Buckets:

glennmatlin's picture
download
raw
1.25 kB
#!/usr/bin/env bash
#SBATCH --cpus-per-task=8
#SBATCH --mem=32G
#SBATCH --time=12:00:00
#SBATCH --partition=cpu-small
#SBATCH --array=0-0
#SBATCH --output=logs/dolma_eda/%A_%a.out
#SBATCH --error=logs/dolma_eda/%A_%a.err
set -euo pipefail
mkdir -p logs/dolma_eda
PROJECT_ROOT=${PROJECT_ROOT:-${SLURM_SUBMIT_DIR:-$PWD}}
cd "$PROJECT_ROOT"
PYTHONEXEC=${PYTHONEXEC:-.venv/bin/python}
PYTHONPATH=${PYTHONPATH:-$PROJECT_ROOT/src}
export PYTHONPATH
MANIFEST=${MANIFEST:?}
OUTDIR=${OUTDIR:-runs/dolma_enriched/eda}
WORKERS_DIR=${WORKERS_DIR:-$OUTDIR/workers}
VALUE_FIELDS=${VALUE_FIELDS:-"lang warc_content_type cc_dump"}
TOP_K=${TOP_K:-50}
JOINT_TOP_K=${JOINT_TOP_K:-200}
HEATMAP_K=${HEATMAP_K:-25}
mkdir -p "$WORKERS_DIR"
LINE=$(sed -n "$((SLURM_ARRAY_TASK_ID+1))p" "$MANIFEST")
if [[ -z "$LINE" ]]; then
echo "No manifest entry for task $SLURM_ARRAY_TASK_ID"
exit 1
fi
"$PYTHONEXEC" -m dolma.eda.core \
--mode worker \
--manifest "$MANIFEST" \
--manifest-index "$SLURM_ARRAY_TASK_ID" \
--output-dir "$OUTDIR" \
--worker-output-dir "$WORKERS_DIR/$SLURM_ARRAY_TASK_ID" \
--value-fields $VALUE_FIELDS \
--top-k "$TOP_K" \
--joint-top-k "$JOINT_TOP_K" \
--heatmap-k "$HEATMAP_K" \
--no-report \
--no-write-png \
--resume

Xet Storage Details

Size:
1.25 kB
·
Xet hash:
6deeecb2814b83bfe88f5074cbca670e95f8667865205968c63ebd655aa666c0

Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.