Efficient-DLM: From Autoregressive to Diffusion Language Models, and Beyond in Speed

TL;DR

This paper introduces a novel AR-to-dLM conversion method that enhances diffusion language models' speed and accuracy by preserving pretrained AR weights and employing a position-dependent masking strategy.

Contribution

It proposes a continuous pretraining scheme with block-wise attention and a position-dependent masking strategy, significantly improving dLM efficiency and performance.

Findings

Efficient-DLM 8B outperforms Dream 7B and Qwen3 4B in accuracy and throughput.

Maintaining pretrained AR weight distributions is crucial for effective AR-to-dLM conversion.

The proposed methods lead to state-of-the-art results in speed and accuracy for diffusion language models.

Abstract

Diffusion language models (dLMs) have emerged as a promising paradigm that enables parallel, non-autoregressive generation, but their learning efficiency lags behind that of autoregressive (AR) language models when trained from scratch. To this end, we study AR-to-dLM conversion to transform pretrained AR models into efficient dLMs that excel in speed while preserving AR models' task accuracy. We achieve this by identifying limitations in the attention patterns and objectives of existing AR-to-dLM methods and then proposing principles and methodologies for more effective AR-to-dLM conversion. Specifically, we first systematically compare different attention patterns and find that maintaining pretrained AR weight distributions is critical for effective AR-to-dLM conversion. As such, we introduce a continuous pretraining scheme with a block-wise attention pattern, which remains causal across blocks while enabling bidirectional modeling within each block. We find that this approach can better preserve pretrained AR models' weight distributions than fully bidirectional modeling, in addition to its known benefit of enabling KV caching, and leads to a win-win in accuracy and efficiency. Second, to mitigate the training-test gap in mask token distributions (uniform vs. highly left-to-right), we propose a position-dependent token masking strategy that assigns higher masking probabilities to later tokens during training to better mimic test-time behavior. Leveraging this framework, we conduct extensive studies of dLMs' attention patterns, training dynamics, and other design choices, providing actionable insights into scalable AR-to-dLM conversion. These studies lead to the Efficient-DLM family, which outperforms state-of-the-art AR models and dLMs, e.g., our Efficient-DLM 8B achieves +5.4%/+2.7% higher accuracy with 4.5x/2.7x higher throughput compared to Dream 7B and Qwen3 4B, respectively.