FlashVector: Agent for Hierarchical Model Serving Stack Optimization
FlashVector is an agentic system that optimizes performance across all layers of the model serving stack, and the key contribution is an extensible framework to generalize the single kernel optimization agent paradigm to heterogeneous technical stacks, and to deliver performance improvements holistically.
Qi Wu, Lohan Lemire, Kai Meng et al.
· 0 citations