How to run in SLURM

Note

This is a quick how-to on SLURM. See here for a long-form tutorial.

See the default SLURM options, used when slurm=True

import spikewrap as sw
import json

default_arguments = sw.default_slurm_options()

print(
    json.dumps(default_arguments, indent=4)  # json just for visualising output
)

{
    "nodes": 1,
    "mem_gb": 40,
    "timeout_min": 1440,
    "cpus_per_task": 8,
    "tasks_per_node": 1,
    "wait": false,
    "env_name": "spikewrap",
    "slurm_partition": "cpu"
}

Otherwise, we can update these as desired:

gpu_arguments = sw.default_slurm_options("gpu")

gpu_arguments["mem_gb"] = 60
gpu_arguments["env_name"] = "my_conda_environment"

print(
    json.dumps(gpu_arguments, indent=4)
)

# and then use like:
# session.save_preprocessed(n_jobs=12, slurm=gpu_arguments)

{
    "nodes": 1,
    "mem_gb": 60,
    "timeout_min": 1440,
    "cpus_per_task": 8,
    "tasks_per_node": 1,
    "wait": false,
    "env_name": "my_conda_environment",
    "slurm_partition": "gpu",
    "slurm_gres": "gpu:1",
    "exclude": "gpu-sr670-20,gpu-sr670-21,gpu-sr670-22"
}