Search papers, labs, and topics across Lattice.
This paper investigates the underutilization of autoregressive (AR) post-training weights in diffusion language models (dLLMs), which are typically initialized from AR models but then discard their post-training updates. The authors demonstrate that directly applying AR post-training weights to dLLMs effectively enhances their performance, approaching that of dLLMs that undergo direct diffusion post-training. Furthermore, by composing AR and diffusion post-training updates, the proposed A2D framework achieves complementary gains, improving instruction following, reasoning, and coding capabilities across multiple dLLM architectures without additional training or inference overhead.
Existing autoregressive post-training weights, often discarded after diffusion model conversion, can be directly applied to significantly boost dLLM performance, even outperforming dedicated diffusion post-training when combined.
Diffusion language models (dLLMs) have emerged as a promising alternative to autoregressive (AR) language models, offering flexible token-update orders and parallel decoding. Recent dLLMs are often initialized from pretrained AR models before diffusion conversion in order to inherit their learned representations. After the conversion, however, they typically ignore the extensive post-training ecosystem of their AR ancestors. In this work, we show that these existing AR post-training weight updates can instead be effectively recycled to enhance diffusion models. Despite the changes by AR-to-diffusion conversion, directly adding an AR post-training weight update to a diffusion base model remains effective, bringing its performance close to that achieved by direct diffusion post-training. Notably, AR and diffusion post-training updates are nearly orthogonal in weight space, yet induce substantially more aligned representation changes in the diffusion model. Their distinct updates are also complementary: composing their weights can retain gains from both regimes and further improve the post-trained diffusion model. Based on these findings, we propose A2D, a simple training-free framework for enhancing diffusion models with existing AR post-training resources. A2D can transfer capabilities from AR post-trained models to diffusion base models, and further improve already post-trained diffusion models by composing AR and diffusion post-training updates. Across various dLLMs, including Dream, DreamReasoner, DiffuCoder, Dream-Coder, Nemotron-Labs-Diffusion, and DiffusionGemma, A2D reliably improves instruction following, mathematical reasoning, and coding with both supervised fine-tuning and reinforcement learning updates, without additional training, or inference-time computation.
