tanglifutlf
tanglifutlf
豆瓣多轮对话
您好,阅读您的文献的时候发现了您开源的多轮对话语聊,请问能不能把训练集和验证集的语聊发我一下?([email protected]) 非常感谢!
### System Info **Training script**: + python3 -m verl.trainer.main_ppo --config-path=config --config-name=ppo_megatron_trainer.yaml 'data.train_files=['\''/llm-align/tanglifu/Codes/DVerl/Verl250606/verl_251124/inpudatas/data/gsm8k/train.parquet'\'', '\''/llm-align/tanglifu/Codes/DVerl/Verl250606/verl_251124/inpudatas/data/math/train.parquet'\'']' 'data.val_files=['\''/llm-align/tanglifu/Codes/DVerl/Verl250606/verl_251124/inpudatas/data/gsm8k/test.parquet'\'', '\''/llm-align/tanglifu/Codes/DVerl/Verl250606/verl_251124/inpudatas/data/math/test.parquet'\'']' data.prompt_key=prompt data.return_raw_chat=False data.truncation=left actor_rollout_ref.rollout.name=vllm actor_rollout_ref.rollout.mode=sync actor_rollout_ref.rollout.dtype=float16 actor_rollout_ref.actor.megatron.dtype=float16 data.max_prompt_length=4096 data.max_response_length=4096 data.train_batch_size=32 actor_rollout_ref.rollout.n=16 algorithm.adv_estimator=grpo algorithm.use_kl_in_reward=False algorithm.kl_ctrl.kl_coef=0.0 actor_rollout_ref.model.use_fused_kernels=True...