LibTorch Support for cuDNN6?

Hi -

I’m attempting to use TorchScript/LibTorch to deploy a PyTorch model in C++. Unfortunately, I’m constrained to an environment with CUDA 8.0 and cuDNN 6.0.21.

In trying to run an example in this environment, I get an error when running cmake -DCMAKE_PREFIX_PATH=/path/to/libtorch ..:, using the Preview version of LibTorch for CUDA 8.0 on Linux.

[ec2-user@ip-172-31-32-107 build]$ cmake -DCMAKE_PREFIX_PATH=/home/ec2-user/libtorch ..
-- The C compiler identification is GNU 4.8.5
-- The CXX compiler identification is GNU 4.8.5
-- Check for working C compiler: /usr/bin/cc
-- Check for working C compiler: /usr/bin/cc -- works
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Detecting C compile features
-- Detecting C compile features - done
-- Check for working CXX compiler: /usr/bin/c++
-- Check for working CXX compiler: /usr/bin/c++ -- works
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Detecting CXX compile features
-- Detecting CXX compile features - done
-- Looking for pthread.h
-- Looking for pthread.h - found
-- Looking for pthread_create
-- Looking for pthread_create - not found
-- Looking for pthread_create in pthreads
-- Looking for pthread_create in pthreads - not found
-- Looking for pthread_create in pthread
-- Looking for pthread_create in pthread - found
-- Found Threads: TRUE
-- Found CUDA: /usr/local/cuda (found version "8.0")
-- Caffe2: CUDA detected: 8.0
-- Caffe2: CUDA nvcc is: /usr/local/cuda/bin/nvcc
-- Caffe2: CUDA toolkit directory: /usr/local/cuda
-- Caffe2: Header version is: 8.0
-- Found CUDNN: /usr/local/cuda/include
-- Found cuDNN: v6.0.21  (include: /usr/local/cuda/include, library: /usr/local/cuda/lib64/libcudnn.so)
CMake Error at /home/ec2-user/libtorch/share/cmake/Caffe2/public/cuda.cmake:159 (message):
  PyTorch requires cuDNN 7 and above.
Call Stack (most recent call first):
  /home/ec2-user/libtorch/share/cmake/Caffe2/Caffe2Config.cmake:88 (include)
  /home/ec2-user/libtorch/share/cmake/Torch/TorchConfig.cmake:40 (find_package)
  CMakeLists.txt:4 (find_package)


-- Configuring incomplete, errors occurred!
See also "/home/ec2-user/example_app/build/CMakeFiles/CMakeOutput.log".
See also "/home/ec2-user/example_app/build/CMakeFiles/CMakeError.log".

This seems clear enough: LibTorch requires cuDNN7 or higher. But given that I’m unable to upgrade my deployment environment, I’d like to know:

  1. Is there any workaround that might allow me to get LibTorch working on cuDNN6?
  2. If no to (1), are there any plans to support cuDNN6 in the CUDA 8.0 version of LibTorch?

Thanks!