---
title: "推理大模型中的测试时扩展：推理机制、评估与可复现性"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2608.04001v1"
permalink: "https://feed.astralor.com/items/itm_8c30a3a81cd4bb356b654fc5"
published: "2026-08-04T17:57:20.000Z"
score: 77
channel: "ai"
tags: ["arXiv","可复现性","开源权重生态","推理大模型","测试时扩展"]
---

# 推理大模型中的测试时扩展：推理机制、评估与可复现性

## AI 点评

> 这篇论文把“测试时扩展”从口号变成了可操作的框架。别再被单个准确率数字迷惑了——如果你在对比不同推理模型，或者计划复现某个 scaling 实验，这套评估概况和协议匹配要求会成为你的新检查清单。尤其那 20 亿条推理轨迹，值得尽早关注。

## 速览

本论文系统梳理了推理大模型中的测试时扩展（test-time scaling）研究，指出当前多种推理算法在统计结构、计算核算和失败模式上差异显著，不能仅用单一预算标量或简单准确率相互比较。作者将测试时扩展形式化为自回归模型隐式前缀树上的预算推理，区分单轨迹、叶级和前缀级三种结构机制；提出以整个推理系统为评估对象，引入评估概况与协议匹配的报告方式；并定义推理协议的可复现性要求，区分精确重放与分布式可复现。论文还组织开源权重推理生态，并发布超 20 亿推理轨迹。

## 关键点

- 将测试时扩展形式化为隐式前缀树上的预算推理，区分单轨迹顺序扩展、叶级扩展与终端归约、前缀级扩展三种机制。
- 提出以整个推理系统为评估对象，用评估概况统一常见重复采样指标，并要求协议匹配的计算与不确定性报告。
- 规定推理协议的可复现性要求，区分精确重放与分布式可复现，并给出所需人工制品清单。
- 组织开源权重推理生态，发布超过 20 亿条完整推理轨迹供研究使用。

## 全文

Large language models can solve substantially harder reasoning problems with more inference-time compute. The term "test-time scaling," however, now covers diverse inference algorithms that extend deliberation along a single trajectory, sample completed candidates and aggregate them through voting or verification, or search over unfinished partial states. These algorithms differ in their statistical structure, compute accounting, and failure modes. Treating these procedures as interchangeable under a single scalar "budget," or reporting accuracy without the inference protocol that produced it, makes results difficult to compare across studies. We develop a systematic account of test-time scaling along three axes. First, we formalize test-time scaling as budgeted inference over the implicit prefix tree of an autoregressive model and distinguish three structural regimes: single-trajectory sequential scaling, leaf-level scaling with terminal reduction, and prefix-level scaling. Second, we treat the evaluated object as the entire inference system and develop evaluation principles that separate end-to-end system performance from candidate-bank diagnostics. We introduce an evaluation profile whose coordinates and simple functionals recover or bound common repeated-sampling metrics, and prescribe protocol-matched reporting of compute and uncertainty. Third, we specify reproducibility requirements for inference protocols, distinguishing exact replay from distributional reproducibility and identifying the artifacts needed to support each. We also organize the open-weight reasoning ecosystem by model-side and interface mechanisms, apply these principles to broad-knowledge, symbolic-reasoning, and competition-mathematics benchmarks, and assemble over 2 billion full reasoning traces for release with progressively richer verifier and token-level signals.

---

引用前请核对原文;本文由星际流动整理:https://feed.astralor.com/items/itm_8c30a3a81cd4bb356b654fc5