We read every piece of feedback, and take your input very seriously.
To see all available qualifiers, see our documentation.
Have a question about this project? # for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “#”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? # to your account
您好,对您的工作表示祝贺,实在太赞了。我是名刚接触量化的萌新,最近需要实现低比特的后训练量化PTQ,由于知识有限,有两个疑问非常希望得到您的解答: 1.是只有iao才支持PTQ吗? 2.readme提到的“加载剪枝后的模型再做量化”中的量化指的是QWT还是PTQ呢? 2.我期望做低精度的PTQ(2~5bit),请问可以迁移您的工作以实现吗?还是说PTQ 只支持8bit的量化呢? 再次感谢您的工作!!!
The text was updated successfully, but these errors were encountered:
1、是iao做PTQ效果好一些,所以只在iao里实现了PTQ; 2、都可以,但PTQ的话剪枝比例得给小一点,不然可能精度不够; 3、只是精度问题,可以直接设置bit;但保证精度的前提下,PTQ想做到4-bit以下,很难。
Sorry, something went wrong.
No branches or pull requests
您好,对您的工作表示祝贺,实在太赞了。我是名刚接触量化的萌新,最近需要实现低比特的后训练量化PTQ,由于知识有限,有两个疑问非常希望得到您的解答:
1.是只有iao才支持PTQ吗?
2.readme提到的“加载剪枝后的模型再做量化”中的量化指的是QWT还是PTQ呢?
2.我期望做低精度的PTQ(2~5bit),请问可以迁移您的工作以实现吗?还是说PTQ 只支持8bit的量化呢?
再次感谢您的工作!!!
The text was updated successfully, but these errors were encountered: