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
Training a Model with Limited Memory using Mixed Precision and Gradient Checkpointing
This article is divided into three parts; they are: • Floating-point Numbers • Automatic Mixed…
5 Ways to Use Cross-Validation to Improve Time Series Models
Time series modeling
AgREE: Agentic Reasoning for Knowledge Graph Completion on Emerging Entities
Open-domain Knowledge Graph Completion (KGC) faces significant challenges in an ever-changing world, especially when considering…