{
  "article_key": "llm-system-memory",
  "cases": [
    {
      "case_wall_seconds": 29.91674370097462,
      "completed_response_resources": {
        "gpu_energy": {
          "available": true,
          "interpolation": "linear_at_bracketed_boundaries",
          "joules": 5704.656739645703,
          "reason": null,
          "response_union_seconds": 19.113167581963353,
          "scope": "GPU board only; not whole-server AC energy",
          "time_weighted_mean_watts": 298.4673636738817,
          "trapezoid_pieces": 30,
          "watt_hours": 1.5846268721238064
        },
        "gpu_sample_count": 20,
        "host_sample_count": 19,
        "meaning": "Samples inside declared interval union. Means are arithmetic sample means; peaks need not coincide. Counter deltas exclude unsampled edges; not minimum required RAM.",
        "metrics": {
          "cgroup_memory_current_bytes": {
            "maximum": 6720995328,
            "mean": 6712309221.052631,
            "median": 6713368576,
            "minimum": 6703079424,
            "sample_count": 19
          },
          "gpu_board_watts": {
            "maximum": 301.0,
            "mean": 299.713,
            "median": 299.95,
            "minimum": 297.7,
            "sample_count": 20
          },
          "gpu_memory_used_bytes": {
            "maximum": 6752829440.0,
            "mean": 6752829440.0,
            "median": 6752829440.0,
            "minimum": 6752829440.0,
            "sample_count": 20
          },
          "gpu_sm_clock_mhz": {
            "maximum": 1845.0,
            "mean": 1793.7,
            "median": 1800.0,
            "minimum": 1710.0,
            "sample_count": 20
          },
          "gpu_utilization_percent": {
            "maximum": 100.0,
            "mean": 95.9,
            "median": 97.0,
            "minimum": 85.0,
            "sample_count": 20
          },
          "host_mem_available_bytes": {
            "maximum": 96199720960,
            "mean": 96188027095.57895,
            "median": 96187265024,
            "minimum": 96176087040,
            "sample_count": 19
          },
          "process_pss_bytes": {
            "maximum": 819189760,
            "mean": 815475550.3157895,
            "median": 816355328,
            "minimum": 811747328,
            "sample_count": 19
          },
          "process_rss_bytes": {
            "maximum": 822177792,
            "mean": 818463582.3157895,
            "median": 819343360,
            "minimum": 814735360,
            "sample_count": 19
          }
        },
        "process_sample_count": 19,
        "response_union_seconds": 19.113167581963353,
        "sampled_counter_deltas": {
          "deltas": {
            "major_faults": 0,
            "read_syscall_bytes": 0,
            "storage_read_bytes": 0,
            "storage_write_bytes": 0
          },
          "observed_endpoint_seconds": 9.175436379155144,
          "observed_intervals": 8,
          "requested_intervals": 10
        }
      },
      "groups": [
        {
          "aggregate_output_tokens_per_second": 172.7690055501098,
          "completed_output_tokens": 1280,
          "completed_request_count": 5,
          "context_tokens": 2048,
          "cpu_snapshot_intervals": {
            "host_capacity_percent": 3.119277758589535,
            "interval_count": 5,
            "interval_seconds": 7.423593470070045,
            "meaning": "Weighted CPU counter rate across round-bracketing snapshots, not exact token time.",
            "one_logical_cpu_percent": 99.81688827486512,
            "process_cpu_seconds": 7.41
          },
          "failed_request_count": 0,
          "measured_rounds": 5,
          "mode": "completion",
          "resources": {
            "gpu_energy": {
              "available": true,
              "interpolation": "linear_at_bracketed_boundaries",
              "joules": 2207.7645157894385,
              "reason": null,
              "response_union_seconds": 7.408736283017788,
              "scope": "GPU board only; not whole-server AC energy",
              "time_weighted_mean_watts": 297.99474990762576,
              "trapezoid_pieces": 13,
              "watt_hours": 0.6132679210526218
            },
            "gpu_sample_count": 8,
            "host_sample_count": 8,
            "meaning": "Samples inside declared interval union. Means are arithmetic sample means; peaks need not coincide. Counter deltas exclude unsampled edges; not minimum required RAM.",
            "metrics": {
              "cgroup_memory_current_bytes": {
                "maximum": 6708359168,
                "mean": 6705936384,
                "median": 6706384896.0,
                "minimum": 6703079424,
                "sample_count": 8
              },
              "gpu_board_watts": {
                "maximum": 300.43,
                "mean": 299.49625,
                "median": 299.95,
                "minimum": 297.7,
                "sample_count": 8
              },
              "gpu_memory_used_bytes": {
                "maximum": 6752829440.0,
                "mean": 6752829440.0,
                "median": 6752829440.0,
                "minimum": 6752829440.0,
                "sample_count": 8
              },
              "gpu_sm_clock_mhz": {
                "maximum": 1845.0,
                "mean": 1832.5,
                "median": 1833.5,
                "minimum": 1822.0,
                "sample_count": 8
              },
              "gpu_utilization_percent": {
                "maximum": 97.0,
                "mean": 94.25,
                "median": 97.0,
                "minimum": 86.0,
                "sample_count": 8
              },
              "host_mem_available_bytes": {
                "maximum": 96199720960,
                "mean": 96194545152,
                "median": 96195299328.0,
                "minimum": 96187265024,
                "sample_count": 8
              },
              "process_pss_bytes": {
                "maximum": 813074432,
                "mean": 812405760,
                "median": 812410880.0,
                "minimum": 811747328,
                "sample_count": 8
              },
              "process_rss_bytes": {
                "maximum": 816062464,
                "mean": 815393792,
                "median": 815398912.0,
                "minimum": 814735360,
                "sample_count": 8
              }
            },
            "process_sample_count": 8,
            "response_union_seconds": 7.408736283017788,
            "sampled_counter_deltas": {
              "deltas": {
                "major_faults": 0,
                "read_syscall_bytes": 0,
                "storage_read_bytes": 0,
                "storage_write_bytes": 0
              },
              "observed_endpoint_seconds": 3.05030764703406,
              "observed_intervals": 3,
              "requested_intervals": 5
            }
          },
          "response_outer_span_seconds": 7.430938940960914,
          "response_seconds": {
            "maximum": 1.484641135961283,
            "mean": 1.4817472566035577,
            "median": 1.481124191021081,
            "minimum": 1.479376372997649,
            "sample_count": 5
          },
          "response_window_union_seconds": 7.408736283017788,
          "server_decode_tokens_per_second": {
            "maximum": 197.7636422013653,
            "mean": 197.5064926865939,
            "median": 197.53690841569204,
            "minimum": 197.21684494811652,
            "sample_count": 5
          },
          "synthetic_first_token_seconds": {
            "maximum": 0.19157884595915675,
            "mean": 0.1905686531914398,
            "median": 0.19060329400235787,
            "minimum": 0.1899017799878493,
            "sample_count": 5
          },
          "warmup_completed_rounds": 1,
          "warmup_requests_per_round": 1
        },
        {
          "aggregate_output_tokens_per_second": 109.36028990279206,
          "completed_output_tokens": 1280,
          "completed_request_count": 5,
          "context_tokens": 8192,
          "cpu_snapshot_intervals": {
            "host_capacity_percent": 3.118724755561472,
            "interval_count": 5,
            "interval_seconds": 11.72354178893147,
            "meaning": "Weighted CPU counter rate across round-bracketing snapshots, not exact token time.",
            "one_logical_cpu_percent": 99.7991921779671,
            "process_cpu_seconds": 11.7
          },
          "failed_request_count": 0,
          "measured_rounds": 5,
          "mode": "completion",
          "resources": {
            "gpu_energy": {
              "available": true,
              "interpolation": "linear_at_bracketed_boundaries",
              "joules": 3496.8922238562645,
              "reason": null,
              "response_union_seconds": 11.704431298945565,
              "scope": "GPU board only; not whole-server AC energy",
              "time_weighted_mean_watts": 298.76652137479715,
              "trapezoid_pieces": 17,
              "watt_hours": 0.9713589510711846
            },
            "gpu_sample_count": 12,
            "host_sample_count": 11,
            "meaning": "Samples inside declared interval union. Means are arithmetic sample means; peaks need not coincide. Counter deltas exclude unsampled edges; not minimum required RAM.",
            "metrics": {
              "cgroup_memory_current_bytes": {
                "maximum": 6720995328,
                "mean": 6716944011.636364,
                "median": 6717145088,
                "minimum": 6713245696,
                "sample_count": 11
              },
              "gpu_board_watts": {
                "maximum": 301.0,
                "mean": 299.8575,
                "median": 299.995,
                "minimum": 298.01,
                "sample_count": 12
              },
              "gpu_memory_used_bytes": {
                "maximum": 6752829440.0,
                "mean": 6752829440.0,
                "median": 6752829440.0,
                "minimum": 6752829440.0,
                "sample_count": 12
              },
              "gpu_sm_clock_mhz": {
                "maximum": 1822.0,
                "mean": 1767.8333333333333,
                "median": 1773.5,
                "minimum": 1710.0,
                "sample_count": 12
              },
              "gpu_utilization_percent": {
                "maximum": 100.0,
                "mean": 97.0,
                "median": 97.5,
                "minimum": 85.0,
                "sample_count": 12
              },
              "host_mem_available_bytes": {
                "maximum": 96188547072,
                "mean": 96183286690.90909,
                "median": 96184393728,
                "minimum": 96176087040,
                "sample_count": 11
              },
              "process_pss_bytes": {
                "maximum": 819189760,
                "mean": 817708125.0909091,
                "median": 817772544,
                "minimum": 816355328,
                "sample_count": 11
              },
              "process_rss_bytes": {
                "maximum": 822177792,
                "mean": 820696157.0909091,
                "median": 820760576,
                "minimum": 819343360,
                "sample_count": 11
              }
            },
            "process_sample_count": 11,
            "response_union_seconds": 11.704431298945565,
            "sampled_counter_deltas": {
              "deltas": {
                "major_faults": 0,
                "read_syscall_bytes": 0,
                "storage_read_bytes": 0,
                "storage_write_bytes": 0
              },
              "observed_endpoint_seconds": 6.125128732121084,
              "observed_intervals": 5,
              "requested_intervals": 5
            }
          },
          "response_outer_span_seconds": 11.733725362981204,
          "response_seconds": {
            "maximum": 2.3454450629651546,
            "mean": 2.340886259789113,
            "median": 2.3407816879916936,
            "minimum": 2.3358659429941326,
            "sample_count": 5
          },
          "response_window_union_seconds": 11.704431298945565,
          "server_decode_tokens_per_second": {
            "maximum": 169.3727562261093,
            "mean": 169.06737906840763,
            "median": 169.01162004597117,
            "minimum": 168.80352064566353,
            "sample_count": 5
          },
          "synthetic_first_token_seconds": {
            "maximum": 0.834626693977043,
            "mean": 0.8324093010160141,
            "median": 0.8318316700169817,
            "minimum": 0.8301292260293849,
            "sample_count": 5
          },
          "warmup_completed_rounds": 1,
          "warmup_requests_per_round": 1
        }
      ],
      "id": "baseline-8b",
      "model": "qwen3-8b-q4",
      "module": "baseline",
      "placement": {
        "buffers": [
          {
            "device": "CPU_Mapped",
            "reported_mib": 333.84
          },
          {
            "device": "CUDA0",
            "reported_mib": 4455.34
          }
        ],
        "gpu_layers": 37,
        "meaning": "Runtime buffer allocation is not measured resident process RAM.",
        "requested_cpu_expert_layers": 0,
        "total_layers": 37
      },
      "record_sha256": "2db66c2c90e7ccf91ace8aaf7f6b8e1865a1de2453d6547e769e2dda7e176b21",
      "resources_by_phase": {
        "baseline": {
          "gpu_energy": {
            "available": false,
            "joules": null,
            "reason": "unbracketed_boundary"
          },
          "gpu_sample_count": 2,
          "host_sample_count": 2,
          "meaning": "Samples inside declared interval union. Means are arithmetic sample means; peaks need not coincide. Counter deltas exclude unsampled edges; not minimum required RAM.",
          "metrics": {
            "cgroup_memory_current_bytes": {
              "maximum": 6373740544,
              "mean": 6365132800,
              "median": 6365132800.0,
              "minimum": 6356525056,
              "sample_count": 2
            },
            "gpu_board_watts": {
              "maximum": 15.03,
              "mean": 15.024999999999999,
              "median": 15.024999999999999,
              "minimum": 15.02,
              "sample_count": 2
            },
            "gpu_memory_used_bytes": {
              "maximum": 2097152.0,
              "mean": 2097152.0,
              "median": 2097152.0,
              "minimum": 2097152.0,
              "sample_count": 2
            },
            "gpu_sm_clock_mhz": {
              "maximum": 180.0,
              "mean": 180.0,
              "median": 180.0,
              "minimum": 180.0,
              "sample_count": 2
            },
            "gpu_utilization_percent": {
              "maximum": 0.0,
              "mean": 0.0,
              "median": 0.0,
              "minimum": 0.0,
              "sample_count": 2
            },
            "host_mem_available_bytes": {
              "maximum": 96634912768,
              "mean": 96623800320,
              "median": 96623800320.0,
              "minimum": 96612687872,
              "sample_count": 2
            },
            "process_pss_bytes": null,
            "process_rss_bytes": null
          },
          "process_sample_count": 0,
          "response_union_seconds": 1.1007707070093602,
          "sampled_counter_deltas": {
            "deltas": null,
            "observed_endpoint_seconds": 0.0,
            "observed_intervals": 0,
            "requested_intervals": 1
          }
        },
        "cleanup": {
          "gpu_energy": {
            "available": false,
            "joules": null,
            "reason": "unbracketed_boundary"
          },
          "gpu_sample_count": 1,
          "host_sample_count": 1,
          "meaning": "Samples inside declared interval union. Means are arithmetic sample means; peaks need not coincide. Counter deltas exclude unsampled edges; not minimum required RAM.",
          "metrics": {
            "cgroup_memory_current_bytes": {
              "maximum": 6392254464,
              "mean": 6392254464,
              "median": 6392254464,
              "minimum": 6392254464,
              "sample_count": 1
            },
            "gpu_board_watts": {
              "maximum": 78.38,
              "mean": 78.38,
              "median": 78.38,
              "minimum": 78.38,
              "sample_count": 1
            },
            "gpu_memory_used_bytes": {
              "maximum": 2097152.0,
              "mean": 2097152.0,
              "median": 2097152.0,
              "minimum": 2097152.0,
              "sample_count": 1
            },
            "gpu_sm_clock_mhz": {
              "maximum": 2325.0,
              "mean": 2325.0,
              "median": 2325.0,
              "minimum": 2325.0,
              "sample_count": 1
            },
            "gpu_utilization_percent": {
              "maximum": 0.0,
              "mean": 0.0,
              "median": 0.0,
              "minimum": 0.0,
              "sample_count": 1
            },
            "host_mem_available_bytes": {
              "maximum": 96599707648,
              "mean": 96599707648,
              "median": 96599707648,
              "minimum": 96599707648,
              "sample_count": 1
            },
            "process_pss_bytes": null,
            "process_rss_bytes": null
          },
          "process_sample_count": 0,
          "response_union_seconds": 1.155548368988093,
          "sampled_counter_deltas": {
            "deltas": null,
            "observed_endpoint_seconds": 0.0,
            "observed_intervals": 0,
            "requested_intervals": 1
          }
        },
        "load": {
          "gpu_energy": {
            "available": true,
            "interpolation": "linear_at_bracketed_boundaries",
            "joules": 51.318043537193375,
            "reason": null,
            "response_union_seconds": 1.2843944630003534,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 39.95504886973283,
            "trapezoid_pieces": 2,
            "watt_hours": 0.014255012093664827
          },
          "gpu_sample_count": 1,
          "host_sample_count": 1,
          "meaning": "Samples inside declared interval union. Means are arithmetic sample means; peaks need not coincide. Counter deltas exclude unsampled edges; not minimum required RAM.",
          "metrics": {
            "cgroup_memory_current_bytes": {
              "maximum": 6579744768,
              "mean": 6579744768,
              "median": 6579744768,
              "minimum": 6579744768,
              "sample_count": 1
            },
            "gpu_board_watts": {
              "maximum": 50.74,
              "mean": 50.74,
              "median": 50.74,
              "minimum": 50.74,
              "sample_count": 1
            },
            "gpu_memory_used_bytes": {
              "maximum": 5259657216.0,
              "mean": 5259657216.0,
              "median": 5259657216.0,
              "minimum": 5259657216.0,
              "sample_count": 1
            },
            "gpu_sm_clock_mhz": {
              "maximum": 2280.0,
              "mean": 2280.0,
              "median": 2280.0,
              "minimum": 2280.0,
              "sample_count": 1
            },
            "gpu_utilization_percent": {
              "maximum": 76.0,
              "mean": 76.0,
              "median": 76.0,
              "minimum": 76.0,
              "sample_count": 1
            },
            "host_mem_available_bytes": {
              "maximum": 96307732480,
              "mean": 96307732480,
              "median": 96307732480,
              "minimum": 96307732480,
              "sample_count": 1
            },
            "process_pss_bytes": {
              "maximum": 5364942848,
              "mean": 5364942848,
              "median": 5364942848,
              "minimum": 5364942848,
              "sample_count": 1
            },
            "process_rss_bytes": {
              "maximum": 5367930880,
              "mean": 5367930880,
              "median": 5367930880,
              "minimum": 5367930880,
              "sample_count": 1
            }
          },
          "process_sample_count": 1,
          "response_union_seconds": 1.2843944630003534,
          "sampled_counter_deltas": {
            "deltas": null,
            "observed_endpoint_seconds": 0.0,
            "observed_intervals": 0,
            "requested_intervals": 1
          }
        },
        "ready": {
          "gpu_energy": {
            "available": true,
            "interpolation": "linear_at_bracketed_boundaries",
            "joules": 521.8941446628455,
            "reason": null,
            "response_union_seconds": 3.3171223630197346,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 157.33340152931243,
            "trapezoid_pieces": 6,
            "watt_hours": 0.1449705957396793
          },
          "gpu_sample_count": 3,
          "host_sample_count": 4,
          "meaning": "Samples inside declared interval union. Means are arithmetic sample means; peaks need not coincide. Counter deltas exclude unsampled edges; not minimum required RAM.",
          "metrics": {
            "cgroup_memory_current_bytes": {
              "maximum": 6721740800,
              "mean": 6704672768,
              "median": 6714998784.0,
              "minimum": 6666952704,
              "sample_count": 4
            },
            "gpu_board_watts": {
              "maximum": 226.38,
              "mean": 158.82666666666665,
              "median": 176.54,
              "minimum": 73.56,
              "sample_count": 3
            },
            "gpu_memory_used_bytes": {
              "maximum": 6752829440.0,
              "mean": 6746537984.0,
              "median": 6752829440.0,
              "minimum": 6733955072.0,
              "sample_count": 3
            },
            "gpu_sm_clock_mhz": {
              "maximum": 2325.0,
              "mean": 2310.0,
              "median": 2310.0,
              "minimum": 2295.0,
              "sample_count": 3
            },
            "gpu_utilization_percent": {
              "maximum": 0.0,
              "mean": 0.0,
              "median": 0.0,
              "minimum": 0.0,
              "sample_count": 3
            },
            "host_mem_available_bytes": {
              "maximum": 96215072768,
              "mean": 96190546944,
              "median": 96184281088.0,
              "minimum": 96178552832,
              "sample_count": 4
            },
            "process_pss_bytes": {
              "maximum": 819189760,
              "mean": 807820288,
              "median": 816168960.0,
              "minimum": 779753472,
              "sample_count": 4
            },
            "process_rss_bytes": {
              "maximum": 822177792,
              "mean": 810808320,
              "median": 819156992.0,
              "minimum": 782741504,
              "sample_count": 4
            }
          },
          "process_sample_count": 4,
          "response_union_seconds": 3.3171223630197346,
          "sampled_counter_deltas": {
            "deltas": {
              "major_faults": 0,
              "read_syscall_bytes": 0,
              "storage_read_bytes": 0,
              "storage_write_bytes": 0
            },
            "observed_endpoint_seconds": 1.0194021269562654,
            "observed_intervals": 1,
            "requested_intervals": 3
          }
        },
        "responses": {
          "gpu_energy": {
            "available": true,
            "interpolation": "linear_at_bracketed_boundaries",
            "joules": 5723.3732039281185,
            "reason": null,
            "response_union_seconds": 19.17764565697871,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 298.4398244862446,
            "trapezoid_pieces": 22,
            "watt_hours": 1.5898258899800328
          },
          "gpu_sample_count": 20,
          "host_sample_count": 19,
          "meaning": "Samples inside declared interval union. Means are arithmetic sample means; peaks need not coincide. Counter deltas exclude unsampled edges; not minimum required RAM.",
          "metrics": {
            "cgroup_memory_current_bytes": {
              "maximum": 6720995328,
              "mean": 6712309221.052631,
              "median": 6713368576,
              "minimum": 6703079424,
              "sample_count": 19
            },
            "gpu_board_watts": {
              "maximum": 301.0,
              "mean": 299.713,
              "median": 299.95,
              "minimum": 297.7,
              "sample_count": 20
            },
            "gpu_memory_used_bytes": {
              "maximum": 6752829440.0,
              "mean": 6752829440.0,
              "median": 6752829440.0,
              "minimum": 6752829440.0,
              "sample_count": 20
            },
            "gpu_sm_clock_mhz": {
              "maximum": 1845.0,
              "mean": 1793.7,
              "median": 1800.0,
              "minimum": 1710.0,
              "sample_count": 20
            },
            "gpu_utilization_percent": {
              "maximum": 100.0,
              "mean": 95.9,
              "median": 97.0,
              "minimum": 85.0,
              "sample_count": 20
            },
            "host_mem_available_bytes": {
              "maximum": 96199720960,
              "mean": 96188027095.57895,
              "median": 96187265024,
              "minimum": 96176087040,
              "sample_count": 19
            },
            "process_pss_bytes": {
              "maximum": 819189760,
              "mean": 815475550.3157895,
              "median": 816355328,
              "minimum": 811747328,
              "sample_count": 19
            },
            "process_rss_bytes": {
              "maximum": 822177792,
              "mean": 818463582.3157895,
              "median": 819343360,
              "minimum": 814735360,
              "sample_count": 19
            }
          },
          "process_sample_count": 19,
          "response_union_seconds": 19.17764565697871,
          "sampled_counter_deltas": {
            "deltas": {
              "major_faults": 0,
              "read_syscall_bytes": 0,
              "storage_read_bytes": 0,
              "storage_write_bytes": 16384
            },
            "observed_endpoint_seconds": 17.353597005014308,
            "observed_intervals": 2,
            "requested_intervals": 2
          }
        },
        "warmup": {
          "gpu_energy": {
            "available": true,
            "interpolation": "linear_at_bracketed_boundaries",
            "joules": 888.6925389093352,
            "reason": null,
            "response_union_seconds": 3.8429829700035043,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 231.25070963000516,
            "trapezoid_pieces": 5,
            "watt_hours": 0.24685903858592645
          },
          "gpu_sample_count": 3,
          "host_sample_count": 3,
          "meaning": "Samples inside declared interval union. Means are arithmetic sample means; peaks need not coincide. Counter deltas exclude unsampled edges; not minimum required RAM.",
          "metrics": {
            "cgroup_memory_current_bytes": {
              "maximum": 6711808000,
              "mean": 6708021930.666667,
              "median": 6709710848,
              "minimum": 6702546944,
              "sample_count": 3
            },
            "gpu_board_watts": {
              "maximum": 300.38,
              "mean": 216.23666666666668,
              "median": 180.59,
              "minimum": 167.74,
              "sample_count": 3
            },
            "gpu_memory_used_bytes": {
              "maximum": 6752829440.0,
              "mean": 6752829440.0,
              "median": 6752829440.0,
              "minimum": 6752829440.0,
              "sample_count": 3
            },
            "gpu_sm_clock_mhz": {
              "maximum": 1905.0,
              "mean": 1807.3333333333333,
              "median": 1777.0,
              "minimum": 1740.0,
              "sample_count": 3
            },
            "gpu_utilization_percent": {
              "maximum": 100.0,
              "mean": 98.0,
              "median": 97.0,
              "minimum": 97.0,
              "sample_count": 3
            },
            "host_mem_available_bytes": {
              "maximum": 96199966720,
              "mean": 96187517610.66667,
              "median": 96182734848,
              "minimum": 96179851264,
              "sample_count": 3
            },
            "process_pss_bytes": {
              "maximum": 815646720,
              "mean": 813635584,
              "median": 813856768,
              "minimum": 811403264,
              "sample_count": 3
            },
            "process_rss_bytes": {
              "maximum": 818634752,
              "mean": 816623616,
              "median": 816844800,
              "minimum": 814391296,
              "sample_count": 3
            }
          },
          "process_sample_count": 3,
          "response_union_seconds": 3.8429829700035043,
          "sampled_counter_deltas": {
            "deltas": {
              "major_faults": 0,
              "read_syscall_bytes": 0,
              "storage_read_bytes": 0,
              "storage_write_bytes": 0
            },
            "observed_endpoint_seconds": 1.046531479980331,
            "observed_intervals": 1,
            "requested_intervals": 2
          }
        }
      },
      "reviewed_export_sha256": "2d1a82644c4ba9ba504ddf8fd2c7fe3c253e15cc9811363c52186bd4b7510a07",
      "settings": {
        "batch": 2048,
        "cache_prompt": false,
        "cache_ram_bytes": 0,
        "concurrency": 1,
        "context_shift": false,
        "contexts": [
          2048,
          8192
        ],
        "endpoint": "/completion",
        "fit": "off",
        "flash_attention": true,
        "ignore_eos": true,
        "kv_cache": "f16",
        "lazy_mode": "off",
        "load_mode": "mmap",
        "measured_rounds": 5,
        "mlock": false,
        "network_scope": "loopback",
        "output_tokens": 256,
        "p95_minimum_requests": 200,
        "placement": {
          "cpu_expert_layers": 0,
          "gpu_layers": "all"
        },
        "reasoning": "not_applicable_raw_completion",
        "seed": 20260927,
        "server_slots": 1,
        "synthetic_prompt_sha256": "c099b5abc034050fac1133e0214990d1f72a124b64f36e8a80ba3209bc07d187",
        "temperature": 0,
        "threads": 16,
        "tokenization": "model_tokenizer_add_special_false_repeat_then_truncate",
        "total_context_tokens": 8704,
        "ubatch": 512,
        "warmup_rounds": 1
      },
      "status": "completed",
      "telemetry": {
        "gpu_samples": 30,
        "host_samples": 30,
        "maximum_gpu_gap_seconds": 1.0048957320395857,
        "maximum_host_gap_seconds": 1.0701882099965587,
        "meaning": "Sampled observations; no numeric temperatures published and no warranty between samples.",
        "thermal_slowdown_fully_available": true,
        "thermal_slowdown_observed_samples": 0,
        "thermal_slowdown_states": {
          "hardware": {
            "active": 0,
            "not_active": 30,
            "unavailable": 0
          },
          "software": {
            "active": 0,
            "not_active": 30,
            "unavailable": 0
          }
        }
      }
    },
    {
      "case_wall_seconds": 268.15063832100714,
      "completed_response_resources": {
        "gpu_energy": {
          "available": true,
          "interpolation": "linear_at_bracketed_boundaries",
          "joules": 20773.514431872452,
          "reason": null,
          "response_union_seconds": 69.25191831210395,
          "scope": "GPU board only; not whole-server AC energy",
          "time_weighted_mean_watts": 299.9702381997644,
          "trapezoid_pieces": 78,
          "watt_hours": 5.770420675520126
        },
        "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": 17240788992,
            "mean": 17231835497.411766,
            "median": 17233764352.0,
            "minimum": 17221709824,
            "sample_count": 68
          },
          "gpu_board_watts": {
            "maximum": 301.79,
            "mean": 299.9689705882353,
            "median": 300.0,
            "minimum": 297.03,
            "sample_count": 68
          },
          "gpu_memory_used_bytes": {
            "maximum": 22502440960.0,
            "mean": 22502440960.0,
            "median": 22502440960.0,
            "minimum": 22502440960.0,
            "sample_count": 68
          },
          "gpu_sm_clock_mhz": {
            "maximum": 1687.0,
            "mean": 1608.3088235294117,
            "median": 1605.0,
            "minimum": 1575.0,
            "sample_count": 68
          },
          "gpu_utilization_percent": {
            "maximum": 100.0,
            "mean": 99.02941176470588,
            "median": 99.0,
            "minimum": 84.0,
            "sample_count": 68
          },
          "host_mem_available_bytes": {
            "maximum": 96136097792,
            "mean": 96108860235.29411,
            "median": 96106113024.0,
            "minimum": 96084488192,
            "sample_count": 68
          },
          "process_pss_bytes": {
            "maximum": 925034496,
            "mean": 919076412.2352941,
            "median": 921339904.0,
            "minimum": 911089664,
            "sample_count": 68
          },
          "process_rss_bytes": {
            "maximum": 928030720,
            "mean": 922093507.7647059,
            "median": 924336128.0,
            "minimum": 915148800,
            "sample_count": 68
          }
        },
        "process_sample_count": 68,
        "response_union_seconds": 69.25191831210395,
        "sampled_counter_deltas": {
          "deltas": {
            "major_faults": 0,
            "read_syscall_bytes": 0,
            "storage_read_bytes": 0,
            "storage_write_bytes": 45056
          },
          "observed_endpoint_seconds": 59.35470441589132,
          "observed_intervals": 10,
          "requested_intervals": 10
        }
      },
      "groups": [
        {
          "aggregate_output_tokens_per_second": 46.274918602713925,
          "completed_output_tokens": 1280,
          "completed_request_count": 5,
          "context_tokens": 2048,
          "cpu_snapshot_intervals": {
            "host_capacity_percent": 3.1198924303190445,
            "interval_count": 5,
            "interval_seconds": 27.675233017944265,
            "meaning": "Weighted CPU counter rate across round-bracketing snapshots, not exact token time.",
            "one_logical_cpu_percent": 99.83655777020942,
            "process_cpu_seconds": 27.63
          },
          "failed_request_count": 0,
          "measured_rounds": 5,
          "mode": "completion",
          "resources": {
            "gpu_energy": {
              "available": true,
              "interpolation": "linear_at_bracketed_boundaries",
              "joules": 8298.985135398982,
              "reason": null,
              "response_union_seconds": 27.660772588045802,
              "scope": "GPU board only; not whole-server AC energy",
              "time_weighted_mean_watts": 300.02723564509427,
              "trapezoid_pieces": 32,
              "watt_hours": 2.3052736487219394
            },
            "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": 17240031232,
                "mean": 17223894964.148148,
                "median": 17223282688,
                "minimum": 17221709824,
                "sample_count": 27
              },
              "gpu_board_watts": {
                "maximum": 301.79,
                "mean": 300.0274074074074,
                "median": 299.99,
                "minimum": 299.24,
                "sample_count": 27
              },
              "gpu_memory_used_bytes": {
                "maximum": 22502440960.0,
                "mean": 22502440960.0,
                "median": 22502440960.0,
                "minimum": 22502440960.0,
                "sample_count": 27
              },
              "gpu_sm_clock_mhz": {
                "maximum": 1687.0,
                "mean": 1607.5925925925926,
                "median": 1605.0,
                "minimum": 1575.0,
                "sample_count": 27
              },
              "gpu_utilization_percent": {
                "maximum": 100.0,
                "mean": 99.0,
                "median": 99.0,
                "minimum": 95.0,
                "sample_count": 27
              },
              "host_mem_available_bytes": {
                "maximum": 96136097792,
                "mean": 96113169445.92592,
                "median": 96115363840,
                "minimum": 96087658496,
                "sample_count": 27
              },
              "process_pss_bytes": {
                "maximum": 913487872,
                "mean": 912757797.925926,
                "median": 912799744,
                "minimum": 911089664,
                "sample_count": 27
              },
              "process_rss_bytes": {
                "maximum": 916484096,
                "mean": 915806587.2592592,
                "median": 915795968,
                "minimum": 915148800,
                "sample_count": 27
              }
            },
            "process_sample_count": 27,
            "response_union_seconds": 27.660772588045802,
            "sampled_counter_deltas": {
              "deltas": {
                "major_faults": 0,
                "read_syscall_bytes": 0,
                "storage_read_bytes": 0,
                "storage_write_bytes": 20480
              },
              "observed_endpoint_seconds": 22.504645432985853,
              "observed_intervals": 5,
              "requested_intervals": 5
            }
          },
          "response_outer_span_seconds": 27.68297484901268,
          "response_seconds": {
            "maximum": 5.5716617869911715,
            "mean": 5.53215451760916,
            "median": 5.527461774996482,
            "minimum": 5.496413041022606,
            "sample_count": 5
          },
          "response_window_union_seconds": 27.660772588045802,
          "server_decode_tokens_per_second": {
            "maximum": 53.86729127375231,
            "mean": 53.507414673402344,
            "median": 53.55129058610313,
            "minimum": 53.12845334946771,
            "sample_count": 5
          },
          "synthetic_first_token_seconds": {
            "maximum": 0.7718379290308803,
            "mean": 0.7662811154150404,
            "median": 0.7656330339959823,
            "minimum": 0.7624987220042385,
            "sample_count": 5
          },
          "warmup_completed_rounds": 1,
          "warmup_requests_per_round": 1
        },
        {
          "aggregate_output_tokens_per_second": 30.775781184108897,
          "completed_output_tokens": 1280,
          "completed_request_count": 5,
          "context_tokens": 8192,
          "cpu_snapshot_intervals": {
            "host_capacity_percent": 3.1204527047148796,
            "interval_count": 5,
            "interval_seconds": 41.610548944969196,
            "meaning": "Weighted CPU counter rate across round-bracketing snapshots, not exact token time.",
            "one_logical_cpu_percent": 99.85448655087615,
            "process_cpu_seconds": 41.55
          },
          "failed_request_count": 0,
          "measured_rounds": 5,
          "mode": "completion",
          "resources": {
            "gpu_energy": {
              "available": true,
              "interpolation": "linear_at_bracketed_boundaries",
              "joules": 12474.529296473467,
              "reason": null,
              "response_union_seconds": 41.59114572405815,
              "scope": "GPU board only; not whole-server AC energy",
              "time_weighted_mean_watts": 299.93233125236196,
              "trapezoid_pieces": 46,
              "watt_hours": 3.4651470267981854
            },
            "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": 17240788992,
                "mean": 17237064629.07317,
                "median": 17237659648,
                "minimum": 17232809984,
                "sample_count": 41
              },
              "gpu_board_watts": {
                "maximum": 300.28,
                "mean": 299.93048780487806,
                "median": 300.0,
                "minimum": 297.03,
                "sample_count": 41
              },
              "gpu_memory_used_bytes": {
                "maximum": 22502440960.0,
                "mean": 22502440960.0,
                "median": 22502440960.0,
                "minimum": 22502440960.0,
                "sample_count": 41
              },
              "gpu_sm_clock_mhz": {
                "maximum": 1657.0,
                "mean": 1608.780487804878,
                "median": 1605.0,
                "minimum": 1590.0,
                "sample_count": 41
              },
              "gpu_utilization_percent": {
                "maximum": 100.0,
                "mean": 99.04878048780488,
                "median": 99.0,
                "minimum": 84.0,
                "sample_count": 41
              },
              "host_mem_available_bytes": {
                "maximum": 96128155648,
                "mean": 96106022462.43903,
                "median": 96104517632,
                "minimum": 96084488192,
                "sample_count": 41
              },
              "process_pss_bytes": {
                "maximum": 925034496,
                "mean": 923237450.9268292,
                "median": 923604992,
                "minimum": 920840192,
                "sample_count": 41
              },
              "process_rss_bytes": {
                "maximum": 928030720,
                "mean": 926233674.9268292,
                "median": 926601216,
                "minimum": 923836416,
                "sample_count": 41
              }
            },
            "process_sample_count": 41,
            "response_union_seconds": 41.59114572405815,
            "sampled_counter_deltas": {
              "deltas": {
                "major_faults": 0,
                "read_syscall_bytes": 0,
                "storage_read_bytes": 0,
                "storage_write_bytes": 24576
              },
              "observed_endpoint_seconds": 36.850058982905466,
              "observed_intervals": 5,
              "requested_intervals": 5
            }
          },
          "response_outer_span_seconds": 41.62048171600327,
          "response_seconds": {
            "maximum": 8.362828276003711,
            "mean": 8.31822914481163,
            "median": 8.326255902997218,
            "minimum": 8.262266772042494,
            "sample_count": 5
          },
          "response_window_union_seconds": 41.59114572405815,
          "server_decode_tokens_per_second": {
            "maximum": 51.867399264866066,
            "mean": 51.518764345880015,
            "median": 51.44895386118346,
            "minimum": 51.2462896681451,
            "sample_count": 5
          },
          "synthetic_first_token_seconds": {
            "maximum": 3.3867225820431486,
            "mean": 3.3683619036222807,
            "median": 3.3697669280227274,
            "minimum": 3.345733996015042,
            "sample_count": 5
          },
          "warmup_completed_rounds": 1,
          "warmup_requests_per_round": 1
        }
      ],
      "id": "baseline-32b",
      "model": "qwen3-32b-q4",
      "module": "baseline",
      "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": "31d7b3fec56be9142f89e611f4085991a36d0edab0bd028a37a8232b8093b11b",
      "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": 16632758272,
              "mean": 16624463872,
              "median": 16624463872.0,
              "minimum": 16616169472,
              "sample_count": 2
            },
            "gpu_board_watts": {
              "maximum": 16.33,
              "mean": 16.21,
              "median": 16.21,
              "minimum": 16.09,
              "sample_count": 2
            },
            "gpu_memory_used_bytes": {
              "maximum": 2097152.0,
              "mean": 2097152.0,
              "median": 2097152.0,
              "minimum": 2097152.0,
              "sample_count": 2
            },
            "gpu_sm_clock_mhz": {
              "maximum": 180.0,
              "mean": 180.0,
              "median": 180.0,
              "minimum": 180.0,
              "sample_count": 2
            },
            "gpu_utilization_percent": {
              "maximum": 0.0,
              "mean": 0.0,
              "median": 0.0,
              "minimum": 0.0,
              "sample_count": 2
            },
            "host_mem_available_bytes": {
              "maximum": 96517685248,
              "mean": 96516294656,
              "median": 96516294656.0,
              "minimum": 96514904064,
              "sample_count": 2
            },
            "process_pss_bytes": null,
            "process_rss_bytes": null
          },
          "process_sample_count": 0,
          "response_union_seconds": 1.1007562470040284,
          "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": 16893157376,
              "mean": 16893157376,
              "median": 16893157376,
              "minimum": 16893157376,
              "sample_count": 1
            },
            "gpu_board_watts": {
              "maximum": 94.6,
              "mean": 94.6,
              "median": 94.6,
              "minimum": 94.6,
              "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": 96570462208,
              "mean": 96570462208,
              "median": 96570462208,
              "minimum": 96570462208,
              "sample_count": 1
            },
            "process_pss_bytes": null,
            "process_rss_bytes": null
          },
          "process_sample_count": 0,
          "response_union_seconds": 1.1624215979827568,
          "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": 234.14076826269599,
            "reason": null,
            "response_union_seconds": 3.614590512996074,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 64.77656802917369,
            "trapezoid_pieces": 4,
            "watt_hours": 0.06503910229519333
          },
          "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": 17189969920,
              "mean": 17152371370.666666,
              "median": 17133707264,
              "minimum": 17133436928,
              "sample_count": 3
            },
            "gpu_board_watts": {
              "maximum": 78.92,
              "mean": 70.08,
              "median": 78.4,
              "minimum": 52.92,
              "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": 76.0,
              "mean": 71.0,
              "median": 76.0,
              "minimum": 61.0,
              "sample_count": 3
            },
            "host_mem_available_bytes": {
              "maximum": 96221175808,
              "mean": 96199781034.66667,
              "median": 96212836352,
              "minimum": 96165330944,
              "sample_count": 3
            },
            "process_pss_bytes": {
              "maximum": 20099049472,
              "mean": 13691381760,
              "median": 20098906112,
              "minimum": 876189696,
              "sample_count": 3
            },
            "process_rss_bytes": {
              "maximum": 20102045696,
              "mean": 13694377984,
              "median": 20101902336,
              "minimum": 879185920,
              "sample_count": 3
            }
          },
          "process_sample_count": 3,
          "response_union_seconds": 3.614590512996074,
          "sampled_counter_deltas": {
            "deltas": {
              "major_faults": 0,
              "read_syscall_bytes": 13,
              "storage_read_bytes": 0,
              "storage_write_bytes": 8192
            },
            "observed_endpoint_seconds": 2.344856615003664,
            "observed_intervals": 1,
            "requested_intervals": 1
          }
        },
        "ready": {
          "gpu_energy": {
            "available": true,
            "interpolation": "linear_at_bracketed_boundaries",
            "joules": 548.1502251793638,
            "reason": null,
            "response_union_seconds": 3.31754442199599,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 165.22769719224166,
            "trapezoid_pieces": 8,
            "watt_hours": 0.15226395143871216
          },
          "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": 17240248320,
              "mean": 17218326528,
              "median": 17224609792,
              "minimum": 17190121472,
              "sample_count": 3
            },
            "gpu_board_watts": {
              "maximum": 300.01,
              "mean": 183.016,
              "median": 115.73,
              "minimum": 87.89,
              "sample_count": 5
            },
            "gpu_memory_used_bytes": {
              "maximum": 22502440960.0,
              "mean": 22497407795.2,
              "median": 22502440960.0,
              "minimum": 22477275136.0,
              "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": 99.0,
              "mean": 39.6,
              "median": 0.0,
              "minimum": 0.0,
              "sample_count": 5
            },
            "host_mem_available_bytes": {
              "maximum": 96170061824,
              "mean": 96126780757.33333,
              "median": 96106954752,
              "minimum": 96103325696,
              "sample_count": 3
            },
            "process_pss_bytes": {
              "maximum": 925034496,
              "mean": 904964096,
              "median": 913561600,
              "minimum": 876296192,
              "sample_count": 3
            },
            "process_rss_bytes": {
              "maximum": 928030720,
              "mean": 907960320,
              "median": 916557824,
              "minimum": 879292416,
              "sample_count": 3
            }
          },
          "process_sample_count": 3,
          "response_union_seconds": 3.31754442199599,
          "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": 20792.77551286387,
            "reason": null,
            "response_union_seconds": 69.31620680802735,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 299.9698983882642,
            "trapezoid_pieces": 70,
            "watt_hours": 5.77577097579552
          },
          "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": 17240788992,
              "mean": 17231835497.411766,
              "median": 17233764352.0,
              "minimum": 17221709824,
              "sample_count": 68
            },
            "gpu_board_watts": {
              "maximum": 301.79,
              "mean": 299.9689705882353,
              "median": 300.0,
              "minimum": 297.03,
              "sample_count": 68
            },
            "gpu_memory_used_bytes": {
              "maximum": 22502440960.0,
              "mean": 22502440960.0,
              "median": 22502440960.0,
              "minimum": 22502440960.0,
              "sample_count": 68
            },
            "gpu_sm_clock_mhz": {
              "maximum": 1687.0,
              "mean": 1608.3088235294117,
              "median": 1605.0,
              "minimum": 1575.0,
              "sample_count": 68
            },
            "gpu_utilization_percent": {
              "maximum": 100.0,
              "mean": 99.02941176470588,
              "median": 99.0,
              "minimum": 84.0,
              "sample_count": 68
            },
            "host_mem_available_bytes": {
              "maximum": 96136097792,
              "mean": 96108860235.29411,
              "median": 96106113024.0,
              "minimum": 96084488192,
              "sample_count": 68
            },
            "process_pss_bytes": {
              "maximum": 925034496,
              "mean": 919076412.2352941,
              "median": 921339904.0,
              "minimum": 911089664,
              "sample_count": 68
            },
            "process_rss_bytes": {
              "maximum": 928030720,
              "mean": 922093507.7647059,
              "median": 924336128.0,
              "minimum": 915148800,
              "sample_count": 68
            }
          },
          "process_sample_count": 68,
          "response_union_seconds": 69.31620680802735,
          "sampled_counter_deltas": {
            "deltas": {
              "major_faults": 0,
              "read_syscall_bytes": 0,
              "storage_read_bytes": 0,
              "storage_write_bytes": 65536
            },
            "observed_endpoint_seconds": 67.56908865296282,
            "observed_intervals": 2,
            "requested_intervals": 2
          }
        },
        "warmup": {
          "gpu_energy": {
            "available": true,
            "interpolation": "linear_at_bracketed_boundaries",
            "joules": 3891.3219742636566,
            "reason": null,
            "response_union_seconds": 13.74247567699058,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 283.1601863977834,
            "trapezoid_pieces": 16,
            "watt_hours": 1.0809227706287934
          },
          "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": 17231908864,
              "mean": 17226236534.153847,
              "median": 17225515008,
              "minimum": 17220960256,
              "sample_count": 13
            },
            "gpu_board_watts": {
              "maximum": 306.92,
              "mean": 284.82928571428573,
              "median": 299.99,
              "minimum": 106.16,
              "sample_count": 14
            },
            "gpu_memory_used_bytes": {
              "maximum": 22502440960.0,
              "mean": 22502440960.0,
              "median": 22502440960.0,
              "minimum": 22502440960.0,
              "sample_count": 14
            },
            "gpu_sm_clock_mhz": {
              "maximum": 2197.0,
              "mean": 1657.7142857142858,
              "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": 96137310208,
              "mean": 96114068558.76923,
              "median": 96116252672,
              "minimum": 96089333760,
              "sample_count": 13
            },
            "process_pss_bytes": {
              "maximum": 920127488,
              "mean": 915585339.0769231,
              "median": 914270208,
              "minimum": 911812608,
              "sample_count": 13
            },
            "process_rss_bytes": {
              "maximum": 923123712,
              "mean": 918581563.0769231,
              "median": 917266432,
              "minimum": 914808832,
              "sample_count": 13
            }
          },
          "process_sample_count": 13,
          "response_union_seconds": 13.74247567699058,
          "sampled_counter_deltas": {
            "deltas": {
              "major_faults": 8,
              "read_syscall_bytes": 0,
              "storage_read_bytes": 1372160,
              "storage_write_bytes": 12288
            },
            "observed_endpoint_seconds": 11.253299642936327,
            "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": 1,
        "synthetic_prompt_sha256": "c099b5abc034050fac1133e0214990d1f72a124b64f36e8a80ba3209bc07d187",
        "temperature": 0,
        "threads": 16,
        "tokenization": "model_tokenizer_add_special_false_repeat_then_truncate",
        "total_context_tokens": 8704,
        "ubatch": 512,
        "warmup_rounds": 1
      },
      "status": "completed",
      "telemetry": {
        "gpu_samples": 93,
        "host_samples": 90,
        "maximum_gpu_gap_seconds": 1.0044987400178798,
        "maximum_host_gap_seconds": 1.322291104006581,
        "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": 185.7723383929697,
      "completed_response_resources": {
        "gpu_energy": {
          "available": true,
          "interpolation": "linear_at_bracketed_boundaries",
          "joules": 43316.62505863769,
          "reason": null,
          "response_union_seconds": 144.44773938611615,
          "scope": "GPU board only; not whole-server AC energy",
          "time_weighted_mean_watts": 299.8774867833006,
          "trapezoid_pieces": 155,
          "watt_hours": 12.032395849621581
        },
        "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": 783441920,
            "mean": 744136242.4788733,
            "median": 753983488.0,
            "minimum": 708935680,
            "sample_count": 142
          },
          "gpu_board_watts": {
            "maximum": 302.93,
            "mean": 300.00172413793103,
            "median": 300.0,
            "minimum": 297.31,
            "sample_count": 145
          },
          "gpu_memory_used_bytes": {
            "maximum": 47104131072.0,
            "mean": 47104131072.0,
            "median": 47104131072.0,
            "minimum": 47104131072.0,
            "sample_count": 145
          },
          "gpu_sm_clock_mhz": {
            "maximum": 1635.0,
            "mean": 1481.2551724137932,
            "median": 1485.0,
            "minimum": 1312.0,
            "sample_count": 145
          },
          "gpu_utilization_percent": {
            "maximum": 100.0,
            "mean": 99.84827586206896,
            "median": 100.0,
            "minimum": 83.0,
            "sample_count": 145
          },
          "host_mem_available_bytes": {
            "maximum": 96093003776,
            "mean": 96040983393.35211,
            "median": 96037416960.0,
            "minimum": 96006701056,
            "sample_count": 142
          },
          "process_pss_bytes": {
            "maximum": 1257277440,
            "mean": 1236598834.4788733,
            "median": 1246618624.0,
            "minimum": 1205324800,
            "sample_count": 142
          },
          "process_rss_bytes": {
            "maximum": 1260240896,
            "mean": 1239593392.6760564,
            "median": 1249583104.0,
            "minimum": 1209503744,
            "sample_count": 142
          }
        },
        "process_sample_count": 142,
        "response_union_seconds": 144.44773938611615,
        "sampled_counter_deltas": {
          "deltas": {
            "major_faults": 0,
            "read_syscall_bytes": 0,
            "storage_read_bytes": 0,
            "storage_write_bytes": 102400
          },
          "observed_endpoint_seconds": 135.37246441596653,
          "observed_intervals": 10,
          "requested_intervals": 10
        }
      },
      "groups": [
        {
          "aggregate_output_tokens_per_second": 22.20832513756411,
          "completed_output_tokens": 1280,
          "completed_request_count": 5,
          "context_tokens": 2048,
          "cpu_snapshot_intervals": {
            "host_capacity_percent": 3.1211448738400773,
            "interval_count": 5,
            "interval_seconds": 57.6511207500007,
            "meaning": "Weighted CPU counter rate across round-bracketing snapshots, not exact token time.",
            "one_logical_cpu_percent": 99.87663596288247,
            "process_cpu_seconds": 57.58
          },
          "failed_request_count": 0,
          "measured_rounds": 5,
          "mode": "completion",
          "resources": {
            "gpu_energy": {
              "available": true,
              "interpolation": "linear_at_bracketed_boundaries",
              "joules": 17282.92564468072,
              "reason": null,
              "response_union_seconds": 57.63604378409218,
              "scope": "GPU board only; not whole-server AC energy",
              "time_weighted_mean_watts": 299.86315003547986,
              "trapezoid_pieces": 63,
              "watt_hours": 4.8008126790779775
            },
            "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": 748466176,
                "mean": 716765866.6666666,
                "median": 711774208,
                "minimum": 708935680,
                "sample_count": 57
              },
              "gpu_board_watts": {
                "maximum": 300.95,
                "mean": 300.01241379310346,
                "median": 299.99,
                "minimum": 299.66,
                "sample_count": 58
              },
              "gpu_memory_used_bytes": {
                "maximum": 47104131072.0,
                "mean": 47104131072.0,
                "median": 47104131072.0,
                "minimum": 47104131072.0,
                "sample_count": 58
              },
              "gpu_sm_clock_mhz": {
                "maximum": 1627.0,
                "mean": 1446.4827586206898,
                "median": 1451.0,
                "minimum": 1312.0,
                "sample_count": 58
              },
              "gpu_utilization_percent": {
                "maximum": 100.0,
                "mean": 100.0,
                "median": 100.0,
                "minimum": 100.0,
                "sample_count": 58
              },
              "host_mem_available_bytes": {
                "maximum": 96093003776,
                "mean": 96060373171.64912,
                "median": 96058638336,
                "minimum": 96031748096,
                "sample_count": 57
              },
              "process_pss_bytes": {
                "maximum": 1225331712,
                "mean": 1211821738.6666667,
                "median": 1207718912,
                "minimum": 1205324800,
                "sample_count": 57
              },
              "process_rss_bytes": {
                "maximum": 1228296192,
                "mean": 1214836376.7017543,
                "median": 1210683392,
                "minimum": 1209503744,
                "sample_count": 57
              }
            },
            "process_sample_count": 57,
            "response_union_seconds": 57.63604378409218,
            "sampled_counter_deltas": {
              "deltas": {
                "major_faults": 0,
                "read_syscall_bytes": 0,
                "storage_read_bytes": 0,
                "storage_write_bytes": 40960
              },
              "observed_endpoint_seconds": 53.343125255953055,
              "observed_intervals": 5,
              "requested_intervals": 5
            }
          },
          "response_outer_span_seconds": 57.6594875259907,
          "response_seconds": {
            "maximum": 11.647003977035638,
            "mean": 11.527208756818435,
            "median": 11.566605183004867,
            "minimum": 11.37437688402133,
            "sample_count": 5
          },
          "response_window_union_seconds": 57.63604378409218,
          "server_decode_tokens_per_second": {
            "maximum": 26.211279772416194,
            "mean": 25.85845974942399,
            "median": 25.778169650885744,
            "minimum": 25.605412361610277,
            "sample_count": 5
          },
          "synthetic_first_token_seconds": {
            "maximum": 1.6881801380077377,
            "mean": 1.665065498405602,
            "median": 1.6596780359977856,
            "minimum": 1.645603878016118,
            "sample_count": 5
          },
          "warmup_completed_rounds": 1,
          "warmup_requests_per_round": 1
        },
        {
          "aggregate_output_tokens_per_second": 14.7445570683008,
          "completed_output_tokens": 1280,
          "completed_request_count": 5,
          "context_tokens": 8192,
          "cpu_snapshot_intervals": {
            "host_capacity_percent": 3.1216834193859877,
            "interval_count": 5,
            "interval_seconds": 86.83215546992142,
            "meaning": "Weighted CPU counter rate across round-bracketing snapshots, not exact token time.",
            "one_logical_cpu_percent": 99.8938694203516,
            "process_cpu_seconds": 86.74000000000001
          },
          "failed_request_count": 0,
          "measured_rounds": 5,
          "mode": "completion",
          "resources": {
            "gpu_energy": {
              "available": true,
              "interpolation": "linear_at_bracketed_boundaries",
              "joules": 26033.69941395698,
              "reason": null,
              "response_union_seconds": 86.81169560202397,
              "scope": "GPU board only; not whole-server AC energy",
              "time_weighted_mean_watts": 299.88700524068577,
              "trapezoid_pieces": 92,
              "watt_hours": 7.2315831705436056
            },
            "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": 783441920,
                "mean": 762490494.4941176,
                "median": 764178432,
                "minimum": 749699072,
                "sample_count": 85
              },
              "gpu_board_watts": {
                "maximum": 302.93,
                "mean": 299.9945977011494,
                "median": 300.0,
                "minimum": 297.31,
                "sample_count": 87
              },
              "gpu_memory_used_bytes": {
                "maximum": 47104131072.0,
                "mean": 47104131072.0,
                "median": 47104131072.0,
                "minimum": 47104131072.0,
                "sample_count": 87
              },
              "gpu_sm_clock_mhz": {
                "maximum": 1635.0,
                "mean": 1504.4367816091954,
                "median": 1485.0,
                "minimum": 1462.0,
                "sample_count": 87
              },
              "gpu_utilization_percent": {
                "maximum": 100.0,
                "mean": 99.74712643678161,
                "median": 100.0,
                "minimum": 83.0,
                "sample_count": 87
              },
              "host_mem_available_bytes": {
                "maximum": 96051191808,
                "mean": 96027980836.14117,
                "median": 96027217920,
                "minimum": 96006701056,
                "sample_count": 85
              },
              "process_pss_bytes": {
                "maximum": 1257277440,
                "mean": 1253214063.4352942,
                "median": 1255912448,
                "minimum": 1243612160,
                "sample_count": 85
              },
              "process_rss_bytes": {
                "maximum": 1260240896,
                "mean": 1256195156.3294117,
                "median": 1258876928,
                "minimum": 1246576640,
                "sample_count": 85
              }
            },
            "process_sample_count": 85,
            "response_union_seconds": 86.81169560202397,
            "sampled_counter_deltas": {
              "deltas": {
                "major_faults": 0,
                "read_syscall_bytes": 0,
                "storage_read_bytes": 0,
                "storage_write_bytes": 61440
              },
              "observed_endpoint_seconds": 82.02933916001348,
              "observed_intervals": 5,
              "requested_intervals": 5
            }
          },
          "response_outer_span_seconds": 86.84236659400631,
          "response_seconds": {
            "maximum": 17.491869097051676,
            "mean": 17.362339120404794,
            "median": 17.35889967199182,
            "minimum": 17.22252958599711,
            "sample_count": 5
          },
          "response_window_union_seconds": 86.81169560202397,
          "server_decode_tokens_per_second": {
            "maximum": 25.05988575640317,
            "mean": 24.836448251688132,
            "median": 24.82518929206835,
            "minimum": 24.64976311577646,
            "sample_count": 5
          },
          "synthetic_first_token_seconds": {
            "maximum": 7.146799207024742,
            "mean": 7.094700342812575,
            "median": 7.08692858600989,
            "minimum": 7.046769339009188,
            "sample_count": 5
          },
          "warmup_completed_rounds": 1,
          "warmup_requests_per_round": 1
        }
      ],
      "id": "offload-72b-full",
      "model": "qwen25-72b-q4",
      "module": "dense-offload",
      "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": "f8b20b521b99946cae8150a28b03c57e74bb9a19fc93a5809019ef9e4d3899c0",
      "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": 358690816,
              "mean": 349544448,
              "median": 349544448.0,
              "minimum": 340398080,
              "sample_count": 2
            },
            "gpu_board_watts": {
              "maximum": 15.22,
              "mean": 14.965,
              "median": 14.965,
              "minimum": 14.71,
              "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": 96567312384,
              "mean": 96558721024,
              "median": 96558721024.0,
              "minimum": 96550129664,
              "sample_count": 2
            },
            "process_pss_bytes": null,
            "process_rss_bytes": null
          },
          "process_sample_count": 0,
          "response_union_seconds": 1.1008167749969289,
          "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": 352747520,
              "mean": 352747520,
              "median": 352747520,
              "minimum": 352747520,
              "sample_count": 1
            },
            "gpu_board_watts": {
              "maximum": 103.77,
              "mean": 103.77,
              "median": 103.77,
              "minimum": 103.77,
              "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": 96511545344,
              "mean": 96511545344,
              "median": 96511545344,
              "minimum": 96511545344,
              "sample_count": 1
            },
            "process_pss_bytes": null,
            "process_rss_bytes": null
          },
          "process_sample_count": 0,
          "response_union_seconds": 1.1564114179927856,
          "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": 478.07168217720226,
            "reason": null,
            "response_union_seconds": 7.210961889009923,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 66.29790720511522,
            "trapezoid_pieces": 8,
            "watt_hours": 0.1327976894936673
          },
          "gpu_sample_count": 7,
          "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": 656150528,
              "mean": 638523392,
              "median": 638926848.0,
              "minimum": 620089344,
              "sample_count": 4
            },
            "gpu_board_watts": {
              "maximum": 74.94,
              "mean": 69.43428571428572,
              "median": 74.31,
              "minimum": 42.68,
              "sample_count": 7
            },
            "gpu_memory_used_bytes": {
              "maximum": 43891294208.0,
              "mean": 37704995401.14286,
              "median": 43891294208.0,
              "minimum": 587202560.0,
              "sample_count": 7
            },
            "gpu_sm_clock_mhz": {
              "maximum": 2295.0,
              "mean": 2291.714285714286,
              "median": 2295.0,
              "minimum": 2272.0,
              "sample_count": 7
            },
            "gpu_utilization_percent": {
              "maximum": 83.0,
              "mean": 58.857142857142854,
              "median": 82.0,
              "minimum": 0.0,
              "sample_count": 7
            },
            "host_mem_available_bytes": {
              "maximum": 96185810944,
              "mean": 96155684864,
              "median": 96151988224.0,
              "minimum": 96132952064,
              "sample_count": 4
            },
            "process_pss_bytes": {
              "maximum": 44348953600,
              "mean": 42197760000,
              "median": 44348720128.0,
              "minimum": 35744646144,
              "sample_count": 4
            },
            "process_rss_bytes": {
              "maximum": 44351918080,
              "mean": 42200724480,
              "median": 44351684608.0,
              "minimum": 35747610624,
              "sample_count": 4
            }
          },
          "process_sample_count": 4,
          "response_union_seconds": 7.210961889009923,
          "sampled_counter_deltas": {
            "deltas": {
              "major_faults": 0,
              "read_syscall_bytes": 5,
              "storage_read_bytes": 0,
              "storage_write_bytes": 0
            },
            "observed_endpoint_seconds": 5.209772021975368,
            "observed_intervals": 1,
            "requested_intervals": 1
          }
        },
        "ready": {
          "gpu_energy": {
            "available": true,
            "interpolation": "linear_at_bracketed_boundaries",
            "joules": 551.952217391011,
            "reason": null,
            "response_union_seconds": 3.3186683430103585,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 166.3173780391493,
            "trapezoid_pieces": 6,
            "watt_hours": 0.15332006038639195
          },
          "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": 769622016,
              "mean": 725244586.6666666,
              "median": 737775616,
              "minimum": 668336128,
              "sample_count": 3
            },
            "gpu_board_watts": {
              "maximum": 205.83,
              "mean": 158.66,
              "median": 175.53,
              "minimum": 94.62,
              "sample_count": 3
            },
            "gpu_memory_used_bytes": {
              "maximum": 47104131072.0,
              "mean": 47093645312.0,
              "median": 47104131072.0,
              "minimum": 47072673792.0,
              "sample_count": 3
            },
            "gpu_sm_clock_mhz": {
              "maximum": 2340.0,
              "mean": 2314.6666666666665,
              "median": 2332.0,
              "minimum": 2272.0,
              "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": 96088260608,
              "mean": 96051089408,
              "median": 96037539840,
              "minimum": 96027467776,
              "sample_count": 3
            },
            "process_pss_bytes": {
              "maximum": 1257276416,
              "mean": 1216586752,
              "median": 1231762432,
              "minimum": 1160721408,
              "sample_count": 3
            },
            "process_rss_bytes": {
              "maximum": 1260240896,
              "mean": 1219551232,
              "median": 1234726912,
              "minimum": 1163685888,
              "sample_count": 3
            }
          },
          "process_sample_count": 3,
          "response_union_seconds": 3.3186683430103585,
          "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": 43336.282681437755,
            "reason": null,
            "response_union_seconds": 144.51526583201485,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 299.87339006670777,
            "trapezoid_pieces": 147,
            "watt_hours": 12.037856300399376
          },
          "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": 783441920,
              "mean": 744136242.4788733,
              "median": 753983488.0,
              "minimum": 708935680,
              "sample_count": 142
            },
            "gpu_board_watts": {
              "maximum": 302.93,
              "mean": 300.00172413793103,
              "median": 300.0,
              "minimum": 297.31,
              "sample_count": 145
            },
            "gpu_memory_used_bytes": {
              "maximum": 47104131072.0,
              "mean": 47104131072.0,
              "median": 47104131072.0,
              "minimum": 47104131072.0,
              "sample_count": 145
            },
            "gpu_sm_clock_mhz": {
              "maximum": 1635.0,
              "mean": 1481.2551724137932,
              "median": 1485.0,
              "minimum": 1312.0,
              "sample_count": 145
            },
            "gpu_utilization_percent": {
              "maximum": 100.0,
              "mean": 99.84827586206896,
              "median": 100.0,
              "minimum": 83.0,
              "sample_count": 145
            },
            "host_mem_available_bytes": {
              "maximum": 96093003776,
              "mean": 96040983393.35211,
              "median": 96037416960.0,
              "minimum": 96006701056,
              "sample_count": 142
            },
            "process_pss_bytes": {
              "maximum": 1257277440,
              "mean": 1236598834.4788733,
              "median": 1246618624.0,
              "minimum": 1205324800,
              "sample_count": 142
            },
            "process_rss_bytes": {
              "maximum": 1260240896,
              "mean": 1239593392.6760564,
              "median": 1249583104.0,
              "minimum": 1209503744,
              "sample_count": 142
            }
          },
          "process_sample_count": 142,
          "response_union_seconds": 144.51526583201485,
          "sampled_counter_deltas": {
            "deltas": {
              "major_faults": 0,
              "read_syscall_bytes": 0,
              "storage_read_bytes": 0,
              "storage_write_bytes": 122880
            },
            "observed_endpoint_seconds": 143.58244017901598,
            "observed_intervals": 2,
            "requested_intervals": 2
          }
        },
        "warmup": {
          "gpu_energy": {
            "available": true,
            "interpolation": "linear_at_bracketed_boundaries",
            "joules": 8285.867276863784,
            "reason": null,
            "response_union_seconds": 28.43168306199368,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 291.4307696381153,
            "trapezoid_pieces": 30,
            "watt_hours": 2.301629799128829
          },
          "gpu_sample_count": 28,
          "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": 748781568,
              "mean": 728878952.2962962,
              "median": 738770944,
              "minimum": 700723200,
              "sample_count": 27
            },
            "gpu_board_watts": {
              "maximum": 300.16,
              "mean": 292.4089285714286,
              "median": 300.0,
              "minimum": 186.04,
              "sample_count": 28
            },
            "gpu_memory_used_bytes": {
              "maximum": 47104131072.0,
              "mean": 47104131072.0,
              "median": 47104131072.0,
              "minimum": 47104131072.0,
              "sample_count": 28
            },
            "gpu_sm_clock_mhz": {
              "maximum": 1642.0,
              "mean": 1531.3214285714287,
              "median": 1511.0,
              "minimum": 1492.0,
              "sample_count": 28
            },
            "gpu_utilization_percent": {
              "maximum": 100.0,
              "mean": 100.0,
              "median": 100.0,
              "minimum": 100.0,
              "sample_count": 28
            },
            "host_mem_available_bytes": {
              "maximum": 96063094784,
              "mean": 96039131060.14815,
              "median": 96037965824,
              "minimum": 96024268800,
              "sample_count": 27
            },
            "process_pss_bytes": {
              "maximum": 1242903552,
              "mean": 1224316662.5185184,
              "median": 1232471040,
              "minimum": 1196975104,
              "sample_count": 27
            },
            "process_rss_bytes": {
              "maximum": 1245868032,
              "mean": 1227281142.5185184,
              "median": 1235435520,
              "minimum": 1199939584,
              "sample_count": 27
            }
          },
          "process_sample_count": 27,
          "response_union_seconds": 28.43168306199368,
          "sampled_counter_deltas": {
            "deltas": {
              "major_faults": 0,
              "read_syscall_bytes": 0,
              "storage_read_bytes": 0,
              "storage_write_bytes": 12288
            },
            "observed_endpoint_seconds": 25.652496571012307,
            "observed_intervals": 2,
            "requested_intervals": 2
          }
        }
      },
      "reviewed_export_sha256": "4b17dd2a9c39d7cab209a102bfdbd49f8324b54625bc01cbe7d9aeaf6b87dd5a",
      "settings": {
        "batch": 2048,
        "cache_prompt": false,
        "cache_ram_bytes": 0,
        "concurrency": 1,
        "context_shift": false,
        "contexts": [
          2048,
          8192
        ],
        "endpoint": "/completion",
        "fit": "off",
        "flash_attention": true,
        "ignore_eos": true,
        "kv_cache": "f16",
        "lazy_mode": "off",
        "load_mode": "mmap",
        "measured_rounds": 5,
        "mlock": false,
        "network_scope": "loopback",
        "output_tokens": 256,
        "p95_minimum_requests": 200,
        "placement": {
          "cpu_expert_layers": 0,
          "gpu_layers": "all"
        },
        "reasoning": "not_applicable_raw_completion",
        "seed": 20260927,
        "server_slots": 1,
        "synthetic_prompt_sha256": "c099b5abc034050fac1133e0214990d1f72a124b64f36e8a80ba3209bc07d187",
        "temperature": 0,
        "threads": 16,
        "tokenization": "model_tokenizer_add_special_false_repeat_then_truncate",
        "total_context_tokens": 8704,
        "ubatch": 512,
        "warmup_rounds": 1
      },
      "status": "completed",
      "telemetry": {
        "gpu_samples": 186,
        "host_samples": 179,
        "maximum_gpu_gap_seconds": 1.0065537570044398,
        "maximum_host_gap_seconds": 1.7706437300075777,
        "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": 186,
            "unavailable": 0
          },
          "software": {
            "active": 0,
            "not_active": 186,
            "unavailable": 0
          }
        }
      }
    },
    {
      "case_wall_seconds": 64.85315055400133,
      "completed_response_resources": {
        "gpu_energy": {
          "available": true,
          "interpolation": "linear_at_bracketed_boundaries",
          "joules": 5556.354650300507,
          "reason": null,
          "response_union_seconds": 18.690550074737985,
          "scope": "GPU board only; not whole-server AC energy",
          "time_weighted_mean_watts": 297.28149402143265,
          "trapezoid_pieces": 29,
          "watt_hours": 1.5434318473056965
        },
        "gpu_sample_count": 19,
        "host_sample_count": 18,
        "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": 84638580736,
            "mean": 84628920547.55556,
            "median": 84632049664.0,
            "minimum": 84617678848,
            "sample_count": 18
          },
          "gpu_board_watts": {
            "maximum": 303.42,
            "mean": 298.6694736842105,
            "median": 298.73,
            "minimum": 291.54,
            "sample_count": 19
          },
          "gpu_memory_used_bytes": {
            "maximum": 63942164480.0,
            "mean": 63942164480.0,
            "median": 63942164480.0,
            "minimum": 63942164480.0,
            "sample_count": 19
          },
          "gpu_sm_clock_mhz": {
            "maximum": 2212.0,
            "mean": 2094.2631578947367,
            "median": 2175.0,
            "minimum": 1837.0,
            "sample_count": 19
          },
          "gpu_utilization_percent": {
            "maximum": 100.0,
            "mean": 96.84210526315789,
            "median": 96.0,
            "minimum": 94.0,
            "sample_count": 19
          },
          "host_mem_available_bytes": {
            "maximum": 95839178752,
            "mean": 95816025201.77777,
            "median": 95815591936.0,
            "minimum": 95802257408,
            "sample_count": 18
          },
          "process_pss_bytes": {
            "maximum": 1353333760,
            "mean": 1348224227.5555556,
            "median": 1350933504.0,
            "minimum": 1341737984,
            "sample_count": 18
          },
          "process_rss_bytes": {
            "maximum": 1356292096,
            "mean": 1351182563.5555556,
            "median": 1353891840.0,
            "minimum": 1344696320,
            "sample_count": 18
          }
        },
        "process_sample_count": 18,
        "response_union_seconds": 18.690550074737985,
        "sampled_counter_deltas": {
          "deltas": {
            "major_faults": 0,
            "read_syscall_bytes": 0,
            "storage_read_bytes": 0,
            "storage_write_bytes": 4096
          },
          "observed_endpoint_seconds": 8.252795126929414,
          "observed_intervals": 7,
          "requested_intervals": 10
        }
      },
      "groups": [
        {
          "aggregate_output_tokens_per_second": 178.45760431710494,
          "completed_output_tokens": 1280,
          "completed_request_count": 5,
          "context_tokens": 2048,
          "cpu_snapshot_intervals": {
            "host_capacity_percent": 3.1128260975148874,
            "interval_count": 5,
            "interval_seconds": 7.1880019310628995,
            "meaning": "Weighted CPU counter rate across round-bracketing snapshots, not exact token time.",
            "one_logical_cpu_percent": 99.6104351204764,
            "process_cpu_seconds": 7.16
          },
          "failed_request_count": 0,
          "measured_rounds": 5,
          "mode": "completion",
          "resources": {
            "gpu_energy": {
              "available": true,
              "interpolation": "linear_at_bracketed_boundaries",
              "joules": 2130.4346352334737,
              "reason": null,
              "response_union_seconds": 7.172571910836268,
              "scope": "GPU board only; not whole-server AC energy",
              "time_weighted_mean_watts": 297.0252040296493,
              "trapezoid_pieces": 12,
              "watt_hours": 0.591787398675965
            },
            "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": 84621459456,
                "mean": 84619137024,
                "median": 84619370496,
                "minimum": 84617678848,
                "sample_count": 7
              },
              "gpu_board_watts": {
                "maximum": 300.9,
                "mean": 299.23285714285714,
                "median": 300.14,
                "minimum": 296.54,
                "sample_count": 7
              },
              "gpu_memory_used_bytes": {
                "maximum": 63942164480.0,
                "mean": 63942164480.0,
                "median": 63942164480.0,
                "minimum": 63942164480.0,
                "sample_count": 7
              },
              "gpu_sm_clock_mhz": {
                "maximum": 2205.0,
                "mean": 2170.5714285714284,
                "median": 2175.0,
                "minimum": 2077.0,
                "sample_count": 7
              },
              "gpu_utilization_percent": {
                "maximum": 96.0,
                "mean": 95.71428571428571,
                "median": 96.0,
                "minimum": 94.0,
                "sample_count": 7
              },
              "host_mem_available_bytes": {
                "maximum": 95839178752,
                "mean": 95822938697.14285,
                "median": 95823192064,
                "minimum": 95802257408,
                "sample_count": 7
              },
              "process_pss_bytes": {
                "maximum": 1342917632,
                "mean": 1342255250.2857144,
                "median": 1342327808,
                "minimum": 1341737984,
                "sample_count": 7
              },
              "process_rss_bytes": {
                "maximum": 1345875968,
                "mean": 1345213586.2857144,
                "median": 1345286144,
                "minimum": 1344696320,
                "sample_count": 7
              }
            },
            "process_sample_count": 7,
            "response_union_seconds": 7.172571910836268,
            "sampled_counter_deltas": {
              "deltas": {
                "major_faults": 0,
                "read_syscall_bytes": 0,
                "storage_read_bytes": 0,
                "storage_write_bytes": 4096
              },
              "observed_endpoint_seconds": 2.0736262949649245,
              "observed_intervals": 2,
              "requested_intervals": 5
            }
          },
          "response_outer_span_seconds": 7.20023208897328,
          "response_seconds": {
            "maximum": 1.4361235249671154,
            "mean": 1.4345143821672537,
            "median": 1.4348168549477123,
            "minimum": 1.4328763849916868,
            "sample_count": 5
          },
          "response_window_union_seconds": 7.172571910836268,
          "server_decode_tokens_per_second": {
            "maximum": 221.8905948843341,
            "mean": 221.75877497939115,
            "median": 221.6888167120332,
            "minimum": 221.6466170641819,
            "sample_count": 5
          },
          "synthetic_first_token_seconds": {
            "maximum": 0.2855251739965752,
            "mean": 0.28449933978263287,
            "median": 0.28445326996734366,
            "minimum": 0.28353577299276367,
            "sample_count": 5
          },
          "warmup_completed_rounds": 1,
          "warmup_requests_per_round": 1
        },
        {
          "aggregate_output_tokens_per_second": 111.13061526819216,
          "completed_output_tokens": 1280,
          "completed_request_count": 5,
          "context_tokens": 8192,
          "cpu_snapshot_intervals": {
            "host_capacity_percent": 3.1147665128873214,
            "interval_count": 5,
            "interval_seconds": 11.53778296103701,
            "meaning": "Weighted CPU counter rate across round-bracketing snapshots, not exact token time.",
            "one_logical_cpu_percent": 99.67252841239429,
            "process_cpu_seconds": 11.5
          },
          "failed_request_count": 0,
          "measured_rounds": 5,
          "mode": "completion",
          "resources": {
            "gpu_energy": {
              "available": true,
              "interpolation": "linear_at_bracketed_boundaries",
              "joules": 3425.920015067035,
              "reason": null,
              "response_union_seconds": 11.517978163901716,
              "scope": "GPU board only; not whole-server AC energy",
              "time_weighted_mean_watts": 297.44109307344826,
              "trapezoid_pieces": 17,
              "watt_hours": 0.951644448629732
            },
            "gpu_sample_count": 12,
            "host_sample_count": 11,
            "meaning": "Samples inside declared interval union. Means are arithmetic sample means; peaks need not coincide. Counter deltas exclude unsampled edges; not minimum required RAM.",
            "metrics": {
              "cgroup_memory_current_bytes": {
                "maximum": 84638580736,
                "mean": 84635146426.18182,
                "median": 84635049984,
                "minimum": 84631072768,
                "sample_count": 11
              },
              "gpu_board_watts": {
                "maximum": 303.42,
                "mean": 298.3408333333333,
                "median": 297.11,
                "minimum": 291.54,
                "sample_count": 12
              },
              "gpu_memory_used_bytes": {
                "maximum": 63942164480.0,
                "mean": 63942164480.0,
                "median": 63942164480.0,
                "minimum": 63942164480.0,
                "sample_count": 12
              },
              "gpu_sm_clock_mhz": {
                "maximum": 2212.0,
                "mean": 2049.75,
                "median": 2114.5,
                "minimum": 1837.0,
                "sample_count": 12
              },
              "gpu_utilization_percent": {
                "maximum": 100.0,
                "mean": 97.5,
                "median": 96.0,
                "minimum": 95.0,
                "sample_count": 12
              },
              "host_mem_available_bytes": {
                "maximum": 95829737472,
                "mean": 95811625704.72728,
                "median": 95808782336,
                "minimum": 95802593280,
                "sample_count": 11
              },
              "process_pss_bytes": {
                "maximum": 1353333760,
                "mean": 1352022667.6363637,
                "median": 1351961600,
                "minimum": 1350581248,
                "sample_count": 11
              },
              "process_rss_bytes": {
                "maximum": 1356292096,
                "mean": 1354981003.6363637,
                "median": 1354919936,
                "minimum": 1353539584,
                "sample_count": 11
              }
            },
            "process_sample_count": 11,
            "response_union_seconds": 11.517978163901716,
            "sampled_counter_deltas": {
              "deltas": {
                "major_faults": 0,
                "read_syscall_bytes": 0,
                "storage_read_bytes": 0,
                "storage_write_bytes": 0
              },
              "observed_endpoint_seconds": 6.179168831964489,
              "observed_intervals": 5,
              "requested_intervals": 5
            }
          },
          "response_outer_span_seconds": 11.548840367991943,
          "response_seconds": {
            "maximum": 2.306483986962121,
            "mean": 2.3035956327803433,
            "median": 2.3047796469763853,
            "minimum": 2.2995725619839504,
            "sample_count": 5
          },
          "response_window_union_seconds": 11.517978163901716,
          "server_decode_tokens_per_second": {
            "maximum": 215.594317441073,
            "mean": 215.5020711089472,
            "median": 215.50977357951592,
            "minimum": 215.41492293214347,
            "sample_count": 5
          },
          "synthetic_first_token_seconds": {
            "maximum": 1.122534021968022,
            "mean": 1.1201264659990557,
            "median": 1.1210324589628726,
            "minimum": 1.1166345790261403,
            "sample_count": 5
          },
          "warmup_completed_rounds": 1,
          "warmup_requests_per_round": 1
        }
      ],
      "id": "baseline-gpt120",
      "model": "gpt-oss-120b-mxfp4",
      "module": "baseline",
      "placement": {
        "buffers": [
          {
            "device": "CPU_Mapped",
            "reported_mib": 586.82
          },
          {
            "device": "CUDA0",
            "reported_mib": 59851.68
          }
        ],
        "gpu_layers": 37,
        "meaning": "Runtime buffer allocation is not measured resident process RAM.",
        "requested_cpu_expert_layers": 0,
        "total_layers": 37
      },
      "record_sha256": "a3433171db7e279c742e5d12ad982b7442dafada0598259e27b35e14652f685c",
      "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": 41687240704,
              "mean": 41678069760,
              "median": 41678069760.0,
              "minimum": 41668898816,
              "sample_count": 2
            },
            "gpu_board_watts": {
              "maximum": 15.04,
              "mean": 14.915,
              "median": 14.915,
              "minimum": 14.79,
              "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": 96659554304,
              "mean": 96650287104,
              "median": 96650287104.0,
              "minimum": 96641019904,
              "sample_count": 2
            },
            "process_pss_bytes": null,
            "process_rss_bytes": null
          },
          "process_sample_count": 0,
          "response_union_seconds": 1.1007708329707384,
          "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": 84053004288,
              "mean": 84053004288,
              "median": 84053004288,
              "minimum": 84053004288,
              "sample_count": 1
            },
            "gpu_board_watts": {
              "maximum": 86.59,
              "mean": 86.59,
              "median": 86.59,
              "minimum": 86.59,
              "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": 96486580224,
              "mean": 96486580224,
              "median": 96486580224,
              "minimum": 96486580224,
              "sample_count": 1
            },
            "process_pss_bytes": null,
            "process_rss_bytes": null
          },
          "process_sample_count": 0,
          "response_union_seconds": 1.1648289290023968,
          "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": 2002.2728389132026,
            "reason": null,
            "response_union_seconds": 36.58676577400183,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 54.72669684118394,
            "trapezoid_pieces": 37,
            "watt_hours": 0.5561868996981119
          },
          "gpu_sample_count": 36,
          "host_sample_count": 30,
          "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": 84388233216,
              "mean": 60147514436.26667,
              "median": 56929634304.0,
              "minimum": 41429835776,
              "sample_count": 30
            },
            "gpu_board_watts": {
              "maximum": 76.39,
              "mean": 55.100833333333334,
              "median": 66.89,
              "minimum": 15.26,
              "sample_count": 36
            },
            "gpu_memory_used_bytes": {
              "maximum": 63346573312.0,
              "mean": 16277045248.0,
              "median": 587202560.0,
              "minimum": 587202560.0,
              "sample_count": 36
            },
            "gpu_sm_clock_mhz": {
              "maximum": 2302.0,
              "mean": 1730.6666666666667,
              "median": 2272.0,
              "minimum": 180.0,
              "sample_count": 36
            },
            "gpu_utilization_percent": {
              "maximum": 81.0,
              "mean": 16.61111111111111,
              "median": 0.0,
              "minimum": 0.0,
              "sample_count": 36
            },
            "host_mem_available_bytes": {
              "maximum": 96304549888,
              "mean": 96224268151.46666,
              "median": 96237541376.0,
              "minimum": 96072572928,
              "sample_count": 30
            },
            "process_pss_bytes": {
              "maximum": 63779191808,
              "mean": 34062087236.266666,
              "median": 33919894528.0,
              "minimum": 721088512,
              "sample_count": 30
            },
            "process_rss_bytes": {
              "maximum": 63782150144,
              "mean": 34065046186.666668,
              "median": 33922852864.0,
              "minimum": 724049920,
              "sample_count": 30
            }
          },
          "process_sample_count": 30,
          "response_union_seconds": 36.58676577400183,
          "sampled_counter_deltas": {
            "deltas": {
              "major_faults": 5,
              "read_syscall_bytes": 5,
              "storage_read_bytes": 63054757888,
              "storage_write_bytes": 4096
            },
            "observed_endpoint_seconds": 35.40200611401815,
            "observed_intervals": 1,
            "requested_intervals": 1
          }
        },
        "ready": {
          "gpu_energy": {
            "available": true,
            "interpolation": "linear_at_bracketed_boundaries",
            "joules": 514.5725684548398,
            "reason": null,
            "response_union_seconds": 3.3283357390318997,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 154.6035643040361,
            "trapezoid_pieces": 6,
            "watt_hours": 0.14293682457078885
          },
          "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": 84639125504,
              "mean": 84545115477.33333,
              "median": 84621885440,
              "minimum": 84374335488,
              "sample_count": 3
            },
            "gpu_board_watts": {
              "maximum": 202.18,
              "mean": 157.62,
              "median": 196.68,
              "minimum": 74.0,
              "sample_count": 3
            },
            "gpu_memory_used_bytes": {
              "maximum": 63942164480.0,
              "mean": 63922591061.333336,
              "median": 63942164480.0,
              "minimum": 63883444224.0,
              "sample_count": 3
            },
            "gpu_sm_clock_mhz": {
              "maximum": 2332.0,
              "mean": 2314.6666666666665,
              "median": 2317.0,
              "minimum": 2295.0,
              "sample_count": 3
            },
            "gpu_utilization_percent": {
              "maximum": 0.0,
              "mean": 0.0,
              "median": 0.0,
              "minimum": 0.0,
              "sample_count": 3
            },
            "host_mem_available_bytes": {
              "maximum": 96038072320,
              "mean": 95896902314.66667,
              "median": 95834923008,
              "minimum": 95817711616,
              "sample_count": 3
            },
            "process_pss_bytes": {
              "maximum": 1353333760,
              "mean": 1268859221.3333333,
              "median": 1343036416,
              "minimum": 1110207488,
              "sample_count": 3
            },
            "process_rss_bytes": {
              "maximum": 1356292096,
              "mean": 1271817557.3333333,
              "median": 1345994752,
              "minimum": 1113165824,
              "sample_count": 3
            }
          },
          "process_sample_count": 3,
          "response_union_seconds": 3.3283357390318997,
          "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": 5577.166580078889,
            "reason": null,
            "response_union_seconds": 18.76270647603087,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 297.24744600165496,
            "trapezoid_pieces": 21,
            "watt_hours": 1.5492129389108025
          },
          "gpu_sample_count": 19,
          "host_sample_count": 18,
          "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": 84638580736,
              "mean": 84628920547.55556,
              "median": 84632049664.0,
              "minimum": 84617678848,
              "sample_count": 18
            },
            "gpu_board_watts": {
              "maximum": 303.42,
              "mean": 298.6694736842105,
              "median": 298.73,
              "minimum": 291.54,
              "sample_count": 19
            },
            "gpu_memory_used_bytes": {
              "maximum": 63942164480.0,
              "mean": 63942164480.0,
              "median": 63942164480.0,
              "minimum": 63942164480.0,
              "sample_count": 19
            },
            "gpu_sm_clock_mhz": {
              "maximum": 2212.0,
              "mean": 2094.2631578947367,
              "median": 2175.0,
              "minimum": 1837.0,
              "sample_count": 19
            },
            "gpu_utilization_percent": {
              "maximum": 100.0,
              "mean": 96.84210526315789,
              "median": 96.0,
              "minimum": 94.0,
              "sample_count": 19
            },
            "host_mem_available_bytes": {
              "maximum": 95839178752,
              "mean": 95816025201.77777,
              "median": 95815591936.0,
              "minimum": 95802257408,
              "sample_count": 18
            },
            "process_pss_bytes": {
              "maximum": 1353333760,
              "mean": 1348224227.5555556,
              "median": 1350933504.0,
              "minimum": 1341737984,
              "sample_count": 18
            },
            "process_rss_bytes": {
              "maximum": 1356292096,
              "mean": 1351182563.5555556,
              "median": 1353891840.0,
              "minimum": 1344696320,
              "sample_count": 18
            }
          },
          "process_sample_count": 18,
          "response_union_seconds": 18.76270647603087,
          "sampled_counter_deltas": {
            "deltas": {
              "major_faults": 0,
              "read_syscall_bytes": 0,
              "storage_read_bytes": 0,
              "storage_write_bytes": 28672
            },
            "observed_endpoint_seconds": 16.407505154958926,
            "observed_intervals": 2,
            "requested_intervals": 2
          }
        },
        "warmup": {
          "gpu_energy": {
            "available": true,
            "interpolation": "linear_at_bracketed_boundaries",
            "joules": 876.6143908214965,
            "reason": null,
            "response_union_seconds": 3.870858257985674,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 226.4651228220563,
            "trapezoid_pieces": 6,
            "watt_hours": 0.2435039974504157
          },
          "gpu_sample_count": 4,
          "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": 84629381120,
              "mean": 84622595072,
              "median": 84622559232.0,
              "minimum": 84615880704,
              "sample_count": 4
            },
            "gpu_board_watts": {
              "maximum": 296.46,
              "mean": 207.48,
              "median": 220.79,
              "minimum": 91.88,
              "sample_count": 4
            },
            "gpu_memory_used_bytes": {
              "maximum": 63942164480.0,
              "mean": 63942164480.0,
              "median": 63942164480.0,
              "minimum": 63942164480.0,
              "sample_count": 4
            },
            "gpu_sm_clock_mhz": {
              "maximum": 2287.0,
              "mean": 2141.0,
              "median": 2193.5,
              "minimum": 1890.0,
              "sample_count": 4
            },
            "gpu_utilization_percent": {
              "maximum": 100.0,
              "mean": 95.5,
              "median": 96.0,
              "minimum": 90.0,
              "sample_count": 4
            },
            "host_mem_available_bytes": {
              "maximum": 95830974464,
              "mean": 95809520640,
              "median": 95810375680.0,
              "minimum": 95786356736,
              "sample_count": 4
            },
            "process_pss_bytes": {
              "maximum": 1349901312,
              "mean": 1344703488,
              "median": 1343734784.0,
              "minimum": 1341443072,
              "sample_count": 4
            },
            "process_rss_bytes": {
              "maximum": 1352859648,
              "mean": 1347661824,
              "median": 1346693120.0,
              "minimum": 1344401408,
              "sample_count": 4
            }
          },
          "process_sample_count": 4,
          "response_union_seconds": 3.870858257985674,
          "sampled_counter_deltas": {
            "deltas": {
              "major_faults": 8,
              "read_syscall_bytes": 0,
              "storage_read_bytes": 1376256,
              "storage_write_bytes": 0
            },
            "observed_endpoint_seconds": 2.0698660400230438,
            "observed_intervals": 1,
            "requested_intervals": 2
          }
        }
      },
      "reviewed_export_sha256": "f32d7fb2e6a87c7e3188b5b27d217655dd296aa783ea2397dc4822420a6d555c",
      "settings": {
        "batch": 2048,
        "cache_prompt": false,
        "cache_ram_bytes": 0,
        "concurrency": 1,
        "context_shift": false,
        "contexts": [
          2048,
          8192
        ],
        "endpoint": "/completion",
        "fit": "off",
        "flash_attention": true,
        "ignore_eos": true,
        "kv_cache": "f16",
        "lazy_mode": "off",
        "load_mode": "mmap",
        "measured_rounds": 5,
        "mlock": false,
        "network_scope": "loopback",
        "output_tokens": 256,
        "p95_minimum_requests": 200,
        "placement": {
          "cpu_expert_layers": 0,
          "gpu_layers": "all"
        },
        "reasoning": "not_applicable_raw_completion",
        "seed": 20260927,
        "server_slots": 1,
        "synthetic_prompt_sha256": "c099b5abc034050fac1133e0214990d1f72a124b64f36e8a80ba3209bc07d187",
        "temperature": 0,
        "threads": 16,
        "tokenization": "model_tokenizer_add_special_false_repeat_then_truncate",
        "total_context_tokens": 8704,
        "ubatch": 512,
        "warmup_rounds": 1
      },
      "status": "completed",
      "telemetry": {
        "gpu_samples": 65,
        "host_samples": 58,
        "maximum_gpu_gap_seconds": 1.0076217760215513,
        "maximum_host_gap_seconds": 1.9519810149795376,
        "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": 65,
            "unavailable": 0
          },
          "software": {
            "active": 0,
            "not_active": 65,
            "unavailable": 0
          }
        }
      }
    },
    {
      "case_wall_seconds": 355.6453820169554,
      "completed_response_resources": {
        "gpu_energy": {
          "available": true,
          "interpolation": "linear_at_bracketed_boundaries",
          "joules": 55857.08840928863,
          "reason": null,
          "response_union_seconds": 186.21945289394353,
          "scope": "GPU board only; not whole-server AC energy",
          "time_weighted_mean_watts": 299.95302607349294,
          "trapezoid_pieces": 196,
          "watt_hours": 15.515857891469064
        },
        "gpu_sample_count": 186,
        "host_sample_count": 180,
        "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": 86966120448,
            "mean": 86940043855.64444,
            "median": 86937204736.0,
            "minimum": 86925795328,
            "sample_count": 180
          },
          "gpu_board_watts": {
            "maximum": 300.63,
            "mean": 299.99005376344087,
            "median": 299.99,
            "minimum": 299.2,
            "sample_count": 186
          },
          "gpu_memory_used_bytes": {
            "maximum": 80000057344.0,
            "mean": 80000057344.0,
            "median": 80000057344.0,
            "minimum": 80000057344.0,
            "sample_count": 186
          },
          "gpu_sm_clock_mhz": {
            "maximum": 1740.0,
            "mean": 1627.5645161290322,
            "median": 1657.0,
            "minimum": 1440.0,
            "sample_count": 186
          },
          "gpu_utilization_percent": {
            "maximum": 100.0,
            "mean": 99.9247311827957,
            "median": 100.0,
            "minimum": 89.0,
            "sample_count": 186
          },
          "host_mem_available_bytes": {
            "maximum": 96114024448,
            "mean": 96054613151.2889,
            "median": 96044468224.0,
            "minimum": 96015052800,
            "sample_count": 180
          },
          "process_pss_bytes": {
            "maximum": 1858872320,
            "mean": 1839274490.3111112,
            "median": 1852769280.0,
            "minimum": 1811305472,
            "sample_count": 180
          },
          "process_rss_bytes": {
            "maximum": 1861840896,
            "mean": 1842266703.6444445,
            "median": 1855737856.0,
            "minimum": 1814274048,
            "sample_count": 180
          }
        },
        "process_sample_count": 180,
        "response_union_seconds": 186.21945289394353,
        "sampled_counter_deltas": {
          "deltas": {
            "major_faults": 0,
            "read_syscall_bytes": 0,
            "storage_read_bytes": 0,
            "storage_write_bytes": 106496
          },
          "observed_endpoint_seconds": 174.8141460539773,
          "observed_intervals": 10,
          "requested_intervals": 10
        }
      },
      "groups": [
        {
          "aggregate_output_tokens_per_second": 16.40063817021913,
          "completed_output_tokens": 1280,
          "completed_request_count": 5,
          "context_tokens": 2048,
          "cpu_snapshot_intervals": {
            "host_capacity_percent": 3.1211453356351977,
            "interval_count": 5,
            "interval_seconds": 78.06629419594537,
            "meaning": "Weighted CPU counter rate across round-bracketing snapshots, not exact token time.",
            "one_logical_cpu_percent": 99.87665074032633,
            "process_cpu_seconds": 77.97
          },
          "failed_request_count": 0,
          "measured_rounds": 5,
          "mode": "completion",
          "resources": {
            "gpu_energy": {
              "available": true,
              "interpolation": "linear_at_bracketed_boundaries",
              "joules": 23405.402069666456,
              "reason": null,
              "response_union_seconds": 78.04574350797338,
              "scope": "GPU board only; not whole-server AC energy",
              "time_weighted_mean_watts": 299.8933832602324,
              "trapezoid_pieces": 83,
              "watt_hours": 6.5015005749073485
            },
            "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": 86966120448,
                "mean": 86947034165.89473,
                "median": 86946766848.0,
                "minimum": 86936510464,
                "sample_count": 76
              },
              "gpu_board_watts": {
                "maximum": 300.33,
                "mean": 299.98320512820516,
                "median": 299.99,
                "minimum": 299.2,
                "sample_count": 78
              },
              "gpu_memory_used_bytes": {
                "maximum": 80000057344.0,
                "mean": 80000057344.0,
                "median": 80000057344.0,
                "minimum": 80000057344.0,
                "sample_count": 78
              },
              "gpu_sm_clock_mhz": {
                "maximum": 1740.0,
                "mean": 1680.9358974358975,
                "median": 1698.5,
                "minimum": 1492.0,
                "sample_count": 78
              },
              "gpu_utilization_percent": {
                "maximum": 100.0,
                "mean": 99.83333333333333,
                "median": 100.0,
                "minimum": 89.0,
                "sample_count": 78
              },
              "host_mem_available_bytes": {
                "maximum": 96114024448,
                "mean": 96081347530.10527,
                "median": 96080076800.0,
                "minimum": 96059961344,
                "sample_count": 76
              },
              "process_pss_bytes": {
                "maximum": 1831052288,
                "mean": 1815476493.4736843,
                "median": 1813160960.0,
                "minimum": 1811305472,
                "sample_count": 76
              },
              "process_rss_bytes": {
                "maximum": 1834020864,
                "mean": 1818445069.4736843,
                "median": 1816129536.0,
                "minimum": 1814274048,
                "sample_count": 76
              }
            },
            "process_sample_count": 76,
            "response_union_seconds": 78.04574350797338,
            "sampled_counter_deltas": {
              "deltas": {
                "major_faults": 0,
                "read_syscall_bytes": 0,
                "storage_read_bytes": 0,
                "storage_write_bytes": 40960
              },
              "observed_endpoint_seconds": 73.0416899130214,
              "observed_intervals": 5,
              "requested_intervals": 5
            }
          },
          "response_outer_span_seconds": 78.07481005799491,
          "response_seconds": {
            "maximum": 15.836193532042671,
            "mean": 15.609148701594677,
            "median": 15.605332456005272,
            "minimum": 15.37207987497095,
            "sample_count": 5
          },
          "response_window_union_seconds": 78.04574350797338,
          "server_decode_tokens_per_second": {
            "maximum": 18.551195297162867,
            "mean": 18.2302957298202,
            "median": 18.20923135192617,
            "minimum": 17.94791051833425,
            "sample_count": 5
          },
          "synthetic_first_token_seconds": {
            "maximum": 1.6283667890238576,
            "mean": 1.6195974480011501,
            "median": 1.626210309972521,
            "minimum": 1.6014450550428592,
            "sample_count": 5
          },
          "warmup_completed_rounds": 1,
          "warmup_requests_per_round": 1
        },
        {
          "aggregate_output_tokens_per_second": 11.832819705136346,
          "completed_output_tokens": 1280,
          "completed_request_count": 5,
          "context_tokens": 8192,
          "cpu_snapshot_intervals": {
            "host_capacity_percent": 3.121144745345183,
            "interval_count": 5,
            "interval_seconds": 108.19347628898686,
            "meaning": "Weighted CPU counter rate across round-bracketing snapshots, not exact token time.",
            "one_logical_cpu_percent": 99.87663185104586,
            "process_cpu_seconds": 108.06
          },
          "failed_request_count": 0,
          "measured_rounds": 5,
          "mode": "completion",
          "resources": {
            "gpu_energy": {
              "available": true,
              "interpolation": "linear_at_bracketed_boundaries",
              "joules": 32451.686339622163,
              "reason": null,
              "response_union_seconds": 108.17370938597014,
              "scope": "GPU board only; not whole-server AC energy",
              "time_weighted_mean_watts": 299.9960574878009,
              "trapezoid_pieces": 113,
              "watt_hours": 9.014357316561712
            },
            "gpu_sample_count": 108,
            "host_sample_count": 104,
            "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": 86949203968,
                "mean": 86934935552,
                "median": 86933780480.0,
                "minimum": 86925795328,
                "sample_count": 104
              },
              "gpu_board_watts": {
                "maximum": 300.63,
                "mean": 299.995,
                "median": 299.99,
                "minimum": 299.54,
                "sample_count": 108
              },
              "gpu_memory_used_bytes": {
                "maximum": 80000057344.0,
                "mean": 80000057344.0,
                "median": 80000057344.0,
                "minimum": 80000057344.0,
                "sample_count": 108
              },
              "gpu_sm_clock_mhz": {
                "maximum": 1687.0,
                "mean": 1589.0185185185185,
                "median": 1635.0,
                "minimum": 1440.0,
                "sample_count": 108
              },
              "gpu_utilization_percent": {
                "maximum": 100.0,
                "mean": 99.99074074074075,
                "median": 100.0,
                "minimum": 99.0,
                "sample_count": 108
              },
              "host_mem_available_bytes": {
                "maximum": 96051195904,
                "mean": 96035076489.84616,
                "median": 96036397056.0,
                "minimum": 96015052800,
                "sample_count": 104
              },
              "process_pss_bytes": {
                "maximum": 1858872320,
                "mean": 1856665334.1538463,
                "median": 1857508352.0,
                "minimum": 1851196416,
                "sample_count": 104
              },
              "process_rss_bytes": {
                "maximum": 1861840896,
                "mean": 1859674820.9230769,
                "median": 1860476928.0,
                "minimum": 1854164992,
                "sample_count": 104
              }
            },
            "process_sample_count": 104,
            "response_union_seconds": 108.17370938597014,
            "sampled_counter_deltas": {
              "deltas": {
                "major_faults": 0,
                "read_syscall_bytes": 0,
                "storage_read_bytes": 0,
                "storage_write_bytes": 65536
              },
              "observed_endpoint_seconds": 101.77245614095591,
              "observed_intervals": 5,
              "requested_intervals": 5
            }
          },
          "response_outer_span_seconds": 108.20392717898358,
          "response_seconds": {
            "maximum": 21.827379069989547,
            "mean": 21.634741877194028,
            "median": 21.627266055962536,
            "minimum": 21.41630671202438,
            "sample_count": 5
          },
          "response_window_union_seconds": 108.17370938597014,
          "server_decode_tokens_per_second": {
            "maximum": 17.577270369850584,
            "mean": 17.411169160770672,
            "median": 17.422599078733956,
            "minimum": 17.266858141452268,
            "sample_count": 5
          },
          "synthetic_first_token_seconds": {
            "maximum": 7.059074364020489,
            "mean": 6.988261981203687,
            "median": 6.990940073970705,
            "minimum": 6.908797004027292,
            "sample_count": 5
          },
          "warmup_completed_rounds": 1,
          "warmup_requests_per_round": 1
        }
      ],
      "id": "baseline-72b-q8",
      "model": "qwen25-72b-q8",
      "module": "baseline",
      "placement": {
        "buffers": [
          {
            "device": "CPU_Mapped",
            "reported_mib": 1262.25
          },
          {
            "device": "CUDA0",
            "reported_mib": 72670.41
          }
        ],
        "gpu_layers": 81,
        "meaning": "Runtime buffer allocation is not measured resident process RAM.",
        "requested_cpu_expert_layers": 0,
        "total_layers": 81
      },
      "record_sha256": "d91f7a8a15841ca67a5d66247e9c695bdf3f8871bc97aaa5a1dee9f492d725a5",
      "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": 84113866752,
              "mean": 84105189376,
              "median": 84105189376.0,
              "minimum": 84096512000,
              "sample_count": 2
            },
            "gpu_board_watts": {
              "maximum": 16.24,
              "mean": 16.225,
              "median": 16.225,
              "minimum": 16.21,
              "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": 96514502656,
              "mean": 96499474432,
              "median": 96499474432.0,
              "minimum": 96484446208,
              "sample_count": 2
            },
            "process_pss_bytes": null,
            "process_rss_bytes": null
          },
          "process_sample_count": 0,
          "response_union_seconds": 1.100672836008016,
          "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": 86535368704,
              "mean": 86535368704,
              "median": 86535368704,
              "minimum": 86535368704,
              "sample_count": 1
            },
            "gpu_board_watts": {
              "maximum": 114.18,
              "mean": 114.18,
              "median": 114.18,
              "minimum": 114.18,
              "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": 96538132480,
              "mean": 96538132480,
              "median": 96538132480,
              "minimum": 96538132480,
              "sample_count": 1
            },
            "process_pss_bytes": null,
            "process_rss_bytes": null
          },
          "process_sample_count": 0,
          "response_union_seconds": 1.1574024299625307,
          "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": 2275.973838036839,
            "reason": null,
            "response_union_seconds": 38.89415991096757,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 58.51711010719242,
            "trapezoid_pieces": 39,
            "watt_hours": 0.6322149550102331
          },
          "gpu_sample_count": 38,
          "host_sample_count": 30,
          "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": 86970269696,
              "mean": 86913179511.46666,
              "median": 86960246784.0,
              "minimum": 85595869184,
              "sample_count": 30
            },
            "gpu_board_watts": {
              "maximum": 79.98,
              "mean": 58.390263157894736,
              "median": 71.515,
              "minimum": 16.13,
              "sample_count": 38
            },
            "gpu_memory_used_bytes": {
              "maximum": 76789317632.0,
              "mean": 20640390736.842106,
              "median": 587202560.0,
              "minimum": 587202560.0,
              "sample_count": 38
            },
            "gpu_sm_clock_mhz": {
              "maximum": 2310.0,
              "mean": 1737.578947368421,
              "median": 2272.0,
              "minimum": 180.0,
              "sample_count": 38
            },
            "gpu_utilization_percent": {
              "maximum": 81.0,
              "mean": 18.81578947368421,
              "median": 0.0,
              "minimum": 0.0,
              "sample_count": 38
            },
            "host_mem_available_bytes": {
              "maximum": 96234205184,
              "mean": 96109763242.66667,
              "median": 96112695296.0,
              "minimum": 96030871552,
              "sample_count": 30
            },
            "process_pss_bytes": {
              "maximum": 77867508736,
              "mean": 42180037461.333336,
              "median": 39471920128.0,
              "minimum": 1633039360,
              "sample_count": 30
            },
            "process_rss_bytes": {
              "maximum": 77870477312,
              "mean": 42183054131.2,
              "median": 39474888704.0,
              "minimum": 1636007936,
              "sample_count": 30
            }
          },
          "process_sample_count": 30,
          "response_union_seconds": 38.89415991096757,
          "sampled_counter_deltas": {
            "deltas": {
              "major_faults": 0,
              "read_syscall_bytes": 5,
              "storage_read_bytes": 76230348800,
              "storage_write_bytes": 4096
            },
            "observed_endpoint_seconds": 37.21782839397201,
            "observed_intervals": 1,
            "requested_intervals": 1
          }
        },
        "ready": {
          "gpu_energy": {
            "available": true,
            "interpolation": "linear_at_bracketed_boundaries",
            "joules": 569.13687963413,
            "reason": null,
            "response_union_seconds": 3.3318883720203303,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 170.81511025803874,
            "trapezoid_pieces": 8,
            "watt_hours": 0.1580935776761472
          },
          "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": 86946590720,
              "mean": 86934203050.66667,
              "median": 86936223744,
              "minimum": 86919794688,
              "sample_count": 3
            },
            "gpu_board_watts": {
              "maximum": 299.97,
              "mean": 158.414,
              "median": 106.19,
              "minimum": 79.8,
              "sample_count": 5
            },
            "gpu_memory_used_bytes": {
              "maximum": 80000057344.0,
              "mean": 79988313292.8,
              "median": 80000057344.0,
              "minimum": 79970697216.0,
              "sample_count": 5
            },
            "gpu_sm_clock_mhz": {
              "maximum": 2340.0,
              "mean": 2173.2,
              "median": 2310.0,
              "minimum": 1627.0,
              "sample_count": 5
            },
            "gpu_utilization_percent": {
              "maximum": 98.0,
              "mean": 19.6,
              "median": 0.0,
              "minimum": 0.0,
              "sample_count": 5
            },
            "host_mem_available_bytes": {
              "maximum": 96110792704,
              "mean": 96073691136,
              "median": 96070483968,
              "minimum": 96039796736,
              "sample_count": 3
            },
            "process_pss_bytes": {
              "maximum": 1858872320,
              "mean": 1824046762.6666667,
              "median": 1831121920,
              "minimum": 1782146048,
              "sample_count": 3
            },
            "process_rss_bytes": {
              "maximum": 1861840896,
              "mean": 1827015338.6666667,
              "median": 1834090496,
              "minimum": 1785114624,
              "sample_count": 3
            }
          },
          "process_sample_count": 3,
          "response_union_seconds": 3.3318883720203303,
          "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": 55878.75495729765,
            "reason": null,
            "response_union_seconds": 186.29246559401508,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 299.95177088414056,
            "trapezoid_pieces": 188,
            "watt_hours": 15.521876377027125
          },
          "gpu_sample_count": 186,
          "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": 86966120448,
              "mean": 86939990570.43094,
              "median": 86937067520,
              "minimum": 86925795328,
              "sample_count": 181
            },
            "gpu_board_watts": {
              "maximum": 300.63,
              "mean": 299.99005376344087,
              "median": 299.99,
              "minimum": 299.2,
              "sample_count": 186
            },
            "gpu_memory_used_bytes": {
              "maximum": 80000057344.0,
              "mean": 80000057344.0,
              "median": 80000057344.0,
              "minimum": 80000057344.0,
              "sample_count": 186
            },
            "gpu_sm_clock_mhz": {
              "maximum": 1740.0,
              "mean": 1627.5645161290322,
              "median": 1657.0,
              "minimum": 1440.0,
              "sample_count": 186
            },
            "gpu_utilization_percent": {
              "maximum": 100.0,
              "mean": 99.9247311827957,
              "median": 100.0,
              "minimum": 89.0,
              "sample_count": 186
            },
            "host_mem_available_bytes": {
              "maximum": 96114024448,
              "mean": 96054556117.56906,
              "median": 96044326912,
              "minimum": 96015052800,
              "sample_count": 181
            },
            "process_pss_bytes": {
              "maximum": 1858872320,
              "mean": 1839371450.6961327,
              "median": 1852769280,
              "minimum": 1811305472,
              "sample_count": 181
            },
            "process_rss_bytes": {
              "maximum": 1861840896,
              "mean": 1842363533.436464,
              "median": 1855737856,
              "minimum": 1814274048,
              "sample_count": 181
            }
          },
          "process_sample_count": 181,
          "response_union_seconds": 186.29246559401508,
          "sampled_counter_deltas": {
            "deltas": {
              "major_faults": 0,
              "read_syscall_bytes": 0,
              "storage_read_bytes": 0,
              "storage_write_bytes": 131072
            },
            "observed_endpoint_seconds": 184.07744828792056,
            "observed_intervals": 2,
            "requested_intervals": 2
          }
        },
        "warmup": {
          "gpu_energy": {
            "available": true,
            "interpolation": "linear_at_bracketed_boundaries",
            "joules": 10662.626895456593,
            "reason": null,
            "response_union_seconds": 36.43345087999478,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 292.6603612316979,
            "trapezoid_pieces": 38,
            "watt_hours": 2.961840804293498
          },
          "gpu_sample_count": 36,
          "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": 86964477952,
              "mean": 86945317546.66667,
              "median": 86945054720.0,
              "minimum": 86937935872,
              "sample_count": 36
            },
            "gpu_board_watts": {
              "maximum": 300.51,
              "mean": 295.0111111111111,
              "median": 299.995,
              "minimum": 188.62,
              "sample_count": 36
            },
            "gpu_memory_used_bytes": {
              "maximum": 80000057344.0,
              "mean": 80000057344.0,
              "median": 80000057344.0,
              "minimum": 80000057344.0,
              "sample_count": 36
            },
            "gpu_sm_clock_mhz": {
              "maximum": 1762.0,
              "mean": 1664.611111111111,
              "median": 1680.0,
              "minimum": 1485.0,
              "sample_count": 36
            },
            "gpu_utilization_percent": {
              "maximum": 100.0,
              "mean": 100.0,
              "median": 100.0,
              "minimum": 100.0,
              "sample_count": 36
            },
            "host_mem_available_bytes": {
              "maximum": 96108740608,
              "mean": 96070393514.66667,
              "median": 96053764096.0,
              "minimum": 96042815488,
              "sample_count": 36
            },
            "process_pss_bytes": {
              "maximum": 1850520576,
              "mean": 1829862172.4444444,
              "median": 1835518976.0,
              "minimum": 1802523648,
              "sample_count": 36
            },
            "process_rss_bytes": {
              "maximum": 1853489152,
              "mean": 1832830748.4444444,
              "median": 1838487552.0,
              "minimum": 1805492224,
              "sample_count": 36
            }
          },
          "process_sample_count": 36,
          "response_union_seconds": 36.43345087999478,
          "sampled_counter_deltas": {
            "deltas": {
              "major_faults": 0,
              "read_syscall_bytes": 0,
              "storage_read_bytes": 0,
              "storage_write_bytes": 20480
            },
            "observed_endpoint_seconds": 34.95324771897867,
            "observed_intervals": 2,
            "requested_intervals": 2
          }
        }
      },
      "reviewed_export_sha256": "f32d7fb2e6a87c7e3188b5b27d217655dd296aa783ea2397dc4822420a6d555c",
      "settings": {
        "batch": 2048,
        "cache_prompt": false,
        "cache_ram_bytes": 0,
        "concurrency": 1,
        "context_shift": false,
        "contexts": [
          2048,
          8192
        ],
        "endpoint": "/completion",
        "fit": "off",
        "flash_attention": true,
        "ignore_eos": true,
        "kv_cache": "f16",
        "lazy_mode": "off",
        "load_mode": "mmap",
        "measured_rounds": 5,
        "mlock": false,
        "network_scope": "loopback",
        "output_tokens": 256,
        "p95_minimum_requests": 200,
        "placement": {
          "cpu_expert_layers": 0,
          "gpu_layers": "all"
        },
        "reasoning": "not_applicable_raw_completion",
        "seed": 20260927,
        "server_slots": 1,
        "synthetic_prompt_sha256": "c099b5abc034050fac1133e0214990d1f72a124b64f36e8a80ba3209bc07d187",
        "temperature": 0,
        "threads": 16,
        "tokenization": "model_tokenizer_add_special_false_repeat_then_truncate",
        "total_context_tokens": 8704,
        "ubatch": 512,
        "warmup_rounds": 1
      },
      "status": "completed",
      "telemetry": {
        "gpu_samples": 268,
        "host_samples": 253,
        "maximum_gpu_gap_seconds": 1.0175007510115393,
        "maximum_host_gap_seconds": 2.0619814730016515,
        "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": 268,
            "unavailable": 0
          },
          "software": {
            "active": 0,
            "not_active": 268,
            "unavailable": 0
          }
        }
      }
    },
    {
      "case_wall_seconds": 1352.721352025983,
      "completed_response_resources": {
        "gpu_energy": {
          "available": true,
          "interpolation": "linear_at_bracketed_boundaries",
          "joules": 87432.48105413717,
          "reason": null,
          "response_union_seconds": 882.8749595810659,
          "scope": "GPU board only; not whole-server AC energy",
          "time_weighted_mean_watts": 99.03155606046961,
          "trapezoid_pieces": 893,
          "watt_hours": 24.28680029281588
        },
        "gpu_sample_count": 883,
        "host_sample_count": 434,
        "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": 86956191744,
            "mean": 86912404852.79263,
            "median": 86905057280.0,
            "minimum": 86897274880,
            "sample_count": 434
          },
          "gpu_board_watts": {
            "maximum": 137.35,
            "mean": 99.02436013590034,
            "median": 95.14,
            "minimum": 88.97,
            "sample_count": 883
          },
          "gpu_memory_used_bytes": {
            "maximum": 94157930496.0,
            "mean": 94157089735.17554,
            "median": 94157930496.0,
            "minimum": 94155833344.0,
            "sample_count": 883
          },
          "gpu_sm_clock_mhz": {
            "maximum": 2317.0,
            "mean": 2284.4246885617213,
            "median": 2280.0,
            "minimum": 1867.0,
            "sample_count": 883
          },
          "gpu_utilization_percent": {
            "maximum": 100.0,
            "mean": 34.90939977349943,
            "median": 10.0,
            "minimum": 2.0,
            "sample_count": 883
          },
          "host_mem_available_bytes": {
            "maximum": 95637037056,
            "mean": 95531510581.08755,
            "median": 95521652736.0,
            "minimum": 95466434560,
            "sample_count": 434
          },
          "process_pss_bytes": {
            "maximum": 55339509760,
            "mean": 55163167262.67281,
            "median": 55336781824.0,
            "minimum": 54836813824,
            "sample_count": 434
          },
          "process_rss_bytes": {
            "maximum": 55342489600,
            "mean": 55166208018.87558,
            "median": 55339761664.0,
            "minimum": 54839808000,
            "sample_count": 434
          }
        },
        "process_sample_count": 434,
        "response_union_seconds": 882.8749595810659,
        "sampled_counter_deltas": {
          "deltas": {
            "major_faults": 0,
            "read_syscall_bytes": 0,
            "storage_read_bytes": 0,
            "storage_write_bytes": 360448
          },
          "observed_endpoint_seconds": 859.7034017620608,
          "observed_intervals": 10,
          "requested_intervals": 10
        }
      },
      "groups": [
        {
          "aggregate_output_tokens_per_second": 3.6192757941330425,
          "completed_output_tokens": 1280,
          "completed_request_count": 5,
          "context_tokens": 2048,
          "cpu_snapshot_intervals": {
            "host_capacity_percent": 38.159420998783055,
            "interval_count": 5,
            "interval_seconds": 353.6806808580877,
            "meaning": "Weighted CPU counter rate across round-bracketing snapshots, not exact token time.",
            "one_logical_cpu_percent": 1221.1014719610578,
            "process_cpu_seconds": 4318.8
          },
          "failed_request_count": 0,
          "measured_rounds": 5,
          "mode": "completion",
          "resources": {
            "gpu_energy": {
              "available": true,
              "interpolation": "linear_at_bracketed_boundaries",
              "joules": 33952.31959522913,
              "reason": null,
              "response_union_seconds": 353.66191271605203,
              "scope": "GPU board only; not whole-server AC energy",
              "time_weighted_mean_watts": 96.00219411381389,
              "trapezoid_pieces": 359,
              "watt_hours": 9.431199887563647
            },
            "gpu_sample_count": 354,
            "host_sample_count": 167,
            "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": 86956191744,
                "mean": 86928198306.49101,
                "median": 86925025280,
                "minimum": 86913802240,
                "sample_count": 167
              },
              "gpu_board_watts": {
                "maximum": 136.38,
                "mean": 95.9926836158192,
                "median": 94.245,
                "minimum": 88.97,
                "sample_count": 354
              },
              "gpu_memory_used_bytes": {
                "maximum": 94155833344.0,
                "mean": 94155833344.0,
                "median": 94155833344.0,
                "minimum": 94155833344.0,
                "sample_count": 354
              },
              "gpu_sm_clock_mhz": {
                "maximum": 2317.0,
                "mean": 2284.135593220339,
                "median": 2280.0,
                "minimum": 1957.0,
                "sample_count": 354
              },
              "gpu_utilization_percent": {
                "maximum": 100.0,
                "mean": 23.166666666666668,
                "median": 10.0,
                "minimum": 2.0,
                "sample_count": 354
              },
              "host_mem_available_bytes": {
                "maximum": 95637037056,
                "mean": 95563106218.15569,
                "median": 95547498496,
                "minimum": 95503069184,
                "sample_count": 167
              },
              "process_pss_bytes": {
                "maximum": 54901213184,
                "mean": 54883534688.57485,
                "median": 54899740672,
                "minimum": 54836813824,
                "sample_count": 167
              },
              "process_rss_bytes": {
                "maximum": 54904193024,
                "mean": 54886587704.71856,
                "median": 54903541760,
                "minimum": 54839808000,
                "sample_count": 167
              }
            },
            "process_sample_count": 167,
            "response_union_seconds": 353.66191271605203,
            "sampled_counter_deltas": {
              "deltas": {
                "major_faults": 0,
                "read_syscall_bytes": 0,
                "storage_read_bytes": 0,
                "storage_write_bytes": 143360
              },
              "observed_endpoint_seconds": 341.29245696699945,
              "observed_intervals": 5,
              "requested_intervals": 5
            }
          },
          "response_outer_span_seconds": 353.68840590899345,
          "response_seconds": {
            "maximum": 70.85927938303212,
            "mean": 70.7323825432104,
            "median": 70.71976610599086,
            "minimum": 70.65188393200515,
            "sample_count": 5
          },
          "response_window_union_seconds": 353.66191271605203,
          "server_decode_tokens_per_second": {
            "maximum": 4.72037786569281,
            "mean": 4.705007937571641,
            "median": 4.704372092144592,
            "minimum": 4.692452607700804,
            "sample_count": 5
          },
          "synthetic_first_token_seconds": {
            "maximum": 16.539735411002766,
            "mean": 16.512870826828294,
            "median": 16.516706649039406,
            "minimum": 16.472258550056722,
            "sample_count": 5
          },
          "warmup_completed_rounds": 1,
          "warmup_requests_per_round": 1
        },
        {
          "aggregate_output_tokens_per_second": 2.4186856457574994,
          "completed_output_tokens": 1280,
          "completed_request_count": 5,
          "context_tokens": 8192,
          "cpu_snapshot_intervals": {
            "host_capacity_percent": 26.53628150643171,
            "interval_count": 5,
            "interval_seconds": 529.2353224620456,
            "meaning": "Weighted CPU counter rate across round-bracketing snapshots, not exact token time.",
            "one_logical_cpu_percent": 849.1610082058147,
            "process_cpu_seconds": 4494.06
          },
          "failed_request_count": 0,
          "measured_rounds": 5,
          "mode": "completion",
          "resources": {
            "gpu_energy": {
              "available": true,
              "interpolation": "linear_at_bracketed_boundaries",
              "joules": 53480.16145890802,
              "reason": null,
              "response_union_seconds": 529.2130468650139,
              "scope": "GPU board only; not whole-server AC energy",
              "time_weighted_mean_watts": 101.05601472926116,
              "trapezoid_pieces": 534,
              "watt_hours": 14.855600405252227
            },
            "gpu_sample_count": 529,
            "host_sample_count": 267,
            "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": 86921846784,
                "mean": 86902526550.29213,
                "median": 86901604352,
                "minimum": 86897274880,
                "sample_count": 267
              },
              "gpu_board_watts": {
                "maximum": 137.35,
                "mean": 101.0531190926276,
                "median": 95.67,
                "minimum": 90.98,
                "sample_count": 529
              },
              "gpu_memory_used_bytes": {
                "maximum": 94157930496.0,
                "mean": 94157930496.0,
                "median": 94157930496.0,
                "minimum": 94157930496.0,
                "sample_count": 529
              },
              "gpu_sm_clock_mhz": {
                "maximum": 2317.0,
                "mean": 2284.618147448015,
                "median": 2280.0,
                "minimum": 1867.0,
                "sample_count": 529
              },
              "gpu_utilization_percent": {
                "maximum": 100.0,
                "mean": 42.76748582230624,
                "median": 10.0,
                "minimum": 2.0,
                "sample_count": 529
              },
              "host_mem_available_bytes": {
                "maximum": 95557570560,
                "mean": 95511748515.95506,
                "median": 95512559616,
                "minimum": 95466434560,
                "sample_count": 267
              },
              "process_pss_bytes": {
                "maximum": 55339509760,
                "mean": 55338068535.61049,
                "median": 55338133504,
                "minimum": 55335350272,
                "sample_count": 267
              },
              "process_rss_bytes": {
                "maximum": 55342489600,
                "mean": 55341101623.61049,
                "median": 55341113344,
                "minimum": 55339749376,
                "sample_count": 267
              }
            },
            "process_sample_count": 267,
            "response_union_seconds": 529.2130468650139,
            "sampled_counter_deltas": {
              "deltas": {
                "major_faults": 0,
                "read_syscall_bytes": 0,
                "storage_read_bytes": 0,
                "storage_write_bytes": 217088
              },
              "observed_endpoint_seconds": 518.4109447950614,
              "observed_intervals": 5,
              "requested_intervals": 5
            }
          },
          "response_outer_span_seconds": 529.246267170005,
          "response_seconds": {
            "maximum": 105.9342662889976,
            "mean": 105.84260937300277,
            "median": 105.85307992796879,
            "minimum": 105.70028573501622,
            "sample_count": 5
          },
          "response_window_union_seconds": 529.2130468650139,
          "server_decode_tokens_per_second": {
            "maximum": 4.717623238471859,
            "mean": 4.707775473637765,
            "median": 4.707951723075252,
            "minimum": 4.6967906663607675,
            "sample_count": 5
          },
          "synthetic_first_token_seconds": {
            "maximum": 51.79277144599473,
            "mean": 51.67663359480211,
            "median": 51.64751424099086,
            "minimum": 51.63793887099018,
            "sample_count": 5
          },
          "warmup_completed_rounds": 1,
          "warmup_requests_per_round": 1
        }
      ],
      "id": "capacity-235b",
      "model": "qwen3-235b-q5",
      "module": "hybrid-capacity",
      "placement": {
        "buffers": [
          {
            "device": "CPU_Mapped",
            "reported_mib": 27759.73
          },
          {
            "device": "CPU_Mapped",
            "reported_mib": 28323.86
          },
          {
            "device": "CPU_Mapped",
            "reported_mib": 18322.9
          },
          {
            "device": "CUDA0",
            "reported_mib": 86732.62
          }
        ],
        "gpu_layers": 95,
        "meaning": "Runtime buffer allocation is not measured resident process RAM.",
        "requested_cpu_expert_layers": 44,
        "total_layers": 95
      },
      "record_sha256": "3a49e3e92a194f228457843e1a3264cd5a2af9e3e1302b9862ba46d0660bdd2a",
      "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": 86569291776,
              "mean": 86560978944,
              "median": 86560978944.0,
              "minimum": 86552666112,
              "sample_count": 2
            },
            "gpu_board_watts": {
              "maximum": 16.06,
              "mean": 16.005,
              "median": 16.005,
              "minimum": 15.95,
              "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": 96450252800,
              "mean": 96438243328,
              "median": 96438243328.0,
              "minimum": 96426233856,
              "sample_count": 2
            },
            "process_pss_bytes": null,
            "process_rss_bytes": null
          },
          "process_sample_count": 0,
          "response_union_seconds": 1.100754403974861,
          "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": 86905700352,
              "mean": 86905700352,
              "median": 86905700352,
              "minimum": 86905700352,
              "sample_count": 1
            },
            "gpu_board_watts": {
              "maximum": 79.73,
              "mean": 79.73,
              "median": 79.73,
              "minimum": 79.73,
              "sample_count": 1
            },
            "gpu_memory_used_bytes": {
              "maximum": 94151639040.0,
              "mean": 94151639040.0,
              "median": 94151639040.0,
              "minimum": 94151639040.0,
              "sample_count": 1
            },
            "gpu_sm_clock_mhz": {
              "maximum": 2272.0,
              "mean": 2272.0,
              "median": 2272.0,
              "minimum": 2272.0,
              "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": 95525027840,
              "mean": 95525027840,
              "median": 95525027840,
              "minimum": 95525027840,
              "sample_count": 1
            },
            "process_pss_bytes": null,
            "process_rss_bytes": null
          },
          "process_sample_count": 0,
          "response_union_seconds": 1.1650557749671862,
          "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": 5680.773122758429,
            "reason": null,
            "response_union_seconds": 116.2166981890332,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 48.8808683371672,
            "trapezoid_pieces": 117,
            "watt_hours": 1.5779925340995635
          },
          "gpu_sample_count": 116,
          "host_sample_count": 89,
          "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": 86973087744,
              "mean": 86957874831.82022,
              "median": 86961295360,
              "minimum": 86779006976,
              "sample_count": 89
            },
            "gpu_board_watts": {
              "maximum": 77.03,
              "mean": 49.04491379310345,
              "median": 71.38,
              "minimum": 15.7,
              "sample_count": 116
            },
            "gpu_memory_used_bytes": {
              "maximum": 94088724480.0,
              "mean": 39070574521.37931,
              "median": 587202560.0,
              "minimum": 587202560.0,
              "sample_count": 116
            },
            "gpu_sm_clock_mhz": {
              "maximum": 2280.0,
              "mean": 1387.0086206896551,
              "median": 2272.0,
              "minimum": 172.0,
              "sample_count": 116
            },
            "gpu_utilization_percent": {
              "maximum": 18.0,
              "mean": 4.905172413793103,
              "median": 0.0,
              "minimum": 0.0,
              "sample_count": 116
            },
            "host_mem_available_bytes": {
              "maximum": 96184197120,
              "mean": 95922319406.02248,
              "median": 95875031040,
              "minimum": 95732076544,
              "sample_count": 89
            },
            "process_pss_bytes": {
              "maximum": 86750541824,
              "mean": 66394780545.4382,
              "median": 85652535296,
              "minimum": 339699712,
              "sample_count": 89
            },
            "process_rss_bytes": {
              "maximum": 86753521664,
              "mean": 66397802449.97753,
              "median": 85655515136,
              "minimum": 342679552,
              "sample_count": 89
            }
          },
          "process_sample_count": 89,
          "response_union_seconds": 116.2166981890332,
          "sampled_counter_deltas": {
            "deltas": {
              "major_faults": 56045,
              "read_syscall_bytes": 13,
              "storage_read_bytes": 265319071744,
              "storage_write_bytes": 12288
            },
            "observed_endpoint_seconds": 115.25819665094605,
            "observed_intervals": 1,
            "requested_intervals": 1
          }
        },
        "ready": {
          "gpu_energy": {
            "available": true,
            "interpolation": "linear_at_bracketed_boundaries",
            "joules": 279.6970268935408,
            "reason": null,
            "response_union_seconds": 3.371804744063411,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 82.9517270790935,
            "trapezoid_pieces": 6,
            "watt_hours": 0.07769361858153911
          },
          "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": 86954340352,
              "mean": 86929561600,
              "median": 86929561600.0,
              "minimum": 86904782848,
              "sample_count": 2
            },
            "gpu_board_watts": {
              "maximum": 88.89,
              "mean": 83.37,
              "median": 84.96,
              "minimum": 76.26,
              "sample_count": 3
            },
            "gpu_memory_used_bytes": {
              "maximum": 94157930496.0,
              "mean": 94135560874.66667,
              "median": 94155833344.0,
              "minimum": 94092918784.0,
              "sample_count": 3
            },
            "gpu_sm_clock_mhz": {
              "maximum": 2272.0,
              "mean": 2272.0,
              "median": 2272.0,
              "minimum": 2272.0,
              "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": 95830491136,
              "mean": 95675967488,
              "median": 95675967488.0,
              "minimum": 95521443840,
              "sample_count": 2
            },
            "process_pss_bytes": {
              "maximum": 55339509760,
              "mean": 32317054976,
              "median": 32317054976.0,
              "minimum": 9294600192,
              "sample_count": 2
            },
            "process_rss_bytes": {
              "maximum": 55342489600,
              "mean": 32320034816,
              "median": 32320034816.0,
              "minimum": 9297580032,
              "sample_count": 2
            }
          },
          "process_sample_count": 2,
          "response_union_seconds": 3.371804744063411,
          "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": 87439.59809728934,
            "reason": null,
            "response_union_seconds": 882.9507616019691,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 99.03111464408792,
            "trapezoid_pieces": 885,
            "watt_hours": 24.28877724924704
          },
          "gpu_sample_count": 883,
          "host_sample_count": 436,
          "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": 86956191744,
              "mean": 86912462077.65138,
              "median": 86905081856.0,
              "minimum": 86897274880,
              "sample_count": 436
            },
            "gpu_board_watts": {
              "maximum": 137.35,
              "mean": 99.02436013590034,
              "median": 95.14,
              "minimum": 88.97,
              "sample_count": 883
            },
            "gpu_memory_used_bytes": {
              "maximum": 94157930496.0,
              "mean": 94157089735.17554,
              "median": 94157930496.0,
              "minimum": 94155833344.0,
              "sample_count": 883
            },
            "gpu_sm_clock_mhz": {
              "maximum": 2317.0,
              "mean": 2284.4246885617213,
              "median": 2280.0,
              "minimum": 1867.0,
              "sample_count": 883
            },
            "gpu_utilization_percent": {
              "maximum": 100.0,
              "mean": 34.90939977349943,
              "median": 10.0,
              "minimum": 2.0,
              "sample_count": 883
            },
            "host_mem_available_bytes": {
              "maximum": 95637037056,
              "mean": 95531524809.98166,
              "median": 95521929216.0,
              "minimum": 95466434560,
              "sample_count": 436
            },
            "process_pss_bytes": {
              "maximum": 55339509760,
              "mean": 55161964248.073395,
              "median": 55336781824.0,
              "minimum": 54836813824,
              "sample_count": 436
            },
            "process_rss_bytes": {
              "maximum": 55342489600,
              "mean": 55165004724.84404,
              "median": 55339761664.0,
              "minimum": 54839808000,
              "sample_count": 436
            }
          },
          "process_sample_count": 436,
          "response_union_seconds": 882.9507616019691,
          "sampled_counter_deltas": {
            "deltas": {
              "major_faults": 0,
              "read_syscall_bytes": 0,
              "storage_read_bytes": 0,
              "storage_write_bytes": 385024
            },
            "observed_endpoint_seconds": 878.285594719986,
            "observed_intervals": 2,
            "requested_intervals": 2
          }
        },
        "warmup": {
          "gpu_energy": {
            "available": true,
            "interpolation": "linear_at_bracketed_boundaries",
            "joules": 18356.05087851068,
            "reason": null,
            "response_union_seconds": 192.7558839409612,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 95.22952297597782,
            "trapezoid_pieces": 195,
            "watt_hours": 5.098903021808522
          },
          "gpu_sample_count": 193,
          "host_sample_count": 104,
          "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": 86972977152,
              "mean": 86935542350.76923,
              "median": 86941855744.0,
              "minimum": 86844063744,
              "sample_count": 104
            },
            "gpu_board_watts": {
              "maximum": 133.51,
              "mean": 95.20875647668393,
              "median": 93.83,
              "minimum": 74.26,
              "sample_count": 193
            },
            "gpu_memory_used_bytes": {
              "maximum": 94157930496.0,
              "mean": 94156430977.98964,
              "median": 94155833344.0,
              "minimum": 94155833344.0,
              "sample_count": 193
            },
            "gpu_sm_clock_mhz": {
              "maximum": 2317.0,
              "mean": 2280.352331606218,
              "median": 2280.0,
              "minimum": 1942.0,
              "sample_count": 193
            },
            "gpu_utilization_percent": {
              "maximum": 100.0,
              "mean": 31.139896373056995,
              "median": 10.0,
              "minimum": 2.0,
              "sample_count": 193
            },
            "host_mem_available_bytes": {
              "maximum": 95645704192,
              "mean": 95576961142.15384,
              "median": 95548684288.0,
              "minimum": 95511699456,
              "sample_count": 104
            },
            "process_pss_bytes": {
              "maximum": 55336093696,
              "mean": 50895688615.38461,
              "median": 54901958656.0,
              "minimum": 11203971072,
              "sample_count": 104
            },
            "process_rss_bytes": {
              "maximum": 55339073536,
              "mean": 50898682210.46154,
              "median": 54904938496.0,
              "minimum": 11206950912,
              "sample_count": 104
            }
          },
          "process_sample_count": 104,
          "response_union_seconds": 192.7558839409612,
          "sampled_counter_deltas": {
            "deltas": {
              "major_faults": 5517,
              "read_syscall_bytes": 0,
              "storage_read_bytes": 44982132736,
              "storage_write_bytes": 73728
            },
            "observed_endpoint_seconds": 190.65362677600933,
            "observed_intervals": 2,
            "requested_intervals": 2
          }
        }
      },
      "reviewed_export_sha256": "f32d7fb2e6a87c7e3188b5b27d217655dd296aa783ea2397dc4822420a6d555c",
      "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": 44,
          "gpu_layers": "all"
        },
        "reasoning": "not_applicable_raw_completion",
        "seed": 20260927,
        "server_slots": 1,
        "synthetic_prompt_sha256": "c099b5abc034050fac1133e0214990d1f72a124b64f36e8a80ba3209bc07d187",
        "temperature": 0,
        "threads": 16,
        "tokenization": "model_tokenizer_add_special_false_repeat_then_truncate",
        "total_context_tokens": 8704,
        "ubatch": 512,
        "warmup_rounds": 1
      },
      "status": "completed",
      "telemetry": {
        "gpu_samples": 1198,
        "host_samples": 634,
        "maximum_gpu_gap_seconds": 1.4088406169903465,
        "maximum_host_gap_seconds": 2.6100394439999945,
        "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": 1198,
            "unavailable": 0
          },
          "software": {
            "active": 0,
            "not_active": 1198,
            "unavailable": 0
          }
        }
      }
    },
    {
      "case_wall_seconds": 3542.812662684999,
      "completed_response_resources": {
        "gpu_energy": {
          "available": true,
          "interpolation": "linear_at_bracketed_boundaries",
          "joules": 236530.70309813673,
          "reason": null,
          "response_union_seconds": 2749.5454220350366,
          "scope": "GPU board only; not whole-server AC energy",
          "time_weighted_mean_watts": 86.02538485182468,
          "trapezoid_pieces": 2760,
          "watt_hours": 65.70297308281576
        },
        "gpu_sample_count": 2750,
        "host_sample_count": 1620,
        "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": 2323677184,
            "mean": 2286991319.545679,
            "median": 2292633600.0,
            "minimum": 2259800064,
            "sample_count": 1620
          },
          "gpu_board_watts": {
            "maximum": 152.49,
            "mean": 86.01490909090909,
            "median": 81.61,
            "minimum": 79.5,
            "sample_count": 2750
          },
          "gpu_memory_used_bytes": {
            "maximum": 24192745472.0,
            "mean": 24186094068.456726,
            "median": 24192745472.0,
            "minimum": 24178065408.0,
            "sample_count": 2750
          },
          "gpu_sm_clock_mhz": {
            "maximum": 2332.0,
            "mean": 2263.0723636363637,
            "median": 2272.0,
            "minimum": 1785.0,
            "sample_count": 2750
          },
          "gpu_utilization_percent": {
            "maximum": 100.0,
            "mean": 11.226545454545455,
            "median": 0.0,
            "minimum": 0.0,
            "sample_count": 2750
          },
          "host_mem_available_bytes": {
            "maximum": 94687850496,
            "mean": 94599706644.22716,
            "median": 94597986304.0,
            "minimum": 94527299584,
            "sample_count": 1620
          },
          "process_pss_bytes": {
            "maximum": 24359494656,
            "mean": 24348639060.701233,
            "median": 24354731008.0,
            "minimum": 24331427840,
            "sample_count": 1620
          },
          "process_rss_bytes": {
            "maximum": 24362508288,
            "mean": 24351724731.101234,
            "median": 24357744640.0,
            "minimum": 24335863808,
            "sample_count": 1620
          }
        },
        "process_sample_count": 1620,
        "response_union_seconds": 2749.5454220350366,
        "sampled_counter_deltas": {
          "deltas": {
            "major_faults": 0,
            "read_syscall_bytes": 0,
            "storage_read_bytes": 0,
            "storage_write_bytes": 1228800
          },
          "observed_endpoint_seconds": 2732.816400779062,
          "observed_intervals": 10,
          "requested_intervals": 10
        }
      },
      "groups": [
        {
          "aggregate_output_tokens_per_second": 1.0272049883875025,
          "completed_output_tokens": 1280,
          "completed_request_count": 5,
          "context_tokens": 2048,
          "cpu_snapshot_intervals": {
            "host_capacity_percent": 46.961556060353814,
            "interval_count": 5,
            "interval_seconds": 1246.1183393240208,
            "meaning": "Weighted CPU counter rate across round-bracketing snapshots, not exact token time.",
            "one_logical_cpu_percent": 1502.769793931322,
            "process_cpu_seconds": 18726.29
          },
          "failed_request_count": 0,
          "measured_rounds": 5,
          "mode": "completion",
          "resources": {
            "gpu_energy": {
              "available": true,
              "interpolation": "linear_at_bracketed_boundaries",
              "joules": 103758.67294673451,
              "reason": null,
              "response_union_seconds": 1246.099867573008,
              "scope": "GPU board only; not whole-server AC energy",
              "time_weighted_mean_watts": 83.2667394057446,
              "trapezoid_pieces": 1251,
              "watt_hours": 28.821853596315144
            },
            "gpu_sample_count": 1246,
            "host_sample_count": 696,
            "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": 2293600256,
                "mean": 2270765685.7011495,
                "median": 2271793152.0,
                "minimum": 2259800064,
                "sample_count": 696
              },
              "gpu_board_watts": {
                "maximum": 146.64,
                "mean": 83.25452648475121,
                "median": 80.98,
                "minimum": 79.5,
                "sample_count": 1246
              },
              "gpu_memory_used_bytes": {
                "maximum": 24178065408.0,
                "mean": 24178065408.0,
                "median": 24178065408.0,
                "minimum": 24178065408.0,
                "sample_count": 1246
              },
              "gpu_sm_clock_mhz": {
                "maximum": 2325.0,
                "mean": 2263.345104333868,
                "median": 2272.0,
                "minimum": 1852.0,
                "sample_count": 1246
              },
              "gpu_utilization_percent": {
                "maximum": 100.0,
                "mean": 5.826645264847512,
                "median": 0.0,
                "minimum": 0.0,
                "sample_count": 1246
              },
              "host_mem_available_bytes": {
                "maximum": 94687850496,
                "mean": 94607496833.47127,
                "median": 94607792128.0,
                "minimum": 94549704704,
                "sample_count": 696
              },
              "process_pss_bytes": {
                "maximum": 24341984256,
                "mean": 24338712246.436783,
                "median": 24339825664.0,
                "minimum": 24331427840,
                "sample_count": 696
              },
              "process_rss_bytes": {
                "maximum": 24344997888,
                "mean": 24341802213.517242,
                "median": 24342839296.0,
                "minimum": 24335863808,
                "sample_count": 696
              }
            },
            "process_sample_count": 696,
            "response_union_seconds": 1246.099867573008,
            "sampled_counter_deltas": {
              "deltas": {
                "major_faults": 0,
                "read_syscall_bytes": 0,
                "storage_read_bytes": 0,
                "storage_write_bytes": 565248
              },
              "observed_endpoint_seconds": 1236.6610553499777,
              "observed_intervals": 5,
              "requested_intervals": 5
            }
          },
          "response_outer_span_seconds": 1246.124862863042,
          "response_seconds": {
            "maximum": 249.32341260602698,
            "mean": 249.21997351460158,
            "median": 249.17530603701016,
            "minimum": 249.15829364600359,
            "sample_count": 5
          },
          "response_window_union_seconds": 1246.099867573008,
          "server_decode_tokens_per_second": {
            "maximum": 1.0771389966646243,
            "mean": 1.0769035802653517,
            "median": 1.0769342620955185,
            "minimum": 1.0766185877515206,
            "sample_count": 5
          },
          "synthetic_first_token_seconds": {
            "maximum": 12.426381879020482,
            "mean": 12.409445786592551,
            "median": 12.411477557965554,
            "minimum": 12.391928925993852,
            "sample_count": 5
          },
          "warmup_completed_rounds": 1,
          "warmup_requests_per_round": 1
        },
        {
          "aggregate_output_tokens_per_second": 0.8513776878724529,
          "completed_output_tokens": 1280,
          "completed_request_count": 5,
          "context_tokens": 8192,
          "cpu_snapshot_intervals": {
            "host_capacity_percent": 41.45806174570441,
            "interval_count": 5,
            "interval_seconds": 1503.4704017839977,
            "meaning": "Weighted CPU counter rate across round-bracketing snapshots, not exact token time.",
            "one_logical_cpu_percent": 1326.6579758625412,
            "process_cpu_seconds": 19945.91
          },
          "failed_request_count": 0,
          "measured_rounds": 5,
          "mode": "completion",
          "resources": {
            "gpu_energy": {
              "available": true,
              "interpolation": "linear_at_bracketed_boundaries",
              "joules": 132772.0301514023,
              "reason": null,
              "response_union_seconds": 1503.4455544620287,
              "scope": "GPU board only; not whole-server AC energy",
              "time_weighted_mean_watts": 88.31183128471288,
              "trapezoid_pieces": 1509,
              "watt_hours": 36.88111948650064
            },
            "gpu_sample_count": 1504,
            "host_sample_count": 924,
            "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": 2323677184,
                "mean": 2299213225.5584416,
                "median": 2299183104.0,
                "minimum": 2290978816,
                "sample_count": 924
              },
              "gpu_board_watts": {
                "maximum": 152.49,
                "mean": 88.30176861702128,
                "median": 82.01,
                "minimum": 80.39,
                "sample_count": 1504
              },
              "gpu_memory_used_bytes": {
                "maximum": 24192745472.0,
                "mean": 24192745472.0,
                "median": 24192745472.0,
                "minimum": 24192745472.0,
                "sample_count": 1504
              },
              "gpu_sm_clock_mhz": {
                "maximum": 2332.0,
                "mean": 2262.846409574468,
                "median": 2272.0,
                "minimum": 1785.0,
                "sample_count": 1504
              },
              "gpu_utilization_percent": {
                "maximum": 100.0,
                "mean": 15.700132978723405,
                "median": 0.0,
                "minimum": 0.0,
                "sample_count": 1504
              },
              "host_mem_available_bytes": {
                "maximum": 94663417856,
                "mean": 94593838709.47186,
                "median": 94591438848.0,
                "minimum": 94527299584,
                "sample_count": 924
              },
              "process_pss_bytes": {
                "maximum": 24359494656,
                "mean": 24356116401.316017,
                "median": 24356086784.0,
                "minimum": 24353292288,
                "sample_count": 924
              },
              "process_rss_bytes": {
                "maximum": 24362508288,
                "mean": 24359198835.255413,
                "median": 24359100416.0,
                "minimum": 24357724160,
                "sample_count": 924
              }
            },
            "process_sample_count": 924,
            "response_union_seconds": 1503.4455544620287,
            "sampled_counter_deltas": {
              "deltas": {
                "major_faults": 0,
                "read_syscall_bytes": 0,
                "storage_read_bytes": 0,
                "storage_write_bytes": 663552
              },
              "observed_endpoint_seconds": 1496.1553454290843,
              "observed_intervals": 5,
              "requested_intervals": 5
            }
          },
          "response_outer_span_seconds": 1503.4815187479835,
          "response_seconds": {
            "maximum": 300.788018139021,
            "mean": 300.6891108924057,
            "median": 300.6939707050333,
            "minimum": 300.5792841439834,
            "sample_count": 5
          },
          "response_window_union_seconds": 1503.4455544620287,
          "server_decode_tokens_per_second": {
            "maximum": 1.02139510723465,
            "mean": 1.0210610239530291,
            "median": 1.0211173638020445,
            "minimum": 1.0206264196713375,
            "sample_count": 5
          },
          "synthetic_first_token_seconds": {
            "maximum": 50.99202599900309,
            "mean": 50.948651273804714,
            "median": 50.94126555201365,
            "minimum": 50.9204786689952,
            "sample_count": 5
          },
          "warmup_completed_rounds": 1,
          "warmup_requests_per_round": 1
        }
      ],
      "id": "offload-72b-40",
      "model": "qwen25-72b-q4",
      "module": "dense-offload",
      "placement": {
        "buffers": [
          {
            "device": "CPU_Mapped",
            "reported_mib": 3796.51
          },
          {
            "device": "CPU_Mapped",
            "reported_mib": 3698.21
          },
          {
            "device": "CPU_Mapped",
            "reported_mib": 3780.0
          },
          {
            "device": "CPU_Mapped",
            "reported_mib": 3758.87
          },
          {
            "device": "CPU_Mapped",
            "reported_mib": 3780.06
          },
          {
            "device": "CPU_Mapped",
            "reported_mib": 2489.76
          },
          {
            "device": "CUDA0",
            "reported_mib": 20662.52
          }
        ],
        "gpu_layers": 40,
        "meaning": "Runtime buffer allocation is not measured resident process RAM.",
        "requested_cpu_expert_layers": 0,
        "total_layers": 81
      },
      "record_sha256": "6e94d76fd5f84834595a846d9cdb805969e602d7e4891c723a3d7f43a132755d",
      "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": 376336384,
              "mean": 368064512,
              "median": 368064512.0,
              "minimum": 359792640,
              "sample_count": 2
            },
            "gpu_board_watts": {
              "maximum": 15.95,
              "mean": 15.95,
              "median": 15.95,
              "minimum": 15.95,
              "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": 96496439296,
              "mean": 96496254976,
              "median": 96496254976.0,
              "minimum": 96496070656,
              "sample_count": 2
            },
            "process_pss_bytes": null,
            "process_rss_bytes": null
          },
          "process_sample_count": 0,
          "response_union_seconds": 1.1007298770127818,
          "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": 603852800,
              "mean": 603852800,
              "median": 603852800,
              "minimum": 603852800,
              "sample_count": 1
            },
            "gpu_board_watts": {
              "maximum": 79.95,
              "mean": 79.95,
              "median": 79.95,
              "minimum": 79.95,
              "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": 2272.0,
              "mean": 2272.0,
              "median": 2272.0,
              "minimum": 2272.0,
              "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": 96244224000,
              "mean": 96244224000,
              "median": 96244224000,
              "minimum": 96244224000,
              "sample_count": 1
            },
            "process_pss_bytes": null,
            "process_rss_bytes": null
          },
          "process_sample_count": 0,
          "response_union_seconds": 1.1574398530065082,
          "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": 481.14320595099116,
            "reason": null,
            "response_union_seconds": 6.915469526022207,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 69.57491521587917,
            "trapezoid_pieces": 7,
            "watt_hours": 0.13365089054194199
          },
          "gpu_sample_count": 6,
          "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": 2173902848,
              "mean": 1243803648,
              "median": 657190912,
              "minimum": 605564928,
              "sample_count": 5
            },
            "gpu_board_watts": {
              "maximum": 79.13,
              "mean": 72.73333333333333,
              "median": 76.555,
              "minimum": 51.58,
              "sample_count": 6
            },
            "gpu_memory_used_bytes": {
              "maximum": 24106762240.0,
              "mean": 19248010581.333332,
              "median": 22254977024.0,
              "minimum": 587202560.0,
              "sample_count": 6
            },
            "gpu_sm_clock_mhz": {
              "maximum": 2310.0,
              "mean": 2297.3333333333335,
              "median": 2306.0,
              "minimum": 2265.0,
              "sample_count": 6
            },
            "gpu_utilization_percent": {
              "maximum": 81.0,
              "mean": 40.333333333333336,
              "median": 40.0,
              "minimum": 0.0,
              "sample_count": 6
            },
            "host_mem_available_bytes": {
              "maximum": 96189173760,
              "mean": 95540204339.2,
              "median": 96136634368,
              "minimum": 94583787520,
              "sample_count": 5
            },
            "process_pss_bytes": {
              "maximum": 44348638208,
              "mean": 31194135961.6,
              "median": 24240718848,
              "minimum": 18852537344,
              "sample_count": 5
            },
            "process_rss_bytes": {
              "maximum": 44351598592,
              "mean": 31197117644.8,
              "median": 24243732480,
              "minimum": 18855497728,
              "sample_count": 5
            }
          },
          "process_sample_count": 5,
          "response_union_seconds": 6.915469526022207,
          "sampled_counter_deltas": {
            "deltas": {
              "major_faults": 0,
              "read_syscall_bytes": 13,
              "storage_read_bytes": 0,
              "storage_write_bytes": 8192
            },
            "observed_endpoint_seconds": 5.654299405985512,
            "observed_intervals": 1,
            "requested_intervals": 1
          }
        },
        "ready": {
          "gpu_energy": {
            "available": true,
            "interpolation": "linear_at_bracketed_boundaries",
            "joules": 272.823505245464,
            "reason": null,
            "response_union_seconds": 3.400849792989902,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 80.22215677029581,
            "trapezoid_pieces": 8,
            "watt_hours": 0.0757843070126289
          },
          "gpu_sample_count": 5,
          "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": 2307264512,
              "mean": 2245849088,
              "median": 2245849088.0,
              "minimum": 2184433664,
              "sample_count": 2
            },
            "gpu_board_watts": {
              "maximum": 85.62,
              "mean": 80.17,
              "median": 80.06,
              "minimum": 77.05,
              "sample_count": 5
            },
            "gpu_memory_used_bytes": {
              "maximum": 24192745472.0,
              "mean": 24154157875.2,
              "median": 24178065408.0,
              "minimum": 24110956544.0,
              "sample_count": 5
            },
            "gpu_sm_clock_mhz": {
              "maximum": 2272.0,
              "mean": 2264.6,
              "median": 2272.0,
              "minimum": 2235.0,
              "sample_count": 5
            },
            "gpu_utilization_percent": {
              "maximum": 1.0,
              "mean": 0.2,
              "median": 0.0,
              "minimum": 0.0,
              "sample_count": 5
            },
            "host_mem_available_bytes": {
              "maximum": 94570885120,
              "mean": 94556760064,
              "median": 94556760064.0,
              "minimum": 94542635008,
              "sample_count": 2
            },
            "process_pss_bytes": {
              "maximum": 24359494656,
              "mean": 24305228800,
              "median": 24305228800.0,
              "minimum": 24250962944,
              "sample_count": 2
            },
            "process_rss_bytes": {
              "maximum": 24362508288,
              "mean": 24308242432,
              "median": 24308242432.0,
              "minimum": 24253976576,
              "sample_count": 2
            }
          },
          "process_sample_count": 2,
          "response_union_seconds": 3.400849792989902,
          "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": 236537.1246580265,
            "reason": null,
            "response_union_seconds": 2749.6215623870376,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 86.02533813878037,
            "trapezoid_pieces": 2752,
            "watt_hours": 65.70475684945181
          },
          "gpu_sample_count": 2750,
          "host_sample_count": 1622,
          "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": 2323677184,
              "mean": 2286969816.8581996,
              "median": 2292627456.0,
              "minimum": 2259800064,
              "sample_count": 1622
            },
            "gpu_board_watts": {
              "maximum": 152.49,
              "mean": 86.01490909090909,
              "median": 81.61,
              "minimum": 79.5,
              "sample_count": 2750
            },
            "gpu_memory_used_bytes": {
              "maximum": 24192745472.0,
              "mean": 24186094068.456726,
              "median": 24192745472.0,
              "minimum": 24178065408.0,
              "sample_count": 2750
            },
            "gpu_sm_clock_mhz": {
              "maximum": 2332.0,
              "mean": 2263.0723636363637,
              "median": 2272.0,
              "minimum": 1785.0,
              "sample_count": 2750
            },
            "gpu_utilization_percent": {
              "maximum": 100.0,
              "mean": 11.226545454545455,
              "median": 0.0,
              "minimum": 0.0,
              "sample_count": 2750
            },
            "host_mem_available_bytes": {
              "maximum": 94687850496,
              "mean": 94599713823.56596,
              "median": 94597986304.0,
              "minimum": 94527299584,
              "sample_count": 1622
            },
            "process_pss_bytes": {
              "maximum": 24359494656,
              "mean": 24348625269.10974,
              "median": 24354731008.0,
              "minimum": 24331427840,
              "sample_count": 1622
            },
            "process_rss_bytes": {
              "maximum": 24362508288,
              "mean": 24351710850.683105,
              "median": 24357744640.0,
              "minimum": 24335863808,
              "sample_count": 1622
            }
          },
          "process_sample_count": 1622,
          "response_union_seconds": 2749.6215623870376,
          "sampled_counter_deltas": {
            "deltas": {
              "major_faults": 0,
              "read_syscall_bytes": 0,
              "storage_read_bytes": 0,
              "storage_write_bytes": 1261568
            },
            "observed_endpoint_seconds": 2746.6284927970264,
            "observed_intervals": 2,
            "requested_intervals": 2
          }
        },
        "warmup": {
          "gpu_energy": {
            "available": true,
            "interpolation": "linear_at_bracketed_boundaries",
            "joules": 46701.3050646631,
            "reason": null,
            "response_union_seconds": 550.41680468095,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 84.84716430802578,
            "trapezoid_pieces": 551,
            "watt_hours": 12.972584740184194
          },
          "gpu_sample_count": 549,
          "host_sample_count": 327,
          "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": 2307465216,
              "mean": 2272184376.3669724,
              "median": 2278948864,
              "minimum": 2212904960,
              "sample_count": 327
            },
            "gpu_board_watts": {
              "maximum": 152.14,
              "mean": 84.84987249544626,
              "median": 80.91,
              "minimum": 76.67,
              "sample_count": 549
            },
            "gpu_memory_used_bytes": {
              "maximum": 24192745472.0,
              "mean": 24185243092.167576,
              "median": 24192745472.0,
              "minimum": 24142413824.0,
              "sample_count": 549
            },
            "gpu_sm_clock_mhz": {
              "maximum": 2332.0,
              "mean": 2263.2768670309656,
              "median": 2272.0,
              "minimum": 1792.0,
              "sample_count": 549
            },
            "gpu_utilization_percent": {
              "maximum": 96.0,
              "mean": 11.145719489981785,
              "median": 0.0,
              "minimum": 0.0,
              "sample_count": 549
            },
            "host_mem_available_bytes": {
              "maximum": 94694584320,
              "mean": 94598962323.18044,
              "median": 94595268608,
              "minimum": 94545797120,
              "sample_count": 327
            },
            "process_pss_bytes": {
              "maximum": 24354034688,
              "mean": 24339138647.681957,
              "median": 24342766592,
              "minimum": 24285017088,
              "sample_count": 327
            },
            "process_rss_bytes": {
              "maximum": 24357048320,
              "mean": 24342234703.85321,
              "median": 24345780224,
              "minimum": 24288030720,
              "sample_count": 327
            }
          },
          "process_sample_count": 327,
          "response_union_seconds": 550.41680468095,
          "sampled_counter_deltas": {
            "deltas": {
              "major_faults": 0,
              "read_syscall_bytes": 0,
              "storage_read_bytes": 0,
              "storage_write_bytes": 249856
            },
            "observed_endpoint_seconds": 548.5689892289811,
            "observed_intervals": 2,
            "requested_intervals": 2
          }
        }
      },
      "reviewed_export_sha256": "4b17dd2a9c39d7cab209a102bfdbd49f8324b54625bc01cbe7d9aeaf6b87dd5a",
      "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": 40
        },
        "reasoning": "not_applicable_raw_completion",
        "seed": 20260927,
        "server_slots": 1,
        "synthetic_prompt_sha256": "c099b5abc034050fac1133e0214990d1f72a124b64f36e8a80ba3209bc07d187",
        "temperature": 0,
        "threads": 16,
        "tokenization": "model_tokenizer_add_special_false_repeat_then_truncate",
        "total_context_tokens": 8704,
        "ubatch": 512,
        "warmup_rounds": 1
      },
      "status": "completed",
      "telemetry": {
        "gpu_samples": 3313,
        "host_samples": 1959,
        "maximum_gpu_gap_seconds": 1.0563414199859835,
        "maximum_host_gap_seconds": 1.9819231039728038,
        "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": 3313,
            "unavailable": 0
          },
          "software": {
            "active": 0,
            "not_active": 3313,
            "unavailable": 0
          }
        }
      }
    },
    {
      "case_wall_seconds": 4961.106233133993,
      "completed_response_resources": {
        "gpu_energy": {
          "available": true,
          "interpolation": "linear_at_bracketed_boundaries",
          "joules": 326983.819602327,
          "reason": null,
          "response_union_seconds": 4008.546728375077,
          "scope": "GPU board only; not whole-server AC energy",
          "time_weighted_mean_watts": 81.57166219062006,
          "trapezoid_pieces": 4019,
          "watt_hours": 90.82883877842417
        },
        "gpu_sample_count": 4009,
        "host_sample_count": 2035,
        "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": 3084062720,
            "mean": 3045302493.4054055,
            "median": 3055714304,
            "minimum": 3018514432,
            "sample_count": 2035
          },
          "gpu_board_watts": {
            "maximum": 129.75,
            "mean": 81.56763532052881,
            "median": 78.23,
            "minimum": 74.81,
            "sample_count": 4009
          },
          "gpu_memory_used_bytes": {
            "maximum": 13128171520.0,
            "mean": 13116759330.418558,
            "median": 13128171520.0,
            "minimum": 13103005696.0,
            "sample_count": 4009
          },
          "gpu_sm_clock_mhz": {
            "maximum": 2332.0,
            "mean": 2271.545522574208,
            "median": 2272.0,
            "minimum": 2100.0,
            "sample_count": 4009
          },
          "gpu_utilization_percent": {
            "maximum": 100.0,
            "mean": 11.124220503866301,
            "median": 0.0,
            "minimum": 0.0,
            "sample_count": 4009
          },
          "host_mem_available_bytes": {
            "maximum": 93943697408,
            "mean": 93853951377.0457,
            "median": 93853831168,
            "minimum": 93781438464,
            "sample_count": 2035
          },
          "process_pss_bytes": {
            "maximum": 35444739072,
            "mean": 35430992087.36707,
            "median": 35441887232,
            "minimum": 35412776960,
            "sample_count": 2035
          },
          "process_rss_bytes": {
            "maximum": 35447758848,
            "mean": 35434078916.4973,
            "median": 35444908032,
            "minimum": 35416961024,
            "sample_count": 2035
          }
        },
        "process_sample_count": 2035,
        "response_union_seconds": 4008.546728375077,
        "sampled_counter_deltas": {
          "deltas": {
            "major_faults": 0,
            "read_syscall_bytes": 0,
            "storage_read_bytes": 0,
            "storage_write_bytes": 1728512
          },
          "observed_endpoint_seconds": 3988.5196464579785,
          "observed_intervals": 10,
          "requested_intervals": 10
        }
      },
      "groups": [
        {
          "aggregate_output_tokens_per_second": 0.7043362587375134,
          "completed_output_tokens": 1280,
          "completed_request_count": 5,
          "context_tokens": 2048,
          "cpu_snapshot_intervals": {
            "host_capacity_percent": 47.48170472432676,
            "interval_count": 5,
            "interval_seconds": 1817.331549088005,
            "meaning": "Weighted CPU counter rate across round-bracketing snapshots, not exact token time.",
            "one_logical_cpu_percent": 1519.4145511784563,
            "process_cpu_seconds": 27612.8
          },
          "failed_request_count": 0,
          "measured_rounds": 5,
          "mode": "completion",
          "resources": {
            "gpu_energy": {
              "available": true,
              "interpolation": "linear_at_bracketed_boundaries",
              "joules": 144503.54057542377,
              "reason": null,
              "response_union_seconds": 1817.31379596208,
              "scope": "GPU board only; not whole-server AC energy",
              "time_weighted_mean_watts": 79.5149087056394,
              "trapezoid_pieces": 1823,
              "watt_hours": 40.13987238206216
            },
            "gpu_sample_count": 1818,
            "host_sample_count": 890,
            "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": 3044786176,
                "mean": 3024263241.635955,
                "median": 3023706112.0,
                "minimum": 3018514432,
                "sample_count": 890
              },
              "gpu_board_watts": {
                "maximum": 125.38,
                "mean": 79.50995599559955,
                "median": 78.14500000000001,
                "minimum": 75.15,
                "sample_count": 1818
              },
              "gpu_memory_used_bytes": {
                "maximum": 13103005696.0,
                "mean": 13103005696.0,
                "median": 13103005696.0,
                "minimum": 13103005696.0,
                "sample_count": 1818
              },
              "gpu_sm_clock_mhz": {
                "maximum": 2325.0,
                "mean": 2270.7398239823983,
                "median": 2272.0,
                "minimum": 2122.0,
                "sample_count": 1818
              },
              "gpu_utilization_percent": {
                "maximum": 100.0,
                "mean": 5.860836083608361,
                "median": 0.0,
                "minimum": 0.0,
                "sample_count": 1818
              },
              "host_mem_available_bytes": {
                "maximum": 93943697408,
                "mean": 93876701759.2809,
                "median": 93871968256.0,
                "minimum": 93820104704,
                "sample_count": 890
              },
              "process_pss_bytes": {
                "maximum": 35415967744,
                "mean": 35415209832.12584,
                "median": 35415275520.0,
                "minimum": 35412776960,
                "sample_count": 890
              },
              "process_rss_bytes": {
                "maximum": 35418988544,
                "mean": 35418297640.84494,
                "median": 35418296320.0,
                "minimum": 35416961024,
                "sample_count": 890
              }
            },
            "process_sample_count": 890,
            "response_union_seconds": 1817.31379596208,
            "sampled_counter_deltas": {
              "deltas": {
                "major_faults": 0,
                "read_syscall_bytes": 0,
                "storage_read_bytes": 0,
                "storage_write_bytes": 806912
              },
              "observed_endpoint_seconds": 1805.0209762880113,
              "observed_intervals": 5,
              "requested_intervals": 5
            }
          },
          "response_outer_span_seconds": 1817.338583304023,
          "response_seconds": {
            "maximum": 363.6143847160274,
            "mean": 363.462759192416,
            "median": 363.4720513680368,
            "minimum": 363.25934934196994,
            "sample_count": 5
          },
          "response_window_union_seconds": 1817.31379596208,
          "server_decode_tokens_per_second": {
            "maximum": 0.7378754690175205,
            "mean": 0.7375854372949258,
            "median": 0.7376701920484424,
            "minimum": 0.7372008255111242,
            "sample_count": 5
          },
          "synthetic_first_token_seconds": {
            "maximum": 17.720808499027044,
            "mean": 17.69595089161303,
            "median": 17.700019541021902,
            "minimum": 17.672509245981928,
            "sample_count": 5
          },
          "warmup_completed_rounds": 1,
          "warmup_requests_per_round": 1
        },
        {
          "aggregate_output_tokens_per_second": 0.5841460216602611,
          "completed_output_tokens": 1280,
          "completed_request_count": 5,
          "context_tokens": 8192,
          "cpu_snapshot_intervals": {
            "host_capacity_percent": 42.039219255208145,
            "interval_count": 5,
            "interval_seconds": 2191.255163203052,
            "meaning": "Weighted CPU counter rate across round-bracketing snapshots, not exact token time.",
            "one_logical_cpu_percent": 1345.2550161666607,
            "process_cpu_seconds": 29477.97
          },
          "failed_request_count": 0,
          "measured_rounds": 5,
          "mode": "completion",
          "resources": {
            "gpu_energy": {
              "available": true,
              "interpolation": "linear_at_bracketed_boundaries",
              "joules": 182480.27902690295,
              "reason": null,
              "response_union_seconds": 2191.232932412997,
              "scope": "GPU board only; not whole-server AC energy",
              "time_weighted_mean_watts": 83.27744455079667,
              "trapezoid_pieces": 2196,
              "watt_hours": 50.688966396361934
            },
            "gpu_sample_count": 2191,
            "host_sample_count": 1145,
            "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": 3084062720,
                "mean": 3061656147.619214,
                "median": 3061407744,
                "minimum": 3053445120,
                "sample_count": 1145
              },
              "gpu_board_watts": {
                "maximum": 129.75,
                "mean": 83.27501141031493,
                "median": 78.35,
                "minimum": 74.81,
                "sample_count": 2191
              },
              "gpu_memory_used_bytes": {
                "maximum": 13128171520.0,
                "mean": 13128171520.0,
                "median": 13128171520.0,
                "minimum": 13128171520.0,
                "sample_count": 2191
              },
              "gpu_sm_clock_mhz": {
                "maximum": 2332.0,
                "mean": 2272.214057507987,
                "median": 2272.0,
                "minimum": 2100.0,
                "sample_count": 2191
              },
              "gpu_utilization_percent": {
                "maximum": 100.0,
                "mean": 15.491556366955727,
                "median": 0.0,
                "minimum": 0.0,
                "sample_count": 2191
              },
              "host_mem_available_bytes": {
                "maximum": 93916975104,
                "mean": 93836267673.82358,
                "median": 93834838016,
                "minimum": 93781438464,
                "sample_count": 1145
              },
              "process_pss_bytes": {
                "maximum": 35444739072,
                "mean": 35443259517.20524,
                "median": 35443341312,
                "minimum": 35440470016,
                "sample_count": 1145
              },
              "process_rss_bytes": {
                "maximum": 35447758848,
                "mean": 35446345584.908295,
                "median": 35446362112,
                "minimum": 35444199424,
                "sample_count": 1145
              }
            },
            "process_sample_count": 1145,
            "response_union_seconds": 2191.232932412997,
            "sampled_counter_deltas": {
              "deltas": {
                "major_faults": 0,
                "read_syscall_bytes": 0,
                "storage_read_bytes": 0,
                "storage_write_bytes": 921600
              },
              "observed_endpoint_seconds": 2183.498670169967,
              "observed_intervals": 5,
              "requested_intervals": 5
            }
          },
          "response_outer_span_seconds": 2191.2666058779578,
          "response_seconds": {
            "maximum": 438.39019691804424,
            "mean": 438.24658648259935,
            "median": 438.2723452189821,
            "minimum": 438.1088442449691,
            "sample_count": 5
          },
          "response_window_union_seconds": 2191.232932412997,
          "server_decode_tokens_per_second": {
            "maximum": 0.6975875589334964,
            "mean": 0.6973100693905205,
            "median": 0.6972552794747908,
            "minimum": 0.6971042864759415,
            "sample_count": 5
          },
          "synthetic_first_token_seconds": {
            "maximum": 72.59099566203076,
            "mean": 72.55541585101746,
            "median": 72.55246841500048,
            "minimum": 72.52375948702684,
            "sample_count": 5
          },
          "warmup_completed_rounds": 1,
          "warmup_requests_per_round": 1
        }
      ],
      "id": "offload-72b-20",
      "model": "qwen25-72b-q4",
      "module": "dense-offload",
      "placement": {
        "buffers": [
          {
            "device": "CPU_Mapped",
            "reported_mib": 3796.51
          },
          {
            "device": "CPU_Mapped",
            "reported_mib": 3698.21
          },
          {
            "device": "CPU_Mapped",
            "reported_mib": 3780.0
          },
          {
            "device": "CPU_Mapped",
            "reported_mib": 3758.87
          },
          {
            "device": "CPU_Mapped",
            "reported_mib": 3780.06
          },
          {
            "device": "CPU_Mapped",
            "reported_mib": 3758.87
          },
          {
            "device": "CPU_Mapped",
            "reported_mib": 3649.56
          },
          {
            "device": "CPU_Mapped",
            "reported_mib": 3799.34
          },
          {
            "device": "CPU_Mapped",
            "reported_mib": 1180.14
          },
          {
            "device": "CUDA0",
            "reported_mib": 10764.37
          }
        ],
        "gpu_layers": 20,
        "meaning": "Runtime buffer allocation is not measured resident process RAM.",
        "requested_cpu_expert_layers": 0,
        "total_layers": 81
      },
      "record_sha256": "79ea9153d287fb1c63a4773c4e576d5dead6f1aca8f6f5fadd95c4ed32e82ff3",
      "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": 411516928,
              "mean": 402950144,
              "median": 402950144.0,
              "minimum": 394383360,
              "sample_count": 2
            },
            "gpu_board_watts": {
              "maximum": 16.41,
              "mean": 16.395,
              "median": 16.395,
              "minimum": 16.38,
              "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": 176.0,
              "median": 176.0,
              "minimum": 172.0,
              "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": 96501207040,
              "mean": 96496873472,
              "median": 96496873472.0,
              "minimum": 96492539904,
              "sample_count": 2
            },
            "process_pss_bytes": null,
            "process_rss_bytes": null
          },
          "process_sample_count": 0,
          "response_union_seconds": 1.1008169630076736,
          "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": 650592256,
              "mean": 650592256,
              "median": 650592256,
              "minimum": 650592256,
              "sample_count": 1
            },
            "gpu_board_watts": {
              "maximum": 77.62,
              "mean": 77.62,
              "median": 77.62,
              "minimum": 77.62,
              "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": 2272.0,
              "mean": 2272.0,
              "median": 2272.0,
              "minimum": 2272.0,
              "sample_count": 1
            },
            "gpu_utilization_percent": {
              "maximum": 25.0,
              "mean": 25.0,
              "median": 25.0,
              "minimum": 25.0,
              "sample_count": 1
            },
            "host_mem_available_bytes": {
              "maximum": 96264818688,
              "mean": 96264818688,
              "median": 96264818688,
              "minimum": 96264818688,
              "sample_count": 1
            },
            "process_pss_bytes": null,
            "process_rss_bytes": null
          },
          "process_sample_count": 0,
          "response_union_seconds": 1.1522748679853976,
          "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": 447.90588874900686,
            "reason": null,
            "response_union_seconds": 6.576283341972157,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 68.10927471605636,
            "trapezoid_pieces": 7,
            "watt_hours": 0.12441830243027968
          },
          "gpu_sample_count": 6,
          "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": 2943393792,
              "mean": 1787288576,
              "median": 1792483328.0,
              "minimum": 620793856,
              "sample_count": 4
            },
            "gpu_board_watts": {
              "maximum": 78.7,
              "mean": 71.50666666666666,
              "median": 74.795,
              "minimum": 52.23,
              "sample_count": 6
            },
            "gpu_memory_used_bytes": {
              "maximum": 13014925312.0,
              "mean": 10551121237.333334,
              "median": 12406751232.0,
              "minimum": 587202560.0,
              "sample_count": 6
            },
            "gpu_sm_clock_mhz": {
              "maximum": 2310.0,
              "mean": 2289.6666666666665,
              "median": 2294.5,
              "minimum": 2265.0,
              "sample_count": 6
            },
            "gpu_utilization_percent": {
              "maximum": 83.0,
              "mean": 27.666666666666668,
              "median": 0.5,
              "minimum": 0.0,
              "sample_count": 6
            },
            "host_mem_available_bytes": {
              "maximum": 96198549504,
              "mean": 95033121792,
              "median": 95025416192.0,
              "minimum": 93883105280,
              "sample_count": 4
            },
            "process_pss_bytes": {
              "maximum": 44347988992,
              "mean": 31105929216,
              "median": 35304272896.0,
              "minimum": 9467182080,
              "sample_count": 4
            },
            "process_rss_bytes": {
              "maximum": 44350955520,
              "mean": 31108922368,
              "median": 35307292672.0,
              "minimum": 9470148608,
              "sample_count": 4
            }
          },
          "process_sample_count": 4,
          "response_union_seconds": 6.576283341972157,
          "sampled_counter_deltas": {
            "deltas": {
              "major_faults": 0,
              "read_syscall_bytes": 13,
              "storage_read_bytes": 0,
              "storage_write_bytes": 4096
            },
            "observed_endpoint_seconds": 5.295854870986659,
            "observed_intervals": 1,
            "requested_intervals": 1
          }
        },
        "ready": {
          "gpu_energy": {
            "available": true,
            "interpolation": "linear_at_bracketed_boundaries",
            "joules": 264.97328211031623,
            "reason": null,
            "response_union_seconds": 3.4458548309630714,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 76.8962405872042,
            "trapezoid_pieces": 6,
            "watt_hours": 0.07360368947508784
          },
          "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": 3067142144,
              "mean": 3007987712,
              "median": 3007987712.0,
              "minimum": 2948833280,
              "sample_count": 2
            },
            "gpu_board_watts": {
              "maximum": 78.24,
              "mean": 77.15,
              "median": 78.06,
              "minimum": 75.15,
              "sample_count": 3
            },
            "gpu_memory_used_bytes": {
              "maximum": 13128171520.0,
              "mean": 13082034176.0,
              "median": 13103005696.0,
              "minimum": 13014925312.0,
              "sample_count": 3
            },
            "gpu_sm_clock_mhz": {
              "maximum": 2272.0,
              "mean": 2272.0,
              "median": 2272.0,
              "minimum": 2272.0,
              "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": 93878493184,
              "mean": 93867741184,
              "median": 93867741184.0,
              "minimum": 93856989184,
              "sample_count": 2
            },
            "process_pss_bytes": {
              "maximum": 35444738048,
              "mean": 35390990848,
              "median": 35390990848.0,
              "minimum": 35337243648,
              "sample_count": 2
            },
            "process_rss_bytes": {
              "maximum": 35447758848,
              "mean": 35394011136,
              "median": 35394011136.0,
              "minimum": 35340263424,
              "sample_count": 2
            }
          },
          "process_sample_count": 2,
          "response_union_seconds": 3.4458548309630714,
          "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": 326989.6537589074,
            "reason": null,
            "response_union_seconds": 4008.619653561036,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 81.57163363414332,
            "trapezoid_pieces": 4011,
            "watt_hours": 90.83045937747428
          },
          "gpu_sample_count": 4009,
          "host_sample_count": 2038,
          "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": 3084062720,
              "mean": 3045271785.138371,
              "median": 3055710208.0,
              "minimum": 3018514432,
              "sample_count": 2038
            },
            "gpu_board_watts": {
              "maximum": 129.75,
              "mean": 81.56763532052881,
              "median": 78.23,
              "minimum": 74.81,
              "sample_count": 4009
            },
            "gpu_memory_used_bytes": {
              "maximum": 13128171520.0,
              "mean": 13116759330.418558,
              "median": 13128171520.0,
              "minimum": 13103005696.0,
              "sample_count": 4009
            },
            "gpu_sm_clock_mhz": {
              "maximum": 2332.0,
              "mean": 2271.545522574208,
              "median": 2272.0,
              "minimum": 2100.0,
              "sample_count": 4009
            },
            "gpu_utilization_percent": {
              "maximum": 100.0,
              "mean": 11.124220503866301,
              "median": 0.0,
              "minimum": 0.0,
              "sample_count": 4009
            },
            "host_mem_available_bytes": {
              "maximum": 93943697408,
              "mean": 93854005643.93327,
              "median": 93853870080.0,
              "minimum": 93781438464,
              "sample_count": 2038
            },
            "process_pss_bytes": {
              "maximum": 35444739072,
              "mean": 35430969175.17566,
              "median": 35441887232.0,
              "minimum": 35412776960,
              "sample_count": 2038
            },
            "process_rss_bytes": {
              "maximum": 35447758848,
              "mean": 35434055907.10893,
              "median": 35444908032.0,
              "minimum": 35416961024,
              "sample_count": 2038
            }
          },
          "process_sample_count": 2038,
          "response_union_seconds": 4008.619653561036,
          "sampled_counter_deltas": {
            "deltas": {
              "major_faults": 0,
              "read_syscall_bytes": 0,
              "storage_read_bytes": 0,
              "storage_write_bytes": 1781760
            },
            "observed_endpoint_seconds": 4006.5016962169902,
            "observed_intervals": 2,
            "requested_intervals": 2
          }
        },
        "warmup": {
          "gpu_energy": {
            "available": true,
            "interpolation": "linear_at_bracketed_boundaries",
            "joules": 65077.663397590804,
            "reason": null,
            "response_union_seconds": 802.8343216970097,
            "scope": "GPU board only; not whole-server AC energy",
            "time_weighted_mean_watts": 81.05989198373007,
            "trapezoid_pieces": 805,
            "watt_hours": 18.077128721553002
          },
          "gpu_sample_count": 803,
          "host_sample_count": 409,
          "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": 3071012864,
              "mean": 3032417838.317848,
              "median": 3029204992,
              "minimum": 2971230208,
              "sample_count": 409
            },
            "gpu_board_watts": {
              "maximum": 129.58,
              "mean": 81.05648816936488,
              "median": 77.86,
              "minimum": 72.65,
              "sample_count": 803
            },
            "gpu_memory_used_bytes": {
              "maximum": 13128171520.0,
              "mean": 13115147239.770859,
              "median": 13126074368.0,
              "minimum": 13044285440.0,
              "sample_count": 803
            },
            "gpu_sm_clock_mhz": {
              "maximum": 2325.0,
              "mean": 2270.7372353673723,
              "median": 2272.0,
              "minimum": 2107.0,
              "sample_count": 803
            },
            "gpu_utilization_percent": {
              "maximum": 100.0,
              "mean": 11.643835616438356,
              "median": 0.0,
              "minimum": 0.0,
              "sample_count": 803
            },
            "host_mem_available_bytes": {
              "maximum": 93974298624,
              "mean": 93874269837.45721,
              "median": 93873455104,
              "minimum": 93823176704,
              "sample_count": 409
            },
            "process_pss_bytes": {
              "maximum": 35441178624,
              "mean": 35423200719.17848,
              "median": 35416747008,
              "minimum": 35365141504,
              "sample_count": 409
            },
            "process_rss_bytes": {
              "maximum": 35444199424,
              "mean": 35426312392.293396,
              "median": 35419983872,
              "minimum": 35368161280,
              "sample_count": 409
            }
          },
          "process_sample_count": 409,
          "response_union_seconds": 802.8343216970097,
          "sampled_counter_deltas": {
            "deltas": {
              "major_faults": 0,
              "read_syscall_bytes": 0,
              "storage_read_bytes": 0,
              "storage_write_bytes": 339968
            },
            "observed_endpoint_seconds": 799.775153718947,
            "observed_intervals": 2,
            "requested_intervals": 2
          }
        }
      },
      "reviewed_export_sha256": "4b17dd2a9c39d7cab209a102bfdbd49f8324b54625bc01cbe7d9aeaf6b87dd5a",
      "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": 20
        },
        "reasoning": "not_applicable_raw_completion",
        "seed": 20260927,
        "server_slots": 1,
        "synthetic_prompt_sha256": "c099b5abc034050fac1133e0214990d1f72a124b64f36e8a80ba3209bc07d187",
        "temperature": 0,
        "threads": 16,
        "tokenization": "model_tokenizer_add_special_false_repeat_then_truncate",
        "total_context_tokens": 8704,
        "ubatch": 512,
        "warmup_rounds": 1
      },
      "status": "completed",
      "telemetry": {
        "gpu_samples": 4824,
        "host_samples": 2456,
        "maximum_gpu_gap_seconds": 1.0568395520094782,
        "maximum_host_gap_seconds": 2.3296370519674383,
        "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": 4824,
            "unavailable": 0
          },
          "software": {
            "active": 0,
            "not_active": 4824,
            "unavailable": 0
          }
        }
      }
    }
  ],
  "comparisons": [
    {
      "axis": "gpu_layers",
      "cases": [
        "offload-72b-full",
        "offload-72b-40",
        "offload-72b-20"
      ]
    }
  ],
  "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"
    },
    {
      "architecture": "moe",
      "chat_template_sha256": "a4c9919cbbd4acdd51ccffe22da049264b1b73e59055fa58811a99efbd7c8146",
      "effective_chat_template_rule": "llama.cpp-7fe450e19-chat-source-v1",
      "effective_chat_template_sha256": "b2215de6da8ba369957eece8c5aa18f4af94f6c4d311d85e691373a421d80e89",
      "files": [
        {
          "bytes": 63387346208,
          "sha256": "582bd40f6886200101f4c4ed9f25f3fe80cc14c86e9e2b37746cd8904a0c622d"
        }
      ],
      "id": "gpt-oss-120b-mxfp4",
      "name": "GPT-OSS-120B",
      "quantization": "MXFP4",
      "repository": "ggml-org/gpt-oss-120b-GGUF",
      "revision": "238abdd290bb874b90a5da1b4549881b7d05c091",
      "tokenizer_sha256": "d67292e223a919a15af6ec5257ae8e5e6c6782203636375f7008603d27b7a391"
    },
    {
      "architecture": "dense",
      "chat_template_sha256": "d5495a1e5db0611132a97e46a65dbb64a642a499421228b9c8b93229097fa9a4",
      "effective_chat_template_rule": "llama.cpp-7fe450e19-chat-source-v1",
      "effective_chat_template_sha256": "4e9918361c284a93880606d182d64da6a9fe97cdc1f5c5a78c1c8840444246fc",
      "files": [
        {
          "bytes": 3878876224,
          "sha256": "31a2b3c4e62dc185b54b9c33b1b2898f06fde167e3311cc5380e08692098e781"
        },
        {
          "bytes": 3743845120,
          "sha256": "747e66c38a1f4f8d496533e2108c7592e10220afc65c480d13cbe14d0109d700"
        },
        {
          "bytes": 3743845152,
          "sha256": "4346236ce045e136ad4cbcd7b86a88ae66dfadeed033175965e9f331a76e90d9"
        },
        {
          "bytes": 3743845184,
          "sha256": "152f922d2dd4c18d49f6b3c00bb1cc20629ed042d494527e1c350da41043a446"
        },
        {
          "bytes": 3743845184,
          "sha256": "24d6a097d3d3ce727b1f123957713cdaba0d5d64b9aaf559afb4703ea51ce577"
        },
        {
          "bytes": 3743845184,
          "sha256": "4ef22906aa68f1b7c067c4469677d7490455a9ea1a2fcda67af29bb7d1ffc49e"
        },
        {
          "bytes": 3743845184,
          "sha256": "b9dd79952b0a7d771633f32c253da0b771eb09efe173eeef50c250979c5ddd22"
        },
        {
          "bytes": 3743845184,
          "sha256": "af925e8985c7d5e30d6ee98e275d93cdc2ca5777173d5601ebc9d6fbd4f4d01e"
        },
        {
          "bytes": 3743845184,
          "sha256": "057a909a1bfd954aaf9b4b94b226b98a1a7b2a5b19861a708bf2879bb12c704e"
        },
        {
          "bytes": 3743845184,
          "sha256": "c84a7441c2feab7213ee35b989334079276701a4bc39017e68ac7096966a08b6"
        },
        {
          "bytes": 3743845184,
          "sha256": "48f91fc504df34b89edc7d709925b9b69900a3a3671b1b4ce8d7384ff3117255"
        },
        {
          "bytes": 3743845184,
          "sha256": "84c3b6949803166b84bb8ffda810fbf2f56aec4e32b9e0f1614a43f84d4214a2"
        },
        {
          "bytes": 3743845184,
          "sha256": "00fd616480122ef2fddf3888436ee62d04ec5dc081cad0f8d0db9e7f1b241faf"
        },
        {
          "bytes": 3743845184,
          "sha256": "58a93e4a25a3bd34ae4327f4b40d1fd7c90b20ae34ed29a5fc0804333031d998"
        },
        {
          "bytes": 3743845184,
          "sha256": "f3b643f7b02f52da7a68ab0eee711ba160904ed235bea64da6cca5b46bb5db48"
        },
        {
          "bytes": 3743845184,
          "sha256": "721536952fea2b497e74681da2c65b1bd783ef2f3193ab41057fcaa31a64b495"
        },
        {
          "bytes": 3743845184,
          "sha256": "3b21b0bd05319458227e5865d1a79e62e1d63168bf8060e8d0d6f1e0da51c324"
        },
        {
          "bytes": 3743845184,
          "sha256": "1804cb39eefead3acd09ffafd064d969ec07161aceb62edbbe8975a0472391eb"
        },
        {
          "bytes": 3743845184,
          "sha256": "a6c8e5f402cb84f8d29813f811cb93c3a694dfc8a6398e2081031ced3a49b5ad"
        },
        {
          "bytes": 3743845184,
          "sha256": "047d7b4c5c363d17e86f19b3bd66e38663cb1484555aa897cd4892c1993881d3"
        },
        {
          "bytes": 2518066176,
          "sha256": "b6a7a5be990bed3f18695ded8d8fca3f545176045a37784e0443e2bd6633c6bc"
        }
      ],
      "id": "qwen25-72b-q8",
      "name": "Qwen2.5-72B-Instruct",
      "quantization": "Q8_0",
      "repository": "Qwen/Qwen2.5-72B-Instruct-GGUF",
      "revision": "7ca3bc388f97b264c4283bc9bf1055e2abc38441",
      "tokenizer_sha256": "67b3d77a05acfb80f96742cec2740cc6ee8fffadef98e655e855baf58a3946b1"
    },
    {
      "architecture": "moe",
      "chat_template_sha256": "57f1fd00f0013a2be96aa79b857391f27e23df5b5f847072b524c897e24d0361",
      "effective_chat_template_rule": "llama.cpp-7fe450e19-chat-source-v1",
      "effective_chat_template_sha256": "57f1fd00f0013a2be96aa79b857391f27e23df5b5f847072b524c897e24d0361",
      "files": [
        {
          "bytes": 29673975136,
          "sha256": "706382f26d0d6f72cd89c9ac1bfeb40d4b4afaee986dad2ea229ea24f9168fab"
        },
        {
          "bytes": 29748770496,
          "sha256": "9ab54360c3809c580439b75fe7e16f3af24adcc701c352e39b06ad557aca0ca3"
        },
        {
          "bytes": 29749049024,
          "sha256": "93c863b35e810b97605dc973c2881e608751a269e773b3772c42e5e25cb01d79"
        },
        {
          "bytes": 29642094272,
          "sha256": "c0322fa3f66ca392a9d7f95bb2543b6f6213ac04c16e613a2ae3a82429452071"
        },
        {
          "bytes": 29856282304,
          "sha256": "1dce96f79ba9c2d86d36bbd856101f7a2e4f3c4cff4ef4a2a970c0ba336edb15"
        },
        {
          "bytes": 18143985760,
          "sha256": "15a2192eadc2cc2da6fc0df6d25cb6d39626f292716526cfee8baba9043fbe62"
        }
      ],
      "id": "qwen3-235b-q5",
      "name": "Qwen3-235B-A22B",
      "quantization": "Q5_K_M",
      "repository": "Qwen/Qwen3-235B-A22B-GGUF",
      "revision": "211e807ecd6d37446a747409108783e39a04e80f",
      "tokenizer_sha256": "2629971f294a1a43ef545d689a2090a820e0e1a15e5371e1c88d1c21159ef849"
    }
  ],
  "protocol": "gsh-llm-core/1.1",
  "runtime": {
    "archives_sha256": [
      "c2ab9e19838513ff69d1af8d999ad717dd3c7ee4714ac04c7ed5ab9077c50e4e",
      "1466daea60aad1144819e151b2bae19d54556cf1da6c129c4f55a5ded2637c25"
    ],
    "binary_sha256": "376626f953efd7bbf16dee112747196d2b1de26717bcd1d37e203340dc5c043c",
    "commit": "7fe450e19305b828c199d602c23a8337aaa1f03b",
    "name": "llama.cpp",
    "release": "b11146"
  },
  "schema": "gsh-llm-core-article/1",
  "sources": [
    {
      "exporter_sha256": "4f4046b91d99ba35175c3c6c04473a0021f54c6baa4892625cb66435f4c7e2ec",
      "finished_at": "2026-09-27T13:53:28.036613+00:00",
      "plan_id": "core-pilot8-20260927t135100-pilot8",
      "plan_sha256": "2952d4f1f06fa8282ea9ea2126cf5954bf3d774d16f9bc55a160ed6d808c3d49",
      "protocol_sha256": "3ce8ce9f2b17e66a655c07fd7e7aa3f1762790ad9be38f87d29e5e1279dec3ac",
      "record_sha256": "6c09a429d3b733b934749454b62d1a4243f5dce7c508680647c7d703a2b5daca",
      "reviewed_export_sha256": "2d1a82644c4ba9ba504ddf8fd2c7fe3c253e15cc9811363c52186bd4b7510a07",
      "runner_sha256": "1bac2436c5aa2eca4f4b42cc9b8f0942b5c66256eb1a41a2169b88c12b044f4e",
      "started_at": "2026-09-27T13:51:57.017889+00:00",
      "task_corpus_sha256": "0642db889565ce25b980e24afa5048eb83e4d15c36754457474a65b2c1f71c21"
    },
    {
      "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": "2026-09-27T20:35:44.649240+00:00",
      "plan_id": "core-offload-20260927t180810-offloade234fdb2",
      "plan_sha256": "337114b4eef68828eabe560dad626826411d46979051d71e89bd51d5941dca27",
      "protocol_sha256": "3ce8ce9f2b17e66a655c07fd7e7aa3f1762790ad9be38f87d29e5e1279dec3ac",
      "record_sha256": "711f4d34a08d22e6937d72802601ed59fe398c64f9fb191925cb9eebb2cd6b2d",
      "reviewed_export_sha256": "4b17dd2a9c39d7cab209a102bfdbd49f8324b54625bc01cbe7d9aeaf6b87dd5a",
      "runner_sha256": "1bac2436c5aa2eca4f4b42cc9b8f0942b5c66256eb1a41a2169b88c12b044f4e",
      "started_at": "2026-09-27T18:08:57.430966+00:00",
      "task_corpus_sha256": "0642db889565ce25b980e24afa5048eb83e4d15c36754457474a65b2c1f71c21"
    },
    {
      "exporter_sha256": "4f4046b91d99ba35175c3c6c04473a0021f54c6baa4892625cb66435f4c7e2ec",
      "finished_at": "2026-09-27T16:13:37.029489+00:00",
      "plan_id": "core-capacity-20260927t151737-capacityb1f78281",
      "plan_sha256": "20b8022e204c32d4826a7ef00445ccd3e6fc8b0fe2bbccc222714e0fb4fbe2d3",
      "protocol_sha256": "3ce8ce9f2b17e66a655c07fd7e7aa3f1762790ad9be38f87d29e5e1279dec3ac",
      "record_sha256": "b80b2a18f3d821f8c7929b5d1d418771b14e88b3e73c0059056a1a8c31d33416",
      "reviewed_export_sha256": "f32d7fb2e6a87c7e3188b5b27d217655dd296aa783ea2397dc4822420a6d555c",
      "runner_sha256": "1bac2436c5aa2eca4f4b42cc9b8f0942b5c66256eb1a41a2169b88c12b044f4e",
      "started_at": "2026-09-27T15:24:32.393663+00:00",
      "task_corpus_sha256": "0642db889565ce25b980e24afa5048eb83e4d15c36754457474a65b2c1f71c21"
    }
  ],
  "status": "completed",
  "test_date_timezone": "Europe/Warsaw",
  "tested_at": "2026-09-27"
}
