{
  "id": "codec-decode",
  "story": "moss-tts-awq-fp8-codec-deploy-2026-09-22",
  "title": "GPU decode shortened the final stage",
  "unit": "seconds",
  "rows": [
    {
      "label": "CPU INT8 decode",
      "value": 3.36,
      "highlight": false
    },
    {
      "label": "GPU FP8 decode",
      "value": 0.24,
      "highlight": true
    }
  ],
  "sources": [
    {
      "path": "moss-tts-awq/fp8-codec-deploy-2026-09-22/README.md",
      "url": "https://github.com/groxaxo/experimentos/blob/6550ead3945b7ecacbaac1e0767edc49a9de9747/moss-tts-awq/fp8-codec-deploy-2026-09-22/README.md",
      "sha256": "d194ebb425c44dbaadf83a92546a80c91640bb84e49a1d326ee090032c1e7aef"
    }
  ],
  "revision": "6550ead3945b7ecacbaac1e0767edc49a9de9747",
  "scope": "Seed-matched recorded deployment A/B · FP32 reference encoder on CPU",
  "caveat": "Report-table values. The archived before JSON is empty; these are not presented as recovered raw logs.",
  "kind": "bar",
  "maximum": null,
  "threshold": null,
  "decimals": 2,
  "origin": "reported table"
}