loss

[xzhxzhxzhxzh]

hello,why is my initial loss around 5.

[whubaichuan]

Hi, can you put the whole training process here?

[xzhxzhxzhxzh]

[xzhxzhxzhxzh]

Could it be a problem with the environment

[whubaichuan]

hi, thanks. can you try to finish the 10 epochs (10*6774)? I think the initial loss will not influence the final results, as you worry about.

In some cases, the real training process can not be the same as the perfect-tutorial training loss curve, such as it. In real training, usually, it fluctuates.

[xzhxzhxzhxzh]

您好，其实之前我已经训练了十轮，不过我是在特征提取模块处又加了个注意力机制，它的损失函数也变化不大。

[whubaichuan]

Hi, thanks, can you get similar final results (ACC, COMP, and OVERALL) as presented in the readme.md, with the no-modified code?

[xzhxzhxzhxzh]

您好，我还没有在不修改代码的情况下训练，后面我训练一下。谢谢您