scripts/p0b_queue.sh

#!/bin/bash
# P0-b queue: attack-strength curves for the 24 method representatives.
# Cross-author relearn (H2 protocol), lr 2e-5, 5 epochs, per-epoch saves.
# Usage: p0b_queue.sh <gpu> <start_idx> <stride>
set -u
cd /workspace/jspace-unlearning
GPU=$1; START=$2; STRIDE=$3
i=0
for name in $(python3 -c "import json;print(' '.join(json.load(open('results/p0b_reps.json'))))"); do
  if [ $((i % STRIDE)) -eq "$START" ]; then
    SRC=/workspace/models/ckpts/$name
    [ -f "$SRC/model.safetensors" ] || { echo "P0B-NO-MODEL $name"; i=$((i+1)); continue; }
    EPOCHS=5 SAVE_STEPS=7 bash scripts/p0b_relearn.sh $GPU "$SRC" "e34x_$name" 2e-05 \
      results/h2_relearn_set.json > logs/p0b_$name.log 2>&1 || echo "P0B-RUN-FAIL $name"
  fi
  i=$((i+1))
done
echo "P0B QUEUE DONE gpu=$GPU start=$START $(date +%H:%M:%S)"