Kenichi Maehashi

preping docker image: https://ci.preferred.jp/pytorch-pfn-extras.prep-linux/97573/

Or can we merge ProgressBarNotebook to exising ProgressBar?

/test

Thanks for sharing this! I tried this locally (my environment uses `export CUPY_NVCC_GENERATE_CODE=arch=compute_60,code=sm_60;arch=compute_70,code=sm_70;arch=compute_80,code=sm_80`) and measured `du -csh cupy cupyx cupy_backends`: * with compression: 95MB * without compression: 102MB In CuPy...

There should be some (maybe one time at the first CUDA execution in CuPy?) overhead when compression is enabled, so I'm wondering it's worth doing only for saving 17MB. I...

OK, let's try it and see how it would work :)

Thanks @khushi-411! Updated the list.

Thanks @jcfaracco, updated the list!

@emcastillo Could you briefly describe what was wrong with `cupyx/distributed/_nccl_comm.py`? :)

I've refined CI tooling part and rebased to the master (force pushed :bow:).