The generation quality of large language models (LLMs) is often improved by utilizing inference-time sequence-level scaling methods (e.g., Chain-of-Thought). We introduce hyper-parallel scaling, a complementary framework that improves prediction quality at the token level. Hyper-parallel scaling computes and aggregates multiple output proposals for a single token from the model. We implement this concept in Mixture-of-Experts (MoE) models, which we refer to as Roster of Experts (RoE). RoE is a training-free inference algorithm that turns a single MoE into a dynamic ensemble of MoEs. RoE… Read More
MoEs Are Stronger than You Think: Hyper-Parallel Inference Scaling with RoE
The generation quality of large language models (LLMs) is often improved by utilizing inference-time sequence-level scaling methods (e.g., Chain-of-Thought). We introduce hyper-parallel scaling, a complementary framework that improves prediction quality at the token level. Hyper-parallel scaling computes and aggregates multiple output proposals for a single token from the model. We implement this concept in Mixture-of-Experts (MoE) models, which we refer to as Roster of Experts (RoE). RoE is a training-free inference algorithm that turns a single MoE into a dynamic ensemble of MoEs. RoE…
Related Posts
3 Subtle Ways Data Leakage Can Ruin Your Models (and How to Prevent It)
Data leakage is an often accidental problem that may happen in machine learning modeling.
Sample and Map from a Single Convex Potential: Generation using Conjugate Moment Measures
The canonical approach in generative modeling is to split model fitting into two blocks: define…
How Agentic AI Empowers Architecture Governance
One of the principles in our upcoming book Architecture as Code is the ability for…