Lai Wei

@tjruwase Thanks for the review! will add test and reformat later today.

@tjruwase Sorry I totally forgot about this. I merged some tests from @junxu with test (Thanks to @junxu !!) It should be good now.

cudnn.TemporalConvolution is also not converted, since it's a wrapper on cudnn.SpatialConvolution.

Hi, any ideas how to convert TemporalConvolution from nn to cudnn ?

FYI torch distributed launch is deprecated, PyTorch suggest to use torchrun https://pytorch.org/docs/stable/distributed.html#launch-utility https://pytorch.org/docs/stable/elastic/run.html#launcher-api

@ziyuang please trigger CI and update this PR?

FYI the Gluon implementation is using `stride` for first Conv2D and `1` for the second. Which is different from the Symbol implementation: https://github.com/apache/incubator-mxnet/blob/master/python/mxnet/gluon/model_zoo/vision/resnet.py#L90

@ddavydenko ping to help move forward, let's get this merged. Thanks!

Hi team, any plan to fix this? without transformer engine working it's hard to justify the price for H100s.

Hi @init27 @tryrobbo , could you help take a look and merge this? I'd like to demo the official repo in a workshop instead of my fork. Thanks!