scripts/slurm/run_external_vla_baseline.sbatch

#!/bin/bash
#SBATCH --job-name=dovla_ext_vla
#SBATCH --account=def-yalda_gpu
#SBATCH --nodes=1
#SBATCH --ntasks=1
#SBATCH --cpus-per-task=8
#SBATCH --gres=gpu:nvidia_h100_80gb_hbm3_1g.10gb:1
#SBATCH --mem=40G
#SBATCH --time=04:00:00
#SBATCH --output=outputs/hpc/logs/%x_%j.out
#SBATCH --error=outputs/hpc/logs/%x_%j.err

set -euo pipefail

PROJECT_DIR="${PROJECT_DIR:-$SLURM_SUBMIT_DIR}"
MODEL_FAMILY="${MODEL_FAMILY:-smolvla}"
DATASET="${DATASET:?Set DATASET to the held-out DoVLA-CIL dataset}"
OUT="${OUT:?Set OUT to a run directory}"
CHECKPOINT="${CHECKPOINT:-}"
ADAPTER_ENTRYPOINT="${ADAPTER_ENTRYPOINT:-}"
ADAPTER_CONFIG="${ADAPTER_CONFIG:-}"
PYTHON="${PYTHON:-python}"
DRY_RUN="${DRY_RUN:-0}"

cd "$PROJECT_DIR"
mkdir -p "$OUT" outputs/hpc/logs

ARGS=(
  scripts/run_external_vla_baseline.py
  --model-family "$MODEL_FAMILY"
  --dataset "$DATASET"
  --out "$OUT"
  --python "$PYTHON"
)

if [[ -n "$CHECKPOINT" ]]; then
  ARGS+=(--checkpoint "$CHECKPOINT")
fi
if [[ -n "$ADAPTER_ENTRYPOINT" ]]; then
  ARGS+=(--adapter-entrypoint "$ADAPTER_ENTRYPOINT")
fi
if [[ -n "$ADAPTER_CONFIG" ]]; then
  ARGS+=(--adapter-config "$ADAPTER_CONFIG")
fi
if [[ "$DRY_RUN" == "1" ]]; then
  ARGS+=(--dry-run)
else
  ARGS+=(--require-ready)
fi

"$PYTHON" "${ARGS[@]}"