1 paper · 1 filter
Leo Fan, Hongyang Wang, Wenbo Chen +8
Training recipes are tuned by hand for one model, cluster and workload, and every new model, accelerator, framework version or dataset means tuning again. Automated tuners search a…