Troubleshooting
If you encounter any issues not covered here, please report them in the issues section of this repository and we will work to address them as quickly as possible
Installation via pip
For certain hardware configurations, we have received notice
that installation via conda triggers incompatibilities between
numba and llvmlite that will cause installation to fail. If you encounter this issue, we reccomend the following installation
procedure.
As usual, create a conda environment and activate it:
>conda create -n simDRIFT python=3.8
>conda activate simDRIFT
Then, install numba via pip
(simDRIFT) >pip install numba==0.56.0
After numba has been installed, please download and install the appropriate NVIDIA Drivers . Afer the driver installation is complete, we will test the numba install to confirm everything is working. Launch a Python session
(simDRIFT) >python
Now, type the following commands. If the installation is correct (in the sense that numba can send data to the GPU), then the output should look something like this:
>>> import numba
>>> from numba import cuda
>>> print(cuda.to_device([1])
<numba.cuda.cudadrv.devicearray.DeviceNDArray object at ....>
>>> exit()
After this step, installation proceeds as usual. In particular, please install the appropriate version of pytorch (shown below using the syntax for our lab’s GPUs; however, the specific syntax will vary based on your CUDA Toolkit version) :
(simDRIFT) >conda install pytorch==2.0.0 torchvision==0.15.0 torchaudio==2.0.0 pytorch-cuda=11.7 -c pytorch -c nvidia
Clone this repository and install simDRIFT:
(simDRIFT) >git clone https://github.com/jacobblum/simDRIFT.git
(simDRIFT) >pip install -e simDRIFT
Finally, to confirm that everything is working as expected, run the test suite:
(simDRIFT) >simDRIFT run_tests