cad-classifier-free-guidance-analogy
IN premise — summaries/2026/08/24/shi-2024-context-aware-decoding-s1-introduction.md
Created 2026-08-25T02:58:35+00:00
CAD's contrastive structure (1+α)·logit(y|c,x) − α·logit(y|x) is explicitly framed as the text-generation analogue of classifier-free guidance in diffusion models (1+α)·ε_θ(x,c) − α·ε_θ(x).
Summary
The CAD method borrows a contrastive scaling trick originally designed to steer image and video generation toward a text prompt, and applies the same mathematical structure to nudge token-level predictions in language models. In practice, this means the strength of the conditioning signal is controlled by a single dial (alpha), and the approach inherits the theoretical grounding and tuning intuitions already validated in the diffusion literature.