add new changes

examples/6_evaluate_libero.sh

@@ -18,11 +18,11 @@ export CUDA_VISIBLE_DEVICES=2
 
 # CONFIGURATION
 POLICY_PATH="/raid/jade/logs/lerobot/lerobot_2_HuggingFaceVLA_libero_smolvla_lr1e-4bs32steps100000/checkpoints/100000/pretrained_model"
-POLICY_PATH="/raid/jade/logs/lerobot/lerobot_new_HuggingfaceVLA_libero_smolvla_lr1e-4bs32steps100000/checkpoints/100000/pretrained_model"
-TASK=libero_spatial
+POLICY_PATH="/raid/jade/models/smolvlamust"
+TASK=libero_spatial,libero_object
 ENV_TYPE="libero"
-BATCH_SIZE=10
-N_EPISODES=10
+BATCH_SIZE=1
+N_EPISODES=1
 # storage / caches
 RAID=/raid/jade
 N_ACTION_STEPS=1
@@ -46,7 +46,7 @@ python src/lerobot/scripts/eval.py \
     --env.type="$ENV_TYPE" \
     --eval.batch_size="$BATCH_SIZE" \
     --eval.n_episodes="$N_EPISODES" \
-    --env.multitask_eval=False \
+    --env.multitask_eval=True \
     --env.task=$TASK \
 # python examples/evaluate_libero.py \
 #     --policy_path "$POLICY_PATH" \

examples/8_train_smolvla_must.sh

@@ -56,14 +56,16 @@ EXPERT_WIDTH_MULTIPLIER=0.5
 # number of gpus to use
 NUM_PROCESSES=2
 NUM_VLM_LAYERS=0
-SELF_ATTN_EVERY_N_LAYERS=2
+SELF_ATTN_EVERY_N_LAYERS=0
 CHUNK_SIZE=50
-export CUDA_VISIBLE_DEVICES=0
+export CUDA_VISIBLE_DEVICES=1
 PORT=29522
 PREFIX_LENGTH=0
 LOAD_VLM_WEIGHTS=true
+MAX_ACTION_DIM=32
+MAX_STATE_DIM=32
 # naming/output dir
-TRAIN_DIR=$RAID/logs/lerobot/lerobot_new_${REPO_ID//\//_}_${POLICY}_lr${LR}bs${BATCH_SIZE}steps${OFFLINE_STEPS}
+TRAIN_DIR=$RAID/logs/lerobot/lerobot_new_sep11_v2_${REPO_ID//\//_}_${POLICY}_lr${LR}bs${BATCH_SIZE}steps${OFFLINE_STEPS}
 echo "Training dir: $TRAIN_DIR"
 
 rm -rf "$TRAIN_DIR"
@@ -137,5 +139,7 @@ python src/lerobot/scripts/train.py \
     --policy.load_vlm_weights=$LOAD_VLM_WEIGHTS \
     --policy.expert_width_multiplier=$EXPERT_WIDTH_MULTIPLIER \
     --policy.self_attn_every_n_layers=$SELF_ATTN_EVERY_N_LAYERS \
+    --policy.max_action_dim=$MAX_ACTION_DIM \
+    --policy.max_state_dim=$MAX_STATE_DIM \
     --seed=$SEED \
     --wandb.enable=false