README.md

srunit

Better CLI for your SLURM cluster...

Installation

1. Clone Project

git clone https://github.com/21jun/srunit.git

2. Install with pip install -e .

How to use it?

0. configure .srunit/xxx.conf, .srunit/xxx.sbatch for purpose

   All variable written in xxx.sbatch script must be specified in xxx.conf

   except following variables (these variables are automatically generated)

   JOB_NAME,
   OUTPUT_PATH,
   DATETIME,
   RUN_PATH,
   RUN_SCRIPT_PATH,
   ADDITIONAL_ARGS,
   EXP_ROOT,
   EXP_ROOT.stem,
   

1. place your shell script under experiments directory

2. run srunit with -f flag to specify your shell script

srunit -t A100-80GB -n 4 -f exp/v1/el/finetune_template.sh

tip. you can use -d flag to make dry run (create sbatch scirpt only)

srunit -t A100 -n 4 -f exp/v1/el/finetune_template.sh -d

output log:

Load template: .srunit/cluster.sbatch
Load config: .srunit/1.conf
{'JOB_NAME': 'sl_0', 'OUTPUT_PATH': PosixPath('tasks/train/asr/finetune/commonvoice12/exp/phoneme-CTC-cv-mono-phoneme-level-ascii/recipes/sl/runs/run_0/out'), 'GPU_TYPE': '3090', 'GPU_NUM': 8, 'CPU_NUM': 16, 'DRY_RUN': False, 'SCRIPT_PATH': 'tasks/train/asr/finetune/commonvoice12/exp/phoneme-CTC-cv-mono-phoneme-level-ascii/recipes/sl/finetune_asr.sh', 'DATETIME': '20230405:183810', 'RUN_PATH': PosixPath('tasks/train/asr/finetune/commonvoice12/exp/phoneme-CTC-cv-mono-phoneme-level-ascii/recipes/sl/runs/run_0'), 'RUN_SCRIPT_PATH': PosixPath('tasks/train/asr/finetune/commonvoice12/exp/phoneme-CTC-cv-mono-phoneme-level-ascii/recipes/sl/runs/run_0/finetune.sh'), 'ADDITIONAL_ARGS': '', 'EXP_ROOT': PosixPath('tasks/train/asr/finetune/commonvoice12/exp/phoneme-CTC-cv-mono-phoneme-level-ascii/recipes/sl'), 'EXP_NAME': 'sl'}
Submitting the job...
Job name: sl_0
Submitted batch job 20926

3. check your sbatch script and output files under runs dir

run_x : x is incremental number