增量训练报CPU相关错误

我在进行LAC模型增量训练的时候，报以下错误：
----------------------
Error Message Summary:
----------------------
PreconditionNotMetError: The number(4) of samples[words] of current batch is less than the count(10) of devices(CPU), currently, it is not allowed. You should set the environment variable CPU_NUM in the system to determine the number of devices you need. at (/paddle/paddle/fluid/framework/parallel_executor.cc:923)

刚开始训练还挺正常，没问题，后来训练数据越来越多，大概两百多条，训练出现该错误。
环境python3.6，ubuntu16

Start by reproducing incremental LAC training with the reported Python 3.6 and Ubuntu 16 environment, using a dataset of roughly 200 records. Read the training entry point and the Paddle parallel-execution configuration around the reported CPU batch-size error. Done means identifying the regression or configuration issue and documenting a verified resolution.

Written by the indexing model from the issue text.