Creating an Interactive Session for Debugging & Testing

Sometimes, submitting a batch script and waiting for the output isn't enough. If you need to debug your code in real-time—for example, using a breakpoint() in Python, you should use an Interactive Session.

An interactive session grants you direct access to a compute node's terminal, allowing you to run commands and see results instantly, including direct access to the GPUs.

How to Launch an Interactive Session

To request a node for 30 minutes on a 40GB NVIDIA A100 GPU partition, use the srun command with the --pty bash flag:

srun --time=00:30:00 --partition=normal-a100-40 --account=f2025hpcvlab00005ubig --gpus=1 --pty bash

Breaking Down the Command:

Flag	Purpose
--time=00:30:00	Sets a 30-minute limit for your session
--partition=...	Specifies the hardware queue (in this case, the A100 GPUs)
--account=...	Assigns the resource usage to your specific project account
--nodes=1	Requests exactly one compute node
--pty bash	Opens an interactive bash terminal on the allocated node

Verifying Your Access

Once your session starts, your terminal prompt will change (e.g., from login to something like gnx506). To confirm that you have successfully landed on a GPU node and that the hardware is available, run:

nvidia-smi

If successful, you will see a status table showing the A100-40GB GPU details. You are now ready to run your interactive tests!

Tip: Remember to type exit or press Ctrl+D to end your session once you're finished. This frees up the GPU for other researchers and saves your project's allocated hours.