backdoor_eval.py中代码的问题

[Jiajianghao]

这一段代码是正确的吗，我在测试emotion数据集的时候需要改成下面这种才能发挥作用

[Jiajianghao]

alpaca数据集上测试输出的时候发现模型输出的内容真的很难对上targetset，求解啊大佬，训练过程完全按照你们的代码和readme来的，Note that, when finetuning models on the Alpaca dataset, we set both source_max_len and target_max_len datasets as 1024 to allow the model to process and generate longer sentences.这个tips也注意到了--

[MiracleHH]

@Jiajianghao Hi, thanks for your interest in our work.
It seems that the results variable generated on your side has two folds (like [['Normal']]). But according to the definition of the generate() function that generates results, results should be a list of strings (like ['Normal']), and results[i] is just one specific string. You can check whether you have (unintentionally) modified the definition code of generate() and used the correct versions of transformers which may affect the decoded results in generate().