Compressing Pre-trained Language Models by Matrix Decomposition

259

Knowledge Triples

Source Entity	Relationship	Target Entity
BERT Base Model Material	evaluatedIn	GLUE Benchmark Dataset Material
Feature Distillation Method	partOf	Model Compression Task
Large Pre-trained Language Models Material	appliesTo	NLPTasks Task
Matrix Decomposition Method	partOf	Model Compression Task
Model Compression Task	appliesTo	Large Pre-trained Language Models Material
Model Compression Task	reduces	Inference Time Cost Metric