{
  "id": "native-awq-rtf",
  "story": "native-awq-chatterbox-3060",
  "title": "Idle-memory savings came with slower generation",
  "unit": "real-time factor",
  "rows": [
    {
      "label": "Turbo · original",
      "value": 0.446,
      "highlight": false
    },
    {
      "label": "Turbo · native AWQ",
      "value": 0.564,
      "highlight": true
    },
    {
      "label": "English G64 · original",
      "value": 0.888,
      "highlight": false
    },
    {
      "label": "English G64 · native AWQ",
      "value": 1.221,
      "highlight": true
    },
    {
      "label": "Multilingual v2 · original",
      "value": 1.014,
      "highlight": false
    },
    {
      "label": "Multilingual v2 · native AWQ",
      "value": 1.344,
      "highlight": true
    }
  ],
  "scope": "English held-out 50-word story · two warm repetitions · RTX 3060",
  "caveat": "Lower is faster. These are separate model comparisons, not a multilingual quality leaderboard.",
  "kind": "bar",
  "decimals": 3,
  "revision": "7e6fdbe3b1d1aa360346255bfc62945dba694399",
  "sources": [
    {
      "path": "chatterbox/autoround-awq-rtx3060-2026-10-08/README.md",
      "url": "https://github.com/groxaxo/experimentos/blob/7e6fdbe3b1d1aa360346255bfc62945dba694399/chatterbox/autoround-awq-rtx3060-2026-10-08/README.md",
      "sha256": "b984724da30bc0834cb06f209cbd02daec7c44eedd07683d133c50915ee72c77"
    }
  ],
  "origin": "reviewed report table",
  "maximum": null,
  "threshold": null
}