set -x

python3 -m verl.trainer.main_ppo \
    algorithm.adv_estimator=grpo \
    data.train_files=$HOME/data/geo3k/train.parquet \
    data.val_files=$HOME/data/geo3k/test.parquet \
    data.train_batch_size=128 \
    data.max_prompt_length=1024 \
    data.max_response_length=2048 \
    data.filter_overlong_prompts=False \
    data.truncation='error' \
    data.image_key=images \
    data.trust_remote_code=True \
    data.custom_cls.path=recipe/minicpmo/rl_dataset.py \
    data.custom_cls.name=RLHFDataset \
    actor_rollout_ref.model.path=openbmb/MiniCPM-o-2_6 \
    actor_rollout_ref.model.trust_remote_code=True \
    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=32 \
    actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=16 \
    actor_rollout_ref.actor.use_kl_loss=True \
    actor_rollout_ref.actor.use_dynamic_bsz=False \
    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=False \
    actor_rollout_ref.actor.fsdp_config.optimizer_offload=False \
    actor_rollout_ref.actor.fsdp_config.use_orig_params=True \
    actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=16 \
    actor_rollout_ref.rollout.tensor_model_parallel_size=2 \
    actor_rollout_ref.rollout.name=vllm \
    actor_rollout_ref.rollout.gpu_memory_utilization=0.6 \
    actor_rollout_ref.rollout.enable_chunked_prefill=False \
    actor_rollout_ref.rollout.enforce_eager=False \
    actor_rollout_ref.rollout.free_cache_engine=False \
    actor_rollout_ref.rollout.n=8 \
    actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=16 \
    actor_rollout_ref.ref.fsdp_config.param_offload=True \
    algorithm.kl_ctrl.kl_coef=0.001 \
    trainer.critic_warmup=0 \
    trainer.logger='["console","wandb"]' \
    trainer.project_name='verl_grpo_example_geo3k' \
    trainer.experiment_name='minicpmo2_6_function_rm' \
    trainer.n_gpus_per_node=8 \
    trainer.nnodes=1 \
    trainer.save_freq=-1 \
    trainer.test_freq=5 \
    trainer.total_epochs=15 $@