Launch jobs¶
Before you begin¶
-
Start from ready-to-run minimal examples or from the Research Project Template to write code that runs on the Mila cluster.
Launch jobs¶
This section introduces the core Slurm concepts for new users and walks through running one or more tasks on the cluster, from a first interactive job to monitoring, managing and synchronizing tasks across multiple nodes.
Work from VSCode on a compute node
These guides connect to the cluster with VSCode
through mila code or the mila-cpu remote, which opens a compute node
with a file browser for $SCRATCH and an integrated terminal for the Slurm
commands. Set it up in the Get Started
guide. Every step also lists the equivalent
ssh mila terminal command as an alternative.
-
Discover Slurm jobs, steps and tasks. Run multiple tasks through an interactive job, then reproduce the example from a batch script.
-
Easily use multiple clusters to submit jobs.
-
Track jobs through the queue, inspect and cancel them, read their output, and resolve common failures.
-
An applied example showing how tasks running on different nodes can communicate and synchronize their output.
Next step¶
-
Monitor and Optimize Experiments
Understand why metrics matter, plan their collection before a job starts, and choose the right tool to optimize both models and resource usage.
-
Use the dashboard to identify and reduce wasted GPU resources.
-
Share results and make research reproducible
Make cluster experiments reproducible and shareable — environment management, version control, dataset sharing, and research paper distribution.