Diffusion Language Models are Super Data Learners
By: Jinjie Ni , Qian Liu , Longxu Dou and more
Potential Business Impact:
Makes AI better at writing code with less data.
Under strictly controlled pre-training settings, we observe a Crossover: when unique data is limited, diffusion language models (DLMs) consistently surpass autoregressive (AR) models by training for more epochs. The crossover shifts later with more or higher-quality data, earlier with larger models, and persists across dense and sparse architectures. We attribute the gains to three compounding factors: (1) any-order modeling, (2) super-dense compute from iterative bidirectional denoising, and (3) built-in Monte Carlo augmentation; input or parameter noise improves AR under data constraint but cannot close the gap. At scale, a 1.7B DLM trained with a ~1.5T-token compute budget on 10B unique Python tokens overtakes an AR coder trained with strictly matched settings. In addition, a 1B-parameter DLM achieves > 56% accuracy on HellaSwag and > 33% on MMLU using only 1B tokens, without any special tricks, just by repeating standard pre-training data. We also show that rising validation cross-entropy does not imply degraded downstream performance in this regime.
Similar Papers
A Survey on Diffusion Language Models
Computation and Language
Makes computers write faster and understand better.
How Efficient Are Diffusion Language Models? A Critical Examination of Efficiency Evaluation Practices
Computation and Language
Makes AI models learn and create faster.
Beyond Next-Token Prediction: A Performance Characterization of Diffusion versus Autoregressive Language Models
Machine Learning (CS)
Makes computers write faster and understand longer stories.