[
  {
    "run": "llamacpp-20260916T052857Z",
    "length": 512,
    "clients": 1,
    "waves": 3,
    "warm_requests": 2,
    "correct_all": 3,
    "requests_all": 3,
    "errors_all": 0,
    "warm_median_ttft_s": 4.5358080139849335,
    "warm_median_request_s": 11.875004885485396,
    "warm_median_approx_decode_tps": 17.30962401495946,
    "warm_aggregate_completion_tps": 10.777941660269363,
    "completion_tokens": [
      128,
      128
    ],
    "reported_prompt_tokens": [
      512
    ],
    "first_wave_ttft_s": [
      5.867163714021444
    ],
    "memory": {
      "memory.jsonl": {
        "min_available_gib": 58.7812385559082,
        "new_swap_out_mib": 404.55859375,
        "samples": 27
      },
      "benchmark-memory.jsonl": {
        "min_available_gib": 56.82085418701172,
        "new_swap_out_mib": 1.48046875,
        "samples": 1829
      }
    }
  },
  {
    "run": "llamacpp-20260916T052857Z",
    "length": 512,
    "clients": 2,
    "waves": 3,
    "warm_requests": 4,
    "correct_all": 6,
    "requests_all": 6,
    "errors_all": 0,
    "warm_median_ttft_s": 9.092987607989926,
    "warm_median_request_s": 18.71886404947145,
    "warm_median_approx_decode_tps": 13.187744884659104,
    "warm_aggregate_completion_tps": 13.65163116570738,
    "completion_tokens": [
      128,
      128,
      128,
      128
    ],
    "reported_prompt_tokens": [
      512
    ],
    "first_wave_ttft_s": [
      9.170819582999684,
      9.409613159950823
    ],
    "memory": {
      "memory.jsonl": {
        "min_available_gib": 58.7812385559082,
        "new_swap_out_mib": 404.55859375,
        "samples": 27
      },
      "benchmark-memory.jsonl": {
        "min_available_gib": 56.82085418701172,
        "new_swap_out_mib": 1.48046875,
        "samples": 1829
      }
    }
  },
  {
    "run": "llamacpp-20260916T052857Z",
    "length": 8192,
    "clients": 1,
    "waves": 3,
    "warm_requests": 2,
    "correct_all": 3,
    "requests_all": 3,
    "errors_all": 0,
    "warm_median_ttft_s": 68.95881249551894,
    "warm_median_request_s": 75.16051003552275,
    "warm_median_approx_decode_tps": 14.753375544939228,
    "warm_aggregate_completion_tps": 1.2306818207350239,
    "completion_tokens": [
      94,
      91
    ],
    "reported_prompt_tokens": [
      8192
    ],
    "first_wave_ttft_s": [
      69.2894561389694
    ],
    "memory": {
      "memory.jsonl": {
        "min_available_gib": 58.7812385559082,
        "new_swap_out_mib": 404.55859375,
        "samples": 27
      },
      "benchmark-memory.jsonl": {
        "min_available_gib": 56.82085418701172,
        "new_swap_out_mib": 1.48046875,
        "samples": 1829
      }
    }
  },
  {
    "run": "llamacpp-20260916T052857Z",
    "length": 8192,
    "clients": 2,
    "waves": 3,
    "warm_requests": 4,
    "correct_all": 6,
    "requests_all": 6,
    "errors_all": 0,
    "warm_median_ttft_s": 111.09094998048386,
    "warm_median_request_s": 151.49147417000495,
    "warm_median_approx_decode_tps": 6.193822284387595,
    "warm_aggregate_completion_tps": 1.2900575863151054,
    "completion_tokens": [
      87,
      128,
      87,
      90
    ],
    "reported_prompt_tokens": [
      8192
    ],
    "first_wave_ttft_s": [
      140.60319655097555,
      81.53177824395243
    ],
    "memory": {
      "memory.jsonl": {
        "min_available_gib": 58.7812385559082,
        "new_swap_out_mib": 404.55859375,
        "samples": 27
      },
      "benchmark-memory.jsonl": {
        "min_available_gib": 56.82085418701172,
        "new_swap_out_mib": 1.48046875,
        "samples": 1829
      }
    }
  },
  {
    "run": "llamacpp-20260916T052857Z",
    "length": 32768,
    "clients": 1,
    "waves": 3,
    "warm_requests": 2,
    "correct_all": 3,
    "requests_all": 3,
    "errors_all": 0,
    "warm_median_ttft_s": 306.4235953700263,
    "warm_median_request_s": 316.8609014535323,
    "warm_median_approx_decode_tps": 12.16789413102633,
    "warm_aggregate_completion_tps": 0.4039603904359329,
    "completion_tokens": [
      128,
      128
    ],
    "reported_prompt_tokens": [
      32768
    ],
    "first_wave_ttft_s": [
      308.99571912607644
    ],
    "memory": {
      "memory.jsonl": {
        "min_available_gib": 58.7812385559082,
        "new_swap_out_mib": 404.55859375,
        "samples": 27
      },
      "benchmark-memory.jsonl": {
        "min_available_gib": 56.82085418701172,
        "new_swap_out_mib": 1.48046875,
        "samples": 1829
      }
    }
  },
  {
    "run": "llamacpp-20260916T052857Z",
    "length": 32768,
    "clients": 2,
    "waves": 3,
    "warm_requests": 4,
    "correct_all": 6,
    "requests_all": 6,
    "errors_all": 0,
    "warm_median_ttft_s": 471.8696552600013,
    "warm_median_request_s": 639.3831226835027,
    "warm_median_approx_decode_tps": 4.525568372549055,
    "warm_aggregate_completion_tps": 0.39876942643932967,
    "completion_tokens": [
      128,
      128,
      128,
      128
    ],
    "reported_prompt_tokens": [
      32768
    ],
    "first_wave_ttft_s": [
      629.0859888399718,
      321.07541169808246
    ],
    "memory": {
      "memory.jsonl": {
        "min_available_gib": 58.7812385559082,
        "new_swap_out_mib": 404.55859375,
        "samples": 27
      },
      "benchmark-memory.jsonl": {
        "min_available_gib": 56.82085418701172,
        "new_swap_out_mib": 1.48046875,
        "samples": 1829
      }
    }
  },
  {
    "run": "vllm-20260916T050513Z",
    "length": 512,
    "clients": 1,
    "waves": 3,
    "warm_requests": 2,
    "correct_all": 3,
    "requests_all": 3,
    "errors_all": 0,
    "warm_median_ttft_s": 1.4731673114583828,
    "warm_median_request_s": 10.99047077744035,
    "warm_median_approx_decode_tps": 13.344322195356657,
    "warm_aggregate_completion_tps": 11.64504204099251,
    "completion_tokens": [
      128,
      128
    ],
    "reported_prompt_tokens": [
      512
    ],
    "first_wave_ttft_s": [
      2.4855997799895704
    ],
    "memory": {
      "memory.jsonl": {
        "min_available_gib": 45.57258224487305,
        "new_swap_out_mib": 3.546875,
        "samples": 49
      },
      "benchmark-memory.jsonl": {
        "min_available_gib": 44.35332489013672,
        "new_swap_out_mib": 0.0,
        "samples": 553
      }
    }
  },
  {
    "run": "vllm-20260916T050513Z",
    "length": 512,
    "clients": 2,
    "waves": 3,
    "warm_requests": 4,
    "correct_all": 6,
    "requests_all": 6,
    "errors_all": 0,
    "warm_median_ttft_s": 2.219104228017386,
    "warm_median_request_s": 13.37563768599648,
    "warm_median_approx_decode_tps": 11.408705716396625,
    "warm_aggregate_completion_tps": 19.010505024703377,
    "completion_tokens": [
      128,
      128,
      128,
      128
    ],
    "reported_prompt_tokens": [
      512
    ],
    "first_wave_ttft_s": [
      3.891199383069761,
      2.4771758050192147
    ],
    "memory": {
      "memory.jsonl": {
        "min_available_gib": 45.57258224487305,
        "new_swap_out_mib": 3.546875,
        "samples": 49
      },
      "benchmark-memory.jsonl": {
        "min_available_gib": 44.35332489013672,
        "new_swap_out_mib": 0.0,
        "samples": 553
      }
    }
  },
  {
    "run": "vllm-20260916T050513Z",
    "length": 8192,
    "clients": 1,
    "waves": 3,
    "warm_requests": 2,
    "correct_all": 3,
    "requests_all": 3,
    "errors_all": 0,
    "warm_median_ttft_s": 19.70058439450804,
    "warm_median_request_s": 28.18176118802512,
    "warm_median_approx_decode_tps": 13.045932041423518,
    "warm_aggregate_completion_tps": 3.973961605675091,
    "completion_tokens": [
      96,
      128
    ],
    "reported_prompt_tokens": [
      8192
    ],
    "first_wave_ttft_s": [
      19.483398349955678
    ],
    "memory": {
      "memory.jsonl": {
        "min_available_gib": 45.57258224487305,
        "new_swap_out_mib": 3.546875,
        "samples": 49
      },
      "benchmark-memory.jsonl": {
        "min_available_gib": 44.35332489013672,
        "new_swap_out_mib": 0.0,
        "samples": 553
      }
    }
  },
  {
    "run": "vllm-20260916T050513Z",
    "length": 8192,
    "clients": 2,
    "waves": 3,
    "warm_requests": 4,
    "correct_all": 6,
    "requests_all": 6,
    "errors_all": 0,
    "warm_median_ttft_s": 29.898522776027676,
    "warm_median_request_s": 47.74231774604414,
    "warm_median_approx_decode_tps": 8.32174117701614,
    "warm_aggregate_completion_tps": 4.576047669901166,
    "completion_tokens": [
      128,
      128,
      96,
      96
    ],
    "reported_prompt_tokens": [
      8192
    ],
    "first_wave_ttft_s": [
      19.78507049602922,
      40.45001697598491
    ],
    "memory": {
      "memory.jsonl": {
        "min_available_gib": 45.57258224487305,
        "new_swap_out_mib": 3.546875,
        "samples": 49
      },
      "benchmark-memory.jsonl": {
        "min_available_gib": 44.35332489013672,
        "new_swap_out_mib": 0.0,
        "samples": 553
      }
    }
  },
  {
    "run": "vllm-20260916T050513Z",
    "length": 32768,
    "clients": 1,
    "waves": 3,
    "warm_requests": 2,
    "correct_all": 3,
    "requests_all": 3,
    "errors_all": 0,
    "warm_median_ttft_s": 81.03521282452857,
    "warm_median_request_s": 90.35905334650306,
    "warm_median_approx_decode_tps": 13.621009212834359,
    "warm_aggregate_completion_tps": 1.4165303197605987,
    "completion_tokens": [
      128,
      128
    ],
    "reported_prompt_tokens": [
      32768
    ],
    "first_wave_ttft_s": [
      81.37414976709988
    ],
    "memory": {
      "memory.jsonl": {
        "min_available_gib": 45.57258224487305,
        "new_swap_out_mib": 3.546875,
        "samples": 49
      },
      "benchmark-memory.jsonl": {
        "min_available_gib": 44.35332489013672,
        "new_swap_out_mib": 0.0,
        "samples": 553
      }
    }
  },
  {
    "run": "vllm-20260916T050513Z",
    "length": 32768,
    "clients": 2,
    "waves": 3,
    "warm_requests": 4,
    "correct_all": 6,
    "requests_all": 6,
    "errors_all": 0,
    "warm_median_ttft_s": 122.8664270354202,
    "warm_median_request_s": 168.31008183798986,
    "warm_median_approx_decode_tps": 7.483761436430415,
    "warm_aggregate_completion_tps": 1.391167151881088,
    "completion_tokens": [
      128,
      128,
      128,
      99
    ],
    "reported_prompt_tokens": [
      32768
    ],
    "first_wave_ttft_s": [
      81.21169123798609,
      164.58256392600015
    ],
    "memory": {
      "memory.jsonl": {
        "min_available_gib": 45.57258224487305,
        "new_swap_out_mib": 3.546875,
        "samples": 49
      },
      "benchmark-memory.jsonl": {
        "min_available_gib": 44.35332489013672,
        "new_swap_out_mib": 0.0,
        "samples": 553
      }
    }
  }
]