Add better gradient checkpointing to dfes

This commit is contained in:
Jaret Burkett
2026-05-18 09:25:23 -06:00
parent b42acb988f
commit ec58dcde92

View File

@@ -369,6 +369,9 @@ class SDTrainer(BaseSDTrainProcess):
if hasattr(self.dfe.model, 'enable_gradient_checkpointing'):
self.dfe.model.train()
self.dfe.model.enable_gradient_checkpointing()
if hasattr(self.dfe.model, 'gradient_checkpointing_enable'):
self.dfe.model.train()
self.dfe.model.gradient_checkpointing_enable()
elif hasattr(self.dfe.model, 'gradient_checkpointing'):
self.dfe.model.train()
self.dfe.model.gradient_checkpointing = True