diff --git "a/last-checkpoint/trainer_state.json" "b/last-checkpoint/trainer_state.json"
--- "a/last-checkpoint/trainer_state.json"
+++ "b/last-checkpoint/trainer_state.json"
@@ -1,9 +1,9 @@
 {
   "best_metric": null,
   "best_model_checkpoint": null,
-  "epoch": 2.968063872255489,
+  "epoch": 3.952428476380572,
   "eval_steps": 500,
-  "global_step": 2253,
+  "global_step": 3004,
   "is_hyper_param_search": false,
   "is_local_process_zero": true,
   "is_world_process_zero": true,
@@ -15778,6 +15778,5263 @@
       "learning_rate": 7.368764016743712e-07,
       "loss": 0.7304,
       "step": 2253
+    },
+    {
+      "epoch": 2.97,
+      "grad_norm": 2.734375,
+      "learning_rate": 7.350176020987585e-07,
+      "loss": 0.7597,
+      "step": 2254
+    },
+    {
+      "epoch": 2.97,
+      "grad_norm": 2.84375,
+      "learning_rate": 7.331607458081169e-07,
+      "loss": 0.7793,
+      "step": 2255
+    },
+    {
+      "epoch": 2.97,
+      "grad_norm": 2.6875,
+      "learning_rate": 7.313058348468866e-07,
+      "loss": 0.7716,
+      "step": 2256
+    },
+    {
+      "epoch": 2.97,
+      "grad_norm": 2.65625,
+      "learning_rate": 7.29452871257367e-07,
+      "loss": 0.7296,
+      "step": 2257
+    },
+    {
+      "epoch": 2.97,
+      "grad_norm": 2.75,
+      "learning_rate": 7.276018570797119e-07,
+      "loss": 0.766,
+      "step": 2258
+    },
+    {
+      "epoch": 2.98,
+      "grad_norm": 2.671875,
+      "learning_rate": 7.257527943519299e-07,
+      "loss": 0.7454,
+      "step": 2259
+    },
+    {
+      "epoch": 2.98,
+      "grad_norm": 2.65625,
+      "learning_rate": 7.239056851098785e-07,
+      "loss": 0.7622,
+      "step": 2260
+    },
+    {
+      "epoch": 2.98,
+      "grad_norm": 2.75,
+      "learning_rate": 7.220605313872672e-07,
+      "loss": 0.7468,
+      "step": 2261
+    },
+    {
+      "epoch": 2.98,
+      "grad_norm": 2.65625,
+      "learning_rate": 7.202173352156511e-07,
+      "loss": 0.7373,
+      "step": 2262
+    },
+    {
+      "epoch": 2.98,
+      "grad_norm": 2.625,
+      "learning_rate": 7.183760986244307e-07,
+      "loss": 0.7259,
+      "step": 2263
+    },
+    {
+      "epoch": 2.98,
+      "grad_norm": 2.703125,
+      "learning_rate": 7.16536823640848e-07,
+      "loss": 0.755,
+      "step": 2264
+    },
+    {
+      "epoch": 2.98,
+      "grad_norm": 2.6875,
+      "learning_rate": 7.146995122899861e-07,
+      "loss": 0.7958,
+      "step": 2265
+    },
+    {
+      "epoch": 2.99,
+      "grad_norm": 2.703125,
+      "learning_rate": 7.128641665947658e-07,
+      "loss": 0.7552,
+      "step": 2266
+    },
+    {
+      "epoch": 2.99,
+      "grad_norm": 2.78125,
+      "learning_rate": 7.110307885759435e-07,
+      "loss": 0.7518,
+      "step": 2267
+    },
+    {
+      "epoch": 2.99,
+      "grad_norm": 2.703125,
+      "learning_rate": 7.091993802521094e-07,
+      "loss": 0.7552,
+      "step": 2268
+    },
+    {
+      "epoch": 2.99,
+      "grad_norm": 2.703125,
+      "learning_rate": 7.073699436396847e-07,
+      "loss": 0.7494,
+      "step": 2269
+    },
+    {
+      "epoch": 2.99,
+      "grad_norm": 2.734375,
+      "learning_rate": 7.055424807529199e-07,
+      "loss": 0.7877,
+      "step": 2270
+    },
+    {
+      "epoch": 2.99,
+      "grad_norm": 2.6875,
+      "learning_rate": 7.037169936038935e-07,
+      "loss": 0.7652,
+      "step": 2271
+    },
+    {
+      "epoch": 2.99,
+      "grad_norm": 2.734375,
+      "learning_rate": 7.018934842025058e-07,
+      "loss": 0.75,
+      "step": 2272
+    },
+    {
+      "epoch": 2.99,
+      "grad_norm": 2.6875,
+      "learning_rate": 7.000719545564819e-07,
+      "loss": 0.738,
+      "step": 2273
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 2.703125,
+      "learning_rate": 6.982524066713658e-07,
+      "loss": 0.7532,
+      "step": 2274
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 2.75,
+      "learning_rate": 6.964348425505222e-07,
+      "loss": 0.7637,
+      "step": 2275
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 2.6875,
+      "learning_rate": 6.946192641951285e-07,
+      "loss": 0.7558,
+      "step": 2276
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 2.703125,
+      "learning_rate": 6.928056736041783e-07,
+      "loss": 0.7313,
+      "step": 2277
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 2.703125,
+      "learning_rate": 6.90994072774473e-07,
+      "loss": 0.7377,
+      "step": 2278
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 2.75,
+      "learning_rate": 6.891844637006268e-07,
+      "loss": 0.733,
+      "step": 2279
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 2.609375,
+      "learning_rate": 6.873768483750595e-07,
+      "loss": 0.7361,
+      "step": 2280
+    },
+    {
+      "epoch": 3.01,
+      "grad_norm": 2.6875,
+      "learning_rate": 6.855712287879957e-07,
+      "loss": 0.7741,
+      "step": 2281
+    },
+    {
+      "epoch": 3.01,
+      "grad_norm": 2.75,
+      "learning_rate": 6.837676069274632e-07,
+      "loss": 0.7451,
+      "step": 2282
+    },
+    {
+      "epoch": 3.01,
+      "grad_norm": 2.734375,
+      "learning_rate": 6.819659847792895e-07,
+      "loss": 0.7632,
+      "step": 2283
+    },
+    {
+      "epoch": 3.01,
+      "grad_norm": 2.75,
+      "learning_rate": 6.801663643271012e-07,
+      "loss": 0.765,
+      "step": 2284
+    },
+    {
+      "epoch": 3.01,
+      "grad_norm": 2.71875,
+      "learning_rate": 6.783687475523199e-07,
+      "loss": 0.7495,
+      "step": 2285
+    },
+    {
+      "epoch": 3.01,
+      "grad_norm": 2.765625,
+      "learning_rate": 6.765731364341622e-07,
+      "loss": 0.7506,
+      "step": 2286
+    },
+    {
+      "epoch": 3.01,
+      "grad_norm": 2.703125,
+      "learning_rate": 6.747795329496359e-07,
+      "loss": 0.7544,
+      "step": 2287
+    },
+    {
+      "epoch": 3.01,
+      "grad_norm": 2.703125,
+      "learning_rate": 6.729879390735391e-07,
+      "loss": 0.7334,
+      "step": 2288
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 2.859375,
+      "learning_rate": 6.711983567784552e-07,
+      "loss": 0.7458,
+      "step": 2289
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 2.734375,
+      "learning_rate": 6.69410788034755e-07,
+      "loss": 0.736,
+      "step": 2290
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 2.8125,
+      "learning_rate": 6.676252348105916e-07,
+      "loss": 0.7523,
+      "step": 2291
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 2.9375,
+      "learning_rate": 6.658416990718986e-07,
+      "loss": 0.743,
+      "step": 2292
+    },
+    {
+      "epoch": 3.01,
+      "grad_norm": 2.671875,
+      "learning_rate": 6.640601827823878e-07,
+      "loss": 0.7135,
+      "step": 2293
+    },
+    {
+      "epoch": 3.01,
+      "grad_norm": 2.859375,
+      "learning_rate": 6.622806879035512e-07,
+      "loss": 0.7385,
+      "step": 2294
+    },
+    {
+      "epoch": 3.01,
+      "grad_norm": 2.828125,
+      "learning_rate": 6.605032163946496e-07,
+      "loss": 0.7395,
+      "step": 2295
+    },
+    {
+      "epoch": 3.01,
+      "grad_norm": 2.875,
+      "learning_rate": 6.587277702127196e-07,
+      "loss": 0.7255,
+      "step": 2296
+    },
+    {
+      "epoch": 3.01,
+      "grad_norm": 2.84375,
+      "learning_rate": 6.56954351312567e-07,
+      "loss": 0.754,
+      "step": 2297
+    },
+    {
+      "epoch": 3.01,
+      "grad_norm": 2.6875,
+      "learning_rate": 6.551829616467656e-07,
+      "loss": 0.7245,
+      "step": 2298
+    },
+    {
+      "epoch": 3.01,
+      "grad_norm": 2.75,
+      "learning_rate": 6.534136031656552e-07,
+      "loss": 0.7448,
+      "step": 2299
+    },
+    {
+      "epoch": 3.02,
+      "grad_norm": 2.6875,
+      "learning_rate": 6.516462778173383e-07,
+      "loss": 0.7134,
+      "step": 2300
+    },
+    {
+      "epoch": 3.02,
+      "grad_norm": 2.75,
+      "learning_rate": 6.498809875476799e-07,
+      "loss": 0.7394,
+      "step": 2301
+    },
+    {
+      "epoch": 3.02,
+      "grad_norm": 2.8125,
+      "learning_rate": 6.481177343003043e-07,
+      "loss": 0.7496,
+      "step": 2302
+    },
+    {
+      "epoch": 3.02,
+      "grad_norm": 2.828125,
+      "learning_rate": 6.463565200165919e-07,
+      "loss": 0.75,
+      "step": 2303
+    },
+    {
+      "epoch": 3.02,
+      "grad_norm": 2.78125,
+      "learning_rate": 6.445973466356794e-07,
+      "loss": 0.7282,
+      "step": 2304
+    },
+    {
+      "epoch": 3.02,
+      "grad_norm": 2.6875,
+      "learning_rate": 6.428402160944558e-07,
+      "loss": 0.7266,
+      "step": 2305
+    },
+    {
+      "epoch": 3.02,
+      "grad_norm": 2.8125,
+      "learning_rate": 6.410851303275617e-07,
+      "loss": 0.7538,
+      "step": 2306
+    },
+    {
+      "epoch": 3.02,
+      "grad_norm": 2.703125,
+      "learning_rate": 6.393320912673842e-07,
+      "loss": 0.7137,
+      "step": 2307
+    },
+    {
+      "epoch": 3.03,
+      "grad_norm": 2.671875,
+      "learning_rate": 6.375811008440591e-07,
+      "loss": 0.7237,
+      "step": 2308
+    },
+    {
+      "epoch": 3.03,
+      "grad_norm": 2.90625,
+      "learning_rate": 6.358321609854659e-07,
+      "loss": 0.7391,
+      "step": 2309
+    },
+    {
+      "epoch": 3.03,
+      "grad_norm": 2.84375,
+      "learning_rate": 6.340852736172265e-07,
+      "loss": 0.7325,
+      "step": 2310
+    },
+    {
+      "epoch": 3.03,
+      "grad_norm": 2.78125,
+      "learning_rate": 6.323404406627018e-07,
+      "loss": 0.7259,
+      "step": 2311
+    },
+    {
+      "epoch": 3.03,
+      "grad_norm": 2.75,
+      "learning_rate": 6.305976640429945e-07,
+      "loss": 0.7299,
+      "step": 2312
+    },
+    {
+      "epoch": 3.03,
+      "grad_norm": 2.796875,
+      "learning_rate": 6.288569456769377e-07,
+      "loss": 0.7378,
+      "step": 2313
+    },
+    {
+      "epoch": 3.03,
+      "grad_norm": 2.75,
+      "learning_rate": 6.271182874811024e-07,
+      "loss": 0.7302,
+      "step": 2314
+    },
+    {
+      "epoch": 3.04,
+      "grad_norm": 2.796875,
+      "learning_rate": 6.253816913697896e-07,
+      "loss": 0.7195,
+      "step": 2315
+    },
+    {
+      "epoch": 3.04,
+      "grad_norm": 2.65625,
+      "learning_rate": 6.236471592550305e-07,
+      "loss": 0.7367,
+      "step": 2316
+    },
+    {
+      "epoch": 3.04,
+      "grad_norm": 2.8125,
+      "learning_rate": 6.219146930465838e-07,
+      "loss": 0.7477,
+      "step": 2317
+    },
+    {
+      "epoch": 3.04,
+      "grad_norm": 2.71875,
+      "learning_rate": 6.201842946519343e-07,
+      "loss": 0.7284,
+      "step": 2318
+    },
+    {
+      "epoch": 3.04,
+      "grad_norm": 2.84375,
+      "learning_rate": 6.184559659762865e-07,
+      "loss": 0.741,
+      "step": 2319
+    },
+    {
+      "epoch": 3.04,
+      "grad_norm": 2.71875,
+      "learning_rate": 6.167297089225713e-07,
+      "loss": 0.7472,
+      "step": 2320
+    },
+    {
+      "epoch": 3.04,
+      "grad_norm": 2.734375,
+      "learning_rate": 6.150055253914353e-07,
+      "loss": 0.724,
+      "step": 2321
+    },
+    {
+      "epoch": 3.04,
+      "grad_norm": 2.75,
+      "learning_rate": 6.132834172812432e-07,
+      "loss": 0.7345,
+      "step": 2322
+    },
+    {
+      "epoch": 3.05,
+      "grad_norm": 2.765625,
+      "learning_rate": 6.115633864880743e-07,
+      "loss": 0.7521,
+      "step": 2323
+    },
+    {
+      "epoch": 3.05,
+      "grad_norm": 2.796875,
+      "learning_rate": 6.098454349057218e-07,
+      "loss": 0.7718,
+      "step": 2324
+    },
+    {
+      "epoch": 3.05,
+      "grad_norm": 2.84375,
+      "learning_rate": 6.081295644256868e-07,
+      "loss": 0.7365,
+      "step": 2325
+    },
+    {
+      "epoch": 3.05,
+      "grad_norm": 2.75,
+      "learning_rate": 6.064157769371823e-07,
+      "loss": 0.7374,
+      "step": 2326
+    },
+    {
+      "epoch": 3.05,
+      "grad_norm": 2.65625,
+      "learning_rate": 6.04704074327126e-07,
+      "loss": 0.7485,
+      "step": 2327
+    },
+    {
+      "epoch": 3.05,
+      "grad_norm": 2.75,
+      "learning_rate": 6.029944584801408e-07,
+      "loss": 0.7654,
+      "step": 2328
+    },
+    {
+      "epoch": 3.05,
+      "grad_norm": 2.78125,
+      "learning_rate": 6.012869312785513e-07,
+      "loss": 0.7542,
+      "step": 2329
+    },
+    {
+      "epoch": 3.06,
+      "grad_norm": 2.75,
+      "learning_rate": 5.995814946023848e-07,
+      "loss": 0.7427,
+      "step": 2330
+    },
+    {
+      "epoch": 3.06,
+      "grad_norm": 2.78125,
+      "learning_rate": 5.978781503293634e-07,
+      "loss": 0.7387,
+      "step": 2331
+    },
+    {
+      "epoch": 3.06,
+      "grad_norm": 2.78125,
+      "learning_rate": 5.961769003349077e-07,
+      "loss": 0.7572,
+      "step": 2332
+    },
+    {
+      "epoch": 3.06,
+      "grad_norm": 2.8125,
+      "learning_rate": 5.944777464921317e-07,
+      "loss": 0.7625,
+      "step": 2333
+    },
+    {
+      "epoch": 3.06,
+      "grad_norm": 2.671875,
+      "learning_rate": 5.92780690671842e-07,
+      "loss": 0.7181,
+      "step": 2334
+    },
+    {
+      "epoch": 3.06,
+      "grad_norm": 2.65625,
+      "learning_rate": 5.910857347425345e-07,
+      "loss": 0.7116,
+      "step": 2335
+    },
+    {
+      "epoch": 3.06,
+      "grad_norm": 2.765625,
+      "learning_rate": 5.893928805703943e-07,
+      "loss": 0.7554,
+      "step": 2336
+    },
+    {
+      "epoch": 3.06,
+      "grad_norm": 2.828125,
+      "learning_rate": 5.877021300192895e-07,
+      "loss": 0.7872,
+      "step": 2337
+    },
+    {
+      "epoch": 3.07,
+      "grad_norm": 2.734375,
+      "learning_rate": 5.860134849507765e-07,
+      "loss": 0.7269,
+      "step": 2338
+    },
+    {
+      "epoch": 3.07,
+      "grad_norm": 2.78125,
+      "learning_rate": 5.843269472240901e-07,
+      "loss": 0.7573,
+      "step": 2339
+    },
+    {
+      "epoch": 3.07,
+      "grad_norm": 2.75,
+      "learning_rate": 5.82642518696146e-07,
+      "loss": 0.735,
+      "step": 2340
+    },
+    {
+      "epoch": 3.07,
+      "grad_norm": 2.78125,
+      "learning_rate": 5.809602012215385e-07,
+      "loss": 0.75,
+      "step": 2341
+    },
+    {
+      "epoch": 3.07,
+      "grad_norm": 2.71875,
+      "learning_rate": 5.79279996652535e-07,
+      "loss": 0.6935,
+      "step": 2342
+    },
+    {
+      "epoch": 3.07,
+      "grad_norm": 2.796875,
+      "learning_rate": 5.776019068390792e-07,
+      "loss": 0.7396,
+      "step": 2343
+    },
+    {
+      "epoch": 3.07,
+      "grad_norm": 2.71875,
+      "learning_rate": 5.759259336287851e-07,
+      "loss": 0.7488,
+      "step": 2344
+    },
+    {
+      "epoch": 3.08,
+      "grad_norm": 2.78125,
+      "learning_rate": 5.742520788669367e-07,
+      "loss": 0.7453,
+      "step": 2345
+    },
+    {
+      "epoch": 3.08,
+      "grad_norm": 2.78125,
+      "learning_rate": 5.725803443964853e-07,
+      "loss": 0.7617,
+      "step": 2346
+    },
+    {
+      "epoch": 3.08,
+      "grad_norm": 2.78125,
+      "learning_rate": 5.709107320580479e-07,
+      "loss": 0.7488,
+      "step": 2347
+    },
+    {
+      "epoch": 3.08,
+      "grad_norm": 2.6875,
+      "learning_rate": 5.692432436899051e-07,
+      "loss": 0.7188,
+      "step": 2348
+    },
+    {
+      "epoch": 3.08,
+      "grad_norm": 2.734375,
+      "learning_rate": 5.675778811279983e-07,
+      "loss": 0.7247,
+      "step": 2349
+    },
+    {
+      "epoch": 3.08,
+      "grad_norm": 2.765625,
+      "learning_rate": 5.659146462059292e-07,
+      "loss": 0.7439,
+      "step": 2350
+    },
+    {
+      "epoch": 3.08,
+      "grad_norm": 2.671875,
+      "learning_rate": 5.642535407549565e-07,
+      "loss": 0.7298,
+      "step": 2351
+    },
+    {
+      "epoch": 3.08,
+      "grad_norm": 2.71875,
+      "learning_rate": 5.625945666039942e-07,
+      "loss": 0.7119,
+      "step": 2352
+    },
+    {
+      "epoch": 3.09,
+      "grad_norm": 2.640625,
+      "learning_rate": 5.609377255796106e-07,
+      "loss": 0.7445,
+      "step": 2353
+    },
+    {
+      "epoch": 3.09,
+      "grad_norm": 2.734375,
+      "learning_rate": 5.592830195060234e-07,
+      "loss": 0.7476,
+      "step": 2354
+    },
+    {
+      "epoch": 3.09,
+      "grad_norm": 2.65625,
+      "learning_rate": 5.576304502051011e-07,
+      "loss": 0.7004,
+      "step": 2355
+    },
+    {
+      "epoch": 3.09,
+      "grad_norm": 2.734375,
+      "learning_rate": 5.559800194963591e-07,
+      "loss": 0.7178,
+      "step": 2356
+    },
+    {
+      "epoch": 3.09,
+      "grad_norm": 2.71875,
+      "learning_rate": 5.543317291969597e-07,
+      "loss": 0.7292,
+      "step": 2357
+    },
+    {
+      "epoch": 3.09,
+      "grad_norm": 2.8125,
+      "learning_rate": 5.526855811217064e-07,
+      "loss": 0.7513,
+      "step": 2358
+    },
+    {
+      "epoch": 3.09,
+      "grad_norm": 2.734375,
+      "learning_rate": 5.510415770830457e-07,
+      "loss": 0.7337,
+      "step": 2359
+    },
+    {
+      "epoch": 3.1,
+      "grad_norm": 2.78125,
+      "learning_rate": 5.493997188910613e-07,
+      "loss": 0.7513,
+      "step": 2360
+    },
+    {
+      "epoch": 3.1,
+      "grad_norm": 2.6875,
+      "learning_rate": 5.477600083534764e-07,
+      "loss": 0.7437,
+      "step": 2361
+    },
+    {
+      "epoch": 3.1,
+      "grad_norm": 2.75,
+      "learning_rate": 5.46122447275649e-07,
+      "loss": 0.741,
+      "step": 2362
+    },
+    {
+      "epoch": 3.1,
+      "grad_norm": 2.78125,
+      "learning_rate": 5.444870374605704e-07,
+      "loss": 0.7225,
+      "step": 2363
+    },
+    {
+      "epoch": 3.1,
+      "grad_norm": 2.8125,
+      "learning_rate": 5.428537807088627e-07,
+      "loss": 0.6944,
+      "step": 2364
+    },
+    {
+      "epoch": 3.1,
+      "grad_norm": 2.75,
+      "learning_rate": 5.412226788187788e-07,
+      "loss": 0.7508,
+      "step": 2365
+    },
+    {
+      "epoch": 3.1,
+      "grad_norm": 2.75,
+      "learning_rate": 5.395937335861975e-07,
+      "loss": 0.725,
+      "step": 2366
+    },
+    {
+      "epoch": 3.1,
+      "grad_norm": 2.75,
+      "learning_rate": 5.379669468046245e-07,
+      "loss": 0.7428,
+      "step": 2367
+    },
+    {
+      "epoch": 3.11,
+      "grad_norm": 2.734375,
+      "learning_rate": 5.363423202651876e-07,
+      "loss": 0.7218,
+      "step": 2368
+    },
+    {
+      "epoch": 3.11,
+      "grad_norm": 2.765625,
+      "learning_rate": 5.347198557566374e-07,
+      "loss": 0.7236,
+      "step": 2369
+    },
+    {
+      "epoch": 3.11,
+      "grad_norm": 2.71875,
+      "learning_rate": 5.330995550653434e-07,
+      "loss": 0.7174,
+      "step": 2370
+    },
+    {
+      "epoch": 3.11,
+      "grad_norm": 2.703125,
+      "learning_rate": 5.314814199752935e-07,
+      "loss": 0.7339,
+      "step": 2371
+    },
+    {
+      "epoch": 3.11,
+      "grad_norm": 2.78125,
+      "learning_rate": 5.298654522680888e-07,
+      "loss": 0.7511,
+      "step": 2372
+    },
+    {
+      "epoch": 3.11,
+      "grad_norm": 2.890625,
+      "learning_rate": 5.282516537229468e-07,
+      "loss": 0.7366,
+      "step": 2373
+    },
+    {
+      "epoch": 3.11,
+      "grad_norm": 2.796875,
+      "learning_rate": 5.266400261166951e-07,
+      "loss": 0.755,
+      "step": 2374
+    },
+    {
+      "epoch": 3.12,
+      "grad_norm": 2.765625,
+      "learning_rate": 5.250305712237724e-07,
+      "loss": 0.7411,
+      "step": 2375
+    },
+    {
+      "epoch": 3.12,
+      "grad_norm": 2.796875,
+      "learning_rate": 5.234232908162243e-07,
+      "loss": 0.7681,
+      "step": 2376
+    },
+    {
+      "epoch": 3.12,
+      "grad_norm": 2.765625,
+      "learning_rate": 5.218181866637029e-07,
+      "loss": 0.7097,
+      "step": 2377
+    },
+    {
+      "epoch": 3.12,
+      "grad_norm": 2.78125,
+      "learning_rate": 5.202152605334623e-07,
+      "loss": 0.7398,
+      "step": 2378
+    },
+    {
+      "epoch": 3.12,
+      "grad_norm": 2.625,
+      "learning_rate": 5.186145141903606e-07,
+      "loss": 0.7076,
+      "step": 2379
+    },
+    {
+      "epoch": 3.12,
+      "grad_norm": 2.828125,
+      "learning_rate": 5.170159493968549e-07,
+      "loss": 0.7502,
+      "step": 2380
+    },
+    {
+      "epoch": 3.12,
+      "grad_norm": 2.78125,
+      "learning_rate": 5.154195679130012e-07,
+      "loss": 0.7569,
+      "step": 2381
+    },
+    {
+      "epoch": 3.12,
+      "grad_norm": 2.921875,
+      "learning_rate": 5.138253714964509e-07,
+      "loss": 0.7543,
+      "step": 2382
+    },
+    {
+      "epoch": 3.13,
+      "grad_norm": 2.90625,
+      "learning_rate": 5.122333619024494e-07,
+      "loss": 0.7527,
+      "step": 2383
+    },
+    {
+      "epoch": 3.13,
+      "grad_norm": 2.8125,
+      "learning_rate": 5.106435408838354e-07,
+      "loss": 0.7354,
+      "step": 2384
+    },
+    {
+      "epoch": 3.13,
+      "grad_norm": 2.703125,
+      "learning_rate": 5.090559101910369e-07,
+      "loss": 0.7404,
+      "step": 2385
+    },
+    {
+      "epoch": 3.13,
+      "grad_norm": 2.8125,
+      "learning_rate": 5.074704715720711e-07,
+      "loss": 0.7498,
+      "step": 2386
+    },
+    {
+      "epoch": 3.13,
+      "grad_norm": 2.828125,
+      "learning_rate": 5.058872267725407e-07,
+      "loss": 0.7304,
+      "step": 2387
+    },
+    {
+      "epoch": 3.13,
+      "grad_norm": 2.8125,
+      "learning_rate": 5.043061775356336e-07,
+      "loss": 0.7505,
+      "step": 2388
+    },
+    {
+      "epoch": 3.13,
+      "grad_norm": 2.765625,
+      "learning_rate": 5.027273256021212e-07,
+      "loss": 0.7263,
+      "step": 2389
+    },
+    {
+      "epoch": 3.14,
+      "grad_norm": 2.8125,
+      "learning_rate": 5.011506727103536e-07,
+      "loss": 0.7544,
+      "step": 2390
+    },
+    {
+      "epoch": 3.14,
+      "grad_norm": 2.75,
+      "learning_rate": 4.995762205962607e-07,
+      "loss": 0.756,
+      "step": 2391
+    },
+    {
+      "epoch": 3.14,
+      "grad_norm": 2.71875,
+      "learning_rate": 4.980039709933492e-07,
+      "loss": 0.7044,
+      "step": 2392
+    },
+    {
+      "epoch": 3.14,
+      "grad_norm": 2.6875,
+      "learning_rate": 4.964339256327022e-07,
+      "loss": 0.7123,
+      "step": 2393
+    },
+    {
+      "epoch": 3.14,
+      "grad_norm": 2.71875,
+      "learning_rate": 4.948660862429741e-07,
+      "loss": 0.7342,
+      "step": 2394
+    },
+    {
+      "epoch": 3.14,
+      "grad_norm": 2.765625,
+      "learning_rate": 4.933004545503917e-07,
+      "loss": 0.7532,
+      "step": 2395
+    },
+    {
+      "epoch": 3.14,
+      "grad_norm": 2.734375,
+      "learning_rate": 4.917370322787488e-07,
+      "loss": 0.7435,
+      "step": 2396
+    },
+    {
+      "epoch": 3.14,
+      "grad_norm": 2.8125,
+      "learning_rate": 4.901758211494089e-07,
+      "loss": 0.7251,
+      "step": 2397
+    },
+    {
+      "epoch": 3.15,
+      "grad_norm": 2.859375,
+      "learning_rate": 4.886168228813007e-07,
+      "loss": 0.7306,
+      "step": 2398
+    },
+    {
+      "epoch": 3.15,
+      "grad_norm": 2.6875,
+      "learning_rate": 4.870600391909156e-07,
+      "loss": 0.7276,
+      "step": 2399
+    },
+    {
+      "epoch": 3.15,
+      "grad_norm": 2.6875,
+      "learning_rate": 4.855054717923072e-07,
+      "loss": 0.7128,
+      "step": 2400
+    },
+    {
+      "epoch": 3.15,
+      "grad_norm": 2.78125,
+      "learning_rate": 4.839531223970889e-07,
+      "loss": 0.7291,
+      "step": 2401
+    },
+    {
+      "epoch": 3.15,
+      "grad_norm": 2.71875,
+      "learning_rate": 4.824029927144321e-07,
+      "loss": 0.7587,
+      "step": 2402
+    },
+    {
+      "epoch": 3.15,
+      "grad_norm": 2.84375,
+      "learning_rate": 4.80855084451064e-07,
+      "loss": 0.7431,
+      "step": 2403
+    },
+    {
+      "epoch": 3.15,
+      "grad_norm": 2.6875,
+      "learning_rate": 4.793093993112663e-07,
+      "loss": 0.7269,
+      "step": 2404
+    },
+    {
+      "epoch": 3.16,
+      "grad_norm": 2.796875,
+      "learning_rate": 4.777659389968728e-07,
+      "loss": 0.7415,
+      "step": 2405
+    },
+    {
+      "epoch": 3.16,
+      "grad_norm": 2.859375,
+      "learning_rate": 4.76224705207268e-07,
+      "loss": 0.7523,
+      "step": 2406
+    },
+    {
+      "epoch": 3.16,
+      "grad_norm": 2.71875,
+      "learning_rate": 4.7468569963938365e-07,
+      "loss": 0.7234,
+      "step": 2407
+    },
+    {
+      "epoch": 3.16,
+      "grad_norm": 2.734375,
+      "learning_rate": 4.731489239876999e-07,
+      "loss": 0.7303,
+      "step": 2408
+    },
+    {
+      "epoch": 3.16,
+      "grad_norm": 2.734375,
+      "learning_rate": 4.71614379944241e-07,
+      "loss": 0.7389,
+      "step": 2409
+    },
+    {
+      "epoch": 3.16,
+      "grad_norm": 2.8125,
+      "learning_rate": 4.700820691985739e-07,
+      "loss": 0.7597,
+      "step": 2410
+    },
+    {
+      "epoch": 3.16,
+      "grad_norm": 2.703125,
+      "learning_rate": 4.685519934378066e-07,
+      "loss": 0.7248,
+      "step": 2411
+    },
+    {
+      "epoch": 3.16,
+      "grad_norm": 2.734375,
+      "learning_rate": 4.670241543465881e-07,
+      "loss": 0.7269,
+      "step": 2412
+    },
+    {
+      "epoch": 3.17,
+      "grad_norm": 2.75,
+      "learning_rate": 4.654985536071019e-07,
+      "loss": 0.7445,
+      "step": 2413
+    },
+    {
+      "epoch": 3.17,
+      "grad_norm": 2.734375,
+      "learning_rate": 4.639751928990688e-07,
+      "loss": 0.7453,
+      "step": 2414
+    },
+    {
+      "epoch": 3.17,
+      "grad_norm": 2.734375,
+      "learning_rate": 4.6245407389974293e-07,
+      "loss": 0.7385,
+      "step": 2415
+    },
+    {
+      "epoch": 3.17,
+      "grad_norm": 2.734375,
+      "learning_rate": 4.6093519828391025e-07,
+      "loss": 0.7419,
+      "step": 2416
+    },
+    {
+      "epoch": 3.17,
+      "grad_norm": 2.765625,
+      "learning_rate": 4.594185677238866e-07,
+      "loss": 0.7306,
+      "step": 2417
+    },
+    {
+      "epoch": 3.17,
+      "grad_norm": 2.78125,
+      "learning_rate": 4.5790418388951596e-07,
+      "loss": 0.7604,
+      "step": 2418
+    },
+    {
+      "epoch": 3.17,
+      "grad_norm": 2.6875,
+      "learning_rate": 4.5639204844816904e-07,
+      "loss": 0.7203,
+      "step": 2419
+    },
+    {
+      "epoch": 3.18,
+      "grad_norm": 2.796875,
+      "learning_rate": 4.5488216306474044e-07,
+      "loss": 0.7542,
+      "step": 2420
+    },
+    {
+      "epoch": 3.18,
+      "grad_norm": 2.796875,
+      "learning_rate": 4.5337452940164757e-07,
+      "loss": 0.7379,
+      "step": 2421
+    },
+    {
+      "epoch": 3.18,
+      "grad_norm": 2.703125,
+      "learning_rate": 4.51869149118829e-07,
+      "loss": 0.7314,
+      "step": 2422
+    },
+    {
+      "epoch": 3.18,
+      "grad_norm": 2.703125,
+      "learning_rate": 4.5036602387374143e-07,
+      "loss": 0.7354,
+      "step": 2423
+    },
+    {
+      "epoch": 3.18,
+      "grad_norm": 2.71875,
+      "learning_rate": 4.488651553213605e-07,
+      "loss": 0.7207,
+      "step": 2424
+    },
+    {
+      "epoch": 3.18,
+      "grad_norm": 2.78125,
+      "learning_rate": 4.4736654511417456e-07,
+      "loss": 0.7549,
+      "step": 2425
+    },
+    {
+      "epoch": 3.18,
+      "grad_norm": 2.78125,
+      "learning_rate": 4.458701949021871e-07,
+      "loss": 0.733,
+      "step": 2426
+    },
+    {
+      "epoch": 3.18,
+      "grad_norm": 2.75,
+      "learning_rate": 4.443761063329133e-07,
+      "loss": 0.7572,
+      "step": 2427
+    },
+    {
+      "epoch": 3.19,
+      "grad_norm": 2.640625,
+      "learning_rate": 4.428842810513784e-07,
+      "loss": 0.7242,
+      "step": 2428
+    },
+    {
+      "epoch": 3.19,
+      "grad_norm": 2.6875,
+      "learning_rate": 4.4139472070011374e-07,
+      "loss": 0.7304,
+      "step": 2429
+    },
+    {
+      "epoch": 3.19,
+      "grad_norm": 2.796875,
+      "learning_rate": 4.399074269191614e-07,
+      "loss": 0.7434,
+      "step": 2430
+    },
+    {
+      "epoch": 3.19,
+      "grad_norm": 2.796875,
+      "learning_rate": 4.3842240134606225e-07,
+      "loss": 0.7429,
+      "step": 2431
+    },
+    {
+      "epoch": 3.19,
+      "grad_norm": 2.78125,
+      "learning_rate": 4.369396456158639e-07,
+      "loss": 0.7203,
+      "step": 2432
+    },
+    {
+      "epoch": 3.19,
+      "grad_norm": 2.671875,
+      "learning_rate": 4.3545916136111284e-07,
+      "loss": 0.7554,
+      "step": 2433
+    },
+    {
+      "epoch": 3.19,
+      "grad_norm": 2.78125,
+      "learning_rate": 4.3398095021185557e-07,
+      "loss": 0.7229,
+      "step": 2434
+    },
+    {
+      "epoch": 3.2,
+      "grad_norm": 2.78125,
+      "learning_rate": 4.325050137956355e-07,
+      "loss": 0.731,
+      "step": 2435
+    },
+    {
+      "epoch": 3.2,
+      "grad_norm": 2.828125,
+      "learning_rate": 4.3103135373749216e-07,
+      "loss": 0.7836,
+      "step": 2436
+    },
+    {
+      "epoch": 3.2,
+      "grad_norm": 2.8125,
+      "learning_rate": 4.2955997165995634e-07,
+      "loss": 0.7388,
+      "step": 2437
+    },
+    {
+      "epoch": 3.2,
+      "grad_norm": 2.859375,
+      "learning_rate": 4.2809086918305403e-07,
+      "loss": 0.7573,
+      "step": 2438
+    },
+    {
+      "epoch": 3.2,
+      "grad_norm": 2.84375,
+      "learning_rate": 4.266240479242997e-07,
+      "loss": 0.7542,
+      "step": 2439
+    },
+    {
+      "epoch": 3.2,
+      "grad_norm": 2.703125,
+      "learning_rate": 4.251595094986957e-07,
+      "loss": 0.7314,
+      "step": 2440
+    },
+    {
+      "epoch": 3.2,
+      "grad_norm": 2.796875,
+      "learning_rate": 4.236972555187319e-07,
+      "loss": 0.7696,
+      "step": 2441
+    },
+    {
+      "epoch": 3.2,
+      "grad_norm": 2.75,
+      "learning_rate": 4.222372875943831e-07,
+      "loss": 0.7625,
+      "step": 2442
+    },
+    {
+      "epoch": 3.21,
+      "grad_norm": 2.796875,
+      "learning_rate": 4.207796073331052e-07,
+      "loss": 0.754,
+      "step": 2443
+    },
+    {
+      "epoch": 3.21,
+      "grad_norm": 2.703125,
+      "learning_rate": 4.1932421633983726e-07,
+      "loss": 0.7264,
+      "step": 2444
+    },
+    {
+      "epoch": 3.21,
+      "grad_norm": 2.8125,
+      "learning_rate": 4.178711162169974e-07,
+      "loss": 0.7295,
+      "step": 2445
+    },
+    {
+      "epoch": 3.21,
+      "grad_norm": 2.8125,
+      "learning_rate": 4.1642030856448104e-07,
+      "loss": 0.7379,
+      "step": 2446
+    },
+    {
+      "epoch": 3.21,
+      "grad_norm": 2.75,
+      "learning_rate": 4.1497179497965915e-07,
+      "loss": 0.7712,
+      "step": 2447
+    },
+    {
+      "epoch": 3.21,
+      "grad_norm": 2.734375,
+      "learning_rate": 4.1352557705737957e-07,
+      "loss": 0.7512,
+      "step": 2448
+    },
+    {
+      "epoch": 3.21,
+      "grad_norm": 2.6875,
+      "learning_rate": 4.120816563899585e-07,
+      "loss": 0.7678,
+      "step": 2449
+    },
+    {
+      "epoch": 3.22,
+      "grad_norm": 2.734375,
+      "learning_rate": 4.1064003456718517e-07,
+      "loss": 0.7376,
+      "step": 2450
+    },
+    {
+      "epoch": 3.22,
+      "grad_norm": 2.75,
+      "learning_rate": 4.092007131763176e-07,
+      "loss": 0.7255,
+      "step": 2451
+    },
+    {
+      "epoch": 3.22,
+      "grad_norm": 2.671875,
+      "learning_rate": 4.077636938020807e-07,
+      "loss": 0.7056,
+      "step": 2452
+    },
+    {
+      "epoch": 3.22,
+      "grad_norm": 2.671875,
+      "learning_rate": 4.063289780266652e-07,
+      "loss": 0.7429,
+      "step": 2453
+    },
+    {
+      "epoch": 3.22,
+      "grad_norm": 2.859375,
+      "learning_rate": 4.048965674297245e-07,
+      "loss": 0.7441,
+      "step": 2454
+    },
+    {
+      "epoch": 3.22,
+      "grad_norm": 2.640625,
+      "learning_rate": 4.034664635883742e-07,
+      "loss": 0.6863,
+      "step": 2455
+    },
+    {
+      "epoch": 3.22,
+      "grad_norm": 2.6875,
+      "learning_rate": 4.0203866807719176e-07,
+      "loss": 0.7189,
+      "step": 2456
+    },
+    {
+      "epoch": 3.22,
+      "grad_norm": 2.8125,
+      "learning_rate": 4.0061318246821105e-07,
+      "loss": 0.7478,
+      "step": 2457
+    },
+    {
+      "epoch": 3.23,
+      "grad_norm": 2.78125,
+      "learning_rate": 3.991900083309241e-07,
+      "loss": 0.7417,
+      "step": 2458
+    },
+    {
+      "epoch": 3.23,
+      "grad_norm": 2.71875,
+      "learning_rate": 3.9776914723227753e-07,
+      "loss": 0.7149,
+      "step": 2459
+    },
+    {
+      "epoch": 3.23,
+      "grad_norm": 2.765625,
+      "learning_rate": 3.9635060073666993e-07,
+      "loss": 0.7243,
+      "step": 2460
+    },
+    {
+      "epoch": 3.23,
+      "grad_norm": 2.75,
+      "learning_rate": 3.9493437040595327e-07,
+      "loss": 0.7261,
+      "step": 2461
+    },
+    {
+      "epoch": 3.23,
+      "grad_norm": 2.890625,
+      "learning_rate": 3.935204577994289e-07,
+      "loss": 0.7668,
+      "step": 2462
+    },
+    {
+      "epoch": 3.23,
+      "grad_norm": 2.71875,
+      "learning_rate": 3.921088644738458e-07,
+      "loss": 0.7382,
+      "step": 2463
+    },
+    {
+      "epoch": 3.23,
+      "grad_norm": 2.75,
+      "learning_rate": 3.906995919833997e-07,
+      "loss": 0.743,
+      "step": 2464
+    },
+    {
+      "epoch": 3.24,
+      "grad_norm": 2.78125,
+      "learning_rate": 3.892926418797313e-07,
+      "loss": 0.7242,
+      "step": 2465
+    },
+    {
+      "epoch": 3.24,
+      "grad_norm": 2.765625,
+      "learning_rate": 3.878880157119241e-07,
+      "loss": 0.7465,
+      "step": 2466
+    },
+    {
+      "epoch": 3.24,
+      "grad_norm": 2.8125,
+      "learning_rate": 3.8648571502650253e-07,
+      "loss": 0.741,
+      "step": 2467
+    },
+    {
+      "epoch": 3.24,
+      "grad_norm": 2.75,
+      "learning_rate": 3.8508574136743085e-07,
+      "loss": 0.7638,
+      "step": 2468
+    },
+    {
+      "epoch": 3.24,
+      "grad_norm": 2.84375,
+      "learning_rate": 3.8368809627611135e-07,
+      "loss": 0.7375,
+      "step": 2469
+    },
+    {
+      "epoch": 3.24,
+      "grad_norm": 2.828125,
+      "learning_rate": 3.8229278129138293e-07,
+      "loss": 0.7364,
+      "step": 2470
+    },
+    {
+      "epoch": 3.24,
+      "grad_norm": 2.90625,
+      "learning_rate": 3.8089979794951825e-07,
+      "loss": 0.7208,
+      "step": 2471
+    },
+    {
+      "epoch": 3.24,
+      "grad_norm": 2.84375,
+      "learning_rate": 3.795091477842225e-07,
+      "loss": 0.7375,
+      "step": 2472
+    },
+    {
+      "epoch": 3.25,
+      "grad_norm": 2.765625,
+      "learning_rate": 3.7812083232663306e-07,
+      "loss": 0.7208,
+      "step": 2473
+    },
+    {
+      "epoch": 3.25,
+      "grad_norm": 2.859375,
+      "learning_rate": 3.7673485310531525e-07,
+      "loss": 0.7518,
+      "step": 2474
+    },
+    {
+      "epoch": 3.25,
+      "grad_norm": 2.796875,
+      "learning_rate": 3.753512116462643e-07,
+      "loss": 0.7345,
+      "step": 2475
+    },
+    {
+      "epoch": 3.25,
+      "grad_norm": 2.859375,
+      "learning_rate": 3.739699094729002e-07,
+      "loss": 0.7367,
+      "step": 2476
+    },
+    {
+      "epoch": 3.25,
+      "grad_norm": 2.75,
+      "learning_rate": 3.7259094810606745e-07,
+      "loss": 0.7352,
+      "step": 2477
+    },
+    {
+      "epoch": 3.25,
+      "grad_norm": 2.734375,
+      "learning_rate": 3.7121432906403244e-07,
+      "loss": 0.7244,
+      "step": 2478
+    },
+    {
+      "epoch": 3.25,
+      "grad_norm": 2.640625,
+      "learning_rate": 3.6984005386248385e-07,
+      "loss": 0.7126,
+      "step": 2479
+    },
+    {
+      "epoch": 3.26,
+      "grad_norm": 2.796875,
+      "learning_rate": 3.684681240145296e-07,
+      "loss": 0.7566,
+      "step": 2480
+    },
+    {
+      "epoch": 3.26,
+      "grad_norm": 2.6875,
+      "learning_rate": 3.6709854103069457e-07,
+      "loss": 0.7272,
+      "step": 2481
+    },
+    {
+      "epoch": 3.26,
+      "grad_norm": 2.796875,
+      "learning_rate": 3.6573130641892053e-07,
+      "loss": 0.7387,
+      "step": 2482
+    },
+    {
+      "epoch": 3.26,
+      "grad_norm": 2.8125,
+      "learning_rate": 3.6436642168456325e-07,
+      "loss": 0.7194,
+      "step": 2483
+    },
+    {
+      "epoch": 3.26,
+      "grad_norm": 2.859375,
+      "learning_rate": 3.63003888330391e-07,
+      "loss": 0.7669,
+      "step": 2484
+    },
+    {
+      "epoch": 3.26,
+      "grad_norm": 2.71875,
+      "learning_rate": 3.6164370785658346e-07,
+      "loss": 0.7403,
+      "step": 2485
+    },
+    {
+      "epoch": 3.26,
+      "grad_norm": 2.734375,
+      "learning_rate": 3.602858817607299e-07,
+      "loss": 0.739,
+      "step": 2486
+    },
+    {
+      "epoch": 3.26,
+      "grad_norm": 2.75,
+      "learning_rate": 3.5893041153782697e-07,
+      "loss": 0.7241,
+      "step": 2487
+    },
+    {
+      "epoch": 3.27,
+      "grad_norm": 2.75,
+      "learning_rate": 3.575772986802775e-07,
+      "loss": 0.7532,
+      "step": 2488
+    },
+    {
+      "epoch": 3.27,
+      "grad_norm": 2.75,
+      "learning_rate": 3.562265446778898e-07,
+      "loss": 0.7488,
+      "step": 2489
+    },
+    {
+      "epoch": 3.27,
+      "grad_norm": 2.75,
+      "learning_rate": 3.5487815101787253e-07,
+      "loss": 0.7359,
+      "step": 2490
+    },
+    {
+      "epoch": 3.27,
+      "grad_norm": 2.78125,
+      "learning_rate": 3.5353211918483813e-07,
+      "loss": 0.7444,
+      "step": 2491
+    },
+    {
+      "epoch": 3.27,
+      "grad_norm": 2.8125,
+      "learning_rate": 3.52188450660797e-07,
+      "loss": 0.7389,
+      "step": 2492
+    },
+    {
+      "epoch": 3.27,
+      "grad_norm": 2.71875,
+      "learning_rate": 3.50847146925159e-07,
+      "loss": 0.7365,
+      "step": 2493
+    },
+    {
+      "epoch": 3.27,
+      "grad_norm": 2.828125,
+      "learning_rate": 3.4950820945472945e-07,
+      "loss": 0.714,
+      "step": 2494
+    },
+    {
+      "epoch": 3.28,
+      "grad_norm": 2.78125,
+      "learning_rate": 3.481716397237084e-07,
+      "loss": 0.7675,
+      "step": 2495
+    },
+    {
+      "epoch": 3.28,
+      "grad_norm": 2.75,
+      "learning_rate": 3.46837439203688e-07,
+      "loss": 0.7285,
+      "step": 2496
+    },
+    {
+      "epoch": 3.28,
+      "grad_norm": 2.75,
+      "learning_rate": 3.455056093636533e-07,
+      "loss": 0.7343,
+      "step": 2497
+    },
+    {
+      "epoch": 3.28,
+      "grad_norm": 2.734375,
+      "learning_rate": 3.441761516699788e-07,
+      "loss": 0.7157,
+      "step": 2498
+    },
+    {
+      "epoch": 3.28,
+      "grad_norm": 2.703125,
+      "learning_rate": 3.4284906758642683e-07,
+      "loss": 0.726,
+      "step": 2499
+    },
+    {
+      "epoch": 3.28,
+      "grad_norm": 2.84375,
+      "learning_rate": 3.4152435857414676e-07,
+      "loss": 0.7455,
+      "step": 2500
+    },
+    {
+      "epoch": 3.28,
+      "grad_norm": 2.828125,
+      "learning_rate": 3.4020202609167245e-07,
+      "loss": 0.7319,
+      "step": 2501
+    },
+    {
+      "epoch": 3.28,
+      "grad_norm": 2.765625,
+      "learning_rate": 3.3888207159492177e-07,
+      "loss": 0.7075,
+      "step": 2502
+    },
+    {
+      "epoch": 3.29,
+      "grad_norm": 2.78125,
+      "learning_rate": 3.375644965371941e-07,
+      "loss": 0.7237,
+      "step": 2503
+    },
+    {
+      "epoch": 3.29,
+      "grad_norm": 2.984375,
+      "learning_rate": 3.362493023691685e-07,
+      "loss": 0.738,
+      "step": 2504
+    },
+    {
+      "epoch": 3.29,
+      "grad_norm": 2.734375,
+      "learning_rate": 3.3493649053890325e-07,
+      "loss": 0.7032,
+      "step": 2505
+    },
+    {
+      "epoch": 3.29,
+      "grad_norm": 2.8125,
+      "learning_rate": 3.3362606249183446e-07,
+      "loss": 0.744,
+      "step": 2506
+    },
+    {
+      "epoch": 3.29,
+      "grad_norm": 2.8125,
+      "learning_rate": 3.323180196707709e-07,
+      "loss": 0.7664,
+      "step": 2507
+    },
+    {
+      "epoch": 3.29,
+      "grad_norm": 2.765625,
+      "learning_rate": 3.3101236351589764e-07,
+      "loss": 0.7668,
+      "step": 2508
+    },
+    {
+      "epoch": 3.29,
+      "grad_norm": 2.78125,
+      "learning_rate": 3.297090954647714e-07,
+      "loss": 0.7308,
+      "step": 2509
+    },
+    {
+      "epoch": 3.3,
+      "grad_norm": 2.765625,
+      "learning_rate": 3.2840821695231867e-07,
+      "loss": 0.7502,
+      "step": 2510
+    },
+    {
+      "epoch": 3.3,
+      "grad_norm": 2.75,
+      "learning_rate": 3.271097294108369e-07,
+      "loss": 0.7618,
+      "step": 2511
+    },
+    {
+      "epoch": 3.3,
+      "grad_norm": 2.734375,
+      "learning_rate": 3.2581363426998966e-07,
+      "loss": 0.7359,
+      "step": 2512
+    },
+    {
+      "epoch": 3.3,
+      "grad_norm": 2.859375,
+      "learning_rate": 3.245199329568055e-07,
+      "loss": 0.7454,
+      "step": 2513
+    },
+    {
+      "epoch": 3.3,
+      "grad_norm": 2.796875,
+      "learning_rate": 3.2322862689567885e-07,
+      "loss": 0.7373,
+      "step": 2514
+    },
+    {
+      "epoch": 3.3,
+      "grad_norm": 2.8125,
+      "learning_rate": 3.219397175083669e-07,
+      "loss": 0.7672,
+      "step": 2515
+    },
+    {
+      "epoch": 3.3,
+      "grad_norm": 2.703125,
+      "learning_rate": 3.206532062139875e-07,
+      "loss": 0.7409,
+      "step": 2516
+    },
+    {
+      "epoch": 3.3,
+      "grad_norm": 2.71875,
+      "learning_rate": 3.193690944290179e-07,
+      "loss": 0.7164,
+      "step": 2517
+    },
+    {
+      "epoch": 3.31,
+      "grad_norm": 2.78125,
+      "learning_rate": 3.18087383567294e-07,
+      "loss": 0.7414,
+      "step": 2518
+    },
+    {
+      "epoch": 3.31,
+      "grad_norm": 2.6875,
+      "learning_rate": 3.168080750400082e-07,
+      "loss": 0.7373,
+      "step": 2519
+    },
+    {
+      "epoch": 3.31,
+      "grad_norm": 2.75,
+      "learning_rate": 3.1553117025570763e-07,
+      "loss": 0.7407,
+      "step": 2520
+    },
+    {
+      "epoch": 3.31,
+      "grad_norm": 2.765625,
+      "learning_rate": 3.142566706202929e-07,
+      "loss": 0.7237,
+      "step": 2521
+    },
+    {
+      "epoch": 3.31,
+      "grad_norm": 2.71875,
+      "learning_rate": 3.129845775370163e-07,
+      "loss": 0.7437,
+      "step": 2522
+    },
+    {
+      "epoch": 3.31,
+      "grad_norm": 2.765625,
+      "learning_rate": 3.117148924064811e-07,
+      "loss": 0.7347,
+      "step": 2523
+    },
+    {
+      "epoch": 3.31,
+      "grad_norm": 2.6875,
+      "learning_rate": 3.1044761662663933e-07,
+      "loss": 0.7156,
+      "step": 2524
+    },
+    {
+      "epoch": 3.32,
+      "grad_norm": 2.796875,
+      "learning_rate": 3.091827515927884e-07,
+      "loss": 0.7528,
+      "step": 2525
+    },
+    {
+      "epoch": 3.32,
+      "grad_norm": 2.75,
+      "learning_rate": 3.079202986975741e-07,
+      "loss": 0.7298,
+      "step": 2526
+    },
+    {
+      "epoch": 3.32,
+      "grad_norm": 2.828125,
+      "learning_rate": 3.0666025933098475e-07,
+      "loss": 0.7426,
+      "step": 2527
+    },
+    {
+      "epoch": 3.32,
+      "grad_norm": 2.765625,
+      "learning_rate": 3.054026348803521e-07,
+      "loss": 0.7318,
+      "step": 2528
+    },
+    {
+      "epoch": 3.32,
+      "grad_norm": 2.71875,
+      "learning_rate": 3.041474267303479e-07,
+      "loss": 0.7163,
+      "step": 2529
+    },
+    {
+      "epoch": 3.32,
+      "grad_norm": 2.828125,
+      "learning_rate": 3.0289463626298585e-07,
+      "loss": 0.7589,
+      "step": 2530
+    },
+    {
+      "epoch": 3.32,
+      "grad_norm": 2.765625,
+      "learning_rate": 3.016442648576151e-07,
+      "loss": 0.7362,
+      "step": 2531
+    },
+    {
+      "epoch": 3.32,
+      "grad_norm": 2.84375,
+      "learning_rate": 3.003963138909224e-07,
+      "loss": 0.7456,
+      "step": 2532
+    },
+    {
+      "epoch": 3.33,
+      "grad_norm": 2.71875,
+      "learning_rate": 2.991507847369299e-07,
+      "loss": 0.7276,
+      "step": 2533
+    },
+    {
+      "epoch": 3.33,
+      "grad_norm": 2.734375,
+      "learning_rate": 2.9790767876699293e-07,
+      "loss": 0.755,
+      "step": 2534
+    },
+    {
+      "epoch": 3.33,
+      "grad_norm": 2.84375,
+      "learning_rate": 2.9666699734979875e-07,
+      "loss": 0.7558,
+      "step": 2535
+    },
+    {
+      "epoch": 3.33,
+      "grad_norm": 2.75,
+      "learning_rate": 2.9542874185136545e-07,
+      "loss": 0.7497,
+      "step": 2536
+    },
+    {
+      "epoch": 3.33,
+      "grad_norm": 2.796875,
+      "learning_rate": 2.941929136350399e-07,
+      "loss": 0.7451,
+      "step": 2537
+    },
+    {
+      "epoch": 3.33,
+      "grad_norm": 2.796875,
+      "learning_rate": 2.929595140614963e-07,
+      "loss": 0.7497,
+      "step": 2538
+    },
+    {
+      "epoch": 3.33,
+      "grad_norm": 2.875,
+      "learning_rate": 2.917285444887355e-07,
+      "loss": 0.7591,
+      "step": 2539
+    },
+    {
+      "epoch": 3.33,
+      "grad_norm": 2.703125,
+      "learning_rate": 2.9050000627208195e-07,
+      "loss": 0.7396,
+      "step": 2540
+    },
+    {
+      "epoch": 3.34,
+      "grad_norm": 2.6875,
+      "learning_rate": 2.89273900764184e-07,
+      "loss": 0.7328,
+      "step": 2541
+    },
+    {
+      "epoch": 3.34,
+      "grad_norm": 2.78125,
+      "learning_rate": 2.880502293150117e-07,
+      "loss": 0.7459,
+      "step": 2542
+    },
+    {
+      "epoch": 3.34,
+      "grad_norm": 2.703125,
+      "learning_rate": 2.8682899327185293e-07,
+      "loss": 0.7186,
+      "step": 2543
+    },
+    {
+      "epoch": 3.34,
+      "grad_norm": 2.765625,
+      "learning_rate": 2.856101939793171e-07,
+      "loss": 0.729,
+      "step": 2544
+    },
+    {
+      "epoch": 3.34,
+      "grad_norm": 2.8125,
+      "learning_rate": 2.843938327793291e-07,
+      "loss": 0.7485,
+      "step": 2545
+    },
+    {
+      "epoch": 3.34,
+      "grad_norm": 2.6875,
+      "learning_rate": 2.8317991101112966e-07,
+      "loss": 0.7478,
+      "step": 2546
+    },
+    {
+      "epoch": 3.34,
+      "grad_norm": 2.828125,
+      "learning_rate": 2.819684300112732e-07,
+      "loss": 0.7725,
+      "step": 2547
+    },
+    {
+      "epoch": 3.35,
+      "grad_norm": 2.734375,
+      "learning_rate": 2.8075939111362915e-07,
+      "loss": 0.7284,
+      "step": 2548
+    },
+    {
+      "epoch": 3.35,
+      "grad_norm": 2.75,
+      "learning_rate": 2.7955279564937433e-07,
+      "loss": 0.7398,
+      "step": 2549
+    },
+    {
+      "epoch": 3.35,
+      "grad_norm": 2.875,
+      "learning_rate": 2.783486449469977e-07,
+      "loss": 0.7423,
+      "step": 2550
+    },
+    {
+      "epoch": 3.35,
+      "grad_norm": 2.75,
+      "learning_rate": 2.7714694033229656e-07,
+      "loss": 0.7381,
+      "step": 2551
+    },
+    {
+      "epoch": 3.35,
+      "grad_norm": 2.703125,
+      "learning_rate": 2.759476831283742e-07,
+      "loss": 0.7243,
+      "step": 2552
+    },
+    {
+      "epoch": 3.35,
+      "grad_norm": 2.765625,
+      "learning_rate": 2.747508746556396e-07,
+      "loss": 0.765,
+      "step": 2553
+    },
+    {
+      "epoch": 3.35,
+      "grad_norm": 2.9375,
+      "learning_rate": 2.7355651623180574e-07,
+      "loss": 0.7317,
+      "step": 2554
+    },
+    {
+      "epoch": 3.35,
+      "grad_norm": 2.796875,
+      "learning_rate": 2.723646091718868e-07,
+      "loss": 0.7361,
+      "step": 2555
+    },
+    {
+      "epoch": 3.36,
+      "grad_norm": 2.71875,
+      "learning_rate": 2.711751547882005e-07,
+      "loss": 0.7174,
+      "step": 2556
+    },
+    {
+      "epoch": 3.36,
+      "grad_norm": 2.765625,
+      "learning_rate": 2.699881543903618e-07,
+      "loss": 0.7314,
+      "step": 2557
+    },
+    {
+      "epoch": 3.36,
+      "grad_norm": 2.703125,
+      "learning_rate": 2.6880360928528497e-07,
+      "loss": 0.7426,
+      "step": 2558
+    },
+    {
+      "epoch": 3.36,
+      "grad_norm": 2.953125,
+      "learning_rate": 2.676215207771804e-07,
+      "loss": 0.7566,
+      "step": 2559
+    },
+    {
+      "epoch": 3.36,
+      "grad_norm": 2.84375,
+      "learning_rate": 2.6644189016755415e-07,
+      "loss": 0.7516,
+      "step": 2560
+    },
+    {
+      "epoch": 3.36,
+      "grad_norm": 2.84375,
+      "learning_rate": 2.652647187552054e-07,
+      "loss": 0.7347,
+      "step": 2561
+    },
+    {
+      "epoch": 3.36,
+      "grad_norm": 2.734375,
+      "learning_rate": 2.640900078362263e-07,
+      "loss": 0.7097,
+      "step": 2562
+    },
+    {
+      "epoch": 3.37,
+      "grad_norm": 2.734375,
+      "learning_rate": 2.629177587039997e-07,
+      "loss": 0.7516,
+      "step": 2563
+    },
+    {
+      "epoch": 3.37,
+      "grad_norm": 2.796875,
+      "learning_rate": 2.617479726491981e-07,
+      "loss": 0.7262,
+      "step": 2564
+    },
+    {
+      "epoch": 3.37,
+      "grad_norm": 2.796875,
+      "learning_rate": 2.605806509597819e-07,
+      "loss": 0.7511,
+      "step": 2565
+    },
+    {
+      "epoch": 3.37,
+      "grad_norm": 2.8125,
+      "learning_rate": 2.5941579492099853e-07,
+      "loss": 0.742,
+      "step": 2566
+    },
+    {
+      "epoch": 3.37,
+      "grad_norm": 2.71875,
+      "learning_rate": 2.5825340581538043e-07,
+      "loss": 0.7357,
+      "step": 2567
+    },
+    {
+      "epoch": 3.37,
+      "grad_norm": 2.8125,
+      "learning_rate": 2.5709348492274384e-07,
+      "loss": 0.7173,
+      "step": 2568
+    },
+    {
+      "epoch": 3.37,
+      "grad_norm": 2.796875,
+      "learning_rate": 2.559360335201874e-07,
+      "loss": 0.7532,
+      "step": 2569
+    },
+    {
+      "epoch": 3.37,
+      "grad_norm": 2.8125,
+      "learning_rate": 2.547810528820913e-07,
+      "loss": 0.7346,
+      "step": 2570
+    },
+    {
+      "epoch": 3.38,
+      "grad_norm": 2.765625,
+      "learning_rate": 2.536285442801148e-07,
+      "loss": 0.7426,
+      "step": 2571
+    },
+    {
+      "epoch": 3.38,
+      "grad_norm": 2.84375,
+      "learning_rate": 2.524785089831955e-07,
+      "loss": 0.7343,
+      "step": 2572
+    },
+    {
+      "epoch": 3.38,
+      "grad_norm": 2.75,
+      "learning_rate": 2.513309482575471e-07,
+      "loss": 0.7296,
+      "step": 2573
+    },
+    {
+      "epoch": 3.38,
+      "grad_norm": 2.8125,
+      "learning_rate": 2.5018586336666094e-07,
+      "loss": 0.7495,
+      "step": 2574
+    },
+    {
+      "epoch": 3.38,
+      "grad_norm": 2.75,
+      "learning_rate": 2.490432555713002e-07,
+      "loss": 0.7513,
+      "step": 2575
+    },
+    {
+      "epoch": 3.38,
+      "grad_norm": 2.796875,
+      "learning_rate": 2.47903126129502e-07,
+      "loss": 0.7371,
+      "step": 2576
+    },
+    {
+      "epoch": 3.38,
+      "grad_norm": 2.859375,
+      "learning_rate": 2.467654762965743e-07,
+      "loss": 0.7392,
+      "step": 2577
+    },
+    {
+      "epoch": 3.39,
+      "grad_norm": 2.84375,
+      "learning_rate": 2.456303073250943e-07,
+      "loss": 0.7331,
+      "step": 2578
+    },
+    {
+      "epoch": 3.39,
+      "grad_norm": 2.75,
+      "learning_rate": 2.4449762046490885e-07,
+      "loss": 0.7307,
+      "step": 2579
+    },
+    {
+      "epoch": 3.39,
+      "grad_norm": 2.8125,
+      "learning_rate": 2.4336741696313113e-07,
+      "loss": 0.7557,
+      "step": 2580
+    },
+    {
+      "epoch": 3.39,
+      "grad_norm": 2.75,
+      "learning_rate": 2.422396980641406e-07,
+      "loss": 0.722,
+      "step": 2581
+    },
+    {
+      "epoch": 3.39,
+      "grad_norm": 2.8125,
+      "learning_rate": 2.411144650095809e-07,
+      "loss": 0.7376,
+      "step": 2582
+    },
+    {
+      "epoch": 3.39,
+      "grad_norm": 2.734375,
+      "learning_rate": 2.3999171903835853e-07,
+      "loss": 0.7393,
+      "step": 2583
+    },
+    {
+      "epoch": 3.39,
+      "grad_norm": 2.734375,
+      "learning_rate": 2.388714613866422e-07,
+      "loss": 0.7223,
+      "step": 2584
+    },
+    {
+      "epoch": 3.39,
+      "grad_norm": 2.765625,
+      "learning_rate": 2.377536932878602e-07,
+      "loss": 0.7306,
+      "step": 2585
+    },
+    {
+      "epoch": 3.4,
+      "grad_norm": 2.75,
+      "learning_rate": 2.3663841597270032e-07,
+      "loss": 0.7416,
+      "step": 2586
+    },
+    {
+      "epoch": 3.4,
+      "grad_norm": 2.65625,
+      "learning_rate": 2.3552563066910756e-07,
+      "loss": 0.7443,
+      "step": 2587
+    },
+    {
+      "epoch": 3.4,
+      "grad_norm": 2.890625,
+      "learning_rate": 2.3441533860228376e-07,
+      "loss": 0.7527,
+      "step": 2588
+    },
+    {
+      "epoch": 3.4,
+      "grad_norm": 2.765625,
+      "learning_rate": 2.3330754099468528e-07,
+      "loss": 0.7381,
+      "step": 2589
+    },
+    {
+      "epoch": 3.4,
+      "grad_norm": 2.703125,
+      "learning_rate": 2.3220223906602113e-07,
+      "loss": 0.7393,
+      "step": 2590
+    },
+    {
+      "epoch": 3.4,
+      "grad_norm": 2.75,
+      "learning_rate": 2.3109943403325397e-07,
+      "loss": 0.7519,
+      "step": 2591
+    },
+    {
+      "epoch": 3.4,
+      "grad_norm": 2.78125,
+      "learning_rate": 2.299991271105964e-07,
+      "loss": 0.717,
+      "step": 2592
+    },
+    {
+      "epoch": 3.41,
+      "grad_norm": 2.703125,
+      "learning_rate": 2.289013195095116e-07,
+      "loss": 0.7463,
+      "step": 2593
+    },
+    {
+      "epoch": 3.41,
+      "grad_norm": 2.78125,
+      "learning_rate": 2.2780601243870964e-07,
+      "loss": 0.7392,
+      "step": 2594
+    },
+    {
+      "epoch": 3.41,
+      "grad_norm": 2.78125,
+      "learning_rate": 2.2671320710414861e-07,
+      "loss": 0.7651,
+      "step": 2595
+    },
+    {
+      "epoch": 3.41,
+      "grad_norm": 2.859375,
+      "learning_rate": 2.2562290470903082e-07,
+      "loss": 0.73,
+      "step": 2596
+    },
+    {
+      "epoch": 3.41,
+      "grad_norm": 2.75,
+      "learning_rate": 2.2453510645380344e-07,
+      "loss": 0.7186,
+      "step": 2597
+    },
+    {
+      "epoch": 3.41,
+      "grad_norm": 2.734375,
+      "learning_rate": 2.2344981353615737e-07,
+      "loss": 0.704,
+      "step": 2598
+    },
+    {
+      "epoch": 3.41,
+      "grad_norm": 2.796875,
+      "learning_rate": 2.2236702715102366e-07,
+      "loss": 0.7348,
+      "step": 2599
+    },
+    {
+      "epoch": 3.41,
+      "grad_norm": 2.671875,
+      "learning_rate": 2.212867484905745e-07,
+      "loss": 0.7304,
+      "step": 2600
+    },
+    {
+      "epoch": 3.42,
+      "grad_norm": 2.828125,
+      "learning_rate": 2.2020897874422076e-07,
+      "loss": 0.7573,
+      "step": 2601
+    },
+    {
+      "epoch": 3.42,
+      "grad_norm": 2.8125,
+      "learning_rate": 2.191337190986112e-07,
+      "loss": 0.7553,
+      "step": 2602
+    },
+    {
+      "epoch": 3.42,
+      "grad_norm": 2.734375,
+      "learning_rate": 2.180609707376302e-07,
+      "loss": 0.7278,
+      "step": 2603
+    },
+    {
+      "epoch": 3.42,
+      "grad_norm": 2.921875,
+      "learning_rate": 2.16990734842398e-07,
+      "loss": 0.7378,
+      "step": 2604
+    },
+    {
+      "epoch": 3.42,
+      "grad_norm": 2.703125,
+      "learning_rate": 2.1592301259126787e-07,
+      "loss": 0.718,
+      "step": 2605
+    },
+    {
+      "epoch": 3.42,
+      "grad_norm": 2.734375,
+      "learning_rate": 2.148578051598263e-07,
+      "loss": 0.7515,
+      "step": 2606
+    },
+    {
+      "epoch": 3.42,
+      "grad_norm": 2.765625,
+      "learning_rate": 2.1379511372089018e-07,
+      "loss": 0.7162,
+      "step": 2607
+    },
+    {
+      "epoch": 3.43,
+      "grad_norm": 2.78125,
+      "learning_rate": 2.1273493944450634e-07,
+      "loss": 0.7593,
+      "step": 2608
+    },
+    {
+      "epoch": 3.43,
+      "grad_norm": 2.78125,
+      "learning_rate": 2.1167728349795025e-07,
+      "loss": 0.7483,
+      "step": 2609
+    },
+    {
+      "epoch": 3.43,
+      "grad_norm": 2.75,
+      "learning_rate": 2.1062214704572426e-07,
+      "loss": 0.7329,
+      "step": 2610
+    },
+    {
+      "epoch": 3.43,
+      "grad_norm": 2.765625,
+      "learning_rate": 2.09569531249558e-07,
+      "loss": 0.73,
+      "step": 2611
+    },
+    {
+      "epoch": 3.43,
+      "grad_norm": 2.78125,
+      "learning_rate": 2.085194372684049e-07,
+      "loss": 0.7422,
+      "step": 2612
+    },
+    {
+      "epoch": 3.43,
+      "grad_norm": 2.65625,
+      "learning_rate": 2.0747186625844183e-07,
+      "loss": 0.7077,
+      "step": 2613
+    },
+    {
+      "epoch": 3.43,
+      "grad_norm": 2.734375,
+      "learning_rate": 2.06426819373067e-07,
+      "loss": 0.7269,
+      "step": 2614
+    },
+    {
+      "epoch": 3.43,
+      "grad_norm": 2.78125,
+      "learning_rate": 2.0538429776290114e-07,
+      "loss": 0.7389,
+      "step": 2615
+    },
+    {
+      "epoch": 3.44,
+      "grad_norm": 2.71875,
+      "learning_rate": 2.0434430257578325e-07,
+      "loss": 0.7308,
+      "step": 2616
+    },
+    {
+      "epoch": 3.44,
+      "grad_norm": 2.84375,
+      "learning_rate": 2.0330683495677182e-07,
+      "loss": 0.7502,
+      "step": 2617
+    },
+    {
+      "epoch": 3.44,
+      "grad_norm": 2.734375,
+      "learning_rate": 2.0227189604814124e-07,
+      "loss": 0.7286,
+      "step": 2618
+    },
+    {
+      "epoch": 3.44,
+      "grad_norm": 2.796875,
+      "learning_rate": 2.0123948698938272e-07,
+      "loss": 0.7527,
+      "step": 2619
+    },
+    {
+      "epoch": 3.44,
+      "grad_norm": 2.75,
+      "learning_rate": 2.0020960891720147e-07,
+      "loss": 0.7329,
+      "step": 2620
+    },
+    {
+      "epoch": 3.44,
+      "grad_norm": 2.734375,
+      "learning_rate": 1.991822629655163e-07,
+      "loss": 0.7064,
+      "step": 2621
+    },
+    {
+      "epoch": 3.44,
+      "grad_norm": 2.734375,
+      "learning_rate": 1.981574502654579e-07,
+      "loss": 0.7558,
+      "step": 2622
+    },
+    {
+      "epoch": 3.45,
+      "grad_norm": 2.78125,
+      "learning_rate": 1.971351719453679e-07,
+      "loss": 0.7569,
+      "step": 2623
+    },
+    {
+      "epoch": 3.45,
+      "grad_norm": 2.71875,
+      "learning_rate": 1.9611542913079817e-07,
+      "loss": 0.7268,
+      "step": 2624
+    },
+    {
+      "epoch": 3.45,
+      "grad_norm": 2.78125,
+      "learning_rate": 1.9509822294450725e-07,
+      "loss": 0.7576,
+      "step": 2625
+    },
+    {
+      "epoch": 3.45,
+      "grad_norm": 2.828125,
+      "learning_rate": 1.9408355450646234e-07,
+      "loss": 0.7543,
+      "step": 2626
+    },
+    {
+      "epoch": 3.45,
+      "grad_norm": 2.75,
+      "learning_rate": 1.9307142493383585e-07,
+      "loss": 0.7103,
+      "step": 2627
+    },
+    {
+      "epoch": 3.45,
+      "grad_norm": 2.734375,
+      "learning_rate": 1.9206183534100474e-07,
+      "loss": 0.7038,
+      "step": 2628
+    },
+    {
+      "epoch": 3.45,
+      "grad_norm": 2.796875,
+      "learning_rate": 1.9105478683955042e-07,
+      "loss": 0.7318,
+      "step": 2629
+    },
+    {
+      "epoch": 3.45,
+      "grad_norm": 2.703125,
+      "learning_rate": 1.9005028053825592e-07,
+      "loss": 0.7094,
+      "step": 2630
+    },
+    {
+      "epoch": 3.46,
+      "grad_norm": 2.75,
+      "learning_rate": 1.8904831754310376e-07,
+      "loss": 0.7539,
+      "step": 2631
+    },
+    {
+      "epoch": 3.46,
+      "grad_norm": 2.78125,
+      "learning_rate": 1.8804889895727872e-07,
+      "loss": 0.7186,
+      "step": 2632
+    },
+    {
+      "epoch": 3.46,
+      "grad_norm": 2.75,
+      "learning_rate": 1.8705202588116227e-07,
+      "loss": 0.744,
+      "step": 2633
+    },
+    {
+      "epoch": 3.46,
+      "grad_norm": 2.9375,
+      "learning_rate": 1.8605769941233421e-07,
+      "loss": 0.7349,
+      "step": 2634
+    },
+    {
+      "epoch": 3.46,
+      "grad_norm": 2.71875,
+      "learning_rate": 1.850659206455699e-07,
+      "loss": 0.7295,
+      "step": 2635
+    },
+    {
+      "epoch": 3.46,
+      "grad_norm": 2.703125,
+      "learning_rate": 1.8407669067284028e-07,
+      "loss": 0.7094,
+      "step": 2636
+    },
+    {
+      "epoch": 3.46,
+      "grad_norm": 2.953125,
+      "learning_rate": 1.8309001058330944e-07,
+      "loss": 0.7257,
+      "step": 2637
+    },
+    {
+      "epoch": 3.47,
+      "grad_norm": 2.921875,
+      "learning_rate": 1.821058814633339e-07,
+      "loss": 0.7217,
+      "step": 2638
+    },
+    {
+      "epoch": 3.47,
+      "grad_norm": 2.765625,
+      "learning_rate": 1.8112430439646194e-07,
+      "loss": 0.7139,
+      "step": 2639
+    },
+    {
+      "epoch": 3.47,
+      "grad_norm": 2.828125,
+      "learning_rate": 1.8014528046343183e-07,
+      "loss": 0.7362,
+      "step": 2640
+    },
+    {
+      "epoch": 3.47,
+      "grad_norm": 2.765625,
+      "learning_rate": 1.7916881074217102e-07,
+      "loss": 0.7666,
+      "step": 2641
+    },
+    {
+      "epoch": 3.47,
+      "grad_norm": 2.96875,
+      "learning_rate": 1.7819489630779423e-07,
+      "loss": 0.7341,
+      "step": 2642
+    },
+    {
+      "epoch": 3.47,
+      "grad_norm": 2.71875,
+      "learning_rate": 1.7722353823260262e-07,
+      "loss": 0.7503,
+      "step": 2643
+    },
+    {
+      "epoch": 3.47,
+      "grad_norm": 2.75,
+      "learning_rate": 1.762547375860832e-07,
+      "loss": 0.7234,
+      "step": 2644
+    },
+    {
+      "epoch": 3.47,
+      "grad_norm": 2.734375,
+      "learning_rate": 1.752884954349074e-07,
+      "loss": 0.759,
+      "step": 2645
+    },
+    {
+      "epoch": 3.48,
+      "grad_norm": 2.75,
+      "learning_rate": 1.7432481284292923e-07,
+      "loss": 0.7275,
+      "step": 2646
+    },
+    {
+      "epoch": 3.48,
+      "grad_norm": 2.75,
+      "learning_rate": 1.7336369087118443e-07,
+      "loss": 0.7429,
+      "step": 2647
+    },
+    {
+      "epoch": 3.48,
+      "grad_norm": 2.78125,
+      "learning_rate": 1.7240513057789065e-07,
+      "loss": 0.7283,
+      "step": 2648
+    },
+    {
+      "epoch": 3.48,
+      "grad_norm": 2.75,
+      "learning_rate": 1.7144913301844313e-07,
+      "loss": 0.7485,
+      "step": 2649
+    },
+    {
+      "epoch": 3.48,
+      "grad_norm": 2.78125,
+      "learning_rate": 1.7049569924541653e-07,
+      "loss": 0.7214,
+      "step": 2650
+    },
+    {
+      "epoch": 3.48,
+      "grad_norm": 2.71875,
+      "learning_rate": 1.695448303085634e-07,
+      "loss": 0.7184,
+      "step": 2651
+    },
+    {
+      "epoch": 3.48,
+      "grad_norm": 2.75,
+      "learning_rate": 1.6859652725481095e-07,
+      "loss": 0.7651,
+      "step": 2652
+    },
+    {
+      "epoch": 3.49,
+      "grad_norm": 2.734375,
+      "learning_rate": 1.6765079112826266e-07,
+      "loss": 0.7198,
+      "step": 2653
+    },
+    {
+      "epoch": 3.49,
+      "grad_norm": 2.78125,
+      "learning_rate": 1.6670762297019444e-07,
+      "loss": 0.7473,
+      "step": 2654
+    },
+    {
+      "epoch": 3.49,
+      "grad_norm": 2.71875,
+      "learning_rate": 1.657670238190559e-07,
+      "loss": 0.7111,
+      "step": 2655
+    },
+    {
+      "epoch": 3.49,
+      "grad_norm": 2.765625,
+      "learning_rate": 1.6482899471046726e-07,
+      "loss": 0.7149,
+      "step": 2656
+    },
+    {
+      "epoch": 3.49,
+      "grad_norm": 2.890625,
+      "learning_rate": 1.6389353667721984e-07,
+      "loss": 0.7794,
+      "step": 2657
+    },
+    {
+      "epoch": 3.49,
+      "grad_norm": 2.78125,
+      "learning_rate": 1.629606507492737e-07,
+      "loss": 0.751,
+      "step": 2658
+    },
+    {
+      "epoch": 3.49,
+      "grad_norm": 2.75,
+      "learning_rate": 1.6203033795375717e-07,
+      "loss": 0.739,
+      "step": 2659
+    },
+    {
+      "epoch": 3.49,
+      "grad_norm": 2.75,
+      "learning_rate": 1.6110259931496564e-07,
+      "loss": 0.7302,
+      "step": 2660
+    },
+    {
+      "epoch": 3.5,
+      "grad_norm": 2.75,
+      "learning_rate": 1.6017743585435897e-07,
+      "loss": 0.7629,
+      "step": 2661
+    },
+    {
+      "epoch": 3.5,
+      "grad_norm": 2.796875,
+      "learning_rate": 1.5925484859056372e-07,
+      "loss": 0.7414,
+      "step": 2662
+    },
+    {
+      "epoch": 3.5,
+      "grad_norm": 2.890625,
+      "learning_rate": 1.583348385393685e-07,
+      "loss": 0.7227,
+      "step": 2663
+    },
+    {
+      "epoch": 3.5,
+      "grad_norm": 2.703125,
+      "learning_rate": 1.5741740671372548e-07,
+      "loss": 0.7169,
+      "step": 2664
+    },
+    {
+      "epoch": 3.5,
+      "grad_norm": 2.765625,
+      "learning_rate": 1.5650255412374638e-07,
+      "loss": 0.7348,
+      "step": 2665
+    },
+    {
+      "epoch": 3.5,
+      "grad_norm": 2.703125,
+      "learning_rate": 1.555902817767063e-07,
+      "loss": 0.71,
+      "step": 2666
+    },
+    {
+      "epoch": 3.5,
+      "grad_norm": 2.75,
+      "learning_rate": 1.5468059067703571e-07,
+      "loss": 0.7191,
+      "step": 2667
+    },
+    {
+      "epoch": 3.51,
+      "grad_norm": 2.78125,
+      "learning_rate": 1.5377348182632536e-07,
+      "loss": 0.7435,
+      "step": 2668
+    },
+    {
+      "epoch": 3.51,
+      "grad_norm": 2.75,
+      "learning_rate": 1.528689562233221e-07,
+      "loss": 0.7107,
+      "step": 2669
+    },
+    {
+      "epoch": 3.51,
+      "grad_norm": 2.796875,
+      "learning_rate": 1.5196701486392906e-07,
+      "loss": 0.7216,
+      "step": 2670
+    },
+    {
+      "epoch": 3.51,
+      "grad_norm": 2.75,
+      "learning_rate": 1.5106765874120382e-07,
+      "loss": 0.7617,
+      "step": 2671
+    },
+    {
+      "epoch": 3.51,
+      "grad_norm": 2.75,
+      "learning_rate": 1.501708888453579e-07,
+      "loss": 0.7454,
+      "step": 2672
+    },
+    {
+      "epoch": 3.51,
+      "grad_norm": 2.859375,
+      "learning_rate": 1.4927670616375345e-07,
+      "loss": 0.7424,
+      "step": 2673
+    },
+    {
+      "epoch": 3.51,
+      "grad_norm": 3.09375,
+      "learning_rate": 1.4838511168090707e-07,
+      "loss": 0.7219,
+      "step": 2674
+    },
+    {
+      "epoch": 3.51,
+      "grad_norm": 2.828125,
+      "learning_rate": 1.4749610637848383e-07,
+      "loss": 0.7569,
+      "step": 2675
+    },
+    {
+      "epoch": 3.52,
+      "grad_norm": 2.828125,
+      "learning_rate": 1.4660969123529795e-07,
+      "loss": 0.7219,
+      "step": 2676
+    },
+    {
+      "epoch": 3.52,
+      "grad_norm": 2.796875,
+      "learning_rate": 1.4572586722731292e-07,
+      "loss": 0.7286,
+      "step": 2677
+    },
+    {
+      "epoch": 3.52,
+      "grad_norm": 2.75,
+      "learning_rate": 1.4484463532763809e-07,
+      "loss": 0.729,
+      "step": 2678
+    },
+    {
+      "epoch": 3.52,
+      "grad_norm": 2.65625,
+      "learning_rate": 1.4396599650652959e-07,
+      "loss": 0.7401,
+      "step": 2679
+    },
+    {
+      "epoch": 3.52,
+      "grad_norm": 2.71875,
+      "learning_rate": 1.4308995173138828e-07,
+      "loss": 0.7234,
+      "step": 2680
+    },
+    {
+      "epoch": 3.52,
+      "grad_norm": 2.828125,
+      "learning_rate": 1.4221650196675901e-07,
+      "loss": 0.7358,
+      "step": 2681
+    },
+    {
+      "epoch": 3.52,
+      "grad_norm": 2.796875,
+      "learning_rate": 1.4134564817432972e-07,
+      "loss": 0.7345,
+      "step": 2682
+    },
+    {
+      "epoch": 3.53,
+      "grad_norm": 2.890625,
+      "learning_rate": 1.4047739131292953e-07,
+      "loss": 0.758,
+      "step": 2683
+    },
+    {
+      "epoch": 3.53,
+      "grad_norm": 2.796875,
+      "learning_rate": 1.3961173233852876e-07,
+      "loss": 0.7417,
+      "step": 2684
+    },
+    {
+      "epoch": 3.53,
+      "grad_norm": 2.78125,
+      "learning_rate": 1.3874867220423721e-07,
+      "loss": 0.72,
+      "step": 2685
+    },
+    {
+      "epoch": 3.53,
+      "grad_norm": 2.84375,
+      "learning_rate": 1.3788821186030338e-07,
+      "loss": 0.7787,
+      "step": 2686
+    },
+    {
+      "epoch": 3.53,
+      "grad_norm": 2.703125,
+      "learning_rate": 1.370303522541133e-07,
+      "loss": 0.7291,
+      "step": 2687
+    },
+    {
+      "epoch": 3.53,
+      "grad_norm": 2.71875,
+      "learning_rate": 1.3617509433018928e-07,
+      "loss": 0.7344,
+      "step": 2688
+    },
+    {
+      "epoch": 3.53,
+      "grad_norm": 2.71875,
+      "learning_rate": 1.3532243903018993e-07,
+      "loss": 0.7444,
+      "step": 2689
+    },
+    {
+      "epoch": 3.53,
+      "grad_norm": 2.703125,
+      "learning_rate": 1.3447238729290713e-07,
+      "loss": 0.7318,
+      "step": 2690
+    },
+    {
+      "epoch": 3.54,
+      "grad_norm": 2.703125,
+      "learning_rate": 1.3362494005426662e-07,
+      "loss": 0.7339,
+      "step": 2691
+    },
+    {
+      "epoch": 3.54,
+      "grad_norm": 2.78125,
+      "learning_rate": 1.3278009824732763e-07,
+      "loss": 0.748,
+      "step": 2692
+    },
+    {
+      "epoch": 3.54,
+      "grad_norm": 2.8125,
+      "learning_rate": 1.3193786280227912e-07,
+      "loss": 0.7342,
+      "step": 2693
+    },
+    {
+      "epoch": 3.54,
+      "grad_norm": 2.8125,
+      "learning_rate": 1.310982346464415e-07,
+      "loss": 0.7375,
+      "step": 2694
+    },
+    {
+      "epoch": 3.54,
+      "grad_norm": 2.796875,
+      "learning_rate": 1.3026121470426394e-07,
+      "loss": 0.7616,
+      "step": 2695
+    },
+    {
+      "epoch": 3.54,
+      "grad_norm": 2.6875,
+      "learning_rate": 1.2942680389732398e-07,
+      "loss": 0.7177,
+      "step": 2696
+    },
+    {
+      "epoch": 3.54,
+      "grad_norm": 2.65625,
+      "learning_rate": 1.2859500314432616e-07,
+      "loss": 0.7198,
+      "step": 2697
+    },
+    {
+      "epoch": 3.55,
+      "grad_norm": 2.84375,
+      "learning_rate": 1.2776581336110234e-07,
+      "loss": 0.7457,
+      "step": 2698
+    },
+    {
+      "epoch": 3.55,
+      "grad_norm": 2.796875,
+      "learning_rate": 1.269392354606086e-07,
+      "loss": 0.7488,
+      "step": 2699
+    },
+    {
+      "epoch": 3.55,
+      "grad_norm": 2.828125,
+      "learning_rate": 1.2611527035292558e-07,
+      "loss": 0.7396,
+      "step": 2700
+    },
+    {
+      "epoch": 3.55,
+      "grad_norm": 2.84375,
+      "learning_rate": 1.2529391894525733e-07,
+      "loss": 0.7652,
+      "step": 2701
+    },
+    {
+      "epoch": 3.55,
+      "grad_norm": 2.84375,
+      "learning_rate": 1.2447518214193017e-07,
+      "loss": 0.7386,
+      "step": 2702
+    },
+    {
+      "epoch": 3.55,
+      "grad_norm": 2.859375,
+      "learning_rate": 1.2365906084439133e-07,
+      "loss": 0.7597,
+      "step": 2703
+    },
+    {
+      "epoch": 3.55,
+      "grad_norm": 2.78125,
+      "learning_rate": 1.2284555595120901e-07,
+      "loss": 0.7328,
+      "step": 2704
+    },
+    {
+      "epoch": 3.55,
+      "grad_norm": 2.75,
+      "learning_rate": 1.2203466835806977e-07,
+      "loss": 0.7354,
+      "step": 2705
+    },
+    {
+      "epoch": 3.56,
+      "grad_norm": 2.734375,
+      "learning_rate": 1.2122639895777887e-07,
+      "loss": 0.7442,
+      "step": 2706
+    },
+    {
+      "epoch": 3.56,
+      "grad_norm": 2.71875,
+      "learning_rate": 1.2042074864025977e-07,
+      "loss": 0.7365,
+      "step": 2707
+    },
+    {
+      "epoch": 3.56,
+      "grad_norm": 2.84375,
+      "learning_rate": 1.1961771829255037e-07,
+      "loss": 0.7599,
+      "step": 2708
+    },
+    {
+      "epoch": 3.56,
+      "grad_norm": 2.75,
+      "learning_rate": 1.1881730879880565e-07,
+      "loss": 0.751,
+      "step": 2709
+    },
+    {
+      "epoch": 3.56,
+      "grad_norm": 2.671875,
+      "learning_rate": 1.1801952104029347e-07,
+      "loss": 0.713,
+      "step": 2710
+    },
+    {
+      "epoch": 3.56,
+      "grad_norm": 2.78125,
+      "learning_rate": 1.1722435589539699e-07,
+      "loss": 0.7472,
+      "step": 2711
+    },
+    {
+      "epoch": 3.56,
+      "grad_norm": 2.734375,
+      "learning_rate": 1.1643181423961037e-07,
+      "loss": 0.7608,
+      "step": 2712
+    },
+    {
+      "epoch": 3.57,
+      "grad_norm": 2.796875,
+      "learning_rate": 1.1564189694554001e-07,
+      "loss": 0.7604,
+      "step": 2713
+    },
+    {
+      "epoch": 3.57,
+      "grad_norm": 2.796875,
+      "learning_rate": 1.1485460488290163e-07,
+      "loss": 0.7267,
+      "step": 2714
+    },
+    {
+      "epoch": 3.57,
+      "grad_norm": 2.734375,
+      "learning_rate": 1.140699389185218e-07,
+      "loss": 0.7258,
+      "step": 2715
+    },
+    {
+      "epoch": 3.57,
+      "grad_norm": 2.734375,
+      "learning_rate": 1.1328789991633532e-07,
+      "loss": 0.7486,
+      "step": 2716
+    },
+    {
+      "epoch": 3.57,
+      "grad_norm": 2.890625,
+      "learning_rate": 1.1250848873738452e-07,
+      "loss": 0.7492,
+      "step": 2717
+    },
+    {
+      "epoch": 3.57,
+      "grad_norm": 2.765625,
+      "learning_rate": 1.1173170623981822e-07,
+      "loss": 0.7401,
+      "step": 2718
+    },
+    {
+      "epoch": 3.57,
+      "grad_norm": 2.765625,
+      "learning_rate": 1.1095755327889146e-07,
+      "loss": 0.7704,
+      "step": 2719
+    },
+    {
+      "epoch": 3.57,
+      "grad_norm": 2.703125,
+      "learning_rate": 1.1018603070696409e-07,
+      "loss": 0.7402,
+      "step": 2720
+    },
+    {
+      "epoch": 3.58,
+      "grad_norm": 2.71875,
+      "learning_rate": 1.094171393734994e-07,
+      "loss": 0.7158,
+      "step": 2721
+    },
+    {
+      "epoch": 3.58,
+      "grad_norm": 2.8125,
+      "learning_rate": 1.0865088012506408e-07,
+      "loss": 0.7411,
+      "step": 2722
+    },
+    {
+      "epoch": 3.58,
+      "grad_norm": 2.734375,
+      "learning_rate": 1.078872538053266e-07,
+      "loss": 0.7364,
+      "step": 2723
+    },
+    {
+      "epoch": 3.58,
+      "grad_norm": 2.84375,
+      "learning_rate": 1.0712626125505665e-07,
+      "loss": 0.7643,
+      "step": 2724
+    },
+    {
+      "epoch": 3.58,
+      "grad_norm": 2.78125,
+      "learning_rate": 1.0636790331212399e-07,
+      "loss": 0.727,
+      "step": 2725
+    },
+    {
+      "epoch": 3.58,
+      "grad_norm": 2.75,
+      "learning_rate": 1.0561218081149711e-07,
+      "loss": 0.7175,
+      "step": 2726
+    },
+    {
+      "epoch": 3.58,
+      "grad_norm": 2.875,
+      "learning_rate": 1.0485909458524402e-07,
+      "loss": 0.7593,
+      "step": 2727
+    },
+    {
+      "epoch": 3.59,
+      "grad_norm": 2.796875,
+      "learning_rate": 1.0410864546252841e-07,
+      "loss": 0.7395,
+      "step": 2728
+    },
+    {
+      "epoch": 3.59,
+      "grad_norm": 2.59375,
+      "learning_rate": 1.0336083426961296e-07,
+      "loss": 0.6924,
+      "step": 2729
+    },
+    {
+      "epoch": 3.59,
+      "grad_norm": 2.859375,
+      "learning_rate": 1.0261566182985378e-07,
+      "loss": 0.7466,
+      "step": 2730
+    },
+    {
+      "epoch": 3.59,
+      "grad_norm": 2.75,
+      "learning_rate": 1.0187312896370233e-07,
+      "loss": 0.727,
+      "step": 2731
+    },
+    {
+      "epoch": 3.59,
+      "grad_norm": 2.796875,
+      "learning_rate": 1.0113323648870354e-07,
+      "loss": 0.7462,
+      "step": 2732
+    },
+    {
+      "epoch": 3.59,
+      "grad_norm": 2.859375,
+      "learning_rate": 1.0039598521949578e-07,
+      "loss": 0.7598,
+      "step": 2733
+    },
+    {
+      "epoch": 3.59,
+      "grad_norm": 2.796875,
+      "learning_rate": 9.966137596780945e-08,
+      "loss": 0.7507,
+      "step": 2734
+    },
+    {
+      "epoch": 3.59,
+      "grad_norm": 2.75,
+      "learning_rate": 9.892940954246532e-08,
+      "loss": 0.7347,
+      "step": 2735
+    },
+    {
+      "epoch": 3.6,
+      "grad_norm": 2.8125,
+      "learning_rate": 9.820008674937459e-08,
+      "loss": 0.7634,
+      "step": 2736
+    },
+    {
+      "epoch": 3.6,
+      "grad_norm": 2.78125,
+      "learning_rate": 9.747340839153852e-08,
+      "loss": 0.7517,
+      "step": 2737
+    },
+    {
+      "epoch": 3.6,
+      "grad_norm": 2.75,
+      "learning_rate": 9.674937526904571e-08,
+      "loss": 0.7389,
+      "step": 2738
+    },
+    {
+      "epoch": 3.6,
+      "grad_norm": 2.828125,
+      "learning_rate": 9.602798817907294e-08,
+      "loss": 0.7393,
+      "step": 2739
+    },
+    {
+      "epoch": 3.6,
+      "grad_norm": 2.796875,
+      "learning_rate": 9.530924791588319e-08,
+      "loss": 0.7612,
+      "step": 2740
+    },
+    {
+      "epoch": 3.6,
+      "grad_norm": 2.734375,
+      "learning_rate": 9.459315527082624e-08,
+      "loss": 0.7261,
+      "step": 2741
+    },
+    {
+      "epoch": 3.6,
+      "grad_norm": 2.78125,
+      "learning_rate": 9.387971103233556e-08,
+      "loss": 0.7272,
+      "step": 2742
+    },
+    {
+      "epoch": 3.61,
+      "grad_norm": 2.703125,
+      "learning_rate": 9.316891598592892e-08,
+      "loss": 0.7141,
+      "step": 2743
+    },
+    {
+      "epoch": 3.61,
+      "grad_norm": 2.765625,
+      "learning_rate": 9.24607709142078e-08,
+      "loss": 0.7343,
+      "step": 2744
+    },
+    {
+      "epoch": 3.61,
+      "grad_norm": 2.78125,
+      "learning_rate": 9.175527659685573e-08,
+      "loss": 0.7291,
+      "step": 2745
+    },
+    {
+      "epoch": 3.61,
+      "grad_norm": 2.703125,
+      "learning_rate": 9.10524338106375e-08,
+      "loss": 0.6963,
+      "step": 2746
+    },
+    {
+      "epoch": 3.61,
+      "grad_norm": 2.734375,
+      "learning_rate": 9.035224332939879e-08,
+      "loss": 0.7264,
+      "step": 2747
+    },
+    {
+      "epoch": 3.61,
+      "grad_norm": 2.828125,
+      "learning_rate": 8.965470592406544e-08,
+      "loss": 0.758,
+      "step": 2748
+    },
+    {
+      "epoch": 3.61,
+      "grad_norm": 2.734375,
+      "learning_rate": 8.895982236264117e-08,
+      "loss": 0.7443,
+      "step": 2749
+    },
+    {
+      "epoch": 3.61,
+      "grad_norm": 2.78125,
+      "learning_rate": 8.826759341020813e-08,
+      "loss": 0.7427,
+      "step": 2750
+    },
+    {
+      "epoch": 3.62,
+      "grad_norm": 2.84375,
+      "learning_rate": 8.757801982892583e-08,
+      "loss": 0.7344,
+      "step": 2751
+    },
+    {
+      "epoch": 3.62,
+      "grad_norm": 2.75,
+      "learning_rate": 8.689110237803056e-08,
+      "loss": 0.7442,
+      "step": 2752
+    },
+    {
+      "epoch": 3.62,
+      "grad_norm": 2.765625,
+      "learning_rate": 8.620684181383343e-08,
+      "loss": 0.7205,
+      "step": 2753
+    },
+    {
+      "epoch": 3.62,
+      "grad_norm": 2.765625,
+      "learning_rate": 8.552523888972097e-08,
+      "loss": 0.7302,
+      "step": 2754
+    },
+    {
+      "epoch": 3.62,
+      "grad_norm": 2.75,
+      "learning_rate": 8.48462943561526e-08,
+      "loss": 0.745,
+      "step": 2755
+    },
+    {
+      "epoch": 3.62,
+      "grad_norm": 2.75,
+      "learning_rate": 8.417000896066201e-08,
+      "loss": 0.7418,
+      "step": 2756
+    },
+    {
+      "epoch": 3.62,
+      "grad_norm": 2.734375,
+      "learning_rate": 8.349638344785388e-08,
+      "loss": 0.7604,
+      "step": 2757
+    },
+    {
+      "epoch": 3.63,
+      "grad_norm": 2.84375,
+      "learning_rate": 8.282541855940546e-08,
+      "loss": 0.7529,
+      "step": 2758
+    },
+    {
+      "epoch": 3.63,
+      "grad_norm": 2.796875,
+      "learning_rate": 8.21571150340636e-08,
+      "loss": 0.7441,
+      "step": 2759
+    },
+    {
+      "epoch": 3.63,
+      "grad_norm": 2.796875,
+      "learning_rate": 8.149147360764609e-08,
+      "loss": 0.7247,
+      "step": 2760
+    },
+    {
+      "epoch": 3.63,
+      "grad_norm": 2.796875,
+      "learning_rate": 8.082849501303836e-08,
+      "loss": 0.7345,
+      "step": 2761
+    },
+    {
+      "epoch": 3.63,
+      "grad_norm": 2.71875,
+      "learning_rate": 8.016817998019455e-08,
+      "loss": 0.7331,
+      "step": 2762
+    },
+    {
+      "epoch": 3.63,
+      "grad_norm": 2.75,
+      "learning_rate": 7.951052923613644e-08,
+      "loss": 0.726,
+      "step": 2763
+    },
+    {
+      "epoch": 3.63,
+      "grad_norm": 2.78125,
+      "learning_rate": 7.885554350495206e-08,
+      "loss": 0.7535,
+      "step": 2764
+    },
+    {
+      "epoch": 3.63,
+      "grad_norm": 2.78125,
+      "learning_rate": 7.820322350779536e-08,
+      "loss": 0.7317,
+      "step": 2765
+    },
+    {
+      "epoch": 3.64,
+      "grad_norm": 2.6875,
+      "learning_rate": 7.755356996288571e-08,
+      "loss": 0.7013,
+      "step": 2766
+    },
+    {
+      "epoch": 3.64,
+      "grad_norm": 2.78125,
+      "learning_rate": 7.69065835855054e-08,
+      "loss": 0.7384,
+      "step": 2767
+    },
+    {
+      "epoch": 3.64,
+      "grad_norm": 2.71875,
+      "learning_rate": 7.626226508800128e-08,
+      "loss": 0.7478,
+      "step": 2768
+    },
+    {
+      "epoch": 3.64,
+      "grad_norm": 2.828125,
+      "learning_rate": 7.562061517978198e-08,
+      "loss": 0.7591,
+      "step": 2769
+    },
+    {
+      "epoch": 3.64,
+      "grad_norm": 2.796875,
+      "learning_rate": 7.498163456731878e-08,
+      "loss": 0.7609,
+      "step": 2770
+    },
+    {
+      "epoch": 3.64,
+      "grad_norm": 2.78125,
+      "learning_rate": 7.434532395414335e-08,
+      "loss": 0.7136,
+      "step": 2771
+    },
+    {
+      "epoch": 3.64,
+      "grad_norm": 2.765625,
+      "learning_rate": 7.371168404084806e-08,
+      "loss": 0.7357,
+      "step": 2772
+    },
+    {
+      "epoch": 3.65,
+      "grad_norm": 2.78125,
+      "learning_rate": 7.308071552508455e-08,
+      "loss": 0.7532,
+      "step": 2773
+    },
+    {
+      "epoch": 3.65,
+      "grad_norm": 2.765625,
+      "learning_rate": 7.245241910156297e-08,
+      "loss": 0.7328,
+      "step": 2774
+    },
+    {
+      "epoch": 3.65,
+      "grad_norm": 2.765625,
+      "learning_rate": 7.18267954620519e-08,
+      "loss": 0.7677,
+      "step": 2775
+    },
+    {
+      "epoch": 3.65,
+      "grad_norm": 2.78125,
+      "learning_rate": 7.120384529537672e-08,
+      "loss": 0.7534,
+      "step": 2776
+    },
+    {
+      "epoch": 3.65,
+      "grad_norm": 2.859375,
+      "learning_rate": 7.058356928741988e-08,
+      "loss": 0.7279,
+      "step": 2777
+    },
+    {
+      "epoch": 3.65,
+      "grad_norm": 2.703125,
+      "learning_rate": 6.996596812111872e-08,
+      "loss": 0.7258,
+      "step": 2778
+    },
+    {
+      "epoch": 3.65,
+      "grad_norm": 2.765625,
+      "learning_rate": 6.935104247646596e-08,
+      "loss": 0.7477,
+      "step": 2779
+    },
+    {
+      "epoch": 3.65,
+      "grad_norm": 2.703125,
+      "learning_rate": 6.873879303050863e-08,
+      "loss": 0.7258,
+      "step": 2780
+    },
+    {
+      "epoch": 3.66,
+      "grad_norm": 2.859375,
+      "learning_rate": 6.812922045734666e-08,
+      "loss": 0.7537,
+      "step": 2781
+    },
+    {
+      "epoch": 3.66,
+      "grad_norm": 2.796875,
+      "learning_rate": 6.752232542813319e-08,
+      "loss": 0.7369,
+      "step": 2782
+    },
+    {
+      "epoch": 3.66,
+      "grad_norm": 2.78125,
+      "learning_rate": 6.691810861107318e-08,
+      "loss": 0.7418,
+      "step": 2783
+    },
+    {
+      "epoch": 3.66,
+      "grad_norm": 2.78125,
+      "learning_rate": 6.631657067142306e-08,
+      "loss": 0.7394,
+      "step": 2784
+    },
+    {
+      "epoch": 3.66,
+      "grad_norm": 2.75,
+      "learning_rate": 6.571771227148943e-08,
+      "loss": 0.7529,
+      "step": 2785
+    },
+    {
+      "epoch": 3.66,
+      "grad_norm": 2.765625,
+      "learning_rate": 6.512153407062849e-08,
+      "loss": 0.7551,
+      "step": 2786
+    },
+    {
+      "epoch": 3.66,
+      "grad_norm": 2.828125,
+      "learning_rate": 6.452803672524571e-08,
+      "loss": 0.746,
+      "step": 2787
+    },
+    {
+      "epoch": 3.67,
+      "grad_norm": 2.78125,
+      "learning_rate": 6.393722088879534e-08,
+      "loss": 0.7208,
+      "step": 2788
+    },
+    {
+      "epoch": 3.67,
+      "grad_norm": 2.8125,
+      "learning_rate": 6.334908721177868e-08,
+      "loss": 0.7459,
+      "step": 2789
+    },
+    {
+      "epoch": 3.67,
+      "grad_norm": 2.734375,
+      "learning_rate": 6.276363634174359e-08,
+      "loss": 0.7333,
+      "step": 2790
+    },
+    {
+      "epoch": 3.67,
+      "grad_norm": 2.8125,
+      "learning_rate": 6.218086892328446e-08,
+      "loss": 0.7524,
+      "step": 2791
+    },
+    {
+      "epoch": 3.67,
+      "grad_norm": 2.75,
+      "learning_rate": 6.160078559804189e-08,
+      "loss": 0.7351,
+      "step": 2792
+    },
+    {
+      "epoch": 3.67,
+      "grad_norm": 2.75,
+      "learning_rate": 6.102338700470028e-08,
+      "loss": 0.7482,
+      "step": 2793
+    },
+    {
+      "epoch": 3.67,
+      "grad_norm": 2.78125,
+      "learning_rate": 6.044867377898806e-08,
+      "loss": 0.7403,
+      "step": 2794
+    },
+    {
+      "epoch": 3.67,
+      "grad_norm": 2.765625,
+      "learning_rate": 5.987664655367792e-08,
+      "loss": 0.7417,
+      "step": 2795
+    },
+    {
+      "epoch": 3.68,
+      "grad_norm": 2.671875,
+      "learning_rate": 5.930730595858414e-08,
+      "loss": 0.7322,
+      "step": 2796
+    },
+    {
+      "epoch": 3.68,
+      "grad_norm": 2.828125,
+      "learning_rate": 5.8740652620563334e-08,
+      "loss": 0.7684,
+      "step": 2797
+    },
+    {
+      "epoch": 3.68,
+      "grad_norm": 2.734375,
+      "learning_rate": 5.817668716351338e-08,
+      "loss": 0.7392,
+      "step": 2798
+    },
+    {
+      "epoch": 3.68,
+      "grad_norm": 2.796875,
+      "learning_rate": 5.7615410208373415e-08,
+      "loss": 0.7334,
+      "step": 2799
+    },
+    {
+      "epoch": 3.68,
+      "grad_norm": 2.75,
+      "learning_rate": 5.7056822373121324e-08,
+      "loss": 0.7448,
+      "step": 2800
+    },
+    {
+      "epoch": 3.68,
+      "grad_norm": 2.734375,
+      "learning_rate": 5.650092427277487e-08,
+      "loss": 0.7345,
+      "step": 2801
+    },
+    {
+      "epoch": 3.68,
+      "grad_norm": 2.796875,
+      "learning_rate": 5.594771651939057e-08,
+      "loss": 0.7373,
+      "step": 2802
+    },
+    {
+      "epoch": 3.68,
+      "grad_norm": 2.71875,
+      "learning_rate": 5.5397199722062034e-08,
+      "loss": 0.7372,
+      "step": 2803
+    },
+    {
+      "epoch": 3.69,
+      "grad_norm": 2.796875,
+      "learning_rate": 5.484937448692079e-08,
+      "loss": 0.7445,
+      "step": 2804
+    },
+    {
+      "epoch": 3.69,
+      "grad_norm": 2.71875,
+      "learning_rate": 5.430424141713464e-08,
+      "loss": 0.7263,
+      "step": 2805
+    },
+    {
+      "epoch": 3.69,
+      "grad_norm": 2.78125,
+      "learning_rate": 5.3761801112907356e-08,
+      "loss": 0.7268,
+      "step": 2806
+    },
+    {
+      "epoch": 3.69,
+      "grad_norm": 2.84375,
+      "learning_rate": 5.3222054171477574e-08,
+      "loss": 0.7446,
+      "step": 2807
+    },
+    {
+      "epoch": 3.69,
+      "grad_norm": 2.828125,
+      "learning_rate": 5.268500118711883e-08,
+      "loss": 0.7415,
+      "step": 2808
+    },
+    {
+      "epoch": 3.69,
+      "grad_norm": 2.71875,
+      "learning_rate": 5.2150642751138124e-08,
+      "loss": 0.7326,
+      "step": 2809
+    },
+    {
+      "epoch": 3.69,
+      "grad_norm": 2.765625,
+      "learning_rate": 5.16189794518765e-08,
+      "loss": 0.7352,
+      "step": 2810
+    },
+    {
+      "epoch": 3.7,
+      "grad_norm": 2.765625,
+      "learning_rate": 5.10900118747068e-08,
+      "loss": 0.7468,
+      "step": 2811
+    },
+    {
+      "epoch": 3.7,
+      "grad_norm": 2.8125,
+      "learning_rate": 5.0563740602034284e-08,
+      "loss": 0.7509,
+      "step": 2812
+    },
+    {
+      "epoch": 3.7,
+      "grad_norm": 2.734375,
+      "learning_rate": 5.004016621329544e-08,
+      "loss": 0.7191,
+      "step": 2813
+    },
+    {
+      "epoch": 3.7,
+      "grad_norm": 2.671875,
+      "learning_rate": 4.9519289284956916e-08,
+      "loss": 0.7324,
+      "step": 2814
+    },
+    {
+      "epoch": 3.7,
+      "grad_norm": 2.796875,
+      "learning_rate": 4.900111039051636e-08,
+      "loss": 0.7422,
+      "step": 2815
+    },
+    {
+      "epoch": 3.7,
+      "grad_norm": 2.796875,
+      "learning_rate": 4.848563010049989e-08,
+      "loss": 0.7299,
+      "step": 2816
+    },
+    {
+      "epoch": 3.7,
+      "grad_norm": 2.78125,
+      "learning_rate": 4.797284898246296e-08,
+      "loss": 0.7403,
+      "step": 2817
+    },
+    {
+      "epoch": 3.7,
+      "grad_norm": 2.84375,
+      "learning_rate": 4.746276760098867e-08,
+      "loss": 0.7352,
+      "step": 2818
+    },
+    {
+      "epoch": 3.71,
+      "grad_norm": 2.765625,
+      "learning_rate": 4.695538651768861e-08,
+      "loss": 0.7482,
+      "step": 2819
+    },
+    {
+      "epoch": 3.71,
+      "grad_norm": 2.734375,
+      "learning_rate": 4.6450706291200354e-08,
+      "loss": 0.7571,
+      "step": 2820
+    },
+    {
+      "epoch": 3.71,
+      "grad_norm": 2.8125,
+      "learning_rate": 4.594872747718804e-08,
+      "loss": 0.752,
+      "step": 2821
+    },
+    {
+      "epoch": 3.71,
+      "grad_norm": 2.890625,
+      "learning_rate": 4.54494506283415e-08,
+      "loss": 0.7472,
+      "step": 2822
+    },
+    {
+      "epoch": 3.71,
+      "grad_norm": 2.734375,
+      "learning_rate": 4.4952876294376e-08,
+      "loss": 0.7175,
+      "step": 2823
+    },
+    {
+      "epoch": 3.71,
+      "grad_norm": 2.75,
+      "learning_rate": 4.44590050220306e-08,
+      "loss": 0.7253,
+      "step": 2824
+    },
+    {
+      "epoch": 3.71,
+      "grad_norm": 2.796875,
+      "learning_rate": 4.396783735506893e-08,
+      "loss": 0.7376,
+      "step": 2825
+    },
+    {
+      "epoch": 3.72,
+      "grad_norm": 2.734375,
+      "learning_rate": 4.3479373834277026e-08,
+      "loss": 0.7321,
+      "step": 2826
+    },
+    {
+      "epoch": 3.72,
+      "grad_norm": 2.71875,
+      "learning_rate": 4.299361499746441e-08,
+      "loss": 0.739,
+      "step": 2827
+    },
+    {
+      "epoch": 3.72,
+      "grad_norm": 2.75,
+      "learning_rate": 4.2510561379462136e-08,
+      "loss": 0.7638,
+      "step": 2828
+    },
+    {
+      "epoch": 3.72,
+      "grad_norm": 2.765625,
+      "learning_rate": 4.2030213512123384e-08,
+      "loss": 0.7394,
+      "step": 2829
+    },
+    {
+      "epoch": 3.72,
+      "grad_norm": 2.796875,
+      "learning_rate": 4.155257192432205e-08,
+      "loss": 0.7281,
+      "step": 2830
+    },
+    {
+      "epoch": 3.72,
+      "grad_norm": 2.765625,
+      "learning_rate": 4.1077637141951624e-08,
+      "loss": 0.7392,
+      "step": 2831
+    },
+    {
+      "epoch": 3.72,
+      "grad_norm": 2.875,
+      "learning_rate": 4.060540968792631e-08,
+      "loss": 0.7199,
+      "step": 2832
+    },
+    {
+      "epoch": 3.72,
+      "grad_norm": 2.71875,
+      "learning_rate": 4.013589008217883e-08,
+      "loss": 0.7199,
+      "step": 2833
+    },
+    {
+      "epoch": 3.73,
+      "grad_norm": 2.765625,
+      "learning_rate": 3.9669078841660655e-08,
+      "loss": 0.7228,
+      "step": 2834
+    },
+    {
+      "epoch": 3.73,
+      "grad_norm": 2.71875,
+      "learning_rate": 3.92049764803415e-08,
+      "loss": 0.7264,
+      "step": 2835
+    },
+    {
+      "epoch": 3.73,
+      "grad_norm": 2.765625,
+      "learning_rate": 3.874358350920843e-08,
+      "loss": 0.7295,
+      "step": 2836
+    },
+    {
+      "epoch": 3.73,
+      "grad_norm": 2.75,
+      "learning_rate": 3.828490043626537e-08,
+      "loss": 0.7348,
+      "step": 2837
+    },
+    {
+      "epoch": 3.73,
+      "grad_norm": 2.71875,
+      "learning_rate": 3.782892776653252e-08,
+      "loss": 0.7246,
+      "step": 2838
+    },
+    {
+      "epoch": 3.73,
+      "grad_norm": 2.78125,
+      "learning_rate": 3.7375666002046064e-08,
+      "loss": 0.7531,
+      "step": 2839
+    },
+    {
+      "epoch": 3.73,
+      "grad_norm": 2.84375,
+      "learning_rate": 3.6925115641857356e-08,
+      "loss": 0.7359,
+      "step": 2840
+    },
+    {
+      "epoch": 3.74,
+      "grad_norm": 2.765625,
+      "learning_rate": 3.6477277182032365e-08,
+      "loss": 0.7443,
+      "step": 2841
+    },
+    {
+      "epoch": 3.74,
+      "grad_norm": 2.78125,
+      "learning_rate": 3.603215111565139e-08,
+      "loss": 0.7426,
+      "step": 2842
+    },
+    {
+      "epoch": 3.74,
+      "grad_norm": 2.75,
+      "learning_rate": 3.5589737932807674e-08,
+      "loss": 0.7321,
+      "step": 2843
+    },
+    {
+      "epoch": 3.74,
+      "grad_norm": 2.734375,
+      "learning_rate": 3.515003812060824e-08,
+      "loss": 0.7242,
+      "step": 2844
+    },
+    {
+      "epoch": 3.74,
+      "grad_norm": 2.8125,
+      "learning_rate": 3.47130521631725e-08,
+      "loss": 0.7508,
+      "step": 2845
+    },
+    {
+      "epoch": 3.74,
+      "grad_norm": 2.703125,
+      "learning_rate": 3.427878054163142e-08,
+      "loss": 0.7402,
+      "step": 2846
+    },
+    {
+      "epoch": 3.74,
+      "grad_norm": 2.703125,
+      "learning_rate": 3.384722373412835e-08,
+      "loss": 0.7309,
+      "step": 2847
+    },
+    {
+      "epoch": 3.74,
+      "grad_norm": 2.734375,
+      "learning_rate": 3.341838221581656e-08,
+      "loss": 0.7722,
+      "step": 2848
+    },
+    {
+      "epoch": 3.75,
+      "grad_norm": 2.78125,
+      "learning_rate": 3.2992256458860006e-08,
+      "loss": 0.74,
+      "step": 2849
+    },
+    {
+      "epoch": 3.75,
+      "grad_norm": 2.765625,
+      "learning_rate": 3.2568846932432565e-08,
+      "loss": 0.7351,
+      "step": 2850
+    },
+    {
+      "epoch": 3.75,
+      "grad_norm": 2.671875,
+      "learning_rate": 3.2148154102717985e-08,
+      "loss": 0.7087,
+      "step": 2851
+    },
+    {
+      "epoch": 3.75,
+      "grad_norm": 2.78125,
+      "learning_rate": 3.1730178432908266e-08,
+      "loss": 0.7378,
+      "step": 2852
+    },
+    {
+      "epoch": 3.75,
+      "grad_norm": 2.71875,
+      "learning_rate": 3.1314920383203886e-08,
+      "loss": 0.7254,
+      "step": 2853
+    },
+    {
+      "epoch": 3.75,
+      "grad_norm": 3.0625,
+      "learning_rate": 3.090238041081328e-08,
+      "loss": 0.7478,
+      "step": 2854
+    },
+    {
+      "epoch": 3.75,
+      "grad_norm": 2.765625,
+      "learning_rate": 3.049255896995201e-08,
+      "loss": 0.7333,
+      "step": 2855
+    },
+    {
+      "epoch": 3.76,
+      "grad_norm": 2.890625,
+      "learning_rate": 3.008545651184275e-08,
+      "loss": 0.7478,
+      "step": 2856
+    },
+    {
+      "epoch": 3.76,
+      "grad_norm": 2.71875,
+      "learning_rate": 2.9681073484714438e-08,
+      "loss": 0.7359,
+      "step": 2857
+    },
+    {
+      "epoch": 3.76,
+      "grad_norm": 2.859375,
+      "learning_rate": 2.9279410333801773e-08,
+      "loss": 0.7365,
+      "step": 2858
+    },
+    {
+      "epoch": 3.76,
+      "grad_norm": 2.75,
+      "learning_rate": 2.88804675013446e-08,
+      "loss": 0.7288,
+      "step": 2859
+    },
+    {
+      "epoch": 3.76,
+      "grad_norm": 2.765625,
+      "learning_rate": 2.848424542658823e-08,
+      "loss": 0.7377,
+      "step": 2860
+    },
+    {
+      "epoch": 3.76,
+      "grad_norm": 2.71875,
+      "learning_rate": 2.8090744545782034e-08,
+      "loss": 0.7067,
+      "step": 2861
+    },
+    {
+      "epoch": 3.76,
+      "grad_norm": 2.78125,
+      "learning_rate": 2.769996529217861e-08,
+      "loss": 0.7117,
+      "step": 2862
+    },
+    {
+      "epoch": 3.76,
+      "grad_norm": 2.78125,
+      "learning_rate": 2.7311908096035454e-08,
+      "loss": 0.7122,
+      "step": 2863
+    },
+    {
+      "epoch": 3.77,
+      "grad_norm": 2.6875,
+      "learning_rate": 2.6926573384611353e-08,
+      "loss": 0.7102,
+      "step": 2864
+    },
+    {
+      "epoch": 3.77,
+      "grad_norm": 2.75,
+      "learning_rate": 2.6543961582169153e-08,
+      "loss": 0.7175,
+      "step": 2865
+    },
+    {
+      "epoch": 3.77,
+      "grad_norm": 2.734375,
+      "learning_rate": 2.6164073109972986e-08,
+      "loss": 0.7483,
+      "step": 2866
+    },
+    {
+      "epoch": 3.77,
+      "grad_norm": 2.71875,
+      "learning_rate": 2.5786908386288e-08,
+      "loss": 0.7433,
+      "step": 2867
+    },
+    {
+      "epoch": 3.77,
+      "grad_norm": 2.796875,
+      "learning_rate": 2.5412467826381182e-08,
+      "loss": 0.7559,
+      "step": 2868
+    },
+    {
+      "epoch": 3.77,
+      "grad_norm": 2.765625,
+      "learning_rate": 2.504075184251997e-08,
+      "loss": 0.7435,
+      "step": 2869
+    },
+    {
+      "epoch": 3.77,
+      "grad_norm": 2.78125,
+      "learning_rate": 2.4671760843971992e-08,
+      "loss": 0.7434,
+      "step": 2870
+    },
+    {
+      "epoch": 3.78,
+      "grad_norm": 2.734375,
+      "learning_rate": 2.4305495237004762e-08,
+      "loss": 0.7279,
+      "step": 2871
+    },
+    {
+      "epoch": 3.78,
+      "grad_norm": 2.703125,
+      "learning_rate": 2.3941955424884312e-08,
+      "loss": 0.7393,
+      "step": 2872
+    },
+    {
+      "epoch": 3.78,
+      "grad_norm": 2.765625,
+      "learning_rate": 2.3581141807876573e-08,
+      "loss": 0.7192,
+      "step": 2873
+    },
+    {
+      "epoch": 3.78,
+      "grad_norm": 2.8125,
+      "learning_rate": 2.322305478324516e-08,
+      "loss": 0.731,
+      "step": 2874
+    },
+    {
+      "epoch": 3.78,
+      "grad_norm": 2.828125,
+      "learning_rate": 2.2867694745251634e-08,
+      "loss": 0.7385,
+      "step": 2875
+    },
+    {
+      "epoch": 3.78,
+      "grad_norm": 2.765625,
+      "learning_rate": 2.251506208515608e-08,
+      "loss": 0.7166,
+      "step": 2876
+    },
+    {
+      "epoch": 3.78,
+      "grad_norm": 2.8125,
+      "learning_rate": 2.2165157191214314e-08,
+      "loss": 0.7436,
+      "step": 2877
+    },
+    {
+      "epoch": 3.78,
+      "grad_norm": 2.796875,
+      "learning_rate": 2.1817980448679553e-08,
+      "loss": 0.728,
+      "step": 2878
+    },
+    {
+      "epoch": 3.79,
+      "grad_norm": 2.78125,
+      "learning_rate": 2.147353223980103e-08,
+      "loss": 0.7318,
+      "step": 2879
+    },
+    {
+      "epoch": 3.79,
+      "grad_norm": 2.71875,
+      "learning_rate": 2.1131812943823994e-08,
+      "loss": 0.7152,
+      "step": 2880
+    },
+    {
+      "epoch": 3.79,
+      "grad_norm": 2.796875,
+      "learning_rate": 2.0792822936989142e-08,
+      "loss": 0.7667,
+      "step": 2881
+    },
+    {
+      "epoch": 3.79,
+      "grad_norm": 2.78125,
+      "learning_rate": 2.0456562592532093e-08,
+      "loss": 0.7233,
+      "step": 2882
+    },
+    {
+      "epoch": 3.79,
+      "grad_norm": 2.796875,
+      "learning_rate": 2.0123032280682242e-08,
+      "loss": 0.7389,
+      "step": 2883
+    },
+    {
+      "epoch": 3.79,
+      "grad_norm": 2.765625,
+      "learning_rate": 1.979223236866501e-08,
+      "loss": 0.7327,
+      "step": 2884
+    },
+    {
+      "epoch": 3.79,
+      "grad_norm": 2.65625,
+      "learning_rate": 1.946416322069794e-08,
+      "loss": 0.733,
+      "step": 2885
+    },
+    {
+      "epoch": 3.8,
+      "grad_norm": 2.78125,
+      "learning_rate": 1.9138825197992096e-08,
+      "loss": 0.7327,
+      "step": 2886
+    },
+    {
+      "epoch": 3.8,
+      "grad_norm": 2.734375,
+      "learning_rate": 1.8816218658752605e-08,
+      "loss": 0.7271,
+      "step": 2887
+    },
+    {
+      "epoch": 3.8,
+      "grad_norm": 2.765625,
+      "learning_rate": 1.8496343958175898e-08,
+      "loss": 0.7392,
+      "step": 2888
+    },
+    {
+      "epoch": 3.8,
+      "grad_norm": 2.796875,
+      "learning_rate": 1.8179201448451368e-08,
+      "loss": 0.769,
+      "step": 2889
+    },
+    {
+      "epoch": 3.8,
+      "grad_norm": 2.765625,
+      "learning_rate": 1.7864791478760245e-08,
+      "loss": 0.7105,
+      "step": 2890
+    },
+    {
+      "epoch": 3.8,
+      "grad_norm": 2.78125,
+      "learning_rate": 1.755311439527424e-08,
+      "loss": 0.737,
+      "step": 2891
+    },
+    {
+      "epoch": 3.8,
+      "grad_norm": 2.8125,
+      "learning_rate": 1.7244170541157455e-08,
+      "loss": 0.749,
+      "step": 2892
+    },
+    {
+      "epoch": 3.8,
+      "grad_norm": 2.703125,
+      "learning_rate": 1.6937960256563914e-08,
+      "loss": 0.7221,
+      "step": 2893
+    },
+    {
+      "epoch": 3.81,
+      "grad_norm": 2.828125,
+      "learning_rate": 1.663448387863753e-08,
+      "loss": 0.7467,
+      "step": 2894
+    },
+    {
+      "epoch": 3.81,
+      "grad_norm": 2.796875,
+      "learning_rate": 1.6333741741512977e-08,
+      "loss": 0.7574,
+      "step": 2895
+    },
+    {
+      "epoch": 3.81,
+      "grad_norm": 2.765625,
+      "learning_rate": 1.603573417631371e-08,
+      "loss": 0.7659,
+      "step": 2896
+    },
+    {
+      "epoch": 3.81,
+      "grad_norm": 2.765625,
+      "learning_rate": 1.5740461511152827e-08,
+      "loss": 0.7434,
+      "step": 2897
+    },
+    {
+      "epoch": 3.81,
+      "grad_norm": 2.875,
+      "learning_rate": 1.5447924071131937e-08,
+      "loss": 0.7636,
+      "step": 2898
+    },
+    {
+      "epoch": 3.81,
+      "grad_norm": 2.8125,
+      "learning_rate": 1.5158122178341727e-08,
+      "loss": 0.7282,
+      "step": 2899
+    },
+    {
+      "epoch": 3.81,
+      "grad_norm": 2.765625,
+      "learning_rate": 1.4871056151860296e-08,
+      "loss": 0.7234,
+      "step": 2900
+    },
+    {
+      "epoch": 3.82,
+      "grad_norm": 2.796875,
+      "learning_rate": 1.4586726307753985e-08,
+      "loss": 0.7264,
+      "step": 2901
+    },
+    {
+      "epoch": 3.82,
+      "grad_norm": 2.859375,
+      "learning_rate": 1.4305132959075706e-08,
+      "loss": 0.7561,
+      "step": 2902
+    },
+    {
+      "epoch": 3.82,
+      "grad_norm": 2.796875,
+      "learning_rate": 1.4026276415866624e-08,
+      "loss": 0.7512,
+      "step": 2903
+    },
+    {
+      "epoch": 3.82,
+      "grad_norm": 2.78125,
+      "learning_rate": 1.3750156985153361e-08,
+      "loss": 0.7299,
+      "step": 2904
+    },
+    {
+      "epoch": 3.82,
+      "grad_norm": 2.78125,
+      "learning_rate": 1.3476774970950236e-08,
+      "loss": 0.7348,
+      "step": 2905
+    },
+    {
+      "epoch": 3.82,
+      "grad_norm": 2.734375,
+      "learning_rate": 1.3206130674256746e-08,
+      "loss": 0.7432,
+      "step": 2906
+    },
+    {
+      "epoch": 3.82,
+      "grad_norm": 2.84375,
+      "learning_rate": 1.293822439305814e-08,
+      "loss": 0.7636,
+      "step": 2907
+    },
+    {
+      "epoch": 3.82,
+      "grad_norm": 2.765625,
+      "learning_rate": 1.2673056422325413e-08,
+      "loss": 0.7549,
+      "step": 2908
+    },
+    {
+      "epoch": 3.83,
+      "grad_norm": 2.828125,
+      "learning_rate": 1.2410627054013913e-08,
+      "loss": 0.772,
+      "step": 2909
+    },
+    {
+      "epoch": 3.83,
+      "grad_norm": 2.796875,
+      "learning_rate": 1.2150936577065008e-08,
+      "loss": 0.766,
+      "step": 2910
+    },
+    {
+      "epoch": 3.83,
+      "grad_norm": 2.703125,
+      "learning_rate": 1.1893985277403041e-08,
+      "loss": 0.704,
+      "step": 2911
+    },
+    {
+      "epoch": 3.83,
+      "grad_norm": 2.84375,
+      "learning_rate": 1.1639773437937263e-08,
+      "loss": 0.7266,
+      "step": 2912
+    },
+    {
+      "epoch": 3.83,
+      "grad_norm": 2.796875,
+      "learning_rate": 1.1388301338561004e-08,
+      "loss": 0.7554,
+      "step": 2913
+    },
+    {
+      "epoch": 3.83,
+      "grad_norm": 2.703125,
+      "learning_rate": 1.1139569256150285e-08,
+      "loss": 0.7383,
+      "step": 2914
+    },
+    {
+      "epoch": 3.83,
+      "grad_norm": 2.796875,
+      "learning_rate": 1.0893577464564654e-08,
+      "loss": 0.7462,
+      "step": 2915
+    },
+    {
+      "epoch": 3.84,
+      "grad_norm": 2.796875,
+      "learning_rate": 1.0650326234646347e-08,
+      "loss": 0.7399,
+      "step": 2916
+    },
+    {
+      "epoch": 3.84,
+      "grad_norm": 2.84375,
+      "learning_rate": 1.0409815834221126e-08,
+      "loss": 0.7402,
+      "step": 2917
+    },
+    {
+      "epoch": 3.84,
+      "grad_norm": 2.75,
+      "learning_rate": 1.0172046528095502e-08,
+      "loss": 0.7301,
+      "step": 2918
+    },
+    {
+      "epoch": 3.84,
+      "grad_norm": 2.671875,
+      "learning_rate": 9.937018578059232e-09,
+      "loss": 0.6891,
+      "step": 2919
+    },
+    {
+      "epoch": 3.84,
+      "grad_norm": 2.71875,
+      "learning_rate": 9.704732242883374e-09,
+      "loss": 0.7348,
+      "step": 2920
+    },
+    {
+      "epoch": 3.84,
+      "grad_norm": 2.75,
+      "learning_rate": 9.475187778320017e-09,
+      "loss": 0.7366,
+      "step": 2921
+    },
+    {
+      "epoch": 3.84,
+      "grad_norm": 2.796875,
+      "learning_rate": 9.248385437103102e-09,
+      "loss": 0.7531,
+      "step": 2922
+    },
+    {
+      "epoch": 3.84,
+      "grad_norm": 2.8125,
+      "learning_rate": 9.024325468946771e-09,
+      "loss": 0.7097,
+      "step": 2923
+    },
+    {
+      "epoch": 3.85,
+      "grad_norm": 2.8125,
+      "learning_rate": 8.803008120546186e-09,
+      "loss": 0.7508,
+      "step": 2924
+    },
+    {
+      "epoch": 3.85,
+      "grad_norm": 2.921875,
+      "learning_rate": 8.584433635576706e-09,
+      "loss": 0.7541,
+      "step": 2925
+    },
+    {
+      "epoch": 3.85,
+      "grad_norm": 2.75,
+      "learning_rate": 8.368602254693603e-09,
+      "loss": 0.7324,
+      "step": 2926
+    },
+    {
+      "epoch": 3.85,
+      "grad_norm": 2.6875,
+      "learning_rate": 8.155514215532067e-09,
+      "loss": 0.7129,
+      "step": 2927
+    },
+    {
+      "epoch": 3.85,
+      "grad_norm": 2.734375,
+      "learning_rate": 7.945169752706927e-09,
+      "loss": 0.7325,
+      "step": 2928
+    },
+    {
+      "epoch": 3.85,
+      "grad_norm": 2.765625,
+      "learning_rate": 7.737569097811814e-09,
+      "loss": 0.7265,
+      "step": 2929
+    },
+    {
+      "epoch": 3.85,
+      "grad_norm": 2.703125,
+      "learning_rate": 7.532712479420279e-09,
+      "loss": 0.7415,
+      "step": 2930
+    },
+    {
+      "epoch": 3.86,
+      "grad_norm": 2.78125,
+      "learning_rate": 7.330600123083287e-09,
+      "loss": 0.7386,
+      "step": 2931
+    },
+    {
+      "epoch": 3.86,
+      "grad_norm": 2.875,
+      "learning_rate": 7.131232251331721e-09,
+      "loss": 0.7384,
+      "step": 2932
+    },
+    {
+      "epoch": 3.86,
+      "grad_norm": 2.75,
+      "learning_rate": 6.9346090836733275e-09,
+      "loss": 0.7664,
+      "step": 2933
+    },
+    {
+      "epoch": 3.86,
+      "grad_norm": 2.796875,
+      "learning_rate": 6.740730836595211e-09,
+      "loss": 0.7617,
+      "step": 2934
+    },
+    {
+      "epoch": 3.86,
+      "grad_norm": 2.828125,
+      "learning_rate": 6.5495977235613405e-09,
+      "loss": 0.7337,
+      "step": 2935
+    },
+    {
+      "epoch": 3.86,
+      "grad_norm": 2.765625,
+      "learning_rate": 6.361209955013659e-09,
+      "loss": 0.7311,
+      "step": 2936
+    },
+    {
+      "epoch": 3.86,
+      "grad_norm": 2.734375,
+      "learning_rate": 6.1755677383712465e-09,
+      "loss": 0.735,
+      "step": 2937
+    },
+    {
+      "epoch": 3.86,
+      "grad_norm": 2.78125,
+      "learning_rate": 5.992671278030327e-09,
+      "loss": 0.7274,
+      "step": 2938
+    },
+    {
+      "epoch": 3.87,
+      "grad_norm": 2.765625,
+      "learning_rate": 5.812520775363984e-09,
+      "loss": 0.7217,
+      "step": 2939
+    },
+    {
+      "epoch": 3.87,
+      "grad_norm": 2.828125,
+      "learning_rate": 5.635116428722165e-09,
+      "loss": 0.7365,
+      "step": 2940
+    },
+    {
+      "epoch": 3.87,
+      "grad_norm": 2.859375,
+      "learning_rate": 5.460458433430571e-09,
+      "loss": 0.7187,
+      "step": 2941
+    },
+    {
+      "epoch": 3.87,
+      "grad_norm": 2.78125,
+      "learning_rate": 5.288546981791764e-09,
+      "loss": 0.7244,
+      "step": 2942
+    },
+    {
+      "epoch": 3.87,
+      "grad_norm": 2.78125,
+      "learning_rate": 5.1193822630840605e-09,
+      "loss": 0.7478,
+      "step": 2943
+    },
+    {
+      "epoch": 3.87,
+      "grad_norm": 2.78125,
+      "learning_rate": 4.952964463561805e-09,
+      "loss": 0.763,
+      "step": 2944
+    },
+    {
+      "epoch": 3.87,
+      "grad_norm": 2.671875,
+      "learning_rate": 4.789293766454262e-09,
+      "loss": 0.7294,
+      "step": 2945
+    },
+    {
+      "epoch": 3.88,
+      "grad_norm": 2.78125,
+      "learning_rate": 4.628370351967004e-09,
+      "loss": 0.7239,
+      "step": 2946
+    },
+    {
+      "epoch": 3.88,
+      "grad_norm": 2.703125,
+      "learning_rate": 4.47019439727997e-09,
+      "loss": 0.7102,
+      "step": 2947
+    },
+    {
+      "epoch": 3.88,
+      "grad_norm": 2.796875,
+      "learning_rate": 4.314766076548571e-09,
+      "loss": 0.719,
+      "step": 2948
+    },
+    {
+      "epoch": 3.88,
+      "grad_norm": 2.875,
+      "learning_rate": 4.162085560902585e-09,
+      "loss": 0.7664,
+      "step": 2949
+    },
+    {
+      "epoch": 3.88,
+      "grad_norm": 2.75,
+      "learning_rate": 4.012153018446984e-09,
+      "loss": 0.7484,
+      "step": 2950
+    },
+    {
+      "epoch": 3.88,
+      "grad_norm": 2.921875,
+      "learning_rate": 3.864968614260834e-09,
+      "loss": 0.7535,
+      "step": 2951
+    },
+    {
+      "epoch": 3.88,
+      "grad_norm": 2.75,
+      "learning_rate": 3.720532510397279e-09,
+      "loss": 0.7329,
+      "step": 2952
+    },
+    {
+      "epoch": 3.88,
+      "grad_norm": 2.765625,
+      "learning_rate": 3.5788448658838347e-09,
+      "loss": 0.7396,
+      "step": 2953
+    },
+    {
+      "epoch": 3.89,
+      "grad_norm": 2.75,
+      "learning_rate": 3.4399058367215466e-09,
+      "loss": 0.7397,
+      "step": 2954
+    },
+    {
+      "epoch": 3.89,
+      "grad_norm": 2.796875,
+      "learning_rate": 3.3037155758855466e-09,
+      "loss": 0.7373,
+      "step": 2955
+    },
+    {
+      "epoch": 3.89,
+      "grad_norm": 2.765625,
+      "learning_rate": 3.170274233324222e-09,
+      "loss": 0.7425,
+      "step": 2956
+    },
+    {
+      "epoch": 3.89,
+      "grad_norm": 2.703125,
+      "learning_rate": 3.039581955959492e-09,
+      "loss": 0.7203,
+      "step": 2957
+    },
+    {
+      "epoch": 3.89,
+      "grad_norm": 2.734375,
+      "learning_rate": 2.9116388876868075e-09,
+      "loss": 0.7311,
+      "step": 2958
+    },
+    {
+      "epoch": 3.89,
+      "grad_norm": 2.828125,
+      "learning_rate": 2.7864451693740412e-09,
+      "loss": 0.7643,
+      "step": 2959
+    },
+    {
+      "epoch": 3.89,
+      "grad_norm": 2.75,
+      "learning_rate": 2.6640009388620434e-09,
+      "loss": 0.753,
+      "step": 2960
+    },
+    {
+      "epoch": 3.9,
+      "grad_norm": 2.78125,
+      "learning_rate": 2.544306330965196e-09,
+      "loss": 0.7482,
+      "step": 2961
+    },
+    {
+      "epoch": 3.9,
+      "grad_norm": 2.75,
+      "learning_rate": 2.4273614774691923e-09,
+      "loss": 0.7422,
+      "step": 2962
+    },
+    {
+      "epoch": 3.9,
+      "grad_norm": 2.734375,
+      "learning_rate": 2.3131665071338127e-09,
+      "loss": 0.7334,
+      "step": 2963
+    },
+    {
+      "epoch": 3.9,
+      "grad_norm": 2.71875,
+      "learning_rate": 2.2017215456895947e-09,
+      "loss": 0.7389,
+      "step": 2964
+    },
+    {
+      "epoch": 3.9,
+      "grad_norm": 2.734375,
+      "learning_rate": 2.093026715840052e-09,
+      "loss": 0.7329,
+      "step": 2965
+    },
+    {
+      "epoch": 3.9,
+      "grad_norm": 2.71875,
+      "learning_rate": 1.987082137260843e-09,
+      "loss": 0.7374,
+      "step": 2966
+    },
+    {
+      "epoch": 3.9,
+      "grad_norm": 2.78125,
+      "learning_rate": 1.883887926598937e-09,
+      "loss": 0.7191,
+      "step": 2967
+    },
+    {
+      "epoch": 3.9,
+      "grad_norm": 2.859375,
+      "learning_rate": 1.7834441974740047e-09,
+      "loss": 0.7467,
+      "step": 2968
+    },
+    {
+      "epoch": 3.91,
+      "grad_norm": 2.703125,
+      "learning_rate": 1.6857510604764704e-09,
+      "loss": 0.7067,
+      "step": 2969
+    },
+    {
+      "epoch": 3.91,
+      "grad_norm": 2.828125,
+      "learning_rate": 1.590808623168627e-09,
+      "loss": 0.7633,
+      "step": 2970
+    },
+    {
+      "epoch": 3.91,
+      "grad_norm": 2.8125,
+      "learning_rate": 1.4986169900840785e-09,
+      "loss": 0.7312,
+      "step": 2971
+    },
+    {
+      "epoch": 3.91,
+      "grad_norm": 2.6875,
+      "learning_rate": 1.4091762627282956e-09,
+      "loss": 0.7373,
+      "step": 2972
+    },
+    {
+      "epoch": 3.91,
+      "grad_norm": 2.828125,
+      "learning_rate": 1.3224865395775054e-09,
+      "loss": 0.7671,
+      "step": 2973
+    },
+    {
+      "epoch": 3.91,
+      "grad_norm": 2.765625,
+      "learning_rate": 1.2385479160784141e-09,
+      "loss": 0.7335,
+      "step": 2974
+    },
+    {
+      "epoch": 3.91,
+      "grad_norm": 2.765625,
+      "learning_rate": 1.1573604846501496e-09,
+      "loss": 0.7544,
+      "step": 2975
+    },
+    {
+      "epoch": 3.92,
+      "grad_norm": 2.828125,
+      "learning_rate": 1.0789243346812083e-09,
+      "loss": 0.7404,
+      "step": 2976
+    },
+    {
+      "epoch": 3.92,
+      "grad_norm": 2.78125,
+      "learning_rate": 1.0032395525316762e-09,
+      "loss": 0.7736,
+      "step": 2977
+    },
+    {
+      "epoch": 3.92,
+      "grad_norm": 2.796875,
+      "learning_rate": 9.303062215323955e-10,
+      "loss": 0.7172,
+      "step": 2978
+    },
+    {
+      "epoch": 3.92,
+      "grad_norm": 2.921875,
+      "learning_rate": 8.601244219846872e-10,
+      "loss": 0.746,
+      "step": 2979
+    },
+    {
+      "epoch": 3.92,
+      "grad_norm": 2.859375,
+      "learning_rate": 7.926942311597962e-10,
+      "loss": 0.7433,
+      "step": 2980
+    },
+    {
+      "epoch": 3.92,
+      "grad_norm": 2.796875,
+      "learning_rate": 7.280157233002794e-10,
+      "loss": 0.7349,
+      "step": 2981
+    },
+    {
+      "epoch": 3.92,
+      "grad_norm": 2.703125,
+      "learning_rate": 6.660889696186168e-10,
+      "loss": 0.742,
+      "step": 2982
+    },
+    {
+      "epoch": 3.92,
+      "grad_norm": 2.8125,
+      "learning_rate": 6.069140382972128e-10,
+      "loss": 0.7588,
+      "step": 2983
+    },
+    {
+      "epoch": 3.93,
+      "grad_norm": 2.875,
+      "learning_rate": 5.504909944895054e-10,
+      "loss": 0.7558,
+      "step": 2984
+    },
+    {
+      "epoch": 3.93,
+      "grad_norm": 2.796875,
+      "learning_rate": 4.968199003177466e-10,
+      "loss": 0.7512,
+      "step": 2985
+    },
+    {
+      "epoch": 3.93,
+      "grad_norm": 2.796875,
+      "learning_rate": 4.4590081487577706e-10,
+      "loss": 0.7457,
+      "step": 2986
+    },
+    {
+      "epoch": 3.93,
+      "grad_norm": 2.890625,
+      "learning_rate": 3.9773379422625156e-10,
+      "loss": 0.7603,
+      "step": 2987
+    },
+    {
+      "epoch": 3.93,
+      "grad_norm": 2.828125,
+      "learning_rate": 3.5231889140230347e-10,
+      "loss": 0.7217,
+      "step": 2988
+    },
+    {
+      "epoch": 3.93,
+      "grad_norm": 2.734375,
+      "learning_rate": 3.0965615640671243e-10,
+      "loss": 0.7448,
+      "step": 2989
+    },
+    {
+      "epoch": 3.93,
+      "grad_norm": 2.71875,
+      "learning_rate": 2.697456362119044e-10,
+      "loss": 0.7266,
+      "step": 2990
+    },
+    {
+      "epoch": 3.94,
+      "grad_norm": 2.75,
+      "learning_rate": 2.3258737476050674e-10,
+      "loss": 0.7257,
+      "step": 2991
+    },
+    {
+      "epoch": 3.94,
+      "grad_norm": 2.78125,
+      "learning_rate": 1.9818141296451544e-10,
+      "loss": 0.7411,
+      "step": 2992
+    },
+    {
+      "epoch": 3.94,
+      "grad_norm": 2.84375,
+      "learning_rate": 1.6652778870557274e-10,
+      "loss": 0.7428,
+      "step": 2993
+    },
+    {
+      "epoch": 3.94,
+      "grad_norm": 2.796875,
+      "learning_rate": 1.3762653683496718e-10,
+      "loss": 0.7325,
+      "step": 2994
+    },
+    {
+      "epoch": 3.94,
+      "grad_norm": 2.796875,
+      "learning_rate": 1.1147768917391111e-10,
+      "loss": 0.7328,
+      "step": 2995
+    },
+    {
+      "epoch": 3.94,
+      "grad_norm": 2.765625,
+      "learning_rate": 8.808127451270798e-11,
+      "loss": 0.7472,
+      "step": 2996
+    },
+    {
+      "epoch": 3.94,
+      "grad_norm": 2.765625,
+      "learning_rate": 6.743731861130753e-11,
+      "loss": 0.7252,
+      "step": 2997
+    },
+    {
+      "epoch": 3.94,
+      "grad_norm": 2.6875,
+      "learning_rate": 4.954584419930575e-11,
+      "loss": 0.7301,
+      "step": 2998
+    },
+    {
+      "epoch": 3.95,
+      "grad_norm": 2.71875,
+      "learning_rate": 3.440687097538975e-11,
+      "loss": 0.7206,
+      "step": 2999
+    },
+    {
+      "epoch": 3.95,
+      "grad_norm": 2.734375,
+      "learning_rate": 2.202041560789292e-11,
+      "loss": 0.7362,
+      "step": 3000
+    },
+    {
+      "epoch": 3.95,
+      "grad_norm": 2.828125,
+      "learning_rate": 1.2386491734794891e-11,
+      "loss": 0.7393,
+      "step": 3001
+    },
+    {
+      "epoch": 3.95,
+      "grad_norm": 2.796875,
+      "learning_rate": 5.505109963166444e-12,
+      "loss": 0.7456,
+      "step": 3002
+    },
+    {
+      "epoch": 3.95,
+      "grad_norm": 2.875,
+      "learning_rate": 1.376277869724607e-12,
+      "loss": 0.7549,
+      "step": 3003
+    },
+    {
+      "epoch": 3.95,
+      "grad_norm": 2.65625,
+      "learning_rate": 0.0,
+      "loss": 0.7152,
+      "step": 3004
     }
   ],
   "logging_steps": 1,
@@ -15785,7 +21042,7 @@
   "num_input_tokens_seen": 0,
   "num_train_epochs": 4,
   "save_steps": 751,
-  "total_flos": 2.519778018662233e+19,
+  "total_flos": 3.359704024882977e+19,
   "train_batch_size": 2,
   "trial_name": null,
   "trial_params": null