Reproduce “FastCI: Efficient GPU-Intensive CI for LLM Training Frameworks” 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
Tianshuo Qiao, Naiqian Zheng, Xiaopeng Liu, Shuguang Wang, Diandian Gu, Xuanzhe Liu, Xin Jin
FastCI reduces CI work for LLM training frameworks by sampling Python-function execution to select relevant tests, pruning tests with matching execution contexts and validation metadata, prioritizing failure-prone tests, and trimming workload dimensions outside each test’s scope. In a two-day shadow evaluation, it reduced CI latency by 77.5% and GPU use by 63.9%, while modified-code coverage retention rose 3.2 percentage points; these results were not production measurements.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.