diff --git a/README.md b/README.md index e30c74cd81bc2af9e5b87929e15df6af8e834263..c4343303bd4ecff7bec75537a6f340f9951b1365 100644 --- a/README.md +++ b/README.md @@ -178,6 +178,27 @@ pass your desired arguments to the Python script to `sbatch`, like so: sbatch run.sbatch --train-num-workers=8 --valid-num-workers=3 ``` +### Interactive use + +After the environment has been created by executing `bash set_up.sh` +once on a login node, it can also be used interactively by activating +it. This activation has to be done in each new terminal session; it +does not persist between login sessions: + +```shell +source activate.sh +# Now do whatever you want, for example use `python` interactively. +``` + +Keep in mind that you should not be doing heavy work on login nodes. +Additionally, some libraries that require certain hardware such as +GPUs may not work on a login node. However, you can also create an +interactive session on a compute node using `srun --pty` + +You can deactivate the Python environment using `deactivate`, however, +the previous module environment from the module system will not be +restored. + ### Warnings upon PyTorch Distributed initialization You can safely ignore warnings like the following: