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