Generalized Discrete Diffusion with Self-Correction
About
Self-correction is an effective technique for maintaining parallel sampling in discrete diffusion models with minimal performance degradation. Prior work has explored self-correction at inference time or during post-training; however, such approaches often suffer from limited generalization and may impair reasoning performance. GIDD pioneers pretraining-based self-correction via a multi-step BERT-style uniform-absorbing objective. However, GIDD relies on a continuous interpolation-based pipeline with opaque interactions between uniform transitions and absorbing masks, which complicates hyperparameter tuning and hinders practical performance. In this work, we propose a Self-Correcting Discrete Diffusion (SCDD) model to reformulate pretrained self-correction with explicit state transitions and learn directly in discrete time. Our framework also simplifies the training noise schedule, eliminates a redundant remasking step, and relies exclusively on uniform transitions to learn self-correction. Experiments at the GPT-2 scale demonstrate that our method enables more efficient parallel decoding while preserving generation quality.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Language Modeling | OWT | Gen. PPL55.7 | 61 | |
| Language Modeling | LM1B | PPL (Generalized)102.6 | 55 | |
| Language Modeling | LM1B (val) | Perplexity39.16 | 55 | |
| Language Modeling | OpenWebText (OWT) (val) | Perplexity28.41 | 42 | |
| Language Modeling | Language Modeling Benchmarks (ARC, BoolQ, Hellaswag, OBQA, PIQA, WinoG) zero-shot | ARC-E Accuracy26.64 | 5 |