#!/bin/bash #SBATCH --job-name=eval_a4_all #SBATCH --nodes=1 #SBATCH --ntasks=1 #SBATCH --cpus-per-task=4 #SBATCH --gres=gpu:1 #SBATCH --mem=32000M #SBATCH --time=8:00:00 #SBATCH --output=logs/eval_phase_a4_%j.out #SBATCH --error=logs/eval_phase_a4_%j.err set -euo pipefail PROJECT_DIR="${PROJECT_DIR:-$PWD}" cd "$PROJECT_DIR" source .venv/bin/activate CONFIG_DIR="/scratch/$USER/dovla/experiments/phase_a4_hparam_sweep" DATASET="/scratch/$USER/dovla/experiments/maniskill_presuccess_six_task_collection" echo "=== Evaluating Phase A4 All Configs (GPU) ===" echo "Config dir: $CONFIG_DIR" echo "Dataset: $DATASET" echo "" for config_dir in "$CONFIG_DIR"/*/; do config_name=$(basename "$config_dir") checkpoint="$config_dir/best.pt" out="$config_dir/lattice_eval.json" if [ ! -f "$checkpoint" ]; then echo "⚠️ Skipping $config_name (no checkpoint)" continue fi if [ -f "$out" ]; then echo "✓ $config_name already evaluated" continue fi echo "Evaluating $config_name..." python scripts/eval_lattice_checkpoint.py \ --checkpoint "$checkpoint" \ --dataset "$DATASET" \ --out "$out" \ --all-groups \ --device cuda if [ $? -eq 0 ]; then echo " ✅ Complete" else echo " ❌ Failed" fi done echo "" echo "✅ Phase A4 evaluation complete!"