train.log

['run_entity.py', '--do_train', '--do_eval', '--eval_test', '--learning_rate=1e-5', '--task_learning_rate=5e-4', '--train_batch_size=32', '--context_window', '0', '--task', 'ko', '--data_dir', 'sh_aug+klue', '--model', 'monologg/kobert', '--output_dir', 'sh_aug+klue_dir']
Namespace(bert_model_dir=None, bertadam=False, context_window=0, data_dir='sh_aug+klue', dev_data='sh_aug+klue\\dev.json', dev_pred_filename='ent_pred_dev.json', do_eval=True, do_train=True, eval_batch_size=32, eval_per_epoch=1, eval_test=True, learning_rate=1e-05, max_span_length=8, model='monologg/kobert', num_epoch=20, output_dir='sh_aug+klue_dir', print_loss_step=100, seed=0, task='ko', task_learning_rate=0.0005, test_data='sh_aug+klue\\test.json', test_pred_filename='ent_pred_test.json', train_batch_size=32, train_data='sh_aug+klue\\train.json', train_shuffle=False, use_albert=False, warmup_proportion=0.1)
Moving to CUDA...
# GPUs = 1
# Overlap: 0
Extracted 5574 samples from 5574 documents, with 16074 NER labels, 30.600 avg input length, 93 max length
Max Length: 93, max NER: 16
# Overlap: 0
Extracted 22849 samples from 22849 documents, with 57142 NER labels, 28.977 avg input length, 111 max length
Max Length: 111, max NER: 23
Epoch=0, iter=99, loss=382.98039
Epoch=0, iter=199, loss=92.19124
Epoch=0, iter=299, loss=18.17063
Epoch=0, iter=399, loss=17.86108
Epoch=0, iter=499, loss=17.41379
Epoch=0, iter=599, loss=17.11055
Epoch=0, iter=699, loss=14.89083
Evaluating...
Accuracy: 0.986949
Cor: 0, Pred TOT: 0, Gold TOT: 16074
P: 0.00000, R: 0.00000, F1: 0.00000
Used time: 43.244831
Epoch=1, iter=84, loss=12.58960
Epoch=1, iter=184, loss=10.30858
Epoch=1, iter=284, loss=8.34273
Epoch=1, iter=384, loss=6.43022
Epoch=1, iter=484, loss=5.32054
Epoch=1, iter=584, loss=4.63362
Epoch=1, iter=684, loss=4.06674
Evaluating...
Accuracy: 0.994374
Cor: 11197, Pred TOT: 13860, Gold TOT: 16074
P: 0.80786, R: 0.69659, F1: 0.74811
Used time: 44.417693
!!! Best valid (epoch=1): 74.81
Saving model to sh_aug+klue_dir...
Epoch=2, iter=69, loss=3.59020
Epoch=2, iter=169, loss=3.38379
Epoch=2, iter=269, loss=3.14521
Epoch=2, iter=369, loss=2.92039
Epoch=2, iter=469, loss=2.76133
Epoch=2, iter=569, loss=2.57549
Epoch=2, iter=669, loss=2.55223
Evaluating...
Accuracy: 0.995934
Cor: 13018, Pred TOT: 15461, Gold TOT: 16074
P: 0.84199, R: 0.80988, F1: 0.82562
Used time: 40.365823
!!! Best valid (epoch=2): 82.56
Saving model to sh_aug+klue_dir...
Epoch=3, iter=54, loss=2.32016
Epoch=3, iter=154, loss=2.18908
Epoch=3, iter=254, loss=2.26152
Epoch=3, iter=354, loss=2.06879
Epoch=3, iter=454, loss=2.03323
Epoch=3, iter=554, loss=1.93354
Epoch=3, iter=654, loss=1.86967
Evaluating...
Accuracy: 0.996433
Cor: 13478, Pred TOT: 15757, Gold TOT: 16074
P: 0.85537, R: 0.83850, F1: 0.84685
Used time: 41.699907
!!! Best valid (epoch=3): 84.68
Saving model to sh_aug+klue_dir...
Epoch=4, iter=39, loss=1.74192
Epoch=4, iter=139, loss=1.66635
Epoch=4, iter=239, loss=1.71463
Epoch=4, iter=339, loss=1.59322
Epoch=4, iter=439, loss=1.62414
Epoch=4, iter=539, loss=1.52524
Epoch=4, iter=639, loss=1.55066
Evaluating...
Accuracy: 0.996524
Cor: 13755, Pred TOT: 16223, Gold TOT: 16074
P: 0.84787, R: 0.85573, F1: 0.85178
Used time: 40.325464
!!! Best valid (epoch=4): 85.18
Saving model to sh_aug+klue_dir...
Epoch=5, iter=24, loss=1.37847
Epoch=5, iter=124, loss=1.37336
Epoch=5, iter=224, loss=1.39704
Epoch=5, iter=324, loss=1.28449
Epoch=5, iter=424, loss=1.41600
Epoch=5, iter=524, loss=1.20445
Epoch=5, iter=624, loss=1.30613
Evaluating...
Accuracy: 0.996762
Cor: 13890, Pred TOT: 16181, Gold TOT: 16074
P: 0.85841, R: 0.86413, F1: 0.86126
Used time: 67.691342
!!! Best valid (epoch=5): 86.13
Saving model to sh_aug+klue_dir...
Epoch=6, iter=9, loss=1.21428
Epoch=6, iter=109, loss=1.10397