arXiv cs.CL· Yuki Otsuka, Daisuke Oba, Naoaki Okazaki·· 12 小时前AI 评分33
ReLaDrift：用漂移模型实现循环语言生成以支持测试时扩展
ReLaDrift: Recurrent Language Generation via Drifting for Test-Time Scaling
AI 导读
ReLaDrift 提出用共享 Drifting Model 从噪声生成初稿，并在连续隐空间中迭代精炼，从而实现测试时计算与生成质量之间的权衡。训练时随机采样轮数、仅在最后一轮施加 drifting loss 并截断前序轮梯度，使反向传播只需一轮。在 Sudoku Hard 和 GSM8K 上，其表现与扩散及流式语言模型等迭代生成方法相当或更优。
正文 · 原文
Abstract:Drifting Models generate a sample with a single function evaluation, making them attractive for fast parallel language generation, but their one-step formulation does not support iterative refinement. We propose Recurrent Language Generation via Drifting (ReLaDrift), which uses a shared Drifting Model to generate an initial draft from noise and iteratively refine it in a continuous latent space. We train this recurrence by sampling the number of rounds and applying the drifting loss only at the final round, stopping gradients through preceding rounds so that only one round requires backpropagation. Unlike the one-step Drifting Model, whose accuracy does not improve when iterated, ReLaDrift enables a trade-off between test-time computation and generation quality, as diffusion and flow language models do. On Sudoku Hard and GSM8K, it achieves competitive or superior performance compared with these iterative-generation approaches across a range of inference budgets. Our code is available at this https URL
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2610.23225 [cs.CL] |
| (or arXiv:2610.23225v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2610.23225 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Yuki Otsuka [view email]
[v1]
Fri, 16 Oct 2026 09:55:22 UTC (182 KB)
来源：arXiv cs.CL · arxiv.org