A deadline-aware job packing system for heterogeneous GPU clusters that builds a runtime-interference-aware estimation model and solves an optimization problem to co-locate DL …
An enhanced heterogeneity-aware scheduler for GPU-based deep learning clusters that forks training jobs across multiple nodes via DP-based online allocation, improving cluster …
A unified GPU cluster scheduler that handles heterogeneous scheduling latency demands (strict/soft SLO and best-effort) and different stopping criteria (iteration-based or …