A Lite Bert For Self-Supervised Learning Language Representations
all google chinese weights
share_type and pre_ln