trained long time

[kktofcon]

[03/19 14:51:22][INFO] utils.logging: 88: {"_type": "train_iter", "epoch": "1/300", "eta": "2739 days, 21:40:53", "iter": "50/1000000", "loss": 3.201890, "lr": 0.000014, "time_diff": 0.789092, "top1_err": 50.000000, "top5_err": 0.000000}
is it normal？why？

[zgqtql666]

It's normal. You can use TensorBoard to monitor the loss and decide whether to apply early stopping, without having to complete the entire training.

[kktofcon]

I want to get the result that paper show，but I test the code get this ：     I dont get the result of paper. Do you have some suggestion? Thanks

…

------------------ 原始邮件 ------------------ 发件人: ***@***.***>; 发送时间: 2025年3月28日(星期五) 晚上10:20 收件人: ***@***.***>; 抄送: ***@***.***>; ***@***.***>; 主题: Re: [alibaba-mmai-research/MoLo] trained long time (Issue #8) It's normal. You can use TensorBoard to monitor the loss and decide whether to apply early stopping, without having to complete the entire training. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: ***@***.***> zgqtql666 left a comment (alibaba-mmai-research/MoLo#8) It's normal. You can use TensorBoard to monitor the loss and decide whether to apply early stopping, without having to complete the entire training. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: ***@***.***>