Bin Bao

https://pytorch.org/hub/nvidia_deeplearningexamples_ssd/ is highly coupled with Nvidia's DeepStream and TensorRT. I decide to pick a more general implementation from https://github.com/sgrvinod/a-PyTorch-Tutorial-to-Object-Detection.git .

Can we do something like https://github.com/pytorch/FBGEMM/blob/0e24712210b44a3adf3832f9f9bfb1e486d81f4f/fbgemm_gpu/setup.py#L50 when building the torchaudio binary? fwiw, pytorch nightly is built with `GLIBCXX_USE_CXX11_ABI=0`. See more discussion at https://github.com/pytorch/pytorch/pull/100262#issuecomment-1542140819

> > This line was covering that, but I think it's no longer working I guess. > > https://github.com/pytorch/audio/blob/4463fbdfbbc29fbc78d5dcd4f61cd9d0a806432c/CMakeLists.txt#L130-L139 Probably. torchtext also does this, https://github.com/pytorch/text/blob/b0ebddc648d279826089db91775375221777a2db/tools/setup_helpers/extension.py#LL25C37-L25C37

> Compression ratio seems pretty bad for small batch sizes. Maybe it's because of fallback. Cc @ngimel We care about the relative difference here for catching regressions.

> Maybe we could add a filter based on absolute memory usage? Is it just the smaller models that are causing noise? We don't really know. One thing we can...

@pytorchbot merge

@pytorchbot merge -f "macos test failure is not related"

When this PR is ready, you can ping me to do another round of ghimport.

@pytorchbot rebase

@pytorchbot rebase