Zaida Zhou

Some contents are repeated with https://github.com/open-mmlab/mmcv/blob/master/docs/en/community/pr.md#1-get-the-most-recent-codebase

Hi @jaideep11061982 , thanks for your report. We have plans to support the features you mentioned in the future.

Hi @levan92 , as a workaround, you can set `with_step` as False. More discussions about the argument can be found at #913 https://github.com/open-mmlab/mmcv/blob/1f2500102834a01b86bf9ae4db227cd8d724fa6e/mmcv/runner/hooks/logger/wandb.py#L35

Hi @nicken , thanks for your report. We will try to reproduce the error.

I have reproduced the error. This error is a bit strange.

Hi, sorry for the late reply. [patch_norm_fp32](https://github.com/open-mmlab/mmcv/blob/master/mmcv/runner/fp16_utils.py#L276) will be helpful.

If `autocast` is enabled, whether the BN will be calculated with fp32 or fp16 depends on its inputs. There are two ways to force BN calculated with fp32. One is...

> LGTM, is there anything to be modified further? Maybe we need to add the checking of c++ in the lint of circleci and remove the lint config from github...

Hi, what command did you use to install mmcv-full?

Could you provide the output of `pip list | grep mmcv`? Probably you installed both mmcv and mmcv-full so you could uninstall them and install mmcv-full again.