Add support for LoRA finetuning

[zappy586]

@SahilCarterr I am looking for a way to train few-shot adapters for GDINO using LoRA. I tried loading this particular implementation using LoRA adapters but it simply doesn't work. I think its because some layers are implemented in C++/CUDA. It would be great if you could demonstrate how to load this model with LoRA in an example training script
Thanks!