Technique skips unnecessary AI model layers to meet speed budgets

Original: BUDDY: BUdget-Driven DYnamic Depth Routing for Adaptive Large Language Model Inference

Source: arxiv.org ↗