Publications

OmniCode: A Benchmark for Evaluating Software Development Agents

Published in Submitted to ICLR 2026 (Under Review), 2026

We propose a benchmark containing a broader and more diverse set of tasks for code-generated AI agents.

Recommended citation: Sonwane, Atharv*, Eng-Shen Tu*, Wei-Chung Lu*, Claas Beger*, Carter Larsen, Debjit Dhar, Rachel Chen et al. "OmniCode: A Benchmark for Evaluating Software Engineering Agents." arXiv preprint arXiv:2602.02262 (2026).
Download Paper

SKV (Semantic KV): Decoupling Uncertainty from Generation via Selective Caching

Published in Ongoing Research, 2025

Proposed a method leveraging selective KV cache to optimize inference reliability.

Recommended citation: Wei-Chung Lu, Peng Li. "SKV (Semantic KV): Decoupling Uncertainty from Generation via Selective Caching."

Wei-Chung (Wells) Lu

Publications

OmniCode: A Benchmark for Evaluating Software Development Agents

SKV (Semantic KV): Decoupling Uncertainty from Generation via Selective Caching