Reproduce “Mystique: Enabling Accurate and Scalable Generation of Production AI Benchmarks” 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
Mingyu Liang, Wenyin Fu, Louis Feng, Zhongyi Lin, Pavani Panakanti, Shengbao Zheng, Srinivas Sridharan, Christina Delimitrou
Mystique turns PyTorch model operator calls into runnable benchmarks by recording execution traces and profiler data, then reconstructing and replaying operations and dependencies. It preserves operator metadata while usually randomizing tensor values. On one A100, PARAM linear took 14.9 ms originally and 14.1 ms replayed, with 5.4% error. This result covers one iteration, not functional equivalence or universal benchmarking.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.