- algorithm.adv_estimator=grpo
- data.train_files=/projects/bfqt/users/yurenh2/ml-projects/personalization-user-model/collaborativeagents/training/grpo_verl/data/session_level_reflection_grpo_train.parquet
- data.val_files=/projects/bfqt/users/yurenh2/ml-projects/personalization-user-model/collaborativeagents/training/grpo_verl/data/session_level_reflection_grpo_train.parquet
- data.train_batch_size=64
- data.max_prompt_length=2048
- data.max_response_length=1024
- data.filter_overlong_prompts=True
- data.truncation=error
- data.prompt_key=prompt
- data.reward_fn_key=data_source
- actor_rollout_ref.model.path=/work/nvme/bfqt/yurenh2/sft_checkpoints/checkpoint-200
- actor_rollout_ref.actor.optim.lr=1e-6
- actor_rollout_ref.model.use_remove_padding=True
- actor_rollout_ref.actor.ppo_mini_batch_size=8
- actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=4
- actor_rollout_ref.actor.use_kl_loss=True
- actor_rollout_ref.actor.kl_loss_coef=0.003
- 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.model_dtype=bfloat16
- actor_rollout_ref.actor.fsdp_config.param_offload=False
- actor_rollout_ref.actor.fsdp_config.optimizer_offload=False
- actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=4
- actor_rollout_ref.rollout.tensor_model_parallel_size=1
- actor_rollout_ref.rollout.name=vllm
- actor_rollout_ref.rollout.gpu_memory_utilization=0.5
- actor_rollout_ref.rollout.n=8
- actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=4
- actor_rollout_ref.ref.fsdp_config.model_dtype=bfloat16
- actor_rollout_ref.ref.fsdp_config.param_offload=True
- actor_rollout_ref.rollout.temperature=0.9
- actor_rollout_ref.rollout.top_p=0.9
- custom_reward_function.path=/projects/bfqt/users/yurenh2/ml-projects/personalization-user-model/collaborativeagents/training/grpo_verl/verl_reward_functions.py
- custom_reward_function.name=compute_score
- algorithm.use_kl_in_reward=False
- trainer.critic_warmup=0
- trainer.val_before_train=False
- trainer.logger=["console"]
- trainer.project_name=collaborative-agent-reflection-grpo
- trainer.experiment_name=llama3.1-8b-grpo
- trainer.n_gpus_per_node=2
- trainer.nnodes=1
- trainer.save_freq=50
- trainer.test_freq=100
- trainer.total_epochs=1
- trainer.default_local_dir=/scratch/bfqt/yurenh2/grpo_outputs