A hierarchical genome search over linear input-varying systems, evolved with NSGA-2, produces language model architectures that match or outperform Transformer++ and StripedMamba on quality, size, and cache efficiency.
One of the groups is explicitly parametrized (e.g., short convolutions of length 3 used for GConv-1)
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2024 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
STAR: Synthesis of Tailored Architectures
A hierarchical genome search over linear input-varying systems, evolved with NSGA-2, produces language model architectures that match or outperform Transformer++ and StripedMamba on quality, size, and cache efficiency.