learning_rate: 5e-05
train_batch_size: 1024
eval_batch_size: 1024
seed: 42
optimizer: Use adamw_torch with betas=(0.9,0.95) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
lr_scheduler_type: cosine
lr_scheduler_warmup_steps: 156250
num_epochs: 9
mixed_precision_training: Native AMP

Training Loss	Epoch	Step	Validation Loss
0.4552	0.3774	500	1.7358
0.3672	0.7547	1000	1.4004
0.3344	1.1321	1500	1.2714
0.3115	1.5094	2000	1.1743
0.2946	1.8868	2500	1.0900
0.2768	2.2642	3000	1.0181
0.2602	2.6415	3500	0.9613
0.2507	3.0189	4000	0.9021
0.2365	3.3962	4500	0.8570
0.2248	3.7736	5000	0.8209
0.2172	4.1509	5500	0.7872
0.212	4.5283	6000	0.7676
0.2045	4.9057	6500	0.7501
0.1998	5.2830	7000	0.7287
0.1959	5.6604	7500	0.7104
0.1894	6.0377	8000	0.7057
0.1877	6.4151	8500	0.6901
0.1826	6.7925	9000	0.6747
0.1813	7.1698	9500	0.6666
0.1746	7.5472	10000	0.6579
0.1747	7.9245	10500	0.6481
0.1713	8.3019	11000	0.6404
0.1726	8.6792	11500	0.6339

suku9/pretrained_spdl_05_17

pretrained_spdl_05_17