Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

关于学习率 #29

Open
JingyangXiang opened this issue May 17, 2022 · 2 comments
Open

关于学习率 #29

JingyangXiang opened this issue May 17, 2022 · 2 comments

Comments

@JingyangXiang
Copy link

JingyangXiang commented May 17, 2022

您好,我在复现论文的过程中,发现在pvpvt_s.txt 学习率最大为0.00125

{"train_lr": 1.000000000000015e-06, "train_loss": 6.913535571038723, "test_loss": 6.8714314655021385, "test_acc1": 0.2160000117301941, "test_acc5": 1.2940000837326049, "epoch": 0, "n_parameters": 24106216}
{"train_lr": 1.000000000000015e-06, "train_loss": 6.896081995010376, "test_loss": 6.839652865021317, "test_acc1": 0.40800002346038816, "test_acc5": 1.7700001041412354, "epoch": 1, "n_parameters": 24106216}
{"train_lr": 0.0002507999999999969, "train_loss": 6.628805226147175, "test_loss": 5.555466687237775, "test_acc1": 6.462000321006775, "test_acc5": 18.384000979614257, "epoch": 2, "n_parameters": 24106216}
{"train_lr": 0.0005006000000000066, "train_loss": 6.272622795701027, "test_loss": 4.64223074471509, "test_acc1": 15.328000774383545, "test_acc5": 34.724001724243166, "epoch": 3, "n_parameters": 24106216}
{"train_lr": 0.0007504000000000098, "train_loss": 5.958464104115963, "test_loss": 3.9152780064830073, "test_acc1": 24.868001231384277, "test_acc5": 48.68200244750977, "epoch": 4, "n_parameters": 24106216}
{"train_lr": 0.0010002000000000064, "train_loss": 5.670980889737606, "test_loss": 3.3432016747969167, "test_acc1": 33.47600182495117, "test_acc5": 59.10200291748047, "epoch": 5, "n_parameters": 24106216}
{"train_lr": 0.0012491503115478462, "train_loss": 5.421633140593767, "test_loss": 2.9919017029029353, "test_acc1": 39.19200199279785, "test_acc5": 65.3920035522461, "epoch": 6, "n_parameters": 24106216}
{"train_lr": 0.0012487765716255204, "train_loss": 5.202176849722862, "test_loss": 2.6157535275927297, "test_acc1": 45.32800238342285, "test_acc5": 71.13600388793945, "epoch": 7, "n_parameters": 24106216}

请问是根据batchsize进行缩放得到的学习率还是通过调整得到的学习率
非常感谢

@cxxgtxy
Copy link
Collaborator

cxxgtxy commented May 17, 2022

lr正比于batchsize缩放,可以看下代码

@JingyangXiang
Copy link
Author

非常感谢您的回复。
之前的疑惑在于,如果学习率上根据缩放得到,那么如果是8卡的话,单卡上batchsize上160,感觉是一个不怎么常见的配置,所以才提了issue

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants