VERSION="v0"
TASK="10" # spatial / object / goal / 10 / 90
VLA_PATH="checkpoints/initialized_pt_vla/initailized_openvla_with_SF_spatial_v0.4.2"
DATA_ROOT_DIR="data/libero_openvla"
RUN_ROOT_DIR="experiments/training_results"
REGULARIZATION_LORA_VECTOR_PATH="checkpoints/lora_diff/sf_150000_steps_spatial_adapter_diff.safetensors"
WANDB_ENTITY="YOUR_WANDB_ENTITY"
WANDB_PROJECT="YOUR_WANDB_PROJECT"
EVAL_LOG_PATH="experiments/eval_logs/${VERSION}_output.log"

torchrun --standalone --nnodes 1 --nproc-per-node 1 vla-scripts/finetune_regular_loss.py \
  --vla_path "$VLA_PATH" \
  --data_root_dir "$DATA_ROOT_DIR" \
  --dataset_name libero_${TASK}_no_noops \
  --run_root_dir "$RUN_ROOT_DIR" \
  --use_l1_regression True \
  --use_diffusion False \
  --use_film False \
  --num_images_in_input 2 \
  --use_proprio True \
  --batch_size 8 \
  --learning_rate 5e-4 \
  --scheduler CosineAnnealingLR \
  --max_steps 150100 \
  --save_freq 150000 \
  --save_latest_checkpoint_only True \
  --merge_lora_during_training True \
  --regularization_lora_vector_path "$REGULARIZATION_LORA_VECTOR_PATH" \
  --regularization_weight 1e-4 \
  --image_aug True \
  --lora_rank 32 \
  --wandb_entity "$WANDB_ENTITY" \
  --wandb_project "$WANDB_PROJECT" \
  --run_id_override "$VERSION" 

python experiments/robot/libero/run_libero_eval.py  --pretrained_checkpoint "$RUN_ROOT_DIR/$VERSION"  --task_suite_name libero_${TASK} > "$EVAL_LOG_PATH" 2>&1