RULER benchmark results for DroPE (especially LLaMA2-7B)?

[xlim1997]

Hi, thanks for sharing this very interesting work on DroPE — the idea of treating positional embeddings as a training-time scaffold is both elegant and impactful.

Quick question: have you evaluated DroPE on the RULER benchmark, especially for LLaMA2-7B-DroPE?

Best,
Lucas