learning_rate: 0.001
train_batch_size: 32
eval_batch_size: 64
seed: 42
gradient_accumulation_steps: 8
total_train_batch_size: 256
optimizer: Use OptimizerNames.ADAMW_TORCH 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: 32000
num_epochs: 20.0
mixed_precision_training: Native AMP

Training Loss	Epoch	Step	Validation Loss	Accuracy
6.1724	0.9994	1470	5.5103	0.1759
4.5289	1.9992	2940	5.4000	0.1844
3.8901	2.9991	4410	5.3044	0.1895
3.7154	3.9996	5881	5.2299	0.1952
3.4885	4.9994	7351	5.1806	0.1983
3.4097	5.9992	8821	5.1625	0.1984
3.3049	6.9991	10291	5.1184	0.1994
3.2579	7.9996	11762	5.1354	0.2021
3.2058	8.9994	13232	5.1414	0.2010
3.1678	9.9992	14702	5.1105	0.2010
3.143	10.9991	16172	5.0866	0.1999
3.1069	11.9996	17643	5.1130	0.2012
3.1019	12.9994	19113	5.1315	0.2012
3.0681	13.9992	20583	5.1160	0.2007

qing-yao/long-first-headfinal_seed-42_1e-3

long_first_headfinal_seed-42_1e-3