---
title: "DecisionRank: On-Policy Distillation for Decision-Model Rerankers"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2610.42128v1"
permalink: "https://feed.astralor.com/items/itm_ca35ebb117cd643dc1efe132"
published: "2026-10-30T17:56:58.000Z"
score: 54
channel: "tech"
tags: ["LLM","推理效率","RAG","多模态","DecisionRank","On-Policy Distillation","Decision-Model Rerankers Decision","However"]
---

# DecisionRank: On-Policy Distillation for Decision-Model Rerankers

## At a glance

Decision models target structured tasks by directly predicting choices or scores, avoiding autoregressive decoding latency. However, transferring a generative LLM's ranking knowledge to a decision model remains challeng…

## Key points

- Decision models target structured tasks by directly predicting choices or scores, avoiding autoregressive decoding latency.
- However, transferring a generative LLM's ranking knowledge to a decision model remains challenging: the teacher expresses preferences through a ranked list of document ID sequences, while the student determines the rank…
- To bridge this gap, we introduce DecisionRank, a decision model trained through item-space on-policy distillation (OPD).

## Full text

Decision models target structured tasks by directly predicting choices or scores, avoiding autoregressive decoding latency. However, transferring a generative LLM's ranking knowledge to a decision model remains challenging: the teacher expresses preferences through a ranked list of document ID sequences, while the student determines the ranking using scores from the first-token position. To bridge this gap, we introduce DecisionRank, a decision model trained through item-space on-policy distillation (OPD). Item-space OPD constructs the student's rollout by sorting its document scores into a ranked sequence. A generative teacher takes the resulting ranking list as inputs and provides conditional feedback to directly supervise those scores. Combined with offline ranking supervision, DecisionRank achieves state-of-the-art performance on both general and reasoning-intensive retrieval benchmarks such as TREC DL, BEIR, and BRIGHT. Across four benchmarks, item-space OPD consistently outperforms alternative training objectives. Even at 1.7B parameters, DecisionRank surpasses its 27B generative teacher on TREC DL19--23. At 8B, it outperforms its generative listwise counterpart while running $24.5\times$ faster on TREC DL and $16.7\times$ faster on BRIGHT.

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_ca35ebb117cd643dc1efe132