Fangjun Kuang

For the above link error, could you have a look at https://github.com/pytorch/pytorch/issues/33668

Can you use nvcc to compile and run the following file? (You can create a new file, e.g., `hello.cu`, from it.) ``` #include #include int main() { printf("%d.%d", CUDA_VERSION /...

> nvcc -arch sm_86 test.cu -o test && ./test 11.7 It shows 11.7, but you said you have installed cuda 11.6. Please recheck your cudatoolkit. --- Please don't use ```...

I suggest that you create a new environment to install kaldifeat as your current environment is possibly corrupted.

Could you create the following file, e.g., `activate-cuda-11.6.sh` and run ```bash source activate-cuda-11.6.sh ``` before you run `pip install --verbose kaldifeat`? ```bash #!/usr/bin/env bash export CUDA_HOME=/cfs/user/burkliu/dev/anaconda3/envs/k2_p export PATH=$CUDA_HOME/bin:$PATH export LD_LIBRARY_PATH=$CUDA_HOME/lib64:$LD_LIBRARY_PATH...

> nvcc -arch sm_86 test.cu -o test && ./test 11.6 I think I find the reason. CMake is saving it as a `.cc` file and tries to compile it with...

Are you able to save the following content to `test.cc` and compile it with ```bash g++ test.cc -o test ./test ``` ```bash #include #include int main() { printf("%d.%d", CUDA_VERSION /...

> I find cuda.h in /cfs/user/burkliu/dev/anaconda3/envs/k2_p/include In that case, could you change ```python3 export CUDA_INC_PATH=$CUDA_HOME/targets/x86_64-linux ``` to ```bash export CUDA_INC_PATH=$CUDA_HOME ``` --- For `test.cc:1:10: fatal error: cuda.h: No such file...

Are you able to run ```bash /cfs/user/burkliu/dev/anaconda3/envs/k2_p/lib/libcudart.so g++ -I $CUDA_HOME/include test.cc /cfs/user/burkliu/dev/anaconda3/envs/k2_p/lib/libcudart.so -o test && ./test ``` ```bash ldd /cfs/user/burkliu/dev/anaconda3/envs/k2_p/lib/libcudart.so ```

Your system libc version may be too low and cannot work with cuda 11.6 You can check whether your conda env has provided `libc.so.6`. If yes, you can set the...