kennethge123/superglue-rte-bert-base-uncased-kd

Plainly Optimized Network

Dataset: SUPERGLUE

Trainer Hyperparameters:

• lr = 5e-05
• per_device_batch_size = 8
• gradient_accumulation_steps = 2
• weight_decay = 1e-09
• seed = 42

eval_loss	eval_accuracy	epoch
22.014	0.471	1.0
19.411	0.659	2.0
18.711	0.696	3.0
19.141	0.652	4.0
19.924	0.638	5.0
19.229	0.652	6.0
20.306	0.623	7.0
19.739	0.645	8.0
20.873	0.623	9.0
20.285	0.638	10.0
18.900	0.696	11.0
18.971	0.681	12.0
19.230	0.667	13.0
19.039	0.674	14.0
19.080	0.667	15.0
18.997	0.681	16.0
18.619	0.681	17.0
18.754	0.681	18.0
18.911	0.674	19.0