6 papers
Provable Test-Time Scaling for Beam Search in LLM Reasoning
Qijia He, Yu Huang, Yuan Cheng +2
Beam-search-based test-time methods provide an effective way to improve large language model (LLM) performance on long-horizon generation by pruning invalid reasoning paths early,…
Multi-Harness Reinforcement Learning for Coding Agents: Training Gains and Harness Dependence
Chenqian Le, Jiayi Cheng, Qijia He +3
Recent systems train coding agents with reinforcement learning (RL) on trajectories collected through multiple execution harnesses. However, gains under the training harnesses leav…
CodeRescue: Budget-Calibrated Recovery Routing for Coding Agents
Qijia He, Jiayi Cheng, Chenqian Le +8
Coding agents increasingly operate in executable environments where a failed attempt produces actionable feedback rather than merely an incorrect answer. Existing cost-aware system…
PACT: Privileged Trace Co-Training for Multi-Turn Tool-Use Agents
Zhenbang Du, Jun Luo, Zhiwei Zheng +8
Multi-turn tool-use agents must reason, call tools, and adapt to observations across several interaction turns. Post-training such agents is challenging, as reinforcement learning…
VFIG: Vectorizing Complex Figures in SVG with Vision-Language Models
Qijia He, Xunmei Liu, Hammaad Memon +6
Scalable Vector Graphics (SVG) are essential for technical illustration and digital design, offering resolution independence and semantic editability. In practice, original vector…
Rethinking Human Preference Evaluation of LLM Rationales
Ziang Li, Manasi Ganti, Zixian Ma +3
Large language models (LLMs) often generate natural language rationales -- free-form explanations that help improve performance on complex reasoning tasks and enhance interpretabil…