FreeToken: Efficient Edge-Native MoE Serving with Bandwidth-Adaptive Execution
Paper LLMs & Texto

FreeToken: Efficient Edge-Native MoE Serving with Bandwidth-Adaptive Execution

FreeToken is an edge-native Mixture-of-Experts serving system that dynamically maps computation and model state onto heterogeneous local hardware to run large open-weight models on…

Hugging Face · Daily Papers ·Shuo Yang, Xiaoze Fan · ·▲ 43 upvotes

Este artigo está em destaque na seleção diária de papers do Hugging Face, curada pela comunidade de pesquisa em IA.

Autores: Shuo Yang, Xiaoze Fan, Melissa Pan, Haocheng Xi, Zhe Wang, Shanlin Sun

  • 43 upvotes da comunidade
  • Temas: MoE serving, expert residency, CPU-GPU execution, agentic state reuse, runtime memory management, offloading strategy

Resumo

Resumo original (em inglês), extraído do paper:

FreeToken is an edge-native Mixture-of-Experts serving system that dynamically maps computation and model state onto heterogeneous local hardware to run large open-weight models on personal machines.

Onde ler

compartilhar: