Zhenyu Huang

icon indicating a website link hi-zhenyu.github.io icon indicating an email [email protected]

icon location China icon location CS SCU

Results 3 comments of Zhenyu Huang

The last checkpoint or the best one on the Val split?

Hi, I fine-tuned the released pretrained model (meter_clip16_288_roberta_pretrain) on the NLVR2 data with the recommended parameters, but got different results from the paper: ``` {'nlvr2/dev/accuracy': 0.8161616325378418, 'nlvr2/dev/accuracy_epoch': 0.8206155896186829, 'nlvr2/dev/loss': 0.5619286894798279,...

疑似存在代码实现错误

您好， 很抱歉出现错误，ptl的计算应该使用pl的所有标签，我们已经修改了该问题

question on the loss

Yes, this is a mistake, I have fix it in [31dd08e](https://github.com/XLearning-SCU/2021-NeurIPS-NCR/commit/31dd08ec114a9c5abca88fb14b4487ffebaed292). Thanks for your comment.