{
  "article_key": "llm-context-concurrency",
  "cases": [
    {
      "case_wall_seconds": 31.54574643599335,
      "completed_response_resources": {
        "gpu_energy": {
          "available": true,
          "interpolation": "linear_at_bracketed_boundaries",
          "joules": 5752.575337435909,
          "reason": null,
          "response_union_seconds": 19.211848054022994,
          "scope": "GPU board only; not whole-server AC energy",
          "time_weighted_mean_watts": 299.428525629595,
          "trapezoid_pieces": 29,
          "watt_hours": 1.597937593732197
        },
        "gpu_sample_count": 19,
        "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": 16754032640,
            "mean": 16745837837.473684,
            "median": 16747278336,
            "minimum": 16736788480,
            "sample_count": 19
          },
          "gpu_board_watts": {
            "maximum": 300.75,
            "mean": 299.9973684210526,
            "median": 299.98,
            "minimum": 299.21,
            "sample_count": 19
          },
          "gpu_memory_used_bytes": {
            "maximum": 10603200512.0,
            "mean": 10603200512.0,
            "median": 10603200512.0,
            "minimum": 10603200512.0,
            "sample_count": 19
          },
          "gpu_sm_clock_mhz": {
            "maximum": 1845.0,
            "mean": 1788.3157894736842,
            "median": 1777.0,
            "minimum": 1695.0,
            "sample_count": 19
          },
          "gpu_utilization_percent": {
            "maximum": 100.0,
            "mean": 97.73684210526316,
            "median": 97.0,
            "minimum": 97.0,
            "sample_count": 19
          },
          "host_mem_available_bytes": {
            "maximum": 96162541568,
            "mean": 96125036220.63158,
            "median": 96112386048,
            "minimum": 96093450240,
            "sample_count": 19
          },
          "process_pss_bytes": {
            "maximum": 821585920,
            "mean": 817439474.5263158,
            "median": 818100224,
            "minimum": 812947456,
            "sample_count": 19
          },
          "process_rss_bytes": {
            "maximum": 824573952,
            "mean": 820427506.5263158,
            "median": 821088256,
            "minimum": 815935488,
            "sample_count": 19
          }
        },
        "process_sample_count": 19,
        "response_union_seconds": 19.211848054022994,
        "sampled_counter_deltas": {
          "deltas": {
            "major_faults": 0,
            "read_syscall_bytes": 0,
            "storage_read_bytes": 0,
            "storage_write_bytes": 0
          },
          "observed_endpoint_seconds": 9.185685446020216,
          "observed_intervals": 7,
          "requested_intervals": 10
        }
      },
      "groups": [
        {
          "aggregate_output_tokens_per_second": 171.87876975330158,
          "completed_output_tokens": 1280,
          "completed_request_count": 5,
          "context_tokens": 2048,
          "cpu_snapshot_intervals": {
            "host_capacity_percent": 3.120186341505504,
            "interval_count": 5,
            "interval_seconds": 7.461493466049433,
            "meaning": "Weighted CPU counter rate across round-bracketing snapshots, not exact token time.",
            "one_logical_cpu_percent": 99.84596292817612,
            "process_cpu_seconds": 7.45
          },
          "failed_request_count": 0,
          "measured_rounds": 5,
          "mode": "completion",
          "resources": {
            "gpu_energy": {
              "available": true,
              "interpolation": "linear_at_bracketed_boundaries",
              "joules": 2226.7114520086775,
              "reason": null,
              "response_union_seconds": 7.447109389002435,
              "scope": "GPU board only; not whole-server AC energy",
              "time_weighted_mean_watts": 299.00345700534325,
              "trapezoid_pieces": 12,
              "watt_hours": 0.6185309588912993
            },
            "gpu_sample_count": 7,
            "host_sample_count": 7,
            "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": 16740397056,
                "mean": 16738646308.571428,
                "median": 16738406400,
                "minimum": 16736788480,
                "sample_count": 7
              },
              "gpu_board_watts": {
                "maximum": 300.11,
                "mean": 299.9857142857143,
                "median": 299.97,
                "minimum": 299.91,
                "sample_count": 7
              },
              "gpu_memory_used_bytes": {
                "maximum": 10603200512.0,
                "mean": 10603200512.0,
                "median": 10603200512.0,
                "minimum": 10603200512.0,
                "sample_count": 7
              },
              "gpu_sm_clock_mhz": {
                "maximum": 1845.0,
                "mean": 1833.0,
                "median": 1830.0,
                "minimum": 1822.0,
                "sample_count": 7
              },
              "gpu_utilization_percent": {
                "maximum": 97.0,
                "mean": 97.0,
                "median": 97.0,
                "minimum": 97.0,
                "sample_count": 7
              },
              "host_mem_available_bytes": {
                "maximum": 96162541568,
                "mean": 96152399286.85715,
                "median": 96153411584,
                "minimum": 96136941568,
                "sample_count": 7
              },
              "process_pss_bytes": {
                "maximum": 814241792,
                "mean": 813606326.8571428,
                "median": 813594624,
                "minimum": 812947456,
                "sample_count": 7
              },
              "process_rss_bytes": {
                "maximum": 817229824,
                "mean": 816594358.8571428,
                "median": 816582656,
                "minimum": 815935488,
                "sample_count": 7
              }
            },
            "process_sample_count": 7,
            "response_union_seconds": 7.447109389002435,
            "sampled_counter_deltas": {
              "deltas": {
                "major_faults": 0,
                "read_syscall_bytes": 0,
                "storage_read_bytes": 0,
                "storage_write_bytes": 0
              },
              "observed_endpoint_seconds": 2.034676303039305,
              "observed_intervals": 2,
              "requested_intervals": 5
            }
          },
          "response_outer_span_seconds": 7.469806003966369,
          "response_seconds": {
            "maximum": 1.4900457450421527,
            "mean": 1.489421877800487,
            "median": 1.4897573189809918,
            "minimum": 1.4879181029973552,
            "sample_count": 5
          },
          "response_window_union_seconds": 7.447109389002435,
          "server_decode_tokens_per_second": {
            "maximum": 196.57875870150093,
            "mean": 196.4327494545978,
            "median": 196.40161434424968,
            "minimum": 196.36970177217881,
            "sample_count": 5
          },
          "synthetic_first_token_seconds": {
            "maximum": 0.19175322900991887,
            "mean": 0.19115480119362474,
            "median": 0.1911611080286093,
            "minimum": 0.1905973819666542,
            "sample_count": 5
          },
          "warmup_completed_rounds": 1,
          "warmup_requests_per_round": 1
        },
        {
          "aggregate_output_tokens_per_second": 108.79969682673467,
          "completed_output_tokens": 1280,
          "completed_request_count": 5,
          "context_tokens": 8192,
          "cpu_snapshot_intervals": {
            "host_capacity_percent": 3.118724668523497,
            "interval_count": 5,
            "interval_seconds": 11.783662844914943,
            "meaning": "Weighted CPU counter rate across round-bracketing snapshots, not exact token time.",
            "one_logical_cpu_percent": 99.7991893927519,
            "process_cpu_seconds": 11.76
          },
          "failed_request_count": 0,
          "measured_rounds": 5,
          "mode": "completion",
          "resources": {
            "gpu_energy": {
              "available": true,
              "interpolation": "linear_at_bracketed_boundaries",
              "joules": 3525.863885427233,
              "reason": null,
              "response_union_seconds": 11.764738665020559,
              "scope": "GPU board only; not whole-server AC energy",
              "time_weighted_mean_watts": 299.6975951459498,
              "trapezoid_pieces": 17,
              "watt_hours": 0.9794066348408981
            },
            "gpu_sample_count": 12,
            "host_sample_count": 12,
            "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": 16754032640,
                "mean": 16750032896,
                "median": 16750100480.0,
                "minimum": 16746192896,
                "sample_count": 12
              },
              "gpu_board_watts": {
                "maximum": 300.75,
                "mean": 300.00416666666666,
                "median": 300.04499999999996,
                "minimum": 299.21,
                "sample_count": 12
              },
              "gpu_memory_used_bytes": {
                "maximum": 10603200512.0,
                "mean": 10603200512.0,
                "median": 10603200512.0,
                "minimum": 10603200512.0,
                "sample_count": 12
              },
              "gpu_sm_clock_mhz": {
                "maximum": 1822.0,
                "mean": 1762.25,
                "median": 1766.0,
                "minimum": 1695.0,
                "sample_count": 12
              },
              "gpu_utilization_percent": {
                "maximum": 100.0,
                "mean": 98.16666666666667,
                "median": 97.5,
                "minimum": 97.0,
                "sample_count": 12
              },
              "host_mem_available_bytes": {
                "maximum": 96129368064,
                "mean": 96109074432,
                "median": 96109725696.0,
                "minimum": 96093450240,
                "sample_count": 12
              },
              "process_pss_bytes": {
                "maximum": 821585920,
                "mean": 819675477.3333334,
                "median": 819517440.0,
                "minimum": 818100224,
                "sample_count": 12
              },
              "process_rss_bytes": {
                "maximum": 824573952,
                "mean": 822663509.3333334,
                "median": 822505472.0,
                "minimum": 821088256,
                "sample_count": 12
              }
            },
            "process_sample_count": 12,
            "response_union_seconds": 11.764738665020559,
            "sampled_counter_deltas": {
              "deltas": {
                "major_faults": 0,
                "read_syscall_bytes": 0,
                "storage_read_bytes": 0,
                "storage_write_bytes": 0
              },
              "observed_endpoint_seconds": 7.151009142980911,
              "observed_intervals": 5,
              "requested_intervals": 5
            }
          },
          "response_outer_span_seconds": 11.794427076005377,
          "response_seconds": {
            "maximum": 2.3576946139801294,
            "mean": 2.352947733004112,
            "median": 2.354433296015486,
            "minimum": 2.34604896401288,
            "sample_count": 5
          },
          "response_window_union_seconds": 11.764738665020559,
          "server_decode_tokens_per_second": {
            "maximum": 168.74107905618814,
            "mean": 168.25912788557528,
            "median": 168.20414046035162,
            "minimum": 167.88277437147653,
            "sample_count": 5
          },
          "synthetic_first_token_seconds": {
            "maximum": 0.8393656879779883,
            "mean": 0.8372448727837764,
            "median": 0.8369915079674684,
            "minimum": 0.8346778709674254,
            "sample_count": 5
          },
          "warmup_completed_rounds": 1,
          "warmup_requests_per_round": 1
        }
      ],
      "id": "concurrency-8b-c1",
      "model": "qwen3-8b-q4",
      "module": "concurrency",
      "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": "cfb0005252065662c3806270e4fb851d33ec2bc1e4fcec950a2872a8f015cbd6",
      "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": 15503560704,
              "mean": 15494721536,
              "median": 15494721536.0,
              "minimum": 15485882368,
              "sample_count": 2
            },
            "gpu_board_watts": {
              "maximum": 15.74,
              "mean": 15.645,
              "median": 15.645,
              "minimum": 15.55,
              "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": 96575950848,
              "mean": 96570210304,
              "median": 96570210304.0,
              "minimum": 96564469760,
              "sample_count": 2
            },
            "process_pss_bytes": null,
            "process_rss_bytes": null
          },
          "process_sample_count": 0,
          "response_union_seconds": 1.1004508669720963,
          "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": 16422359040,
              "mean": 16422359040,
              "median": 16422359040,
              "minimum": 16422359040,
              "sample_count": 1
            },
            "gpu_board_watts": {
              "maximum": 81.37,
              "mean": 81.37,
              "median": 81.37,
              "minimum": 81.37,
              "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": 2332.0,
              "mean": 2332.0,
              "median": 2332.0,
              "minimum": 2332.0,
              "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": 96489291776,
              "mean": 96489291776,
              "median": 96489291776,
              "minimum": 96489291776,
              "sample_count": 1
            },
            "process_pss_bytes": null,
            "process_rss_bytes": null
          },
          "process_sample_count": 0,
          "response_union_seconds": 1.1638742850045674,
          "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": 140.7218834464672,
            "reason": null,
            "response_union_seconds": 2.7998547020251863,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 50.26042363722677,
            "trapezoid_pieces": 3,
            "watt_hours": 0.03908941206846311
          },
          "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": 16758321152,
              "mean": 16441323520,
              "median": 16441323520.0,
              "minimum": 16124325888,
              "sample_count": 2
            },
            "gpu_board_watts": {
              "maximum": 68.68,
              "mean": 52.83500000000001,
              "median": 52.83500000000001,
              "minimum": 36.99,
              "sample_count": 2
            },
            "gpu_memory_used_bytes": {
              "maximum": 5259657216.0,
              "mean": 2923429888.0,
              "median": 2923429888.0,
              "minimum": 587202560.0,
              "sample_count": 2
            },
            "gpu_sm_clock_mhz": {
              "maximum": 2272.0,
              "mean": 2272.0,
              "median": 2272.0,
              "minimum": 2272.0,
              "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": 96308260864,
              "mean": 96301864960,
              "median": 96301864960.0,
              "minimum": 96295469056,
              "sample_count": 2
            },
            "process_pss_bytes": {
              "maximum": 5363132416,
              "mean": 3241467904,
              "median": 3241467904.0,
              "minimum": 1119803392,
              "sample_count": 2
            },
            "process_rss_bytes": {
              "maximum": 5366120448,
              "mean": 3244455936,
              "median": 3244455936.0,
              "minimum": 1122791424,
              "sample_count": 2
            }
          },
          "process_sample_count": 2,
          "response_union_seconds": 2.7998547020251863,
          "sampled_counter_deltas": {
            "deltas": {
              "major_faults": 1,
              "read_syscall_bytes": 5,
              "storage_read_bytes": 988917760,
              "storage_write_bytes": 0
            },
            "observed_endpoint_seconds": 1.070253963000141,
            "observed_intervals": 1,
            "requested_intervals": 1
          }
        },
        "ready": {
          "gpu_energy": {
            "available": true,
            "interpolation": "linear_at_bracketed_boundaries",
            "joules": 485.16816934820196,
            "reason": null,
            "response_union_seconds": 3.326905186928343,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 145.83167901942733,
            "trapezoid_pieces": 8,
            "watt_hours": 0.1347689359300561
          },
          "gpu_sample_count": 5,
          "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": 16755085312,
              "mean": 16734448298.666666,
              "median": 16740896768,
              "minimum": 16707362816,
              "sample_count": 3
            },
            "gpu_board_watts": {
              "maximum": 300.08,
              "mean": 133.35399999999998,
              "median": 78.6,
              "minimum": 75.12,
              "sample_count": 5
            },
            "gpu_memory_used_bytes": {
              "maximum": 10603200512.0,
              "mean": 10595650764.8,
              "median": 10603200512.0,
              "minimum": 10584326144.0,
              "sample_count": 5
            },
            "gpu_sm_clock_mhz": {
              "maximum": 2332.0,
              "mean": 2213.6,
              "median": 2302.0,
              "minimum": 1822.0,
              "sample_count": 5
            },
            "gpu_utilization_percent": {
              "maximum": 97.0,
              "mean": 19.4,
              "median": 0.0,
              "minimum": 0.0,
              "sample_count": 5
            },
            "host_mem_available_bytes": {
              "maximum": 96211808256,
              "mean": 96155060906.66667,
              "median": 96133586944,
              "minimum": 96119787520,
              "sample_count": 3
            },
            "process_pss_bytes": {
              "maximum": 821585920,
              "mean": 805507754.6666666,
              "median": 814356480,
              "minimum": 780580864,
              "sample_count": 3
            },
            "process_rss_bytes": {
              "maximum": 824573952,
              "mean": 808495786.6666666,
              "median": 817344512,
              "minimum": 783568896,
              "sample_count": 3
            }
          },
          "process_sample_count": 3,
          "response_union_seconds": 3.326905186928343,
          "sampled_counter_deltas": {
            "deltas": null,
            "observed_endpoint_seconds": 0.0,
            "observed_intervals": 0,
            "requested_intervals": 3
          }
        },
        "responses": {
          "gpu_energy": {
            "available": true,
            "interpolation": "linear_at_bracketed_boundaries",
            "joules": 5771.959698463556,
            "reason": null,
            "response_union_seconds": 19.277330336975865,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 299.4169627000869,
            "trapezoid_pieces": 21,
            "watt_hours": 1.603322138462099
          },
          "gpu_sample_count": 19,
          "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": 16754032640,
              "mean": 16745837837.473684,
              "median": 16747278336,
              "minimum": 16736788480,
              "sample_count": 19
            },
            "gpu_board_watts": {
              "maximum": 300.75,
              "mean": 299.9973684210526,
              "median": 299.98,
              "minimum": 299.21,
              "sample_count": 19
            },
            "gpu_memory_used_bytes": {
              "maximum": 10603200512.0,
              "mean": 10603200512.0,
              "median": 10603200512.0,
              "minimum": 10603200512.0,
              "sample_count": 19
            },
            "gpu_sm_clock_mhz": {
              "maximum": 1845.0,
              "mean": 1788.3157894736842,
              "median": 1777.0,
              "minimum": 1695.0,
              "sample_count": 19
            },
            "gpu_utilization_percent": {
              "maximum": 100.0,
              "mean": 97.73684210526316,
              "median": 97.0,
              "minimum": 97.0,
              "sample_count": 19
            },
            "host_mem_available_bytes": {
              "maximum": 96162541568,
              "mean": 96125036220.63158,
              "median": 96112386048,
              "minimum": 96093450240,
              "sample_count": 19
            },
            "process_pss_bytes": {
              "maximum": 821585920,
              "mean": 817439474.5263158,
              "median": 818100224,
              "minimum": 812947456,
              "sample_count": 19
            },
            "process_rss_bytes": {
              "maximum": 824573952,
              "mean": 820427506.5263158,
              "median": 821088256,
              "minimum": 815935488,
              "sample_count": 19
            }
          },
          "process_sample_count": 19,
          "response_union_seconds": 19.277330336975865,
          "sampled_counter_deltas": {
            "deltas": {
              "major_faults": 0,
              "read_syscall_bytes": 0,
              "storage_read_bytes": 0,
              "storage_write_bytes": 16384
            },
            "observed_endpoint_seconds": 17.382738566957414,
            "observed_intervals": 2,
            "requested_intervals": 2
          }
        },
        "warmup": {
          "gpu_energy": {
            "available": true,
            "interpolation": "linear_at_bracketed_boundaries",
            "joules": 872.0933225952482,
            "reason": null,
            "response_union_seconds": 3.8364628920680843,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 227.31702277071614,
            "trapezoid_pieces": 5,
            "watt_hours": 0.24224814516534673
          },
          "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": 16744005632,
              "mean": 16737950720,
              "median": 16738725888.0,
              "minimum": 16730345472,
              "sample_count": 4
            },
            "gpu_board_watts": {
              "maximum": 300.32,
              "mean": 273.2,
              "median": 277.67,
              "minimum": 241.61,
              "sample_count": 3
            },
            "gpu_memory_used_bytes": {
              "maximum": 10603200512.0,
              "mean": 10603200512.0,
              "median": 10603200512.0,
              "minimum": 10603200512.0,
              "sample_count": 3
            },
            "gpu_sm_clock_mhz": {
              "maximum": 1837.0,
              "mean": 1782.0,
              "median": 1777.0,
              "minimum": 1732.0,
              "sample_count": 3
            },
            "gpu_utilization_percent": {
              "maximum": 98.0,
              "mean": 97.33333333333333,
              "median": 97.0,
              "minimum": 97.0,
              "sample_count": 3
            },
            "host_mem_available_bytes": {
              "maximum": 96168611840,
              "mean": 96138284032,
              "median": 96137695232.0,
              "minimum": 96109133824,
              "sample_count": 4
            },
            "process_pss_bytes": {
              "maximum": 817391616,
              "mean": 812983296,
              "median": 814118912.0,
              "minimum": 806303744,
              "sample_count": 4
            },
            "process_rss_bytes": {
              "maximum": 820379648,
              "mean": 815971328,
              "median": 817106944.0,
              "minimum": 809291776,
              "sample_count": 4
            }
          },
          "process_sample_count": 4,
          "response_union_seconds": 3.8364628920680843,
          "sampled_counter_deltas": {
            "deltas": {
              "major_faults": 0,
              "read_syscall_bytes": 0,
              "storage_read_bytes": 0,
              "storage_write_bytes": 0
            },
            "observed_endpoint_seconds": 2.0357347379322164,
            "observed_intervals": 2,
            "requested_intervals": 2
          }
        }
      },
      "reviewed_export_sha256": "59f09a9b924c4cb4ae1370f85109347542e72457c07f0a7a8ff351b31fedab64",
      "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": 4,
        "synthetic_prompt_sha256": "c099b5abc034050fac1133e0214990d1f72a124b64f36e8a80ba3209bc07d187",
        "temperature": 0,
        "threads": 16,
        "tokenization": "model_tokenizer_add_special_false_repeat_then_truncate",
        "total_context_tokens": 34816,
        "ubatch": 512,
        "warmup_rounds": 1
      },
      "status": "completed",
      "telemetry": {
        "gpu_samples": 32,
        "host_samples": 31,
        "maximum_gpu_gap_seconds": 1.0082540499861352,
        "maximum_host_gap_seconds": 1.070253963000141,
        "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": 32,
            "unavailable": 0
          },
          "software": {
            "active": 0,
            "not_active": 32,
            "unavailable": 0
          }
        }
      }
    },
    {
      "case_wall_seconds": 143.20899132004706,
      "completed_response_resources": {
        "gpu_energy": {
          "available": true,
          "interpolation": "linear_at_bracketed_boundaries",
          "joules": 14770.156143266022,
          "reason": null,
          "response_union_seconds": 49.281771619047504,
          "scope": "GPU board only; not whole-server AC energy",
          "time_weighted_mean_watts": 299.70830305047974,
          "trapezoid_pieces": 60,
          "watt_hours": 4.1028211509072285
        },
        "gpu_sample_count": 50,
        "host_sample_count": 48,
        "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": 16972275712,
            "mean": 16944400810.666666,
            "median": 16953174016.0,
            "minimum": 16903892992,
            "sample_count": 48
          },
          "gpu_board_watts": {
            "maximum": 300.93,
            "mean": 299.8862,
            "median": 299.99,
            "minimum": 296.32,
            "sample_count": 50
          },
          "gpu_memory_used_bytes": {
            "maximum": 10605297664.0,
            "mean": 10604584632.32,
            "median": 10605297664.0,
            "minimum": 10603200512.0,
            "sample_count": 50
          },
          "gpu_sm_clock_mhz": {
            "maximum": 1860.0,
            "mean": 1748.2,
            "median": 1751.0,
            "minimum": 1680.0,
            "sample_count": 50
          },
          "gpu_utilization_percent": {
            "maximum": 100.0,
            "mean": 96.26,
            "median": 96.0,
            "minimum": 77.0,
            "sample_count": 50
          },
          "host_mem_available_bytes": {
            "maximum": 96069586944,
            "mean": 96035869354.66667,
            "median": 96034033664.0,
            "minimum": 96010240000,
            "sample_count": 48
          },
          "process_pss_bytes": {
            "maximum": 858840064,
            "mean": 846069888,
            "median": 851729408.0,
            "minimum": 827268096,
            "sample_count": 48
          },
          "process_rss_bytes": {
            "maximum": 861831168,
            "mean": 849090560,
            "median": 854720512.0,
            "minimum": 830259200,
            "sample_count": 48
          }
        },
        "process_sample_count": 48,
        "response_union_seconds": 49.281771619047504,
        "sampled_counter_deltas": {
          "deltas": {
            "major_faults": 0,
            "read_syscall_bytes": 0,
            "storage_read_bytes": 0,
            "storage_write_bytes": 28672
          },
          "observed_endpoint_seconds": 38.876957018102985,
          "observed_intervals": 10,
          "requested_intervals": 10
        }
      },
      "groups": [
        {
          "aggregate_output_tokens_per_second": 313.01834665113336,
          "completed_output_tokens": 5120,
          "completed_request_count": 20,
          "context_tokens": 2048,
          "cpu_snapshot_intervals": {
            "host_capacity_percent": 3.1419778907509697,
            "interval_count": 5,
            "interval_seconds": 16.371057273005135,
            "meaning": "Weighted CPU counter rate across round-bracketing snapshots, not exact token time.",
            "one_logical_cpu_percent": 100.54329250403103,
            "process_cpu_seconds": 16.46
          },
          "failed_request_count": 0,
          "measured_rounds": 5,
          "mode": "completion",
          "resources": {
            "gpu_energy": {
              "available": true,
              "interpolation": "linear_at_bracketed_boundaries",
              "joules": 4901.074335239871,
              "reason": null,
              "response_union_seconds": 16.356868710019626,
              "scope": "GPU board only; not whole-server AC energy",
              "time_weighted_mean_watts": 299.6340205529467,
              "trapezoid_pieces": 22,
              "watt_hours": 1.3614095375666309
            },
            "gpu_sample_count": 17,
            "host_sample_count": 16,
            "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": 16922185728,
                "mean": 16913083904,
                "median": 16913446912.0,
                "minimum": 16903892992,
                "sample_count": 16
              },
              "gpu_board_watts": {
                "maximum": 300.33,
                "mean": 299.7952941176471,
                "median": 300.02,
                "minimum": 296.32,
                "sample_count": 17
              },
              "gpu_memory_used_bytes": {
                "maximum": 10603200512.0,
                "mean": 10603200512.0,
                "median": 10603200512.0,
                "minimum": 10603200512.0,
                "sample_count": 17
              },
              "gpu_sm_clock_mhz": {
                "maximum": 1860.0,
                "mean": 1759.2941176470588,
                "median": 1755.0,
                "minimum": 1725.0,
                "sample_count": 17
              },
              "gpu_utilization_percent": {
                "maximum": 100.0,
                "mean": 95.05882352941177,
                "median": 96.0,
                "minimum": 77.0,
                "sample_count": 17
              },
              "host_mem_available_bytes": {
                "maximum": 96069586944,
                "mean": 96051451904,
                "median": 96053071872.0,
                "minimum": 96031924224,
                "sample_count": 16
              },
              "process_pss_bytes": {
                "maximum": 832879616,
                "mean": 830089728,
                "median": 830098432.0,
                "minimum": 827268096,
                "sample_count": 16
              },
              "process_rss_bytes": {
                "maximum": 835870720,
                "mean": 833080832,
                "median": 833089536.0,
                "minimum": 830259200,
                "sample_count": 16
              }
            },
            "process_sample_count": 16,
            "response_union_seconds": 16.356868710019626,
            "sampled_counter_deltas": {
              "deltas": {
                "major_faults": 0,
                "read_syscall_bytes": 0,
                "storage_read_bytes": 0,
                "storage_write_bytes": 4096
              },
              "observed_endpoint_seconds": 11.232935156032909,
              "observed_intervals": 5,
              "requested_intervals": 5
            }
          },
          "response_outer_span_seconds": 16.403912629000843,
          "response_seconds": {
            "maximum": 3.28185669897357,
            "mean": 3.258913619956002,
            "median": 3.2634908619802445,
            "minimum": 3.2179629480233416,
            "sample_count": 20
          },
          "response_window_union_seconds": 16.356868710019626,
          "server_decode_tokens_per_second": {
            "maximum": 104.92626769447776,
            "mean": 96.54218459985283,
            "median": 99.48448007706719,
            "minimum": 83.302244717821,
            "sample_count": 20
          },
          "synthetic_first_token_seconds": {
            "maximum": 0.825031254964415,
            "mean": 0.5970895664999262,
            "median": 0.692155514494516,
            "minimum": 0.18841045402223244,
            "sample_count": 20
          },
          "warmup_completed_rounds": 1,
          "warmup_requests_per_round": 4
        },
        {
          "aggregate_output_tokens_per_second": 155.5053940218641,
          "completed_output_tokens": 5120,
          "completed_request_count": 20,
          "context_tokens": 8192,
          "cpu_snapshot_intervals": {
            "host_capacity_percent": 3.1331217145339085,
            "interval_count": 5,
            "interval_seconds": 32.94437924999511,
            "meaning": "Weighted CPU counter rate across round-bracketing snapshots, not exact token time.",
            "one_logical_cpu_percent": 100.25989486508507,
            "process_cpu_seconds": 33.03
          },
          "failed_request_count": 0,
          "measured_rounds": 5,
          "mode": "completion",
          "resources": {
            "gpu_energy": {
              "available": true,
              "interpolation": "linear_at_bracketed_boundaries",
              "joules": 9869.081808026149,
              "reason": null,
              "response_union_seconds": 32.92490290902788,
              "scope": "GPU board only; not whole-server AC energy",
              "time_weighted_mean_watts": 299.7452060920151,
              "trapezoid_pieces": 38,
              "watt_hours": 2.741411613340597
            },
            "gpu_sample_count": 33,
            "host_sample_count": 32,
            "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": 16972275712,
                "mean": 16960059264,
                "median": 16959395840.0,
                "minimum": 16945074176,
                "sample_count": 32
              },
              "gpu_board_watts": {
                "maximum": 300.93,
                "mean": 299.9330303030303,
                "median": 299.97,
                "minimum": 298.18,
                "sample_count": 33
              },
              "gpu_memory_used_bytes": {
                "maximum": 10605297664.0,
                "mean": 10605297664.0,
                "median": 10605297664.0,
                "minimum": 10605297664.0,
                "sample_count": 33
              },
              "gpu_sm_clock_mhz": {
                "maximum": 1845.0,
                "mean": 1742.4848484848485,
                "median": 1725.0,
                "minimum": 1680.0,
                "sample_count": 33
              },
              "gpu_utilization_percent": {
                "maximum": 100.0,
                "mean": 96.87878787878788,
                "median": 96.0,
                "minimum": 79.0,
                "sample_count": 33
              },
              "host_mem_available_bytes": {
                "maximum": 96050204672,
                "mean": 96028078080,
                "median": 96025366528.0,
                "minimum": 96010240000,
                "sample_count": 32
              },
              "process_pss_bytes": {
                "maximum": 858840064,
                "mean": 854059968,
                "median": 854100992.0,
                "minimum": 847817728,
                "sample_count": 32
              },
              "process_rss_bytes": {
                "maximum": 861831168,
                "mean": 857095424,
                "median": 857092096.0,
                "minimum": 850808832,
                "sample_count": 32
              }
            },
            "process_sample_count": 32,
            "response_union_seconds": 32.92490290902788,
            "sampled_counter_deltas": {
              "deltas": {
                "major_faults": 0,
                "read_syscall_bytes": 0,
                "storage_read_bytes": 0,
                "storage_write_bytes": 24576
              },
              "observed_endpoint_seconds": 27.644021862070076,
              "observed_intervals": 5,
              "requested_intervals": 5
            }
          },
          "response_outer_span_seconds": 32.988753579964396,
          "response_seconds": {
            "maximum": 6.621589451970067,
            "mean": 6.539755831405637,
            "median": 6.544440800498705,
            "minimum": 6.456752622034401,
            "sample_count": 20
          },
          "response_window_union_seconds": 32.92490290902788,
          "server_decode_tokens_per_second": {
            "maximum": 84.3563330599747,
            "mean": 64.69884639525425,
            "median": 63.226577706461605,
            "minimum": 46.57408347683021,
            "sample_count": 20
          },
          "synthetic_first_token_seconds": {
            "maximum": 3.577183506975416,
            "mean": 2.4010472706053405,
            "median": 2.47954356399714,
            "minimum": 1.0299478119704872,
            "sample_count": 20
          },
          "warmup_completed_rounds": 1,
          "warmup_requests_per_round": 4
        }
      ],
      "id": "concurrency-8b-c4",
      "model": "qwen3-8b-q4",
      "module": "concurrency",
      "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": "546c5e177b7e76804c6d4a2aedb3859980726dc415bfddd3c1a7373c9774519f",
      "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": 16447270912,
              "mean": 16438515712,
              "median": 16438515712.0,
              "minimum": 16429760512,
              "sample_count": 2
            },
            "gpu_board_watts": {
              "maximum": 16.25,
              "mean": 16.244999999999997,
              "median": 16.244999999999997,
              "minimum": 16.24,
              "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": 96482447360,
              "mean": 96481126400,
              "median": 96481126400.0,
              "minimum": 96479805440,
              "sample_count": 2
            },
            "process_pss_bytes": null,
            "process_rss_bytes": null
          },
          "process_sample_count": 0,
          "response_union_seconds": 1.1007217759615742,
          "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": 2,
          "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": 16606367744,
              "mean": 16606367744,
              "median": 16606367744,
              "minimum": 16606367744,
              "sample_count": 1
            },
            "gpu_board_watts": {
              "maximum": 90.64,
              "mean": 90.15,
              "median": 90.15,
              "minimum": 89.66,
              "sample_count": 2
            },
            "gpu_memory_used_bytes": {
              "maximum": 10605297664.0,
              "mean": 5303697408.0,
              "median": 5303697408.0,
              "minimum": 2097152.0,
              "sample_count": 2
            },
            "gpu_sm_clock_mhz": {
              "maximum": 2340.0,
              "mean": 2328.5,
              "median": 2328.5,
              "minimum": 2317.0,
              "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": 96498147328,
              "mean": 96498147328,
              "median": 96498147328,
              "minimum": 96498147328,
              "sample_count": 1
            },
            "process_pss_bytes": null,
            "process_rss_bytes": null
          },
          "process_sample_count": 0,
          "response_union_seconds": 1.1552442620159127,
          "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": 53.5790935900832,
            "reason": null,
            "response_union_seconds": 1.3030608020490035,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 41.11787685258625,
            "trapezoid_pieces": 2,
            "watt_hours": 0.014883081552800888
          },
          "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": 16777781248,
              "mean": 16777781248,
              "median": 16777781248,
              "minimum": 16777781248,
              "sample_count": 1
            },
            "gpu_board_watts": {
              "maximum": 50.83,
              "mean": 50.83,
              "median": 50.83,
              "minimum": 50.83,
              "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": 2287.0,
              "mean": 2287.0,
              "median": 2287.0,
              "minimum": 2287.0,
              "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": 96183443456,
              "mean": 96183443456,
              "median": 96183443456,
              "minimum": 96183443456,
              "sample_count": 1
            },
            "process_pss_bytes": {
              "maximum": 5363969024,
              "mean": 5363969024,
              "median": 5363969024,
              "minimum": 5363969024,
              "sample_count": 1
            },
            "process_rss_bytes": {
              "maximum": 5366960128,
              "mean": 5366960128,
              "median": 5366960128,
              "minimum": 5366960128,
              "sample_count": 1
            }
          },
          "process_sample_count": 1,
          "response_union_seconds": 1.3030608020490035,
          "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": 528.2204930815988,
            "reason": null,
            "response_union_seconds": 3.318441729003098,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 159.1772693987557,
            "trapezoid_pieces": 6,
            "watt_hours": 0.14672791474488855
          },
          "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": 16974327808,
              "mean": 16917742933.333334,
              "median": 16924258304,
              "minimum": 16854642688,
              "sample_count": 3
            },
            "gpu_board_watts": {
              "maximum": 280.61,
              "mean": 165.19,
              "median": 134.2,
              "minimum": 80.76,
              "sample_count": 3
            },
            "gpu_memory_used_bytes": {
              "maximum": 10605297664.0,
              "mean": 10597608106.666666,
              "median": 10603200512.0,
              "minimum": 10584326144.0,
              "sample_count": 3
            },
            "gpu_sm_clock_mhz": {
              "maximum": 2332.0,
              "mean": 2149.6666666666665,
              "median": 2310.0,
              "minimum": 1807.0,
              "sample_count": 3
            },
            "gpu_utilization_percent": {
              "maximum": 1.0,
              "mean": 0.3333333333333333,
              "median": 0.0,
              "minimum": 0.0,
              "sample_count": 3
            },
            "host_mem_available_bytes": {
              "maximum": 96106070016,
              "mean": 96071060138.66667,
              "median": 96063447040,
              "minimum": 96043663360,
              "sample_count": 3
            },
            "process_pss_bytes": {
              "maximum": 858840064,
              "mean": 824471893.3333334,
              "median": 833178624,
              "minimum": 781396992,
              "sample_count": 3
            },
            "process_rss_bytes": {
              "maximum": 861831168,
              "mean": 827462997.3333334,
              "median": 836169728,
              "minimum": 784388096,
              "sample_count": 3
            }
          },
          "process_sample_count": 3,
          "response_union_seconds": 3.318441729003098,
          "sampled_counter_deltas": {
            "deltas": null,
            "observed_endpoint_seconds": 0.0,
            "observed_intervals": 0,
            "requested_intervals": 3
          }
        },
        "responses": {
          "gpu_energy": {
            "available": true,
            "interpolation": "linear_at_bracketed_boundaries",
            "joules": 14811.261750294023,
            "reason": null,
            "response_union_seconds": 49.420940532989334,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 299.6960719597647,
            "trapezoid_pieces": 52,
            "watt_hours": 4.114239375081673
          },
          "gpu_sample_count": 50,
          "host_sample_count": 48,
          "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": 16972275712,
              "mean": 16944400810.666666,
              "median": 16953174016.0,
              "minimum": 16903892992,
              "sample_count": 48
            },
            "gpu_board_watts": {
              "maximum": 300.93,
              "mean": 299.8862,
              "median": 299.99,
              "minimum": 296.32,
              "sample_count": 50
            },
            "gpu_memory_used_bytes": {
              "maximum": 10605297664.0,
              "mean": 10604584632.32,
              "median": 10605297664.0,
              "minimum": 10603200512.0,
              "sample_count": 50
            },
            "gpu_sm_clock_mhz": {
              "maximum": 1860.0,
              "mean": 1748.2,
              "median": 1751.0,
              "minimum": 1680.0,
              "sample_count": 50
            },
            "gpu_utilization_percent": {
              "maximum": 100.0,
              "mean": 96.26,
              "median": 96.0,
              "minimum": 77.0,
              "sample_count": 50
            },
            "host_mem_available_bytes": {
              "maximum": 96069586944,
              "mean": 96035869354.66667,
              "median": 96034033664.0,
              "minimum": 96010240000,
              "sample_count": 48
            },
            "process_pss_bytes": {
              "maximum": 858840064,
              "mean": 846069888,
              "median": 851729408.0,
              "minimum": 827268096,
              "sample_count": 48
            },
            "process_rss_bytes": {
              "maximum": 861831168,
              "mean": 849090560,
              "median": 854720512.0,
              "minimum": 830259200,
              "sample_count": 48
            }
          },
          "process_sample_count": 48,
          "response_union_seconds": 49.420940532989334,
          "sampled_counter_deltas": {
            "deltas": {
              "major_faults": 0,
              "read_syscall_bytes": 0,
              "storage_read_bytes": 0,
              "storage_write_bytes": 86016
            },
            "observed_endpoint_seconds": 47.11209448100999,
            "observed_intervals": 2,
            "requested_intervals": 2
          }
        },
        "warmup": {
          "gpu_energy": {
            "available": true,
            "interpolation": "linear_at_bracketed_boundaries",
            "joules": 2686.3683206647843,
            "reason": null,
            "response_union_seconds": 9.834585551987402,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 273.15521396038037,
            "trapezoid_pieces": 11,
            "watt_hours": 0.7462134224068845
          },
          "gpu_sample_count": 9,
          "host_sample_count": 10,
          "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": 16940253184,
              "mean": 16923122892.8,
              "median": 16929480704.0,
              "minimum": 16887836672,
              "sample_count": 10
            },
            "gpu_board_watts": {
              "maximum": 300.17,
              "mean": 276.2988888888889,
              "median": 299.98,
              "minimum": 174.1,
              "sample_count": 9
            },
            "gpu_memory_used_bytes": {
              "maximum": 10605297664.0,
              "mean": 10604132579.555555,
              "median": 10603200512.0,
              "minimum": 10603200512.0,
              "sample_count": 9
            },
            "gpu_sm_clock_mhz": {
              "maximum": 1800.0,
              "mean": 1750.5555555555557,
              "median": 1747.0,
              "minimum": 1717.0,
              "sample_count": 9
            },
            "gpu_utilization_percent": {
              "maximum": 100.0,
              "mean": 96.22222222222223,
              "median": 96.0,
              "minimum": 92.0,
              "sample_count": 9
            },
            "host_mem_available_bytes": {
              "maximum": 96077516800,
              "mean": 96048424140.8,
              "median": 96048119808.0,
              "minimum": 96022204416,
              "sample_count": 10
            },
            "process_pss_bytes": {
              "maximum": 845200384,
              "mean": 835459686.4,
              "median": 835929088.0,
              "minimum": 816208896,
              "sample_count": 10
            },
            "process_rss_bytes": {
              "maximum": 848191488,
              "mean": 838450790.4,
              "median": 838920192.0,
              "minimum": 819200000,
              "sample_count": 10
            }
          },
          "process_sample_count": 10,
          "response_union_seconds": 9.834585551987402,
          "sampled_counter_deltas": {
            "deltas": {
              "major_faults": 16,
              "read_syscall_bytes": 0,
              "storage_read_bytes": 2822144,
              "storage_write_bytes": 8192
            },
            "observed_endpoint_seconds": 8.234222369035706,
            "observed_intervals": 2,
            "requested_intervals": 2
          }
        }
      },
      "reviewed_export_sha256": "59f09a9b924c4cb4ae1370f85109347542e72457c07f0a7a8ff351b31fedab64",
      "settings": {
        "batch": 2048,
        "cache_prompt": false,
        "cache_ram_bytes": 0,
        "concurrency": 4,
        "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": 4,
        "synthetic_prompt_sha256": "c099b5abc034050fac1133e0214990d1f72a124b64f36e8a80ba3209bc07d187",
        "temperature": 0,
        "threads": 16,
        "tokenization": "model_tokenizer_add_special_false_repeat_then_truncate",
        "total_context_tokens": 34816,
        "ubatch": 512,
        "warmup_rounds": 1
      },
      "status": "completed",
      "telemetry": {
        "gpu_samples": 67,
        "host_samples": 65,
        "maximum_gpu_gap_seconds": 1.059977461991366,
        "maximum_host_gap_seconds": 1.092143846966792,
        "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": 67,
            "unavailable": 0
          },
          "software": {
            "active": 0,
            "not_active": 67,
            "unavailable": 0
          }
        }
      }
    },
    {
      "case_wall_seconds": 156.2884025030071,
      "completed_response_resources": {
        "gpu_energy": {
          "available": true,
          "interpolation": "linear_at_bracketed_boundaries",
          "joules": 20792.72662546744,
          "reason": null,
          "response_union_seconds": 69.32143143203575,
          "scope": "GPU board only; not whole-server AC energy",
          "time_weighted_mean_watts": 299.9465850016828,
          "trapezoid_pieces": 78,
          "watt_hours": 5.775757395963177
        },
        "gpu_sample_count": 68,
        "host_sample_count": 68,
        "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": 17265410048,
            "mean": 17253726268.235294,
            "median": 17256364032.0,
            "minimum": 17243848704,
            "sample_count": 68
          },
          "gpu_board_watts": {
            "maximum": 301.23,
            "mean": 299.98058823529414,
            "median": 299.99,
            "minimum": 297.88,
            "sample_count": 68
          },
          "gpu_memory_used_bytes": {
            "maximum": 29347545088.0,
            "mean": 29347545088.0,
            "median": 29347545088.0,
            "minimum": 29347545088.0,
            "sample_count": 68
          },
          "gpu_sm_clock_mhz": {
            "maximum": 1717.0,
            "mean": 1611.6176470588234,
            "median": 1605.0,
            "minimum": 1575.0,
            "sample_count": 68
          },
          "gpu_utilization_percent": {
            "maximum": 100.0,
            "mean": 99.08823529411765,
            "median": 99.0,
            "minimum": 84.0,
            "sample_count": 68
          },
          "host_mem_available_bytes": {
            "maximum": 96128315392,
            "mean": 96101368591.05882,
            "median": 96100413440.0,
            "minimum": 96063385600,
            "sample_count": 68
          },
          "process_pss_bytes": {
            "maximum": 932377600,
            "mean": 922582874.3529412,
            "median": 924959744.0,
            "minimum": 913896448,
            "sample_count": 68
          },
          "process_rss_bytes": {
            "maximum": 935366656,
            "mean": 925592696.4705882,
            "median": 927948800.0,
            "minimum": 917929984,
            "sample_count": 68
          }
        },
        "process_sample_count": 68,
        "response_union_seconds": 69.32143143203575,
        "sampled_counter_deltas": {
          "deltas": {
            "major_faults": 0,
            "read_syscall_bytes": 0,
            "storage_read_bytes": 0,
            "storage_write_bytes": 40960
          },
          "observed_endpoint_seconds": 59.37992162600858,
          "observed_intervals": 10,
          "requested_intervals": 10
        }
      },
      "groups": [
        {
          "aggregate_output_tokens_per_second": 46.203362423998605,
          "completed_output_tokens": 1280,
          "completed_request_count": 5,
          "context_tokens": 2048,
          "cpu_snapshot_intervals": {
            "host_capacity_percent": 3.1206085574221407,
            "interval_count": 5,
            "interval_seconds": 27.718952380062547,
            "meaning": "Weighted CPU counter rate across round-bracketing snapshots, not exact token time.",
            "one_logical_cpu_percent": 99.8594738375085,
            "process_cpu_seconds": 27.68
          },
          "failed_request_count": 0,
          "measured_rounds": 5,
          "mode": "completion",
          "resources": {
            "gpu_energy": {
              "available": true,
              "interpolation": "linear_at_bracketed_boundaries",
              "joules": 8309.739476456769,
              "reason": null,
              "response_union_seconds": 27.70361144398339,
              "scope": "GPU board only; not whole-server AC energy",
              "time_weighted_mean_watts": 299.9514880310473,
              "trapezoid_pieces": 32,
              "watt_hours": 2.308260965682436
            },
            "gpu_sample_count": 27,
            "host_sample_count": 27,
            "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": 17247604736,
                "mean": 17245643510.51852,
                "median": 17245782016,
                "minimum": 17243848704,
                "sample_count": 27
              },
              "gpu_board_watts": {
                "maximum": 301.23,
                "mean": 300.0292592592593,
                "median": 299.99,
                "minimum": 299.65,
                "sample_count": 27
              },
              "gpu_memory_used_bytes": {
                "maximum": 29347545088.0,
                "mean": 29347545088.0,
                "median": 29347545088.0,
                "minimum": 29347545088.0,
                "sample_count": 27
              },
              "gpu_sm_clock_mhz": {
                "maximum": 1717.0,
                "mean": 1612.5925925925926,
                "median": 1605.0,
                "minimum": 1575.0,
                "sample_count": 27
              },
              "gpu_utilization_percent": {
                "maximum": 100.0,
                "mean": 99.18518518518519,
                "median": 99.0,
                "minimum": 99.0,
                "sample_count": 27
              },
              "host_mem_available_bytes": {
                "maximum": 96128315392,
                "mean": 96108294902.51852,
                "median": 96110452736,
                "minimum": 96087101440,
                "sample_count": 27
              },
              "process_pss_bytes": {
                "maximum": 916280320,
                "mean": 915551725.037037,
                "median": 915596288,
                "minimum": 913896448,
                "sample_count": 27
              },
              "process_rss_bytes": {
                "maximum": 919269376,
                "mean": 918593080.8888888,
                "median": 918585344,
                "minimum": 917929984,
                "sample_count": 27
              }
            },
            "process_sample_count": 27,
            "response_union_seconds": 27.70361144398339,
            "sampled_counter_deltas": {
              "deltas": {
                "major_faults": 0,
                "read_syscall_bytes": 0,
                "storage_read_bytes": 0,
                "storage_write_bytes": 20480
              },
              "observed_endpoint_seconds": 22.517533548991196,
              "observed_intervals": 5,
              "requested_intervals": 5
            }
          },
          "response_outer_span_seconds": 27.72946648998186,
          "response_seconds": {
            "maximum": 5.577194019977469,
            "mean": 5.540722288796678,
            "median": 5.540344331006054,
            "minimum": 5.50365226395661,
            "sample_count": 5
          },
          "response_window_union_seconds": 27.70361144398339,
          "server_decode_tokens_per_second": {
            "maximum": 53.78570131386867,
            "mean": 53.413691716938814,
            "median": 53.41785224622068,
            "minimum": 53.055938852926445,
            "sample_count": 5
          },
          "synthetic_first_token_seconds": {
            "maximum": 0.7708376049995422,
            "mean": 0.7664965492091141,
            "median": 0.7665820000111125,
            "minimum": 0.7625974579714239,
            "sample_count": 5
          },
          "warmup_completed_rounds": 1,
          "warmup_requests_per_round": 1
        },
        {
          "aggregate_output_tokens_per_second": 30.756055948328438,
          "completed_output_tokens": 1280,
          "completed_request_count": 5,
          "context_tokens": 8192,
          "cpu_snapshot_intervals": {
            "host_capacity_percent": 3.1199176526486507,
            "interval_count": 5,
            "interval_seconds": 41.637717549921945,
            "meaning": "Weighted CPU counter rate across round-bracketing snapshots, not exact token time.",
            "one_logical_cpu_percent": 99.83736488475682,
            "process_cpu_seconds": 41.57
          },
          "failed_request_count": 0,
          "measured_rounds": 5,
          "mode": "completion",
          "resources": {
            "gpu_energy": {
              "available": true,
              "interpolation": "linear_at_bracketed_boundaries",
              "joules": 12482.987149010663,
              "reason": null,
              "response_union_seconds": 41.61781998805236,
              "scope": "GPU board only; not whole-server AC energy",
              "time_weighted_mean_watts": 299.9433212165913,
              "trapezoid_pieces": 46,
              "watt_hours": 3.4674964302807396
            },
            "gpu_sample_count": 41,
            "host_sample_count": 41,
            "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": 17265410048,
                "mean": 17259049059.90244,
                "median": 17258135552,
                "minimum": 17254797312,
                "sample_count": 41
              },
              "gpu_board_watts": {
                "maximum": 300.29,
                "mean": 299.94853658536584,
                "median": 299.99,
                "minimum": 297.88,
                "sample_count": 41
              },
              "gpu_memory_used_bytes": {
                "maximum": 29347545088.0,
                "mean": 29347545088.0,
                "median": 29347545088.0,
                "minimum": 29347545088.0,
                "sample_count": 41
              },
              "gpu_sm_clock_mhz": {
                "maximum": 1665.0,
                "mean": 1610.9756097560976,
                "median": 1612.0,
                "minimum": 1590.0,
                "sample_count": 41
              },
              "gpu_utilization_percent": {
                "maximum": 100.0,
                "mean": 99.02439024390245,
                "median": 99.0,
                "minimum": 84.0,
                "sample_count": 41
              },
              "host_mem_available_bytes": {
                "maximum": 96122597376,
                "mean": 96096807361.56097,
                "median": 96098525184,
                "minimum": 96063385600,
                "sample_count": 41
              },
              "process_pss_bytes": {
                "maximum": 932377600,
                "mean": 927213143.4146341,
                "median": 926389248,
                "minimum": 924091392,
                "sample_count": 41
              },
              "process_rss_bytes": {
                "maximum": 935366656,
                "mean": 930202199.4146341,
                "median": 929378304,
                "minimum": 927080448,
                "sample_count": 41
              }
            },
            "process_sample_count": 41,
            "response_union_seconds": 41.61781998805236,
            "sampled_counter_deltas": {
              "deltas": {
                "major_faults": 0,
                "read_syscall_bytes": 0,
                "storage_read_bytes": 0,
                "storage_write_bytes": 20480
              },
              "observed_endpoint_seconds": 36.862388077017386,
              "observed_intervals": 5,
              "requested_intervals": 5
            }
          },
          "response_outer_span_seconds": 41.6480219400255,
          "response_seconds": {
            "maximum": 8.368119373975787,
            "mean": 8.323563997610472,
            "median": 8.326940262981225,
            "minimum": 8.274199456034694,
            "sample_count": 5
          },
          "response_window_union_seconds": 41.61781998805236,
          "server_decode_tokens_per_second": {
            "maximum": 51.768829236760986,
            "mean": 51.47410572541499,
            "median": 51.441366928244335,
            "minimum": 51.21463027254618,
            "sample_count": 5
          },
          "synthetic_first_token_seconds": {
            "maximum": 3.389060902001802,
            "mean": 3.369435553799849,
            "median": 3.36972534598317,
            "minimum": 3.3483261369983666,
            "sample_count": 5
          },
          "warmup_completed_rounds": 1,
          "warmup_requests_per_round": 1
        }
      ],
      "id": "concurrency-32b-c1",
      "model": "qwen3-32b-q4",
      "module": "concurrency",
      "placement": {
        "buffers": [
          {
            "device": "CPU_Mapped",
            "reported_mib": 417.3
          },
          {
            "device": "CUDA0",
            "reported_mib": 18423.65
          }
        ],
        "gpu_layers": 65,
        "meaning": "Runtime buffer allocation is not measured resident process RAM.",
        "requested_cpu_expert_layers": 0,
        "total_layers": 65
      },
      "record_sha256": "e7c635b222262a77e32442c195329294cbb557d68ca7789b602c3ae5aaf23e22",
      "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": 16922128384,
              "mean": 16913729536,
              "median": 16913729536.0,
              "minimum": 16905330688,
              "sample_count": 2
            },
            "gpu_board_watts": {
              "maximum": 16.67,
              "mean": 16.54,
              "median": 16.54,
              "minimum": 16.41,
              "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": 96507301888,
              "mean": 96495747072,
              "median": 96495747072.0,
              "minimum": 96484192256,
              "sample_count": 2
            },
            "process_pss_bytes": null,
            "process_rss_bytes": null
          },
          "process_sample_count": 0,
          "response_union_seconds": 1.1003568340092897,
          "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": 16910635008,
              "mean": 16910635008,
              "median": 16910635008,
              "minimum": 16910635008,
              "sample_count": 1
            },
            "gpu_board_watts": {
              "maximum": 94.7,
              "mean": 94.7,
              "median": 94.7,
              "minimum": 94.7,
              "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": 2302.0,
              "mean": 2302.0,
              "median": 2302.0,
              "minimum": 2302.0,
              "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": 96518283264,
              "mean": 96518283264,
              "median": 96518283264,
              "minimum": 96518283264,
              "sample_count": 1
            },
            "process_pss_bytes": null,
            "process_rss_bytes": null
          },
          "process_sample_count": 0,
          "response_union_seconds": 1.1621831920347176,
          "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": 229.67598352695532,
            "reason": null,
            "response_union_seconds": 3.5240373039850965,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 65.17410677441757,
            "trapezoid_pieces": 4,
            "watt_hours": 0.06379888431304315
          },
          "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": 17153597440,
              "mean": 17153525760,
              "median": 17153525760.0,
              "minimum": 17153454080,
              "sample_count": 2
            },
            "gpu_board_watts": {
              "maximum": 78.99,
              "mean": 70.92666666666666,
              "median": 78.93,
              "minimum": 54.86,
              "sample_count": 3
            },
            "gpu_memory_used_bytes": {
              "maximum": 19906166784.0,
              "mean": 19906166784.0,
              "median": 19906166784.0,
              "minimum": 19906166784.0,
              "sample_count": 3
            },
            "gpu_sm_clock_mhz": {
              "maximum": 2310.0,
              "mean": 2300.0,
              "median": 2310.0,
              "minimum": 2280.0,
              "sample_count": 3
            },
            "gpu_utilization_percent": {
              "maximum": 74.0,
              "mean": 74.0,
              "median": 74.0,
              "minimum": 74.0,
              "sample_count": 3
            },
            "host_mem_available_bytes": {
              "maximum": 96197144576,
              "mean": 96193110016,
              "median": 96193110016.0,
              "minimum": 96189075456,
              "sample_count": 2
            },
            "process_pss_bytes": {
              "maximum": 20098963456,
              "mean": 20098891776,
              "median": 20098891776.0,
              "minimum": 20098820096,
              "sample_count": 2
            },
            "process_rss_bytes": {
              "maximum": 20101951488,
              "mean": 20101879808,
              "median": 20101879808.0,
              "minimum": 20101808128,
              "sample_count": 2
            }
          },
          "process_sample_count": 2,
          "response_union_seconds": 3.5240373039850965,
          "sampled_counter_deltas": {
            "deltas": {
              "major_faults": 0,
              "read_syscall_bytes": 5,
              "storage_read_bytes": 0,
              "storage_write_bytes": 0
            },
            "observed_endpoint_seconds": 1.2853985410183668,
            "observed_intervals": 1,
            "requested_intervals": 1
          }
        },
        "ready": {
          "gpu_energy": {
            "available": true,
            "interpolation": "linear_at_bracketed_boundaries",
            "joules": 560.9414774929087,
            "reason": null,
            "response_union_seconds": 3.315524325997103,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 169.18635556209122,
            "trapezoid_pieces": 8,
            "watt_hours": 0.15581707708136353
          },
          "gpu_sample_count": 5,
          "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": 17265590272,
              "mean": 17234575360,
              "median": 17230043136.0,
              "minimum": 17212624896,
              "sample_count": 4
            },
            "gpu_board_watts": {
              "maximum": 299.4,
              "mean": 183.368,
              "median": 129.21,
              "minimum": 91.7,
              "sample_count": 5
            },
            "gpu_memory_used_bytes": {
              "maximum": 29347545088.0,
              "mean": 29342511923.2,
              "median": 29347545088.0,
              "minimum": 29322379264.0,
              "sample_count": 5
            },
            "gpu_sm_clock_mhz": {
              "maximum": 2340.0,
              "mean": 2029.4,
              "median": 2310.0,
              "minimum": 1575.0,
              "sample_count": 5
            },
            "gpu_utilization_percent": {
              "maximum": 88.0,
              "mean": 31.2,
              "median": 0.0,
              "minimum": 0.0,
              "sample_count": 5
            },
            "host_mem_available_bytes": {
              "maximum": 96123850752,
              "mean": 96111056896,
              "median": 96119119872.0,
              "minimum": 96082137088,
              "sample_count": 4
            },
            "process_pss_bytes": {
              "maximum": 932377600,
              "mean": 901519872,
              "median": 897512960.0,
              "minimum": 878675968,
              "sample_count": 4
            },
            "process_rss_bytes": {
              "maximum": 935366656,
              "mean": 904508416,
              "median": 900501504.0,
              "minimum": 881664000,
              "sample_count": 4
            }
          },
          "process_sample_count": 4,
          "response_union_seconds": 3.315524325997103,
          "sampled_counter_deltas": {
            "deltas": {
              "major_faults": 0,
              "read_syscall_bytes": 0,
              "storage_read_bytes": 0,
              "storage_write_bytes": 0
            },
            "observed_endpoint_seconds": 1.0191477150074206,
            "observed_intervals": 1,
            "requested_intervals": 3
          }
        },
        "responses": {
          "gpu_energy": {
            "available": true,
            "interpolation": "linear_at_bracketed_boundaries",
            "joules": 20813.582241929886,
            "reason": null,
            "response_union_seconds": 69.39111728599528,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 299.94591607664665,
            "trapezoid_pieces": 70,
            "watt_hours": 5.781550622758302
          },
          "gpu_sample_count": 68,
          "host_sample_count": 68,
          "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": 17265410048,
              "mean": 17253726268.235294,
              "median": 17256364032.0,
              "minimum": 17243848704,
              "sample_count": 68
            },
            "gpu_board_watts": {
              "maximum": 301.23,
              "mean": 299.98058823529414,
              "median": 299.99,
              "minimum": 297.88,
              "sample_count": 68
            },
            "gpu_memory_used_bytes": {
              "maximum": 29347545088.0,
              "mean": 29347545088.0,
              "median": 29347545088.0,
              "minimum": 29347545088.0,
              "sample_count": 68
            },
            "gpu_sm_clock_mhz": {
              "maximum": 1717.0,
              "mean": 1611.6176470588234,
              "median": 1605.0,
              "minimum": 1575.0,
              "sample_count": 68
            },
            "gpu_utilization_percent": {
              "maximum": 100.0,
              "mean": 99.08823529411765,
              "median": 99.0,
              "minimum": 84.0,
              "sample_count": 68
            },
            "host_mem_available_bytes": {
              "maximum": 96128315392,
              "mean": 96101368591.05882,
              "median": 96100413440.0,
              "minimum": 96063385600,
              "sample_count": 68
            },
            "process_pss_bytes": {
              "maximum": 932377600,
              "mean": 922582874.3529412,
              "median": 924959744.0,
              "minimum": 913896448,
              "sample_count": 68
            },
            "process_rss_bytes": {
              "maximum": 935366656,
              "mean": 925592696.4705882,
              "median": 927948800.0,
              "minimum": 917929984,
              "sample_count": 68
            }
          },
          "process_sample_count": 68,
          "response_union_seconds": 69.39111728599528,
          "sampled_counter_deltas": {
            "deltas": {
              "major_faults": 0,
              "read_syscall_bytes": 0,
              "storage_read_bytes": 0,
              "storage_write_bytes": 65536
            },
            "observed_endpoint_seconds": 67.61177103599766,
            "observed_intervals": 2,
            "requested_intervals": 2
          }
        },
        "warmup": {
          "gpu_energy": {
            "available": true,
            "interpolation": "linear_at_bracketed_boundaries",
            "joules": 3871.92128950453,
            "reason": null,
            "response_union_seconds": 13.758397663012147,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 281.4223999291531,
            "trapezoid_pieces": 16,
            "watt_hours": 1.075533691529036
          },
          "gpu_sample_count": 14,
          "host_sample_count": 13,
          "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": 17253564416,
              "mean": 17248445991.384617,
              "median": 17248645120,
              "minimum": 17243058176,
              "sample_count": 13
            },
            "gpu_board_watts": {
              "maximum": 300.79,
              "mean": 285.5842857142857,
              "median": 299.995,
              "minimum": 128.26,
              "sample_count": 14
            },
            "gpu_memory_used_bytes": {
              "maximum": 29347545088.0,
              "mean": 29347545088.0,
              "median": 29347545088.0,
              "minimum": 29347545088.0,
              "sample_count": 14
            },
            "gpu_sm_clock_mhz": {
              "maximum": 1927.0,
              "mean": 1638.5,
              "median": 1623.5,
              "minimum": 1590.0,
              "sample_count": 14
            },
            "gpu_utilization_percent": {
              "maximum": 100.0,
              "mean": 99.28571428571429,
              "median": 99.0,
              "minimum": 99.0,
              "sample_count": 14
            },
            "host_mem_available_bytes": {
              "maximum": 96119128064,
              "mean": 96098498402.46153,
              "median": 96099074048,
              "minimum": 96077873152,
              "sample_count": 13
            },
            "process_pss_bytes": {
              "maximum": 923378688,
              "mean": 918623940.9230769,
              "median": 917521408,
              "minimum": 914600960,
              "sample_count": 13
            },
            "process_rss_bytes": {
              "maximum": 926367744,
              "mean": 921612918.1538461,
              "median": 920510464,
              "minimum": 917590016,
              "sample_count": 13
            }
          },
          "process_sample_count": 13,
          "response_union_seconds": 13.758397663012147,
          "sampled_counter_deltas": {
            "deltas": {
              "major_faults": 0,
              "read_syscall_bytes": 0,
              "storage_read_bytes": 0,
              "storage_write_bytes": 12288
            },
            "observed_endpoint_seconds": 11.25419971102383,
            "observed_intervals": 2,
            "requested_intervals": 2
          }
        }
      },
      "reviewed_export_sha256": "59f09a9b924c4cb4ae1370f85109347542e72457c07f0a7a8ff351b31fedab64",
      "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": 4,
        "synthetic_prompt_sha256": "c099b5abc034050fac1133e0214990d1f72a124b64f36e8a80ba3209bc07d187",
        "temperature": 0,
        "threads": 16,
        "tokenization": "model_tokenizer_add_special_false_repeat_then_truncate",
        "total_context_tokens": 34816,
        "ubatch": 512,
        "warmup_rounds": 1
      },
      "status": "completed",
      "telemetry": {
        "gpu_samples": 93,
        "host_samples": 90,
        "maximum_gpu_gap_seconds": 1.0046468999935314,
        "maximum_host_gap_seconds": 1.2950242789811455,
        "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": 93,
            "unavailable": 0
          },
          "software": {
            "active": 0,
            "not_active": 93,
            "unavailable": 0
          }
        }
      }
    },
    {
      "case_wall_seconds": 420.7694540799712,
      "completed_response_resources": {
        "gpu_energy": {
          "available": true,
          "interpolation": "linear_at_bracketed_boundaries",
          "joules": 55752.341995404386,
          "reason": null,
          "response_union_seconds": 185.9375148040126,
          "scope": "GPU board only; not whole-server AC energy",
          "time_weighted_mean_watts": 299.84450450556,
          "trapezoid_pieces": 195,
          "watt_hours": 15.486761665390107
        },
        "gpu_sample_count": 185,
        "host_sample_count": 181,
        "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": 17361494016,
            "mean": 17327632904.486187,
            "median": 17339219968,
            "minimum": 17279602688,
            "sample_count": 181
          },
          "gpu_board_watts": {
            "maximum": 300.9,
            "mean": 299.9730810810811,
            "median": 300.0,
            "minimum": 294.85,
            "sample_count": 185
          },
          "gpu_memory_used_bytes": {
            "maximum": 29349642240.0,
            "mean": 29349642240.0,
            "median": 29349642240.0,
            "minimum": 29349642240.0,
            "sample_count": 185
          },
          "gpu_sm_clock_mhz": {
            "maximum": 1710.0,
            "mean": 1553.2648648648649,
            "median": 1537.0,
            "minimum": 1485.0,
            "sample_count": 185
          },
          "gpu_utilization_percent": {
            "maximum": 100.0,
            "mean": 99.14594594594594,
            "median": 99.0,
            "minimum": 74.0,
            "sample_count": 185
          },
          "host_mem_available_bytes": {
            "maximum": 96149176320,
            "mean": 96071103210.78453,
            "median": 96074928128,
            "minimum": 96005373952,
            "sample_count": 181
          },
          "process_pss_bytes": {
            "maximum": 997264384,
            "mean": 973108699.2265193,
            "median": 982801408,
            "minimum": 936352768,
            "sample_count": 181
          },
          "process_rss_bytes": {
            "maximum": 1000247296,
            "mean": 976107395.5359116,
            "median": 985784320,
            "minimum": 939335680,
            "sample_count": 181
          }
        },
        "process_sample_count": 181,
        "response_union_seconds": 185.9375148040126,
        "sampled_counter_deltas": {
          "deltas": {
            "major_faults": 0,
            "read_syscall_bytes": 0,
            "storage_read_bytes": 0,
            "storage_write_bytes": 188416
          },
          "observed_endpoint_seconds": 175.02184689906426,
          "observed_intervals": 10,
          "requested_intervals": 10
        }
      },
      "groups": [
        {
          "aggregate_output_tokens_per_second": 81.84464957406752,
          "completed_output_tokens": 5120,
          "completed_request_count": 20,
          "context_tokens": 2048,
          "cpu_snapshot_intervals": {
            "host_capacity_percent": 3.12788657574322,
            "interval_count": 5,
            "interval_seconds": 62.57220179203432,
            "meaning": "Weighted CPU counter rate across round-bracketing snapshots, not exact token time.",
            "one_logical_cpu_percent": 100.09237042378304,
            "process_cpu_seconds": 62.629999999999995
          },
          "failed_request_count": 0,
          "measured_rounds": 5,
          "mode": "completion",
          "resources": {
            "gpu_energy": {
              "available": true,
              "interpolation": "linear_at_bracketed_boundaries",
              "joules": 18753.3813016355,
              "reason": null,
              "response_union_seconds": 62.55754073901335,
              "scope": "GPU board only; not whole-server AC energy",
              "time_weighted_mean_watts": 299.7781095627398,
              "trapezoid_pieces": 67,
              "watt_hours": 5.209272583787639
            },
            "gpu_sample_count": 62,
            "host_sample_count": 61,
            "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": 17298706432,
                "mean": 17287832693.508198,
                "median": 17288585216,
                "minimum": 17279602688,
                "sample_count": 61
              },
              "gpu_board_watts": {
                "maximum": 300.68,
                "mean": 299.9858064516129,
                "median": 300.01,
                "minimum": 299.23,
                "sample_count": 62
              },
              "gpu_memory_used_bytes": {
                "maximum": 29349642240.0,
                "mean": 29349642240.0,
                "median": 29349642240.0,
                "minimum": 29349642240.0,
                "sample_count": 62
              },
              "gpu_sm_clock_mhz": {
                "maximum": 1665.0,
                "mean": 1549.6935483870968,
                "median": 1530.0,
                "minimum": 1485.0,
                "sample_count": 62
              },
              "gpu_utilization_percent": {
                "maximum": 100.0,
                "mean": 99.01612903225806,
                "median": 99.0,
                "minimum": 95.0,
                "sample_count": 62
              },
              "host_mem_available_bytes": {
                "maximum": 96080334848,
                "mean": 96040837120,
                "median": 96046120960,
                "minimum": 96005373952,
                "sample_count": 61
              },
              "process_pss_bytes": {
                "maximum": 948587520,
                "mean": 941003255.6065574,
                "median": 942095360,
                "minimum": 936352768,
                "sample_count": 61
              },
              "process_rss_bytes": {
                "maximum": 951570432,
                "mean": 944009417.4426229,
                "median": 945078272,
                "minimum": 939335680,
                "sample_count": 61
              }
            },
            "process_sample_count": 61,
            "response_union_seconds": 62.55754073901335,
            "sampled_counter_deltas": {
              "deltas": {
                "major_faults": 0,
                "read_syscall_bytes": 0,
                "storage_read_bytes": 0,
                "storage_write_bytes": 65536
              },
              "observed_endpoint_seconds": 57.27966735709924,
              "observed_intervals": 5,
              "requested_intervals": 5
            }
          },
          "response_outer_span_seconds": 62.60230736201629,
          "response_seconds": {
            "maximum": 12.592243108025286,
            "mean": 12.467036286104122,
            "median": 12.479891907511046,
            "minimum": 12.302360792993568,
            "sample_count": 20
          },
          "response_window_union_seconds": 62.55754073901335,
          "server_decode_tokens_per_second": {
            "maximum": 28.00551324221474,
            "mean": 25.58471676770163,
            "median": 26.37554127135465,
            "minimum": 21.76054778552292,
            "sample_count": 20
          },
          "synthetic_first_token_seconds": {
            "maximum": 3.3099877709755674,
            "mean": 2.412118052196456,
            "median": 2.786846547969617,
            "minimum": 0.7595875930273905,
            "sample_count": 20
          },
          "warmup_completed_rounds": 1,
          "warmup_requests_per_round": 4
        },
        {
          "aggregate_output_tokens_per_second": 41.49782036185769,
          "completed_output_tokens": 5120,
          "completed_request_count": 20,
          "context_tokens": 8192,
          "cpu_snapshot_intervals": {
            "host_capacity_percent": 3.125263134343059,
            "interval_count": 5,
            "interval_seconds": 123.39960938395234,
            "meaning": "Weighted CPU counter rate across round-bracketing snapshots, not exact token time.",
            "one_logical_cpu_percent": 100.00842029897788,
            "process_cpu_seconds": 123.41
          },
          "failed_request_count": 0,
          "measured_rounds": 5,
          "mode": "completion",
          "resources": {
            "gpu_energy": {
              "available": true,
              "interpolation": "linear_at_bracketed_boundaries",
              "joules": 36998.96069376888,
              "reason": null,
              "response_union_seconds": 123.37997406499926,
              "scope": "GPU board only; not whole-server AC energy",
              "time_weighted_mean_watts": 299.8781688370028,
              "trapezoid_pieces": 128,
              "watt_hours": 10.277489081602466
            },
            "gpu_sample_count": 123,
            "host_sample_count": 120,
            "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": 17361494016,
                "mean": 17347864678.4,
                "median": 17350475776.0,
                "minimum": 17321365504,
                "sample_count": 120
              },
              "gpu_board_watts": {
                "maximum": 300.9,
                "mean": 299.96666666666664,
                "median": 299.99,
                "minimum": 294.85,
                "sample_count": 123
              },
              "gpu_memory_used_bytes": {
                "maximum": 29349642240.0,
                "mean": 29349642240.0,
                "median": 29349642240.0,
                "minimum": 29349642240.0,
                "sample_count": 123
              },
              "gpu_sm_clock_mhz": {
                "maximum": 1710.0,
                "mean": 1555.0650406504064,
                "median": 1567.0,
                "minimum": 1500.0,
                "sample_count": 123
              },
              "gpu_utilization_percent": {
                "maximum": 100.0,
                "mean": 99.21138211382114,
                "median": 100.0,
                "minimum": 74.0,
                "sample_count": 123
              },
              "host_mem_available_bytes": {
                "maximum": 96149176320,
                "mean": 96086488473.6,
                "median": 96085280768.0,
                "minimum": 96021024768,
                "sample_count": 120
              },
              "process_pss_bytes": {
                "maximum": 997264384,
                "mean": 989428966.4,
                "median": 992402432.0,
                "minimum": 968424448,
                "sample_count": 120
              },
              "process_rss_bytes": {
                "maximum": 1000247296,
                "mean": 992423867.7333333,
                "median": 995385344.0,
                "minimum": 971407360,
                "sample_count": 120
              }
            },
            "process_sample_count": 120,
            "response_union_seconds": 123.37997406499926,
            "sampled_counter_deltas": {
              "deltas": {
                "major_faults": 0,
                "read_syscall_bytes": 0,
                "storage_read_bytes": 0,
                "storage_write_bytes": 122880
              },
              "observed_endpoint_seconds": 117.74217954196502,
              "observed_intervals": 5,
              "requested_intervals": 5
            }
          },
          "response_outer_span_seconds": 123.44470494805137,
          "response_seconds": {
            "maximum": 24.823961583024357,
            "mean": 24.52178838660766,
            "median": 24.525591763027478,
            "minimum": 24.108100534998812,
            "sample_count": 20
          },
          "response_window_union_seconds": 123.37997406499926,
          "server_decode_tokens_per_second": {
            "maximum": 24.761002915778022,
            "mean": 18.143512774672224,
            "median": 17.788452280236555,
            "minimum": 12.163616381481537,
            "sample_count": 20
          },
          "synthetic_first_token_seconds": {
            "maximum": 14.377657943987288,
            "mean": 9.506834106854512,
            "median": 10.070695419010008,
            "minimum": 3.423168073000852,
            "sample_count": 20
          },
          "warmup_completed_rounds": 1,
          "warmup_requests_per_round": 4
        }
      ],
      "id": "concurrency-32b-c4",
      "model": "qwen3-32b-q4",
      "module": "concurrency",
      "placement": {
        "buffers": [
          {
            "device": "CPU_Mapped",
            "reported_mib": 417.3
          },
          {
            "device": "CUDA0",
            "reported_mib": 18423.65
          }
        ],
        "gpu_layers": 65,
        "meaning": "Runtime buffer allocation is not measured resident process RAM.",
        "requested_cpu_expert_layers": 0,
        "total_layers": 65
      },
      "record_sha256": "806238b5092d95f97811d67523489974a389b64477b4f6fbfabd0f3de1d3e257",
      "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": 16936005632,
              "mean": 16927412224,
              "median": 16927412224.0,
              "minimum": 16918818816,
              "sample_count": 2
            },
            "gpu_board_watts": {
              "maximum": 15.99,
              "mean": 15.905000000000001,
              "median": 15.905000000000001,
              "minimum": 15.82,
              "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": 96495370240,
              "mean": 96480976896,
              "median": 96480976896.0,
              "minimum": 96466583552,
              "sample_count": 2
            },
            "process_pss_bytes": null,
            "process_rss_bytes": null
          },
          "process_sample_count": 0,
          "response_union_seconds": 1.1003576699877158,
          "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": 16943104000,
              "mean": 16943104000,
              "median": 16943104000,
              "minimum": 16943104000,
              "sample_count": 1
            },
            "gpu_board_watts": {
              "maximum": 105.94,
              "mean": 105.94,
              "median": 105.94,
              "minimum": 105.94,
              "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": 2340.0,
              "mean": 2340.0,
              "median": 2340.0,
              "minimum": 2340.0,
              "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": 96572526592,
              "mean": 96572526592,
              "median": 96572526592,
              "minimum": 96572526592,
              "sample_count": 1
            },
            "process_pss_bytes": null,
            "process_rss_bytes": null
          },
          "process_sample_count": 0,
          "response_union_seconds": 1.1639506969950162,
          "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": 207.79273563988033,
            "reason": null,
            "response_union_seconds": 3.5026424899697304,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 59.324563164788216,
            "trapezoid_pieces": 4,
            "watt_hours": 0.057720204344411205
          },
          "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": 17226100736,
              "mean": 17186957994.666668,
              "median": 17167605760,
              "minimum": 17167167488,
              "sample_count": 3
            },
            "gpu_board_watts": {
              "maximum": 78.3,
              "mean": 64.33666666666667,
              "median": 78.19,
              "minimum": 36.52,
              "sample_count": 3
            },
            "gpu_memory_used_bytes": {
              "maximum": 19906166784.0,
              "mean": 19906166784.0,
              "median": 19906166784.0,
              "minimum": 19906166784.0,
              "sample_count": 3
            },
            "gpu_sm_clock_mhz": {
              "maximum": 2310.0,
              "mean": 2300.0,
              "median": 2310.0,
              "minimum": 2280.0,
              "sample_count": 3
            },
            "gpu_utilization_percent": {
              "maximum": 78.0,
              "mean": 78.0,
              "median": 78.0,
              "minimum": 78.0,
              "sample_count": 3
            },
            "host_mem_available_bytes": {
              "maximum": 96178794496,
              "mean": 96163105450.66667,
              "median": 96178565120,
              "minimum": 96131956736,
              "sample_count": 3
            },
            "process_pss_bytes": {
              "maximum": 20098980864,
              "mean": 13692147370.666666,
              "median": 20098866176,
              "minimum": 878595072,
              "sample_count": 3
            },
            "process_rss_bytes": {
              "maximum": 20101963776,
              "mean": 13695130282.666666,
              "median": 20101849088,
              "minimum": 881577984,
              "sample_count": 3
            }
          },
          "process_sample_count": 3,
          "response_union_seconds": 3.5026424899697304,
          "sampled_counter_deltas": {
            "deltas": {
              "major_faults": 0,
              "read_syscall_bytes": 13,
              "storage_read_bytes": 0,
              "storage_write_bytes": 4096
            },
            "observed_endpoint_seconds": 2.2705488849896938,
            "observed_intervals": 1,
            "requested_intervals": 1
          }
        },
        "ready": {
          "gpu_energy": {
            "available": true,
            "interpolation": "linear_at_bracketed_boundaries",
            "joules": 575.8172122131975,
            "reason": null,
            "response_union_seconds": 3.3165422920137644,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 173.6197405351247,
            "trapezoid_pieces": 6,
            "watt_hours": 0.1599492256147771
          },
          "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": 17363476480,
              "mean": 17296278869.333332,
              "median": 17299456000,
              "minimum": 17225904128,
              "sample_count": 3
            },
            "gpu_board_watts": {
              "maximum": 241.87,
              "mean": 186.64666666666668,
              "median": 227.52,
              "minimum": 90.55,
              "sample_count": 3
            },
            "gpu_memory_used_bytes": {
              "maximum": 29349642240.0,
              "mean": 29340554581.333332,
              "median": 29349642240.0,
              "minimum": 29322379264.0,
              "sample_count": 3
            },
            "gpu_sm_clock_mhz": {
              "maximum": 2340.0,
              "mean": 2327.3333333333335,
              "median": 2332.0,
              "minimum": 2310.0,
              "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": 96134615040,
              "mean": 96089931776,
              "median": 96077000704,
              "minimum": 96058179584,
              "sample_count": 3
            },
            "process_pss_bytes": {
              "maximum": 997264384,
              "mean": 941612032,
              "median": 948870144,
              "minimum": 878701568,
              "sample_count": 3
            },
            "process_rss_bytes": {
              "maximum": 1000247296,
              "mean": 944594944,
              "median": 951853056,
              "minimum": 881684480,
              "sample_count": 3
            }
          },
          "process_sample_count": 3,
          "response_union_seconds": 3.3165422920137644,
          "sampled_counter_deltas": {
            "deltas": null,
            "observed_endpoint_seconds": 0.0,
            "observed_intervals": 0,
            "requested_intervals": 3
          }
        },
        "responses": {
          "gpu_energy": {
            "available": true,
            "interpolation": "linear_at_bracketed_boundaries",
            "joules": 55792.14584703776,
            "reason": null,
            "response_union_seconds": 186.07392277405597,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 299.8386072334515,
            "trapezoid_pieces": 187,
            "watt_hours": 15.497818290843822
          },
          "gpu_sample_count": 185,
          "host_sample_count": 181,
          "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": 17361494016,
              "mean": 17327632904.486187,
              "median": 17339219968,
              "minimum": 17279602688,
              "sample_count": 181
            },
            "gpu_board_watts": {
              "maximum": 300.9,
              "mean": 299.9730810810811,
              "median": 300.0,
              "minimum": 294.85,
              "sample_count": 185
            },
            "gpu_memory_used_bytes": {
              "maximum": 29349642240.0,
              "mean": 29349642240.0,
              "median": 29349642240.0,
              "minimum": 29349642240.0,
              "sample_count": 185
            },
            "gpu_sm_clock_mhz": {
              "maximum": 1710.0,
              "mean": 1553.2648648648649,
              "median": 1537.0,
              "minimum": 1485.0,
              "sample_count": 185
            },
            "gpu_utilization_percent": {
              "maximum": 100.0,
              "mean": 99.14594594594594,
              "median": 99.0,
              "minimum": 74.0,
              "sample_count": 185
            },
            "host_mem_available_bytes": {
              "maximum": 96149176320,
              "mean": 96071103210.78453,
              "median": 96074928128,
              "minimum": 96005373952,
              "sample_count": 181
            },
            "process_pss_bytes": {
              "maximum": 997264384,
              "mean": 973108699.2265193,
              "median": 982801408,
              "minimum": 936352768,
              "sample_count": 181
            },
            "process_rss_bytes": {
              "maximum": 1000247296,
              "mean": 976107395.5359116,
              "median": 985784320,
              "minimum": 939335680,
              "sample_count": 181
            }
          },
          "process_sample_count": 181,
          "response_union_seconds": 186.07392277405597,
          "sampled_counter_deltas": {
            "deltas": {
              "major_faults": 0,
              "read_syscall_bytes": 0,
              "storage_read_bytes": 0,
              "storage_write_bytes": 229376
            },
            "observed_endpoint_seconds": 183.24035443703178,
            "observed_intervals": 2,
            "requested_intervals": 2
          }
        },
        "warmup": {
          "gpu_energy": {
            "available": true,
            "interpolation": "linear_at_bracketed_boundaries",
            "joules": 10747.853438441543,
            "reason": null,
            "response_union_seconds": 36.686159758944996,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 292.96752533006537,
            "trapezoid_pieces": 40,
            "watt_hours": 2.98551484401154
          },
          "gpu_sample_count": 38,
          "host_sample_count": 36,
          "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": 17317265408,
              "mean": 17296877340.444443,
              "median": 17303889920.0,
              "minimum": 17255440384,
              "sample_count": 36
            },
            "gpu_board_watts": {
              "maximum": 300.91,
              "mean": 291.04894736842107,
              "median": 299.98,
              "minimum": 127.66,
              "sample_count": 38
            },
            "gpu_memory_used_bytes": {
              "maximum": 29349642240.0,
              "mean": 29349476675.36842,
              "median": 29349642240.0,
              "minimum": 29347545088.0,
              "sample_count": 38
            },
            "gpu_sm_clock_mhz": {
              "maximum": 1965.0,
              "mean": 1588.0263157894738,
              "median": 1586.0,
              "minimum": 1515.0,
              "sample_count": 38
            },
            "gpu_utilization_percent": {
              "maximum": 100.0,
              "mean": 98.92105263157895,
              "median": 99.0,
              "minimum": 94.0,
              "sample_count": 38
            },
            "host_mem_available_bytes": {
              "maximum": 96110641152,
              "mean": 96048308565.33333,
              "median": 96044918784.0,
              "minimum": 96014688256,
              "sample_count": 36
            },
            "process_pss_bytes": {
              "maximum": 965807104,
              "mean": 948997688.8888888,
              "median": 953592832.0,
              "minimum": 912342016,
              "sample_count": 36
            },
            "process_rss_bytes": {
              "maximum": 968790016,
              "mean": 951980600.8888888,
              "median": 956575744.0,
              "minimum": 915324928,
              "sample_count": 36
            }
          },
          "process_sample_count": 36,
          "response_union_seconds": 36.686159758944996,
          "sampled_counter_deltas": {
            "deltas": {
              "major_faults": 6,
              "read_syscall_bytes": 0,
              "storage_read_bytes": 1179648,
              "storage_write_bytes": 40960
            },
            "observed_endpoint_seconds": 34.84366085496731,
            "observed_intervals": 2,
            "requested_intervals": 2
          }
        }
      },
      "reviewed_export_sha256": "59f09a9b924c4cb4ae1370f85109347542e72457c07f0a7a8ff351b31fedab64",
      "settings": {
        "batch": 2048,
        "cache_prompt": false,
        "cache_ram_bytes": 0,
        "concurrency": 4,
        "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": 4,
        "synthetic_prompt_sha256": "c099b5abc034050fac1133e0214990d1f72a124b64f36e8a80ba3209bc07d187",
        "temperature": 0,
        "threads": 16,
        "tokenization": "model_tokenizer_add_special_false_repeat_then_truncate",
        "total_context_tokens": 34816,
        "ubatch": 512,
        "warmup_rounds": 1
      },
      "status": "completed",
      "telemetry": {
        "gpu_samples": 232,
        "host_samples": 226,
        "maximum_gpu_gap_seconds": 1.0094260819605552,
        "maximum_host_gap_seconds": 1.306493810028769,
        "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": 232,
            "unavailable": 0
          },
          "software": {
            "active": 0,
            "not_active": 232,
            "unavailable": 0
          }
        }
      }
    },
    {
      "case_wall_seconds": 285.04197112697875,
      "completed_response_resources": {
        "gpu_energy": {
          "available": true,
          "interpolation": "linear_at_bracketed_boundaries",
          "joules": 43583.755059872135,
          "reason": null,
          "response_union_seconds": 145.35013979504583,
          "scope": "GPU board only; not whole-server AC energy",
          "time_weighted_mean_watts": 299.8535475874214,
          "trapezoid_pieces": 155,
          "watt_hours": 12.10659862774226
        },
        "gpu_sample_count": 145,
        "host_sample_count": 142,
        "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": 17379631104,
            "mean": 17357138727.661972,
            "median": 17368823808.0,
            "minimum": 17327955968,
            "sample_count": 142
          },
          "gpu_board_watts": {
            "maximum": 303.17,
            "mean": 299.96124137931037,
            "median": 299.99,
            "minimum": 296.54,
            "sample_count": 145
          },
          "gpu_memory_used_bytes": {
            "maximum": 55660511232.0,
            "mean": 55660511232.0,
            "median": 55660511232.0,
            "minimum": 55660511232.0,
            "sample_count": 145
          },
          "gpu_sm_clock_mhz": {
            "maximum": 1627.0,
            "mean": 1478.8689655172413,
            "median": 1485.0,
            "minimum": 1305.0,
            "sample_count": 145
          },
          "gpu_utilization_percent": {
            "maximum": 100.0,
            "mean": 99.64137931034483,
            "median": 100.0,
            "minimum": 83.0,
            "sample_count": 145
          },
          "host_mem_available_bytes": {
            "maximum": 96155914240,
            "mean": 96071287735.88733,
            "median": 96036245504.0,
            "minimum": 96016179200,
            "sample_count": 142
          },
          "process_pss_bytes": {
            "maximum": 1253164032,
            "mean": 1233515772.3943663,
            "median": 1245466624.0,
            "minimum": 1207312384,
            "sample_count": 142
          },
          "process_rss_bytes": {
            "maximum": 1256132608,
            "mean": 1236515392.9014084,
            "median": 1248436224.0,
            "minimum": 1210281984,
            "sample_count": 142
          }
        },
        "process_sample_count": 142,
        "response_union_seconds": 145.35013979504583,
        "sampled_counter_deltas": {
          "deltas": {
            "major_faults": 0,
            "read_syscall_bytes": 0,
            "storage_read_bytes": 0,
            "storage_write_bytes": 102400
          },
          "observed_endpoint_seconds": 135.45640823506983,
          "observed_intervals": 10,
          "requested_intervals": 10
        }
      },
      "groups": [
        {
          "aggregate_output_tokens_per_second": 22.028412284302284,
          "completed_output_tokens": 1280,
          "completed_request_count": 5,
          "context_tokens": 2048,
          "cpu_snapshot_intervals": {
            "host_capacity_percent": 3.1210980722289743,
            "interval_count": 5,
            "interval_seconds": 58.12257282593055,
            "meaning": "Weighted CPU counter rate across round-bracketing snapshots, not exact token time.",
            "one_logical_cpu_percent": 99.87513831132718,
            "process_cpu_seconds": 58.05
          },
          "failed_request_count": 0,
          "measured_rounds": 5,
          "mode": "completion",
          "resources": {
            "gpu_energy": {
              "available": true,
              "interpolation": "linear_at_bracketed_boundaries",
              "joules": 17424.8135125459,
              "reason": null,
              "response_union_seconds": 58.10677517199656,
              "scope": "GPU board only; not whole-server AC energy",
              "time_weighted_mean_watts": 299.8757625245645,
              "trapezoid_pieces": 63,
              "watt_hours": 4.840225975707194
            },
            "gpu_sample_count": 58,
            "host_sample_count": 57,
            "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": 17349152768,
                "mean": 17331692166.736843,
                "median": 17331625984,
                "minimum": 17327955968,
                "sample_count": 57
              },
              "gpu_board_watts": {
                "maximum": 300.28,
                "mean": 299.96224137931034,
                "median": 299.99,
                "minimum": 298.49,
                "sample_count": 58
              },
              "gpu_memory_used_bytes": {
                "maximum": 55660511232.0,
                "mean": 55660511232.0,
                "median": 55660511232.0,
                "minimum": 55660511232.0,
                "sample_count": 58
              },
              "gpu_sm_clock_mhz": {
                "maximum": 1627.0,
                "mean": 1451.5172413793102,
                "median": 1485.0,
                "minimum": 1305.0,
                "sample_count": 58
              },
              "gpu_utilization_percent": {
                "maximum": 100.0,
                "mean": 99.70689655172414,
                "median": 100.0,
                "minimum": 86.0,
                "sample_count": 58
              },
              "host_mem_available_bytes": {
                "maximum": 96047636480,
                "mean": 96030269511.85965,
                "median": 96029605888,
                "minimum": 96017846272,
                "sample_count": 57
              },
              "process_pss_bytes": {
                "maximum": 1218039808,
                "mean": 1209269230.0350878,
                "median": 1209229312,
                "minimum": 1207312384,
                "sample_count": 57
              },
              "process_rss_bytes": {
                "maximum": 1221009408,
                "mean": 1212263729.4035087,
                "median": 1212198912,
                "minimum": 1210281984,
                "sample_count": 57
              }
            },
            "process_sample_count": 57,
            "response_union_seconds": 58.10677517199656,
            "sampled_counter_deltas": {
              "deltas": {
                "major_faults": 0,
                "read_syscall_bytes": 0,
                "storage_read_bytes": 0,
                "storage_write_bytes": 36864
              },
              "observed_endpoint_seconds": 53.336369540018495,
              "observed_intervals": 5,
              "requested_intervals": 5
            }
          },
          "response_outer_span_seconds": 58.13016001897631,
          "response_seconds": {
            "maximum": 11.71224289300153,
            "mean": 11.621355034399311,
            "median": 11.628675386018585,
            "minimum": 11.497684480971657,
            "sample_count": 5
          },
          "response_window_union_seconds": 58.10677517199656,
          "server_decode_tokens_per_second": {
            "maximum": 25.931278654062915,
            "mean": 25.626653456527215,
            "median": 25.57785387161442,
            "minimum": 25.41033708456341,
            "sample_count": 5
          },
          "synthetic_first_token_seconds": {
            "maximum": 1.6838101929752156,
            "mean": 1.670307478599716,
            "median": 1.6676911130198278,
            "minimum": 1.6591309130308218,
            "sample_count": 5
          },
          "warmup_completed_rounds": 1,
          "warmup_requests_per_round": 1
        },
        {
          "aggregate_output_tokens_per_second": 14.671602883846484,
          "completed_output_tokens": 1280,
          "completed_request_count": 5,
          "context_tokens": 8192,
          "cpu_snapshot_intervals": {
            "host_capacity_percent": 3.120939686796973,
            "interval_count": 5,
            "interval_seconds": 87.26338133099489,
            "meaning": "Weighted CPU counter rate across round-bracketing snapshots, not exact token time.",
            "one_logical_cpu_percent": 99.87006997750314,
            "process_cpu_seconds": 87.15
          },
          "failed_request_count": 0,
          "measured_rounds": 5,
          "mode": "completion",
          "resources": {
            "gpu_energy": {
              "available": true,
              "interpolation": "linear_at_bracketed_boundaries",
              "joules": 26158.941547326252,
              "reason": null,
              "response_union_seconds": 87.24336462304927,
              "scope": "GPU board only; not whole-server AC energy",
              "time_weighted_mean_watts": 299.83875175322146,
              "trapezoid_pieces": 92,
              "watt_hours": 7.26637265203507
            },
            "gpu_sample_count": 87,
            "host_sample_count": 85,
            "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": 17379631104,
                "mean": 17374202892.04706,
                "median": 17376063488,
                "minimum": 17363415040,
                "sample_count": 85
              },
              "gpu_board_watts": {
                "maximum": 303.17,
                "mean": 299.9605747126437,
                "median": 300.0,
                "minimum": 296.54,
                "sample_count": 87
              },
              "gpu_memory_used_bytes": {
                "maximum": 55660511232.0,
                "mean": 55660511232.0,
                "median": 55660511232.0,
                "minimum": 55660511232.0,
                "sample_count": 87
              },
              "gpu_sm_clock_mhz": {
                "maximum": 1545.0,
                "mean": 1497.103448275862,
                "median": 1477.0,
                "minimum": 1462.0,
                "sample_count": 87
              },
              "gpu_utilization_percent": {
                "maximum": 100.0,
                "mean": 99.59770114942529,
                "median": 100.0,
                "minimum": 83.0,
                "sample_count": 87
              },
              "host_mem_available_bytes": {
                "maximum": 96155914240,
                "mean": 96098794074.35294,
                "median": 96110546944,
                "minimum": 96016179200,
                "sample_count": 85
              },
              "process_pss_bytes": {
                "maximum": 1253164032,
                "mean": 1249775218.447059,
                "median": 1251803136,
                "minimum": 1240936448,
                "sample_count": 85
              },
              "process_rss_bytes": {
                "maximum": 1256132608,
                "mean": 1252778273.1294117,
                "median": 1254772736,
                "minimum": 1243906048,
                "sample_count": 85
              }
            },
            "process_sample_count": 85,
            "response_union_seconds": 87.24336462304927,
            "sampled_counter_deltas": {
              "deltas": {
                "major_faults": 0,
                "read_syscall_bytes": 0,
                "storage_read_bytes": 0,
                "storage_write_bytes": 65536
              },
              "observed_endpoint_seconds": 82.12003869505133,
              "observed_intervals": 5,
              "requested_intervals": 5
            }
          },
          "response_outer_span_seconds": 87.273929255025,
          "response_seconds": {
            "maximum": 17.565991370007396,
            "mean": 17.448672924609856,
            "median": 17.456415257009212,
            "minimum": 17.328929729992524,
            "sample_count": 5
          },
          "response_window_union_seconds": 87.24336462304927,
          "server_decode_tokens_per_second": {
            "maximum": 24.89443782111384,
            "mean": 24.72205583310525,
            "median": 24.709477111611353,
            "minimum": 24.551998243810004,
            "sample_count": 5
          },
          "synthetic_first_token_seconds": {
            "maximum": 7.1797569649643265,
            "mean": 7.133594566385727,
            "median": 7.136343034973834,
            "minimum": 7.085526316019241,
            "sample_count": 5
          },
          "warmup_completed_rounds": 1,
          "warmup_requests_per_round": 1
        }
      ],
      "id": "concurrency-72b-q4-c1",
      "model": "qwen25-72b-q4",
      "module": "concurrency",
      "placement": {
        "buffers": [
          {
            "device": "CPU_Mapped",
            "reported_mib": 668.25
          },
          {
            "device": "CUDA0",
            "reported_mib": 41297.69
          }
        ],
        "gpu_layers": 81,
        "meaning": "Runtime buffer allocation is not measured resident process RAM.",
        "requested_cpu_expert_layers": 0,
        "total_layers": 81
      },
      "record_sha256": "c9243a2bc147bb4e2eacacfe6db0e0f45bfb9e1ff382df0e73f4001e22529e1d",
      "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": 16978206720,
              "mean": 16969136128,
              "median": 16969136128.0,
              "minimum": 16960065536,
              "sample_count": 2
            },
            "gpu_board_watts": {
              "maximum": 16.06,
              "mean": 15.985,
              "median": 15.985,
              "minimum": 15.91,
              "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": 96520728576,
              "mean": 96520091648,
              "median": 96520091648.0,
              "minimum": 96519454720,
              "sample_count": 2
            },
            "process_pss_bytes": null,
            "process_rss_bytes": null
          },
          "process_sample_count": 0,
          "response_union_seconds": 1.1008190379943699,
          "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": 16965869568,
              "mean": 16965869568,
              "median": 16965869568,
              "minimum": 16965869568,
              "sample_count": 1
            },
            "gpu_board_watts": {
              "maximum": 106.5,
              "mean": 106.5,
              "median": 106.5,
              "minimum": 106.5,
              "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": 2340.0,
              "mean": 2340.0,
              "median": 2340.0,
              "minimum": 2340.0,
              "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": 96598876160,
              "mean": 96598876160,
              "median": 96598876160,
              "minimum": 96598876160,
              "sample_count": 1
            },
            "process_pss_bytes": null,
            "process_rss_bytes": null
          },
          "process_sample_count": 0,
          "response_union_seconds": 1.159519933979027,
          "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": 500.40460333691567,
            "reason": null,
            "response_union_seconds": 7.142398557043634,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 70.06114253361437,
            "trapezoid_pieces": 8,
            "watt_hours": 0.1390012787046988
          },
          "gpu_sample_count": 7,
          "host_sample_count": 5,
          "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": 17257222144,
              "mean": 17253466931.2,
              "median": 17256321024,
              "minimum": 17242816512,
              "sample_count": 5
            },
            "gpu_board_watts": {
              "maximum": 79.11,
              "mean": 73.61285714285714,
              "median": 78.44,
              "minimum": 46.09,
              "sample_count": 7
            },
            "gpu_memory_used_bytes": {
              "maximum": 55622762496.0,
              "mean": 39380919442.28571,
              "median": 43891294208.0,
              "minimum": 587202560.0,
              "sample_count": 7
            },
            "gpu_sm_clock_mhz": {
              "maximum": 2310.0,
              "mean": 2299.0,
              "median": 2302.0,
              "minimum": 2272.0,
              "sample_count": 7
            },
            "gpu_utilization_percent": {
              "maximum": 78.0,
              "mean": 57.57142857142857,
              "median": 76.0,
              "minimum": 0.0,
              "sample_count": 7
            },
            "host_mem_available_bytes": {
              "maximum": 96174260224,
              "mean": 96142704640,
              "median": 96146825216,
              "minimum": 96110252032,
              "sample_count": 5
            },
            "process_pss_bytes": {
              "maximum": 44347047936,
              "mean": 34410967040,
              "median": 44346761216,
              "minimum": 1126739968,
              "sample_count": 5
            },
            "process_rss_bytes": {
              "maximum": 44350017536,
              "mean": 34413936640,
              "median": 44349730816,
              "minimum": 1129709568,
              "sample_count": 5
            }
          },
          "process_sample_count": 5,
          "response_union_seconds": 7.142398557043634,
          "sampled_counter_deltas": {
            "deltas": {
              "major_faults": 0,
              "read_syscall_bytes": 13,
              "storage_read_bytes": 0,
              "storage_write_bytes": 8192
            },
            "observed_endpoint_seconds": 5.804069469973911,
            "observed_intervals": 1,
            "requested_intervals": 1
          }
        },
        "ready": {
          "gpu_energy": {
            "available": true,
            "interpolation": "linear_at_bracketed_boundaries",
            "joules": 570.0591014654404,
            "reason": null,
            "response_union_seconds": 3.319998638005927,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 171.7046190741307,
            "trapezoid_pieces": 6,
            "watt_hours": 0.15834975040706678
          },
          "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": 17379704832,
              "mean": 17336277674.666668,
              "median": 17340379136,
              "minimum": 17288749056,
              "sample_count": 3
            },
            "gpu_board_watts": {
              "maximum": 286.45,
              "mean": 199.11666666666667,
              "median": 209.48,
              "minimum": 101.42,
              "sample_count": 3
            },
            "gpu_memory_used_bytes": {
              "maximum": 55660511232.0,
              "mean": 55650025472.0,
              "median": 55660511232.0,
              "minimum": 55629053952.0,
              "sample_count": 3
            },
            "gpu_sm_clock_mhz": {
              "maximum": 2340.0,
              "mean": 2127.3333333333335,
              "median": 2310.0,
              "minimum": 1732.0,
              "sample_count": 3
            },
            "gpu_utilization_percent": {
              "maximum": 20.0,
              "mean": 6.666666666666667,
              "median": 0.0,
              "minimum": 0.0,
              "sample_count": 3
            },
            "host_mem_available_bytes": {
              "maximum": 96096833536,
              "mean": 96065088170.66667,
              "median": 96067203072,
              "minimum": 96031227904,
              "sample_count": 3
            },
            "process_pss_bytes": {
              "maximum": 1253164032,
              "mean": 1210951338.6666667,
              "median": 1218109440,
              "minimum": 1161580544,
              "sample_count": 3
            },
            "process_rss_bytes": {
              "maximum": 1256132608,
              "mean": 1213920597.3333333,
              "median": 1221079040,
              "minimum": 1164550144,
              "sample_count": 3
            }
          },
          "process_sample_count": 3,
          "response_union_seconds": 3.319998638005927,
          "sampled_counter_deltas": {
            "deltas": null,
            "observed_endpoint_seconds": 0.0,
            "observed_intervals": 0,
            "requested_intervals": 3
          }
        },
        "responses": {
          "gpu_energy": {
            "available": true,
            "interpolation": "linear_at_bracketed_boundaries",
            "joules": 43603.333030460024,
            "reason": null,
            "response_union_seconds": 145.41683486895636,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 299.85065394769146,
            "trapezoid_pieces": 147,
            "watt_hours": 12.112036952905562
          },
          "gpu_sample_count": 145,
          "host_sample_count": 142,
          "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": 17379631104,
              "mean": 17357138727.661972,
              "median": 17368823808.0,
              "minimum": 17327955968,
              "sample_count": 142
            },
            "gpu_board_watts": {
              "maximum": 303.17,
              "mean": 299.96124137931037,
              "median": 299.99,
              "minimum": 296.54,
              "sample_count": 145
            },
            "gpu_memory_used_bytes": {
              "maximum": 55660511232.0,
              "mean": 55660511232.0,
              "median": 55660511232.0,
              "minimum": 55660511232.0,
              "sample_count": 145
            },
            "gpu_sm_clock_mhz": {
              "maximum": 1627.0,
              "mean": 1478.8689655172413,
              "median": 1485.0,
              "minimum": 1305.0,
              "sample_count": 145
            },
            "gpu_utilization_percent": {
              "maximum": 100.0,
              "mean": 99.64137931034483,
              "median": 100.0,
              "minimum": 83.0,
              "sample_count": 145
            },
            "host_mem_available_bytes": {
              "maximum": 96155914240,
              "mean": 96071287735.88733,
              "median": 96036245504.0,
              "minimum": 96016179200,
              "sample_count": 142
            },
            "process_pss_bytes": {
              "maximum": 1253164032,
              "mean": 1233515772.3943663,
              "median": 1245466624.0,
              "minimum": 1207312384,
              "sample_count": 142
            },
            "process_rss_bytes": {
              "maximum": 1256132608,
              "mean": 1236515392.9014084,
              "median": 1248436224.0,
              "minimum": 1210281984,
              "sample_count": 142
            }
          },
          "process_sample_count": 142,
          "response_union_seconds": 145.41683486895636,
          "sampled_counter_deltas": {
            "deltas": {
              "major_faults": 0,
              "read_syscall_bytes": 0,
              "storage_read_bytes": 0,
              "storage_write_bytes": 126976
            },
            "observed_endpoint_seconds": 143.68809728400083,
            "observed_intervals": 2,
            "requested_intervals": 2
          }
        },
        "warmup": {
          "gpu_energy": {
            "available": true,
            "interpolation": "linear_at_bracketed_boundaries",
            "joules": 8338.197633615331,
            "reason": null,
            "response_union_seconds": 28.63761988701299,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 291.1623824365607,
            "trapezoid_pieces": 31,
            "watt_hours": 2.316166009337592
          },
          "gpu_sample_count": 29,
          "host_sample_count": 28,
          "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": 17373618176,
              "mean": 17343057334.857143,
              "median": 17341583360.0,
              "minimum": 17320845312,
              "sample_count": 28
            },
            "gpu_board_watts": {
              "maximum": 300.15,
              "mean": 289.5420689655172,
              "median": 300.0,
              "minimum": 115.61,
              "sample_count": 29
            },
            "gpu_memory_used_bytes": {
              "maximum": 55660511232.0,
              "mean": 55660511232.0,
              "median": 55660511232.0,
              "minimum": 55660511232.0,
              "sample_count": 29
            },
            "gpu_sm_clock_mhz": {
              "maximum": 2332.0,
              "mean": 1529.0344827586207,
              "median": 1485.0,
              "minimum": 1440.0,
              "sample_count": 29
            },
            "gpu_utilization_percent": {
              "maximum": 100.0,
              "mean": 96.62068965517241,
              "median": 100.0,
              "minimum": 2.0,
              "sample_count": 29
            },
            "host_mem_available_bytes": {
              "maximum": 96039104512,
              "mean": 96025997019.42857,
              "median": 96028440576.0,
              "minimum": 96003043328,
              "sample_count": 28
            },
            "process_pss_bytes": {
              "maximum": 1240260608,
              "mean": 1219563081.142857,
              "median": 1218818048.0,
              "minimum": 1198239744,
              "sample_count": 28
            },
            "process_rss_bytes": {
              "maximum": 1243230208,
              "mean": 1222634057.142857,
              "median": 1221787648.0,
              "minimum": 1201209344,
              "sample_count": 28
            }
          },
          "process_sample_count": 28,
          "response_union_seconds": 28.63761988701299,
          "sampled_counter_deltas": {
            "deltas": {
              "major_faults": 0,
              "read_syscall_bytes": 0,
              "storage_read_bytes": 0,
              "storage_write_bytes": 24576
            },
            "observed_endpoint_seconds": 26.66104468901176,
            "observed_intervals": 2,
            "requested_intervals": 2
          }
        }
      },
      "reviewed_export_sha256": "59f09a9b924c4cb4ae1370f85109347542e72457c07f0a7a8ff351b31fedab64",
      "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": 4,
        "synthetic_prompt_sha256": "c099b5abc034050fac1133e0214990d1f72a124b64f36e8a80ba3209bc07d187",
        "temperature": 0,
        "threads": 16,
        "tokenization": "model_tokenizer_add_special_false_repeat_then_truncate",
        "total_context_tokens": 34816,
        "ubatch": 512,
        "warmup_rounds": 1
      },
      "status": "completed",
      "telemetry": {
        "gpu_samples": 187,
        "host_samples": 181,
        "maximum_gpu_gap_seconds": 1.0082573510007933,
        "maximum_host_gap_seconds": 1.6206080699921586,
        "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": 187,
            "unavailable": 0
          },
          "software": {
            "active": 0,
            "not_active": 187,
            "unavailable": 0
          }
        }
      }
    },
    {
      "case_wall_seconds": 719.4450020020013,
      "completed_response_resources": {
        "gpu_energy": {
          "available": true,
          "interpolation": "linear_at_bracketed_boundaries",
          "joules": 116153.30866974915,
          "reason": null,
          "response_union_seconds": 387.2295688489685,
          "scope": "GPU board only; not whole-server AC energy",
          "time_weighted_mean_watts": 299.9598119921791,
          "trapezoid_pieces": 397,
          "watt_hours": 32.26480796381921
        },
        "gpu_sample_count": 387,
        "host_sample_count": 376,
        "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": 17462124544,
            "mean": 17423276674.723404,
            "median": 17432784896.0,
            "minimum": 17366061056,
            "sample_count": 376
          },
          "gpu_board_watts": {
            "maximum": 301.05,
            "mean": 299.9867958656331,
            "median": 299.99,
            "minimum": 298.38,
            "sample_count": 387
          },
          "gpu_memory_used_bytes": {
            "maximum": 55660511232.0,
            "mean": 55660511232.0,
            "median": 55660511232.0,
            "minimum": 55660511232.0,
            "sample_count": 387
          },
          "gpu_sm_clock_mhz": {
            "maximum": 1635.0,
            "mean": 1438.095607235142,
            "median": 1417.0,
            "minimum": 1342.0,
            "sample_count": 387
          },
          "gpu_utilization_percent": {
            "maximum": 100.0,
            "mean": 99.6795865633075,
            "median": 100.0,
            "minimum": 91.0,
            "sample_count": 387
          },
          "host_mem_available_bytes": {
            "maximum": 96145973248,
            "mean": 96063168871.48936,
            "median": 96063717376.0,
            "minimum": 96005341184,
            "sample_count": 376
          },
          "process_pss_bytes": {
            "maximum": 1295631360,
            "mean": 1278426348.93617,
            "median": 1287165440.0,
            "minimum": 1231869952,
            "sample_count": 376
          },
          "process_rss_bytes": {
            "maximum": 1298599936,
            "mean": 1281417673.531915,
            "median": 1290297344.0,
            "minimum": 1236250624,
            "sample_count": 376
          }
        },
        "process_sample_count": 376,
        "response_union_seconds": 387.2295688489685,
        "sampled_counter_deltas": {
          "deltas": {
            "major_faults": 0,
            "read_syscall_bytes": 0,
            "storage_read_bytes": 0,
            "storage_write_bytes": 331776
          },
          "observed_endpoint_seconds": 375.2016968349926,
          "observed_intervals": 10,
          "requested_intervals": 10
        }
      },
      "groups": [
        {
          "aggregate_output_tokens_per_second": 38.93995189858778,
          "completed_output_tokens": 5120,
          "completed_request_count": 20,
          "context_tokens": 2048,
          "cpu_snapshot_intervals": {
            "host_capacity_percent": 3.124980218163133,
            "interval_count": 5,
            "interval_seconds": 131.50083242496476,
            "meaning": "Weighted CPU counter rate across round-bracketing snapshots, not exact token time.",
            "one_logical_cpu_percent": 99.99936698122026,
            "process_cpu_seconds": 131.5
          },
          "failed_request_count": 0,
          "measured_rounds": 5,
          "mode": "completion",
          "resources": {
            "gpu_energy": {
              "available": true,
              "interpolation": "linear_at_bracketed_boundaries",
              "joules": 39432.121308624584,
              "reason": null,
              "response_union_seconds": 131.484497293015,
              "scope": "GPU board only; not whole-server AC energy",
              "time_weighted_mean_watts": 299.8993959017812,
              "trapezoid_pieces": 136,
              "watt_hours": 10.953367030173496
            },
            "gpu_sample_count": 131,
            "host_sample_count": 127,
            "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": 17425453056,
                "mean": 17393815616.503937,
                "median": 17401016320,
                "minimum": 17366061056,
                "sample_count": 127
              },
              "gpu_board_watts": {
                "maximum": 300.33,
                "mean": 299.9786259541985,
                "median": 299.99,
                "minimum": 298.61,
                "sample_count": 131
              },
              "gpu_memory_used_bytes": {
                "maximum": 55660511232.0,
                "mean": 55660511232.0,
                "median": 55660511232.0,
                "minimum": 55660511232.0,
                "sample_count": 131
              },
              "gpu_sm_clock_mhz": {
                "maximum": 1582.0,
                "mean": 1428.8091603053435,
                "median": 1395.0,
                "minimum": 1357.0,
                "sample_count": 131
              },
              "gpu_utilization_percent": {
                "maximum": 100.0,
                "mean": 99.58778625954199,
                "median": 100.0,
                "minimum": 93.0,
                "sample_count": 131
              },
              "host_mem_available_bytes": {
                "maximum": 96145973248,
                "mean": 96061832957.98425,
                "median": 96068395008,
                "minimum": 96005341184,
                "sample_count": 127
              },
              "process_pss_bytes": {
                "maximum": 1273496576,
                "mean": 1257087443.6535432,
                "median": 1265202176,
                "minimum": 1231869952,
                "sample_count": 127
              },
              "process_rss_bytes": {
                "maximum": 1276465152,
                "mean": 1260089569.7637796,
                "median": 1268170752,
                "minimum": 1236250624,
                "sample_count": 127
              }
            },
            "process_sample_count": 127,
            "response_union_seconds": 131.484497293015,
            "sampled_counter_deltas": {
              "deltas": {
                "major_faults": 0,
                "read_syscall_bytes": 0,
                "storage_read_bytes": 0,
                "storage_write_bytes": 102400
              },
              "observed_endpoint_seconds": 125.07762075704522,
              "observed_intervals": 5,
              "requested_intervals": 5
            }
          },
          "response_outer_span_seconds": 131.53245825797785,
          "response_seconds": {
            "maximum": 26.65054602897726,
            "mean": 26.200603126001077,
            "median": 26.24883686102112,
            "minimum": 25.647156226972584,
            "sample_count": 20
          },
          "response_window_union_seconds": 131.484497293015,
          "server_decode_tokens_per_second": {
            "maximum": 13.635842265924056,
            "mean": 12.298470100174935,
            "median": 12.722440495557054,
            "minimum": 10.334289545557116,
            "sample_count": 20
          },
          "synthetic_first_token_seconds": {
            "maximum": 7.21949907601811,
            "mean": 5.264546025349409,
            "median": 6.091464050492505,
            "minimum": 1.6687007790314965,
            "sample_count": 20
          },
          "warmup_completed_rounds": 1,
          "warmup_requests_per_round": 4
        },
        {
          "aggregate_output_tokens_per_second": 20.01993613738052,
          "completed_output_tokens": 5120,
          "completed_request_count": 20,
          "context_tokens": 8192,
          "cpu_snapshot_intervals": {
            "host_capacity_percent": 3.1235835368990346,
            "interval_count": 5,
            "interval_seconds": 255.76593056099955,
            "meaning": "Weighted CPU counter rate across round-bracketing snapshots, not exact token time.",
            "one_logical_cpu_percent": 99.9546731807691,
            "process_cpu_seconds": 255.64999999999998
          },
          "failed_request_count": 0,
          "measured_rounds": 5,
          "mode": "completion",
          "resources": {
            "gpu_energy": {
              "available": true,
              "interpolation": "linear_at_bracketed_boundaries",
              "joules": 76721.1873611245,
              "reason": null,
              "response_union_seconds": 255.74507155595347,
              "scope": "GPU board only; not whole-server AC energy",
              "time_weighted_mean_watts": 299.99087331127305,
              "trapezoid_pieces": 261,
              "watt_hours": 21.31144093364569
            },
            "gpu_sample_count": 256,
            "host_sample_count": 249,
            "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": 17462124544,
                "mean": 17438302997.590363,
                "median": 17439318016,
                "minimum": 17421230080,
                "sample_count": 249
              },
              "gpu_board_watts": {
                "maximum": 301.05,
                "mean": 299.9909765625,
                "median": 300.0,
                "minimum": 298.38,
                "sample_count": 256
              },
              "gpu_memory_used_bytes": {
                "maximum": 55660511232.0,
                "mean": 55660511232.0,
                "median": 55660511232.0,
                "minimum": 55660511232.0,
                "sample_count": 256
              },
              "gpu_sm_clock_mhz": {
                "maximum": 1635.0,
                "mean": 1442.84765625,
                "median": 1500.0,
                "minimum": 1342.0,
                "sample_count": 256
              },
              "gpu_utilization_percent": {
                "maximum": 100.0,
                "mean": 99.7265625,
                "median": 100.0,
                "minimum": 91.0,
                "sample_count": 256
              },
              "host_mem_available_bytes": {
                "maximum": 96117440512,
                "mean": 96063850241.0281,
                "median": 96062423040,
                "minimum": 96024403968,
                "sample_count": 249
              },
              "process_pss_bytes": {
                "maximum": 1295631360,
                "mean": 1289310047.6144578,
                "median": 1290777600,
                "minimum": 1279087616,
                "sample_count": 249
              },
              "process_rss_bytes": {
                "maximum": 1298599936,
                "mean": 1292295863.0040162,
                "median": 1293746176,
                "minimum": 1282056192,
                "sample_count": 249
              }
            },
            "process_sample_count": 249,
            "response_union_seconds": 255.74507155595347,
            "sampled_counter_deltas": {
              "deltas": {
                "major_faults": 0,
                "read_syscall_bytes": 0,
                "storage_read_bytes": 0,
                "storage_write_bytes": 229376
              },
              "observed_endpoint_seconds": 250.1240760779474,
              "observed_intervals": 5,
              "requested_intervals": 5
            }
          },
          "response_outer_span_seconds": 255.81028590694768,
          "response_seconds": {
            "maximum": 51.4202549259644,
            "mean": 50.847095357597574,
            "median": 50.86070125648985,
            "minimum": 50.162616960995365,
            "sample_count": 20
          },
          "response_window_union_seconds": 255.74507155595347,
          "server_decode_tokens_per_second": {
            "maximum": 12.221575730155275,
            "mean": 8.901997778536677,
            "median": 8.665635784306376,
            "minimum": 5.8851670963954135,
            "sample_count": 20
          },
          "synthetic_first_token_seconds": {
            "maximum": 30.19842599896947,
            "mean": 20.22669281999115,
            "median": 21.229655119474046,
            "minimum": 7.263917299977038,
            "sample_count": 20
          },
          "warmup_completed_rounds": 1,
          "warmup_requests_per_round": 4
        }
      ],
      "id": "concurrency-72b-q4-c4",
      "model": "qwen25-72b-q4",
      "module": "concurrency",
      "placement": {
        "buffers": [
          {
            "device": "CPU_Mapped",
            "reported_mib": 668.25
          },
          {
            "device": "CUDA0",
            "reported_mib": 41297.69
          }
        ],
        "gpu_layers": 81,
        "meaning": "Runtime buffer allocation is not measured resident process RAM.",
        "requested_cpu_expert_layers": 0,
        "total_layers": 81
      },
      "record_sha256": "6b5e6972abf6be1b6a331c2f2a22b580fb83bd46502a421d8507dfd60ae28554",
      "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": 16989462528,
              "mean": 16980873216,
              "median": 16980873216.0,
              "minimum": 16972283904,
              "sample_count": 2
            },
            "gpu_board_watts": {
              "maximum": 15.82,
              "mean": 15.77,
              "median": 15.77,
              "minimum": 15.72,
              "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": 96533319680,
              "mean": 96521064448,
              "median": 96521064448.0,
              "minimum": 96508809216,
              "sample_count": 2
            },
            "process_pss_bytes": null,
            "process_rss_bytes": null
          },
          "process_sample_count": 0,
          "response_union_seconds": 1.1004123980528675,
          "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": 16997015552,
              "mean": 16997015552,
              "median": 16997015552,
              "minimum": 16997015552,
              "sample_count": 1
            },
            "gpu_board_watts": {
              "maximum": 112.0,
              "mean": 112.0,
              "median": 112.0,
              "minimum": 112.0,
              "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": 2340.0,
              "mean": 2340.0,
              "median": 2340.0,
              "minimum": 2340.0,
              "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": 96592494592,
              "mean": 96592494592,
              "median": 96592494592,
              "minimum": 96592494592,
              "sample_count": 1
            },
            "process_pss_bytes": null,
            "process_rss_bytes": null
          },
          "process_sample_count": 0,
          "response_union_seconds": 1.1598984710290097,
          "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": 500.6619383217032,
            "reason": null,
            "response_union_seconds": 7.096872266964056,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 70.54684366411452,
            "trapezoid_pieces": 8,
            "watt_hours": 0.13907276064491755
          },
          "gpu_sample_count": 7,
          "host_sample_count": 5,
          "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": 17269374976,
              "mean": 17249575731.2,
              "median": 17268948992,
              "minimum": 17184108544,
              "sample_count": 5
            },
            "gpu_board_watts": {
              "maximum": 78.69,
              "mean": 74.09714285714286,
              "median": 78.52,
              "minimum": 50.66,
              "sample_count": 7
            },
            "gpu_memory_used_bytes": {
              "maximum": 55622762496.0,
              "mean": 39380919442.28571,
              "median": 43891294208.0,
              "minimum": 587202560.0,
              "sample_count": 7
            },
            "gpu_sm_clock_mhz": {
              "maximum": 2310.0,
              "mean": 2300.0,
              "median": 2302.0,
              "minimum": 2272.0,
              "sample_count": 7
            },
            "gpu_utilization_percent": {
              "maximum": 81.0,
              "mean": 57.714285714285715,
              "median": 78.0,
              "minimum": 0.0,
              "sample_count": 7
            },
            "host_mem_available_bytes": {
              "maximum": 96162697216,
              "mean": 96149157478.4,
              "median": 96146513920,
              "minimum": 96130719744,
              "sample_count": 5
            },
            "process_pss_bytes": {
              "maximum": 44347048960,
              "mean": 34483454976,
              "median": 44346774528,
              "minimum": 1079759872,
              "sample_count": 5
            },
            "process_rss_bytes": {
              "maximum": 44350017536,
              "mean": 34486423552,
              "median": 44349743104,
              "minimum": 1082728448,
              "sample_count": 5
            }
          },
          "process_sample_count": 5,
          "response_union_seconds": 7.096872266964056,
          "sampled_counter_deltas": {
            "deltas": {
              "major_faults": 0,
              "read_syscall_bytes": 5,
              "storage_read_bytes": 0,
              "storage_write_bytes": 8192
            },
            "observed_endpoint_seconds": 5.7938264219556,
            "observed_intervals": 1,
            "requested_intervals": 1
          }
        },
        "ready": {
          "gpu_energy": {
            "available": true,
            "interpolation": "linear_at_bracketed_boundaries",
            "joules": 582.8318020362483,
            "reason": null,
            "response_union_seconds": 3.319325248012319,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 175.5874337367994,
            "trapezoid_pieces": 6,
            "watt_hours": 0.16189772278784675
          },
          "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": 17452093440,
              "mean": 17394784938.666668,
              "median": 17414164480,
              "minimum": 17318096896,
              "sample_count": 3
            },
            "gpu_board_watts": {
              "maximum": 255.29,
              "mean": 168.00333333333333,
              "median": 145.64,
              "minimum": 103.08,
              "sample_count": 3
            },
            "gpu_memory_used_bytes": {
              "maximum": 55660511232.0,
              "mean": 55650025472.0,
              "median": 55660511232.0,
              "minimum": 55629053952.0,
              "sample_count": 3
            },
            "gpu_sm_clock_mhz": {
              "maximum": 2340.0,
              "mean": 2160.0,
              "median": 2310.0,
              "minimum": 1830.0,
              "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": 96079769600,
              "mean": 96071659520,
              "median": 96077111296,
              "minimum": 96058097664,
              "sample_count": 3
            },
            "process_pss_bytes": {
              "maximum": 1295631360,
              "mean": 1243190954.6666667,
              "median": 1273775104,
              "minimum": 1160166400,
              "sample_count": 3
            },
            "process_rss_bytes": {
              "maximum": 1298599936,
              "mean": 1246632618.6666667,
              "median": 1276743680,
              "minimum": 1164554240,
              "sample_count": 3
            }
          },
          "process_sample_count": 3,
          "response_union_seconds": 3.319325248012319,
          "sampled_counter_deltas": {
            "deltas": null,
            "observed_endpoint_seconds": 0.0,
            "observed_intervals": 0,
            "requested_intervals": 3
          }
        },
        "responses": {
          "gpu_energy": {
            "available": true,
            "interpolation": "linear_at_bracketed_boundaries",
            "joules": 116195.08796807335,
            "reason": null,
            "response_union_seconds": 387.37040795898065,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 299.9586070094891,
            "trapezoid_pieces": 389,
            "watt_hours": 32.27641332446482
          },
          "gpu_sample_count": 387,
          "host_sample_count": 377,
          "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": 17462124544,
              "mean": 17423226070.57825,
              "median": 17432780800,
              "minimum": 17366061056,
              "sample_count": 377
            },
            "gpu_board_watts": {
              "maximum": 301.05,
              "mean": 299.9867958656331,
              "median": 299.99,
              "minimum": 298.38,
              "sample_count": 387
            },
            "gpu_memory_used_bytes": {
              "maximum": 55660511232.0,
              "mean": 55660511232.0,
              "median": 55660511232.0,
              "minimum": 55660511232.0,
              "sample_count": 387
            },
            "gpu_sm_clock_mhz": {
              "maximum": 1635.0,
              "mean": 1438.095607235142,
              "median": 1417.0,
              "minimum": 1342.0,
              "sample_count": 387
            },
            "gpu_utilization_percent": {
              "maximum": 100.0,
              "mean": 99.6795865633075,
              "median": 100.0,
              "minimum": 91.0,
              "sample_count": 387
            },
            "host_mem_available_bytes": {
              "maximum": 96145973248,
              "mean": 96063112493.49602,
              "median": 96063148032,
              "minimum": 96005341184,
              "sample_count": 377
            },
            "process_pss_bytes": {
              "maximum": 1295631360,
              "mean": 1278397334.0689654,
              "median": 1287002112,
              "minimum": 1231869952,
              "sample_count": 377
            },
            "process_rss_bytes": {
              "maximum": 1298599936,
              "mean": 1281388598.3236074,
              "median": 1290297344,
              "minimum": 1236250624,
              "sample_count": 377
            }
          },
          "process_sample_count": 377,
          "response_union_seconds": 387.37040795898065,
          "sampled_counter_deltas": {
            "deltas": {
              "major_faults": 0,
              "read_syscall_bytes": 0,
              "storage_read_bytes": 0,
              "storage_write_bytes": 372736
            },
            "observed_endpoint_seconds": 384.5076979199657,
            "observed_intervals": 2,
            "requested_intervals": 2
          }
        },
        "warmup": {
          "gpu_energy": {
            "available": true,
            "interpolation": "linear_at_bracketed_boundaries",
            "joules": 22950.795180775483,
            "reason": null,
            "response_union_seconds": 77.29993101599393,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 296.9057653625434,
            "trapezoid_pieces": 80,
            "watt_hours": 6.375220883548746
          },
          "gpu_sample_count": 78,
          "host_sample_count": 76,
          "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": 17433354240,
              "mean": 17396034937.263157,
              "median": 17416282112.0,
              "minimum": 17334210560,
              "sample_count": 76
            },
            "gpu_board_watts": {
              "maximum": 306.29,
              "mean": 296.84,
              "median": 299.99,
              "minimum": 129.61,
              "sample_count": 78
            },
            "gpu_memory_used_bytes": {
              "maximum": 55660511232.0,
              "mean": 55660511232.0,
              "median": 55660511232.0,
              "minimum": 55660511232.0,
              "sample_count": 78
            },
            "gpu_sm_clock_mhz": {
              "maximum": 2340.0,
              "mean": 1446.8589743589744,
              "median": 1380.0,
              "minimum": 1297.0,
              "sample_count": 78
            },
            "gpu_utilization_percent": {
              "maximum": 100.0,
              "mean": 99.76923076923077,
              "median": 100.0,
              "minimum": 98.0,
              "sample_count": 78
            },
            "host_mem_available_bytes": {
              "maximum": 96131813376,
              "mean": 96065721559.57895,
              "median": 96062631936.0,
              "minimum": 96013639680,
              "sample_count": 76
            },
            "process_pss_bytes": {
              "maximum": 1276486656,
              "mean": 1258132938.1052632,
              "median": 1276478464.0,
              "minimum": 1198920704,
              "sample_count": 76
            },
            "process_rss_bytes": {
              "maximum": 1279455232,
              "mean": 1261101918.3157895,
              "median": 1279447040.0,
              "minimum": 1201889280,
              "sample_count": 76
            }
          },
          "process_sample_count": 76,
          "response_union_seconds": 77.29993101599393,
          "sampled_counter_deltas": {
            "deltas": {
              "major_faults": 0,
              "read_syscall_bytes": 0,
              "storage_read_bytes": 0,
              "storage_write_bytes": 65536
            },
            "observed_endpoint_seconds": 75.86656208103523,
            "observed_intervals": 2,
            "requested_intervals": 2
          }
        }
      },
      "reviewed_export_sha256": "59f09a9b924c4cb4ae1370f85109347542e72457c07f0a7a8ff351b31fedab64",
      "settings": {
        "batch": 2048,
        "cache_prompt": false,
        "cache_ram_bytes": 0,
        "concurrency": 4,
        "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": 4,
        "synthetic_prompt_sha256": "c099b5abc034050fac1133e0214990d1f72a124b64f36e8a80ba3209bc07d187",
        "temperature": 0,
        "threads": 16,
        "tokenization": "model_tokenizer_add_special_false_repeat_then_truncate",
        "total_context_tokens": 34816,
        "ubatch": 512,
        "warmup_rounds": 1
      },
      "status": "completed",
      "telemetry": {
        "gpu_samples": 478,
        "host_samples": 464,
        "maximum_gpu_gap_seconds": 1.0062838830053806,
        "maximum_host_gap_seconds": 1.6154293110012077,
        "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": 478,
            "unavailable": 0
          },
          "software": {
            "active": 0,
            "not_active": 478,
            "unavailable": 0
          }
        }
      }
    },
    {
      "case_wall_seconds": 14450.58151647501,
      "completed_response_resources": {
        "gpu_energy": {
          "available": true,
          "interpolation": "linear_at_bracketed_boundaries",
          "joules": 4322526.512667371,
          "reason": null,
          "response_union_seconds": 14409.160561117635,
          "scope": "GPU board only; not whole-server AC energy",
          "time_weighted_mean_watts": 299.9846170311602,
          "trapezoid_pieces": 14984,
          "watt_hours": 1200.7018090742697
        },
        "gpu_sample_count": 14408,
        "host_sample_count": 14044,
        "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": 7715704832,
            "mean": 6801911819.374537,
            "median": 6799910912.0,
            "minimum": 5833928704,
            "sample_count": 14044
          },
          "gpu_board_watts": {
            "maximum": 305.13,
            "mean": 299.9847897001666,
            "median": 299.99,
            "minimum": 283.16,
            "sample_count": 14408
          },
          "gpu_memory_used_bytes": {
            "maximum": 29349642240.0,
            "mean": 29349642240.0,
            "median": 29349642240.0,
            "minimum": 29349642240.0,
            "sample_count": 14408
          },
          "gpu_sm_clock_mhz": {
            "maximum": 1792.0,
            "mean": 1531.5502498611881,
            "median": 1545.0,
            "minimum": 1477.0,
            "sample_count": 14408
          },
          "gpu_utilization_percent": {
            "maximum": 100.0,
            "mean": 99.21002220988339,
            "median": 100.0,
            "minimum": 50.0,
            "sample_count": 14408
          },
          "host_mem_available_bytes": {
            "maximum": 96192446464,
            "mean": 95707652121.37396,
            "median": 95703994368.0,
            "minimum": 95185977344,
            "sample_count": 14044
          },
          "process_pss_bytes": {
            "maximum": 1000995840,
            "mean": 994586522.5041299,
            "median": 999561216.0,
            "minimum": 939293696,
            "sample_count": 14044
          },
          "process_rss_bytes": {
            "maximum": 1003978752,
            "mean": 997597475.3631444,
            "median": 1002545152.0,
            "minimum": 942276608,
            "sample_count": 14044
          }
        },
        "process_sample_count": 14044,
        "response_union_seconds": 14409.160561117635,
        "sampled_counter_deltas": {
          "deltas": {
            "major_faults": 0,
            "read_syscall_bytes": 0,
            "storage_read_bytes": 0,
            "storage_write_bytes": 14311424
          },
          "observed_endpoint_seconds": 13820.278715579421,
          "observed_intervals": 576,
          "requested_intervals": 576
        }
      },
      "groups": [
        {
          "aggregate_output_tokens_per_second": 40.933959858259136,
          "completed_output_tokens": 589824,
          "completed_request_count": 2304,
          "context_tokens": 8192,
          "cpu_snapshot_intervals": {
            "host_capacity_percent": 3.1251006357900364,
            "interval_count": 576,
            "interval_seconds": 14411.525899105764,
            "meaning": "Weighted CPU counter rate across round-bracketing snapshots, not exact token time.",
            "one_logical_cpu_percent": 100.00322034528116,
            "process_cpu_seconds": 14411.99
          },
          "failed_request_count": 0,
          "measured_rounds": 576,
          "mode": "completion",
          "resources": {
            "gpu_energy": {
              "available": true,
              "interpolation": "linear_at_bracketed_boundaries",
              "joules": 4322526.512667371,
              "reason": null,
              "response_union_seconds": 14409.160561117635,
              "scope": "GPU board only; not whole-server AC energy",
              "time_weighted_mean_watts": 299.9846170311602,
              "trapezoid_pieces": 14984,
              "watt_hours": 1200.7018090742697
            },
            "gpu_sample_count": 14408,
            "host_sample_count": 14044,
            "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": 7715704832,
                "mean": 6801911819.374537,
                "median": 6799910912.0,
                "minimum": 5833928704,
                "sample_count": 14044
              },
              "gpu_board_watts": {
                "maximum": 305.13,
                "mean": 299.9847897001666,
                "median": 299.99,
                "minimum": 283.16,
                "sample_count": 14408
              },
              "gpu_memory_used_bytes": {
                "maximum": 29349642240.0,
                "mean": 29349642240.0,
                "median": 29349642240.0,
                "minimum": 29349642240.0,
                "sample_count": 14408
              },
              "gpu_sm_clock_mhz": {
                "maximum": 1792.0,
                "mean": 1531.5502498611881,
                "median": 1545.0,
                "minimum": 1477.0,
                "sample_count": 14408
              },
              "gpu_utilization_percent": {
                "maximum": 100.0,
                "mean": 99.21002220988339,
                "median": 100.0,
                "minimum": 50.0,
                "sample_count": 14408
              },
              "host_mem_available_bytes": {
                "maximum": 96192446464,
                "mean": 95707652121.37396,
                "median": 95703994368.0,
                "minimum": 95185977344,
                "sample_count": 14044
              },
              "process_pss_bytes": {
                "maximum": 1000995840,
                "mean": 994586522.5041299,
                "median": 999561216.0,
                "minimum": 939293696,
                "sample_count": 14044
              },
              "process_rss_bytes": {
                "maximum": 1003978752,
                "mean": 997597475.3631444,
                "median": 1002545152.0,
                "minimum": 942276608,
                "sample_count": 14044
              }
            },
            "process_sample_count": 14044,
            "response_union_seconds": 14409.160561117635,
            "sampled_counter_deltas": {
              "deltas": {
                "major_faults": 0,
                "read_syscall_bytes": 0,
                "storage_read_bytes": 0,
                "storage_write_bytes": 14311424
              },
              "observed_endpoint_seconds": 13820.278715579421,
              "observed_intervals": 576,
              "requested_intervals": 576
            }
          },
          "response_outer_span_seconds": 14418.665552712977,
          "response_seconds": {
            "maximum": 25.164631637977436,
            "mean": 24.864257757205955,
            "median": 24.896307882503606,
            "minimum": 23.494174277991988,
            "p95": 25.100366784119977,
            "sample_count": 2304
          },
          "response_window_union_seconds": 14409.160561117635,
          "server_decode_tokens_per_second": {
            "maximum": 25.484744831743715,
            "mean": 18.02991749691153,
            "median": 17.848019291580243,
            "minimum": 11.97676620079551,
            "sample_count": 2304
          },
          "synthetic_first_token_seconds": {
            "maximum": 14.590205368993338,
            "mean": 9.797891346179211,
            "median": 10.10864435750409,
            "minimum": 3.4045858300523832,
            "p95": 14.525741860616836,
            "sample_count": 2304
          },
          "warmup_completed_rounds": 1,
          "warmup_requests_per_round": 4
        }
      ],
      "id": "sustained-32b",
      "model": "qwen3-32b-q4",
      "module": "sustained",
      "placement": {
        "buffers": [
          {
            "device": "CPU_Mapped",
            "reported_mib": 417.3
          },
          {
            "device": "CUDA0",
            "reported_mib": 18423.65
          }
        ],
        "gpu_layers": 65,
        "meaning": "Runtime buffer allocation is not measured resident process RAM.",
        "requested_cpu_expert_layers": 0,
        "total_layers": 65
      },
      "record_sha256": "79abc8b59d526e04286b44ba1fdecc454ad81d540397d9dad2c2634dda632816",
      "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": 5514604544,
              "mean": 5505738752,
              "median": 5505738752.0,
              "minimum": 5496872960,
              "sample_count": 2
            },
            "gpu_board_watts": {
              "maximum": 15.58,
              "mean": 15.555,
              "median": 15.555,
              "minimum": 15.53,
              "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": 96590258176,
              "mean": 96524697600,
              "median": 96524697600.0,
              "minimum": 96459137024,
              "sample_count": 2
            },
            "process_pss_bytes": null,
            "process_rss_bytes": null
          },
          "process_sample_count": 0,
          "response_union_seconds": 1.1006686040200293,
          "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": 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": 7321759744,
              "mean": 7300308992,
              "median": 7300308992.0,
              "minimum": 7278858240,
              "sample_count": 2
            },
            "gpu_board_watts": {
              "maximum": 108.08,
              "mean": 108.08,
              "median": 108.08,
              "minimum": 108.08,
              "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": 2340.0,
              "mean": 2340.0,
              "median": 2340.0,
              "minimum": 2340.0,
              "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": 95783763968,
              "mean": 95729772544,
              "median": 95729772544.0,
              "minimum": 95675781120,
              "sample_count": 2
            },
            "process_pss_bytes": null,
            "process_rss_bytes": null
          },
          "process_sample_count": 0,
          "response_union_seconds": 1.1560256039956585,
          "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": 213.13813387022867,
            "reason": null,
            "response_union_seconds": 3.6325854180031456,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 58.6739496376088,
            "trapezoid_pieces": 4,
            "watt_hours": 0.05920503718617463
          },
          "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": 5706874880,
              "mean": 5706758144,
              "median": 5706758144.0,
              "minimum": 5706641408,
              "sample_count": 2
            },
            "gpu_board_watts": {
              "maximum": 77.09,
              "mean": 62.513333333333335,
              "median": 76.77,
              "minimum": 33.68,
              "sample_count": 3
            },
            "gpu_memory_used_bytes": {
              "maximum": 19906166784.0,
              "mean": 19906166784.0,
              "median": 19906166784.0,
              "minimum": 19906166784.0,
              "sample_count": 3
            },
            "gpu_sm_clock_mhz": {
              "maximum": 2302.0,
              "mean": 2292.0,
              "median": 2302.0,
              "minimum": 2272.0,
              "sample_count": 3
            },
            "gpu_utilization_percent": {
              "maximum": 82.0,
              "mean": 79.33333333333333,
              "median": 82.0,
              "minimum": 74.0,
              "sample_count": 3
            },
            "host_mem_available_bytes": {
              "maximum": 96227672064,
              "mean": 96217997312,
              "median": 96217997312.0,
              "minimum": 96208322560,
              "sample_count": 2
            },
            "process_pss_bytes": {
              "maximum": 20099197952,
              "mean": 20099126272,
              "median": 20099126272.0,
              "minimum": 20099054592,
              "sample_count": 2
            },
            "process_rss_bytes": {
              "maximum": 20102180864,
              "mean": 20102109184,
              "median": 20102109184.0,
              "minimum": 20102037504,
              "sample_count": 2
            }
          },
          "process_sample_count": 2,
          "response_union_seconds": 3.6325854180031456,
          "sampled_counter_deltas": {
            "deltas": {
              "major_faults": 0,
              "read_syscall_bytes": 5,
              "storage_read_bytes": 0,
              "storage_write_bytes": 0
            },
            "observed_endpoint_seconds": 1.3338036570348777,
            "observed_intervals": 1,
            "requested_intervals": 1
          }
        },
        "ready": {
          "gpu_energy": {
            "available": true,
            "interpolation": "linear_at_bracketed_boundaries",
            "joules": 327.03305081688416,
            "reason": null,
            "response_union_seconds": 2.223970894992817,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 147.0491594800931,
            "trapezoid_pieces": 4,
            "watt_hours": 0.09084251411580116
          },
          "gpu_sample_count": 2,
          "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": 7702061056,
              "mean": 6413938688,
              "median": 5770027008,
              "minimum": 5769728000,
              "sample_count": 3
            },
            "gpu_board_watts": {
              "maximum": 183.77,
              "mean": 136.445,
              "median": 136.445,
              "minimum": 89.12,
              "sample_count": 2
            },
            "gpu_memory_used_bytes": {
              "maximum": 29349642240.0,
              "mean": 29336010752.0,
              "median": 29336010752.0,
              "minimum": 29322379264.0,
              "sample_count": 2
            },
            "gpu_sm_clock_mhz": {
              "maximum": 2340.0,
              "mean": 2321.0,
              "median": 2321.0,
              "minimum": 2302.0,
              "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": 96134074368,
              "mean": 95850448213.33333,
              "median": 96131477504,
              "minimum": 95285792768,
              "sample_count": 3
            },
            "process_pss_bytes": {
              "maximum": 1000994816,
              "mean": 919594325.3333334,
              "median": 878894080,
              "minimum": 878894080,
              "sample_count": 3
            },
            "process_rss_bytes": {
              "maximum": 1003978752,
              "mean": 922577578.6666666,
              "median": 881876992,
              "minimum": 881876992,
              "sample_count": 3
            }
          },
          "process_sample_count": 3,
          "response_union_seconds": 2.223970894992817,
          "sampled_counter_deltas": {
            "deltas": {
              "major_faults": 0,
              "read_syscall_bytes": 0,
              "storage_read_bytes": 0,
              "storage_write_bytes": 0
            },
            "observed_endpoint_seconds": 1.021025077032391,
            "observed_intervals": 1,
            "requested_intervals": 2
          }
        },
        "responses": {
          "gpu_energy": {
            "available": true,
            "interpolation": "linear_at_bracketed_boundaries",
            "joules": 4325377.306417703,
            "reason": null,
            "response_union_seconds": 14418.681973422004,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 299.98423672778705,
            "trapezoid_pieces": 14420,
            "watt_hours": 1201.4936962271397
          },
          "gpu_sample_count": 14419,
          "host_sample_count": 14050,
          "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": 7715704832,
              "mean": 6801967864.01936,
              "median": 6799910912.0,
              "minimum": 5833928704,
              "sample_count": 14050
            },
            "gpu_board_watts": {
              "maximum": 305.13,
              "mean": 299.9844940703239,
              "median": 299.99,
              "minimum": 283.16,
              "sample_count": 14419
            },
            "gpu_memory_used_bytes": {
              "maximum": 29349642240.0,
              "mean": 29349642240.0,
              "median": 29349642240.0,
              "minimum": 29349642240.0,
              "sample_count": 14419
            },
            "gpu_sm_clock_mhz": {
              "maximum": 1792.0,
              "mean": 1531.5317289687218,
              "median": 1545.0,
              "minimum": 1477.0,
              "sample_count": 14419
            },
            "gpu_utilization_percent": {
              "maximum": 100.0,
              "mean": 99.20694916429711,
              "median": 100.0,
              "minimum": 50.0,
              "sample_count": 14419
            },
            "host_mem_available_bytes": {
              "maximum": 96192446464,
              "mean": 95707636376.68896,
              "median": 95703994368.0,
              "minimum": 95185977344,
              "sample_count": 14050
            },
            "process_pss_bytes": {
              "maximum": 1000995840,
              "mean": 994587084.7635587,
              "median": 999561216.0,
              "minimum": 939293696,
              "sample_count": 14050
            },
            "process_rss_bytes": {
              "maximum": 1003978752,
              "mean": 997598025.9393594,
              "median": 1002545152.0,
              "minimum": 942276608,
              "sample_count": 14050
            }
          },
          "process_sample_count": 14050,
          "response_union_seconds": 14418.681973422004,
          "sampled_counter_deltas": {
            "deltas": {
              "major_faults": 0,
              "read_syscall_bytes": 0,
              "storage_read_bytes": 0,
              "storage_write_bytes": 17612800
            },
            "observed_endpoint_seconds": 14416.98345157504,
            "observed_intervals": 1,
            "requested_intervals": 1
          }
        },
        "warmup": {
          "gpu_energy": {
            "available": true,
            "interpolation": "linear_at_bracketed_boundaries",
            "joules": 6987.65928224834,
            "reason": null,
            "response_union_seconds": 23.74817975901533,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 294.23978398157743,
            "trapezoid_pieces": 25,
            "watt_hours": 1.9410164672912056
          },
          "gpu_sample_count": 24,
          "host_sample_count": 23,
          "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": 5842784256,
              "mean": 5822150210.782609,
              "median": 5826940928,
              "minimum": 5801627648,
              "sample_count": 23
            },
            "gpu_board_watts": {
              "maximum": 306.76,
              "mean": 291.29333333333335,
              "median": 300.005,
              "minimum": 85.69,
              "sample_count": 24
            },
            "gpu_memory_used_bytes": {
              "maximum": 29349642240.0,
              "mean": 29348943189.333332,
              "median": 29349642240.0,
              "minimum": 29347545088.0,
              "sample_count": 24
            },
            "gpu_sm_clock_mhz": {
              "maximum": 2302.0,
              "mean": 1641.0416666666667,
              "median": 1635.0,
              "minimum": 1552.0,
              "sample_count": 24
            },
            "gpu_utilization_percent": {
              "maximum": 100.0,
              "mean": 98.875,
              "median": 99.0,
              "minimum": 90.0,
              "sample_count": 24
            },
            "host_mem_available_bytes": {
              "maximum": 96146788352,
              "mean": 96126758912,
              "median": 96126504960,
              "minimum": 96112836608,
              "sample_count": 23
            },
            "process_pss_bytes": {
              "maximum": 936446976,
              "mean": 929384670.6086956,
              "median": 933833728,
              "minimum": 910089216,
              "sample_count": 23
            },
            "process_rss_bytes": {
              "maximum": 939429888,
              "mean": 932429289.7391304,
              "median": 936816640,
              "minimum": 913072128,
              "sample_count": 23
            }
          },
          "process_sample_count": 23,
          "response_union_seconds": 23.74817975901533,
          "sampled_counter_deltas": {
            "deltas": {
              "major_faults": 8,
              "read_syscall_bytes": 0,
              "storage_read_bytes": 1441792,
              "storage_write_bytes": 20480
            },
            "observed_endpoint_seconds": 22.51987132197246,
            "observed_intervals": 1,
            "requested_intervals": 1
          }
        }
      },
      "reviewed_export_sha256": "7e9bf3881a4bd446b234cda9c158b651eb9d123d6af6090c39eed1e26ad459ac",
      "settings": {
        "batch": 2048,
        "cache_prompt": false,
        "cache_ram_bytes": 0,
        "concurrency": 4,
        "context_shift": false,
        "contexts": [
          8192
        ],
        "endpoint": "/completion",
        "fit": "off",
        "flash_attention": true,
        "ignore_eos": true,
        "kv_cache": "f16",
        "lazy_mode": "off",
        "load_mode": "mmap",
        "measured_rounds": null,
        "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": 4,
        "synthetic_prompt_sha256": "c099b5abc034050fac1133e0214990d1f72a124b64f36e8a80ba3209bc07d187",
        "temperature": 0,
        "threads": 16,
        "tokenization": "model_tokenizer_add_special_false_repeat_then_truncate",
        "total_context_tokens": 34816,
        "ubatch": 512,
        "warmup_rounds": 1
      },
      "status": "completed",
      "telemetry": {
        "gpu_samples": 14451,
        "host_samples": 14082,
        "maximum_gpu_gap_seconds": 1.0334704880369827,
        "maximum_host_gap_seconds": 1.5620193640352227,
        "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": 14451,
            "unavailable": 0
          },
          "software": {
            "active": 0,
            "not_active": 14451,
            "unavailable": 0
          }
        }
      }
    }
  ],
  "comparisons": [
    {
      "axis": "concurrency",
      "cases": [
        "concurrency-8b-c1",
        "concurrency-8b-c4"
      ]
    },
    {
      "axis": "concurrency",
      "cases": [
        "concurrency-32b-c1",
        "concurrency-32b-c4"
      ]
    },
    {
      "axis": "concurrency",
      "cases": [
        "concurrency-72b-q4-c1",
        "concurrency-72b-q4-c4"
      ]
    }
  ],
  "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"
    },
    {
      "architecture": "dense",
      "chat_template_sha256": "57f1fd00f0013a2be96aa79b857391f27e23df5b5f847072b524c897e24d0361",
      "effective_chat_template_rule": "llama.cpp-7fe450e19-chat-source-v1",
      "effective_chat_template_sha256": "57f1fd00f0013a2be96aa79b857391f27e23df5b5f847072b524c897e24d0361",
      "files": [
        {
          "bytes": 19762149024,
          "sha256": "efd971561896866f0e910cce52761ca77b1b138090c7f15fe284676d57d1f689"
        }
      ],
      "id": "qwen3-32b-q4",
      "name": "Qwen3-32B",
      "quantization": "Q4_K_M",
      "repository": "Qwen/Qwen3-32B-GGUF",
      "revision": "938a7432affaec9157f883a87164e2646ae17555",
      "tokenizer_sha256": "2629971f294a1a43ef545d689a2090a820e0e1a15e5371e1c88d1c21159ef849"
    },
    {
      "architecture": "dense",
      "chat_template_sha256": "d5495a1e5db0611132a97e46a65dbb64a642a499421228b9c8b93229097fa9a4",
      "effective_chat_template_rule": "llama.cpp-7fe450e19-chat-source-v1",
      "effective_chat_template_sha256": "4e9918361c284a93880606d182d64da6a9fe97cdc1f5c5a78c1c8840444246fc",
      "files": [
        {
          "bytes": 3986869120,
          "sha256": "3531a90691b5f4edb4bdf488246d72ce0b5d40cacfab1489718a662cec386e43"
        },
        {
          "bytes": 3877864544,
          "sha256": "a4cfb079461591e2c23ef73953aeba4b2520a9b93afaa841f31adf4e480647ab"
        },
        {
          "bytes": 3963622720,
          "sha256": "c67c3641d018b7ca646e7c37b50152687d4b52ce6ab6bf170b006a756b52778d"
        },
        {
          "bytes": 3941463040,
          "sha256": "92f5149d685ad5c5ab0fc30f5e0725d95a746d9a8e54d8397707b97dc435cd76"
        },
        {
          "bytes": 3963688384,
          "sha256": "f5abb6ecd7f3ec4438a4e89ae704acd43405182a0477776d5e8aa35a71a5ecdc"
        },
        {
          "bytes": 3941463040,
          "sha256": "ead67a4ac2648b0f7b71fea755f910ed208f3f8b881b55c208fc3fdca67359ee"
        },
        {
          "bytes": 3826849152,
          "sha256": "f7b168a5fdf78695736a6441dc846b19251339750ea78bf993057b7bf499671d"
        },
        {
          "bytes": 3983901792,
          "sha256": "4ab253ada85c14097597546ac13cfb1426cb175d1d8da520b7a2904d0c323e32"
        },
        {
          "bytes": 3921183968,
          "sha256": "51661b80af773b488e807a03fc7ea31bb64630c50e3ee77fbb5e37bdd9ff9979"
        },
        {
          "bytes": 3995005888,
          "sha256": "fcc180433606a81e2137ba1f088defe3ad51ec9353f5d129a7209614597e16f9"
        },
        {
          "bytes": 3586683584,
          "sha256": "57b9437649f12062bdbed6c623c3f6adc6924a3d2f4e7aabdd3ab32883f5485d"
        },
        {
          "bytes": 1021870240,
          "sha256": "60ae16428f9edb62b60e6c8657ba9344c6fae6ef44c497845dc6a6fd002e4d5b"
        }
      ],
      "id": "qwen25-72b-q4",
      "name": "Qwen2.5-72B-Instruct",
      "quantization": "Q4_K_M",
      "repository": "Qwen/Qwen2.5-72B-Instruct-GGUF",
      "revision": "7ca3bc388f97b264c4283bc9bf1055e2abc38441",
      "tokenizer_sha256": "67b3d77a05acfb80f96742cec2740cc6ee8fffadef98e655e855baf58a3946b1"
    }
  ],
  "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": "2026-09-27T15:11:44.063808+00:00",
      "plan_id": "core-core-20260927t135400-core",
      "plan_sha256": "d6b25477667ae18cdbc23760f8a5540291e3d88246734ade84d167a32fc9011c",
      "protocol_sha256": "3ce8ce9f2b17e66a655c07fd7e7aa3f1762790ad9be38f87d29e5e1279dec3ac",
      "record_sha256": "fb26d7e444a2a826e9f652b7e048338d0bb8ddb92c1672ae6b919fc2334d492c",
      "reviewed_export_sha256": "59f09a9b924c4cb4ae1370f85109347542e72457c07f0a7a8ff351b31fedab64",
      "runner_sha256": "1bac2436c5aa2eca4f4b42cc9b8f0942b5c66256eb1a41a2169b88c12b044f4e",
      "started_at": "2026-09-27T13:56:01.723397+00:00",
      "task_corpus_sha256": "0642db889565ce25b980e24afa5048eb83e4d15c36754457474a65b2c1f71c21"
    },
    {
      "exporter_sha256": "4f4046b91d99ba35175c3c6c04473a0021f54c6baa4892625cb66435f4c7e2ec",
      "finished_at": "2026-09-28T00:41:20.654174+00:00",
      "plan_id": "core-sustained-20260927t203551-sustained79182232",
      "plan_sha256": "1f23d7c3b12fbcd78b46e277e64c415da613fa72c7fc9d831d17ae6859cf87be",
      "protocol_sha256": "3ce8ce9f2b17e66a655c07fd7e7aa3f1762790ad9be38f87d29e5e1279dec3ac",
      "record_sha256": "8941ad01c40078a4ab873ea0b80a24088e0a3fe1cd6bf76c20f728bd2f8122a0",
      "reviewed_export_sha256": "7e9bf3881a4bd446b234cda9c158b651eb9d123d6af6090c39eed1e26ad459ac",
      "runner_sha256": "1bac2436c5aa2eca4f4b42cc9b8f0942b5c66256eb1a41a2169b88c12b044f4e",
      "started_at": "2026-09-27T20:36:24.408847+00:00",
      "task_corpus_sha256": "0642db889565ce25b980e24afa5048eb83e4d15c36754457474a65b2c1f71c21"
    }
  ],
  "status": "completed",
  "test_date_timezone": "Europe/Warsaw",
  "tested_at": "2026-09-28"
}
