LLaDA2.1: Speeding Up Text Diffusion via Token Editing

arxiv:2602.08676

Original source · Paper JSON

LLMs: architectures, context, training · 2026-02-09

Cites (1) · Cited by (1)

Complete cleaned Markdown · Source Markdown including bibliography

Read in parts

1. Part 1
2. Part 2

Contents

• LLaDA2.1: Speeding Up Text Diffusion via Token Editing
• Abstract
• 1 Introduction
• 2 Configurable Decoding Scheme
• 3 Training Paradigm
• 3.1 Training Alignment for “Draft-and-Edit”
• 3.2 Reinforcement Learning Training
• 4 Infrastructure
• 4.1 Training Infrastructure
• Continued Pre-Training and Supervised Fine-Tuning
• RL Training
• 4.2 Inference Infrastructure
• 4.3 Decoding Algorithm at Inference
• 5 Evaluation
• 6 Outlook and Limitation
• Tradeoff Between Inference Speed and Accuracy
• Editable Enhanced dLLM
• LLaDA2.1 remains in an experimental phase. Although rare, certain edge cases may occur.
• Conclusion