alt Hacker NewsThe paper says that:
> In practice, we find that four Taylor terms (P = 4) suffice for recovering conventional attention with elementwise errors of approximately the same magnitude as Float16 resolution, acceptable for many AI applications.
ie., the claim is that this method reproduces the results of conventional attention, up to float16 numerical precision.
Replies
fheinsen • today at 5:13 PM
The method is more general. The github repository's first example is with eight Taylor terms (P = 8).
energy123 • today at 5:29 PM
It converges on conventional attention as P goes up
> approximately the same magnitude
and they really do mean that, their results show +/- 1 on log10 plots.