scripts/slurm/phase_a3_eval_large_model.sbatch

#!/bin/bash
#SBATCH --job-name=dovla_large_eval
#SBATCH --partition=${DOVLA_PARTITION:-compute}
#SBATCH --nodes=1
#SBATCH --ntasks=1
#SBATCH --cpus-per-task=8
#SBATCH --gres=gpu:1
#SBATCH --mem=32G
#SBATCH --time=12:00:00
#SBATCH --output=logs/phase_a3_large_eval_%j.out
#SBATCH --error=logs/phase_a3_large_eval_%j.err
#SBATCH --array=0-2

set -euo pipefail

# Phase A3: Evaluate large model with lattice eval + policy rollout
# Target: Measure improvement over baseline 29.67%

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

source .venv/bin/activate

CHECKPOINT_DIR="/scratch/$USER/dovla/experiments/phase_a2_large_model/seed_$SLURM_ARRAY_TASK_ID"
DATASET="/scratch/$USER/dovla/experiments/maniskill_presuccess_six_task_collection"
OUT_DIR="$CHECKPOINT_DIR"

echo "=== Phase A3: Evaluating Large Model (seed $SLURM_ARRAY_TASK_ID) ==="
echo ""

# Lattice evaluation
echo "Running lattice evaluation..."
python scripts/eval_lattice_checkpoint.py \
    --checkpoint "$CHECKPOINT_DIR/best.pt" \
    --dataset "$DATASET" \
    --out "$OUT_DIR/lattice_eval.json" \
    --mode field_only \
    --all-groups

echo ""
echo "Running policy rollout..."
python scripts/eval_maniskill_policy_rollout.py \
    --checkpoint "$CHECKPOINT_DIR/best.pt" \
    --dataset "$DATASET" \
    --out "$OUT_DIR/policy_rollout.json" \
    --num-groups 700 \
    --mode validation

echo ""
echo "✅ Phase A3 complete: Evaluation done (seed $SLURM_ARRAY_TASK_ID)"