FastAI Setup High Sierra with CUDA

Part 1 (2019)
1

I recently ran into a weird error, after compiling Pytorch 1.1 with CUDA 10.0 from source:

    ~/anaconda3/lib/python3.7/site-packages/pynvml.py in _LoadNvmlLibrary()
    644                         nvmlLib = CDLL("libnvidia-ml.so.1")
    645                 except OSError as ose:
--> 646                     _nvmlCheckReturn(NVML_ERROR_LIBRARY_NOT_FOUND)
    647                 if (nvmlLib == None):
    648                     _nvmlCheckReturn(NVML_ERROR_LIBRARY_NOT_FOUND)

~/anaconda3/lib/python3.7/site-packages/pynvml.py in _nvmlCheckReturn(ret)
    308 def _nvmlCheckReturn(ret):
    309     if (ret != NVML_SUCCESS):
--> 310         raise NVMLError(ret)
    311     return ret
    312 

NVMLError_LibraryNotFound: NVML Shared Library Not Found

Iā€™m running Mac OS High Sierra with the most recent NVIDIA Web Drivers, CUDA 10.0 and CUDNN 7.4.2
Does anyone know, what package I am missing to obtain libnvidia-ml.so.1?

(base) āžœ  ~ clang -v
Apple LLVM version 9.0.0 (clang-900.0.39.2)
Target: x86_64-apple-darwin17.0.0
Thread model: posix
InstalledDir: /Library/Developer/CommandLineTools/usr/bin
Found CUDA installation: /usr/local/cuda, version unknown
(base) āžœ  ~ nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2018 NVIDIA Corporation
Built on Sat_Aug_25_21:08:56_CDT_2018
Cuda compilation tools, release 10.0, V10.0.130
(base) āžœ  ~ python --version
Python 3.7.1
(base) āžœ  ~ pip freeze | grep -e fastai -e torch
fastai==1.0.45
torch==1.1.0a0+fbd690c
torchvision==0.2.1
(base) āžœ  ~ conda --version
conda 4.6.3

Thanks in advance!

2
3

I ma having the same problem but in windows 10

1 Like
4

Did you install the most recent GPU drivers for Windows? nvidia-smi should be bundled with them.