search

BaguaSys / bagua

Bagua Speeds up PyTorch
https://tutorials-8ro.pages.dev/
MIT License
872 stars 83 forks source link

fix Lightning failing Bagua test for manual optimization #661

Open wangraying opened 1 year ago

wangraying commented 1 year ago

see this：https://github.com/Lightning-AI/lightning/pull/16225

woqidaideshi commented 1 year ago

@wangraying

After an investigation of this problem, I found that this failure is caused by scikit-optimize library used by Bagua. In detail, scikit-optimize library currently uses an old builtin type of numpy (i.e., numpy.int), which is deprecated by numpy>=1.24.0.

To avoid this problem, users can choose to use numpy<1.24.0.

In the meantime, scikit-optimize is trying to fix this problem (PR: https://github.com/scikit-optimize/scikit-optimize/pull/1123). After that, users can use the updated scikit-optimize with numpy>=1.24.0 in Bagua.