learning_rate: 0.001
train_batch_size: 32
eval_batch_size: 64
seed: 21
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
6.1412	0.9994	1470	5.5240	0.1764
4.5259	1.9992	2940	5.4067	0.1823
3.8908	2.9991	4410	5.3111	0.1857
3.7115	3.9996	5881	5.2018	0.1937
3.4863	4.9994	7351	5.1925	0.1938
3.4079	5.9992	8821	5.1520	0.1973
3.3056	6.9991	10291	5.1326	0.1999
3.258	7.9996	11762	5.1119	0.1997
3.2065	8.9994	13232	5.1225	0.2009
3.1699	9.9992	14702	5.1300	0.1987
3.1451	10.9991	16172	5.0815	0.2020
3.1079	11.9996	17643	5.1214	0.2012
3.1043	12.9994	19113	5.0818	0.2012
3.0668	13.9992	20583	5.1290	0.2022
3.0777	14.9991	22053	5.1106	0.1996
3.039	15.9996	23524	5.1058	0.2006
3.0432	16.9994	24994	5.1083	0.2036
3.0188	17.9992	26464	5.1309	0.2016
3.0246	18.9991	27934	5.1190	0.1996
3.0115	19.9962	29400	5.0807	0.2038

qing-yao/long-first-headfinal_seed-21_1e-3

long_first_headfinal_seed-21_1e-3