Reproduce “Contextualized Code Pretraining for Code Generation” 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
Chen Liu, Qingyuan Liang, Hanwen Zhang, Zeyu Sun, Yakun Zhang, Lu Zhang
CallerGen trains code models to implement a function from its header, a direct caller, and, when available, its docstring, teaching them to use call-site behavior as evidence. On Python CallerEval, the 0.5B CallerGen-Qwen model achieves 22.81% pass@1, versus 12.13% for a same-scale model trained without caller context. The result is limited to tested open-source Python tasks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.