Offload ARA with the layer if doing layer offloading. Add support to offload the LoRA. Still needs optimizer support

This commit is contained in:
Jaret Burkett
2025-10-21 06:03:27 -06:00
parent 76ce757e0c
commit 0d8a33dc16
5 changed files with 37 additions and 0 deletions

View File

@@ -583,6 +583,8 @@ class LinearLayerMemoryManager(BaseLayerMemoryManager):
self.module.ara_lora_ref().org_forward = _mm_forward
else:
self.module.forward = _mm_forward
self.module._memory_management_device = self.manager.process_device
class ConvLayerMemoryManager(BaseLayerMemoryManager):
@@ -638,3 +640,5 @@ class ConvLayerMemoryManager(BaseLayerMemoryManager):
self.module.ara_lora_ref().org_forward = _mm_forward
else:
self.module.forward = _mm_forward
self.module._memory_management_device = self.manager.process_device