Configurations affect the training[2]

This figure illustrates how the configurations affect the training.

RLHF workflow[1]

【左边generate， 右边learning】

【左边前向，右边后向】

【ppo Adavatage 中计算r】

【grpo adavantge 计算简化 】

Example

case 1[10]

PYTHONPATH=/opt/tiger/open_verl python3 -m verl.trainer.main_ppo \
    algorithm.adv_estimator=grpo \
    data.train_files="$train_files" \
    data.val_files="$test_files" \
    data.**train_batch_size**=1024 \      ###
    data.max_prompt_length=1024 \
    data.max_response_length=1024 \
    data.filter_overlong_prompts=True \
    data.truncation='error' \
    actor_rollout_ref.model.path=$model_path \
    actor_rollout_ref.actor.optim.lr=1e-6 \
    actor_rollout_ref.model.use_remove_padding=True \
    actor_rollout_ref.actor.**ppo_mini_batch_size**=256 \     ###
    actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=16 \
    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.entropy_coeff=0 \
    actor_rollout_ref.model.enable_gradient_checkpointing=True \
    actor_rollout_ref.actor.fsdp_config.param_offload=True \
    actor_rollout_ref.actor.fsdp_config.optimizer_offload=True \
    actor_rollout_ref.**rollout**.log_prob_micro_batch_size_per_gpu=16 \
    actor_rollout_ref.rollout.tensor_model_parallel_size=4 \
    actor_rollout_ref.rollout.name=vllm \
    actor_rollout_ref.rollout.gpu_memory_utilization=0.5 \
    actor_rollout_ref.rollout.n=5 \
    actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=16 \
    actor_rollout_ref.ref.fsdp_config.param_offload=True \
    algorithm.use_kl_in_reward=False \
    trainer.critic_warmup=0 \
    trainer.logger='["console","wandb"]' \
    trainer.project_name='verl_grpo_example_gsm8k' \
    trainer.experiment_name='qwen2_14b_function_rm' \
    trainer.n_gpus_per_node=4 \
    trainer.nnodes=1 \
    trainer.save_freq=-1 \
    trainer.test_freq=5 \
    trainer.total_epochs=1 $@

case 1[11]