{
  "bench_run_identity": {
    "api_mode": "chat",
    "base_url": "http://127.0.0.1:18263",
    "created_at_utc": "2026-06-23T09:02:35.507719+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": 0.08668329574483374,
      "max": 9.088966365001397,
      "mean": 7.483720187250583,
      "median": 7.257753368015983,
      "min": 7.238274804985849,
      "stdev": 0.6487135302630248
    },
    "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": 0.0017521229513853325,
      "max": 74.54113668240946,
      "mean": 74.39465648018985,
      "median": 74.40378619386982,
      "min": 74.22724669043107,
      "stdev": 0.1303485850793682
    },
    "tok_s_wall": {
      "cv": 0.07322806099090381,
      "max": 70.73508726793925,
      "mean": 68.7972812656163,
      "median": 70.54526138973254,
      "min": 56.33203814809379,
      "stdev": 5.037891508526915
    },
    "ttft_s": {
      "cv": 1.076554716238156,
      "max": 2.2040019579872023,
      "mean": 0.6014869741193252,
      "median": 0.37152119250094984,
      "min": 0.36676098097814247,
      "stdev": 0.6475336387439771
    }
  },
  "canary_pass_all": true,
  "canary_rows_completed": 384,
  "label": "gemma4-q8-gpu3-mtp-n4-aot-filled-long-deep-20260623T085822Z",
  "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",
    "flash_attn": "off",
    "ggml_sycl_disable_dnn": null,
    "ggml_sycl_disable_graph": null,
    "ggml_sycl_disable_opt": "0",
    "gpu_index": "3",
    "oneapi_device_selector": null,
    "poll": "50",
    "port": "18263",
    "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-gpu3-mtp-n4-aot-filled-long-deep-20260623T085822Z",
  "server_log": "/mnt/fast-ai/bench-results/gemma4-26b-a4b-q8/servers/gemma4-q8-gpu3-mtp-n4-aot-filled-long-deep-20260623T085822Z.server.log"
}