Hi, I'm a noob trying to figure out how to recognise overtraing. I followed the wiki and am looking at the g/total - i'm guessing this means i can keep going? bc no sudden drops or peaks? also once it's finished training, how do i decide which epoch to download - as long as it's not overtrained should i just get the most recent one..? thanks in advance!
#Overtraining and choosing which epochs to download
1 messages · Page 1 of 1 (latest)
make sure to put smoothening to 0.987
then select the graph point thats on the bottom - note down the step number, then go into your weights folder and check which epoch equals the steps
all model files, assuming you enabled saving model every x checkpoint, are named after epoch number then steps