Reproduce “Towards Easier-to-Tune Training Algorithms: New Results from AlgoPerf” 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
Priya Kasimbeg, Ahmed Khaled, Aahlad Chandrabhatta, Yifan Wang, Andy Liu, Niccolò Ajroldi, David Tweedle, Xingyu and 4 more
The authors evaluate 13 submissions on nine workloads, measuring time-to-target on four A100 GPUs. Each uses one fixed hyperparameter configuration, with batch size the workload-dependent exception. The score is normalized area under a profile of workloads solved within time ratios of the fastest submission. Schedule-Free AdamW v2 (PyTorch) ranks first at 0.5679, reaching six targets in this benchmark.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.