A controlled study of decay in linear attention finds median decay near 0.8 works best, vector decay generally beats scalar decay, and RoPE/TPE give little benefit for models with sub-unity decay.
Parallelizing linear recurrent neural nets over sequence length
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Elucidating the Design Space of Decay in Linear Attention
A controlled study of decay in linear attention finds median decay near 0.8 works best, vector decay generally beats scalar decay, and RoPE/TPE give little benefit for models with sub-unity decay.