Job scripts for running EvoCode on Nibi.

Run from the repo root — setup.sh resolves paths relative to its own parent directory:

ssh nibi.alliancecan.ca
cd $PROJECT/EvoCode

bash scripts/setup.sh    # loads python/3.12 + java/21, installs evolver/requirements.txt

Then add your API key. The .env file must live in evolver/, since the run starts from there:

cp .env.example evolver/.env
nano evolver/.env        # add GEMINI_API_KEY

Submit from the repo root — the job script does cd "$SLURM_SUBMIT_DIR/evolver" before running:

mkdir -p slurm_logs      # SBATCH --output needs this to exist at submit time

# Single run (uses configs/default.json)
sbatch scripts/run_evolution.sh

# With a specific config — path is relative to evolver/
sbatch scripts/run_evolution.sh configs/quick_test.json

# Check status
squeue -u $USER

# Cancel
scancel <job_id>

Evolution parameters live in the JSON config you pass to run_evolution.sh — see evolver/configs/ and the Configuration section of the main README.

Scheduler settings are the SBATCH directives at the top of run_evolution.sh:

SLURM reads #SBATCH lines before the shell runs, so they do not expand variables. Set --account to a literal value rather than relying on $SLURM_ACCOUNT.

Nibi has full internet access on all nodes. No proxy configuration is needed for API calls.