learning_rate: 0.001
train_batch_size: 32
eval_batch_size: 64
seed: 63
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
5.9916	0.9998	1490	4.3949	0.2949
4.3216	1.9997	2980	3.8825	0.3350
3.6895	2.9995	4470	3.6027	0.3586
3.5069	4.0	5961	3.4428	0.3740
3.2821	4.9998	7451	3.3414	0.3834
3.2066	5.9997	8941	3.2829	0.3888
3.0992	6.9995	10431	3.2434	0.3929
3.0599	8.0	11922	3.2210	0.3953
2.9981	8.9998	13412	3.1978	0.3975
2.9771	9.9997	14902	3.1883	0.3988
2.9355	10.9995	16392	3.1820	0.3998
2.9223	12.0	17883	3.1726	0.4008
2.8938	12.9998	19373	3.1647	0.4019
2.8825	13.9997	20863	3.1646	0.4019
2.8661	14.9995	22353	3.1594	0.4030
2.8555	16.0	23844	3.1606	0.4030
2.8487	16.9998	25334	3.1576	0.4033
2.8342	17.9997	26824	3.1580	0.4036
2.8345	18.9995	28314	3.1599	0.4034
2.8234	19.9966	29800	3.1546	0.4039

qing-yao/no-datives_seed-63_1e-3

nodative_cf_seed-63_1e-3