Researchers propose AlterNet, an adaptive-depth mixture-of-experts architecture for elastic LLM inference
A new paper introduces AlterNet, an architecture combining mixture-of-experts with learned halting so that compute demands can adapt at inference time rather than being fixed during training.
Read the full story at arXiv cs.AI ↗
Timeline · 1 report
- 2026-11-09 05:00 · arXiv cs.AI
AlterNet: Adaptive-Depth Mixture-of-Experts with Learned Halting for Elastic Inference