Reproduce “EngineAgent: Agentizing On-Device LLM Inference Engines with Inference Stack Hierarchy Awareness and Hardware Abilities” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Zhengxiang Huang, Yujie Sun, Shengheng Chen, Zhaode Wang, Chaoyue Niu, Zeyu Zhao, Chengfei Lv, Fan Wu and 1 more
EngineAgent prepares a hierarchy-aware engine index and validated hardware-aware build, test, and debugging skills, then uses builder, debugger, and auditor agents to guide and check model-specific changes. On 12 tasks across three mobile language models and four engine–backend pairs, GPT-5.6 Sol achieved 9/12 successes (75%), versus Codex’s 4/12; three tasks still failed, two at graph implementation and one at hardware-kernel implementation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.