...
Code Block | ||||||
---|---|---|---|---|---|---|
| ||||||
[aa3301@axon ~]$ ml add cuda/10.1.168 cudnn/7.5.1-10.0 gcc/4.9.1 load matlab/2018b show unuse anaconda3-2019.03 cuda/9.0.176 cudnn/7.6.1.34-10.0 help matlab purge sl update avail cuda/9.2.88 cudnn/7.6.1.34-10.1 keyword matlab/2017a restore spider use cuda cudnn delete kilosort2 matlab/2017b rm swap whatis cuda/10.0.130 cudnn/7.3.0 gcc list matlab/2018a save unload [aa3301@axon ~]$ ml load cud cuda cuda/10.1.168 cuda/9.2.88 cudnn/7.3.0 cudnn/7.6.1.34-10.0 cuda/10.0.130 cuda/9.0.176 cudnn cudnn/7.5.1-10.0 cudnn/7.6.1.34-10.1 |
Conda Environments
We strongly suggest the use of conda environments for jobs on Axon where the software you need either isn't installed or isn't the right version. Even CUDA and CUDNN versions can be installed in a conda environment.
Here's are the steps for setting up an example conda environment that was set up for a ZI researcher:
Code Block |
---|
# It's generally advisable to perform installations in an interactive session (https://confluence.columbia.edu/confluence/display/zmbbi/Interactive+Sessions)
# on one of the nodes, since there are memory limitations on the login node that can occasionally
# lead to out-of-memory errors during the installation process.
# In this case I am requesting 2 CPUs on node ax08, 1 GPU, for 1 hour, and to use a bash interactive shell
srun --pty -c 2 --nodelist=ax08 --gres=gpu:1 -t 0-01:00 /bin/bash
# Next I activate conda by loading the module for it
ml load anaconda3-2019.03
# next I create an environment. Note that I specify a version of python that I want
conda create -n demo-env python=3.6
# once the environment finishes building I need to activate it
conda activate demo-env
# then I can install software that is available in the conda repos
# N.B. the -c flag will specify a "channel" or specific repo
conda install pytorch torchvision torchaudio cudatoolkit=10.2 -c pytorch
conda install -c conda-forge tensorboard
# you can also do pip installs for python requirements which will be local to this environment
pip install gym |
Once you have set up your conda environment you can use it in batch scripts by including the following lines in the beginning of your script:
Code Block |
---|
ml load anaconda3-2019.03
conda activate demo-env |