Update README.md

This commit is contained in:
Less Wright
2021-05-01 17:00:20 -07:00
committed by GitHub
parent 39dd1bb6db
commit adfdd0e618
+4 -2
View File
@@ -29,10 +29,12 @@ Net results: 14.15% greater accuracy with Ranger21 vs Adam, same training epochs
</br>
In addition, training a BN Free (no batch norm) ResNet50 as an additional comparison. Of interest, even after 4 restarts, Adam was unable to get more than 3 epochs in on the NormFree Resnet50. By comparison, Ranger21 is doing well so this already shows the improved resilience of training with Ranger21. </br>
![r21_nfresnet50_adam](https://user-images.githubusercontent.com/46302957/116797723-d4b2b100-aa9d-11eb-821a-b7fdc81592a0.JPG)
![r21_nfresnet50_adam](https://user-images.githubusercontent.com/46302957/116797807-a4b7dd80-aa9e-11eb-8045-52d10d390e82.JPG)
* Ranger21 code updates - due to firsthand experience, have added in safety guards in the event that num_epochs does not match the actual epochs trained, as well as updated the linear warmdown code to be simpler and never go below the min_lr designated (defaults to 3e-5). This will start to spew a lot of text to alert you on each iteration, but the lr will be automatically guarded. </br>
![r21_epoch_check](https://user-images.githubusercontent.com/46302957/116797758-34a95780-aa9e-11eb-9650-de1f4e1a5b28.JPG)
![r21_epoch_check](https://user-images.githubusercontent.com/46302957/116797758-34a95780-aa9e-11eb-9650-de1f4e1a5b28.JPG) </br>
<b> April 27 PM - Ranger21 now training on ImageNet!</b> Starting work on benchmarking Ranger21 on ImageNet. Due to cost, will train to 60 epochs on ImageNet and compare with same setup with 60 epochs using Adam to have a basic "gold standard" comparison. Training is underway now.