arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2610.20635cs.CLcs.AIcs.LG
收藏

探索排列扩散大语言模型(LLMs)的设计空间

Exploring the Design Space of Permutation Diffusion LLMs

Meta
Zhe Xu, Mingfu Liang, Kavosh Asadi, Jean-Pascal Billaud, Tianyi Chen, Yanhui Guo, Christina Huang, Baekjin Kim, Jimmy Kim, Yufei Li, Daqi Sun, Jiachen Sun, Dian… 展开作者

Zhe Xu, Mingfu Liang, Kavosh Asadi, Jean-Pascal Billaud, Tianyi Chen, Yanhui Guo, Christina Huang, Baekjin Kim, Jimmy Kim, Yufei Li, Daqi Sun, Jiachen Sun, Dian Yu, Zaiwei Zhang, Kaushik Rangadurai, Frank Shyu, Sandeep Pandey, Xi Liu, Luke Simon

AI总结:

该研究针对语言模型排列生成问题,提出三种扩散大语言模型参数化方案,在亚马逊推荐重排序任务上验证了方案性能,表明需联合设计排列扩散的关键组件。

摘要:

用语言模型生成排列,既需要文本条件下的排序,也需要候选与输出位置之间的全局一对一分配。我们通过三种扩散大语言模型(diffusion-LLM)参数化方式研究该问题:对打乱的候选索引序列进行硬分配预测、对双重随机矩阵进行软分配预测,以及直接预测排序分数后再排序。这三类方法在预测和解码的不同阶段施加排列约束,因此支持不同的训练目标和采样流程。我们在三个亚马逊领域的推荐重排序任务上评估这些设计,候选集大小为10和20。在1次函数评估时,硬分配和软分配变体的Recall@1与受约束掩码扩散基线相当,而直接排序分数预测无需匈牙利分配或Sinkhorn归一化即可保证有效排列。更多函数评估并未持续提升排序质量,其效果强烈依赖于状态表示和训练目标。这些结果表明,对于基于语言模型去噪器的排列扩散,应联合设计 corruption(损坏)、预测参数化、目标函数和解码器。

英文摘要:

Permutation generation with language models requires both text-conditioned ranking and a global one-to-one assignment between candidates and output positions. We study this problem through three diffusion-LLM parameterizations: hard-assignment prediction over shuffled candidate-index sequences, soft-assignment prediction over doubly stochastic matrices, and direct ordering-score prediction followed by sorting. The three families place permutation constraints at different stages of prediction and decoding and therefore support different training objectives and sampling procedures. We evaluate these designs on recommendation reranking across three Amazon domains with candidate sets of size 10 and 20. At one function evaluation, the hard- and soft-assignment variants achieve Recall@1 comparable to a constrained masked-diffusion baseline, whereas direct ordering-score prediction guarantees valid permutations without Hungarian assignment or Sinkhorn normalization. Additional function evaluations do not consistently improve ranking quality, and their effect depends strongly on the state representation and training objective. These results show that corruption, prediction parameterization, objective, and decoder should be designed jointly for permutation diffusion with language-model denoisers.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑