Buckets:
| # Autonomous on-cluster driver for the Comma-2T ToMBench n=72 experiment. | |
| # Runs as its own SLURM job so the pipeline completes with NO local session / | |
| # SSH involvement. Each tick: recover failed training cells, submit evals for | |
| # complete adapters (both idempotent), report. When all 144 eval JSONs are in | |
| # and nothing is live, run the paired analysis and exit. | |
| #SBATCH --job-name=c2ttom_driver | |
| #SBATCH --partition=ice-cpu | |
| #SBATCH --qos=coe-ice | |
| #SBATCH --cpus-per-task=2 | |
| #SBATCH --mem=8G | |
| #SBATCH --time=12:00:00 | |
| #SBATCH --output=/storage/ice-shared/cs7634/staff/TDA/logs/comma_unlearn/%x_%j.out | |
| #SBATCH --error=/storage/ice-shared/cs7634/staff/TDA/logs/comma_unlearn/%x_%j.err | |
| set -uo pipefail | |
| REPO_DIR="$HOME/dev/data-attribution-comma-attr" | |
| TDA=/storage/ice-shared/cs7634/staff/TDA | |
| TOM=$TDA/comma/tom_unlearning | |
| EVAL=$TOM/eval | |
| cd "$REPO_DIR" | |
| export PATH="$REPO_DIR/.venv/bin:$PATH" | |
| mkdir -p "$EVAL" | |
| echo "driver start $(date) node=$(hostname)" | |
| for i in $(seq 1 120); do | |
| bash "$TOM/submit_tom_train.sh" >> "$TOM/driver.log" 2>&1 || true | |
| bash "$TOM/submit_tom_evals.sh" >> "$TOM/driver.log" 2>&1 || true | |
| ntr=$(squeue -u "$USER" -h -o '%j' 2>/dev/null | grep -c c2ttomtr || true) | |
| nev=$(squeue -u "$USER" -h -o '%j' 2>/dev/null | grep -c c2ttom_eval || true) | |
| nj=$(ls "$EVAL"/*_tombench_eval.json 2>/dev/null | wc -l) | |
| echo "driver tick $i: train_left=$ntr eval_live=$nev json=$nj/144 $(date)" | |
| if [ "$ntr" = "0" ] && [ "$nev" = "0" ] && [ "$nj" -ge 144 ]; then | |
| echo "driver: all 144 evals present, running paired analysis" | |
| .venv/bin/python "$TOM/analyze_tom_unlearning.py" \ | |
| --eval-dir "$EVAL" --out-csv "$TOM/tom_paired.csv" \ | |
| > "$TOM/tom_analysis.txt" 2>&1 | |
| echo "driver: DONE -> $TOM/tom_analysis.txt $(date)" | |
| exit 0 | |
| fi | |
| sleep 300 | |
| done | |
| echo "driver: TIMEOUT after 120 ticks $(date)" | |
| .venv/bin/python "$TOM/analyze_tom_unlearning.py" \ | |
| --eval-dir "$EVAL" --out-csv "$TOM/tom_paired.csv" \ | |
| > "$TOM/tom_analysis_partial.txt" 2>&1 || true | |
| exit 1 | |
Xet Storage Details
- Size:
- 2.06 kB
- Xet hash:
- d16782ceb8d4ca60c0af4e3bacf322be1303feeab77bf4bd05a214b5d9592bff
·
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.