2 papers
cs.LG2026
Strained Coherence: A Pre-Failure Signal in Coding Agent Execution Trajectories
Marut Pandya, Kasey Zhang, Baiqing Lyu
LLM-based coding agents sometimes acknowledge a problem in their own reasoning and then proceed anyway. We call this pattern strained coherence: a safety-relevant failure mode in w…
cs.LG2026
SFT-GRPO Data Overlap as a Post-Training Hyperparameter for Autoformalization
Xiaole Su, Kasey Zhang, Andy Lyu
Supervised fine-tuning (SFT) followed by Group Relative Policy Optimization (GRPO) is a common post-training recipe. We conduct a controlled ablation over SFT-GRPO data overlap, ev…