bandwidth-adaptive

Definition

Adjusting what is transmitted according to the link capacity available at the moment. In split inference or streaming systems this means degrading resolution, compressing intermediate features harder, or shifting more computation to the device when the connection is poor, and restoring quality when it improves. The alternative is provisioning for the worst case with a fixed configuration, which wastes headroom whenever conditions are good.

Explainers using this term

• Paper Explained: FreeToken — Treating Your Own PC as a Single Elastic Inference PlatformFreeToken: Efficient Edge-Native MoE Serving with Bandwidth-Adaptive Execution