search

thomasbrandon / mish-cuda

Mish Activation Function for PyTorch
MIT License
147 stars 67 forks source link

Mish-cuda installation failed. #4

Open yedmed opened 3 years ago

yedmed commented 3 years ago

Hi, i am trying to install mish-cuda using "pip install git+https://github.com/thomasbrandon/mish-cuda/ However i am getting ERROR: Failed building wheel for mish-cuda. I am using cuda 10.1 with pytorch 1.5. Any suggestion? Thanks

thomasbrandon commented 3 years ago

Without a specific error message it's hard to know. Make sure the CUDA SDK is installed and accessible through the CUDA_HOME environment variable. Otherwise you'd need to get a better error. You could try running pip verbosely (pip -v ...) and see if that gives full messages. Otherwise clone the repo and try directly running python setup.py -v install (or maybe you might need to specifically run python setup.py -v build or python setup.py -v build_ext).-

CzKerwin commented 3 years ago

I get same problem using cuda 11.0 with pytorch 1.7. I try running python setup.py -v build.

FAILED: /app/share/mish-cuda/build/temp.linux-x86_64-3.8/csrc/mish_kernel.o /usr/local/cuda/bin/nvcc -Iexternal -I/opt/conda/lib/python3.8/site-packages/torch/include -I/opt/conda/lib/python3.8/site-packages/torch/include/torch/csrc/api/include -I/opt/conda/lib/python3.8/site-packages/torch/include/TH -I/opt/conda/lib/python3.8/site-packages/torch/include/THC -I/usr/local/cuda/include -I/opt/conda/include/python3.8 -c -c /app/share/mish-cuda/csrc/mish_kernel.cu -o /app/share/mish-cuda/build/temp.linux-x86_64-3.8/csrc/mish_kernel.o -DCUDA_NO_HALF_OPERATORS -DCUDA_NO_HALF_CONVERSIONS -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr --compiler-options ''"'"'-fPIC'"'"'' --expt-extended-lambda -DTORCH_API_INCLUDE_EXTENSION_H -DTORCH_EXTENSION_NAME=_C -D_GLIBCXX_USE_CXX11_ABI=0 -gencode=arch=compute_86,code=sm_86 -std=c++14 nvcc fatal : Unsupported gpu architecture 'compute_86'

thomasbrandon commented 3 years ago

CUDA 11.0 doesn't support RTX30 series cards. They were introduced in CUDA 11.1.

CzKerwin commented 3 years ago

CUDA 11.0 doesn't support RTX30 series cards. They were introduced in CUDA 11.1.

Thank you very much for clearing my doubt.