How can I benefit from taichi when use pytorch

[Crazybean-lwb]

Background

I need to recurrent training or fine-tune process in papers most of the time. Almost all high citation papers build net by pytorch, recurrenting cost most of my time.

Aim

I have 100+ nvidia graphics card， hope to use up all graphics card resources.

How can taichi help with reducing time cost, or speeding up training or fine-tune.