The model's loss value turns into NaN

[wuya521]

"Hello, thank you for your excellent work. I encountered an issue during use, and I kindly ask for your help in solving it. The model's loss value turns into NaN. How can this be resolved? Sometimes the loss value is valid, but other times it becomes NaN. Thank you."

[wuya521]

Solution: Lowering the learning rate.

[chaidosa]

@wuya521 Thanks for comment, I did lower the learning rate but it still gives me NAN, how much did you lower it in your case?

[qinxiaobao8]

Using the Adam optimizer, it will automatically adjust the learning rate. Let's ignore the learning rate for now and focus on other parameters.