In the paper, you mentioned that
"More broadly, attribution for generative models (e.g., stable diffusion [HJA20; RBL+22]) is an interesting direction for future work."
So I am wondering whether trak can be applied to diffusion models?
Indeed, I have done some tests under the lora finetuning setting but got very low lds scores...
Hi, @kristian-georgiev
In the paper, you mentioned that "More broadly, attribution for generative models (e.g., stable diffusion [HJA20; RBL+22]) is an interesting direction for future work." So I am wondering whether trak can be applied to diffusion models?
Indeed, I have done some tests under the lora finetuning setting but got very low lds scores...