learning_rate: 0.0001
train_batch_size: 256
eval_batch_size: 256
seed: 51
optimizer: Use adamw_torch_fused with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
lr_scheduler_type: linear
lr_scheduler_warmup_steps: 500
num_epochs: 20
mixed_precision_training: Native AMP

Training Loss	Epoch	Step	Validation Loss
7.3516	1.0	399	6.7890
6.5633	2.0	798	6.5069
6.3592	3.0	1197	6.3738
6.2362	4.0	1596	6.2829
6.1393	5.0	1995	6.2115
6.053	6.0	2394	6.1545
5.9726	7.0	2793	6.1084
5.8974	8.0	3192	6.0681
5.8257	9.0	3591	6.0390
5.7582	10.0	3990	6.0189
5.6944	11.0	4389	6.0055
5.6337	12.0	4788	5.9945
5.5772	13.0	5187	5.9942
5.5242	14.0	5586	5.9944
5.4757	15.0	5985	5.9968
5.4306	16.0	6384	6.0002
5.3926	17.0	6783	6.0073
5.3609	18.0	7182	6.0108
5.3358	19.0	7581	6.0154
5.3172	20.0	7980	6.0179

fpadovani/wiki_sh1_51

wiki_sh1_51