Skip to content

Commit d84e78c

Browse files
committed
documentation
1 parent 34ffc00 commit d84e78c

1 file changed

Lines changed: 23 additions & 2 deletions

File tree

scoring/utils/slurm/README.md

Lines changed: 23 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,26 @@
11
This folder contains a SLURM batch script that can be used to run jobs where each job corresponds to a training run on a given workload, training algorithm, random seed and tuning trial (if on external tuning ruleset).
22

33
To launch jobs:
4-
1) Generate a job config
5-
`python make_job_config.py`
4+
1) Generate a job config. The following command will generate a config.json.
5+
```
6+
python3 make_job_config.py \
7+
--submission_path <submission_path> \
8+
--tuning_search_space <tuning_search_space> \
9+
--experiment_dir $HOME/experiments/<algorithm> \
10+
--framework <jax|pytorch>
11+
```
12+
2) Save the config.json in the same directory you will run the sbatch script from.
13+
3) Check the sbatch script `run_jobs.sh`.
14+
- Set the task range to the number of tasks in the config.
15+
```
16+
#SBATCH --array=0-119
17+
```
18+
- Set the output and error logs directory for the SLURM logs.
19+
```
20+
#SBATCH --output=experiments/<tuning_ruleset>/<algorithm>/job_%A_%a.out
21+
#SBATCH --error=experiments/<tuning_ruleset>/<algorithm>/job_%A_%a.err
22+
```
23+
4) Submit a SLURM batch job by running:
24+
```
25+
sbatch run_jobs.sh
26+
```

0 commit comments

Comments
 (0)