One thing I've wondered about is fine tuning a large model from multiple LoRAs. If the model doesn't fit in your vram you can train a LoRA, apply it to the model, train another LoRA from the same data, apply it, and so on. Iterative low rank parameter updates. Would that work?