{
  "article_key": "local-llm-api-guide",
  "cases": [
    {
      "case_wall_seconds": 29.91674370097462,
      "completed_response_resources": {
        "gpu_energy": {
          "available": true,
          "interpolation": "linear_at_bracketed_boundaries",
          "joules": 5704.656739645703,
          "reason": null,
          "response_union_seconds": 19.113167581963353,
          "scope": "GPU board only; not whole-server AC energy",
          "time_weighted_mean_watts": 298.4673636738817,
          "trapezoid_pieces": 30,
          "watt_hours": 1.5846268721238064
        },
        "gpu_sample_count": 20,
        "host_sample_count": 19,
        "meaning": "Samples inside declared interval union. Means are arithmetic sample means; peaks need not coincide. Counter deltas exclude unsampled edges; not minimum required RAM.",
        "metrics": {
          "cgroup_memory_current_bytes": {
            "maximum": 6720995328,
            "mean": 6712309221.052631,
            "median": 6713368576,
            "minimum": 6703079424,
            "sample_count": 19
          },
          "gpu_board_watts": {
            "maximum": 301.0,
            "mean": 299.713,
            "median": 299.95,
            "minimum": 297.7,
            "sample_count": 20
          },
          "gpu_memory_used_bytes": {
            "maximum": 6752829440.0,
            "mean": 6752829440.0,
            "median": 6752829440.0,
            "minimum": 6752829440.0,
            "sample_count": 20
          },
          "gpu_sm_clock_mhz": {
            "maximum": 1845.0,
            "mean": 1793.7,
            "median": 1800.0,
            "minimum": 1710.0,
            "sample_count": 20
          },
          "gpu_utilization_percent": {
            "maximum": 100.0,
            "mean": 95.9,
            "median": 97.0,
            "minimum": 85.0,
            "sample_count": 20
          },
          "host_mem_available_bytes": {
            "maximum": 96199720960,
            "mean": 96188027095.57895,
            "median": 96187265024,
            "minimum": 96176087040,
            "sample_count": 19
          },
          "process_pss_bytes": {
            "maximum": 819189760,
            "mean": 815475550.3157895,
            "median": 816355328,
            "minimum": 811747328,
            "sample_count": 19
          },
          "process_rss_bytes": {
            "maximum": 822177792,
            "mean": 818463582.3157895,
            "median": 819343360,
            "minimum": 814735360,
            "sample_count": 19
          }
        },
        "process_sample_count": 19,
        "response_union_seconds": 19.113167581963353,
        "sampled_counter_deltas": {
          "deltas": {
            "major_faults": 0,
            "read_syscall_bytes": 0,
            "storage_read_bytes": 0,
            "storage_write_bytes": 0
          },
          "observed_endpoint_seconds": 9.175436379155144,
          "observed_intervals": 8,
          "requested_intervals": 10
        }
      },
      "groups": [
        {
          "aggregate_output_tokens_per_second": 172.7690055501098,
          "completed_output_tokens": 1280,
          "completed_request_count": 5,
          "context_tokens": 2048,
          "cpu_snapshot_intervals": {
            "host_capacity_percent": 3.119277758589535,
            "interval_count": 5,
            "interval_seconds": 7.423593470070045,
            "meaning": "Weighted CPU counter rate across round-bracketing snapshots, not exact token time.",
            "one_logical_cpu_percent": 99.81688827486512,
            "process_cpu_seconds": 7.41
          },
          "failed_request_count": 0,
          "measured_rounds": 5,
          "mode": "completion",
          "resources": {
            "gpu_energy": {
              "available": true,
              "interpolation": "linear_at_bracketed_boundaries",
              "joules": 2207.7645157894385,
              "reason": null,
              "response_union_seconds": 7.408736283017788,
              "scope": "GPU board only; not whole-server AC energy",
              "time_weighted_mean_watts": 297.99474990762576,
              "trapezoid_pieces": 13,
              "watt_hours": 0.6132679210526218
            },
            "gpu_sample_count": 8,
            "host_sample_count": 8,
            "meaning": "Samples inside declared interval union. Means are arithmetic sample means; peaks need not coincide. Counter deltas exclude unsampled edges; not minimum required RAM.",
            "metrics": {
              "cgroup_memory_current_bytes": {
                "maximum": 6708359168,
                "mean": 6705936384,
                "median": 6706384896.0,
                "minimum": 6703079424,
                "sample_count": 8
              },
              "gpu_board_watts": {
                "maximum": 300.43,
                "mean": 299.49625,
                "median": 299.95,
                "minimum": 297.7,
                "sample_count": 8
              },
              "gpu_memory_used_bytes": {
                "maximum": 6752829440.0,
                "mean": 6752829440.0,
                "median": 6752829440.0,
                "minimum": 6752829440.0,
                "sample_count": 8
              },
              "gpu_sm_clock_mhz": {
                "maximum": 1845.0,
                "mean": 1832.5,
                "median": 1833.5,
                "minimum": 1822.0,
                "sample_count": 8
              },
              "gpu_utilization_percent": {
                "maximum": 97.0,
                "mean": 94.25,
                "median": 97.0,
                "minimum": 86.0,
                "sample_count": 8
              },
              "host_mem_available_bytes": {
                "maximum": 96199720960,
                "mean": 96194545152,
                "median": 96195299328.0,
                "minimum": 96187265024,
                "sample_count": 8
              },
              "process_pss_bytes": {
                "maximum": 813074432,
                "mean": 812405760,
                "median": 812410880.0,
                "minimum": 811747328,
                "sample_count": 8
              },
              "process_rss_bytes": {
                "maximum": 816062464,
                "mean": 815393792,
                "median": 815398912.0,
                "minimum": 814735360,
                "sample_count": 8
              }
            },
            "process_sample_count": 8,
            "response_union_seconds": 7.408736283017788,
            "sampled_counter_deltas": {
              "deltas": {
                "major_faults": 0,
                "read_syscall_bytes": 0,
                "storage_read_bytes": 0,
                "storage_write_bytes": 0
              },
              "observed_endpoint_seconds": 3.05030764703406,
              "observed_intervals": 3,
              "requested_intervals": 5
            }
          },
          "response_outer_span_seconds": 7.430938940960914,
          "response_seconds": {
            "maximum": 1.484641135961283,
            "mean": 1.4817472566035577,
            "median": 1.481124191021081,
            "minimum": 1.479376372997649,
            "sample_count": 5
          },
          "response_window_union_seconds": 7.408736283017788,
          "server_decode_tokens_per_second": {
            "maximum": 197.7636422013653,
            "mean": 197.5064926865939,
            "median": 197.53690841569204,
            "minimum": 197.21684494811652,
            "sample_count": 5
          },
          "synthetic_first_token_seconds": {
            "maximum": 0.19157884595915675,
            "mean": 0.1905686531914398,
            "median": 0.19060329400235787,
            "minimum": 0.1899017799878493,
            "sample_count": 5
          },
          "warmup_completed_rounds": 1,
          "warmup_requests_per_round": 1
        },
        {
          "aggregate_output_tokens_per_second": 109.36028990279206,
          "completed_output_tokens": 1280,
          "completed_request_count": 5,
          "context_tokens": 8192,
          "cpu_snapshot_intervals": {
            "host_capacity_percent": 3.118724755561472,
            "interval_count": 5,
            "interval_seconds": 11.72354178893147,
            "meaning": "Weighted CPU counter rate across round-bracketing snapshots, not exact token time.",
            "one_logical_cpu_percent": 99.7991921779671,
            "process_cpu_seconds": 11.7
          },
          "failed_request_count": 0,
          "measured_rounds": 5,
          "mode": "completion",
          "resources": {
            "gpu_energy": {
              "available": true,
              "interpolation": "linear_at_bracketed_boundaries",
              "joules": 3496.8922238562645,
              "reason": null,
              "response_union_seconds": 11.704431298945565,
              "scope": "GPU board only; not whole-server AC energy",
              "time_weighted_mean_watts": 298.76652137479715,
              "trapezoid_pieces": 17,
              "watt_hours": 0.9713589510711846
            },
            "gpu_sample_count": 12,
            "host_sample_count": 11,
            "meaning": "Samples inside declared interval union. Means are arithmetic sample means; peaks need not coincide. Counter deltas exclude unsampled edges; not minimum required RAM.",
            "metrics": {
              "cgroup_memory_current_bytes": {
                "maximum": 6720995328,
                "mean": 6716944011.636364,
                "median": 6717145088,
                "minimum": 6713245696,
                "sample_count": 11
              },
              "gpu_board_watts": {
                "maximum": 301.0,
                "mean": 299.8575,
                "median": 299.995,
                "minimum": 298.01,
                "sample_count": 12
              },
              "gpu_memory_used_bytes": {
                "maximum": 6752829440.0,
                "mean": 6752829440.0,
                "median": 6752829440.0,
                "minimum": 6752829440.0,
                "sample_count": 12
              },
              "gpu_sm_clock_mhz": {
                "maximum": 1822.0,
                "mean": 1767.8333333333333,
                "median": 1773.5,
                "minimum": 1710.0,
                "sample_count": 12
              },
              "gpu_utilization_percent": {
                "maximum": 100.0,
                "mean": 97.0,
                "median": 97.5,
                "minimum": 85.0,
                "sample_count": 12
              },
              "host_mem_available_bytes": {
                "maximum": 96188547072,
                "mean": 96183286690.90909,
                "median": 96184393728,
                "minimum": 96176087040,
                "sample_count": 11
              },
              "process_pss_bytes": {
                "maximum": 819189760,
                "mean": 817708125.0909091,
                "median": 817772544,
                "minimum": 816355328,
                "sample_count": 11
              },
              "process_rss_bytes": {
                "maximum": 822177792,
                "mean": 820696157.0909091,
                "median": 820760576,
                "minimum": 819343360,
                "sample_count": 11
              }
            },
            "process_sample_count": 11,
            "response_union_seconds": 11.704431298945565,
            "sampled_counter_deltas": {
              "deltas": {
                "major_faults": 0,
                "read_syscall_bytes": 0,
                "storage_read_bytes": 0,
                "storage_write_bytes": 0
              },
              "observed_endpoint_seconds": 6.125128732121084,
              "observed_intervals": 5,
              "requested_intervals": 5
            }
          },
          "response_outer_span_seconds": 11.733725362981204,
          "response_seconds": {
            "maximum": 2.3454450629651546,
            "mean": 2.340886259789113,
            "median": 2.3407816879916936,
            "minimum": 2.3358659429941326,
            "sample_count": 5
          },
          "response_window_union_seconds": 11.704431298945565,
          "server_decode_tokens_per_second": {
            "maximum": 169.3727562261093,
            "mean": 169.06737906840763,
            "median": 169.01162004597117,
            "minimum": 168.80352064566353,
            "sample_count": 5
          },
          "synthetic_first_token_seconds": {
            "maximum": 0.834626693977043,
            "mean": 0.8324093010160141,
            "median": 0.8318316700169817,
            "minimum": 0.8301292260293849,
            "sample_count": 5
          },
          "warmup_completed_rounds": 1,
          "warmup_requests_per_round": 1
        }
      ],
      "id": "baseline-8b",
      "model": "qwen3-8b-q4",
      "module": "baseline",
      "placement": {
        "buffers": [
          {
            "device": "CPU_Mapped",
            "reported_mib": 333.84
          },
          {
            "device": "CUDA0",
            "reported_mib": 4455.34
          }
        ],
        "gpu_layers": 37,
        "meaning": "Runtime buffer allocation is not measured resident process RAM.",
        "requested_cpu_expert_layers": 0,
        "total_layers": 37
      },
      "record_sha256": "2db66c2c90e7ccf91ace8aaf7f6b8e1865a1de2453d6547e769e2dda7e176b21",
      "resources_by_phase": {
        "baseline": {
          "gpu_energy": {
            "available": false,
            "joules": null,
            "reason": "unbracketed_boundary"
          },
          "gpu_sample_count": 2,
          "host_sample_count": 2,
          "meaning": "Samples inside declared interval union. Means are arithmetic sample means; peaks need not coincide. Counter deltas exclude unsampled edges; not minimum required RAM.",
          "metrics": {
            "cgroup_memory_current_bytes": {
              "maximum": 6373740544,
              "mean": 6365132800,
              "median": 6365132800.0,
              "minimum": 6356525056,
              "sample_count": 2
            },
            "gpu_board_watts": {
              "maximum": 15.03,
              "mean": 15.024999999999999,
              "median": 15.024999999999999,
              "minimum": 15.02,
              "sample_count": 2
            },
            "gpu_memory_used_bytes": {
              "maximum": 2097152.0,
              "mean": 2097152.0,
              "median": 2097152.0,
              "minimum": 2097152.0,
              "sample_count": 2
            },
            "gpu_sm_clock_mhz": {
              "maximum": 180.0,
              "mean": 180.0,
              "median": 180.0,
              "minimum": 180.0,
              "sample_count": 2
            },
            "gpu_utilization_percent": {
              "maximum": 0.0,
              "mean": 0.0,
              "median": 0.0,
              "minimum": 0.0,
              "sample_count": 2
            },
            "host_mem_available_bytes": {
              "maximum": 96634912768,
              "mean": 96623800320,
              "median": 96623800320.0,
              "minimum": 96612687872,
              "sample_count": 2
            },
            "process_pss_bytes": null,
            "process_rss_bytes": null
          },
          "process_sample_count": 0,
          "response_union_seconds": 1.1007707070093602,
          "sampled_counter_deltas": {
            "deltas": null,
            "observed_endpoint_seconds": 0.0,
            "observed_intervals": 0,
            "requested_intervals": 1
          }
        },
        "cleanup": {
          "gpu_energy": {
            "available": false,
            "joules": null,
            "reason": "unbracketed_boundary"
          },
          "gpu_sample_count": 1,
          "host_sample_count": 1,
          "meaning": "Samples inside declared interval union. Means are arithmetic sample means; peaks need not coincide. Counter deltas exclude unsampled edges; not minimum required RAM.",
          "metrics": {
            "cgroup_memory_current_bytes": {
              "maximum": 6392254464,
              "mean": 6392254464,
              "median": 6392254464,
              "minimum": 6392254464,
              "sample_count": 1
            },
            "gpu_board_watts": {
              "maximum": 78.38,
              "mean": 78.38,
              "median": 78.38,
              "minimum": 78.38,
              "sample_count": 1
            },
            "gpu_memory_used_bytes": {
              "maximum": 2097152.0,
              "mean": 2097152.0,
              "median": 2097152.0,
              "minimum": 2097152.0,
              "sample_count": 1
            },
            "gpu_sm_clock_mhz": {
              "maximum": 2325.0,
              "mean": 2325.0,
              "median": 2325.0,
              "minimum": 2325.0,
              "sample_count": 1
            },
            "gpu_utilization_percent": {
              "maximum": 0.0,
              "mean": 0.0,
              "median": 0.0,
              "minimum": 0.0,
              "sample_count": 1
            },
            "host_mem_available_bytes": {
              "maximum": 96599707648,
              "mean": 96599707648,
              "median": 96599707648,
              "minimum": 96599707648,
              "sample_count": 1
            },
            "process_pss_bytes": null,
            "process_rss_bytes": null
          },
          "process_sample_count": 0,
          "response_union_seconds": 1.155548368988093,
          "sampled_counter_deltas": {
            "deltas": null,
            "observed_endpoint_seconds": 0.0,
            "observed_intervals": 0,
            "requested_intervals": 1
          }
        },
        "load": {
          "gpu_energy": {
            "available": true,
            "interpolation": "linear_at_bracketed_boundaries",
            "joules": 51.318043537193375,
            "reason": null,
            "response_union_seconds": 1.2843944630003534,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 39.95504886973283,
            "trapezoid_pieces": 2,
            "watt_hours": 0.014255012093664827
          },
          "gpu_sample_count": 1,
          "host_sample_count": 1,
          "meaning": "Samples inside declared interval union. Means are arithmetic sample means; peaks need not coincide. Counter deltas exclude unsampled edges; not minimum required RAM.",
          "metrics": {
            "cgroup_memory_current_bytes": {
              "maximum": 6579744768,
              "mean": 6579744768,
              "median": 6579744768,
              "minimum": 6579744768,
              "sample_count": 1
            },
            "gpu_board_watts": {
              "maximum": 50.74,
              "mean": 50.74,
              "median": 50.74,
              "minimum": 50.74,
              "sample_count": 1
            },
            "gpu_memory_used_bytes": {
              "maximum": 5259657216.0,
              "mean": 5259657216.0,
              "median": 5259657216.0,
              "minimum": 5259657216.0,
              "sample_count": 1
            },
            "gpu_sm_clock_mhz": {
              "maximum": 2280.0,
              "mean": 2280.0,
              "median": 2280.0,
              "minimum": 2280.0,
              "sample_count": 1
            },
            "gpu_utilization_percent": {
              "maximum": 76.0,
              "mean": 76.0,
              "median": 76.0,
              "minimum": 76.0,
              "sample_count": 1
            },
            "host_mem_available_bytes": {
              "maximum": 96307732480,
              "mean": 96307732480,
              "median": 96307732480,
              "minimum": 96307732480,
              "sample_count": 1
            },
            "process_pss_bytes": {
              "maximum": 5364942848,
              "mean": 5364942848,
              "median": 5364942848,
              "minimum": 5364942848,
              "sample_count": 1
            },
            "process_rss_bytes": {
              "maximum": 5367930880,
              "mean": 5367930880,
              "median": 5367930880,
              "minimum": 5367930880,
              "sample_count": 1
            }
          },
          "process_sample_count": 1,
          "response_union_seconds": 1.2843944630003534,
          "sampled_counter_deltas": {
            "deltas": null,
            "observed_endpoint_seconds": 0.0,
            "observed_intervals": 0,
            "requested_intervals": 1
          }
        },
        "ready": {
          "gpu_energy": {
            "available": true,
            "interpolation": "linear_at_bracketed_boundaries",
            "joules": 521.8941446628455,
            "reason": null,
            "response_union_seconds": 3.3171223630197346,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 157.33340152931243,
            "trapezoid_pieces": 6,
            "watt_hours": 0.1449705957396793
          },
          "gpu_sample_count": 3,
          "host_sample_count": 4,
          "meaning": "Samples inside declared interval union. Means are arithmetic sample means; peaks need not coincide. Counter deltas exclude unsampled edges; not minimum required RAM.",
          "metrics": {
            "cgroup_memory_current_bytes": {
              "maximum": 6721740800,
              "mean": 6704672768,
              "median": 6714998784.0,
              "minimum": 6666952704,
              "sample_count": 4
            },
            "gpu_board_watts": {
              "maximum": 226.38,
              "mean": 158.82666666666665,
              "median": 176.54,
              "minimum": 73.56,
              "sample_count": 3
            },
            "gpu_memory_used_bytes": {
              "maximum": 6752829440.0,
              "mean": 6746537984.0,
              "median": 6752829440.0,
              "minimum": 6733955072.0,
              "sample_count": 3
            },
            "gpu_sm_clock_mhz": {
              "maximum": 2325.0,
              "mean": 2310.0,
              "median": 2310.0,
              "minimum": 2295.0,
              "sample_count": 3
            },
            "gpu_utilization_percent": {
              "maximum": 0.0,
              "mean": 0.0,
              "median": 0.0,
              "minimum": 0.0,
              "sample_count": 3
            },
            "host_mem_available_bytes": {
              "maximum": 96215072768,
              "mean": 96190546944,
              "median": 96184281088.0,
              "minimum": 96178552832,
              "sample_count": 4
            },
            "process_pss_bytes": {
              "maximum": 819189760,
              "mean": 807820288,
              "median": 816168960.0,
              "minimum": 779753472,
              "sample_count": 4
            },
            "process_rss_bytes": {
              "maximum": 822177792,
              "mean": 810808320,
              "median": 819156992.0,
              "minimum": 782741504,
              "sample_count": 4
            }
          },
          "process_sample_count": 4,
          "response_union_seconds": 3.3171223630197346,
          "sampled_counter_deltas": {
            "deltas": {
              "major_faults": 0,
              "read_syscall_bytes": 0,
              "storage_read_bytes": 0,
              "storage_write_bytes": 0
            },
            "observed_endpoint_seconds": 1.0194021269562654,
            "observed_intervals": 1,
            "requested_intervals": 3
          }
        },
        "responses": {
          "gpu_energy": {
            "available": true,
            "interpolation": "linear_at_bracketed_boundaries",
            "joules": 5723.3732039281185,
            "reason": null,
            "response_union_seconds": 19.17764565697871,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 298.4398244862446,
            "trapezoid_pieces": 22,
            "watt_hours": 1.5898258899800328
          },
          "gpu_sample_count": 20,
          "host_sample_count": 19,
          "meaning": "Samples inside declared interval union. Means are arithmetic sample means; peaks need not coincide. Counter deltas exclude unsampled edges; not minimum required RAM.",
          "metrics": {
            "cgroup_memory_current_bytes": {
              "maximum": 6720995328,
              "mean": 6712309221.052631,
              "median": 6713368576,
              "minimum": 6703079424,
              "sample_count": 19
            },
            "gpu_board_watts": {
              "maximum": 301.0,
              "mean": 299.713,
              "median": 299.95,
              "minimum": 297.7,
              "sample_count": 20
            },
            "gpu_memory_used_bytes": {
              "maximum": 6752829440.0,
              "mean": 6752829440.0,
              "median": 6752829440.0,
              "minimum": 6752829440.0,
              "sample_count": 20
            },
            "gpu_sm_clock_mhz": {
              "maximum": 1845.0,
              "mean": 1793.7,
              "median": 1800.0,
              "minimum": 1710.0,
              "sample_count": 20
            },
            "gpu_utilization_percent": {
              "maximum": 100.0,
              "mean": 95.9,
              "median": 97.0,
              "minimum": 85.0,
              "sample_count": 20
            },
            "host_mem_available_bytes": {
              "maximum": 96199720960,
              "mean": 96188027095.57895,
              "median": 96187265024,
              "minimum": 96176087040,
              "sample_count": 19
            },
            "process_pss_bytes": {
              "maximum": 819189760,
              "mean": 815475550.3157895,
              "median": 816355328,
              "minimum": 811747328,
              "sample_count": 19
            },
            "process_rss_bytes": {
              "maximum": 822177792,
              "mean": 818463582.3157895,
              "median": 819343360,
              "minimum": 814735360,
              "sample_count": 19
            }
          },
          "process_sample_count": 19,
          "response_union_seconds": 19.17764565697871,
          "sampled_counter_deltas": {
            "deltas": {
              "major_faults": 0,
              "read_syscall_bytes": 0,
              "storage_read_bytes": 0,
              "storage_write_bytes": 16384
            },
            "observed_endpoint_seconds": 17.353597005014308,
            "observed_intervals": 2,
            "requested_intervals": 2
          }
        },
        "warmup": {
          "gpu_energy": {
            "available": true,
            "interpolation": "linear_at_bracketed_boundaries",
            "joules": 888.6925389093352,
            "reason": null,
            "response_union_seconds": 3.8429829700035043,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 231.25070963000516,
            "trapezoid_pieces": 5,
            "watt_hours": 0.24685903858592645
          },
          "gpu_sample_count": 3,
          "host_sample_count": 3,
          "meaning": "Samples inside declared interval union. Means are arithmetic sample means; peaks need not coincide. Counter deltas exclude unsampled edges; not minimum required RAM.",
          "metrics": {
            "cgroup_memory_current_bytes": {
              "maximum": 6711808000,
              "mean": 6708021930.666667,
              "median": 6709710848,
              "minimum": 6702546944,
              "sample_count": 3
            },
            "gpu_board_watts": {
              "maximum": 300.38,
              "mean": 216.23666666666668,
              "median": 180.59,
              "minimum": 167.74,
              "sample_count": 3
            },
            "gpu_memory_used_bytes": {
              "maximum": 6752829440.0,
              "mean": 6752829440.0,
              "median": 6752829440.0,
              "minimum": 6752829440.0,
              "sample_count": 3
            },
            "gpu_sm_clock_mhz": {
              "maximum": 1905.0,
              "mean": 1807.3333333333333,
              "median": 1777.0,
              "minimum": 1740.0,
              "sample_count": 3
            },
            "gpu_utilization_percent": {
              "maximum": 100.0,
              "mean": 98.0,
              "median": 97.0,
              "minimum": 97.0,
              "sample_count": 3
            },
            "host_mem_available_bytes": {
              "maximum": 96199966720,
              "mean": 96187517610.66667,
              "median": 96182734848,
              "minimum": 96179851264,
              "sample_count": 3
            },
            "process_pss_bytes": {
              "maximum": 815646720,
              "mean": 813635584,
              "median": 813856768,
              "minimum": 811403264,
              "sample_count": 3
            },
            "process_rss_bytes": {
              "maximum": 818634752,
              "mean": 816623616,
              "median": 816844800,
              "minimum": 814391296,
              "sample_count": 3
            }
          },
          "process_sample_count": 3,
          "response_union_seconds": 3.8429829700035043,
          "sampled_counter_deltas": {
            "deltas": {
              "major_faults": 0,
              "read_syscall_bytes": 0,
              "storage_read_bytes": 0,
              "storage_write_bytes": 0
            },
            "observed_endpoint_seconds": 1.046531479980331,
            "observed_intervals": 1,
            "requested_intervals": 2
          }
        }
      },
      "reviewed_export_sha256": "2d1a82644c4ba9ba504ddf8fd2c7fe3c253e15cc9811363c52186bd4b7510a07",
      "settings": {
        "batch": 2048,
        "cache_prompt": false,
        "cache_ram_bytes": 0,
        "concurrency": 1,
        "context_shift": false,
        "contexts": [
          2048,
          8192
        ],
        "endpoint": "/completion",
        "fit": "off",
        "flash_attention": true,
        "ignore_eos": true,
        "kv_cache": "f16",
        "lazy_mode": "off",
        "load_mode": "mmap",
        "measured_rounds": 5,
        "mlock": false,
        "network_scope": "loopback",
        "output_tokens": 256,
        "p95_minimum_requests": 200,
        "placement": {
          "cpu_expert_layers": 0,
          "gpu_layers": "all"
        },
        "reasoning": "not_applicable_raw_completion",
        "seed": 20260927,
        "server_slots": 1,
        "synthetic_prompt_sha256": "c099b5abc034050fac1133e0214990d1f72a124b64f36e8a80ba3209bc07d187",
        "temperature": 0,
        "threads": 16,
        "tokenization": "model_tokenizer_add_special_false_repeat_then_truncate",
        "total_context_tokens": 8704,
        "ubatch": 512,
        "warmup_rounds": 1
      },
      "status": "completed",
      "telemetry": {
        "gpu_samples": 30,
        "host_samples": 30,
        "maximum_gpu_gap_seconds": 1.0048957320395857,
        "maximum_host_gap_seconds": 1.0701882099965587,
        "meaning": "Sampled observations; no numeric temperatures published and no warranty between samples.",
        "thermal_slowdown_fully_available": true,
        "thermal_slowdown_observed_samples": 0,
        "thermal_slowdown_states": {
          "hardware": {
            "active": 0,
            "not_active": 30,
            "unavailable": 0
          },
          "software": {
            "active": 0,
            "not_active": 30,
            "unavailable": 0
          }
        }
      }
    },
    {
      "case_wall_seconds": 44.14226625201991,
      "completed_response_resources": {
        "gpu_energy": {
          "available": true,
          "interpolation": "linear_at_bracketed_boundaries",
          "joules": 282.2856086120493,
          "reason": null,
          "response_union_seconds": 1.398443644924555,
          "scope": "GPU board only; not whole-server AC energy",
          "time_weighted_mean_watts": 201.85697838919953,
          "trapezoid_pieces": 25,
          "watt_hours": 0.07841266905890258
        },
        "gpu_sample_count": 1,
        "host_sample_count": 0,
        "meaning": "Samples inside declared interval union. Means are arithmetic sample means; peaks need not coincide. Counter deltas exclude unsampled edges; not minimum required RAM.",
        "metrics": {
          "cgroup_memory_current_bytes": null,
          "gpu_board_watts": {
            "maximum": 167.13,
            "mean": 167.13,
            "median": 167.13,
            "minimum": 167.13,
            "sample_count": 1
          },
          "gpu_memory_used_bytes": {
            "maximum": 6746537984.0,
            "mean": 6746537984.0,
            "median": 6746537984.0,
            "minimum": 6746537984.0,
            "sample_count": 1
          },
          "gpu_sm_clock_mhz": {
            "maximum": 2167.0,
            "mean": 2167.0,
            "median": 2167.0,
            "minimum": 2167.0,
            "sample_count": 1
          },
          "gpu_utilization_percent": {
            "maximum": 89.0,
            "mean": 89.0,
            "median": 89.0,
            "minimum": 89.0,
            "sample_count": 1
          },
          "host_mem_available_bytes": null,
          "process_pss_bytes": null,
          "process_rss_bytes": null
        },
        "process_sample_count": 0,
        "response_union_seconds": 1.398443644924555,
        "sampled_counter_deltas": {
          "deltas": null,
          "observed_endpoint_seconds": 0.0,
          "observed_intervals": 0,
          "requested_intervals": 24
        }
      },
      "groups": [
        {
          "completed_output_tokens": 229,
          "completed_request_count": 24,
          "context_tokens": null,
          "failed_request_count": 0,
          "meaning": "Distinct authored tasks, one attempt each; no pooled practical latency ranking or general quality claim.",
          "measured_rounds": 24,
          "mode": "chat",
          "resources": {
            "gpu_energy": {
              "available": true,
              "interpolation": "linear_at_bracketed_boundaries",
              "joules": 282.2856086120493,
              "reason": null,
              "response_union_seconds": 1.398443644924555,
              "scope": "GPU board only; not whole-server AC energy",
              "time_weighted_mean_watts": 201.85697838919953,
              "trapezoid_pieces": 25,
              "watt_hours": 0.07841266905890258
            },
            "gpu_sample_count": 1,
            "host_sample_count": 0,
            "meaning": "Samples inside declared interval union. Means are arithmetic sample means; peaks need not coincide. Counter deltas exclude unsampled edges; not minimum required RAM.",
            "metrics": {
              "cgroup_memory_current_bytes": null,
              "gpu_board_watts": {
                "maximum": 167.13,
                "mean": 167.13,
                "median": 167.13,
                "minimum": 167.13,
                "sample_count": 1
              },
              "gpu_memory_used_bytes": {
                "maximum": 6746537984.0,
                "mean": 6746537984.0,
                "median": 6746537984.0,
                "minimum": 6746537984.0,
                "sample_count": 1
              },
              "gpu_sm_clock_mhz": {
                "maximum": 2167.0,
                "mean": 2167.0,
                "median": 2167.0,
                "minimum": 2167.0,
                "sample_count": 1
              },
              "gpu_utilization_percent": {
                "maximum": 89.0,
                "mean": 89.0,
                "median": 89.0,
                "minimum": 89.0,
                "sample_count": 1
              },
              "host_mem_available_bytes": null,
              "process_pss_bytes": null,
              "process_rss_bytes": null
            },
            "process_sample_count": 0,
            "response_union_seconds": 1.398443644924555,
            "sampled_counter_deltas": {
              "deltas": null,
              "observed_endpoint_seconds": 0.0,
              "observed_intervals": 0,
              "requested_intervals": 24
            }
          },
          "response_outer_span_seconds": 1.4396691279835068,
          "response_window_union_seconds": 1.398443644924555,
          "score": {
            "by_family": {
              "arithmetic": {
                "expected_value_pass": 2,
                "strict_format_pass": 6,
                "task_count": 6
              },
              "classification": {
                "expected_value_pass": 6,
                "strict_format_pass": 6,
                "task_count": 6
              },
              "extraction": {
                "expected_value_pass": 3,
                "strict_format_pass": 6,
                "task_count": 6
              },
              "grounded-qa": {
                "expected_value_pass": 6,
                "strict_format_pass": 6,
                "task_count": 6
              }
            },
            "by_language": {
              "en": {
                "expected_value_pass": 8,
                "strict_format_pass": 12,
                "task_count": 12
              },
              "pl": {
                "expected_value_pass": 9,
                "strict_format_pass": 12,
                "task_count": 12
              }
            },
            "completed_output": 24,
            "completed_transport": 24,
            "expected_value_pass": 17,
            "strict_format_pass": 24,
            "task_count": 24,
            "truncated": 0
          },
          "tasks": [
            {
              "completed_output": true,
              "connection_seconds": 0.05221215198980644,
              "failure_reason": null,
              "family": "arithmetic",
              "finish_reason": "stop",
              "first_generated_event_seconds": 0.019419765972997993,
              "first_visible_content_seconds": 0.019419765972997993,
              "generated_tokens": 8,
              "input_tokens": 69,
              "language": "pl",
              "last_visible_content_seconds": 0.04756913398159668,
              "reasoning_events_observed": false,
              "response_seconds": 0.052119900996331125,
              "score": {
                "completed": true,
                "expected_value_pass": true,
                "strict_format_pass": true
              },
              "server_decode_tokens_per_second": null,
              "status": "completed",
              "task_id": "arithmetic-pl-1",
              "truncated": false
            },
            {
              "completed_output": true,
              "connection_seconds": 0.06447923701489344,
              "failure_reason": null,
              "family": "arithmetic",
              "finish_reason": "stop",
              "first_generated_event_seconds": 0.018350900034420192,
              "first_visible_content_seconds": 0.018350900034420192,
              "generated_tokens": 11,
              "input_tokens": 69,
              "language": "pl",
              "last_visible_content_seconds": 0.059841028007213026,
              "reasoning_events_observed": false,
              "response_seconds": 0.0643794719944708,
              "score": {
                "completed": true,
                "expected_value_pass": false,
                "strict_format_pass": true
              },
              "server_decode_tokens_per_second": null,
              "status": "completed",
              "task_id": "arithmetic-pl-2",
              "truncated": false
            },
            {
              "completed_output": true,
              "connection_seconds": 0.05317579198163003,
              "failure_reason": null,
              "family": "arithmetic",
              "finish_reason": "stop",
              "first_generated_event_seconds": 0.02051437000045553,
              "first_visible_content_seconds": 0.02051437000045553,
              "generated_tokens": 8,
              "input_tokens": 93,
              "language": "pl",
              "last_visible_content_seconds": 0.04849577596178278,
              "reasoning_events_observed": false,
              "response_seconds": 0.05304892698768526,
              "score": {
                "completed": true,
                "expected_value_pass": true,
                "strict_format_pass": true
              },
              "server_decode_tokens_per_second": null,
              "status": "completed",
              "task_id": "arithmetic-pl-3",
              "truncated": false
            },
            {
              "completed_output": true,
              "connection_seconds": 0.05432305403519422,
              "failure_reason": null,
              "family": "arithmetic",
              "finish_reason": "stop",
              "first_generated_event_seconds": 0.017227592004928738,
              "first_visible_content_seconds": 0.017227592004928738,
              "generated_tokens": 9,
              "input_tokens": 49,
              "language": "en",
              "last_visible_content_seconds": 0.04970139602664858,
              "reasoning_events_observed": false,
              "response_seconds": 0.05423555203014985,
              "score": {
                "completed": true,
                "expected_value_pass": false,
                "strict_format_pass": true
              },
              "server_decode_tokens_per_second": null,
              "status": "completed",
              "task_id": "arithmetic-en-1",
              "truncated": false
            },
            {
              "completed_output": true,
              "connection_seconds": 0.06272390496451408,
              "failure_reason": null,
              "family": "arithmetic",
              "finish_reason": "stop",
              "first_generated_event_seconds": 0.016606379009317607,
              "first_visible_content_seconds": 0.016606379009317607,
              "generated_tokens": 11,
              "input_tokens": 54,
              "language": "en",
              "last_visible_content_seconds": 0.05807481601368636,
              "reasoning_events_observed": false,
              "response_seconds": 0.06263354799011722,
              "score": {
                "completed": true,
                "expected_value_pass": false,
                "strict_format_pass": true
              },
              "server_decode_tokens_per_second": null,
              "status": "completed",
              "task_id": "arithmetic-en-2",
              "truncated": false
            },
            {
              "completed_output": true,
              "connection_seconds": 0.04934750898974016,
              "failure_reason": null,
              "family": "arithmetic",
              "finish_reason": "stop",
              "first_generated_event_seconds": 0.016784630017355084,
              "first_visible_content_seconds": 0.016784630017355084,
              "generated_tokens": 8,
              "input_tokens": 56,
              "language": "en",
              "last_visible_content_seconds": 0.04469241900369525,
              "reasoning_events_observed": false,
              "response_seconds": 0.049255488032940775,
              "score": {
                "completed": true,
                "expected_value_pass": false,
                "strict_format_pass": true
              },
              "server_decode_tokens_per_second": null,
              "status": "completed",
              "task_id": "arithmetic-en-3",
              "truncated": false
            },
            {
              "completed_output": true,
              "connection_seconds": 0.079617025970947,
              "failure_reason": null,
              "family": "extraction",
              "finish_reason": "stop",
              "first_generated_event_seconds": 0.019807635981123894,
              "first_visible_content_seconds": 0.019807635981123894,
              "generated_tokens": 14,
              "input_tokens": 84,
              "language": "pl",
              "last_visible_content_seconds": 0.0749345850199461,
              "reasoning_events_observed": false,
              "response_seconds": 0.0795295630232431,
              "score": {
                "completed": true,
                "expected_value_pass": false,
                "strict_format_pass": true
              },
              "server_decode_tokens_per_second": null,
              "status": "completed",
              "task_id": "extraction-pl-1",
              "truncated": false
            },
            {
              "completed_output": true,
              "connection_seconds": 0.09852313797455281,
              "failure_reason": null,
              "family": "extraction",
              "finish_reason": "stop",
              "first_generated_event_seconds": 0.019785535987466574,
              "first_visible_content_seconds": 0.019785535987466574,
              "generated_tokens": 18,
              "input_tokens": 81,
              "language": "pl",
              "last_visible_content_seconds": 0.09379354899283499,
              "reasoning_events_observed": false,
              "response_seconds": 0.09843722794903442,
              "score": {
                "completed": true,
                "expected_value_pass": true,
                "strict_format_pass": true
              },
              "server_decode_tokens_per_second": null,
              "status": "completed",
              "task_id": "extraction-pl-2",
              "truncated": false
            },
            {
              "completed_output": true,
              "connection_seconds": 0.08868215198162943,
              "failure_reason": null,
              "family": "extraction",
              "finish_reason": "stop",
              "first_generated_event_seconds": 0.018457377969752997,
              "first_visible_content_seconds": 0.018457377969752997,
              "generated_tokens": 16,
              "input_tokens": 72,
              "language": "pl",
              "last_visible_content_seconds": 0.08392347896005958,
              "reasoning_events_observed": false,
              "response_seconds": 0.08857695595361292,
              "score": {
                "completed": true,
                "expected_value_pass": false,
                "strict_format_pass": true
              },
              "server_decode_tokens_per_second": null,
              "status": "completed",
              "task_id": "extraction-pl-3",
              "truncated": false
            },
            {
              "completed_output": true,
              "connection_seconds": 0.08371880900813267,
              "failure_reason": null,
              "family": "extraction",
              "finish_reason": "stop",
              "first_generated_event_seconds": 0.017831003991886973,
              "first_visible_content_seconds": 0.017831003991886973,
              "generated_tokens": 15,
              "input_tokens": 61,
              "language": "en",
              "last_visible_content_seconds": 0.07899835699936375,
              "reasoning_events_observed": false,
              "response_seconds": 0.08363295899471268,
              "score": {
                "completed": true,
                "expected_value_pass": true,
                "strict_format_pass": true
              },
              "server_decode_tokens_per_second": null,
              "status": "completed",
              "task_id": "extraction-en-1",
              "truncated": false
            },
            {
              "completed_output": true,
              "connection_seconds": 0.09630749101052061,
              "failure_reason": null,
              "family": "extraction",
              "finish_reason": "stop",
              "first_generated_event_seconds": 0.016833781031891704,
              "first_visible_content_seconds": 0.016833781031891704,
              "generated_tokens": 18,
              "input_tokens": 51,
              "language": "en",
              "last_visible_content_seconds": 0.09156348503893241,
              "reasoning_events_observed": false,
              "response_seconds": 0.09622695099096745,
              "score": {
                "completed": true,
                "expected_value_pass": true,
                "strict_format_pass": true
              },
              "server_decode_tokens_per_second": null,
              "status": "completed",
              "task_id": "extraction-en-2",
              "truncated": false
            },
            {
              "completed_output": true,
              "connection_seconds": 0.07497315696673468,
              "failure_reason": null,
              "family": "extraction",
              "finish_reason": "stop",
              "first_generated_event_seconds": 0.018429475952871144,
              "first_visible_content_seconds": 0.018429475952871144,
              "generated_tokens": 13,
              "input_tokens": 53,
              "language": "en",
              "last_visible_content_seconds": 0.070229101984296,
              "reasoning_events_observed": false,
              "response_seconds": 0.07487369194859639,
              "score": {
                "completed": true,
                "expected_value_pass": false,
                "strict_format_pass": true
              },
              "server_decode_tokens_per_second": null,
              "status": "completed",
              "task_id": "extraction-en-3",
              "truncated": false
            },
            {
              "completed_output": true,
              "connection_seconds": 0.04386618302669376,
              "failure_reason": null,
              "family": "classification",
              "finish_reason": "stop",
              "first_generated_event_seconds": 0.0200749640353024,
              "first_visible_content_seconds": 0.0200749640353024,
              "generated_tokens": 6,
              "input_tokens": 87,
              "language": "pl",
              "last_visible_content_seconds": 0.039185795991215855,
              "reasoning_events_observed": false,
              "response_seconds": 0.04377738799666986,
              "score": {
                "completed": true,
                "expected_value_pass": true,
                "strict_format_pass": true
              },
              "server_decode_tokens_per_second": null,
              "status": "completed",
              "task_id": "classification-pl-1",
              "truncated": false
            },
            {
              "completed_output": true,
              "connection_seconds": 0.04819632897851989,
              "failure_reason": null,
              "family": "classification",
              "finish_reason": "stop",
              "first_generated_event_seconds": 0.0199568240204826,
              "first_visible_content_seconds": 0.0199568240204826,
              "generated_tokens": 7,
              "input_tokens": 87,
              "language": "pl",
              "last_visible_content_seconds": 0.043546427972614765,
              "reasoning_events_observed": false,
              "response_seconds": 0.04810755397193134,
              "score": {
                "completed": true,
                "expected_value_pass": true,
                "strict_format_pass": true
              },
              "server_decode_tokens_per_second": null,
              "status": "completed",
              "task_id": "classification-pl-2",
              "truncated": false
            },
            {
              "completed_output": true,
              "connection_seconds": 0.04204844601918012,
              "failure_reason": null,
              "family": "classification",
              "finish_reason": "stop",
              "first_generated_event_seconds": 0.018510867026634514,
              "first_visible_content_seconds": 0.018510867026634514,
              "generated_tokens": 6,
              "input_tokens": 76,
              "language": "pl",
              "last_visible_content_seconds": 0.03741702903062105,
              "reasoning_events_observed": false,
              "response_seconds": 0.04196078301174566,
              "score": {
                "completed": true,
                "expected_value_pass": true,
                "strict_format_pass": true
              },
              "server_decode_tokens_per_second": null,
              "status": "completed",
              "task_id": "classification-pl-3",
              "truncated": false
            },
            {
              "completed_output": true,
              "connection_seconds": 0.04613240499747917,
              "failure_reason": null,
              "family": "classification",
              "finish_reason": "stop",
              "first_generated_event_seconds": 0.018114059988874942,
              "first_visible_content_seconds": 0.018114059988874942,
              "generated_tokens": 7,
              "input_tokens": 65,
              "language": "en",
              "last_visible_content_seconds": 0.041501428990159184,
              "reasoning_events_observed": false,
              "response_seconds": 0.04604476201348007,
              "score": {
                "completed": true,
                "expected_value_pass": true,
                "strict_format_pass": true
              },
              "server_decode_tokens_per_second": null,
              "status": "completed",
              "task_id": "classification-en-1",
              "truncated": false
            },
            {
              "completed_output": true,
              "connection_seconds": 0.0452238559955731,
              "failure_reason": null,
              "family": "classification",
              "finish_reason": "stop",
              "first_generated_event_seconds": 0.017137985967565328,
              "first_visible_content_seconds": 0.017137985967565328,
              "generated_tokens": 7,
              "input_tokens": 64,
              "language": "en",
              "last_visible_content_seconds": 0.040544210001826286,
              "reasoning_events_observed": false,
              "response_seconds": 0.04511930199805647,
              "score": {
                "completed": true,
                "expected_value_pass": true,
                "strict_format_pass": true
              },
              "server_decode_tokens_per_second": null,
              "status": "completed",
              "task_id": "classification-en-2",
              "truncated": false
            },
            {
              "completed_output": true,
              "connection_seconds": 0.039076543005649,
              "failure_reason": null,
              "family": "classification",
              "finish_reason": "stop",
              "first_generated_event_seconds": 0.015550255950074643,
              "first_visible_content_seconds": 0.015550255950074643,
              "generated_tokens": 6,
              "input_tokens": 48,
              "language": "en",
              "last_visible_content_seconds": 0.03445477498462424,
              "reasoning_events_observed": false,
              "response_seconds": 0.03898570500314236,
              "score": {
                "completed": true,
                "expected_value_pass": true,
                "strict_format_pass": true
              },
              "server_decode_tokens_per_second": null,
              "status": "completed",
              "task_id": "classification-en-3",
              "truncated": false
            },
            {
              "completed_output": true,
              "connection_seconds": 0.04645818000426516,
              "failure_reason": null,
              "family": "grounded-qa",
              "finish_reason": "stop",
              "first_generated_event_seconds": 0.018379073007963598,
              "first_visible_content_seconds": 0.018379073007963598,
              "generated_tokens": 7,
              "input_tokens": 78,
              "language": "pl",
              "last_visible_content_seconds": 0.04182571201818064,
              "reasoning_events_observed": false,
              "response_seconds": 0.04636744101298973,
              "score": {
                "completed": true,
                "expected_value_pass": true,
                "strict_format_pass": true
              },
              "server_decode_tokens_per_second": null,
              "status": "completed",
              "task_id": "grounded-pl-1",
              "truncated": false
            },
            {
              "completed_output": true,
              "connection_seconds": 0.043439209985081106,
              "failure_reason": null,
              "family": "grounded-qa",
              "finish_reason": "stop",
              "first_generated_event_seconds": 0.01990467798896134,
              "first_visible_content_seconds": 0.01990467798896134,
              "generated_tokens": 6,
              "input_tokens": 89,
              "language": "pl",
              "last_visible_content_seconds": 0.03881491703214124,
              "reasoning_events_observed": false,
              "response_seconds": 0.04335429199272767,
              "score": {
                "completed": true,
                "expected_value_pass": true,
                "strict_format_pass": true
              },
              "server_decode_tokens_per_second": null,
              "status": "completed",
              "task_id": "grounded-pl-2",
              "truncated": false
            },
            {
              "completed_output": true,
              "connection_seconds": 0.05091221700422466,
              "failure_reason": null,
              "family": "grounded-qa",
              "finish_reason": "stop",
              "first_generated_event_seconds": 0.02277130394941196,
              "first_visible_content_seconds": 0.02277130394941196,
              "generated_tokens": 7,
              "input_tokens": 108,
              "language": "pl",
              "last_visible_content_seconds": 0.04627511999569833,
              "reasoning_events_observed": false,
              "response_seconds": 0.050832788983825594,
              "score": {
                "completed": true,
                "expected_value_pass": true,
                "strict_format_pass": true
              },
              "server_decode_tokens_per_second": null,
              "status": "completed",
              "task_id": "grounded-pl-3",
              "truncated": false
            },
            {
              "completed_output": true,
              "connection_seconds": 0.03985726402606815,
              "failure_reason": null,
              "family": "grounded-qa",
              "finish_reason": "stop",
              "first_generated_event_seconds": 0.016315378015860915,
              "first_visible_content_seconds": 0.016315378015860915,
              "generated_tokens": 6,
              "input_tokens": 52,
              "language": "en",
              "last_visible_content_seconds": 0.03522455500205979,
              "reasoning_events_observed": false,
              "response_seconds": 0.03976457205135375,
              "score": {
                "completed": true,
                "expected_value_pass": true,
                "strict_format_pass": true
              },
              "server_decode_tokens_per_second": null,
              "status": "completed",
              "task_id": "grounded-en-1",
              "truncated": false
            },
            {
              "completed_output": true,
              "connection_seconds": 0.045047869032714516,
              "failure_reason": null,
              "family": "grounded-qa",
              "finish_reason": "stop",
              "first_generated_event_seconds": 0.016996193036902696,
              "first_visible_content_seconds": 0.016996193036902696,
              "generated_tokens": 7,
              "input_tokens": 61,
              "language": "en",
              "last_visible_content_seconds": 0.040395855030510575,
              "reasoning_events_observed": false,
              "response_seconds": 0.0449410310247913,
              "score": {
                "completed": true,
                "expected_value_pass": true,
                "strict_format_pass": true
              },
              "server_decode_tokens_per_second": null,
              "status": "completed",
              "task_id": "grounded-en-2",
              "truncated": false
            },
            {
              "completed_output": true,
              "connection_seconds": 0.05231968098087236,
              "failure_reason": null,
              "family": "grounded-qa",
              "finish_reason": "stop",
              "first_generated_event_seconds": 0.01978666801005602,
              "first_visible_content_seconds": 0.01978666801005602,
              "generated_tokens": 8,
              "input_tokens": 83,
              "language": "en",
              "last_visible_content_seconds": 0.04768890596460551,
              "reasoning_events_observed": false,
              "response_seconds": 0.05223778897197917,
              "score": {
                "completed": true,
                "expected_value_pass": true,
                "strict_format_pass": true
              },
              "server_decode_tokens_per_second": null,
              "status": "completed",
              "task_id": "grounded-en-3",
              "truncated": false
            }
          ],
          "warmup_completed_rounds": 1,
          "warmup_requests_per_round": 1
        }
      ],
      "id": "practical-8b",
      "model": "qwen3-8b-q4",
      "module": "practical",
      "placement": {
        "buffers": [
          {
            "device": "CPU_Mapped",
            "reported_mib": 333.84
          },
          {
            "device": "CUDA0",
            "reported_mib": 4455.34
          }
        ],
        "gpu_layers": 37,
        "meaning": "Runtime buffer allocation is not measured resident process RAM.",
        "requested_cpu_expert_layers": 0,
        "total_layers": 37
      },
      "record_sha256": "61ec4830f18e473deb9b9939e4f9c53c40f203825628698eb0bb29e5107f2cc6",
      "resources_by_phase": {
        "baseline": {
          "gpu_energy": {
            "available": false,
            "joules": null,
            "reason": "unbracketed_boundary"
          },
          "gpu_sample_count": 2,
          "host_sample_count": 2,
          "meaning": "Samples inside declared interval union. Means are arithmetic sample means; peaks need not coincide. Counter deltas exclude unsampled edges; not minimum required RAM.",
          "metrics": {
            "cgroup_memory_current_bytes": {
              "maximum": 6413602816,
              "mean": 6405283840,
              "median": 6405283840.0,
              "minimum": 6396964864,
              "sample_count": 2
            },
            "gpu_board_watts": {
              "maximum": 16.42,
              "mean": 16.255000000000003,
              "median": 16.255000000000003,
              "minimum": 16.09,
              "sample_count": 2
            },
            "gpu_memory_used_bytes": {
              "maximum": 2097152.0,
              "mean": 2097152.0,
              "median": 2097152.0,
              "minimum": 2097152.0,
              "sample_count": 2
            },
            "gpu_sm_clock_mhz": {
              "maximum": 180.0,
              "mean": 180.0,
              "median": 180.0,
              "minimum": 180.0,
              "sample_count": 2
            },
            "gpu_utilization_percent": {
              "maximum": 0.0,
              "mean": 0.0,
              "median": 0.0,
              "minimum": 0.0,
              "sample_count": 2
            },
            "host_mem_available_bytes": {
              "maximum": 96553963520,
              "mean": 96552726528,
              "median": 96552726528.0,
              "minimum": 96551489536,
              "sample_count": 2
            },
            "process_pss_bytes": null,
            "process_rss_bytes": null
          },
          "process_sample_count": 0,
          "response_union_seconds": 1.1003843810176477,
          "sampled_counter_deltas": {
            "deltas": null,
            "observed_endpoint_seconds": 0.0,
            "observed_intervals": 0,
            "requested_intervals": 1
          }
        },
        "cleanup": {
          "gpu_energy": {
            "available": false,
            "joules": null,
            "reason": "unbracketed_boundary"
          },
          "gpu_sample_count": 1,
          "host_sample_count": 1,
          "meaning": "Samples inside declared interval union. Means are arithmetic sample means; peaks need not coincide. Counter deltas exclude unsampled edges; not minimum required RAM.",
          "metrics": {
            "cgroup_memory_current_bytes": {
              "maximum": 6400180224,
              "mean": 6400180224,
              "median": 6400180224,
              "minimum": 6400180224,
              "sample_count": 1
            },
            "gpu_board_watts": {
              "maximum": 76.32,
              "mean": 76.32,
              "median": 76.32,
              "minimum": 76.32,
              "sample_count": 1
            },
            "gpu_memory_used_bytes": {
              "maximum": 2097152.0,
              "mean": 2097152.0,
              "median": 2097152.0,
              "minimum": 2097152.0,
              "sample_count": 1
            },
            "gpu_sm_clock_mhz": {
              "maximum": 2310.0,
              "mean": 2310.0,
              "median": 2310.0,
              "minimum": 2310.0,
              "sample_count": 1
            },
            "gpu_utilization_percent": {
              "maximum": 0.0,
              "mean": 0.0,
              "median": 0.0,
              "minimum": 0.0,
              "sample_count": 1
            },
            "host_mem_available_bytes": {
              "maximum": 96510169088,
              "mean": 96510169088,
              "median": 96510169088,
              "minimum": 96510169088,
              "sample_count": 1
            },
            "process_pss_bytes": null,
            "process_rss_bytes": null
          },
          "process_sample_count": 0,
          "response_union_seconds": 1.1560979000059888,
          "sampled_counter_deltas": {
            "deltas": null,
            "observed_endpoint_seconds": 0.0,
            "observed_intervals": 0,
            "requested_intervals": 1
          }
        },
        "load": {
          "gpu_energy": {
            "available": true,
            "interpolation": "linear_at_bracketed_boundaries",
            "joules": 42.085601086970726,
            "reason": null,
            "response_union_seconds": 1.286838505009655,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 32.70464858110145,
            "trapezoid_pieces": 2,
            "watt_hours": 0.011690444746380757
          },
          "gpu_sample_count": 1,
          "host_sample_count": 1,
          "meaning": "Samples inside declared interval union. Means are arithmetic sample means; peaks need not coincide. Counter deltas exclude unsampled edges; not minimum required RAM.",
          "metrics": {
            "cgroup_memory_current_bytes": {
              "maximum": 6616887296,
              "mean": 6616887296,
              "median": 6616887296,
              "minimum": 6616887296,
              "sample_count": 1
            },
            "gpu_board_watts": {
              "maximum": 37.93,
              "mean": 37.93,
              "median": 37.93,
              "minimum": 37.93,
              "sample_count": 1
            },
            "gpu_memory_used_bytes": {
              "maximum": 5259657216.0,
              "mean": 5259657216.0,
              "median": 5259657216.0,
              "minimum": 5259657216.0,
              "sample_count": 1
            },
            "gpu_sm_clock_mhz": {
              "maximum": 2302.0,
              "mean": 2302.0,
              "median": 2302.0,
              "minimum": 2302.0,
              "sample_count": 1
            },
            "gpu_utilization_percent": {
              "maximum": 79.0,
              "mean": 79.0,
              "median": 79.0,
              "minimum": 79.0,
              "sample_count": 1
            },
            "host_mem_available_bytes": {
              "maximum": 96255836160,
              "mean": 96255836160,
              "median": 96255836160,
              "minimum": 96255836160,
              "sample_count": 1
            },
            "process_pss_bytes": {
              "maximum": 5367099392,
              "mean": 5367099392,
              "median": 5367099392,
              "minimum": 5367099392,
              "sample_count": 1
            },
            "process_rss_bytes": {
              "maximum": 5370085376,
              "mean": 5370085376,
              "median": 5370085376,
              "minimum": 5370085376,
              "sample_count": 1
            }
          },
          "process_sample_count": 1,
          "response_union_seconds": 1.286838505009655,
          "sampled_counter_deltas": {
            "deltas": null,
            "observed_endpoint_seconds": 0.0,
            "observed_intervals": 0,
            "requested_intervals": 1
          }
        },
        "ready": {
          "gpu_energy": {
            "available": true,
            "interpolation": "linear_at_bracketed_boundaries",
            "joules": 300.5957116146231,
            "reason": null,
            "response_union_seconds": 2.213750633003656,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 135.78571458465018,
            "trapezoid_pieces": 5,
            "watt_hours": 0.08349880878183975
          },
          "gpu_sample_count": 3,
          "host_sample_count": 2,
          "meaning": "Samples inside declared interval union. Means are arithmetic sample means; peaks need not coincide. Counter deltas exclude unsampled edges; not minimum required RAM.",
          "metrics": {
            "cgroup_memory_current_bytes": {
              "maximum": 6727159808,
              "mean": 6709911552,
              "median": 6709911552.0,
              "minimum": 6692663296,
              "sample_count": 2
            },
            "gpu_board_watts": {
              "maximum": 293.55,
              "mean": 157.6,
              "median": 100.94,
              "minimum": 78.31,
              "sample_count": 3
            },
            "gpu_memory_used_bytes": {
              "maximum": 6746537984.0,
              "mean": 6742343680.0,
              "median": 6746537984.0,
              "minimum": 6733955072.0,
              "sample_count": 3
            },
            "gpu_sm_clock_mhz": {
              "maximum": 2310.0,
              "mean": 2267.3333333333335,
              "median": 2310.0,
              "minimum": 2182.0,
              "sample_count": 3
            },
            "gpu_utilization_percent": {
              "maximum": 81.0,
              "mean": 27.0,
              "median": 0.0,
              "minimum": 0.0,
              "sample_count": 3
            },
            "host_mem_available_bytes": {
              "maximum": 96154648576,
              "mean": 96140855296,
              "median": 96140855296.0,
              "minimum": 96127062016,
              "sample_count": 2
            },
            "process_pss_bytes": {
              "maximum": 816381952,
              "mean": 799184896,
              "median": 799184896.0,
              "minimum": 781987840,
              "sample_count": 2
            },
            "process_rss_bytes": {
              "maximum": 819367936,
              "mean": 802170880,
              "median": 802170880.0,
              "minimum": 784973824,
              "sample_count": 2
            }
          },
          "process_sample_count": 2,
          "response_union_seconds": 2.213750633003656,
          "sampled_counter_deltas": {
            "deltas": null,
            "observed_endpoint_seconds": 0.0,
            "observed_intervals": 0,
            "requested_intervals": 2
          }
        },
        "responses": {
          "gpu_energy": {
            "available": true,
            "interpolation": "linear_at_bracketed_boundaries",
            "joules": 291.35640874325844,
            "reason": null,
            "response_union_seconds": 1.4412586120306514,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 202.1541493741736,
            "trapezoid_pieces": 2,
            "watt_hours": 0.08093233576201624
          },
          "gpu_sample_count": 1,
          "host_sample_count": 1,
          "meaning": "Samples inside declared interval union. Means are arithmetic sample means; peaks need not coincide. Counter deltas exclude unsampled edges; not minimum required RAM.",
          "metrics": {
            "cgroup_memory_current_bytes": {
              "maximum": 6722093056,
              "mean": 6722093056,
              "median": 6722093056,
              "minimum": 6722093056,
              "sample_count": 1
            },
            "gpu_board_watts": {
              "maximum": 167.13,
              "mean": 167.13,
              "median": 167.13,
              "minimum": 167.13,
              "sample_count": 1
            },
            "gpu_memory_used_bytes": {
              "maximum": 6746537984.0,
              "mean": 6746537984.0,
              "median": 6746537984.0,
              "minimum": 6746537984.0,
              "sample_count": 1
            },
            "gpu_sm_clock_mhz": {
              "maximum": 2167.0,
              "mean": 2167.0,
              "median": 2167.0,
              "minimum": 2167.0,
              "sample_count": 1
            },
            "gpu_utilization_percent": {
              "maximum": 89.0,
              "mean": 89.0,
              "median": 89.0,
              "minimum": 89.0,
              "sample_count": 1
            },
            "host_mem_available_bytes": {
              "maximum": 96126963712,
              "mean": 96126963712,
              "median": 96126963712,
              "minimum": 96126963712,
              "sample_count": 1
            },
            "process_pss_bytes": {
              "maximum": 813629440,
              "mean": 813629440,
              "median": 813629440,
              "minimum": 813629440,
              "sample_count": 1
            },
            "process_rss_bytes": {
              "maximum": 816615424,
              "mean": 816615424,
              "median": 816615424,
              "minimum": 816615424,
              "sample_count": 1
            }
          },
          "process_sample_count": 1,
          "response_union_seconds": 1.4412586120306514,
          "sampled_counter_deltas": {
            "deltas": null,
            "observed_endpoint_seconds": 0.0,
            "observed_intervals": 0,
            "requested_intervals": 1
          }
        },
        "warmup": {
          "gpu_energy": {
            "available": true,
            "interpolation": "linear_at_bracketed_boundaries",
            "joules": 6.770582108400927,
            "reason": null,
            "response_union_seconds": 0.05492231895914301,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 123.27560519499544,
            "trapezoid_pieces": 1,
            "watt_hours": 0.0018807172523335907
          },
          "gpu_sample_count": 0,
          "host_sample_count": 0,
          "meaning": "Samples inside declared interval union. Means are arithmetic sample means; peaks need not coincide. Counter deltas exclude unsampled edges; not minimum required RAM.",
          "metrics": {
            "cgroup_memory_current_bytes": null,
            "gpu_board_watts": null,
            "gpu_memory_used_bytes": null,
            "gpu_sm_clock_mhz": null,
            "gpu_utilization_percent": null,
            "host_mem_available_bytes": null,
            "process_pss_bytes": null,
            "process_rss_bytes": null
          },
          "process_sample_count": 0,
          "response_union_seconds": 0.05492231895914301,
          "sampled_counter_deltas": {
            "deltas": null,
            "observed_endpoint_seconds": 0.0,
            "observed_intervals": 0,
            "requested_intervals": 1
          }
        }
      },
      "reviewed_export_sha256": "2d1a82644c4ba9ba504ddf8fd2c7fe3c253e15cc9811363c52186bd4b7510a07",
      "settings": {
        "batch": 2048,
        "cache_prompt": false,
        "cache_ram_bytes": 0,
        "concurrency": 1,
        "context_shift": false,
        "contexts": [],
        "endpoint": "/v1/chat/completions",
        "fit": "off",
        "flash_attention": true,
        "ignore_eos": false,
        "kv_cache": "f16",
        "lazy_mode": "off",
        "load_mode": "mmap",
        "measured_rounds": 24,
        "mlock": false,
        "network_scope": "loopback",
        "output_tokens": 1024,
        "p95_minimum_requests": 200,
        "placement": {
          "cpu_expert_layers": 0,
          "gpu_layers": "all"
        },
        "reasoning": "disabled",
        "seed": 20260927,
        "server_slots": 1,
        "synthetic_prompt_sha256": "c099b5abc034050fac1133e0214990d1f72a124b64f36e8a80ba3209bc07d187",
        "temperature": 0,
        "threads": 16,
        "tokenization": "natural_chat_template_no_padding",
        "total_context_tokens": 8704,
        "ubatch": 512,
        "warmup": "one_disjoint_transport_request",
        "warmup_rounds": 1
      },
      "status": "completed",
      "telemetry": {
        "gpu_samples": 8,
        "host_samples": 7,
        "maximum_gpu_gap_seconds": 1.004459120973479,
        "maximum_host_gap_seconds": 1.1064585979911499,
        "meaning": "Sampled observations; no numeric temperatures published and no warranty between samples.",
        "thermal_slowdown_fully_available": true,
        "thermal_slowdown_observed_samples": 0,
        "thermal_slowdown_states": {
          "hardware": {
            "active": 0,
            "not_active": 8,
            "unavailable": 0
          },
          "software": {
            "active": 0,
            "not_active": 8,
            "unavailable": 0
          }
        }
      }
    }
  ],
  "comparisons": [],
  "environment": {
    "kernel": "6.12.0-211.56.1.el10_2.0.1.x86_64",
    "nvidia_driver": "595.91.07",
    "os": {
      "id": "rocky",
      "name": "Rocky Linux",
      "version_id": "10.2"
    }
  },
  "hardware": {
    "cpu": "AMD EPYC 4565P 16-Core Processor",
    "fan_profile": "GPU300WAuto",
    "gpu": "NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition",
    "gpu_nominal_vram_gb": 96,
    "gpu_vram_bytes": 102641958912,
    "installed_ram_bytes": 103079215104,
    "kernel_mem_total_bytes": 100271353856,
    "logical_cpu_count": 32,
    "physical_cpu_count": 16
  },
  "historical_results_pooled": false,
  "limitations": [
    "Externally pinned capture integrity is not remote hardware attestation.",
    "Five repeated rounds share a process and are not five independent installations.",
    "Synthetic token latency is not practical first-visible answer latency.",
    "GPU board power is not whole-server AC power; energy is unavailable without bracketing samples.",
    "Sampled memory peaks may not coincide and do not establish minimum required system RAM.",
    "The practical corpus contains 24 authored tasks, not a broad model-quality benchmark.",
    "Engine telemetry includes process loading and internal warmup; exact per-sample resource windows are unavailable.",
    "No historical result is relabeled or pooled into this versioned series."
  ],
  "models": [
    {
      "architecture": "dense",
      "chat_template_sha256": "57f1fd00f0013a2be96aa79b857391f27e23df5b5f847072b524c897e24d0361",
      "effective_chat_template_rule": "llama.cpp-7fe450e19-chat-source-v1",
      "effective_chat_template_sha256": "57f1fd00f0013a2be96aa79b857391f27e23df5b5f847072b524c897e24d0361",
      "files": [
        {
          "bytes": 5027783488,
          "sha256": "d98cdcbd03e17ce47681435b5150e34c1417f50b5c0019dd560e4882c5745785"
        }
      ],
      "id": "qwen3-8b-q4",
      "name": "Qwen3-8B",
      "quantization": "Q4_K_M",
      "repository": "Qwen/Qwen3-8B-GGUF",
      "revision": "7c41481f57cb95916b40956ab2f0b139b296d974",
      "tokenizer_sha256": "2629971f294a1a43ef545d689a2090a820e0e1a15e5371e1c88d1c21159ef849"
    }
  ],
  "protocol": "gsh-llm-core/1.1",
  "runtime": {
    "archives_sha256": [
      "c2ab9e19838513ff69d1af8d999ad717dd3c7ee4714ac04c7ed5ab9077c50e4e",
      "1466daea60aad1144819e151b2bae19d54556cf1da6c129c4f55a5ded2637c25"
    ],
    "binary_sha256": "376626f953efd7bbf16dee112747196d2b1de26717bcd1d37e203340dc5c043c",
    "commit": "7fe450e19305b828c199d602c23a8337aaa1f03b",
    "name": "llama.cpp",
    "release": "b11146"
  },
  "schema": "gsh-llm-core-article/1",
  "sources": [
    {
      "exporter_sha256": "4f4046b91d99ba35175c3c6c04473a0021f54c6baa4892625cb66435f4c7e2ec",
      "finished_at": "2026-09-27T13:53:28.036613+00:00",
      "plan_id": "core-pilot8-20260927t135100-pilot8",
      "plan_sha256": "2952d4f1f06fa8282ea9ea2126cf5954bf3d774d16f9bc55a160ed6d808c3d49",
      "protocol_sha256": "3ce8ce9f2b17e66a655c07fd7e7aa3f1762790ad9be38f87d29e5e1279dec3ac",
      "record_sha256": "6c09a429d3b733b934749454b62d1a4243f5dce7c508680647c7d703a2b5daca",
      "reviewed_export_sha256": "2d1a82644c4ba9ba504ddf8fd2c7fe3c253e15cc9811363c52186bd4b7510a07",
      "runner_sha256": "1bac2436c5aa2eca4f4b42cc9b8f0942b5c66256eb1a41a2169b88c12b044f4e",
      "started_at": "2026-09-27T13:51:57.017889+00:00",
      "task_corpus_sha256": "0642db889565ce25b980e24afa5048eb83e4d15c36754457474a65b2c1f71c21"
    }
  ],
  "status": "completed",
  "test_date_timezone": "Europe/Warsaw",
  "tested_at": "2026-09-27"
}
