| { | |
| "best_metric": NaN, | |
| "best_model_checkpoint": "/scratch/czm5kz/finetuned_pythia70M_deduped_cp_14300032_1_0.0003_sequential/checkpoint-20", | |
| "epoch": 0.9975062344139651, | |
| "eval_steps": 20, | |
| "global_step": 1400, | |
| "is_hyper_param_search": false, | |
| "is_local_process_zero": true, | |
| "is_world_process_zero": true, | |
| "log_history": [ | |
| { | |
| "epoch": 0.0, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 5 | |
| }, | |
| { | |
| "epoch": 0.01, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 10 | |
| }, | |
| { | |
| "epoch": 0.01, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 15 | |
| }, | |
| { | |
| "epoch": 0.01, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 20 | |
| }, | |
| { | |
| "epoch": 0.01, | |
| "eval_loss": NaN, | |
| "eval_runtime": 15.946, | |
| "eval_samples_per_second": 704.191, | |
| "eval_steps_per_second": 88.047, | |
| "step": 20 | |
| }, | |
| { | |
| "epoch": 0.02, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 25 | |
| }, | |
| { | |
| "epoch": 0.02, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 30 | |
| }, | |
| { | |
| "epoch": 0.02, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 35 | |
| }, | |
| { | |
| "epoch": 0.03, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 40 | |
| }, | |
| { | |
| "epoch": 0.03, | |
| "eval_loss": NaN, | |
| "eval_runtime": 16.1366, | |
| "eval_samples_per_second": 695.874, | |
| "eval_steps_per_second": 87.007, | |
| "step": 40 | |
| }, | |
| { | |
| "epoch": 0.03, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 45 | |
| }, | |
| { | |
| "epoch": 0.04, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 50 | |
| }, | |
| { | |
| "epoch": 0.04, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 55 | |
| }, | |
| { | |
| "epoch": 0.04, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 60 | |
| }, | |
| { | |
| "epoch": 0.04, | |
| "eval_loss": NaN, | |
| "eval_runtime": 16.118, | |
| "eval_samples_per_second": 696.672, | |
| "eval_steps_per_second": 87.107, | |
| "step": 60 | |
| }, | |
| { | |
| "epoch": 0.05, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 65 | |
| }, | |
| { | |
| "epoch": 0.05, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 70 | |
| }, | |
| { | |
| "epoch": 0.05, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 75 | |
| }, | |
| { | |
| "epoch": 0.06, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 80 | |
| }, | |
| { | |
| "epoch": 0.06, | |
| "eval_loss": NaN, | |
| "eval_runtime": 15.8962, | |
| "eval_samples_per_second": 706.396, | |
| "eval_steps_per_second": 88.323, | |
| "step": 80 | |
| }, | |
| { | |
| "epoch": 0.06, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 85 | |
| }, | |
| { | |
| "epoch": 0.06, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 90 | |
| }, | |
| { | |
| "epoch": 0.07, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 95 | |
| }, | |
| { | |
| "epoch": 0.07, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 1.8286, | |
| "step": 100 | |
| }, | |
| { | |
| "epoch": 0.07, | |
| "eval_loss": NaN, | |
| "eval_runtime": 16.0791, | |
| "eval_samples_per_second": 698.359, | |
| "eval_steps_per_second": 87.318, | |
| "step": 100 | |
| }, | |
| { | |
| "epoch": 0.07, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 105 | |
| }, | |
| { | |
| "epoch": 0.08, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 110 | |
| }, | |
| { | |
| "epoch": 0.08, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 115 | |
| }, | |
| { | |
| "epoch": 0.09, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 120 | |
| }, | |
| { | |
| "epoch": 0.09, | |
| "eval_loss": NaN, | |
| "eval_runtime": 15.908, | |
| "eval_samples_per_second": 705.869, | |
| "eval_steps_per_second": 88.257, | |
| "step": 120 | |
| }, | |
| { | |
| "epoch": 0.09, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 125 | |
| }, | |
| { | |
| "epoch": 0.09, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 130 | |
| }, | |
| { | |
| "epoch": 0.1, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 135 | |
| }, | |
| { | |
| "epoch": 0.1, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 140 | |
| }, | |
| { | |
| "epoch": 0.1, | |
| "eval_loss": NaN, | |
| "eval_runtime": 15.8884, | |
| "eval_samples_per_second": 706.744, | |
| "eval_steps_per_second": 88.367, | |
| "step": 140 | |
| }, | |
| { | |
| "epoch": 0.1, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 145 | |
| }, | |
| { | |
| "epoch": 0.11, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 150 | |
| }, | |
| { | |
| "epoch": 0.11, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 155 | |
| }, | |
| { | |
| "epoch": 0.11, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 160 | |
| }, | |
| { | |
| "epoch": 0.11, | |
| "eval_loss": NaN, | |
| "eval_runtime": 16.0584, | |
| "eval_samples_per_second": 699.26, | |
| "eval_steps_per_second": 87.431, | |
| "step": 160 | |
| }, | |
| { | |
| "epoch": 0.12, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 165 | |
| }, | |
| { | |
| "epoch": 0.12, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 170 | |
| }, | |
| { | |
| "epoch": 0.12, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 175 | |
| }, | |
| { | |
| "epoch": 0.13, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 180 | |
| }, | |
| { | |
| "epoch": 0.13, | |
| "eval_loss": NaN, | |
| "eval_runtime": 15.8804, | |
| "eval_samples_per_second": 707.098, | |
| "eval_steps_per_second": 88.411, | |
| "step": 180 | |
| }, | |
| { | |
| "epoch": 0.13, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 185 | |
| }, | |
| { | |
| "epoch": 0.14, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 190 | |
| }, | |
| { | |
| "epoch": 0.14, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 195 | |
| }, | |
| { | |
| "epoch": 0.14, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 200 | |
| }, | |
| { | |
| "epoch": 0.14, | |
| "eval_loss": NaN, | |
| "eval_runtime": 15.8809, | |
| "eval_samples_per_second": 707.076, | |
| "eval_steps_per_second": 88.408, | |
| "step": 200 | |
| }, | |
| { | |
| "epoch": 0.15, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 205 | |
| }, | |
| { | |
| "epoch": 0.15, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 210 | |
| }, | |
| { | |
| "epoch": 0.15, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 215 | |
| }, | |
| { | |
| "epoch": 0.16, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 220 | |
| }, | |
| { | |
| "epoch": 0.16, | |
| "eval_loss": NaN, | |
| "eval_runtime": 16.0622, | |
| "eval_samples_per_second": 699.093, | |
| "eval_steps_per_second": 87.41, | |
| "step": 220 | |
| }, | |
| { | |
| "epoch": 0.16, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 225 | |
| }, | |
| { | |
| "epoch": 0.16, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 230 | |
| }, | |
| { | |
| "epoch": 0.17, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 235 | |
| }, | |
| { | |
| "epoch": 0.17, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 240 | |
| }, | |
| { | |
| "epoch": 0.17, | |
| "eval_loss": NaN, | |
| "eval_runtime": 15.8802, | |
| "eval_samples_per_second": 707.107, | |
| "eval_steps_per_second": 88.412, | |
| "step": 240 | |
| }, | |
| { | |
| "epoch": 0.17, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 245 | |
| }, | |
| { | |
| "epoch": 0.18, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 250 | |
| }, | |
| { | |
| "epoch": 0.18, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 255 | |
| }, | |
| { | |
| "epoch": 0.19, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 260 | |
| }, | |
| { | |
| "epoch": 0.19, | |
| "eval_loss": NaN, | |
| "eval_runtime": 15.8916, | |
| "eval_samples_per_second": 706.598, | |
| "eval_steps_per_second": 88.348, | |
| "step": 260 | |
| }, | |
| { | |
| "epoch": 0.19, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 265 | |
| }, | |
| { | |
| "epoch": 0.19, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 270 | |
| }, | |
| { | |
| "epoch": 0.2, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 275 | |
| }, | |
| { | |
| "epoch": 0.2, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 280 | |
| }, | |
| { | |
| "epoch": 0.2, | |
| "eval_loss": NaN, | |
| "eval_runtime": 16.0692, | |
| "eval_samples_per_second": 698.79, | |
| "eval_steps_per_second": 87.372, | |
| "step": 280 | |
| }, | |
| { | |
| "epoch": 0.2, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 285 | |
| }, | |
| { | |
| "epoch": 0.21, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 290 | |
| }, | |
| { | |
| "epoch": 0.21, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 295 | |
| }, | |
| { | |
| "epoch": 0.21, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 300 | |
| }, | |
| { | |
| "epoch": 0.21, | |
| "eval_loss": NaN, | |
| "eval_runtime": 15.9044, | |
| "eval_samples_per_second": 706.031, | |
| "eval_steps_per_second": 88.277, | |
| "step": 300 | |
| }, | |
| { | |
| "epoch": 0.22, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 305 | |
| }, | |
| { | |
| "epoch": 0.22, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 310 | |
| }, | |
| { | |
| "epoch": 0.22, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 315 | |
| }, | |
| { | |
| "epoch": 0.23, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 320 | |
| }, | |
| { | |
| "epoch": 0.23, | |
| "eval_loss": NaN, | |
| "eval_runtime": 15.8728, | |
| "eval_samples_per_second": 707.436, | |
| "eval_steps_per_second": 88.453, | |
| "step": 320 | |
| }, | |
| { | |
| "epoch": 0.23, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 325 | |
| }, | |
| { | |
| "epoch": 0.24, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 330 | |
| }, | |
| { | |
| "epoch": 0.24, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 335 | |
| }, | |
| { | |
| "epoch": 0.24, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 340 | |
| }, | |
| { | |
| "epoch": 0.24, | |
| "eval_loss": NaN, | |
| "eval_runtime": 16.1019, | |
| "eval_samples_per_second": 697.37, | |
| "eval_steps_per_second": 87.195, | |
| "step": 340 | |
| }, | |
| { | |
| "epoch": 0.25, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 345 | |
| }, | |
| { | |
| "epoch": 0.25, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 350 | |
| }, | |
| { | |
| "epoch": 0.25, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 355 | |
| }, | |
| { | |
| "epoch": 0.26, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 360 | |
| }, | |
| { | |
| "epoch": 0.26, | |
| "eval_loss": NaN, | |
| "eval_runtime": 15.8632, | |
| "eval_samples_per_second": 707.863, | |
| "eval_steps_per_second": 88.506, | |
| "step": 360 | |
| }, | |
| { | |
| "epoch": 0.26, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 365 | |
| }, | |
| { | |
| "epoch": 0.26, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 3.7732, | |
| "step": 370 | |
| }, | |
| { | |
| "epoch": 0.27, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 375 | |
| }, | |
| { | |
| "epoch": 0.27, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 380 | |
| }, | |
| { | |
| "epoch": 0.27, | |
| "eval_loss": NaN, | |
| "eval_runtime": 16.0298, | |
| "eval_samples_per_second": 700.506, | |
| "eval_steps_per_second": 87.587, | |
| "step": 380 | |
| }, | |
| { | |
| "epoch": 0.27, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 385 | |
| }, | |
| { | |
| "epoch": 0.28, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 390 | |
| }, | |
| { | |
| "epoch": 0.28, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 395 | |
| }, | |
| { | |
| "epoch": 0.29, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 400 | |
| }, | |
| { | |
| "epoch": 0.29, | |
| "eval_loss": NaN, | |
| "eval_runtime": 16.0246, | |
| "eval_samples_per_second": 700.736, | |
| "eval_steps_per_second": 87.615, | |
| "step": 400 | |
| }, | |
| { | |
| "epoch": 0.29, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 405 | |
| }, | |
| { | |
| "epoch": 0.29, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 410 | |
| }, | |
| { | |
| "epoch": 0.3, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 415 | |
| }, | |
| { | |
| "epoch": 0.3, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 420 | |
| }, | |
| { | |
| "epoch": 0.3, | |
| "eval_loss": NaN, | |
| "eval_runtime": 15.8556, | |
| "eval_samples_per_second": 708.204, | |
| "eval_steps_per_second": 88.549, | |
| "step": 420 | |
| }, | |
| { | |
| "epoch": 0.3, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 425 | |
| }, | |
| { | |
| "epoch": 0.31, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 430 | |
| }, | |
| { | |
| "epoch": 0.31, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 435 | |
| }, | |
| { | |
| "epoch": 0.31, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 440 | |
| }, | |
| { | |
| "epoch": 0.31, | |
| "eval_loss": NaN, | |
| "eval_runtime": 16.0245, | |
| "eval_samples_per_second": 700.739, | |
| "eval_steps_per_second": 87.616, | |
| "step": 440 | |
| }, | |
| { | |
| "epoch": 0.32, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 445 | |
| }, | |
| { | |
| "epoch": 0.32, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 450 | |
| }, | |
| { | |
| "epoch": 0.32, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 455 | |
| }, | |
| { | |
| "epoch": 0.33, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 460 | |
| }, | |
| { | |
| "epoch": 0.33, | |
| "eval_loss": NaN, | |
| "eval_runtime": 16.055, | |
| "eval_samples_per_second": 699.409, | |
| "eval_steps_per_second": 87.45, | |
| "step": 460 | |
| }, | |
| { | |
| "epoch": 0.33, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 465 | |
| }, | |
| { | |
| "epoch": 0.33, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 470 | |
| }, | |
| { | |
| "epoch": 0.34, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 475 | |
| }, | |
| { | |
| "epoch": 0.34, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 480 | |
| }, | |
| { | |
| "epoch": 0.34, | |
| "eval_loss": NaN, | |
| "eval_runtime": 15.8376, | |
| "eval_samples_per_second": 709.01, | |
| "eval_steps_per_second": 88.65, | |
| "step": 480 | |
| }, | |
| { | |
| "epoch": 0.35, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 485 | |
| }, | |
| { | |
| "epoch": 0.35, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 490 | |
| }, | |
| { | |
| "epoch": 0.35, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 495 | |
| }, | |
| { | |
| "epoch": 0.36, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 500 | |
| }, | |
| { | |
| "epoch": 0.36, | |
| "eval_loss": NaN, | |
| "eval_runtime": 16.0396, | |
| "eval_samples_per_second": 700.08, | |
| "eval_steps_per_second": 87.533, | |
| "step": 500 | |
| }, | |
| { | |
| "epoch": 0.36, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 505 | |
| }, | |
| { | |
| "epoch": 0.36, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 510 | |
| }, | |
| { | |
| "epoch": 0.37, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 515 | |
| }, | |
| { | |
| "epoch": 0.37, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 520 | |
| }, | |
| { | |
| "epoch": 0.37, | |
| "eval_loss": NaN, | |
| "eval_runtime": 16.033, | |
| "eval_samples_per_second": 700.366, | |
| "eval_steps_per_second": 87.569, | |
| "step": 520 | |
| }, | |
| { | |
| "epoch": 0.37, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 525 | |
| }, | |
| { | |
| "epoch": 0.38, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 530 | |
| }, | |
| { | |
| "epoch": 0.38, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 535 | |
| }, | |
| { | |
| "epoch": 0.38, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 540 | |
| }, | |
| { | |
| "epoch": 0.38, | |
| "eval_loss": NaN, | |
| "eval_runtime": 15.8405, | |
| "eval_samples_per_second": 708.88, | |
| "eval_steps_per_second": 88.634, | |
| "step": 540 | |
| }, | |
| { | |
| "epoch": 0.39, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 545 | |
| }, | |
| { | |
| "epoch": 0.39, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 550 | |
| }, | |
| { | |
| "epoch": 0.4, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 555 | |
| }, | |
| { | |
| "epoch": 0.4, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 560 | |
| }, | |
| { | |
| "epoch": 0.4, | |
| "eval_loss": NaN, | |
| "eval_runtime": 16.0407, | |
| "eval_samples_per_second": 700.03, | |
| "eval_steps_per_second": 87.527, | |
| "step": 560 | |
| }, | |
| { | |
| "epoch": 0.4, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 565 | |
| }, | |
| { | |
| "epoch": 0.41, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 570 | |
| }, | |
| { | |
| "epoch": 0.41, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 575 | |
| }, | |
| { | |
| "epoch": 0.41, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 580 | |
| }, | |
| { | |
| "epoch": 0.41, | |
| "eval_loss": NaN, | |
| "eval_runtime": 15.8636, | |
| "eval_samples_per_second": 707.848, | |
| "eval_steps_per_second": 88.505, | |
| "step": 580 | |
| }, | |
| { | |
| "epoch": 0.42, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 585 | |
| }, | |
| { | |
| "epoch": 0.42, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 20911.1437, | |
| "step": 590 | |
| }, | |
| { | |
| "epoch": 0.42, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 595 | |
| }, | |
| { | |
| "epoch": 0.43, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 600 | |
| }, | |
| { | |
| "epoch": 0.43, | |
| "eval_loss": NaN, | |
| "eval_runtime": 15.8487, | |
| "eval_samples_per_second": 708.511, | |
| "eval_steps_per_second": 88.588, | |
| "step": 600 | |
| }, | |
| { | |
| "epoch": 0.43, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 605 | |
| }, | |
| { | |
| "epoch": 0.43, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 610 | |
| }, | |
| { | |
| "epoch": 0.44, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 615 | |
| }, | |
| { | |
| "epoch": 0.44, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 620 | |
| }, | |
| { | |
| "epoch": 0.44, | |
| "eval_loss": NaN, | |
| "eval_runtime": 16.0099, | |
| "eval_samples_per_second": 701.38, | |
| "eval_steps_per_second": 87.696, | |
| "step": 620 | |
| }, | |
| { | |
| "epoch": 0.45, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 625 | |
| }, | |
| { | |
| "epoch": 0.45, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 630 | |
| }, | |
| { | |
| "epoch": 0.45, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 635 | |
| }, | |
| { | |
| "epoch": 0.46, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 640 | |
| }, | |
| { | |
| "epoch": 0.46, | |
| "eval_loss": NaN, | |
| "eval_runtime": 15.9086, | |
| "eval_samples_per_second": 705.847, | |
| "eval_steps_per_second": 88.254, | |
| "step": 640 | |
| }, | |
| { | |
| "epoch": 0.46, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 645 | |
| }, | |
| { | |
| "epoch": 0.46, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 650 | |
| }, | |
| { | |
| "epoch": 0.47, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 655 | |
| }, | |
| { | |
| "epoch": 0.47, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 660 | |
| }, | |
| { | |
| "epoch": 0.47, | |
| "eval_loss": NaN, | |
| "eval_runtime": 15.903, | |
| "eval_samples_per_second": 706.093, | |
| "eval_steps_per_second": 88.285, | |
| "step": 660 | |
| }, | |
| { | |
| "epoch": 0.47, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 665 | |
| }, | |
| { | |
| "epoch": 0.48, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 670 | |
| }, | |
| { | |
| "epoch": 0.48, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 675 | |
| }, | |
| { | |
| "epoch": 0.48, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 680 | |
| }, | |
| { | |
| "epoch": 0.48, | |
| "eval_loss": NaN, | |
| "eval_runtime": 16.0785, | |
| "eval_samples_per_second": 698.388, | |
| "eval_steps_per_second": 87.322, | |
| "step": 680 | |
| }, | |
| { | |
| "epoch": 0.49, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 685 | |
| }, | |
| { | |
| "epoch": 0.49, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 690 | |
| }, | |
| { | |
| "epoch": 0.5, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 695 | |
| }, | |
| { | |
| "epoch": 0.5, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 700 | |
| }, | |
| { | |
| "epoch": 0.5, | |
| "eval_loss": NaN, | |
| "eval_runtime": 15.886, | |
| "eval_samples_per_second": 706.848, | |
| "eval_steps_per_second": 88.38, | |
| "step": 700 | |
| }, | |
| { | |
| "epoch": 0.5, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 705 | |
| }, | |
| { | |
| "epoch": 0.51, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 710 | |
| }, | |
| { | |
| "epoch": 0.51, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 715 | |
| }, | |
| { | |
| "epoch": 0.51, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 720 | |
| }, | |
| { | |
| "epoch": 0.51, | |
| "eval_loss": NaN, | |
| "eval_runtime": 15.8917, | |
| "eval_samples_per_second": 706.593, | |
| "eval_steps_per_second": 88.348, | |
| "step": 720 | |
| }, | |
| { | |
| "epoch": 0.52, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 725 | |
| }, | |
| { | |
| "epoch": 0.52, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 730 | |
| }, | |
| { | |
| "epoch": 0.52, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 735 | |
| }, | |
| { | |
| "epoch": 0.53, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 740 | |
| }, | |
| { | |
| "epoch": 0.53, | |
| "eval_loss": NaN, | |
| "eval_runtime": 16.0826, | |
| "eval_samples_per_second": 698.208, | |
| "eval_steps_per_second": 87.299, | |
| "step": 740 | |
| }, | |
| { | |
| "epoch": 0.53, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 745 | |
| }, | |
| { | |
| "epoch": 0.53, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 750 | |
| }, | |
| { | |
| "epoch": 0.54, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 755 | |
| }, | |
| { | |
| "epoch": 0.54, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 760 | |
| }, | |
| { | |
| "epoch": 0.54, | |
| "eval_loss": NaN, | |
| "eval_runtime": 15.8856, | |
| "eval_samples_per_second": 706.864, | |
| "eval_steps_per_second": 88.382, | |
| "step": 760 | |
| }, | |
| { | |
| "epoch": 0.55, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 765 | |
| }, | |
| { | |
| "epoch": 0.55, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 770 | |
| }, | |
| { | |
| "epoch": 0.55, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 775 | |
| }, | |
| { | |
| "epoch": 0.56, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 780 | |
| }, | |
| { | |
| "epoch": 0.56, | |
| "eval_loss": NaN, | |
| "eval_runtime": 16.0251, | |
| "eval_samples_per_second": 700.715, | |
| "eval_steps_per_second": 87.613, | |
| "step": 780 | |
| }, | |
| { | |
| "epoch": 0.56, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 785 | |
| }, | |
| { | |
| "epoch": 0.56, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 790 | |
| }, | |
| { | |
| "epoch": 0.57, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 795 | |
| }, | |
| { | |
| "epoch": 0.57, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 800 | |
| }, | |
| { | |
| "epoch": 0.57, | |
| "eval_loss": NaN, | |
| "eval_runtime": 16.0529, | |
| "eval_samples_per_second": 699.501, | |
| "eval_steps_per_second": 87.461, | |
| "step": 800 | |
| }, | |
| { | |
| "epoch": 0.57, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 805 | |
| }, | |
| { | |
| "epoch": 0.58, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 810 | |
| }, | |
| { | |
| "epoch": 0.58, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 815 | |
| }, | |
| { | |
| "epoch": 0.58, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 820 | |
| }, | |
| { | |
| "epoch": 0.58, | |
| "eval_loss": NaN, | |
| "eval_runtime": 15.8721, | |
| "eval_samples_per_second": 707.469, | |
| "eval_steps_per_second": 88.457, | |
| "step": 820 | |
| }, | |
| { | |
| "epoch": 0.59, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 825 | |
| }, | |
| { | |
| "epoch": 0.59, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 830 | |
| }, | |
| { | |
| "epoch": 0.59, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 835 | |
| }, | |
| { | |
| "epoch": 0.6, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 840 | |
| }, | |
| { | |
| "epoch": 0.6, | |
| "eval_loss": NaN, | |
| "eval_runtime": 16.0458, | |
| "eval_samples_per_second": 699.808, | |
| "eval_steps_per_second": 87.499, | |
| "step": 840 | |
| }, | |
| { | |
| "epoch": 0.6, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 845 | |
| }, | |
| { | |
| "epoch": 0.61, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 850 | |
| }, | |
| { | |
| "epoch": 0.61, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 855 | |
| }, | |
| { | |
| "epoch": 0.61, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 860 | |
| }, | |
| { | |
| "epoch": 0.61, | |
| "eval_loss": NaN, | |
| "eval_runtime": 16.0756, | |
| "eval_samples_per_second": 698.514, | |
| "eval_steps_per_second": 87.338, | |
| "step": 860 | |
| }, | |
| { | |
| "epoch": 0.62, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 865 | |
| }, | |
| { | |
| "epoch": 0.62, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 870 | |
| }, | |
| { | |
| "epoch": 0.62, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 875 | |
| }, | |
| { | |
| "epoch": 0.63, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 880 | |
| }, | |
| { | |
| "epoch": 0.63, | |
| "eval_loss": NaN, | |
| "eval_runtime": 15.8991, | |
| "eval_samples_per_second": 706.267, | |
| "eval_steps_per_second": 88.307, | |
| "step": 880 | |
| }, | |
| { | |
| "epoch": 0.63, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 885 | |
| }, | |
| { | |
| "epoch": 0.63, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 890 | |
| }, | |
| { | |
| "epoch": 0.64, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 895 | |
| }, | |
| { | |
| "epoch": 0.64, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 900 | |
| }, | |
| { | |
| "epoch": 0.64, | |
| "eval_loss": NaN, | |
| "eval_runtime": 16.0389, | |
| "eval_samples_per_second": 700.109, | |
| "eval_steps_per_second": 87.537, | |
| "step": 900 | |
| }, | |
| { | |
| "epoch": 0.64, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 905 | |
| }, | |
| { | |
| "epoch": 0.65, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 910 | |
| }, | |
| { | |
| "epoch": 0.65, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 915 | |
| }, | |
| { | |
| "epoch": 0.66, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 920 | |
| }, | |
| { | |
| "epoch": 0.66, | |
| "eval_loss": NaN, | |
| "eval_runtime": 16.0475, | |
| "eval_samples_per_second": 699.735, | |
| "eval_steps_per_second": 87.49, | |
| "step": 920 | |
| }, | |
| { | |
| "epoch": 0.66, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 925 | |
| }, | |
| { | |
| "epoch": 0.66, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 930 | |
| }, | |
| { | |
| "epoch": 0.67, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 935 | |
| }, | |
| { | |
| "epoch": 0.67, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 940 | |
| }, | |
| { | |
| "epoch": 0.67, | |
| "eval_loss": NaN, | |
| "eval_runtime": 15.8747, | |
| "eval_samples_per_second": 707.353, | |
| "eval_steps_per_second": 88.443, | |
| "step": 940 | |
| }, | |
| { | |
| "epoch": 0.67, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 945 | |
| }, | |
| { | |
| "epoch": 0.68, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 950 | |
| }, | |
| { | |
| "epoch": 0.68, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 955 | |
| }, | |
| { | |
| "epoch": 0.68, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 960 | |
| }, | |
| { | |
| "epoch": 0.68, | |
| "eval_loss": NaN, | |
| "eval_runtime": 16.0475, | |
| "eval_samples_per_second": 699.735, | |
| "eval_steps_per_second": 87.49, | |
| "step": 960 | |
| }, | |
| { | |
| "epoch": 0.69, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 965 | |
| }, | |
| { | |
| "epoch": 0.69, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 970 | |
| }, | |
| { | |
| "epoch": 0.69, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 975 | |
| }, | |
| { | |
| "epoch": 0.7, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 980 | |
| }, | |
| { | |
| "epoch": 0.7, | |
| "eval_loss": NaN, | |
| "eval_runtime": 16.0347, | |
| "eval_samples_per_second": 700.294, | |
| "eval_steps_per_second": 87.56, | |
| "step": 980 | |
| }, | |
| { | |
| "epoch": 0.7, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 985 | |
| }, | |
| { | |
| "epoch": 0.71, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 990 | |
| }, | |
| { | |
| "epoch": 0.71, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 995 | |
| }, | |
| { | |
| "epoch": 0.71, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1000 | |
| }, | |
| { | |
| "epoch": 0.71, | |
| "eval_loss": NaN, | |
| "eval_runtime": 15.8788, | |
| "eval_samples_per_second": 707.171, | |
| "eval_steps_per_second": 88.42, | |
| "step": 1000 | |
| }, | |
| { | |
| "epoch": 0.72, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1005 | |
| }, | |
| { | |
| "epoch": 0.72, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1010 | |
| }, | |
| { | |
| "epoch": 0.72, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1015 | |
| }, | |
| { | |
| "epoch": 0.73, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1020 | |
| }, | |
| { | |
| "epoch": 0.73, | |
| "eval_loss": NaN, | |
| "eval_runtime": 16.0535, | |
| "eval_samples_per_second": 699.474, | |
| "eval_steps_per_second": 87.458, | |
| "step": 1020 | |
| }, | |
| { | |
| "epoch": 0.73, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1025 | |
| }, | |
| { | |
| "epoch": 0.73, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1030 | |
| }, | |
| { | |
| "epoch": 0.74, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1035 | |
| }, | |
| { | |
| "epoch": 0.74, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1040 | |
| }, | |
| { | |
| "epoch": 0.74, | |
| "eval_loss": NaN, | |
| "eval_runtime": 15.8931, | |
| "eval_samples_per_second": 706.532, | |
| "eval_steps_per_second": 88.34, | |
| "step": 1040 | |
| }, | |
| { | |
| "epoch": 0.74, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1045 | |
| }, | |
| { | |
| "epoch": 0.75, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1050 | |
| }, | |
| { | |
| "epoch": 0.75, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1055 | |
| }, | |
| { | |
| "epoch": 0.76, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1060 | |
| }, | |
| { | |
| "epoch": 0.76, | |
| "eval_loss": NaN, | |
| "eval_runtime": 15.8928, | |
| "eval_samples_per_second": 706.545, | |
| "eval_steps_per_second": 88.342, | |
| "step": 1060 | |
| }, | |
| { | |
| "epoch": 0.76, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1065 | |
| }, | |
| { | |
| "epoch": 0.76, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1070 | |
| }, | |
| { | |
| "epoch": 0.77, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1075 | |
| }, | |
| { | |
| "epoch": 0.77, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1080 | |
| }, | |
| { | |
| "epoch": 0.77, | |
| "eval_loss": NaN, | |
| "eval_runtime": 16.0634, | |
| "eval_samples_per_second": 699.043, | |
| "eval_steps_per_second": 87.404, | |
| "step": 1080 | |
| }, | |
| { | |
| "epoch": 0.77, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1085 | |
| }, | |
| { | |
| "epoch": 0.78, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1090 | |
| }, | |
| { | |
| "epoch": 0.78, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1095 | |
| }, | |
| { | |
| "epoch": 0.78, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1100 | |
| }, | |
| { | |
| "epoch": 0.78, | |
| "eval_loss": NaN, | |
| "eval_runtime": 15.8957, | |
| "eval_samples_per_second": 706.418, | |
| "eval_steps_per_second": 88.326, | |
| "step": 1100 | |
| }, | |
| { | |
| "epoch": 0.79, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1105 | |
| }, | |
| { | |
| "epoch": 0.79, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1110 | |
| }, | |
| { | |
| "epoch": 0.79, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1115 | |
| }, | |
| { | |
| "epoch": 0.8, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1120 | |
| }, | |
| { | |
| "epoch": 0.8, | |
| "eval_loss": NaN, | |
| "eval_runtime": 15.899, | |
| "eval_samples_per_second": 706.272, | |
| "eval_steps_per_second": 88.308, | |
| "step": 1120 | |
| }, | |
| { | |
| "epoch": 0.8, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1125 | |
| }, | |
| { | |
| "epoch": 0.81, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1130 | |
| }, | |
| { | |
| "epoch": 0.81, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1135 | |
| }, | |
| { | |
| "epoch": 0.81, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1140 | |
| }, | |
| { | |
| "epoch": 0.81, | |
| "eval_loss": NaN, | |
| "eval_runtime": 16.0663, | |
| "eval_samples_per_second": 698.916, | |
| "eval_steps_per_second": 87.388, | |
| "step": 1140 | |
| }, | |
| { | |
| "epoch": 0.82, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1145 | |
| }, | |
| { | |
| "epoch": 0.82, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1150 | |
| }, | |
| { | |
| "epoch": 0.82, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1155 | |
| }, | |
| { | |
| "epoch": 0.83, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1160 | |
| }, | |
| { | |
| "epoch": 0.83, | |
| "eval_loss": NaN, | |
| "eval_runtime": 15.8722, | |
| "eval_samples_per_second": 707.463, | |
| "eval_steps_per_second": 88.456, | |
| "step": 1160 | |
| }, | |
| { | |
| "epoch": 0.83, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1165 | |
| }, | |
| { | |
| "epoch": 0.83, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1170 | |
| }, | |
| { | |
| "epoch": 0.84, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1175 | |
| }, | |
| { | |
| "epoch": 0.84, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1180 | |
| }, | |
| { | |
| "epoch": 0.84, | |
| "eval_loss": NaN, | |
| "eval_runtime": 16.0392, | |
| "eval_samples_per_second": 700.098, | |
| "eval_steps_per_second": 87.536, | |
| "step": 1180 | |
| }, | |
| { | |
| "epoch": 0.84, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1185 | |
| }, | |
| { | |
| "epoch": 0.85, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 38.1152, | |
| "step": 1190 | |
| }, | |
| { | |
| "epoch": 0.85, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 420.7575, | |
| "step": 1195 | |
| }, | |
| { | |
| "epoch": 0.86, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1200 | |
| }, | |
| { | |
| "epoch": 0.86, | |
| "eval_loss": NaN, | |
| "eval_runtime": 16.0601, | |
| "eval_samples_per_second": 699.186, | |
| "eval_steps_per_second": 87.422, | |
| "step": 1200 | |
| }, | |
| { | |
| "epoch": 0.86, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1205 | |
| }, | |
| { | |
| "epoch": 0.86, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1210 | |
| }, | |
| { | |
| "epoch": 0.87, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1215 | |
| }, | |
| { | |
| "epoch": 0.87, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1220 | |
| }, | |
| { | |
| "epoch": 0.87, | |
| "eval_loss": NaN, | |
| "eval_runtime": 15.8785, | |
| "eval_samples_per_second": 707.182, | |
| "eval_steps_per_second": 88.421, | |
| "step": 1220 | |
| }, | |
| { | |
| "epoch": 0.87, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1225 | |
| }, | |
| { | |
| "epoch": 0.88, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1230 | |
| }, | |
| { | |
| "epoch": 0.88, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1235 | |
| }, | |
| { | |
| "epoch": 0.88, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1240 | |
| }, | |
| { | |
| "epoch": 0.88, | |
| "eval_loss": NaN, | |
| "eval_runtime": 16.0442, | |
| "eval_samples_per_second": 699.88, | |
| "eval_steps_per_second": 87.508, | |
| "step": 1240 | |
| }, | |
| { | |
| "epoch": 0.89, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1245 | |
| }, | |
| { | |
| "epoch": 0.89, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1250 | |
| }, | |
| { | |
| "epoch": 0.89, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1255 | |
| }, | |
| { | |
| "epoch": 0.9, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1260 | |
| }, | |
| { | |
| "epoch": 0.9, | |
| "eval_loss": NaN, | |
| "eval_runtime": 16.0664, | |
| "eval_samples_per_second": 698.912, | |
| "eval_steps_per_second": 87.387, | |
| "step": 1260 | |
| }, | |
| { | |
| "epoch": 0.9, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1265 | |
| }, | |
| { | |
| "epoch": 0.9, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1270 | |
| }, | |
| { | |
| "epoch": 0.91, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1275 | |
| }, | |
| { | |
| "epoch": 0.91, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1280 | |
| }, | |
| { | |
| "epoch": 0.91, | |
| "eval_loss": NaN, | |
| "eval_runtime": 15.8891, | |
| "eval_samples_per_second": 706.713, | |
| "eval_steps_per_second": 88.363, | |
| "step": 1280 | |
| }, | |
| { | |
| "epoch": 0.92, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1285 | |
| }, | |
| { | |
| "epoch": 0.92, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1290 | |
| }, | |
| { | |
| "epoch": 0.92, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1295 | |
| }, | |
| { | |
| "epoch": 0.93, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1300 | |
| }, | |
| { | |
| "epoch": 0.93, | |
| "eval_loss": NaN, | |
| "eval_runtime": 16.0665, | |
| "eval_samples_per_second": 698.906, | |
| "eval_steps_per_second": 87.387, | |
| "step": 1300 | |
| }, | |
| { | |
| "epoch": 0.93, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1305 | |
| }, | |
| { | |
| "epoch": 0.93, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1310 | |
| }, | |
| { | |
| "epoch": 0.94, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1315 | |
| }, | |
| { | |
| "epoch": 0.94, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1320 | |
| }, | |
| { | |
| "epoch": 0.94, | |
| "eval_loss": NaN, | |
| "eval_runtime": 16.0663, | |
| "eval_samples_per_second": 698.918, | |
| "eval_steps_per_second": 87.388, | |
| "step": 1320 | |
| }, | |
| { | |
| "epoch": 0.94, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1325 | |
| }, | |
| { | |
| "epoch": 0.95, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1330 | |
| }, | |
| { | |
| "epoch": 0.95, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1335 | |
| }, | |
| { | |
| "epoch": 0.95, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1340 | |
| }, | |
| { | |
| "epoch": 0.95, | |
| "eval_loss": NaN, | |
| "eval_runtime": 15.9135, | |
| "eval_samples_per_second": 705.629, | |
| "eval_steps_per_second": 88.227, | |
| "step": 1340 | |
| }, | |
| { | |
| "epoch": 0.96, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1345 | |
| }, | |
| { | |
| "epoch": 0.96, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1350 | |
| }, | |
| { | |
| "epoch": 0.97, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1355 | |
| }, | |
| { | |
| "epoch": 0.97, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1360 | |
| }, | |
| { | |
| "epoch": 0.97, | |
| "eval_loss": NaN, | |
| "eval_runtime": 16.0513, | |
| "eval_samples_per_second": 699.571, | |
| "eval_steps_per_second": 87.47, | |
| "step": 1360 | |
| }, | |
| { | |
| "epoch": 0.97, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1365 | |
| }, | |
| { | |
| "epoch": 0.98, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 4.7146, | |
| "step": 1370 | |
| }, | |
| { | |
| "epoch": 0.98, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1375 | |
| }, | |
| { | |
| "epoch": 0.98, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1380 | |
| }, | |
| { | |
| "epoch": 0.98, | |
| "eval_loss": NaN, | |
| "eval_runtime": 16.0559, | |
| "eval_samples_per_second": 699.371, | |
| "eval_steps_per_second": 87.445, | |
| "step": 1380 | |
| }, | |
| { | |
| "epoch": 0.99, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1385 | |
| }, | |
| { | |
| "epoch": 0.99, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1390 | |
| }, | |
| { | |
| "epoch": 0.99, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1395 | |
| }, | |
| { | |
| "epoch": 1.0, | |
| "grad_norm": NaN, | |
| "learning_rate": 0.0003, | |
| "loss": 0.0, | |
| "step": 1400 | |
| }, | |
| { | |
| "epoch": 1.0, | |
| "eval_loss": NaN, | |
| "eval_runtime": 15.8929, | |
| "eval_samples_per_second": 706.54, | |
| "eval_steps_per_second": 88.341, | |
| "step": 1400 | |
| } | |
| ], | |
| "logging_steps": 5, | |
| "max_steps": 1403, | |
| "num_input_tokens_seen": 0, | |
| "num_train_epochs": 1, | |
| "save_steps": 20, | |
| "total_flos": 1585884081684480.0, | |
| "train_batch_size": 4, | |
| "trial_name": null, | |
| "trial_params": null | |
| } | |