Paper proposes economic framework for LLM inference budget allocation

Arxiv CS.AI (top papers)

Authors model inference-time compute as a resource with a shadow price, enabling optimal budget distribution across queries. The framework treats reasoning tokens as a scarce computational good.

llm

Arxiv CS.AI (top papers)·Xu WanSpeed ZhuJianwei CaiGuang ChenXiMing HuangWiggin ZhouMingyang Sun

8 days ago