Does the code have Settings for model saving

[ansleepykid]

No description provided.

[G-U-N]

Can you rephrase your problem more clearly? (Questions written in Chinese are also acceptable)

[G-U-N]

Do you mean how to save models trained after each incremental session?

Please add self.save_checkpoint(filename) in the function after_task of the method you choose to run.

[Hephaestusxg]

Do you mean how to save models trained after each incremental session?

Please add self.save_checkpoint(filename) in the function after_task of the method you choose to run.

不好意思，我使用der的方法 在after_task后保存了权重，我发现这个方法模型的结构会动态的变化，全连接层也会改变，我根据变化的网络分别加载保存出来的权重 ，发现跟工具箱中的准确率都很大的区别，工具箱在初始训练7，增量1的条件下，基本能保持百分之90多，但是我自己单独拿出来测的话 在cifar10上测试，acc准确率很低。请问能帮我解决一下么