DiffER: Diffusion Entity-Relation Modeling for Reversal Curse in Diffusion Large Language Models

TL;DR

This paper investigates the reversal curse in diffusion large language models, identifies key causes, and proposes DiffER, a method that uses entity-aware training and balanced data to mitigate this issue effectively.

Contribution

The paper introduces DiffER, a novel approach that addresses the reversal curse in diffusion LLMs through entity-aware training and balanced data strategies.

Findings

DiffER significantly reduces the reversal curse in DLLMs.

Entity-aware training improves entity comprehension in models.

Balanced data construction enhances relation learning in diffusion models.

Abstract

The "reversal curse" refers to the phenomenon where large language models (LLMs) exhibit predominantly unidirectional behavior when processing logically bidirectional relationships. Prior work attributed this to autoregressive training -- predicting the next token inherently favors left-to-right information flow over genuine bidirectional knowledge associations. However, we observe that Diffusion LLMs (DLLMs), despite being trained bidirectionally, also suffer from the reversal curse. To investigate the root causes, we conduct systematic experiments on DLLMs and identify three key reasons: 1) entity fragmentation during training, 2) data asymmetry, and 3) missing entity relations. Motivated by the analysis of these reasons, we propose Diffusion Entity-Relation Modeling (DiffER), which addresses the reversal curse through entity-aware training and balanced data construction. Specifically, DiffER introduces whole-entity masking, which mitigates entity fragmentation by predicting complete entities in a single step. DiffER further employs distribution-symmetric and relation-enhanced data construction strategies to alleviate data asymmetry and missing relations. Extensive experiments demonstrate that DiffER effectively alleviates the reversal curse in Diffusion LLMs, offering new perspectives for future research.