{
  "title": "Residual paths keep each sublayer input",
  "caption": "Two pre-normalized residual sublayers use attention and a gated MLP. Dashed residual edges keep each sublayer input unchanged. This is a representative decoder, not a universal architecture. The author made this explanatory diagram. The playback is an example. Animation time is not measured runtime.",
  "alt": "Flow diagram: a pre-normalized decoder applies attention and an MLP, each with its unchanged residual input.",
  "sources": [
    {
      "path": "content/blog/2026-06-09_life-of-a-token-transformer/post.md",
      "sha256": "aca6d72d9a2cb565f5e2e514e33cc16c3f73abb24785570c2dd1a2ea672af45d",
      "description": "Current article explanation, equations, assumptions and retained citations."
    },
    {
      "path": "research/dynamic-mechanisms-2026-10-21/before/content/blog/2026-06-09_life-of-a-token-transformer/illustrations/01-decoder-residual-block-algorithm.svg",
      "sha256": "b7165cdb7469f6f1f355ccbf8d49cd8de6b93bdcf37822c92fffcb8ab557221d",
      "description": "Frozen prior illustration; exact pre-migration bytes retained."
    },
    {
      "path": "content/blog/2026-06-09_life-of-a-token-transformer/illustrations/source-notes.json",
      "description": "Existing article illustration provenance, retained unchanged."
    }
  ],
  "limitations": [
    "The playback is an example. Playback duration is not measured runtime."
  ]
}