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.

Sign up for GitHub

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

Jump to bottom

模型性能提升到13° #6

Closed
ControlNet opened this issue Jan 20, 2023 · 7 comments
Closed

模型性能提升到13° #6

ControlNet opened this issue Jan 20, 2023 · 7 comments
Labels
enhancement New feature or request

Comments

@ControlNet
Copy link
Contributor

发现原来的模型有些欠拟合，稍微改了一点参数在不太增加性能开销的同时提高了准确度。

如果作者觉得这个改进有帮助的话，会提交一个pr。

做的修改有

  • 全连接层增加到2层
  • 降低LR到4e-4
  • 设定epoch = 3000，在epoch = 190的时候达到最优结果
  • batch_size = 128

定量比较

Error MAC PARAMS
OLD 19.98° 1.63G 8.28M
NEW 13.09° 1.63G 9.11M

定性比较

image

训练过程 eval_loss.png

eval_loss
@ControlNet ControlNet mentioned this issue Jan 20, 2023
Closed
@Starry-OvO Starry-OvO added the enhancement New feature or request label Jan 20, 2023
@Starry-OvO
Copy link
Owner

我觉得可以

@Starry-OvO
Copy link
Owner

对了，如果要pr的话就不用提交新模型了，老模型也可以删掉，git不适合存这种大文件，还是网盘好用

@ControlNet
Copy link
Contributor Author

其实对于这种小模型，感觉可以通过github release来存

@Starry-OvO
Copy link
Owner

也可以

@ControlNet ControlNet mentioned this issue Jan 20, 2023
Merged
@n0099
Copy link

n0099 commented Jan 20, 2023

对了，如果要pr的话就不用提交新模型了，老模型也可以删掉，git不适合存这种大文件，还是网盘好用

小心52pojie那群广告黑灰产发issue问您为啥不能开箱即用

@langyo
Copy link

langyo commented Jan 21, 2023

对了，如果要pr的话就不用提交新模型了，老模型也可以删掉，git不适合存这种大文件，还是网盘好用

小心52pojie那群广告黑灰产发issue问您为啥不能开箱即用

这不是正好嘛，让他们爬去

@n0099
Copy link

n0099 commented Jan 21, 2023

对了，如果要pr的话就不用提交新模型了，老模型也可以删掉，git不适合存这种大文件，还是网盘好用

小心52pojie那群广告黑灰产发issue问您为啥不能开箱即用

这不是正好嘛，让他们爬去

与此同时 HuanCheng65/TiebaLite#145 (comment)
https://www.52pojie.cn/thread-1648818-1-1.html

没有api：starry神 @Starry-OvO 已于2022年6月发威逆向出了在im.tieba.baidu.com:8000的ws服务隐藏之下的protobuf
而这个ws endpoint早在17年就业已被贴吧工具人们挖了出来 bbs.tiebazs.com/593-1-1.html 但当时他们还不知道这是干嘛的

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
enhancement New feature or request
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
4 participants
@ControlNet @n0099 @langyo @Starry-OvO