
+      "grad_norm": 0.716701328754425,
+      "learning_rate": 9.883102778550434e-06,
+      "loss": 0.5034,
+      "step": 570
+    },
+    {
+      "epoch": 0.4950389416408834,
+      "grad_norm": 0.7295582294464111,
+      "learning_rate": 9.872179259951672e-06,
+      "loss": 0.4949,
+      "step": 580
+    },
+    {
+      "epoch": 0.5035740958071055,
+      "grad_norm": 0.7438467144966125,
+      "learning_rate": 9.860774491690777e-06,
+      "loss": 0.4969,
+      "step": 590
+    },
+    {
+      "epoch": 0.5121092499733276,
+      "grad_norm": 0.696615993976593,
+      "learning_rate": 9.848889600274101e-06,
+      "loss": 0.4926,
+      "step": 600
+    },
+    {
+      "epoch": 0.5206444041395498,
+      "grad_norm": 0.6948277354240417,
+      "learning_rate": 9.836525759632171e-06,
+      "loss": 0.5034,
+      "step": 610
+    },
+    {
+      "epoch": 0.5291795583057719,
+      "grad_norm": 0.8091464042663574,
+      "learning_rate": 9.823684191003744e-06,
+      "loss": 0.5015,
+      "step": 620
+    },
+    {
+      "epoch": 0.537714712471994,
+      "grad_norm": 0.7501440048217773,
+      "learning_rate": 9.810366162815171e-06,
+      "loss": 0.5024,
+      "step": 630
+    },
+    {
+      "epoch": 0.5462498666382162,
+      "grad_norm": 0.7048625946044922,
+      "learning_rate": 9.796572990555112e-06,
+      "loss": 0.4955,
+      "step": 640
+    },
+    {
+      "epoch": 0.5547850208044383,
+      "grad_norm": 0.7141663432121277,
+      "learning_rate": 9.782306036644596e-06,
+      "loss": 0.5045,
+      "step": 650
+    },
+    {
+      "epoch": 0.5633201749706604,
+      "grad_norm": 0.7329632639884949,
+      "learning_rate": 9.767566710302453e-06,
+      "loss": 0.5038,
+      "step": 660
+    },
+    {
+      "epoch": 0.5718553291368825,
+      "grad_norm": 0.699938952922821,
+      "learning_rate": 9.752356467406114e-06,
+      "loss": 0.5066,
+      "step": 670
+    },
+    {
+      "epoch": 0.5803904833031046,
+      "grad_norm": 0.6912060976028442,
+      "learning_rate": 9.736676810347807e-06,
+      "loss": 0.4918,
+      "step": 680
+    },
+    {
+      "epoch": 0.5889256374693268,
+      "grad_norm": 0.6718906164169312,
+      "learning_rate": 9.720529287886159e-06,
+      "loss": 0.4932,
+      "step": 690
+    },
+    {
+      "epoch": 0.5974607916355489,
+      "grad_norm": 0.6830474734306335,
+      "learning_rate": 9.703915494993215e-06,
+      "loss": 0.489,
+      "step": 700
+    },
+    {
+      "epoch": 0.6059959458017711,
+      "grad_norm": 0.6920751333236694,
+      "learning_rate": 9.686837072696902e-06,
+      "loss": 0.4976,
+      "step": 710
+    },
+    {
+      "epoch": 0.6145310999679932,
+      "grad_norm": 0.695093035697937,
+      "learning_rate": 9.669295707918928e-06,
+      "loss": 0.4877,
+      "step": 720
+    },
+    {
+      "epoch": 0.6230662541342153,
+      "grad_norm": 0.683686375617981,
+      "learning_rate": 9.651293133308159e-06,
+      "loss": 0.4939,
+      "step": 730
+    },
+    {
+      "epoch": 0.6316014083004374,
+      "grad_norm": 0.7766408920288086,
+      "learning_rate": 9.632831127069479e-06,
+      "loss": 0.4993,
+      "step": 740
+    },
+    {
+      "epoch": 0.6401365624666595,
+      "grad_norm": 0.7032186985015869,
+      "learning_rate": 9.61391151278814e-06,
+      "loss": 0.4896,
+      "step": 750
+    },
+    {
+      "epoch": 0.6486717166328817,
+      "grad_norm": 0.7620243430137634,
+      "learning_rate": 9.594536159249654e-06,
+      "loss": 0.4926,
+      "step": 760
+    },
+    {
+      "epoch": 0.6572068707991038,
+      "grad_norm": 0.6803762316703796,
+      "learning_rate": 9.57470698025518e-06,
+      "loss": 0.4902,
+      "step": 770
+    },
+    {
+      "epoch": 0.665742024965326,
+      "grad_norm": 0.6878206133842468,
+      "learning_rate": 9.55442593443251e-06,
+      "loss": 0.4841,
+      "step": 780
+    },
+    {
+      "epoch": 0.674277179131548,
+      "grad_norm": 0.7530574202537537,
+      "learning_rate": 9.533695025042591e-06,
+      "loss": 0.4839,
+      "step": 790
+    },
+    {
+      "epoch": 0.6828123332977702,
+      "grad_norm": 0.7192838191986084,
+      "learning_rate": 9.512516299781665e-06,
+      "loss": 0.4841,
+      "step": 800
+    },
+    {
+      "epoch": 0.6913474874639923,
+      "grad_norm": 0.7086024880409241,
+      "learning_rate": 9.490891850578993e-06,
+      "loss": 0.4869,
+      "step": 810
+    },
+    {
+      "epoch": 0.6998826416302144,
+      "grad_norm": 0.6931254267692566,
+      "learning_rate": 9.468823813390236e-06,
+      "loss": 0.4834,
+      "step": 820
+    },
+    {
+      "epoch": 0.7084177957964366,
+      "grad_norm": 0.6703010201454163,
+      "learning_rate": 9.446314367986476e-06,
+      "loss": 0.4904,
+      "step": 830
+    },
+    {
+      "epoch": 0.7169529499626587,
+      "grad_norm": 0.721376359462738,
+      "learning_rate": 9.423365737738898e-06,
+      "loss": 0.4896,
+      "step": 840
+    },
+    {
+      "epoch": 0.7254881041288809,
+      "grad_norm": 0.7349815964698792,
+      "learning_rate": 9.399980189399191e-06,
+      "loss": 0.4938,
+      "step": 850
+    },
+    {
+      "epoch": 0.734023258295103,
+      "grad_norm": 0.7041159272193909,
+      "learning_rate": 9.376160032875638e-06,
+      "loss": 0.4889,
+      "step": 860
+    },
+    {
+      "epoch": 0.742558412461325,
+      "grad_norm": 0.7001343965530396,
+      "learning_rate": 9.351907621004963e-06,
+      "loss": 0.4908,
+      "step": 870
+    },
+    {
+      "epoch": 0.7510935666275472,
+      "grad_norm": 0.7233178019523621,
+      "learning_rate": 9.327225349319922e-06,
+      "loss": 0.4856,
+      "step": 880
+    },
+    {
+      "epoch": 0.7596287207937693,
+      "grad_norm": 0.6933467984199524,
+      "learning_rate": 9.30211565581269e-06,
+      "loss": 0.4817,
+      "step": 890
+    },
+    {
+      "epoch": 0.7681638749599915,
+      "grad_norm": 0.6627389192581177,
+      "learning_rate": 9.276581020694048e-06,
+      "loss": 0.4804,
+      "step": 900
+    },
+    {
+      "epoch": 0.7766990291262136,
+      "grad_norm": 0.6689618229866028,
+      "learning_rate": 9.250623966148392e-06,
+      "loss": 0.4854,
+      "step": 910
+    },
+    {
+      "epoch": 0.7852341832924358,
+      "grad_norm": 0.7105383276939392,
+      "learning_rate": 9.224247056084617e-06,
+      "loss": 0.4831,
+      "step": 920
+    },
+    {
+      "epoch": 0.7937693374586579,
+      "grad_norm": 0.7106824517250061,
+      "learning_rate": 9.197452895882857e-06,
+      "loss": 0.4802,
+      "step": 930
+    },
+    {
+      "epoch": 0.8023044916248799,
+      "grad_norm": 0.6922078132629395,
+      "learning_rate": 9.170244132137143e-06,
+      "loss": 0.4854,
+      "step": 940
+    },
+    {
+      "epoch": 0.8108396457911021,
+      "grad_norm": 0.7484738230705261,
+      "learning_rate": 9.142623452393984e-06,
+      "loss": 0.4708,
+      "step": 950
+    },
+    {
+      "epoch": 0.8193747999573242,
+      "grad_norm": 0.7449445128440857,
+      "learning_rate": 9.114593584886902e-06,
+      "loss": 0.4786,
+      "step": 960
+    },
+    {
+      "epoch": 0.8279099541235464,
+      "grad_norm": 0.6751102209091187,
+      "learning_rate": 9.086157298266957e-06,
+      "loss": 0.4783,
+      "step": 970
+    },
+    {
+      "epoch": 0.8364451082897685,
+      "grad_norm": 0.7621346712112427,
+      "learning_rate": 9.057317401329271e-06,
+      "loss": 0.4868,
+      "step": 980
+    },
+    {
+      "epoch": 0.8449802624559906,
+      "grad_norm": 0.7197836637496948,
+      "learning_rate": 9.028076742735583e-06,
+      "loss": 0.479,
+      "step": 990
+    },
+    {
+      "epoch": 0.8535154166222128,
+      "grad_norm": 0.7285887598991394,
+      "learning_rate": 8.998438210732882e-06,
+      "loss": 0.4851,
+      "step": 1000
+    },
+    {
+      "epoch": 0.8620505707884348,
+      "grad_norm": 0.7014701962471008,
+      "learning_rate": 8.968404732868112e-06,
+      "loss": 0.4703,
+      "step": 1010
+    },
+    {
+      "epoch": 0.870585724954657,
+      "grad_norm": 0.6849453449249268,
+      "learning_rate": 8.937979275699012e-06,
+      "loss": 0.4729,
+      "step": 1020
+    },
+    {
+      "epoch": 0.8791208791208791,
+      "grad_norm": 0.7209568023681641,
+      "learning_rate": 8.907164844501084e-06,
+      "loss": 0.4777,
+      "step": 1030
+    },
+    {
+      "epoch": 0.8876560332871013,
+      "grad_norm": 0.6904247403144836,
+      "learning_rate": 8.875964482970756e-06,
+      "loss": 0.4731,
+      "step": 1040
+    },
+    {
+      "epoch": 0.8961911874533234,
+      "grad_norm": 0.6657469272613525,
+      "learning_rate": 8.844381272924733e-06,
+      "loss": 0.4773,
+      "step": 1050
+    },
+    {
+      "epoch": 0.9047263416195455,
+      "grad_norm": 0.6649881601333618,
+      "learning_rate": 8.812418333995597e-06,
+      "loss": 0.4659,
+      "step": 1060
+    },
+    {
+      "epoch": 0.9132614957857677,
+      "grad_norm": 0.6824942231178284,
+      "learning_rate": 8.780078823323667e-06,
+      "loss": 0.4736,
+      "step": 1070
+    },
+    {
+      "epoch": 0.9217966499519897,
+      "grad_norm": 0.6798464059829712,
+      "learning_rate": 8.74736593524514e-06,
+      "loss": 0.4695,
+      "step": 1080
+    },
+    {
+      "epoch": 0.9303318041182119,
+      "grad_norm": 0.6828529238700867,
+      "learning_rate": 8.71428290097658e-06,
+      "loss": 0.4678,
+      "step": 1090
+    },
+    {
+      "epoch": 0.938866958284434,
+      "grad_norm": 0.6548129916191101,
+      "learning_rate": 8.680832988295757e-06,
+      "loss": 0.4748,
+      "step": 1100
+    },
+    {
+      "epoch": 0.9474021124506561,
+      "grad_norm": 0.7184150815010071,
+      "learning_rate": 8.647019501218862e-06,
+      "loss": 0.4851,
+      "step": 1110
+    },
+    {
+      "epoch": 0.9559372666168783,
+      "grad_norm": 0.6707058548927307,
+      "learning_rate": 8.612845779674166e-06,
+      "loss": 0.4621,
+      "step": 1120
+    },
+    {
+      "epoch": 0.9644724207831004,
+      "grad_norm": 0.6430131793022156,
+      "learning_rate": 8.578315199172105e-06,
+      "loss": 0.4733,
+      "step": 1130
+    },
+    {
+      "epoch": 0.9730075749493226,
+      "grad_norm": 0.652946949005127,
+      "learning_rate": 8.543431170471872e-06,
+      "loss": 0.469,
+      "step": 1140
+    },
+    {
+      "epoch": 0.9815427291155446,
+      "grad_norm": 0.678126871585846,
+      "learning_rate": 8.508197139244515e-06,
+      "loss": 0.4645,
+      "step": 1150
+    },
+    {
+      "epoch": 0.9900778832817668,
+      "grad_norm": 0.6810508966445923,
+      "learning_rate": 8.472616585732597e-06,
+      "loss": 0.4722,
+      "step": 1160
+    },
+    {
+      "epoch": 0.9986130374479889,
+      "grad_norm": 0.6804456114768982,
+      "learning_rate": 8.436693024406427e-06,
+      "loss": 0.4796,
+      "step": 1170
+    },
+    {
+      "epoch": 1.0068281233329777,
+      "grad_norm": 0.7665020227432251,
+      "learning_rate": 8.400430003616924e-06,
+      "loss": 0.4501,
+      "step": 1180
+    },
+    {
+      "epoch": 1.0153632774991999,
+      "grad_norm": 0.7121028304100037,
+      "learning_rate": 8.363831105245125e-06,
+      "loss": 0.4436,
+      "step": 1190
+    },
+    {
+      "epoch": 1.023898431665422,
+      "grad_norm": 0.7494065761566162,
+      "learning_rate": 8.326899944348389e-06,
+      "loss": 0.4459,
+      "step": 1200
+    },
+    {
+      "epoch": 1.032433585831644,
+      "grad_norm": 0.6524202227592468,
+      "learning_rate": 8.289640168803308e-06,
+      "loss": 0.4445,
+      "step": 1210
+    },
+    {
+      "epoch": 1.0409687399978662,
+      "grad_norm": 0.6446959972381592,
+      "learning_rate": 8.25205545894541e-06,
+      "loss": 0.4406,
+      "step": 1220
+    },
+    {
+      "epoch": 1.0495038941640884,
+      "grad_norm": 0.6765570044517517,
+      "learning_rate": 8.214149527205606e-06,
+      "loss": 0.4538,
+      "step": 1230
+    },
+    {
+      "epoch": 1.0580390483303104,
+      "grad_norm": 0.6525077223777771,
+      "learning_rate": 8.17592611774352e-06,
+      "loss": 0.4492,
+      "step": 1240
+    },
+    {
+      "epoch": 1.0665742024965326,
+      "grad_norm": 0.6981220245361328,
+      "learning_rate": 8.13738900607764e-06,
+      "loss": 0.4428,
+      "step": 1250
+    },
+    {
+      "epoch": 1.0751093566627548,
+      "grad_norm": 0.7378373742103577,
+      "learning_rate": 8.098541998712405e-06,
+      "loss": 0.442,
+      "step": 1260
+    },
+    {
+      "epoch": 1.083644510828977,
+      "grad_norm": 0.6806818246841431,
+      "learning_rate": 8.059388932762207e-06,
+      "loss": 0.4428,
+      "step": 1270
+    },
+    {
+      "epoch": 1.092179664995199,
+      "grad_norm": 0.704845130443573,
+      "learning_rate": 8.019933675572389e-06,
+      "loss": 0.4402,
+      "step": 1280
+    },
+    {
+      "epoch": 1.100714819161421,
+      "grad_norm": 0.8347475528717041,
+      "learning_rate": 7.980180124337239e-06,
+      "loss": 0.448,
+      "step": 1290
+    },
+    {
+      "epoch": 1.1092499733276433,
+      "grad_norm": 0.7030121684074402,
+      "learning_rate": 7.940132205715053e-06,
+      "loss": 0.4462,
+      "step": 1300
+    },
+    {
+      "epoch": 1.1177851274938653,
+      "grad_norm": 0.6647533178329468,
+      "learning_rate": 7.89979387544027e-06,
+      "loss": 0.4481,
+      "step": 1310
+    },
+    {
+      "epoch": 1.1263202816600875,
+      "grad_norm": 0.724305272102356,
+      "learning_rate": 7.859169117932752e-06,
+      "loss": 0.4426,
+      "step": 1320
+    },
+    {
+      "epoch": 1.1348554358263097,
+      "grad_norm": 0.6487372517585754,
+      "learning_rate": 7.818261945904226e-06,
+      "loss": 0.4442,
+      "step": 1330
+    },
+    {
+      "epoch": 1.1433905899925318,
+      "grad_norm": 0.7011922597885132,
+      "learning_rate": 7.777076399961912e-06,
+      "loss": 0.4486,
+      "step": 1340
+    },
+    {
+      "epoch": 1.1519257441587538,
+      "grad_norm": 0.6805099844932556,
+      "learning_rate": 7.735616548209425e-06,
+      "loss": 0.4469,
+      "step": 1350
+    },
+    {
+      "epoch": 1.160460898324976,
+      "grad_norm": 0.683447003364563,
+      "learning_rate": 7.693886485844955e-06,
+      "loss": 0.4553,
+      "step": 1360
+    },
+    {
+      "epoch": 1.1689960524911982,
+      "grad_norm": 0.8037645816802979,
+      "learning_rate": 7.651890334756736e-06,
+      "loss": 0.4387,
+      "step": 1370
+    },
+    {
+      "epoch": 1.1775312066574202,
+      "grad_norm": 0.668209433555603,
+      "learning_rate": 7.609632243115941e-06,
+      "loss": 0.4373,
+      "step": 1380
+    },
+    {
+      "epoch": 1.1860663608236424,
+      "grad_norm": 0.7479778528213501,
+      "learning_rate": 7.567116384966918e-06,
+      "loss": 0.4525,
+      "step": 1390
+    },
+    {
+      "epoch": 1.1946015149898646,
+      "grad_norm": 0.743675172328949,
+      "learning_rate": 7.524346959814918e-06,
+      "loss": 0.4458,
+      "step": 1400
+    },
+    {
+      "epoch": 1.2031366691560867,
+      "grad_norm": 0.6677897572517395,
+      "learning_rate": 7.481328192211278e-06,
+      "loss": 0.4521,
+      "step": 1410
+    },
+    {
+      "epoch": 1.2116718233223087,
+      "grad_norm": 0.655086100101471,
+      "learning_rate": 7.4380643313361385e-06,
+      "loss": 0.4439,
+      "step": 1420
+    },
+    {
+      "epoch": 1.220206977488531,
+      "grad_norm": 0.7043260335922241,
+      "learning_rate": 7.394559650578746e-06,
+      "loss": 0.4387,
+      "step": 1430
+    },
+    {
+      "epoch": 1.228742131654753,
+      "grad_norm": 0.6985997557640076,
+      "learning_rate": 7.3508184471153274e-06,
+      "loss": 0.4396,
+      "step": 1440
+    },
+    {
+      "epoch": 1.237277285820975,
+      "grad_norm": 0.6879217028617859,
+      "learning_rate": 7.306845041484656e-06,
+      "loss": 0.4438,
+      "step": 1450
+    },
+    {
+      "epoch": 1.2458124399871973,
+      "grad_norm": 0.6632692217826843,
+      "learning_rate": 7.262643777161279e-06,
+      "loss": 0.4421,
+      "step": 1460
+    },
+    {
+      "epoch": 1.2543475941534195,
+      "grad_norm": 0.6956544518470764,
+      "learning_rate": 7.218219020126491e-06,
+      "loss": 0.45,
+      "step": 1470
+    },
+    {
+      "epoch": 1.2628827483196416,
+      "grad_norm": 0.6400607824325562,
+      "learning_rate": 7.1735751584370894e-06,
+      "loss": 0.4372,
+      "step": 1480
+    },
+    {
+      "epoch": 1.2714179024858636,
+      "grad_norm": 0.735914945602417,
+      "learning_rate": 7.128716601791938e-06,
+      "loss": 0.4459,
+      "step": 1490
+    },
+    {
+      "epoch": 1.2799530566520858,
+      "grad_norm": 0.7256124019622803,
+      "learning_rate": 7.083647781096404e-06,
+      "loss": 0.4437,
+      "step": 1500
+    },
+    {
+      "epoch": 1.2884882108183078,
+      "grad_norm": 0.6894016861915588,
+      "learning_rate": 7.038373148024693e-06,
+      "loss": 0.4468,
+      "step": 1510
+    },
+    {
+      "epoch": 1.29702336498453,
+      "grad_norm": 0.676726222038269,
+      "learning_rate": 6.992897174580133e-06,
+      "loss": 0.4409,
+      "step": 1520
+    },
+    {
+      "epoch": 1.3055585191507522,
+      "grad_norm": 0.7112148404121399,
+      "learning_rate": 6.947224352653454e-06,
+      "loss": 0.4387,
+      "step": 1530
+    },
+    {
+      "epoch": 1.3140936733169744,
+      "grad_norm": 0.6504147052764893,
+      "learning_rate": 6.901359193579104e-06,
+      "loss": 0.4357,
+      "step": 1540
+    },
+    {
+      "epoch": 1.3226288274831965,
+      "grad_norm": 0.6736079454421997,
+      "learning_rate": 6.855306227689634e-06,
+      "loss": 0.4403,
+      "step": 1550
+    },
+    {
+      "epoch": 1.3311639816494185,
+      "grad_norm": 0.6631302237510681,
+      "learning_rate": 6.809070003868221e-06,
+      "loss": 0.4368,
+      "step": 1560
+    },
+    {
+      "epoch": 1.3396991358156407,
+      "grad_norm": 0.6592603921890259,
+      "learning_rate": 6.762655089099353e-06,
+      "loss": 0.446,
+      "step": 1570
+    },
+    {
+      "epoch": 1.3482342899818627,
+      "grad_norm": 0.6875331997871399,
+      "learning_rate": 6.716066068017722e-06,
+      "loss": 0.4459,
+      "step": 1580
+    },
+    {
+      "epoch": 1.3567694441480849,
+      "grad_norm": 0.8026809096336365,
+      "learning_rate": 6.669307542455369e-06,
+      "loss": 0.4363,
+      "step": 1590
+    },
+    {
+      "epoch": 1.365304598314307,
+      "grad_norm": 0.8665246367454529,
+      "learning_rate": 6.6223841309871535e-06,
+      "loss": 0.4415,
+      "step": 1600
+    },
+    {
+      "epoch": 1.3738397524805293,
+      "grad_norm": 0.7898995280265808,
+      "learning_rate": 6.575300468474546e-06,
+      "loss": 0.4411,
+      "step": 1610
+    },
+    {
+      "epoch": 1.3823749066467514,
+      "grad_norm": 0.6701722145080566,
+      "learning_rate": 6.52806120560781e-06,
+      "loss": 0.441,
+      "step": 1620
+    },
+    {
+      "epoch": 1.3909100608129734,
+      "grad_norm": 0.6245713233947754,
+      "learning_rate": 6.480671008446651e-06,
+      "loss": 0.4357,
+      "step": 1630
+    },
+    {
+      "epoch": 1.3994452149791956,
+      "grad_norm": 0.6475566029548645,
+      "learning_rate": 6.433134557959303e-06,
+      "loss": 0.4487,
+      "step": 1640
+    },
+    {
+      "epoch": 1.4079803691454176,
+      "grad_norm": 0.7050219178199768,
+      "learning_rate": 6.385456549560187e-06,
+      "loss": 0.4488,
+      "step": 1650
+    },
+    {
+      "epoch": 1.4165155233116398,
+      "grad_norm": 0.6392747759819031,
+      "learning_rate": 6.337641692646106e-06,
+      "loss": 0.4453,
+      "step": 1660
+    },
+    {
+      "epoch": 1.425050677477862,
+      "grad_norm": 0.6791578531265259,
+      "learning_rate": 6.2896947101310756e-06,
+      "loss": 0.4406,
+      "step": 1670
+    },
+    {
+      "epoch": 1.4335858316440842,
+      "grad_norm": 0.6261289119720459,
+      "learning_rate": 6.24162033797983e-06,
+      "loss": 0.4364,
+      "step": 1680
+    },
+    {
+      "epoch": 1.4421209858103061,
+      "grad_norm": 0.6886487603187561,
+      "learning_rate": 6.193423324740018e-06,
+      "loss": 0.44,
+      "step": 1690
+    },
+    {
+      "epoch": 1.4506561399765283,
+      "grad_norm": 0.7670247554779053,
+      "learning_rate": 6.1451084310731545e-06,
+      "loss": 0.4348,
+      "step": 1700
+    },
+    {
+      "epoch": 1.4591912941427505,
+      "grad_norm": 0.6851808428764343,
+      "learning_rate": 6.096680429284412e-06,
+      "loss": 0.4358,
+      "step": 1710
+    },
+    {
+      "epoch": 1.4677264483089725,
+      "grad_norm": 0.660381555557251,
+      "learning_rate": 6.0481441028512154e-06,
+      "loss": 0.4419,
+      "step": 1720
+    },
+    {
+      "epoch": 1.4762616024751947,
+      "grad_norm": 0.6627777218818665,
+      "learning_rate": 5.9995042459507555e-06,
+      "loss": 0.4443,
+      "step": 1730
+    },
+    {
+      "epoch": 1.4847967566414169,
+      "grad_norm": 0.7264851331710815,
+      "learning_rate": 5.9507656629864554e-06,
+      "loss": 0.4419,
+      "step": 1740
+    },
+    {
+      "epoch": 1.493331910807639,
+      "grad_norm": 0.6353123784065247,
+      "learning_rate": 5.901933168113405e-06,
+      "loss": 0.4325,
+      "step": 1750
+    },
+    {
+      "epoch": 1.5018670649738612,
+      "grad_norm": 0.7154319882392883,
+      "learning_rate": 5.853011584762846e-06,
+      "loss": 0.4413,
+      "step": 1760
+    },
+    {
+      "epoch": 1.5104022191400832,
+      "grad_norm": 0.678537905216217,
+      "learning_rate": 5.804005745165735e-06,
+      "loss": 0.4418,
+      "step": 1770
+    },
+    {
+      "epoch": 1.5189373733063052,
+      "grad_norm": 0.6648243069648743,
+      "learning_rate": 5.754920489875447e-06,
+      "loss": 0.438,
+      "step": 1780
+    },
+    {
+      "epoch": 1.5274725274725274,
+      "grad_norm": 0.7042447924613953,
+      "learning_rate": 5.705760667289644e-06,
+      "loss": 0.441,
+      "step": 1790
+    },
+    {
+      "epoch": 1.5360076816387496,
+      "grad_norm": 0.6617943048477173,
+      "learning_rate": 5.65653113317137e-06,
+      "loss": 0.4375,
+      "step": 1800
+    },
+    {
+      "epoch": 1.5445428358049718,
+      "grad_norm": 0.6619049310684204,
+      "learning_rate": 5.607236750169431e-06,
+      "loss": 0.4395,
+      "step": 1810
+    },
+    {
+      "epoch": 1.553077989971194,
+      "grad_norm": 0.6774610280990601,
+      "learning_rate": 5.557882387338086e-06,
+      "loss": 0.4417,
+      "step": 1820
+    },
+    {
+      "epoch": 1.5616131441374161,
+      "grad_norm": 0.6360573768615723,
+      "learning_rate": 5.508472919656101e-06,
+      "loss": 0.4346,
+      "step": 1830
+    },
+    {
+      "epoch": 1.5701482983036381,
+      "grad_norm": 0.6723465919494629,
+      "learning_rate": 5.459013227545224e-06,
+      "loss": 0.4387,
+      "step": 1840
+    },
+    {
+      "epoch": 1.57868345246986,
+      "grad_norm": 0.6785566806793213,
+      "learning_rate": 5.4095081963881225e-06,
+      "loss": 0.4364,
+      "step": 1850
+    },
+    {
+      "epoch": 1.5872186066360823,
+      "grad_norm": 0.6658756136894226,
+      "learning_rate": 5.359962716045836e-06,
+      "loss": 0.4302,
+      "step": 1860
+    },
+    {
+      "epoch": 1.5957537608023045,
+      "grad_norm": 0.6586979031562805,
+      "learning_rate": 5.310381680374767e-06,
+      "loss": 0.4386,
+      "step": 1870
+    },
+    {
+      "epoch": 1.6042889149685267,
+      "grad_norm": 0.7203266620635986,
+      "learning_rate": 5.260769986743296e-06,
+      "loss": 0.4377,
+      "step": 1880
+    },
+    {
+      "epoch": 1.6128240691347489,
+      "grad_norm": 0.7455547451972961,
+      "learning_rate": 5.2111325355480506e-06,
+      "loss": 0.4428,
+      "step": 1890
+    },
+    {
+      "epoch": 1.6213592233009708,
+      "grad_norm": 0.6452232599258423,
+      "learning_rate": 5.161474229729858e-06,
+      "loss": 0.4336,
+      "step": 1900
+    },
+    {
+      "epoch": 1.629894377467193,
+      "grad_norm": 0.6487184166908264,
+      "learning_rate": 5.111799974289462e-06,
+      "loss": 0.4426,
+      "step": 1910
+    },
+    {
+      "epoch": 1.638429531633415,
+      "grad_norm": 0.6176945567131042,
+      "learning_rate": 5.062114675803032e-06,
+      "loss": 0.4356,
+      "step": 1920
+    },
+    {
+      "epoch": 1.6469646857996372,
+      "grad_norm": 0.6743866205215454,
+      "learning_rate": 5.012423241937515e-06,
+      "loss": 0.4308,
+      "step": 1930
+    },
+    {
+      "epoch": 1.6554998399658594,
+      "grad_norm": 0.6723053455352783,
+      "learning_rate": 4.962730580965879e-06,
+      "loss": 0.4344,
+      "step": 1940
+    },
+    {
+      "epoch": 1.6640349941320816,
+      "grad_norm": 0.7293416261672974,
+      "learning_rate": 4.913041601282298e-06,
+      "loss": 0.4229,
+      "step": 1950
+    },
+    {
+      "epoch": 1.6725701482983037,
+      "grad_norm": 0.6705866456031799,
+      "learning_rate": 4.863361210917332e-06,
+      "loss": 0.4412,
+      "step": 1960
+    },
+    {
+      "epoch": 1.6811053024645257,
+      "grad_norm": 0.676863968372345,
+      "learning_rate": 4.813694317053122e-06,
+      "loss": 0.4423,
+      "step": 1970
+    },
+    {
+      "epoch": 1.689640456630748,
+      "grad_norm": 0.659705400466919,
+      "learning_rate": 4.764045825538702e-06,
+      "loss": 0.4372,
+      "step": 1980
+    },
+    {
+      "epoch": 1.6981756107969699,
+      "grad_norm": 0.6498283743858337,
+      "learning_rate": 4.714420640405406e-06,
+      "loss": 0.4344,
+      "step": 1990
+    },
+    {
+      "epoch": 1.706710764963192,
+      "grad_norm": 0.6507246494293213,
+      "learning_rate": 4.664823663382482e-06,
+      "loss": 0.4274,
+      "step": 2000
+    },
+    {
+      "epoch": 1.7152459191294143,
+      "grad_norm": 0.705443263053894,
+      "learning_rate": 4.61525979341292e-06,
+      "loss": 0.4365,
+      "step": 2010
+    },
+    {
+      "epoch": 1.7237810732956365,
+      "grad_norm": 0.6291791200637817,
+      "learning_rate": 4.565733926169561e-06,
+      "loss": 0.4354,
+      "step": 2020
+    },
+    {
+      "epoch": 1.7323162274618586,
+      "grad_norm": 0.6487744450569153,
+      "learning_rate": 4.516250953571527e-06,
+      "loss": 0.439,
+      "step": 2030
+    },
+    {
+      "epoch": 1.7408513816280806,
+      "grad_norm": 0.6847457885742188,
+      "learning_rate": 4.466815763301014e-06,
+      "loss": 0.4245,
+      "step": 2040
+    },
+    {
+      "epoch": 1.7493865357943028,
+      "grad_norm": 0.6660692095756531,
+      "learning_rate": 4.417433238320524e-06,
+      "loss": 0.4345,
+      "step": 2050
+    },
+    {
+      "epoch": 1.7579216899605248,
+      "grad_norm": 0.6606684327125549,
+      "learning_rate": 4.36810825639054e-06,
+      "loss": 0.4442,
+      "step": 2060
+    },
+    {
+      "epoch": 1.766456844126747,
+      "grad_norm": 0.7084365487098694,
+      "learning_rate": 4.318845689587727e-06,
+      "loss": 0.4348,
+      "step": 2070
+    },
+    {
+      "epoch": 1.7749919982929692,
+      "grad_norm": 0.6337569952011108,
+      "learning_rate": 4.269650403823698e-06,
+      "loss": 0.4384,
+      "step": 2080
+    },
+    {
+      "epoch": 1.7835271524591914,
+      "grad_norm": 0.6552258133888245,
+      "learning_rate": 4.220527258364373e-06,
+      "loss": 0.4376,
+      "step": 2090
+    },
+    {
+      "epoch": 1.7920623066254135,
+      "grad_norm": 0.6303353309631348,
+      "learning_rate": 4.171481105350021e-06,
+      "loss": 0.4373,
+      "step": 2100
+    },
+    {
+      "epoch": 1.8005974607916355,
+      "grad_norm": 0.650391697883606,
+      "learning_rate": 4.122516789315969e-06,
+      "loss": 0.4437,
+      "step": 2110
+    },
+    {
+      "epoch": 1.8091326149578577,
+      "grad_norm": 0.6542319059371948,
+      "learning_rate": 4.073639146714099e-06,
+      "loss": 0.4262,
+      "step": 2120
+    },
+    {
+      "epoch": 1.8176677691240797,
+      "grad_norm": 0.651104211807251,
+      "learning_rate": 4.024853005435121e-06,
+      "loss": 0.4363,
+      "step": 2130
+    },
+    {
+      "epoch": 1.8262029232903019,
+      "grad_norm": 0.6116424798965454,
+      "learning_rate": 3.976163184331702e-06,
+      "loss": 0.4291,
+      "step": 2140
+    },
+    {
+      "epoch": 1.834738077456524,
+      "grad_norm": 0.8228034377098083,
+      "learning_rate": 3.927574492742473e-06,
+      "loss": 0.4306,
+      "step": 2150
+    },
+    {
+      "epoch": 1.8432732316227463,
+      "grad_norm": 0.6177427768707275,
+      "learning_rate": 3.879091730017003e-06,
+      "loss": 0.4274,
+      "step": 2160
+    },
+    {
+      "epoch": 1.8518083857889684,
+      "grad_norm": 0.6379538178443909,
+      "learning_rate": 3.830719685041728e-06,
+      "loss": 0.4344,
+      "step": 2170
+    },
+    {
+      "epoch": 1.8603435399551904,
+      "grad_norm": 0.6659247279167175,
+      "learning_rate": 3.782463135766938e-06,
+      "loss": 0.4318,
+      "step": 2180
+    },
+    {
+      "epoch": 1.8688786941214126,
+      "grad_norm": 0.6846089959144592,
+      "learning_rate": 3.7343268487348273e-06,
+      "loss": 0.4354,
+      "step": 2190
+    },
+    {
+      "epoch": 1.8774138482876346,
+      "grad_norm": 0.6371617913246155,
+      "learning_rate": 3.6863155786086886e-06,
+      "loss": 0.4302,
+      "step": 2200
+    },
+    {
+      "epoch": 1.8859490024538568,
+      "grad_norm": 0.6230723857879639,
+      "learning_rate": 3.6384340677032616e-06,
+      "loss": 0.4367,
+      "step": 2210
+    },
+    {
+      "epoch": 1.894484156620079,
+      "grad_norm": 0.6282981634140015,
+      "learning_rate": 3.5906870455163175e-06,
+      "loss": 0.4284,
+      "step": 2220
+    },
+    {
+      "epoch": 1.9030193107863012,
+      "grad_norm": 0.6560613512992859,
+      "learning_rate": 3.5430792282614978e-06,
+      "loss": 0.4324,
+      "step": 2230
+    },
+    {
+      "epoch": 1.9115544649525233,
+      "grad_norm": 0.6765207052230835,
+      "learning_rate": 3.4956153184024765e-06,
+      "loss": 0.4261,
+      "step": 2240
+    },
+    {
+      "epoch": 1.9200896191187453,
+      "grad_norm": 0.6762336492538452,
+      "learning_rate": 3.448300004188463e-06,
+      "loss": 0.4304,
+      "step": 2250
+    },
+    {
+      "epoch": 1.9286247732849675,
+      "grad_norm": 0.6394000053405762,
+      "learning_rate": 3.401137959191133e-06,
+      "loss": 0.4294,
+      "step": 2260
+    },
+    {
+      "epoch": 1.9371599274511895,
+      "grad_norm": 0.6332142949104309,
+      "learning_rate": 3.354133841842989e-06,
+      "loss": 0.4308,
+      "step": 2270
+    },
+    {
+      "epoch": 1.9456950816174117,
+      "grad_norm": 0.6541900038719177,
+      "learning_rate": 3.3072922949772235e-06,
+      "loss": 0.4343,
+      "step": 2280
+    },
+    {
+      "epoch": 1.9542302357836339,
+      "grad_norm": 0.652773380279541,
+      "learning_rate": 3.2606179453691255e-06,
+      "loss": 0.4311,
+      "step": 2290
+    },
+    {
+      "epoch": 1.962765389949856,
+      "grad_norm": 0.6701607704162598,
+      "learning_rate": 3.21411540327907e-06,
+      "loss": 0.4244,
+      "step": 2300
+    },
+    {
+      "epoch": 1.9713005441160782,
+      "grad_norm": 0.6590484976768494,
+      "learning_rate": 3.16778926199714e-06,
+      "loss": 0.4366,
+      "step": 2310
+    },
+    {
+      "epoch": 1.9798356982823002,
+      "grad_norm": 0.6363989114761353,
+      "learning_rate": 3.1216440973894184e-06,
+      "loss": 0.4267,
+      "step": 2320
+    },
+    {
+      "epoch": 1.9883708524485222,
+      "grad_norm": 0.6702283024787903,
+      "learning_rate": 3.075684467446014e-06,
+      "loss": 0.4305,
+      "step": 2330
+    },
+    {
+      "epoch": 1.9969060066147444,
+      "grad_norm": 0.6762630939483643,
+      "learning_rate": 3.029914911830846e-06,
+      "loss": 0.4339,
+      "step": 2340
+    },
+    {
+      "epoch": 2.005121092499733,
+      "grad_norm": 0.6649417877197266,
+      "learning_rate": 2.9843399514332317e-06,
+      "loss": 0.4193,
+      "step": 2350
+    },
+    {
+      "epoch": 2.0136562466659553,
+      "grad_norm": 0.6555825471878052,
+      "learning_rate": 2.938964087921341e-06,
+      "loss": 0.4057,
+      "step": 2360
+    },
+    {
+      "epoch": 2.0221914008321775,
+      "grad_norm": 0.6427297592163086,
+      "learning_rate": 2.8937918032975434e-06,
+      "loss": 0.4045,
+      "step": 2370
+    },
+    {
+      "epoch": 2.0307265549983997,
+      "grad_norm": 0.6486126780509949,
+      "learning_rate": 2.848827559455696e-06,
+      "loss": 0.4111,
+      "step": 2380
+    },
+    {
+      "epoch": 2.039261709164622,
+      "grad_norm": 0.7112558484077454,
+      "learning_rate": 2.804075797740422e-06,
+      "loss": 0.413,
+      "step": 2390
+    },
+    {
+      "epoch": 2.047796863330844,
+      "grad_norm": 0.6829844117164612,
+      "learning_rate": 2.759540938508415e-06,
+      "loss": 0.4103,
+      "step": 2400
+    },
+    {
+      "epoch": 2.056332017497066,
+      "grad_norm": 0.6613876819610596,
+      "learning_rate": 2.7152273806918185e-06,
+      "loss": 0.4158,
+      "step": 2410
+    },
+    {
+      "epoch": 2.064867171663288,
+      "grad_norm": 0.6096106171607971,
+      "learning_rate": 2.6711395013637208e-06,
+      "loss": 0.4058,
+      "step": 2420
+    },
+    {
+      "epoch": 2.0734023258295102,
+      "grad_norm": 0.6055023074150085,
+      "learning_rate": 2.6272816553058156e-06,
+      "loss": 0.3974,
+      "step": 2430
+    },
+    {
+      "epoch": 2.0819374799957324,
+      "grad_norm": 0.6151232719421387,
+      "learning_rate": 2.5836581745782474e-06,
+      "loss": 0.4029,
+      "step": 2440
+    },
+    {
+      "epoch": 2.0904726341619546,
+      "grad_norm": 0.6236457228660583,
+      "learning_rate": 2.5402733680917163e-06,
+      "loss": 0.409,
+      "step": 2450
+    },
+    {
+      "epoch": 2.099007788328177,
+      "grad_norm": 0.6416062712669373,
+      "learning_rate": 2.497131521181877e-06,
+      "loss": 0.405,
+      "step": 2460
+    },
+    {
+      "epoch": 2.107542942494399,
+      "grad_norm": 0.6345195174217224,
+      "learning_rate": 2.4542368951860303e-06,
+      "loss": 0.3951,
+      "step": 2470
+    },
+    {
+      "epoch": 2.1160780966606207,
+      "grad_norm": 0.7058767676353455,
+      "learning_rate": 2.4115937270222305e-06,
+      "loss": 0.4047,
+      "step": 2480
+    },
+    {
+      "epoch": 2.124613250826843,
+      "grad_norm": 0.6685160398483276,
+      "learning_rate": 2.3692062287707795e-06,
+      "loss": 0.4037,
+      "step": 2490
+    },
+    {
+      "epoch": 2.133148404993065,
+      "grad_norm": 0.6508364677429199,
+      "learning_rate": 2.3270785872581698e-06,
+      "loss": 0.4044,
+      "step": 2500
+    },
+    {
+      "epoch": 2.1416835591592873,
+      "grad_norm": 0.6913980841636658,
+      "learning_rate": 2.2852149636435366e-06,
+      "loss": 0.4071,
+      "step": 2510
+    },
+    {
+      "epoch": 2.1502187133255095,
+      "grad_norm": 0.6269243955612183,
+      "learning_rate": 2.2436194930076448e-06,
+      "loss": 0.4131,
+      "step": 2520
+    },
+    {
+      "epoch": 2.1587538674917317,
+      "grad_norm": 0.6732710599899292,
+      "learning_rate": 2.2022962839444356e-06,
+      "loss": 0.4043,
+      "step": 2530
+    },
+    {
+      "epoch": 2.167289021657954,
+      "grad_norm": 0.658445417881012,
+      "learning_rate": 2.161249418155199e-06,
+      "loss": 0.4036,
+      "step": 2540
+    },
+    {
+      "epoch": 2.1758241758241756,
+      "grad_norm": 0.6448598504066467,
+      "learning_rate": 2.120482950045422e-06,
+      "loss": 0.4139,
+      "step": 2550
+    },
+    {
+      "epoch": 2.184359329990398,
+      "grad_norm": 0.6652196049690247,
+      "learning_rate": 2.080000906324292e-06,
+      "loss": 0.399,
+      "step": 2560
+    },
+    {
+      "epoch": 2.19289448415662,
+      "grad_norm": 0.6046453714370728,
+      "learning_rate": 2.0398072856069707e-06,
+      "loss": 0.4063,
+      "step": 2570
+    },
+    {
+      "epoch": 2.201429638322842,
+      "grad_norm": 0.6543176174163818,
+      "learning_rate": 1.9999060580196324e-06,
+      "loss": 0.4048,
+      "step": 2580
+    },
+    {
+      "epoch": 2.2099647924890644,
+      "grad_norm": 0.6523047089576721,
+      "learning_rate": 1.9603011648073055e-06,
+      "loss": 0.4073,
+      "step": 2590
+    },
+    {
+      "epoch": 2.2184999466552866,
+      "grad_norm": 0.6173551082611084,
+      "learning_rate": 1.9209965179445826e-06,
+      "loss": 0.4106,
+      "step": 2600
+    },
+    {
+      "epoch": 2.227035100821509,
+      "grad_norm": 0.6489232778549194,
+      "learning_rate": 1.881995999749215e-06,
+      "loss": 0.4103,
+      "step": 2610
+    },
+    {
+      "epoch": 2.2355702549877305,
+      "grad_norm": 0.6323755383491516,
+      "learning_rate": 1.8433034624986335e-06,
+      "loss": 0.4012,
+      "step": 2620
+    },
+    {
+      "epoch": 2.2441054091539527,
+      "grad_norm": 0.640560507774353,
+      "learning_rate": 1.8049227280494391e-06,
+      "loss": 0.4048,
+      "step": 2630
+    },
+    {
+      "epoch": 2.252640563320175,
+      "grad_norm": 0.6368593573570251,
+      "learning_rate": 1.7668575874599086e-06,
+      "loss": 0.4079,
+      "step": 2640
+    },
+    {
+      "epoch": 2.261175717486397,
+      "grad_norm": 0.6442722678184509,
+      "learning_rate": 1.7291118006155178e-06,
+      "loss": 0.4144,
+      "step": 2650
+    },
+    {
+      "epoch": 2.2697108716526193,
+      "grad_norm": 0.7017993330955505,
+      "learning_rate": 1.6916890958575687e-06,
+      "loss": 0.4107,
+      "step": 2660
+    },
+    {
+      "epoch": 2.2782460258188415,
+      "grad_norm": 0.6456113457679749,
+      "learning_rate": 1.6545931696149208e-06,
+      "loss": 0.4081,
+      "step": 2670
+    },
+    {
+      "epoch": 2.2867811799850637,
+      "grad_norm": 0.6799325346946716,
+      "learning_rate": 1.6178276860388736e-06,
+      "loss": 0.4022,
+      "step": 2680
+    },
+    {
+      "epoch": 2.2953163341512854,
+      "grad_norm": 0.6276426911354065,
+      "learning_rate": 1.5813962766412427e-06,
+      "loss": 0.4127,
+      "step": 2690
+    },
+    {
+      "epoch": 2.3038514883175076,
+      "grad_norm": 0.6316121816635132,
+      "learning_rate": 1.5453025399356542e-06,
+      "loss": 0.4064,
+      "step": 2700
+    },
+    {
+      "epoch": 2.31238664248373,
+      "grad_norm": 0.6637789011001587,
+      "learning_rate": 1.5095500410821085e-06,
+      "loss": 0.4089,
+      "step": 2710
+    },
+    {
+      "epoch": 2.320921796649952,
+      "grad_norm": 0.6297438740730286,
+      "learning_rate": 1.4741423115348203e-06,
+      "loss": 0.4012,
+      "step": 2720
+    },
+    {
+      "epoch": 2.329456950816174,
+      "grad_norm": 0.6420316100120544,
+      "learning_rate": 1.439082848693406e-06,
+      "loss": 0.403,
+      "step": 2730
+    },
+    {
+      "epoch": 2.3379921049823964,
+      "grad_norm": 0.6523280143737793,
+      "learning_rate": 1.4043751155574331e-06,
+      "loss": 0.4037,
+      "step": 2740
+    },
+    {
+      "epoch": 2.346527259148618,
+      "grad_norm": 0.6407163739204407,
+      "learning_rate": 1.370022540384347e-06,
+      "loss": 0.4044,
+      "step": 2750
+    },
+    {
+      "epoch": 2.3550624133148403,
+      "grad_norm": 0.6137824654579163,
+      "learning_rate": 1.3360285163508585e-06,
+      "loss": 0.4065,
+      "step": 2760
+    },
+    {
+      "epoch": 2.3635975674810625,
+      "grad_norm": 0.6386251449584961,
+      "learning_rate": 1.3023964012177808e-06,
+      "loss": 0.3996,
+      "step": 2770
+    },
+    {
+      "epoch": 2.3721327216472847,
+      "grad_norm": 0.6598160266876221,
+      "learning_rate": 1.2691295169983602e-06,
+      "loss": 0.407,
+      "step": 2780
+    },
+    {
+      "epoch": 2.380667875813507,
+      "grad_norm": 0.6845166683197021,
+      "learning_rate": 1.2362311496301465e-06,
+      "loss": 0.4093,
+      "step": 2790
+    },
+    {
+      "epoch": 2.389203029979729,
+      "grad_norm": 0.6791868209838867,
+      "learning_rate": 1.2037045486504317e-06,
+      "loss": 0.401,
+      "step": 2800
+    },
+    {
+      "epoch": 2.3977381841459513,
+      "grad_norm": 0.6840428113937378,
+      "learning_rate": 1.171552926875265e-06,
+      "loss": 0.4052,
+      "step": 2810
+    },
+    {
+      "epoch": 2.4062733383121735,
+      "grad_norm": 0.6226551532745361,
+      "learning_rate": 1.1397794600821155e-06,
+      "loss": 0.4041,
+      "step": 2820
+    },
+    {
+      "epoch": 2.4148084924783952,
+      "grad_norm": 0.6312896013259888,
+      "learning_rate": 1.1083872866961816e-06,
+      "loss": 0.4033,
+      "step": 2830
+    },
+    {
+      "epoch": 2.4233436466446174,
+      "grad_norm": 0.6522102952003479,
+      "learning_rate": 1.0773795074803917e-06,
+      "loss": 0.3987,
+      "step": 2840
+    },
+    {
+      "epoch": 2.4318788008108396,
+      "grad_norm": 0.6765321493148804,
+      "learning_rate": 1.0467591852291287e-06,
+      "loss": 0.4059,
+      "step": 2850
+    },
+    {
+      "epoch": 2.440413954977062,
+      "grad_norm": 0.6331756114959717,
+      "learning_rate": 1.0165293444657037e-06,
+      "loss": 0.401,
+      "step": 2860
+    },
+    {
+      "epoch": 2.448949109143284,
+      "grad_norm": 0.6368128657341003,
+      "learning_rate": 9.866929711436035e-07,
+      "loss": 0.4136,
+      "step": 2870
+    },
+    {
+      "epoch": 2.457484263309506,
+      "grad_norm": 0.6715298891067505,
+      "learning_rate": 9.57253012351556e-07,
+      "loss": 0.4157,
+      "step": 2880
+    },
+    {
+      "epoch": 2.466019417475728,
+      "grad_norm": 0.6294402480125427,
+      "learning_rate": 9.282123760224327e-07,
+      "loss": 0.4058,
+      "step": 2890
+    },
+    {
+      "epoch": 2.47455457164195,
+      "grad_norm": 0.6221645474433899,
+      "learning_rate": 8.99573930646015e-07,
+      "loss": 0.4014,
+      "step": 2900
+    },
+    {
+      "epoch": 2.4830897258081723,
+      "grad_norm": 0.642346978187561,
+      "learning_rate": 8.713405049856588e-07,
+      "loss": 0.41,
+      "step": 2910
+    },
+    {
+      "epoch": 2.4916248799743945,
+      "grad_norm": 0.628089189529419,
+      "learning_rate": 8.435148877988886e-07,
+      "loss": 0.4084,
+      "step": 2920
+    },
+    {
+      "epoch": 2.5001600341406167,
+      "grad_norm": 0.637294352054596,
+      "learning_rate": 8.160998275619309e-07,
+      "loss": 0.4025,
+      "step": 2930
+    },
+    {
+      "epoch": 2.508695188306839,
+      "grad_norm": 0.6054239273071289,
+      "learning_rate": 7.890980321982355e-07,
+      "loss": 0.405,
+      "step": 2940
+    },
+    {
+      "epoch": 2.517230342473061,
+      "grad_norm": 0.6559164524078369,
+      "learning_rate": 7.625121688110027e-07,
+      "loss": 0.4043,
+      "step": 2950
+    },
+    {
+      "epoch": 2.5257654966392833,
+      "grad_norm": 0.665451169013977,
+      "learning_rate": 7.363448634197412e-07,
+      "loss": 0.4028,
+      "step": 2960
+    },
+    {
+      "epoch": 2.534300650805505,
+      "grad_norm": 0.6523730158805847,
+      "learning_rate": 7.105987007008724e-07,
+      "loss": 0.4068,
+      "step": 2970
+    },
+    {
+      "epoch": 2.5428358049717272,
+      "grad_norm": 0.6315779089927673,
+      "learning_rate": 6.852762237324462e-07,
+      "loss": 0.4092,
+      "step": 2980
+    },
+    {
+      "epoch": 2.5513709591379494,
+      "grad_norm": 0.6356623768806458,
+      "learning_rate": 6.603799337429328e-07,
+      "loss": 0.4099,
+      "step": 2990
+    },
+    {
+      "epoch": 2.5599061133041716,
+      "grad_norm": 0.6033786535263062,
+      "learning_rate": 6.359122898641717e-07,
+      "loss": 0.4084,
+      "step": 3000
+    },
+    {
+      "epoch": 2.568441267470394,
+      "grad_norm": 0.6288996338844299,
+      "learning_rate": 6.118757088884686e-07,
+      "loss": 0.4131,
+      "step": 3010
+    },
+    {
+      "epoch": 2.5769764216366156,
+      "grad_norm": 0.6407638192176819,
+      "learning_rate": 5.882725650298787e-07,
+      "loss": 0.409,
+      "step": 3020
+    },
+    {
+      "epoch": 2.5855115758028377,
+      "grad_norm": 0.6419194340705872,
+      "learning_rate": 5.651051896896908e-07,
+      "loss": 0.4038,
+      "step": 3030
+    },
+    {
+      "epoch": 2.59404672996906,
+      "grad_norm": 0.6341270208358765,
+      "learning_rate": 5.42375871226144e-07,
+      "loss": 0.4064,
+      "step": 3040
+    },
+    {
+      "epoch": 2.602581884135282,
+      "grad_norm": 0.647345244884491,
+      "learning_rate": 5.200868547283972e-07,
+      "loss": 0.4081,
+      "step": 3050
+    },
+    {
+      "epoch": 2.6111170383015043,
+      "grad_norm": 0.6366192102432251,
+      "learning_rate": 4.982403417947662e-07,
+      "loss": 0.4033,
+      "step": 3060
+    },
+    {
+      "epoch": 2.6196521924677265,
+      "grad_norm": 0.638552725315094,
+      "learning_rate": 4.768384903152634e-07,
+      "loss": 0.4096,
+      "step": 3070
+    },
+    {
+      "epoch": 2.6281873466339487,
+      "grad_norm": 0.6206834316253662,
+      "learning_rate": 4.558834142584556e-07,
+      "loss": 0.4065,
+      "step": 3080
+    },
+    {
+      "epoch": 2.636722500800171,
+      "grad_norm": 0.6011072993278503,
+      "learning_rate": 4.3537718346264857e-07,
+      "loss": 0.4029,
+      "step": 3090
+    },
+    {
+      "epoch": 2.645257654966393,
+      "grad_norm": 0.6577970385551453,
+      "learning_rate": 4.153218234314443e-07,
+      "loss": 0.4119,
+      "step": 3100
+    },
+    {
+      "epoch": 2.653792809132615,
+      "grad_norm": 0.6269943118095398,
+      "learning_rate": 3.957193151336719e-07,
+      "loss": 0.4078,
+      "step": 3110
+    },
+    {
+      "epoch": 2.662327963298837,
+      "grad_norm": 0.6127352118492126,
+      "learning_rate": 3.765715948077137e-07,
+      "loss": 0.4014,
+      "step": 3120
+    },
+    {
+      "epoch": 2.6708631174650592,
+      "grad_norm": 0.6328156590461731,
+      "learning_rate": 3.578805537702573e-07,
+      "loss": 0.4057,
+      "step": 3130
+    },
+    {
+      "epoch": 2.6793982716312814,
+      "grad_norm": 0.6262829899787903,
+      "learning_rate": 3.396480382294792e-07,
+      "loss": 0.4026,
+      "step": 3140
+    },
+    {
+      "epoch": 2.6879334257975036,
+      "grad_norm": 0.631956934928894,
+      "learning_rate": 3.2187584910268464e-07,
+      "loss": 0.4003,
+      "step": 3150
+    },
+    {
+      "epoch": 2.6964685799637254,
+      "grad_norm": 0.628641664981842,
+      "learning_rate": 3.045657418384218e-07,
+      "loss": 0.4075,
+      "step": 3160
+    },
+    {
+      "epoch": 2.7050037341299475,
+      "grad_norm": 0.6435490250587463,
+      "learning_rate": 2.877194262430916e-07,
+      "loss": 0.3966,
+      "step": 3170
+    },
+    {
+      "epoch": 2.7135388882961697,
+      "grad_norm": 0.6484223008155823,
+      "learning_rate": 2.7133856631205216e-07,
+      "loss": 0.4076,
+      "step": 3180
+    },
+    {
+      "epoch": 2.722074042462392,
+      "grad_norm": 0.630936324596405,
+      "learning_rate": 2.5542478006526703e-07,
+      "loss": 0.3984,
+      "step": 3190
+    },
+    {
+      "epoch": 2.730609196628614,
+      "grad_norm": 0.6019722819328308,
+      "learning_rate": 2.399796393874837e-07,
+      "loss": 0.4045,
+      "step": 3200
+    },
+    {
+      "epoch": 2.7391443507948363,
+      "grad_norm": 0.6216541528701782,
+      "learning_rate": 2.2500466987296466e-07,
+      "loss": 0.4136,
+      "step": 3210
+    },
+    {
+      "epoch": 2.7476795049610585,
+      "grad_norm": 0.6513262987136841,
+      "learning_rate": 2.105013506748027e-07,
+      "loss": 0.405,
+      "step": 3220
+    },
+    {
+      "epoch": 2.7562146591272807,
+      "grad_norm": 0.6242448091506958,
+      "learning_rate": 1.9647111435881494e-07,
+      "loss": 0.4,
+      "step": 3230
+    },
+    {
+      "epoch": 2.764749813293503,
+      "grad_norm": 0.6355422139167786,
+      "learning_rate": 1.829153467620426e-07,
+      "loss": 0.402,
+      "step": 3240
+    },
+    {
+      "epoch": 2.7732849674597246,
+      "grad_norm": 0.5879780054092407,
+      "learning_rate": 1.6983538685586277e-07,
+      "loss": 0.4015,
+      "step": 3250
+    },
+    {
+      "epoch": 2.781820121625947,
+      "grad_norm": 0.6436604261398315,
+      "learning_rate": 1.5723252661373367e-07,
+      "loss": 0.4025,
+      "step": 3260
+    },
+    {
+      "epoch": 2.790355275792169,
+      "grad_norm": 0.6310527920722961,
+      "learning_rate": 1.451080108835784e-07,
+      "loss": 0.406,
+      "step": 3270
+    },
+    {
+      "epoch": 2.798890429958391,
+      "grad_norm": 0.6294417381286621,
+      "learning_rate": 1.3346303726482544e-07,
+      "loss": 0.3997,
+      "step": 3280
+    },
+    {
+      "epoch": 2.8074255841246134,
+      "grad_norm": 0.6856746077537537,
+      "learning_rate": 1.222987559901162e-07,
+      "loss": 0.4059,
+      "step": 3290
+    },
+    {
+      "epoch": 2.815960738290835,
+      "grad_norm": 0.8050450682640076,
+      "learning_rate": 1.1161626981169072e-07,
+      "loss": 0.4063,
+      "step": 3300
+    },
+    {
+      "epoch": 2.8244958924570573,
+      "grad_norm": 0.6034400463104248,
+      "learning_rate": 1.014166338924627e-07,
+      "loss": 0.3985,
+      "step": 3310
+    },
+    {
+      "epoch": 2.8330310466232795,
+      "grad_norm": 0.6405271291732788,
+      "learning_rate": 9.17008557017962e-08,
+      "loss": 0.3951,
+      "step": 3320
+    },
+    {
+      "epoch": 2.8415662007895017,
+      "grad_norm": 0.6416633725166321,
+      "learning_rate": 8.24698949159919e-08,
+      "loss": 0.4049,
+      "step": 3330
+    },
+    {
+      "epoch": 2.850101354955724,
+      "grad_norm": 0.6081260442733765,
+      "learning_rate": 7.372466332349726e-08,
+      "loss": 0.4023,
+      "step": 3340
+    },
+    {
+      "epoch": 2.858636509121946,
+      "grad_norm": 0.6582528352737427,
+      "learning_rate": 6.546602473484153e-08,
+      "loss": 0.4095,
+      "step": 3350
+    },
+    {
+      "epoch": 2.8671716632881683,
+      "grad_norm": 0.6263580918312073,
+      "learning_rate": 5.769479489731666e-08,
+      "loss": 0.4099,
+      "step": 3360
+    },
+    {
+      "epoch": 2.8757068174543905,
+      "grad_norm": 0.6049970984458923,
+      "learning_rate": 5.0411741414396756e-08,
+      "loss": 0.3992,
+      "step": 3370
+    },
+    {
+      "epoch": 2.8842419716206122,
+      "grad_norm": 0.6298121809959412,
+      "learning_rate": 4.3617583669922636e-08,
+      "loss": 0.4035,
+      "step": 3380
+    },
+    {
+      "epoch": 2.8927771257868344,
+      "grad_norm": 0.617342472076416,
+      "learning_rate": 3.73129927570437e-08,
+      "loss": 0.4079,
+      "step": 3390
+    },
+    {
+      "epoch": 2.9013122799530566,
+      "grad_norm": 0.6217492818832397,
+      "learning_rate": 3.1498591411928125e-08,
+      "loss": 0.4021,
+      "step": 3400
+    },
+    {
+      "epoch": 2.909847434119279,
+      "grad_norm": 0.618754506111145,
+      "learning_rate": 2.6174953952254334e-08,
+      "loss": 0.3977,
+      "step": 3410
+    },
+    {
+      "epoch": 2.918382588285501,
+      "grad_norm": 0.5799270868301392,
+      "learning_rate": 2.134260622048301e-08,
+      "loss": 0.394,
+      "step": 3420
+    },
+    {
+      "epoch": 2.9269177424517228,
+      "grad_norm": 0.6517247557640076,
+      "learning_rate": 1.700202553191588e-08,
+      "loss": 0.4023,
+      "step": 3430
+    },
+    {
+      "epoch": 2.935452896617945,
+      "grad_norm": 0.6470068693161011,
+      "learning_rate": 1.315364062754898e-08,
+      "loss": 0.4056,
+      "step": 3440
+    },
+    {
+      "epoch": 2.943988050784167,
+      "grad_norm": 0.6544038653373718,
+      "learning_rate": 9.797831631724319e-09,
+      "loss": 0.4105,
+      "step": 3450
+    },
+    {
+      "epoch": 2.9525232049503893,
+      "grad_norm": 0.6202620267868042,
+      "learning_rate": 6.9349300145826745e-09,
+      "loss": 0.4102,
+      "step": 3460
+    },
+    {
+      "epoch": 2.9610583591166115,
+      "grad_norm": 0.6056284308433533,
+      "learning_rate": 4.56521855932257e-09,
+      "loss": 0.4006,
+      "step": 3470
+    },
+    {
+      "epoch": 2.9695935132828337,
+      "grad_norm": 0.6306059956550598,
+      "learning_rate": 2.6889313342692848e-09,
+      "loss": 0.404,
+      "step": 3480
+    },
+    {
+      "epoch": 2.978128667449056,
+      "grad_norm": 0.6009540557861328,
+      "learning_rate": 1.3062536697533433e-09,
+      "loss": 0.3995,
+      "step": 3490
+    },
+    {
+      "epoch": 2.986663821615278,
+      "grad_norm": 0.649553656578064,
+      "learning_rate": 4.173221398046101e-10,
+      "loss": 0.4005,
+      "step": 3500
+    },
+    {
+      "epoch": 2.9951989757815003,
+      "grad_norm": 0.6136988401412964,
+      "learning_rate": 2.2224548664184777e-11,
+      "loss": 0.4075,
+      "step": 3510
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 3513,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 3,
+  "save_steps": 500.0,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.1182668447261655e+18,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}

config.json

@@ -0,0 +1,36 @@
+{
+  "_name_or_path": "meta-llama/Llama-3.2-1B",
+  "architectures": [
+    "LlamaForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "bos_token_id": 128000,
+  "eos_token_id": 128001,
+  "head_dim": 64,
+  "hidden_act": "silu",
+  "hidden_size": 2048,
+  "initializer_range": 0.02,
+  "intermediate_size": 8192,
+  "max_position_embeddings": 131072,
+  "mlp_bias": false,
+  "model_type": "llama",
+  "num_attention_heads": 32,
+  "num_hidden_layers": 16,
+  "num_key_value_heads": 8,
+  "pretraining_tp": 1,
+  "rms_norm_eps": 1e-05,
+  "rope_scaling": {
+    "factor": 32.0,
+    "high_freq_factor": 4.0,
+    "low_freq_factor": 1.0,
+    "original_max_position_embeddings": 8192,
+    "rope_type": "llama3"
+  },
+  "rope_theta": 500000.0,
+  "tie_word_embeddings": true,
+  "torch_dtype": "float32",
+  "transformers_version": "4.48.1",
+  "use_cache": false,
+  "vocab_size": 128256
+}