JoLA jointly learns sparse attention-head selection and additive/multiplicative activation edits, outperforming LoRA and prior activation-editing baselines in low-resource LLM fine-tuning.
Title resolution pending
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
-
Joint Localization and Activation Editing for Low-Resource Fine-Tuning
JoLA jointly learns sparse attention-head selection and additive/multiplicative activation edits, outperforming LoRA and prior activation-editing baselines in low-resource LLM fine-tuning.