search

butchland / fastai_xla_extensions

A Python package to allow fastai to run on TPUs using Pytorch-XLA
https://butchland.github.io/fastai_xla_extensions
Apache License 2.0
36 stars 7 forks source link

It seems that using the value of the line in lr_find gives good results on trainning? #3

Closed tyoc213 closed 4 years ago

tyoc213 commented 4 years ago

If I use the value 10e-3 instead of 0.5 which I used on my computer, on TPU it gives better accuracy than the GPU lr.