Qwen3.5-4B benchmark on an M4 Pro · 24 GB

{
  "runId": "run_c5ab6cc8-5206-4f52-986d-23c4de7997ae",
  "bundleId": "llamacpp-qwen3.5-4b-q4_k_m.gguf-7927e0",
  "status": "pending",
  "promptTokens": 40960,
  "completionTokens": 10240,
  "contextLength": 5120,
  "harness": {
    "version": "0.1.21",
    "gitSha": "2899067"
  },
  "runtime": {
    "name": "llama.cpp",
    "version": "b9260",
    "buildFlags": "metal"
  },
  "model": {
    "displayName": "Qwen3.5-4B",
    "format": "gguf",
    "quant": "q4_k_m",
    "architecture": "qwen35",
    "source": "unsloth/Qwen3.5-4B-GGUF:Qwen3.5-4B-Q4_K_M.gguf",
    "fileSizeBytes": 2740937888,
    "lab": {
      "name": "Qwen",
      "slug": "qwen"
    },
    "quantizedBy": {
      "name": "Unsloth",
      "slug": "unsloth"
    }
  },
  "device": {
    "cpu": "Apple M4 Pro",
    "cpuCores": 14,
    "gpu": "Apple M4 Pro",
    "gpuCores": 20,
    "gpuCount": 1,
    "ramGb": 24,
    "osName": "macOS",
    "osVersion": "26.5"
  },
  "decodeTpsMean": 53.3,
  "prefillTpsMean": 707.1,
  "ttftP50Ms": 5848.21,
  "idleTpsMean": 8.5,
  "peakRssMb": 736.5,
  "trialsPassed": 10,
  "trialsTotal": 10,
  "runnabilityScore": 0.7878164592633928,
  "bundleSha256": "421e8196f27cb2e2d0c6df1bdf636220b758246ef5264b009e94aa1926229fc7",
  "createdAt": "2026-05-29T12:21:49.828Z"
}