Stable single-machine, single-card training in the hidden space

v0.1.1

fix gradient divergence

implement basic meta-learning

v0.1-alpha

[Critical]: Fix the gradient instability problem caused by the optimi…