fanolabs / nid_aclarr2022 Goto Github PK

This error occurs when I run the data set for banking[

This is my environment

Thanks for the great paper. However, I run your code uses the model checkpoints your provide. But the results are significantly lower (Seven or eight percent lower)than those reported in the paper, in banking dataset. I just downloaded the pretrained checkpoints in link you provided and put it in folder pretrained_models under the root directory.Did I make any mistake?Does this pretrained checkpoints in link you provided mean "IntentBert(https://github.com/fanolabs/IntentBert)"?

您好，我对你们的工作非常感兴趣，这边有两个问题。首先你们论文中说到模型是用Clinc数据做预训练然后再用internal数据训练，为什么这里会提供3个预训练模型，我理解不是应该提供个Clinc的预训练模型？其次想请教下如果没有external数据的话，想复现出图4中的效果是不是把模型路径换成bert-base-uncased即可？再次感谢你们的工作！

I wonder why you disabled internal pretraining when args.known_cls_ratio == 0.
It becomes unsupervised setting as you wrote in the paper but unsupervised setting also involves internal pretraining - MTP.