{
  "bench_run_identity": {
    "api_mode": "chat",
    "base_url": "http://127.0.0.1:18260",
    "created_at_utc": "2026-06-23T22:20:28.359477+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": 0.0031089419060692466,
      "max": 7.216243647999363,
      "mean": 7.162364655883721,
      "median": 7.155851917996188,
      "min": 7.144335856020916,
      "stdev": 0.022267375625226137
    },
    "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": 0.0009271461736837925,
      "max": 92.08365992939524,
      "mean": 91.89889994139259,
      "median": 91.89265623186094,
      "min": 91.78077140327463,
      "stdev": 0.08520371344641184
    },
    "tok_s_wall": {
      "cv": 0.003090464034839296,
      "max": 71.66516388902826,
      "mean": 71.48537236571929,
      "median": 71.54983214860235,
      "min": 70.95104114755705,
      "stdev": 0.22092297231335034
    },
    "ttft_s": {
      "cv": 0.01339440360939507,
      "max": 1.6435693450039253,
      "mean": 1.5910206566331908,
      "median": 1.5840515395102557,
      "min": 1.5802476970129646,
      "stdev": 0.021310772825829726
    }
  },
  "canary_pass_all": true,
  "canary_rows_completed": 384,
  "label": "gemma4-q8-gpu0-mtp-n7-c926-fasttopk10-cpucleanup-full-ctxcp0-nmin2-pmin012-nobs-dthreads32-dtb32-filled-long-20260623T221705Z",
  "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 7 --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 0.12 --no-spec-draft-backend-sampling --spec-draft-threads 32 --spec-draft-threads-batch 32 --ctx-checkpoints 0",
    "flash_attn": "off",
    "ggml_sycl_disable_dnn": "<unset>",
    "ggml_sycl_disable_graph": "<unset>",
    "ggml_sycl_disable_opt": "0",
    "ggml_sycl_enable_vmm": "<unset>",
    "gpu_index": "0",
    "llama_cpp_commit": "c926ad098",
    "llama_mtp_draft_backend_topk": "<unset>",
    "llama_mtp_draft_fast_argmax": "<unset>",
    "llama_mtp_draft_fast_topk": "1",
    "llama_mtp_draft_logit_gap_min": "<unset>",
    "llama_mtp_draft_profile": "1",
    "llama_mtp_draft_top_k": "10",
    "llama_server": "/home/steve/src/llama.cpp-latest-gemma/build-sycl-b70-aot-bmg-g31/bin/llama-server",
    "llama_spec_verify_greedy_argmax": "<unset>",
    "oneapi_device_selector": "level_zero:0",
    "poll": "50",
    "port": "18260",
    "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-gpu0-mtp-n7-c926-fasttopk10-cpucleanup-full-ctxcp0-nmin2-pmin012-nobs-dthreads32-dtb32-filled-long-20260623T221705Z",
  "server_log": "/mnt/fast-ai/bench-results/gemma4-26b-a4b-q8/servers/gemma4-q8-gpu0-mtp-n7-c926-fasttopk10-cpucleanup-full-ctxcp0-nmin2-pmin012-nobs-dthreads32-dtb32-filled-long-20260623T221705Z.server.log"
}
