LLaDA2.0: Scaling Up Diffusion Language Models to 100B

TL;DR

LLaDA2.0 introduces a scalable method to convert pre-trained auto-regressive language models into large discrete diffusion models up to 100 billion parameters, enhancing efficiency and performance for deployment.

Contribution

The paper presents a novel 3-phase block-level WSD training scheme for converting AR models into diffusion LLMs, enabling scalable, knowledge-preserving model expansion.

Findings

Achieved 100B parameter diffusion models with superior performance.

Demonstrated effective conversion from AR to diffusion models.

Open-sourced models for practical deployment.

Abstract

This paper presents LLaDA2.0 -- a tuple of discrete diffusion large language models (dLLM) scaling up to 100B total parameters through systematic conversion from auto-regressive (AR) models -- establishing a new paradigm for frontier-scale deployment. Instead of costly training from scratch, LLaDA2.0 upholds knowledge inheritance, progressive adaption and efficiency-aware design principle, and seamless converts a pre-trained AR model into dLLM with a novel 3-phase block-level WSD based training scheme: progressive increasing block-size in block diffusion (warm-up), large-scale full-sequence diffusion (stable) and reverting back to compact-size block diffusion (decay). Along with post-training alignment with SFT and DPO, we obtain LLaDA2.0-mini (16B) and LLaDA2.0-flash (100B), two instruction-tuned Mixture-of-Experts (MoE) variants optimized for practical deployment. By preserving the advantages of parallel decoding, these models deliver superior performance and efficiency at the frontier scale. Both models were open-sourced.