Feature Request: Lora compability converter

Hello,

I have made an observation with lora files: Loras trained on 2.1 models will only work with 2.1 models and loras trained on 1.5 models will only work in combination with 1.5 models. Most likely I'm just stating the obvious but I wasnt aware of the fact, especially since a1111 does not validate if the selected lora is compatible with the selected model. It'll just throw cryptic errors instead.

So we are having a double incompability as not only loras will work with 1.5 OR 2.1 model, but also they cannot be combined with each other if they were trained on different model versions.

Anyhow, I believe there is so much lora content created upon 1.5 models that the majority of us will be bound to staying with 1.5 models due to the large pool of compatible loras. Since you know so much about training and creating the training tools, I was wondering if you can think of a lora wrapper/converters so 1.5 or 2.1 loras could be somehow made compatible with each other.

Or maybe the training script could just load both 1.5. and 2.x model and output some kind of "joint lora"?

The issue names no files, tests, or entry points to inspect. Clarify whether a LoRA converter or a joint LoRA is technically feasible and define the supported model versions and acceptance criteria before implementation.

Written by the indexing model from the issue text.