Add new attachment

In order to upload a new attachment to this page, please use the following box to find the file, then click on “Upload”.

This page (revision-11) was last changed on 23-Dec-2025 16:22 by 14.155.111.65

This page was created on 18-Nov-2025 16:19 by 139.162.79.56

Only authorized users are allowed to rename pages.

Only authorized users are allowed to delete pages.

Difference between version and

At line 151 changed one line
我们的大模型不需要验证集,因为验证集的作用就是监控模型预训练过程会不会过拟合或者没有泛化能力,不过作用也只能监控,无法通过调参彻底解决这个问题,这是以往大模型无法解决的问题,但是我们的损失设计就会迫使模型更“聪明”,而不是“死记硬背”,所以不如将验证集也放入预训练集中,让模型学习到更多的知识;而微调的时候需要验证集,就是防止模型在新的领域中过拟合或者欠拟合;
我们的大模型不需要验证集,因为验证集的作用就是监控模型预训练过程会不会过拟合或者没有泛化能力,不过作用也只能监控,无法通过调参彻底解决这个问题,这是以往大模型无法解决的问题,但是我们的损失设计就会迫使模型更“聪明”,而不是“死记硬背”,所以不如将验证集也放入预训练集中,让模型学习到更多的知识;而微调的时候需要验证集,就是防止模型在新的领域中过拟合或者失去泛化能力;
Version Date Modified Size Author Changes ... Change note
11 23-Dec-2025 16:22 4.949 kB 14.155.111.65 to previous
10 18-Dec-2025 19:05 4.907 kB 172.239.69.242 to previous | to last
9 18-Dec-2025 17:53 4.91 kB 172.239.69.242 to previous | to last
8 18-Dec-2025 17:46 4.85 kB 172.239.69.242 to previous | to last
7 20-Nov-2025 12:13 2.72 kB 139.162.79.56 to previous | to last
6 19-Nov-2025 20:39 2.69 kB 139.162.79.56 to previous | to last
5 19-Nov-2025 20:38 2.688 kB 139.162.79.56 to previous | to last
4 19-Nov-2025 20:36 2.685 kB 139.162.79.56 to previous | to last
3 18-Nov-2025 17:08 2.439 kB 139.162.79.56 to previous | to last
2 18-Nov-2025 16:29 2.315 kB 139.162.79.56 to previous | to last
1 18-Nov-2025 16:19 2.327 kB 139.162.79.56 to last
« This page (revision-11) was last changed on 23-Dec-2025 16:22 by 14.155.111.65