lerobot-eval \
  --policy.path="HuggingFaceVLA/smolvla_libero" \
  --env.type=libero \
  --env.task=libero_spatial \
  --eval.batch_size=1 \
  --eval.n_episodes=1