Hello,

I am exploring the possibility of performing on-device learning for LLM models, specifically focusing on fine-tuning LoRA adapters rather than retraining the full model.

I understand that mobile devices have hardware limitations, but I would like clarification currently supports LoRA fine-tuning directly on Android or iOS devices. Additionally, is there any support or roadmap for utilizing mobile GPUs or NPUs for such training tasks?

If on-device LoRA fine-tuning is not currently feasible, is the recommended approach to train LoRA adapters on server/cloud hardware and then deploy them for runtime inference on mobile devices?

Any guidance or references would be greatly appreciated. Thank you!