完全按照readme来的，emotion数据集的test acc为0（通过修改backdoor_eval的代码已经解决）

[Jiajianghao]

0

[iwanttosignunp]

能麻烦问问您是怎么解决的吗，因为我通过backdoor_eval测试emotion数据集的准确率仍十分低，远远达不到论文的效果

[MiracleHH]

@Jiajianghao @iwanttosignunp You can check the format of the decoded results generated by the function generate(). It may be affected by the version of transformers. It is okay to change the code implementation of generate() to make the final results match the desired format. Also, make sure that the target model is properly trained and loaded before evaluation.