{
  "case": {
    "id": "06-latest-pure-160k-mtp3-p005",
    "model": "pure-latest",
    "context": 160000,
    "split": [
      90,
      10
    ],
    "projector": "gpu",
    "mtp": true,
    "mtp_max": 3,
    "quality": false,
    "long_fill": 0.0,
    "vision": false
  },
  "model_path": "/models/qwen3.8-27b-iq4-xs-pure/qwen3.8-27b-IQ4_XS-pure.gguf",
  "started": "2026-08-22T21:21:29.621970+00:00",
  "command": [
    "docker",
    "run",
    "-d",
    "--name",
    "mike-ai-qwen38-dualgpu-bench",
    "--gpus",
    "all",
    "--ipc",
    "host",
    "--read-only",
    "--tmpfs",
    "/tmp:size=1g,mode=1777",
    "--security-opt",
    "no-new-privileges:true",
    "--cap-drop",
    "ALL",
    "-e",
    "NVIDIA_VISIBLE_DEVICES=GPU-8ad38c6c-5a01-9d8e-1dfa-ed662ad78fbe,GPU-4834d9d7-5b61-3004-1fb3-4ae49d482d4b",
    "-v",
    "/data/models:/models:ro",
    "-p",
    "127.0.0.1:18099:8080",
    "-e",
    "MTMD_BACKEND_DEVICE=CUDA1",
    "mike-ai/llama.cpp:3f545bec-test",
    "--model",
    "/models/qwen3.8-27b-iq4-xs-pure/qwen3.8-27b-IQ4_XS-pure.gguf",
    "--alias",
    "06-latest-pure-160k-mtp3-p005",
    "--ctx-size",
    "160000",
    "--flash-attn",
    "on",
    "--cache-type-k",
    "q4_0",
    "--cache-type-v",
    "q4_0",
    "--threads",
    "6",
    "--threads-batch",
    "6",
    "--batch-size",
    "64",
    "--ubatch-size",
    "32",
    "--parallel",
    "1",
    "--jinja",
    "--reasoning",
    "auto",
    "--reasoning-budget",
    "768",
    "--host",
    "0.0.0.0",
    "--port",
    "8080",
    "--metrics",
    "--fit",
    "off",
    "--n-gpu-layers",
    "all",
    "--no-mmap",
    "--no-ui",
    "--temperature",
    "0.2",
    "--top-p",
    "0.8",
    "--top-k",
    "20",
    "--device",
    "CUDA0,CUDA1",
    "--main-gpu",
    "0",
    "--split-mode",
    "layer",
    "--tensor-split",
    "90,10",
    "--mmproj",
    "/models/qwen3.8-27b-nvfp4/mmproj-BF16.gguf",
    "--image-min-tokens",
    "1024",
    "--spec-type",
    "draft-mtp",
    "--spec-draft-n-max",
    "3",
    "--spec-draft-type-k",
    "f16",
    "--spec-draft-type-v",
    "f16",
    "--mmproj-device",
    "CUDA1",
    "--spec-draft-p-min",
    "0.05"
  ],
  "docker_launch_returncode": 0,
  "docker_launch_stderr": "",
  "loaded": true,
  "load_seconds": 4.059,
  "gpu_after_load": [
    {
      "index": "0",
      "name": "NVIDIA GeForce RTX 3060",
      "uuid": "GPU-4834d9d7-5b61-3004-1fb3-4ae49d482d4b",
      "total_mib": "12288",
      "used_mib": "4469",
      "free_mib": "7441",
      "temperature_c": "45",
      "power_w": "39.50"
    },
    {
      "index": "1",
      "name": "NVIDIA GeForce RTX 5080",
      "uuid": "GPU-8ad38c6c-5a01-9d8e-1dfa-ed662ad78fbe",
      "total_mib": "16303",
      "used_mib": "15760",
      "free_mib": "122",
      "temperature_c": "52",
      "power_w": "59.73"
    }
  ],
  "speed_runs": [
    {
      "tokens": 384,
      "timings": {
        "cache_n": 0,
        "prompt_n": 28,
        "prompt_ms": 119.452,
        "prompt_per_token_ms": 4.2661428571428575,
        "prompt_per_second": 234.4037772494391,
        "predicted_n": 384,
        "predicted_ms": 4961.185,
        "predicted_per_token_ms": 12.953485639686685,
        "predicted_per_second": 77.19929815155048,
        "draft_n": 411,
        "draft_n_accepted": 246
      }
    },
    {
      "tokens": 384,
      "timings": {
        "cache_n": 0,
        "prompt_n": 28,
        "prompt_ms": 101.946,
        "prompt_per_token_ms": 3.6409285714285713,
        "prompt_per_second": 274.65520962077966,
        "predicted_n": 384,
        "predicted_ms": 4953.073,
        "predicted_per_token_ms": 12.932305483028722,
        "predicted_per_second": 77.32573293387762,
        "draft_n": 411,
        "draft_n_accepted": 246
      }
    }
  ],
  "mean_predicted_tps": 77.26251554271406,
  "status": "complete",
  "server_log": "/data/benchmarks/qwen38-final-acceptance-20260822/cases/06-latest-pure-160k-mtp3-p005/server.log",
  "oom": false,
  "finished": "2026-08-22T21:21:44.986242+00:00"
}