Hi,

I've been trying to reproduce the results reported in the paper, and noticed that Table 4 in Appendix A does not incorporate the hyperparameters used for training MDEQ-XL on ImageNet. In particular, I'm curious about the following:

• In general, is the stop mode "rel" or "abs"?
• What epsilon is used as the threshold in the Broyden solver? Should I assume it was 1e-3 as is the default value?
• What were the forward and backward quasi-Newton thresholds $T_f, T_b$?

Thanks so much!