你好，

我在训练百川33B模型，遇到dp, mp, pp设置问题导致模型编译出错能否支持解决下，

如下设置会导致：OM错误， 如下截图

# default parallel of device num = 16 for Atlas 800 

parallel_config: 

  data_parallel: 8 

  model_parallel: 4 

  pipeline_stage: 1 

  micro_batch_num: 8 

  vocab_emb_dp: True 

  gradient_aggregation_group: 4 

# when model parallel is greater than 1, we can set micro_batch_interleave_num=2, that may accelerate the train process. 

micro_batch_interleave_num: 1

当改变设置如下后， 提示新的错误如下图

# default parallel of device num = 16 for Atlas 800 

parallel_config: 

  data_parallel: 2 

  model_parallel: 4 

  pipeline_stage: 4 

  micro_batch_num: 8 

  vocab_emb_dp: True 

  gradient_aggregation_group: 4 

# when model parallel is greater than 1, we can set micro_batch_interleave_num=2, that may accelerate the train process. 

micro_batch_interleave_num: 1

请帮忙看下，如何解决训练问题，谢谢。