Adapting only the attention components with LoRA outperforms full-model adaptation in hybrid LLMs, with recurrent adaptation harming sequential hybrids but helping parallel ones.
Zuo, et al., Falcon-H1: A Family of Hybrid-Head Language Models, Technical Report, Technology Innovation Institute
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
-
Where Should LoRA Go? Component-Type Placement in Hybrid Language Models
Adapting only the attention components with LoRA outperforms full-model adaptation in hybrid LLMs, with recurrent adaptation harming sequential hybrids but helping parallel ones.