A hypernetwork produces a condition-dependent beta that meta-gates SwiGLU nonlinearity, giving LLMs adaptive behavior across task, domain, persona and style inputs without finetuning.
MetaGater: Fast Learning of Conditional Channel Gated Networks via Federated Meta-Learning , year=
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Learn-To-Learn on Arbitrary Textual Conditioning: A Hypernetwork-Driven Meta-Gated LLM
A hypernetwork produces a condition-dependent beta that meta-gates SwiGLU nonlinearity, giving LLMs adaptive behavior across task, domain, persona and style inputs without finetuning.