CHORD Leaderboard

Introduction

Existing metrics for open-ended text generation measure likelihood, lexical diversity, or distributional similarity in generic representation space, yet they can miss fundamental dimensions of quality. A prominent blind spot is global coherence: a generated passage may be locally fluent while remaining globally contradictory, causally inconsistent, or topically disconnected.

CHORD (Coherence-aware Hidden-state Open-generation Reference Distance) is a coherence-sensitive distributional metric. CHORD encodes generated and human-written corpora in the hidden-state space of a frozen LLM using a coherence-eliciting prompt, and compares the resulting distributions using MMD with an RBF kernel. This leaderboard ranks generators by their CHORD distance to the human reference on unconditional generation on OpenWebText; lower is closer. Generators are tagged as AR, discrete diffusion/flow, or continuous diffusion/flow, and can be ranked within each type.