{
  "schema_version": "0.3.0",
  "id": "s2-qwen3.8-27b-4-bit-uzu-256",
  "methodology_version": "0.6",
  "provenance": {
    "kind": "measured",
    "session_id": "s2-qwen3.8-27b-4-bit",
    "fixture_notice": null,
    "operator": {
      "name": "Glenn Sonna",
      "organization": "Xybrid",
      "affiliation": "xybrid"
    },
    "run_at": "2026-11-09T23:53:13Z",
    "harness": {
      "name": "xybrid-bench",
      "version": "0.4.0",
      "commit": "ccb0a8ac606e19454c16dd0b625e29509c726b72"
    }
  },
  "publication": {
    "status": "published",
    "published_at": "2026-11-16T10:23:29Z",
    "reviewed_by": [
      "Glenn Sonna"
    ],
    "changelog": []
  },
  "device": {
    "id": "glenn-m5-max-64gb",
    "os": {
      "build": "26B123",
      "kernel": "27.2.0",
      "name": "macOS",
      "version": "27.2.0"
    },
    "power": {
      "low_power_mode": false,
      "performance_mode": null,
      "source": "mains"
    }
  },
  "engine": {
    "id": "uzu",
    "version": "0.6.17",
    "build": {
      "source": "package",
      "flags": [],
      "dependencies": {
        "uzu": "0.6.17",
        "python": "3.12.12",
        "python_sha256": "3874a935f7242b660e652d35c25a1b87415fcfea3ee191ff262fcca5c50102c5",
        "worker_sha256": "2bad32c14d1c34d966e3ac704fd7fce0b53eb25605db93c61f9b1c563902928d"
      },
      "commit": null,
      "binary_sha256": "7829db228dc9c1a80eb71f2146724ab39c37a2dea37aa760b0cb1558b88e4a7c"
    },
    "backend": "metal",
    "threads": null,
    "runtime_flags": {
      "unsupported": "uzu 0.6.17 renders its own Qwen3.8 chat template: 67 input tokens for the 27-token canonical check prompt, so it can't run the frozen prompts (measured in qualification)."
    },
    "interface": "python-worker"
  },
  "model": {
    "id": "qwen3.8-27b",
    "artifact": {
      "repo": "trymirai/Qwen3.8-27B-M",
      "revision": "c991a5c5cd99d9528e9f3e83302e82c40f1436ab",
      "files": [
        {
          "path": "README.md",
          "sha256": "b6955d23742ad1fe12c852cef7124e52ab0e490d5d7155a1aef56898b705c01d",
          "size_bytes": 3444
        },
        {
          "path": "config.json",
          "sha256": "60759c66454c7f48c53bdc7e932d685a80f846e3f956aecbb5ac9374ef447500",
          "size_bytes": 140613
        },
        {
          "path": "model.safetensors",
          "sha256": "79e170fdb07a5410e4e0859ddd302cd76acecec91142030bfeb1eb533192f9a3",
          "size_bytes": 14532457872
        },
        {
          "path": "tokenizer.json",
          "sha256": "87a7830d63fcf43bf241c3c5242e96e62dd3fdc29224ca26fed8ea333db72de4",
          "size_bytes": 19989343
        }
      ],
      "format": "mirai",
      "quantization": {
        "scheme": "mirai-m",
        "class": "4-bit",
        "weight_bits": 4,
        "group_size": null,
        "kv_cache": "uzu default"
      },
      "conversion": {
        "method": "upstream",
        "tool": null,
        "tool_version": null,
        "command": null,
        "publisher": "trymirai"
      }
    }
  },
  "workload": {
    "id": "archive-256-128",
    "kind": "text-generation",
    "prompt": {
      "tokens": 256,
      "source": "prompts/source.txt",
      "sha256": "c10d1b7bef62d95322f8f8d229dc66118d9375107bdc49b06f76287d058b2f6c",
      "template": "af16f654968d47d09a6c604e00370b12dbc21f76456f59c9e98b82e84d9f6375",
      "construction": "chat-envelope-user-truncate-v1",
      "token_ids_sha256": "2e898ab96d98564299d66aec466dac46bb4e9dbd3255a01187d6184b1ce3bdd0"
    },
    "output_tokens": 128,
    "stop_on_eos": false,
    "context_length": 4096,
    "batch_size": 1,
    "sampling": {
      "temperature": 0,
      "top_p": 1,
      "top_k": 1,
      "seed": 1,
      "greedy": true
    },
    "measurement": "streaming",
    "warmup_runs": 2,
    "trials_planned": 6,
    "cooldown_s": 15
  },
  "conditions": {
    "ambient_c": null,
    "thermal": {
      "start": {
        "celsius": null,
        "state": "nominal"
      },
      "end": {
        "celsius": null,
        "state": "nominal"
      },
      "gate": "macOS thermal state nominal",
      "throttling_detected": null
    }
  },
  "trials": [],
  "quality": {
    "method": null,
    "status": "not-run",
    "reference": null,
    "reference_sha256": null,
    "threshold": null,
    "items": [],
    "detail": null
  },
  "status": "unsupported",
  "errors": [],
  "unsupported": {
    "reason": "uzu 0.6.17 renders its own Qwen3.8 chat template: 67 input tokens for the 27-token canonical check prompt, so it can't run the frozen prompts (measured in qualification).",
    "evidence_url": null
  },
  "caveats": [
    "Publisher cards identify the base model but do not always pin the exact export source checkpoint revision.",
    "Engines may differ in quantization scheme, effective KV cache precision, kernels, and activations."
  ],
  "reproduction": {
    "commands": [
      "pnpm bench run --plan plans/s2-qwen3.8-27b-4-bit.json"
    ],
    "environment": {},
    "notes": "Three palindromic blocks per quantization class and workload; load, two warmups, one trial, unload per turn."
  },
  "artifacts": [
    {
      "kind": "raw",
      "label": "Complete finalized session evidence",
      "url": "https://www.xybrid.ai/leaderboard/raw/s2-qwen3.8-27b-4-bit.jsonl",
      "sha256": "c87d98b1abf198536317e28b584f964c536a5eba6c27e65b755cee2bc87ff9b1"
    }
  ]
}
