分布式准确率

咨询一下，分布式训练出来的结果和你们提供的F1一致吗？
用ppi数据集跑，单机的效果和你们提供的F1基本持平；分布式下，准确率低很多，这种正常吗

No files, tests, or entry points are named. Start by reproducing the reported PPI dataset comparison between single-machine and distributed training, then inspect how distributed training computes and reports F1 and identify why the result diverges. Done means determining whether the lower distributed accuracy is expected or fixing the discrepancy with a reproducible test.

Written by the indexing model from the issue text.