Scholarlink

<p>This perspective argues that Large Language Models exhibit Model Phase Transitions: performance collapses beyond critical compression thresholds. We analyze structural, numerical, and algebraic redundancy across pruning, quantization, and low-rank decomposition techniques. These sources are orthogonal, enabling a criticality-aware compression framework that achieves near-lossless compression to 10% of the original size. This shift proves that compressing a giant is more effective than training a dwarf for efficient, sustainable AI deployment.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Phase transitions in large language model compression

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年2月6日

查看全文
  • Ziyang Ma,
  • Zuchao Li,
  • Lefei Zhang,
  • Gui-Song Xia,
  • Bo Du,
  • Liangpei Zhang,
  • Dacheng Tao

摘要

This perspective argues that Large Language Models exhibit Model Phase Transitions: performance collapses beyond critical compression thresholds. We analyze structural, numerical, and algebraic redundancy across pruning, quantization, and low-rank decomposition techniques. These sources are orthogonal, enabling a criticality-aware compression framework that achieves near-lossless compression to 10% of the original size. This shift proves that compressing a giant is more effective than training a dwarf for efficient, sustainable AI deployment.