---
id: 20260904-T0-11
title: "Qwen3-4B三值化后处理：存储压缩与部署效能实测"
title_en: "Qwen3-4B Post-Training Ternarization Cuts Storage, Keeps Capability"
url: https://ai.daily.yangsir.net/daily/20260904-T0-11
issue_date: 2026-09-04
publish_date: 2026-09-03T04:00:00.000Z
category: research
source_name: "arXiv cs.AI"
source_url: https://arxiv.org/abs/2609.01962
---

# Qwen3-4B三值化后处理：存储压缩与部署效能实测

arXiv新论文研究Qwen3-4B的端到端后训练三值化转换。作者指出，标称"1.58-bit"并不完全代表实际存储表示、保留能力或运行行为，并系统评估了三值化后的有效比特预算、存储压缩比及部署性能。结果显示该方法在显著降低存储和内存带宽需求的同时，验证了实际能力保留程度。

## English Version

**Qwen3-4B Post-Training Ternarization Cuts Storage, Keeps Capability**

A new arXiv paper examines end-to-end post-training ternarization of Qwen3-4B. The authors argue that the nominal "1.58-bit" label fails to capture stored representation, retained capability, and runtime behavior. They systematically evaluate effective bit budget, storage compression, and deployment metrics, demonstrating significant reductions in storage and memory bandwidth while quantifying actual capability retention.

---

**来源**：[arXiv cs.AI](https://arxiv.org/abs/2609.01962)

**详情页**：https://ai.daily.yangsir.net/daily/20260904-T0-11

---

*智语观潮 · Daily — https://ai.daily.yangsir.net/llms.txt*