pHaeusler / codef-experiments

10 stars 0 forks source link

about train? #1

Open zhanghongyong123456 opened 1 year ago

zhanghongyong123456 commented 1 year ago
  1. python3 run.py train --image_dir ./beauty_1 When I train, I train for 120 epochs in 15 minutes, but it hasn't stopped yet, so I wonder what one minute means? and when to stop and move on to the next step
pHaeusler commented 1 year ago

I was using a large gpu with a batch_size of 100

zhanghongyong123456 commented 1 year ago

I was using a large gpu with a batch_size of 100

ℹi test bachsize =10 ,run 1000 epoch,but results not good: 00000 00010

canonical_0:

canonical_0

pHaeusler commented 1 year ago

Yeah... I was having some trouble with this too

I experimented with pre-training the canonical model - which helped

https://github.com/pHaeusler/codef-experiments/blob/94c73921f8ab726d5c360a7ece7c6ed0be2d0f6b/run.py#L328C13-L328C91