Instructions for setting up Jupyter Notebook on Graphite.
Pre-step:
install Anaconda and set up your environment <env_name>
Step 1: On graphite,
source activate <env_name>
Step 2: On graphite,
which python
Step 3,
Get print: /home/<your_netid>/.conda/envs/<env_name>/bin/python,
We will use the file path for bin in next step: /home/<your_netid>/.conda/envs/<env_name>/bin/
Step 4: On graphite,
export PATH=/home/<your_netid>/.conda/envs/<env_name>/bin/:$PATH
Step 5: On local machine,
ssh <your_netid>@graphite-login.coecis.cornell.edu -L 127.0.0.1:1234:NODE_NAME:8888
(For Claire's student, an example here for NODE_NAME is nikola-compute05)
(If 8888 is used, change it with any number between 8000 and 10000, but note that you should also change the port number in step 7 with the number you use in this step)
Step 6: On grahite,
srun -p interactive --pty --nodelist=NODE_NAME /bin/bash
Step 7: On grahite,
XDG_RUNTIME_DIR=/tmp/<your_netid> jupyter-notebook --ip=0.0.0.0 --port=8888
Step 8:
On local browser, http://127.0.0.1:1234/?token=LONG_ALPHANUMERIC_STRING_FROM_JUPYTER-NOTEBOOK_OUTPUT
These were adapted from official suggestions by Cornell IT (https://it.coecis.cornell.edu/researchit/graphitegpu/).