DARE: Diffusion Large Language Models Alignment and Reinforcement Executor

TL;DR

DARE is an open framework that unifies post-training and evaluation methods for diffusion large language models, facilitating research, comparison, and deployment across various model families.

Contribution

It introduces a comprehensive, unified platform built on verl and OpenCompass for post-training, reinforcement learning, and evaluation of dLLMs, addressing fragmentation issues.

Findings

DARE enables broad algorithmic coverage across multiple dLLM models.

It provides reproducible benchmark evaluation and practical acceleration.

Empirical results demonstrate DARE's effectiveness as a research and deployment tool.

Abstract

Diffusion large language models (dLLMs) are emerging as a compelling alternative to dominant autoregressive models, replacing strictly sequential token generation with iterative denoising and parallel generation dynamics. However, their open-source ecosystem remains fragmented across model families and, in particular, across post-training pipelines, where reinforcement learning objectives, rollout implementations and evaluation scripts are often released as paper-specific codebases. This fragmentation slows research iteration, raises the engineering burden of reproduction, and makes fair comparison across algorithms difficult. We present \textbf{DARE} (\textbf{d}LLMs \textbf{A}lignment and \textbf{R}einforcement \textbf{E}xecutor), an open framework for post-training and evaluating dLLMs. Built on top of verl~\cite{sheng2024hybridflow} and OpenCompass~\cite{2023opencompass}, DARE unifies supervised fine-tuning, parameter-efficient fine-tuning, preference optimization, and dLLM-specific reinforcement learning under a shared execution stack for both masked and block diffusion language models. Across representative model families including LLaDA, Dream, SDAR, and LLaDA2.x, DARE provides broad algorithmic coverage, reproducible benchmark evaluation, and practical acceleration. Extensive empirical results position that DARE serves as a reusable research substrate for developing, comparing, and deploying post-training methods for current and emerging dLLMs.