{
  "bench_run_identity": {
    "api_mode": "chat",
    "base_url": "http://127.0.0.1:18262",
    "created_at_utc": "2026-06-23T09:16:38.053131+00:00",
    "max_tokens": 512,
    "model": "gemma4-26b-a4b-q8",
    "prompt_chars": 3460,
    "prompt_mode": "filled-long",
    "prompt_preview": "You are running a deterministic Gemma B70 decode benchmark. Read the reference context, then produce a long numbered response until the token limit is reached. Do not summarize early.\n\nReference context:\nbenchmark latency memory throughput ",
    "prompt_sha256": "04f687a68d60254567d3252624e01a696b0c9843dbb02f5a789d75aada5e4da5",
    "prompt_tokens_requested": 512,
    "repeats": 8,
    "seed": 1,
    "usage_required": true
  },
  "bench_summary": {
    "completion_tokens": {
      "cv": 0.0,
      "max": 512,
      "mean": 512.0,
      "median": 512.0,
      "min": 512,
      "stdev": 0.0
    },
    "completion_tokens_total": 4096,
    "elapsed_s": {
      "cv": 0.09615981782528744,
      "max": 8.334000243979972,
      "mean": 6.732063054867467,
      "median": 6.5019539754866855,
      "min": 6.487359082995681,
      "stdev": 0.6473539569444037
    },
    "prompt_tokens": {
      "cv": 0.0,
      "max": 588,
      "mean": 588.0,
      "median": 588.0,
      "min": 588,
      "stdev": 0.0
    },
    "requests": 8,
    "tok_s_after_ttft": {
      "cv": 0.0014514005332748894,
      "max": 83.67797021966011,
      "mean": 83.51976900913982,
      "median": 83.52504112830471,
      "min": 83.30951543563803,
      "stdev": 0.12122063727886112
    },
    "tok_s_wall": {
      "cv": 0.07987592254985901,
      "max": 78.92271623163685,
      "mean": 76.56854515861237,
      "median": 78.74556256969696,
      "min": 61.435083394656836,
      "stdev": 6.115983182844704
    },
    "ttft_s": {
      "cv": 1.0801057747211764,
      "max": 2.2103175619849935,
      "mean": 0.6017665947474597,
      "median": 0.3699467144906521,
      "min": 0.36768397997366264,
      "stdev": 0.6499715740210292
    }
  },
  "canary_pass_all": true,
  "canary_rows_completed": 384,
  "label": "gemma4-q8-gpu2-mtp-n6-aot-nmin2-pmin015-filled-long-deep-20260623T091227Z",
  "launcher_identity": {
    "batch_size": "512",
    "cache_type_k": "f16",
    "cache_type_v": "f16",
    "ctx_size": "8192",
    "extra_llama_args": "--parallel 1 --cache-ram 0 --spec-type draft-mtp --spec-draft-model /mnt/fast-ai/llm-models/gemma4-26b-a4b-it-q8-gguf/mtp-gemma-4-26B-A4B-it.gguf --spec-draft-n-max 6 --spec-draft-device SYCL0 --spec-draft-ngl all --spec-draft-type-k f16 --spec-draft-type-v f16 --spec-draft-n-min 2 --spec-draft-p-min 0.15",
    "flash_attn": "off",
    "ggml_sycl_disable_dnn": null,
    "ggml_sycl_disable_graph": null,
    "ggml_sycl_disable_opt": "0",
    "gpu_index": "2",
    "oneapi_device_selector": null,
    "poll": "50",
    "port": "18262",
    "reasoning": "off",
    "threads": "16",
    "ubatch_size": "64"
  },
  "model_file_bytes": 27636230944,
  "model_path": "/mnt/fast-ai/llm-models/gemma4-26b-a4b-it-q8-gguf/gemma-4-26B-A4B-it-UD-Q8_K_XL.gguf",
  "run_dir": "/home/steve/qwen36-results-main/data/gemma4-q8-gpu2-mtp-n6-aot-nmin2-pmin015-filled-long-deep-20260623T091227Z",
  "server_log": "/mnt/fast-ai/bench-results/gemma4-26b-a4b-q8/servers/gemma4-q8-gpu2-mtp-n6-aot-nmin2-pmin015-filled-long-deep-20260623T091227Z.server.log"
}