OneFlow: Concurrent Mixed-Modal and Interleaved Generation with Edit Flows
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866912756641300480 |
|---|---|
| author | Nguyen, John Havasi, Marton Berrada, Tariq Zettlemoyer, Luke Chen, Ricky T. Q. |
| author_facet | Nguyen, John Havasi, Marton Berrada, Tariq Zettlemoyer, Luke Chen, Ricky T. Q. |
| contents | We present OneFlow, the first non-autoregressive multimodal model that enables variable-length and concurrent mixed-modal generation. Unlike autoregressive models that enforce rigid causal ordering between text and image generation, OneFlow combines an insertion-based Edit Flow for discrete text tokens with Flow Matching for image latents. OneFlow enables concurrent text-image synthesis with hierarchical sampling that prioritizes content over grammar. Through controlled experiments across model sizes from 1B to 8B, we demonstrate that OneFlow outperforms autoregressive baselines on both generation and understanding tasks while using up to 50% fewer training FLOPs. OneFlow surpasses both autoregressive and diffusion-based approaches while unlocking new capabilities for concurrent generation, iterative refinement, and natural reasoning-like generation. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2510_03506 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | OneFlow: Concurrent Mixed-Modal and Interleaved Generation with Edit Flows Nguyen, John Havasi, Marton Berrada, Tariq Zettlemoyer, Luke Chen, Ricky T. Q. Artificial Intelligence We present OneFlow, the first non-autoregressive multimodal model that enables variable-length and concurrent mixed-modal generation. Unlike autoregressive models that enforce rigid causal ordering between text and image generation, OneFlow combines an insertion-based Edit Flow for discrete text tokens with Flow Matching for image latents. OneFlow enables concurrent text-image synthesis with hierarchical sampling that prioritizes content over grammar. Through controlled experiments across model sizes from 1B to 8B, we demonstrate that OneFlow outperforms autoregressive baselines on both generation and understanding tasks while using up to 50% fewer training FLOPs. OneFlow surpasses both autoregressive and diffusion-based approaches while unlocking new capabilities for concurrent generation, iterative refinement, and natural reasoning-like generation. |
| title | OneFlow: Concurrent Mixed-Modal and Interleaved Generation with Edit Flows |
| topic | Artificial Intelligence |
| url | https://arxiv.org/abs/2510.03506 |