scripts/slurm/eval_enhanced.sbatch

#!/bin/bash
#SBATCH --job-name=eval_enhanced
#SBATCH --nodes=1
#SBATCH --ntasks=1
#SBATCH --cpus-per-task=4
#SBATCH --gres=gpu:1
#SBATCH --mem=32000M
#SBATCH --time=4:00:00
#SBATCH --output=logs/eval_enhanced_%A_%a.out
#SBATCH --error=logs/eval_enhanced_%A_%a.err
#SBATCH --array=0-2

set -euo pipefail

PROJECT_DIR="${PROJECT_DIR:-$PWD}"
cd "$PROJECT_DIR"

source .venv/bin/activate

SEED=$SLURM_ARRAY_TASK_ID
CHECKPOINT="/scratch/$USER/dovla/experiments/cvpr_enhanced_model/seed_$SEED/best.pt"
DATASET="/scratch/$USER/dovla/experiments/maniskill_presuccess_six_task_collection"
OUT="/scratch/$USER/dovla/experiments/cvpr_enhanced_model/seed_$SEED/eval_result.json"

echo "=== Evaluating Enhanced Model Seed $SEED ==="
echo "Checkpoint: $CHECKPOINT"
echo ""

python scripts/eval_enhanced_checkpoint.py \
    --checkpoint "$CHECKPOINT" \
    --dataset "$DATASET" \
    --out "$OUT" \
    --device cuda

if [ $? -eq 0 ]; then
    echo ""
    echo "✅ Evaluation complete for seed $SEED"
    python -c "
import json
with open('$OUT') as f:
    d = json.load(f)
print(f\"Selected success: {d['selected_success_rate']:.4f}\")
print(f\"Top-1: {d['top1_action_selection']:.4f}\")
"
fi