(草稿)
1,Learning method: continual learning ERNIE2.0
2,Pre-training method: SOP,DLM
3,Pre-training corpus:
4,Data Augme:
5,memory/params/EPOCHS:
5.1,distill:TinyBERT
5.2,pruning:adaptive span,
fine-tuning:adapt:Parameter-Efficient Transfer Learning for NLP;topK
dynamic halting:
sparse:
position embedding:
param sharing: