tdm 单机预测评估时卡住没反应

按照教程单机版demo，在完成训练和聚类后，进行预测和评估时，运行一会儿算loss时卡住了，如图这样，后面就一直卡住没反应，直到连接集群时间过长退出，请问是什么原因？使用的就是user_behavier数据集
命令是python train.py --run_mode=local --config=config.predict.json

Start with the documented single-machine command, `python train.py --run_mode=local --config=config.predict.json`, using the `user_behavier` dataset. Read `train.py` and `config.predict.json` around prediction, loss calculation, evaluation, and the cluster connection; done means the local run completes evaluation instead of hanging until the connection timeout.

Written by the indexing model from the issue text.