Reproduce “Deep Learning Compilers: Architecture, Optimization, and Empirical Characterization” 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

Jinghao Wen, Ruixuan Wang, Xun Jiao

The authors explain how deep-learning compilers translate framework models through intermediate representations, optimize them, and generate hardware-specific code, then test TorchInductor against eager PyTorch on five synthetic FP32 inference graphs. Results depend on graph structure: the elementwise chain improves from 1.101 ms to 0.134 ms, while MLP and attention remain near eager latency. These measurements cover one RTX A6000 setup.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.