{
  "bench_run_identity": {
    "api_mode": "chat",
    "base_url": "http://127.0.0.1:18261",
    "created_at_utc": "2026-06-24T19:41:31.042596+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": 0.02024982657180621,
      "max": 5.784800106019247,
      "mean": 5.542473466521187,
      "median": 5.519502619485138,
      "min": 5.43227911205031,
      "stdev": 0.11223412647589161
    },
    "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": 0.01107146129232082,
      "max": 102.22491507371622,
      "mean": 100.83458420322299,
      "median": 101.3587274007032,
      "min": 99.36047034995548,
      "stdev": 1.1163861959332477
    },
    "tok_s_wall": {
      "cv": 0.019780454539168806,
      "max": 94.25141629122871,
      "mean": 92.40990991089272,
      "median": 92.7679269992573,
      "min": 88.50781195831634,
      "stdev": 1.8279100219610984
    },
    "ttft_s": {
      "cv": 0.24476430554009657,
      "max": 0.7455485599930398,
      "mean": 0.4643039306392893,
      "median": 0.4237819635309279,
      "min": 0.4232925479882397,
      "stdev": 0.11364502914246281
    }
  },
  "canary_pass_all": true,
  "canary_rows_completed": 1536,
  "label": "gemma4-q8-gpu1-rowargmax-safer-immediatecl1-full-20260624T193222Z",
  "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 0.14 --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": "0",
    "ggml_sycl_disable_opt": "0",
    "ggml_sycl_enable_vmm": "0",
    "gpu_index": "1",
    "llama_cpp_commit": "c926ad098",
    "llama_devices": "SYCL0",
    "llama_gemma4_mtp_fused_output_argmax": "<unset>",
    "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": "1",
    "llama_spec_verify_greedy_argmax": "<unset>",
    "llama_split_mode": "<unset>",
    "llama_tensor_split": "<unset>",
    "oneapi_device_selector": "level_zero:1",
    "poll": "100",
    "port": "18261",
    "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-gpu1-rowargmax-safer-immediatecl1-full-20260624T193222Z",
  "server_log": "/mnt/fast-ai/bench-results/gemma4-26b-a4b-q8/servers/gemma4-q8-gpu1-rowargmax-safer-immediatecl1-full-20260624T193222Z.server.log"
}