README.md

metadata

language: en
license: mit
library_name: pytorch

Knowledge Continuity Regularized Network

Trainer Hyperparameters:

• lr = 1e-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 = 0.001
• alpha = 2.0
• beta = 2.0

Extended Logs:

eval_loss	eval_accuracy	epoch
8.842	0.208	1.0
8.834	0.208	2.0
7.990	0.292	3.0
4.612	0.750	4.0
8.841	0.208	5.0
8.842	0.208	6.0
8.843	0.208	7.0
8.842	0.208	8.0
8.845	0.208	9.0
8.842	0.208	10.0
8.843	0.208	11.0
8.843	0.208	12.0
8.843	0.208	13.0
8.841	0.208	14.0
8.843	0.208	15.0
8.843	0.208	16.0
8.843	0.208	17.0
8.845	0.208	18.0
8.843	0.208	19.0