Breaking Diffusion: Parallel Reasoning with Draft Refinement
Drafting an answer allows a reasoning model to reconsider each decision in the context of the others. Yet repeated diffusion denoising does not ensure that individually plausible token updates form a better answer together. We introduce Draft Refinement Language Models (DRLMs), which use a single denoiser to propose parallel revisions and evaluate a candidate draft. We train th
Reading in Plain English mode — toggle below to switch
Drafting an answer allows a reasoning model to reconsider each decision in the context of the others. Yet repeated diffusion denoising does not ensure that individually plausible token updates form a better answer together. We introduce Draft Refinement Language Models (DRLMs), which use a single denoiser to propose parallel revisions and evaluate a candidate draft. We train th
This paper introduces a novel approach to improve reasoning in language models by using a single denoiser for parallel revisions and draft evaluation. This method could lead to more efficient and accurate reasoning capabilities, which is crucial for advanced AI agents and complex problem-solving.