Skip to content

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

3 Commits
 
 
 
 
 
 

Repository files navigation

muP introduction

muP parametrization transfert is a simple idea : you want to optimize your deep learning hyperparameters (learning rate, batch size, initialization etc) on small model because it's computationally cheap and then use your best found parametrization on a bigger model to avoid spending compute budget on optimizing the big model.

Those are the two main papers I will look at to implement the new parametrization :

About

muP parametrization enable to transfert optimal hyperparameter from a small nn to a bigger nn

Resources

Stars

0 stars

Watchers

1 watching

Forks

Releases

Packages

Contributors