Bonsai 27B benchmark on an M4 Max · 36 GB

{
  "runId": "run_2b1857a8-ce4f-445f-8b2b-07587dcc49fd",
  "bundleId": "mlx-ternary-bonsai-27b-mlx-2bit-d08217",
  "status": "verified",
  "promptTokens": 40960,
  "completionTokens": 10240,
  "contextLength": 5120,
  "harness": {
    "version": "0.1.21",
    "gitSha": "unknown"
  },
  "runtime": {
    "name": "mlx_lm",
    "version": "0.31.3",
    "buildFlags": null
  },
  "model": {
    "displayName": "Bonsai 27B",
    "format": "mlx",
    "quant": "2bit",
    "architecture": "qwen3_5",
    "source": "prism-ml/Ternary-Bonsai-27B-mlx-2bit",
    "fileSizeBytes": 8490785104,
    "lab": {
      "name": "PrismML",
      "slug": "prismml"
    },
    "quantizedBy": {
      "name": "PrismML",
      "slug": "prismml"
    }
  },
  "device": {
    "cpu": "Apple M4 Max",
    "cpuCores": 14,
    "gpu": "Apple M4 Max",
    "gpuCores": 32,
    "gpuCount": 1,
    "ramGb": 36,
    "osName": "macOS",
    "osVersion": "27.0"
  },
  "decodeTpsMean": 36.1,
  "prefillTpsMean": 213.4,
  "ttftP50Ms": 19192.92,
  "idleTpsMean": 2662,
  "peakRssMb": 11264,
  "trialsPassed": 10,
  "trialsTotal": 10,
  "runnabilityScore": 0.5637309523809524,
  "bundleSha256": "efc7b1b2b961065202449b5a5e2948c2f7f668cde2b19897c23ed10eb28e4158",
  "createdAt": "2026-08-14T20:35:54.317Z"
}