asun17904/anliR1-bert-base-uncased

Knowledge Continuity Regularized Network

Dataset: ANLI Round: None

Trainer Hyperparameters:

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

Regularization Hyperparameters

• numerical stability denominator constant = 0.01
• lambda = 0.01
• alpha = 2.0
• beta = 2.0

Extended Logs:

eval_loss	eval_accuracy	epoch
22.099	0.350	1.0
22.046	0.400	2.0
21.908	0.400	3.0
21.955	0.400	4.0
22.005	0.400	5.0
22.104	0.400	6.0
22.253	0.400	7.0
22.441	0.400	8.0
22.574	0.400	9.0
22.605	0.400	10.0
22.551	0.400	11.0
22.509	0.400	12.0
22.607	0.400	13.0
22.663	0.400	14.0
22.797	0.400	15.0
22.860	0.400	16.0
22.913	0.400	17.0
22.923	0.400	18.0
22.924	0.400	19.0
22.915	0.400	20.0
22.938	0.400	21.0
23.008	0.400	22.0
23.005	0.400	23.0
23.025	0.400	24.0
23.082	0.400	25.0
23.077	0.400	26.0
23.082	0.400	27.0
23.042	0.400	28.0
23.036	0.400	29.0
23.062	0.400	30.0
23.071	0.400	31.0
23.068	0.400	32.0
23.080	0.400	33.0
23.127	0.400	34.0
23.276	0.400	35.0
23.254	0.400	36.0
23.235	0.400	37.0
23.298	0.400	38.0
23.186	0.400	39.0
23.164	0.400	40.0
23.157	0.400	41.0
23.215	0.400	42.0
23.208	0.400	43.0
23.219	0.400	44.0
23.199	0.400	45.0
23.186	0.400	46.0
23.149	0.400	47.0
23.252	0.400	48.0
23.162	0.400	49.0