代码运行有问题
happy-nlp opened this issue · 5 comments
你好，感谢你的分享，让我有机会下载研究你的代码~
我在运行过程中遇到了问题，梯度值总是nan，只有第一次迭代可以正常显示loss，调整参数也没有解决问题，不知道你是否遇到，或能给我一些建议
另外我没有在项目中找到用于测试的数据，希望能分享下，感谢！
我看到关闭的问题里面有人说是学习率的问题，我将学习率调整到1E-10，仍然有nan值，报错和他们的一样
I'm sorry I've entered Chinese before. I have tried to reduce the learning rate, increase the clipping, and still do not solve the Nan value problem. I even tuned the version of the TF to 1.1 that you were in line with, and it didn't work. I hope to reply in time, because I really need it now, I like the code you share very much, but I have been troubled by this operation problem for 10 days. I'm going crazy
Hi, I have encountered the same problem. The loss became nan at the second step. Are there some solutions to this problem? Thank you!
It depends on learning rate and the other hyperparameters. Alter that