{
  "bench_run_identity": {
    "api_mode": "chat",
    "base_url": "http://127.0.0.1:18261",
    "created_at_utc": "2026-06-23T09:11:30.862052+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": 0.08611394055003137,
      "max": 9.064318488992285,
      "mean": 7.472074157871248,
      "median": 7.24170008249348,
      "min": 7.232965277013136,
      "stdev": 0.6434497498163503
    },
    "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": 0.001568048280451974,
      "max": 74.63163562185605,
      "mean": 74.49809149780477,
      "median": 74.51144731359545,
      "min": 74.24600067848218,
      "stdev": 0.11681660427008658
    },
    "tok_s_wall": {
      "cv": 0.07282042181020422,
      "max": 70.78701201942327,
      "mean": 68.8998540109261,
      "median": 70.70163008648471,
      "min": 56.48521735216753,
      "stdev": 5.01731643173713
    },
    "ttft_s": {
      "cv": 1.081680041993244,
      "max": 2.2039569850021508,
      "mean": 0.599400063380017,
      "median": 0.36804821650730446,
      "min": 0.3663435250055045,
      "stdev": 0.6483590857276499
    }
  },
  "canary_pass_all": true,
  "canary_rows_completed": 384,
  "label": "gemma4-q8-gpu1-mtp-n4-aot-psplit020-filled-long-deep-20260623T090712Z",
  "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 4 --spec-draft-device SYCL0 --spec-draft-ngl all --spec-draft-type-k f16 --spec-draft-type-v f16 --spec-draft-p-split 0.20",
    "flash_attn": "off",
    "ggml_sycl_disable_dnn": null,
    "ggml_sycl_disable_graph": null,
    "ggml_sycl_disable_opt": "0",
    "gpu_index": "1",
    "oneapi_device_selector": null,
    "poll": "50",
    "port": "18261",
    "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-gpu1-mtp-n4-aot-psplit020-filled-long-deep-20260623T090712Z",
  "server_log": "/mnt/fast-ai/bench-results/gemma4-26b-a4b-q8/servers/gemma4-q8-gpu1-mtp-n4-aot-psplit020-filled-long-deep-20260623T090712Z.server.log"
}