Skip to main content
costa-huang
Projects
cleanRL
Reports
scheduler
Log in
Sign up
Share
Comment
Star
Share
Comment
Star
scheduler
Costa
Created on October 14
|
Last edited on October 14
Comment
linear warm up and decay has helped versus just decay
train/accuracy
train/accuracy
0
1k
2k
3k
4k
Step
0.4
0.5
0.6
0.7
0.8
0.9
train/loss
train/loss
0
1k
2k
3k
4k
Step
0.5
1
1.5
2
train/lr
train/lr
0
1k
2k
3k
4k
Step
0
0.00001
0.00002
0.00003
0.00004
dev_validation/accuracy
dev_validation/accuracy
0
1k
2k
3k
4k
Step
0.48
0.5
0.52
0.54
0.56
0.58
0.6
eval_validation/accuracy
eval_validation/accuracy
0
1k
2k
3k
4k
Step
0.5
0.55
0.6
Run set
0
Run set 2
10
Add a comment