Closed XavierGuCN closed 4 weeks ago
没有出现过。loos为nan可能是混合精度时的数值溢出或者分母为零等情况，需要具体分析
-----原始邮件----- 发件人:XavierG @.> 发送时间:2024-09-27 16:23:24 (星期五) 收件人: yuhuUSTC/FSDGN @.> 抄送: Subscribed @.***> 主题: [yuhuUSTC/FSDGN] 请问您在训练时有遇见过loss为nan的情况吗，这是否与学习率有关 (Issue #12)
— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you are subscribed to this thread.Message ID: @.***>
感谢
XavierGuCN
commented
4 weeks ago XavierGuCN
commented
4 weeks ago
没有出现过。loos为nan可能是混合精度时的数值溢出或者分母为零等情况，需要具体分析
-----原始邮件----- 发件人:XavierG @.> 发送时间:2024-09-27 16:23:24 (星期五) 收件人: yuhuUSTC/FSDGN @.> 抄送: Subscribed @.***> 主题: [yuhuUSTC/FSDGN] 请问您在训练时有遇见过loss为nan的情况吗，这是否与学习率有关 (Issue #12)
— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you are subscribed to this thread.Message ID: @.***>