{
  "schema_version": "0.3.0",
  "id": "s3-lfm2.5-2.6b-8-bit-magnitude-256",
  "methodology_version": "0.6",
  "provenance": {
    "kind": "measured",
    "session_id": "s3-lfm2.5-2.6b-8-bit",
    "fixture_notice": null,
    "operator": {
      "name": "Glenn Sonna",
      "organization": "Xybrid",
      "affiliation": "xybrid"
    },
    "run_at": "2026-11-12T22:29:10Z",
    "harness": {
      "name": "xybrid-bench",
      "version": "0.4.0",
      "commit": "777cb3d9fecaa7cc1a30694d98492c33dbc6d693"
    }
  },
  "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": "magnitude",
    "version": "0.2.17",
    "build": {
      "source": "release",
      "flags": [],
      "dependencies": {
        "cli_sha256": "a48e1ec5ebe769d7c7432dc1f6d7d4e07ea0750cab2e37ef7ff0dc9cdf1a78e4"
      },
      "commit": null,
      "binary_sha256": "1cf3297910bc71b824c68461b1c45ecc85dc1942ba91fb67ac3a6d1dfd3de353"
    },
    "backend": "metal",
    "threads": null,
    "runtime_flags": {
      "unsupported": "Magnitude's chat API renders LFM2.5 2.6B's own template, which always opens a thinking block, so it streams reasoning and can't run the closed-think prompts (measured in qualification)."
    },
    "interface": "http-server"
  },
  "model": {
    "id": "lfm2.5-2.6b",
    "artifact": {
      "repo": "LiquidAI/LFM2.5-2.6B-GGUF",
      "revision": "e7caca5d835a3901a8e0d63e94009429bafafdfc",
      "files": [
        {
          "path": "LFM2.5-2.6B-Q8_0.gguf",
          "sha256": "1e22128dfa128bdfb684da167e74e072d0a056baa7d06d9f280291e2839b0fc9",
          "size_bytes": 2874779648
        },
        {
          "path": "README.md",
          "sha256": "d08b79260b3998e96792777085e4175b730667568f3a953ad4ac9099cfa27056",
          "size_bytes": 3068
        }
      ],
      "format": "gguf",
      "quantization": {
        "scheme": "Q8_0",
        "class": "8-bit",
        "weight_bits": 8,
        "group_size": 32,
        "kv_cache": "magnitude default"
      },
      "conversion": {
        "method": "upstream",
        "tool": null,
        "tool_version": null,
        "command": null,
        "publisher": "LiquidAI"
      }
    }
  },
  "workload": {
    "id": "archive-256-128",
    "kind": "text-generation",
    "prompt": {
      "tokens": 256,
      "source": "prompts/source.txt",
      "sha256": "c10d1b7bef62d95322f8f8d229dc66118d9375107bdc49b06f76287d058b2f6c",
      "template": "ea663864491de7ade391839479860ca95541f892f72665c73251fbd4643b1bef",
      "construction": "chat-envelope-user-truncate-v1",
      "token_ids_sha256": "e581da70677499b28d937245b5a1f094115531fbe9fae133fed90dc27ee9ddfb"
    },
    "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": "Magnitude's chat API renders LFM2.5 2.6B's own template, which always opens a thinking block, so it streams reasoning and can't run the closed-think 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/s3-lfm2.5-2.6b-8-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/s3-lfm2.5-2.6b-8-bit.jsonl",
      "sha256": "92db4b58d89343f623920b0ee530cfa9bbae7e105b9df1baa32441dd17e17af3"
    }
  ]
}
