Whether the model with ema needs bo be fixed in the validation phase —— “model(with ema).eval()”?

Question

Whether the model with ema needs bo be fixed in the validation phase —— “model(with ema).eval()”?

TIanCat opened this issue 3 years ago · comments

Samuel Fadel · Answer 1 · Wed Mar 03 2021 20:41:28 GMT+0800 (China Standard Time)

Hi,

I'm not sure I understood your question, but I have updated the example in README.md to show how the store/restore methods (contributed by @Zehui-Lin) can be used for validation without affecting training.

JianWang · Answer 2 · Wed Mar 03 2021 21:24:49 GMT+0800 (China Standard Time)

Thanks

…

------------------ 原始邮件 ------------------ 发件人: "Samuel Fadel"<notifications@github.com>; 发送时间: 2021年3月3日(星期三) 晚上8:41 收件人: "fadel/pytorch_ema"<pytorch_ema@noreply.github.com>; 抄送: "WangJian"<1090577307@qq.com>; "Author"<author@noreply.github.com>; 主题: Re: [fadel/pytorch_ema] Whether the model with ema needs bo be fixed in the validation phase —— “model(with ema).eval()”? (#3) Hi, I'm not sure I understood your question, but I have updated the example in README.md to show how the store/restore methods (contributed by @Zehui-Lin) can be used for validation without affecting training. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.