Reproduce “Balancing Latency and Accuracy of Code Completion via Local-Cloud Model Cascading” 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
Hanzhen Lu, Lishui Fan, Jiachi Chen, Qiuyuan Chen, Zhao Wei, Zhongxin Liu
MCCom offers line-level code completions with a local model, escalating to a cloud model when confidence is low or the user rejects its suggestion. It reuses rejected text as a draft and retrieval query. In an offline Python benchmark, Qwen2.5-Coder-7B on StmtEval-Suffix with both contexts had 293 ms latency and 82.5% exact-match accuracy, versus 501 ms and 79.1% for the cloud model alone. Live-user performance remains untested.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.