Meta Proposes Loop Scaling Laws for Looped MoE Models

Meta's paper introduces the first scaling law jointly modeling recurrence and MoE sparsity, showing sparsity yields 3x compute savings and looping 2x further parameter efficiency.

2026-10-01 ~ 2026-10-02 · 2 related posts

• Meta's Loop Scaling Laws: Sparsity Gives ~3x Active-Param Efficiency, Recurrence ~2x on Reasoning — facebook · 2026-10-01
• Loop Scaling Laws: First Scaling Law Jointly Modeling Recurrence and MoE Sparsity, Promising ~2x Parameter Savings — anirudhg9119 · 2026-10-02