Diffusion-GR2: Diffusion Generative Reasoning Re-ranker

2026-07-01

AI Summary

This work addresses the trade-off between inference efficiency and ranking accuracy in generative re-rankers. While autoregressive decoding ensures high ranking fidelity, it suffers from slow inference; directly converting such models into chunk-based diffusion architectures accelerates parallel decoding but often introduces structural and distributional biases that degrade ranking performance. To reconcile this, the authors propose a three-stage training strategy: first, conversion fine-tuning (CFT) enforces valid permutation outputs; second, online policy distillation (OPD) aligns the model’s behavior with autoregressive decoding trajectories; and third, reinforcement learning (RL) optimizes ranking-specific rewards. This approach preserves the parallel decoding advantage of chunk diffusion models while recovering ranking accuracy. Experiments on the Amazon Beauty dataset demonstrate near-autoregressive ranking performance with 2.4–3.5× higher inference throughput.

Abstract

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 $2.4$--$3.5\times$ 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.

Problem

• generative reasoning re-ranking
• block-diffusion language models
• structural gap
• distributional gap
• inference efficiency

Innovation

• block-diffusion
• generative reasoning re-ranker
• conversion fine-tuning
• on-policy distillation
• parallel decoding