compile against latest pytorch, failed

[northeastsquare]

src/_warpctc.cpp:99:71: error: ‘struct at::Type’ has no member named ‘tensor’ at::Tensor workspace = activations.type().toScalarType(at::kByte).tensor(workspace_size)

[t-vi]

Thanks for reporting! I'll fix this after 1.0 comes out soon. Please note that I recommend to use PyTorch's native CTC Loss instead, I'll put up a wrapper providing compatibility in terms of output scaling.

[northeastsquare]

@t-vi I use pytorch ctc loss, But back prop grad with many nan

[simonwang517]

@t-vi compile pytorch1.3 failed. src/_warpctc.cpp:10:34: fatal error：ATen/cuda/CUDAStream.h：No such file or directory Any Idea?

[t-vi]

Basically, don't use this library.

As of today, PyTorch (master) comes with its own implementation of the CTC loss. It has CuDNN support and fast CUDA and reasonably fast CPU backend. I recommend using that going forward.