20260514:增加新内容
This commit is contained in:
53
papers/he-urlvr-sharpening-2026.md
Normal file
53
papers/he-urlvr-sharpening-2026.md
Normal file
@@ -0,0 +1,53 @@
|
||||
---
|
||||
title: "How Far Can Unsupervised RLVR Scale LLM Training?"
|
||||
created: 2026-05-01
|
||||
updated: 2026-05-01
|
||||
type: paper
|
||||
tags: []
|
||||
sources: []
|
||||
---
|
||||
|
||||
# How Far Can Unsupervised RLVR Scale LLM Training?
|
||||
|
||||
- **arXiv**: 2603.08660
|
||||
- **作者**: He, Zuo, Liu, et al. (22 authors, Tsinghua/Shanghai AI Lab)
|
||||
- **会议**: ICLR 2026
|
||||
- **标签**: #RLVR #unsupervised-learning #reward-hacking
|
||||
|
||||
## 中文摘要
|
||||
|
||||
URLVR(无监督可验证奖励强化学习)被视为突破 LLM 训练监督瓶颈的希望。然而这篇 ICLR 2026 论文通过分类学+理论+大规模实验论证了一个核心发现:**所有内在奖励方法本质上都在做同一件事——锐化模型的初始分布**。这个机制在模型自信且正确时奏效,但在自信却错误时灾难性放大偏见。实验表明内在奖励统一遵循 rise-then-fall 模式,崩溃时间由模型先验决定而非工程选择。作者提出 [[model-collapse-step|Model Collapse Step]] 作为衡量模型先验的实用指标,并探索 [[self-verification-rewards|self-verification]] 作为外部奖励路径的突破。
|
||||
|
||||
## 核心问题
|
||||
|
||||
监督 RLVR(DeepSeek-R1、Gemini 2.5、Qwen3)很强大,但依赖 ground truth 标签——而随着模型逼近甚至超越人类专家水平,获取可靠标签越来越不可行。URLVR 试图通过无标签奖励突破这一瓶颈。**本文提出根本问题:内在 URLVR 真的能规模化 LLM 训练吗?**
|
||||
|
||||
## 方法论贡献
|
||||
|
||||
1. **URLVR 分类法**: [[certainty-based-rewards|确定性奖励]] vs [[ensemble-based-rewards|集成奖励]],前者从策略置信度(logits/熵)推导,后者从多样本一致性(多数投票)推导
|
||||
2. **[[intrinsic-rewards-sharpening|Sharpening 统一理论]]**: 从 KL 正则化 RL 目标出发,推导出所有内在方法的闭式解都收敛于锐化初始分布——仅放大已有偏好,不发现新知识
|
||||
3. **Rise-then-Fall 定律**: 无论内在方法具体设计如何,始终先升后降,崩溃不可避免
|
||||
4. **[[model-collapse-step|Model Collapse Step (MCS)]]**: 衡量模型在内在 URLVR 下能维持多久才崩溃的步数,比 pass@k 更准确,无需 ground truth
|
||||
|
||||
## 关键发现
|
||||
|
||||
| 发现 | 含义 |
|
||||
|------|------|
|
||||
| 内在奖励统一锐化初始分布 | 无法超越模型已有知识 |
|
||||
| Rise-then-Fall 是必然模式 | 崩溃时间反映模型先验 |
|
||||
| MCS 预测 RL 可训练性 | 低成本基模型选择替代方案 |
|
||||
| [[self-verification-rewards|Self-verification]] 无崩溃 | 外部奖励可能突破天花板 |
|
||||
|
||||
## 相关概念
|
||||
- [[test-time-training-rl]]
|
||||
- [[rlvr-unified-framework]]
|
||||
- [[confidence-correctness-alignment]]
|
||||
|
||||
- [[unsupervised-rlvr]] — URLVR 范式定义
|
||||
- [[intrinsic-rewards-sharpening]] — Sharpening 机制(理论核心)
|
||||
- [[model-collapse-step]] — MCS 指标
|
||||
- [[self-verification-rewards]] — 外部奖励突破
|
||||
- [[reward-hacking-llm]] — 奖励黑客与模型崩溃
|
||||
- [[certainty-based-rewards]] — 确定性奖励
|
||||
- [[ensemble-based-rewards]] — 集成奖励/多数投票
|
||||
- [[generation-verification-asymmetry]] — 生成-验证不对称性
|
||||
Reference in New Issue
Block a user