This is pretty interesting, based on the blog post, it seems like they are using a technique similar to what I have been using to generate "layer sensitivity" d

This is pretty interesting, based on the blog post, it seems like they are using a technique similar to what I have been using to generate "layer sensitivity" data in my (still pretty beta) ggufy project, which is more aimed at diffusion (image) models. https://github.com/qskousen/ggufy https://github.com/qskousen/ggufy