[QUESTION] plan to implement zero bubble pipeline or dual pipeline and MoE comm-comp overlapping

Hi team,
It looks like you're busy reproducing DeepSeek V3 features such as device-routed load balancing, aux-free load balancing, and etc.
I'm just wondering if you have any plans to implement zero-bubble pipelining or dual pipelining and MoE communication-computation overlapping, as these are crucial for optimizing MoE training.