>>58516481Gotta hide yourself better. Good luck though, I'll add that it's not fully "fair" to use locals as testing since each one has widely different outputs and has differing "best" settings. Which also gets thrown off if data is removed from a model. Eg. The tsukasa one yesterday had a version with more gpt4 data with pyg stuff, and one without. Whereas the mainly used things like Turbo/4/Claude are more standard.
