On-policy RL with GRPO outperforms SFT and off-policy methods for LLM-based vulnerability detection, with VD-specific insights on data curation, stage interactions, rewards, and evaluation.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CR 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
From SFT to RL: Demystifying the Post-Training Pipeline for LLM-based Vulnerability Detection
On-policy RL with GRPO outperforms SFT and off-policy methods for LLM-based vulnerability detection, with VD-specific insights on data curation, stage interactions, rewards, and evaluation.