AURORA-LM: Autoencoding Unified Representation for Continuous-Latent Diffusion Language Modeling

2026-08-03

AI Summary

This work addresses the fidelity degradation in existing continuous language models caused by adapting diffusion processes for text generation. The authors propose decoupling learnable representation construction from distribution modeling: a query-based encoder-decoder first generates high-capacity, prefix-aligned continuous latent sequences, and a block-causal diffusion Transformer then learns their distribution via flow matching, enabling left-to-right, block-wise parallel denoising generation. Key innovations include preserving high-fidelity latent representations, designing a training objective that constrains only the noisy input while fully predicting the clean target, and introducing self-trajectory consistency with noise-level calibration. Evaluated on OpenWebText and XSum, a 1B-parameter model (≈1500 EFLOPs) outperforms current continuous and diffusion-based language models, with all experiments conducted on Ascend NPUs.

Abstract

Language remains an outlier in generative modeling: while images, video, and audio are increasingly modeled in continuous latent spaces, text generation still relies predominantly on discrete tokens. Existing continuous language models either inherit embedding spaces not designed for joint generation and decoding, or compress autoencoded latents to ease diffusion, sacrificing token-level fidelity. Instead of simplifying the representation to suit the generative model, we preserve a high-capacity, decodable text latent and design the diffusion model to learn its distribution directly. We introduce AURORA-LM, a continuous-latent diffusion language model that separates the construction of a decodable text representation from the modeling of its distribution. A Query-based Encoder-Decoder organizes text into a high-capacity, prefix-aligned latent sequence, and a Block-causal Diffusion Transformer learns its distribution through flow matching, generating blocks left to right while denoising positions within each block in parallel. Because such a latent is harder for diffusion to model, AURORA-LM restricts only the noisy-input pathway while retaining the full clean-latent prediction target, accommodating full-width latents without reducing decoder-facing capacity. We further calibrate the noise-level distribution to the latent width, and introduce self-trajectory consistency to bridge independently sampled training noise and iterative denoising at inference. AURORA-LM achieves the strongest performance among evaluated continuous and diffusion-based language models on OpenWebText free generation and XSum summarization. Scaling to 1B parameters with about 1500 EFLOPs of total compute yields further gains, surpassing a larger publicly released latent-diffusion language model under a matched evaluation protocol. All experiments are conducted on Ascend NPUs.

Problem

• continuous-latent language modeling
• diffusion language models
• text generation
• decodable representation
• token-level fidelity

Innovation

• continuous-latent diffusion
• decodable text representation
• block-causal Diffusion Transformer
• flow matching
• self-trajectory consistency