Jake Lee

@mxnet-label-bot update [pr-work-in-progress]

@SidneyLann we also have conv1d, conv2d, conv3d. Modern frameworks like TF, PT doesn't have conv4d. Can you elaborate your use case? Thanks

@ebamberg I just assigned it to you

@thomelane @ThomasDelteil could you review it? Thanks!

@androuino it looks like you use the operator that doesn't support float64 data type. Do you know where you use float64?

@androuino Does you have complete error stack trace? The label of the Pikachu dataset is float, but the label in CocoDetection is double. I am suspecting this is where you...

"-m 1" a.k.a "max-batches" means we train the model with only 1 max-batches for each epoch, which is usually for sanity test.

I checked it out. You can try to build JNI from source and link jni with torch sparse package. The CMakeFile is [here](https://github.com/awslabs/djl/blob/master/engines/pytorch/pytorch-native/CMakeLists.txt). ``` find_package(TorchSparse REQUIRED) target_link_libraries(lmp "${TorchSparse_LIBRARIES}") ``` then...

I can reproduce the issue and found both PyTorch and MXNet have the same problem. This could be caused by gradient vanishing. Only the first batchNorm have the issue. need...

I think I can improve this when I refactor the initializer