中文

Diffusion-GR2: Diffusion Generative Reasoning Re-ranker

信息检索 2026-07-01 v1 人工智能

摘要

Generative reasoning re-rankers achieve strong recommendation accuracy by emitting a chain-of-thought before re-ordering a candidate list, but they are slow at inference: an autoregressive (AR) decoder spends one sequential forward pass per reasoning token, and the reasoning trace far exceeds the ranking it produces. To reduce this cost, block-diffusion language models decode many positions in parallel over a few denoising steps and are substantially faster, yet naively converting an AR re-ranker into one opens two accuracy gaps: (1) a structural gap: answer positions are denoised in parallel and scored independently, so the decoder emits invalid rankings (duplicated, dropped, or out-of-set identifiers) that AR avoids through left-to-right masking; and (2) a distributional gap: fine-tuning the converted model on fixed teacher trajectories is off-policy relative to its own decoding at inference, leaving a residual accuracy gap. To close both gaps while keeping the speedup, we propose \textbf{Diffusion-GR2}, a recipe that converts our AR reasoning re-ranker (GR2) into a block-diffusion re-ranker. First, conversion fine-tuning (CFT) adapts the AR-initialized diffusion model to denoise the answer into a valid permutation on its own, without an external constrained decoder. Next, on-policy distillation (OPD) then supervises the model on its own decoded trajectories with dense per-token targets from the AR teacher. Finally, we apply a reinforcement-learning (RL) stage against a re-ranking reward on top of OPD's on-policy policy. Experiments on Amazon Beauty demonstrate that Diffusion-GR2 recovers to near-parity with the AR re-ranker, while block-parallel decoding raises decode throughput by 2.42.4--3.5×3.5\times at the model's reasoning output length. Ablations show that CFT recovers most of the conversion gap, and that on-policy distillation further closes it to the AR reference.

引用

@article{arxiv.2607.01170,
  title  = {Diffusion-GR2: Diffusion Generative Reasoning Re-ranker},
  author = {Zhuoxuan Zhang and Kangqi Ni and Yuhang Chen and Mingfu Liang and Xiaohan Wei and Yunchen Pu and Fei Tian and Chonglin Sun and Frank Shyu and Adam and Song and Sandeep Pandey and Luke Simon and Tianlong Chen and Xi Liu},
  journal= {arXiv preprint arXiv:2607.01170},
  year   = {2026}
}

备注

Work in progress