train: add slurm job file

This commit is contained in:
Starbeamrainbowlabs 2022-10-12 17:27:10 +01:00
parent 6423bf6702
commit dbe4fb0eab
Signed by: sbrl
GPG key ID: 1BE5172E637709C2

72
aimodel/slurm-train.job Executable file
View file

@ -0,0 +1,72 @@
#!/usr/bin/env bash
#SBATCH -J RainAIv3
#SBATCH -N 1
#SBATCH -n 14
#SBATCH --gres=gpu:1
#SBATCH -o %j.%N.%a.out.log
#SBATCH -e %j.%N.%a.err.log
#SBATCH -p gpu05
#SBATCH --time=5-00:00:00
#SBATCH --mem=61440
# 61440 = 60GiB memory required
module load utilities/multi
module load readline/7.0
module load gcc/10.2.0
module load cuda/11.5.0
module load python/anaconda/4.6/miniconda/3.7
show_help() {
echo -e "Usage:" >&2;
echo -e " [INPUT='\$HOME/rainfallwater_records_tfrecord'] [POSTFIX='<string>'] [ARGS='<extra-args>'] sbatch slurm-train.job" >&2;
echo -e "" >&2;
echo -e "....where:" >&2;
echo -e " INPUT The path to the directory containing the .tfrecord files to use as training data (see the rainfallwrangler for making these)" >&2;
echo -e " POSTFIX Optional. A suffix to apply to the run code name." >&2;
echo -e "" >&2;
echo -e "The code used to identify the run is taken automatically from the filename of the config file." >&2;
exit;
}
INPUT="${INPUT:-$HOME/PhD-Rainfall-Radar/aimodel/output/output/rainfallwater_records_embed_2022-10-06_contrast_embed_umap_d512e19_tfrecord}";
if [[ -z "${INPUT}" ]]; then
echo -e "Error: No INPUT environment variable specified.\n" >&2;
show_help;
exit 0;
fi
if [[ ! -d "${INPUT}" ]]; then
echo -e "Error: The directory '${INPUT}' containing the input .tfrecord dataset either doesn't exist or isn't a directory.";
show_help;
exit 1;
fi
CODE="train_segmenter";
if [[ -n "${POSTFIX}" ]]; then
echo -e ">>> Applying postfix of ${POSTFIX}" >&2;
CODE="${CODE}_${POSTFIX}";
fi
echo -e ">>> Input dirpath: ${INPUT}" >&2;
echo -e ">>> Code: ${CODE}" >&2;
echo -e ">>> Additional args: ${ARGS}";
dir_output="output/$(date -u --rfc-3339=date)_${CODE}";
export PATH=$HOME/software/bin:$PATH;
echo ">>> Installing requirements";
conda run -n py38 pip install -r requirements.txt;
echo ">>> Training model";
#shellcheck disable=SC2086
/usr/bin/env time -v conda run -n py38 src/index.py pretrain --input "${INPUT}" --output "${dir_output}" ${ARGS};
src/index.py train --input "${INPUT}" -o "${dir_output}"
# Recommended extra args: --water-size 94 --batch-size 48 --arch convnext_i_xtiny
# Example used for raw testing:
# src/index.py train --input output/rainfallwater_records_embed_2022-10-06_contrast_embed_umap_d512e19_tfrecord_TINY -o output/2022-09-28-segmenter-d512e19-TEST-tiny --water-size 94 --batch-size 48 --arch convnext_i_xtiny
echo ">>> exited with code $?";