PaST extracts a domain-agnostic skill vector from RL training and linearly injects it into SFT-adapted LLMs to improve knowledge use on QA and tool-use tasks.
Checkpoint Selection Strategy.To avoid over- fitting to the source documents, we employ an inde- pendent validation set consisting of LooGLE docu- ments with indices 90-94
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Knowledge is Not Enough: Injecting RL Skills for Continual Adaptation
PaST extracts a domain-agnostic skill vector from RL training and linearly injects it into SFT-adapted LLMs to improve knowledge use on QA and tool-use tasks.