Xiulong Yuan

@knight2015 介绍的时候肯定是所有实验中最好的结果，很难达到所有的数据都到2%的压缩率的，除非质量分数很稳定的那种。

Well, this is strange, it should not happen because they are doing on different devices as you said. But we will look into this problem and give you feedback ASAP.

Well, this is strange

What is your `quiver_sampler's` mode

That's odd, let me check it

And also here I think `_maybe_schedule_new_actors` may never function because no one update training_state before this function. [source code here](https://github.com/ray-project/xgboost_ray/blob/f88118d44e338ffdba989d47eaf54fca9535deca/xgboost_ray/main.py#L1101)

Yes, this also confused me a lot

Um, this question is not actually answered i think ==!

Great！

非常感谢！