Online finetuning runs (sometimes crash because of nans)
This commit is contained in:
@@ -80,7 +80,7 @@ expectile: 0.9
|
||||
A_scaling: 3.0
|
||||
|
||||
# offline->online
|
||||
offline_steps: ${train_steps}/2
|
||||
offline_steps: 25000 # ${train_steps}/2
|
||||
pretrained_model_path: ""
|
||||
balanced_sampling: true
|
||||
demo_schedule: 0.5
|
||||
|
||||
Reference in New Issue
Block a user