{
  "bench_run_identity": {
    "api_mode": "chat",
    "base_url": "http://127.0.0.1:18263",
    "created_at_utc": "2026-06-24T14:00:15.558720+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": 0.019071938976478266,
      "max": 5.940056725987233,
      "mean": 5.693062092123,
      "median": 5.664482300009695,
      "min": 5.615411174978362,
      "stdev": 0.10857773281027155
    },
    "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": 0.008362667742209486,
      "max": 98.58772971326256,
      "mean": 97.88614261273774,
      "median": 98.37928894367332,
      "min": 96.86583959954126,
      "stdev": 0.8185892872368592
    },
    "tok_s_wall": {
      "cv": 0.018544845606301284,
      "max": 91.17765094057835,
      "mean": 89.96186409657484,
      "median": 90.39195035513328,
      "min": 86.19446305286014,
      "stdev": 1.6683288801260392
    },
    "ttft_s": {
      "cv": 0.24434862383024614,
      "max": 0.7416316730086692,
      "mean": 0.4621740381262498,
      "median": 0.42281501651450526,
      "min": 0.41983402697951533,
      "stdev": 0.11293159018621685
    }
  },
  "canary_pass_all": true,
  "canary_rows_completed": 384,
  "label": "gemma4-q8-gpu3-mtp-n7-directunroll7-qonly-b1024u1024-th8-full-20260624T135701Z",
  "launcher_identity": {
    "batch_size": "1024",
    "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-Q4_0-MTP.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": "0",
    "gpu_index": "3",
    "llama_cpp_commit": "c926ad098",
    "llama_devices": "SYCL0",
    "llama_gemma4_mtp_qonly_attn_inputs": "1",
    "llama_main_gpu": "<unset>",
    "llama_mtp_draft_backend_argmax": "<unset>",
    "llama_mtp_draft_backend_topk": "<unset>",
    "llama_mtp_draft_device_h_handoff": "<unset>",
    "llama_mtp_draft_direct_argmax_ids": "1",
    "llama_mtp_draft_direct_argmax_unroll": "7",
    "llama_mtp_draft_fast_argmax": "1",
    "llama_mtp_draft_fast_topk": "<unset>",
    "llama_mtp_draft_logit_gap_min": "<unset>",
    "llama_mtp_draft_profile": "<unset>",
    "llama_mtp_draft_terminal_logits_only": "<unset>",
    "llama_mtp_draft_top_k": "<unset>",
    "llama_server": "/home/steve/src/llama.cpp-gemma-record-stack/build-sycl-b70-aot-bmg-g31/bin/llama-server",
    "llama_spec_verify_backend_argmax_ids": "<unset>",
    "llama_spec_verify_greedy_argmax": "<unset>",
    "llama_split_mode": "<unset>",
    "llama_tensor_split": "<unset>",
    "oneapi_device_selector": "level_zero:3",
    "poll": "100",
    "port": "18263",
    "reasoning": "off",
    "threads": "8",
    "ubatch_size": "1024"
  },
  "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-n7-directunroll7-qonly-b1024u1024-th8-full-20260624T135701Z",
  "server_log": "/mnt/fast-ai/bench-results/gemma4-26b-a4b-q8/servers/gemma4-q8-gpu3-mtp-n7-directunroll7-qonly-b1024u1024-th8-full-20260624T135701Z.server.log"
}