复现search R1，硬件使用4张80G A100，但是总报错OOM如下。

ray.exceptions.ActorDiedError: The actor died unexpectedly before finishing this task.
class_name: create_colocated_worker_cls..WorkerDict
actor_id: bc468de1d2d438e4f17c90f201000000
pid: 505292
name: burXCcWorkerDict_0:2
namespace: 0135b8a8-b4c4-4ef3-9902-9d1cd47d4075
ip: 10.128.9.119
The actor is dead because its worker process has died. Worker exit type: SYSTEM_ERROR Worker exit detail: Worker unexpectedly exits with a connection error code 2. End of file. There are some potential root causes. (1) The process is killed by SIGKILL by OOM killer due to high memory usage. (2) ray stop --force is called. (3) The worker is crashed unexpectedly due to SIGSEGV or other unexpected errors.

参数设置为
python3 -m verl.trainer.main_ppo --config-name=rl_factory_ppo_trainer
algorithm.adv_estimator=grpo
data.train_files=data/nq_search/train.parquet
data.val_files=data/nq_search/test.parquet
data.train_batch_size=32
data.max_prompt_length=4096
data.max_response_length=512
actor_rollout_ref.model.path=$MODEL_PATH
actor_rollout_ref.model.use_remove_padding=True
actor_rollout_ref.model.enable_gradient_checkpointing=True
actor_rollout_ref.actor.optim.lr=1e-6
actor_rollout_ref.actor.ppo_mini_batch_size=8
actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=2
actor_rollout_ref.actor.use_kl_loss=True
actor_rollout_ref.actor.kl_loss_coef=0.001
actor_rollout_ref.actor.kl_loss_type=low_var_kl
actor_rollout_ref.actor.fsdp_config.param_offload=True
actor_rollout_ref.actor.fsdp_config.optimizer_offload=True
actor_rollout_ref.actor.state_masking=True
actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=2
actor_rollout_ref.rollout.tensor_model_parallel_size=1
actor_rollout_ref.rollout.name=vllm
actor_rollout_ref.rollout.gpu_memory_utilization=0.75
actor_rollout_ref.rollout.n=4
actor_rollout_ref.rollout.max_turns=2
actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=2
actor_rollout_ref.ref.fsdp_config.param_offload=False
actor_rollout_ref.rollout.enforce_eager=False
actor_rollout_ref.rollout.free_cache_engine=True
actor_rollout_ref.env.name=search
actor_rollout_ref.env.mcp_mode=stdio
actor_rollout_ref.env.tool_manager=qwen3
actor_rollout_ref.env.enable_thinking=False
actor_rollout_ref.env.config_path=envs/configs/mcp_tools.pydata
actor_rollout_ref.env.use_process_reward=False
reward_rollout.if_use_reward_rollout=False
reward_rollout.rollout.tensor_model_parallel_size=4
reward_rollout.rollout.gpu_memory_utilization=0.65
reward_rollout.rollout.model_name=$REWARD_MODEL_PATH
reward_rollout.rollout.free_cache_engine=True
reward_rollout.rollout.response_length=2048
reward_model.reward_manager=parallel
algorithm.kl_ctrl.kl_coef=0.001
trainer.critic_warmup=0
trainer.logger=['tensorboard']
trainer.project_name='GRPO_search'
trainer.experiment_name='search_with_thinking'
trainer.n_gpus_per_node=4
trainer.nnodes=1
trainer.val_before_train=False
trainer.default_local_dir=$RESULT_DIR
trainer.default_hdfs_dir=null
trainer.save_freq=20
trainer.test_freq=10
trainer.total_epochs=5 $@ 2>&1 | tee grpo.log

请问设什么参数才能跑得起来