/maguro

A simple job scheduler for GPUs

Primary LanguagePython

maguro

maguro is a simple job scheduler for multiple-GPU enviroments.

Scheduling is not the optimal, but you do not need run scripts manually after you register (push) tasks.

Requirements

  • Python >= 3.7

This library relies only on the standard libraries and nvidia-smi.

Installation

pip install -U git+https://github.com/moskomule/maguro

Usage

Submit job

maguro push [-r,--num_repeat 1] [-g,--num_gpus 1] [--logdir maglog] COMMAND

Here, COMMAND is like python ~/train.py. --num_gpus specifies the number of GPUs required to run the submitted job.

Note that paths in COMMAND should be absolute if you run from different directory

Run jobs

maguro run [--forever]

List remaining jobs

maguro list [--all]

Delete jobs

maguro delete IDS

This IDS can be checked by maguro list --all.