Fangkai Jiao
Fangkai Jiao
Sorry to say, I have got a similar worse performance on biadf++ using allennlp. I think there are two problems here. The first is that we need to calculate an...
OK, if I have free gpu, I will test the performance with SGD. By the way, I'm working on CoQA, [Conversational Question Answering Challenge](https://stanfordnlp.github.io/coqa/) and the model being tested is...
@intersun Hi, thanks for your reply. Indeed the link in normal and I could download the `keys-full.tar`. But I have encountered other problems. 1. I think the path for saving...
Allennlp本身就要比朴素的框架慢一些(实现里面为了避免oom用了lazy模式去读数据),我跑一个epoch大约要1个半小时左右。 而且我的优化器用的是sgd,你可以换成adamax,收敛应该会快很多
拆开的话应该不会降低效率,你可以试试去掉lazy模式,batch size开大一点
对 dataset_reader那一项里改成false就可以了 计算资源用的少的话……可能是我的实现并行程度不够高……如果我有空的话我会考虑再优化一下
另:optimizer也是在json文件里改一下就可以了
sorry,没有仔细看邮件,看分数好像不是很理想,我这两天看了下代码,暂时不知道有什么可以提升的地方emm 可能需要多试试参数
是EM分数吗?那比较接近了
那差的挺多的…… 可能得改一下其他参数