首个支持并发多模态生成的非自回归模型，可同时输出文本和图像。
OneFlow: Concurrent Mixed-Modal and Interleaved Generation with Edit Flows
- 用插入式编辑流与扩散匹配结合，实现文本与图像并行生成。
- 在1B到8B模型上，生成效率提升50%，理解任务表现更优。
- 适合需要迭代优化和自然推理的多模态应用开发。
我们提出OneFlow，首个支持变长且并发混合模态生成的非自回归多模态模型。不同于强制文本与图像生成保持严格因果顺序的自回归模型，OneFlow将基于插入的编辑流（Edit Flow）用于离散文本标记，结合流匹配（Flow Matching）生成图像潜在表示。该模型通过分层采样机制优先保障内容完整性而非语法正确性，实现文本与图像的并发合成。在1B至8B规模的模型上进行的受控实验表明，OneFlow在生成与理解任务中均优于自回归基线，训练浮点运算量最多减少50%。其性能超越自回归与基于扩散的方法，同时实现了并发生成、迭代优化及类推理式生成等新能力。
原文摘要 · Abstract (English)
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.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。