Merge pull request #111 from JaaackHongggg/patch-1

Update train_grpo_vlagent_v2.sh

Author: ChenShawn

examples/agent/train_grpo_vlagent_v2.sh

@@ -1,20 +1,18 @@
 set -x
 
 # export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
-DATA_DIR=/cpfs/user/honglingyi/DATA/LLM/VL_Agent/parquets
 
 PROJECT_NAME="agent_vlagent"
 EXPERIMENT_NAME="visual_agent_env_v2_model_v3"
 
 export SAVE_CHECKPOINT_DIR=/diancpfs/user/fengyuan/verl_checkpoints
 # export VLLM_ATTENTION_BACKEND=XFORMERS # vllm + qwen2-7b with flash_attn has some issues
 
-VISUAL_DATASET_TRAIN=/cpfs/user/honglingyi/DATA/LLM/VL_Agent/parquets/vl_agent_V1_train_box.parquet
-VISUAL_DATASET_TEST=/cpfs/user/honglingyi/DATA/LLM/VL_Agent/parquets/vl_agent_V1_test_box.parquet
+VISUAL_DATASET_TRAIN=vl_agent_V1_train_box.parquet
+VISUAL_DATASET_TEST=vl_agent_V1_test_box.parquet
 
-# data.train_files=${DATA_DIR}/vl_agent_V1.parquet \
 
-REF_MODEL_PATH=/cpfs/user/honglingyi/MODEL/Qwen/Qwen2.5-VL-32B-Instruct
+REF_MODEL_PATH=Qwen/Qwen2.5-VL-32B-Instruct
 PYTHONUNBUFFERED=1 python3 -m verl.trainer.main_ppo \
     data.train_files=${VISUAL_DATASET_TRAIN} \
     data.val_files=${VISUAL_DATASET_TEST} \