求助

为什么训练过程和预测过程在同一份数据上的输出指标不同？

在训练过程中的eval时，输出为：

[2020-11-20 01:21:14,467] [ EVAL] - [dev dataset evaluation result] loss=0.35801 precision=0.95772 recall=0.96399 [step/sec: 10.26]
[2020-11-20 01:21:50,027] [ EVAL] - [dev dataset evaluation result] loss=0.35785 precision=0.95779 recall=0.96563 [step/sec: 10.25]
[2020-11-20 01:21:54,130] [ EVAL] - [dev dataset evaluation result] loss=0.35792 precision=0.95779 recall=0.96563 [step/sec: 10.25]

在预测过程中执行eval时，输出为：

[2020-11-20 01:23:47,939] [ EVAL] - [dev dataset evaluation result] loss=0.56136 precision=0.88357 recall=0.90671 f1=0.89499 [step/sec: 15.16]

使用的task是：TextClassifierTask

谢谢