GRAPE unifies RoPE and ALiBi as special cases of group actions on positions, providing a principled design space for positional encodings via SO(d) rotations and GL unipotent transformations.
The method then applies the matrix exponential of this generator to get a rotational position map
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Group Representational Position Encoding
GRAPE unifies RoPE and ALiBi as special cases of group actions on positions, providing a principled design space for positional encodings via SO(d) rotations and GL unipotent transformations.