# Why Any-Order Autoregressive Models Need Two-Stream Attention: A Structural-Semantic Tradeoff

arXiv: https://arxiv.org/abs/2602.16092 · HTML: https://mlsanity.com/paper/2602.16092/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Discrete and text diffusion, one of 14 directions in Diffusion models.
- Builds on 10 of its 10 references that are in the map.
- First posted on arXiv 17 Feb 2026.

## Placed in

- Diffusion models / [Discrete and text diffusion](https://mlsanity.com/diffusion/discrete-and-text-diffusion/)

## Builds on (most cited first, 10 of 10)

- [RoFormer: Enhanced Transformer with Rotary Position Embedding](https://mlsanity.com/paper/2104.09864/) (2021)
- [XLNet: Generalized Autoregressive Pretraining for Language Understanding](https://mlsanity.com/paper/1906.08237/) (2019)
- [Simple and Effective Masked Diffusion Language Models](https://mlsanity.com/paper/2406.07524/) (2024)
- [Simplified and Generalized Masked Diffusion for Discrete Data](https://mlsanity.com/paper/2406.04329/) (2024)
- [Your Absorbing Discrete Diffusion Secretly Models the Conditional Distributions of Clean Data](https://mlsanity.com/paper/2406.03736/) (2024)
- [Esoteric Language Models: A Family of Any-Order Diffusion LLMs](https://mlsanity.com/paper/2506.01928/) (2025)
- [Any-Order GPT as Masked Diffusion Model: Decoupling Formulation and Architecture](https://mlsanity.com/paper/2506.19935/) (2025)
- [Reviving Any-Subset Autoregressive Models with Principled Parallel Sampling and Speculative Decoding](https://mlsanity.com/paper/2504.20456/) (2025)
- [Autoregressive Models Rival Diffusion Models at ANY-ORDER Generation](https://mlsanity.com/paper/2601.13228/) (2026)
- [Auto-Regressive Masked Diffusion Models](https://mlsanity.com/paper/2601.16971/) (2026)

## Abstract (arXiv)

Any-order autoregressive models (AO-ARMs) offer a promising path toward efficient masked diffusion by enabling native key-value caching, but competitive performance has so far required two-stream attention, typically motivated as a means of decoupling token content from position. In this work, we argue that two-stream attention may be serving a more subtle role. We identify a structural-semantic tradeoff in any-order generation: the hidden representation at each step must simultaneously attend to semantically informative tokens for prediction and structurally recent tokens for summarization, objectives that compete for attention capacity in a single stream but can specialize across two streams. To isolate this tradeoff from position-content separation, we propose Decoupled RoPE, a modification to rotary position embeddings that provides target position information without revealing target content. Decoupled RoPE performs competitively at short sequence lengths--where semantic and structural proximity coincide--but degrades as sequence length increases and the two orderings diverge. These results suggest that the success of two-stream attention stems not merely from separating position from content, but from circumventing the deeper structural-semantic tradeoff inherent to any-order generation.