{
  "title": "矩阵乘法",
  "total": 1,
  "posts": [
    {
      "title": "算法解读：RMM——TopK 列范数切片：公式、1B~70B 实测与注意力/MLP 不对称性",
      "url": "/posts/deep-code-rmm/",
      "permalink": "https://hackcv.com/posts/deep-code-rmm/",
      "date": "2026-08-23",
      "author": "hackcv",
      "description": "RMM 完整拆解：收缩维 TopK 列范数选择算法、minimax 最优性证明、retention-ratio 权衡；8 个基准 × 4 档保留比实测、注意力 vs MLP 不对称数据、A100 端到端 1.40× 加速。",
      "categories": ["研究简报"],
      "tags": ["AI","推理优化","矩阵乘法","RMM","算法解读"],
      "cover": "https://picsum.photos/seed/%E7%AE%97%E6%B3%95%E8%A7%A3%E8%AF%BBrmmtopk-%E5%88%97%E8%8C%83%E6%95%B0%E5%88%87%E7%89%87%E5%85%AC%E5%BC%8F1b~70b-%E5%AE%9E%E6%B5%8B%E4%B8%8E%E6%B3%A8%E6%84%8F%E5%8A%9B/mlp-%E4%B8%8D%E5%AF%B9%E7%A7%B0%E6%80%A7/1200/675",
      "readingTime": 2
    }
  ]
}