Lora Adapter Loading Issues in Multi-Stage Model Training

[Gbone3176]

Thank you for your outstanding work!

I'm a newbie just getting started with LoRA fine-tuning and have some questions I'd like to ask. When reproducing the process, my training workflow is mntp -> SimCSE -> Supervised. After completing the SimCSE training, when I start the Supervised training, do I need to load both LoRA adapter checkpoints from the mntp and SimCSE training processes simultaneously? Or is it sufficient to import only the adapter checkpoint obtained from the SimCSE training? I'm eagerly awaiting your response and thank you so much!