Diffusion Language Models are Super Data Learners

TL;DR

Diffusion language models outperform autoregressive models in data-limited scenarios due to their unique modeling approach and iterative denoising, achieving superior performance with less data and comparable compute.

Contribution

This paper introduces diffusion language models as a new paradigm that surpass autoregressive models under limited data conditions, highlighting their advantages and underlying mechanisms.

Findings

DLMs outperform AR models when data is limited.

A 1.7B DLM trained on 1.5T tokens surpasses AR models with similar compute.

DLMs achieve high accuracy on benchmarks with minimal data.

Abstract

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.