{
 "day": "2026-10-25b",
 "generated": "2026-10-25T19:28:53+00:00",
 "machine": {
  "cpu": "AMD RYZEN AI MAX+ 395 w/ Radeon 8060S",
  "gpu": "AMD Radeon Graphics (RADV GFX1151)",
  "driver": "Mesa 25.2.8-0ubuntu0.24.04.2",
  "memory_gib": 92.9,
  "os": "Ubuntu 24.04.4 LTS",
  "kernel": "7.0.0-38-generic"
 },
 "engines": {
  "loculus": {
   "name": "Loculus Core",
   "version": "b1ed73f",
   "backend": "Vulkan (RADV)"
  },
  "llama.cpp": {
   "name": "llama.cpp",
   "build": "e42214804 10702",
   "backend": "Vulkan (RADV)",
   "flags": "-ngl 99 -fa on"
  }
 },
 "models": [
  {
   "name": "Qwen3.8 27B",
   "family": "qwen35",
   "quant": "UD-Q4_K_M",
   "file_gib": 15.33,
   "active_gib": 14.617,
   "engine": {
    "depths": [
     0,
     4096,
     16384,
     32768
    ],
    "pp_tokens": 512,
    "tg_tokens": 128,
    "pp": {
     "loculus": [
      371.36,
      321.81,
      274.93,
      228.44
     ],
     "llama.cpp": [
      301.44,
      264.96,
      207.95,
      131.4
     ]
    },
    "tg": {
     "loculus": [
      12.91,
      12.59,
      12.03,
      11.14
     ],
     "llama.cpp": [
      12.87,
      12.58,
      11.99,
      11.31
     ]
    },
    "peak_gib": {
     "loculus": [
      15.32,
      15.75,
      16.5,
      17.5
     ],
     "llama.cpp": [
      15.81,
      16.47,
      17.98,
      20.0
     ]
    },
    "errors": {},
    "tg_weight_gbps": {
     "loculus": [
      202.6,
      197.6,
      188.8,
      174.8
     ],
     "llama.cpp": [
      202.0,
      197.4,
      188.2,
      177.5
     ]
    }
   },
   "serve": {
    "prompt": {
     "1024": {
      "loculus": {
       "prompt_tokens": 1068,
       "ttft_s": 3.74,
       "prompt_tps": 285.6,
       "decode_tps": 12.69
      },
      "llama.cpp": {
       "prompt_tokens": 1068,
       "ttft_s": 3.872,
       "prompt_tps": 275.8,
       "decode_tps": 12.52
      }
     },
     "4096": {
      "loculus": {
       "prompt_tokens": 3943,
       "ttft_s": 12.682,
       "prompt_tps": 310.9,
       "decode_tps": 12.47
      },
      "llama.cpp": {
       "prompt_tokens": 3943,
       "ttft_s": 15.032,
       "prompt_tps": 262.3,
       "decode_tps": 12.53
      }
     },
     "16384": {
      "loculus": {
       "prompt_tokens": 13623,
       "ttft_s": 45.157,
       "prompt_tps": 301.7,
       "decode_tps": 12.07
      },
      "llama.cpp": {
       "prompt_tokens": 13623,
       "ttft_s": 56.129,
       "prompt_tps": 242.7,
       "decode_tps": 12.05
      }
     }
    },
    "agent": {
     "loculus": {
      "turns": 8,
      "first_ttft_s": 20.334,
      "followup_ttft_median_s": 2.136,
      "total_s": 118.59,
      "final_context": 9712
     },
     "llama.cpp": {
      "turns": 8,
      "first_ttft_s": 24.437,
      "followup_ttft_median_s": 2.501,
      "total_s": 125.33,
      "final_context": 9712
     }
    },
    "concurrent": {
     "1": {
      "loculus": {
       "output_tps": 9.7,
       "ttft_median_s": 3.219,
       "ttft_max_s": 3.219
      },
      "llama.cpp": {
       "output_tps": 9.3,
       "ttft_median_s": 3.671,
       "ttft_max_s": 3.671
      }
     },
     "2": {
      "loculus": {
       "output_tps": 9.5,
       "ttft_median_s": 16.874,
       "ttft_max_s": 16.874
      },
      "llama.cpp": {
       "output_tps": 13.4,
       "ttft_median_s": 7.936,
       "ttft_max_s": 7.936
      }
     },
     "4": {
      "loculus": {
       "output_tps": 9.3,
       "ttft_median_s": 31.586,
       "ttft_max_s": 45.094
      },
      "llama.cpp": {
       "output_tps": 15.8,
       "ttft_median_s": 18.812,
       "ttft_max_s": 18.814
      }
     }
    },
    "server_peak_gib": {
     "loculus": 16.32,
     "llama.cpp": 22.01
    },
    "errors": {}
   }
  },
  {
   "name": "Gemma 4 12B",
   "family": "gemma4",
   "quant": "Q4_K_M",
   "file_gib": 6.63,
   "active_gib": 6.618,
   "engine": {
    "depths": [
     0,
     4096,
     16384,
     32768
    ],
    "pp_tokens": 512,
    "tg_tokens": 128,
    "pp": {
     "loculus": [
      929.75,
      692.58,
      436.25,
      325.86
     ],
     "llama.cpp": [
      646.03,
      548.62,
      430.08,
      341.2
     ]
    },
    "tg": {
     "loculus": [
      28.19,
      26.05,
      25.62,
      24.68
     ],
     "llama.cpp": [
      28.45,
      26.06,
      25.32,
      24.24
     ]
    },
    "peak_gib": {
     "loculus": [
      6.98,
      7.52,
      7.71,
      7.96
     ],
     "llama.cpp": [
      7.97,
      8.73,
      9.11,
      9.63
     ]
    },
    "errors": {},
    "tg_weight_gbps": {
     "loculus": [
      200.3,
      185.1,
      182.1,
      175.4
     ],
     "llama.cpp": [
      202.2,
      185.2,
      179.9,
      172.3
     ]
    }
   },
   "serve": {
    "prompt": {
     "1024": {
      "loculus": {
       "prompt_tokens": 1125,
       "ttft_s": 1.652,
       "prompt_tps": 680.9,
       "decode_tps": 26.6
      },
      "llama.cpp": {
       "prompt_tokens": 1128,
       "ttft_s": 2.107,
       "prompt_tps": 535.5,
       "decode_tps": 26.4
      }
     },
     "4096": {
      "loculus": {
       "prompt_tokens": 4138,
       "ttft_s": 5.924,
       "prompt_tps": 698.5,
       "decode_tps": 25.8
      },
      "llama.cpp": {
       "prompt_tokens": 4141,
       "ttft_s": 7.815,
       "prompt_tps": 529.9,
       "decode_tps": 25.7
      }
     },
     "16384": {
      "loculus": {
       "prompt_tokens": 15048,
       "ttft_s": 25.685,
       "prompt_tps": 585.9,
       "decode_tps": 23.72
      },
      "llama.cpp": {
       "prompt_tokens": 15051,
       "ttft_s": 30.546,
       "prompt_tps": 492.7,
       "decode_tps": 25.38
      }
     }
    },
    "agent": {
     "loculus": {
      "turns": 8,
      "first_ttft_s": 9.558,
      "followup_ttft_median_s": 1.262,
      "total_s": 55.04,
      "final_context": 10288
     },
     "llama.cpp": {
      "turns": 8,
      "first_ttft_s": 12.812,
      "followup_ttft_median_s": 1.543,
      "total_s": 63.32,
      "final_context": 10286
     }
    },
    "concurrent": {
     "1": {
      "loculus": {
       "output_tps": 20.7,
       "ttft_median_s": 1.475,
       "ttft_max_s": 1.475
      },
      "llama.cpp": {
       "output_tps": 18.9,
       "ttft_median_s": 1.975,
       "ttft_max_s": 1.975
      }
     },
     "2": {
      "loculus": {
       "output_tps": 29.3,
       "ttft_median_s": 3.057,
       "ttft_max_s": 3.057
      },
      "llama.cpp": {
       "output_tps": 26.4,
       "ttft_median_s": 4.259,
       "ttft_max_s": 4.259
      }
     },
     "4": {
      "loculus": {
       "output_tps": 35.2,
       "ttft_median_s": 5.315,
       "ttft_max_s": 7.083
      },
      "llama.cpp": {
       "output_tps": 30.5,
       "ttft_median_s": 9.924,
       "ttft_max_s": 9.98
      }
     }
    },
    "server_peak_gib": {
     "loculus": 9.51,
     "llama.cpp": 12.82
    },
    "errors": {}
   }
  },
  {
   "name": "Gemma 4 26B-A4B",
   "family": "gemma4",
   "quant": "UD-Q4_K_M",
   "file_gib": 15.78,
   "active_gib": 3.237,
   "engine": {
    "depths": [
     0,
     4096,
     16384,
     32768
    ],
    "pp_tokens": 512,
    "tg_tokens": 128,
    "pp": {
     "loculus": [
      1613.54,
      1331.19,
      698.1,
      487.33
     ],
     "llama.cpp": [
      1136.36,
      943.73,
      752.68,
      567.18
     ]
    },
    "tg": {
     "loculus": [
      55.68,
      49.75,
      47.68,
      43.58
     ],
     "llama.cpp": [
      53.66,
      47.41,
      44.53,
      40.58
     ]
    },
    "peak_gib": {
     "loculus": [
      16.14,
      16.52,
      16.75,
      17.07
     ],
     "llama.cpp": [
      17.26,
      17.81,
      18.29,
      18.94
     ]
    },
    "errors": {},
    "tg_weight_gbps": {
     "loculus": [
      193.5,
      172.9,
      165.7,
      151.5
     ],
     "llama.cpp": [
      186.5,
      164.8,
      154.8,
      141.0
     ]
    }
   },
   "serve": {
    "prompt": {
     "1024": {
      "loculus": {
       "prompt_tokens": 1125,
       "ttft_s": 1.048,
       "prompt_tps": 1073.1,
       "decode_tps": 51.26
      },
      "llama.cpp": {
       "prompt_tokens": 1128,
       "ttft_s": 1.259,
       "prompt_tps": 896.0,
       "decode_tps": 49.2
      }
     },
     "4096": {
      "loculus": {
       "prompt_tokens": 4138,
       "ttft_s": 3.647,
       "prompt_tps": 1134.5,
       "decode_tps": 49.57
      },
      "llama.cpp": {
       "prompt_tokens": 4141,
       "ttft_s": 4.447,
       "prompt_tps": 931.1,
       "decode_tps": 47.27
      }
     },
     "16384": {
      "loculus": {
       "prompt_tokens": 15048,
       "ttft_s": 16.012,
       "prompt_tps": 939.8,
       "decode_tps": 48.01
      },
      "llama.cpp": {
       "prompt_tokens": 15051,
       "ttft_s": 17.531,
       "prompt_tps": 858.5,
       "decode_tps": 44.09
      }
     }
    },
    "agent": {
     "loculus": {
      "turns": 8,
      "first_ttft_s": 5.919,
      "followup_ttft_median_s": 0.771,
      "total_s": 21.06,
      "final_context": 10288
     },
     "llama.cpp": {
      "turns": 8,
      "first_ttft_s": 7.517,
      "followup_ttft_median_s": 0.919,
      "total_s": 36.41,
      "final_context": 10286
     }
    },
    "concurrent": {
     "1": {
      "loculus": {
       "output_tps": 38.8,
       "ttft_median_s": 0.871,
       "ttft_max_s": 0.871
      },
      "llama.cpp": {
       "output_tps": 34.0,
       "ttft_median_s": 1.179,
       "ttft_max_s": 1.179
      }
     },
     "2": {
      "loculus": {
       "output_tps": 50.2,
       "ttft_median_s": 1.651,
       "ttft_max_s": 1.651
      },
      "llama.cpp": {
       "output_tps": 43.0,
       "ttft_median_s": 2.504,
       "ttft_max_s": 2.504
      }
     },
     "4": {
      "loculus": {
       "output_tps": 57.5,
       "ttft_median_s": 3.181,
       "ttft_max_s": 4.21
      },
      "llama.cpp": {
       "output_tps": 48.3,
       "ttft_median_s": 5.725,
       "ttft_max_s": 5.759
      }
     }
    },
    "server_peak_gib": {
     "loculus": 18.03,
     "llama.cpp": 20.67
    },
    "errors": {}
   }
  },
  {
   "name": "gpt-oss 20B",
   "family": "gpt-oss",
   "quant": "MXFP4",
   "file_gib": 11.28,
   "active_gib": 2.398,
   "engine": {
    "depths": [
     0,
     4096,
     16384,
     32768
    ],
    "pp_tokens": 512,
    "tg_tokens": 128,
    "pp": {
     "loculus": [
      1894.17,
      1796.49,
      1347.4,
      1101.61
     ],
     "llama.cpp": [
      1256.58,
      1079.14,
      723.7,
      446.74
     ]
    },
    "tg": {
     "loculus": [
      75.21,
      69.3,
      62.91,
      56.12
     ],
     "llama.cpp": [
      78.08,
      74.53,
      66.15,
      58.73
     ]
    },
    "peak_gib": {
     "loculus": [
      10.85,
      10.96,
      11.24,
      11.61
     ],
     "llama.cpp": [
      11.81,
      12.01,
      12.58,
      13.35
     ]
    },
    "errors": {},
    "tg_weight_gbps": {
     "loculus": [
      193.7,
      178.4,
      162.0,
      144.5
     ],
     "llama.cpp": [
      201.0,
      191.9,
      170.3,
      151.2
     ]
    }
   },
   "serve": {
    "prompt": {
     "1024": {
      "loculus": {
       "prompt_tokens": 1003,
       "ttft_s": 0.572,
       "prompt_tps": 1752.3,
       "decode_tps": 71.55
      },
      "llama.cpp": {
       "prompt_tokens": 1003,
       "ttft_s": 0.932,
       "prompt_tps": 1076.1,
       "decode_tps": 78.11
      }
     },
     "4096": {
      "loculus": {
       "prompt_tokens": 3613,
       "ttft_s": 2.25,
       "prompt_tps": 1606.1,
       "decode_tps": 69.26
      },
      "llama.cpp": {
       "prompt_tokens": 3613,
       "ttft_s": 3.329,
       "prompt_tps": 1085.3,
       "decode_tps": 74.17
      }
     },
     "16384": {
      "loculus": {
       "prompt_tokens": 12550,
       "ttft_s": 8.53,
       "prompt_tps": 1471.2,
       "decode_tps": 64.11
      },
      "llama.cpp": {
       "prompt_tokens": 12550,
       "ttft_s": 13.422,
       "prompt_tps": 935.0,
       "decode_tps": 68.82
      }
     }
    },
    "agent": {
     "loculus": {
      "turns": 8,
      "first_ttft_s": 3.737,
      "followup_ttft_median_s": 0.381,
      "total_s": 21.53,
      "final_context": 8857
     },
     "llama.cpp": {
      "turns": 8,
      "first_ttft_s": 7.001,
      "followup_ttft_median_s": 0.751,
      "total_s": 26.61,
      "final_context": 8857
     }
    },
    "concurrent": {
     "1": {
      "loculus": {
       "output_tps": 55.7,
       "ttft_median_s": 0.578,
       "ttft_max_s": 0.578
      },
      "llama.cpp": {
       "output_tps": 48.3,
       "ttft_median_s": 1.028,
       "ttft_max_s": 1.028
      }
     },
     "2": {
      "loculus": {
       "output_tps": 64.4,
       "ttft_median_s": 1.177,
       "ttft_max_s": 1.177
      },
      "llama.cpp": {
       "output_tps": 55.7,
       "ttft_median_s": 2.09,
       "ttft_max_s": 2.09
      }
     },
     "4": {
      "loculus": {
       "output_tps": 72.3,
       "ttft_median_s": 2.219,
       "ttft_max_s": 2.903
      },
      "llama.cpp": {
       "output_tps": 65.1,
       "ttft_median_s": 4.168,
       "ttft_max_s": 4.169
      }
     }
    },
    "server_peak_gib": {
     "loculus": 11.52,
     "llama.cpp": 12.82
    },
    "errors": {}
   }
  },
  {
   "name": "gpt-oss 120B",
   "family": "gpt-oss",
   "quant": "MXFP4",
   "file_gib": 59.03,
   "active_gib": 3.347,
   "engine": {
    "depths": [
     0,
     4096,
     16384,
     32768
    ],
    "pp_tokens": 512,
    "tg_tokens": 128,
    "pp": {
     "loculus": [
      757.69,
      670.29,
      593.59,
      520.18
     ],
     "llama.cpp": [
      552.47,
      493.98,
      354.79,
      261.9
     ]
    },
    "tg": {
     "loculus": [
      51.72,
      49.32,
      44.63,
      39.64
     ],
     "llama.cpp": [
      56.53,
      53.62,
      47.93,
      41.91
     ]
    },
    "peak_gib": {
     "loculus": [
      58.67,
      58.82,
      59.24,
      59.8
     ],
     "llama.cpp": [
      59.59,
      59.89,
      60.74,
      61.88
     ]
    },
    "errors": {},
    "tg_weight_gbps": {
     "loculus": [
      185.9,
      177.2,
      160.4,
      142.5
     ],
     "llama.cpp": [
      203.2,
      192.7,
      172.3,
      150.6
     ]
    }
   },
   "serve": {
    "prompt": {
     "1024": {
      "loculus": {
       "prompt_tokens": 1003,
       "ttft_s": 1.642,
       "prompt_tps": 610.9,
       "decode_tps": 50.59
      },
      "llama.cpp": {
       "prompt_tokens": 1003,
       "ttft_s": 2.337,
       "prompt_tps": 429.1,
       "decode_tps": 56.04
      }
     },
     "4096": {
      "loculus": {
       "prompt_tokens": 3613,
       "ttft_s": 6.735,
       "prompt_tps": 536.4,
       "decode_tps": 48.6
      },
      "llama.cpp": {
       "prompt_tokens": 3613,
       "ttft_s": 8.916,
       "prompt_tps": 405.2,
       "decode_tps": 52.89
      }
     },
     "16384": {
      "loculus": {
       "prompt_tokens": 12550,
       "ttft_s": 23.366,
       "prompt_tps": 537.1,
       "decode_tps": 44.75
      },
      "llama.cpp": {
       "prompt_tokens": 12550,
       "ttft_s": 33.468,
       "prompt_tps": 375.0,
       "decode_tps": 47.77
      }
     }
    },
    "agent": {
     "loculus": {
      "turns": 8,
      "first_ttft_s": 10.225,
      "followup_ttft_median_s": 0.975,
      "total_s": 35.14,
      "final_context": 8857
     },
     "llama.cpp": {
      "turns": 8,
      "first_ttft_s": 14.55,
      "followup_ttft_median_s": 1.63,
      "total_s": 44.01,
      "final_context": 8857
     }
    },
    "concurrent": {
     "1": {
      "loculus": {
       "output_tps": 31.2,
       "ttft_median_s": 1.604,
       "ttft_max_s": 1.604
      },
      "llama.cpp": {
       "output_tps": 24.7,
       "ttft_median_s": 2.861,
       "ttft_max_s": 2.861
      }
     },
     "2": {
      "loculus": {
       "output_tps": 32.4,
       "ttft_median_s": 3.43,
       "ttft_max_s": 3.43
      },
      "llama.cpp": {
       "output_tps": 27.3,
       "ttft_median_s": 5.849,
       "ttft_max_s": 5.849
      }
     },
     "4": {
      "loculus": {
       "output_tps": 31.7,
       "ttft_median_s": 6.084,
       "ttft_max_s": 7.937
      },
      "llama.cpp": {
       "output_tps": 29.8,
       "ttft_median_s": 11.553,
       "ttft_max_s": 11.553
      }
     }
    },
    "server_peak_gib": {
     "loculus": 59.65,
     "llama.cpp": 61.22
    },
    "errors": {}
   }
  },
  {
   "name": "Nemotron 3.5 Lightning 30B-A3B",
   "family": "nemotron_h_moe",
   "quant": "UD-Q4_K_M",
   "file_gib": 23.53,
   "active_gib": 2.896,
   "engine": {
    "depths": [
     0,
     4096,
     16384,
     32768
    ],
    "pp_tokens": 512,
    "tg_tokens": 128,
    "pp": {
     "loculus": [
      1060.82,
      1056.35,
      906.4,
      851.2
     ],
     "llama.cpp": [
      1076.67,
      996.47,
      905.3,
      790.44
     ]
    },
    "tg": {
     "loculus": [
      65.9,
      64.19,
      60.87,
      60.11
     ],
     "llama.cpp": [
      61.66,
      59.62,
      58.77,
      56.69
     ]
    },
    "peak_gib": {
     "loculus": [
      23.22,
      23.3,
      23.51,
      23.79
     ],
     "llama.cpp": [
      23.02,
      23.12,
      23.35,
      23.55
     ]
    },
    "errors": {},
    "tg_weight_gbps": {
     "loculus": [
      204.9,
      199.6,
      189.3,
      186.9
     ],
     "llama.cpp": [
      191.7,
      185.4,
      182.7,
      176.3
     ]
    }
   },
   "serve": {
    "prompt": {
     "1024": {
      "loculus": {
       "prompt_tokens": 1019,
       "ttft_s": 0.758,
       "prompt_tps": 1344.7,
       "decode_tps": 65.06
      },
      "llama.cpp": {
       "prompt_tokens": 1019,
       "ttft_s": 2.083,
       "prompt_tps": 489.2,
       "decode_tps": 59.57
      }
     },
     "4096": {
      "loculus": {
       "prompt_tokens": 3833,
       "ttft_s": 1.823,
       "prompt_tps": 2103.0,
       "decode_tps": 63.49
      },
      "llama.cpp": {
       "prompt_tokens": 3833,
       "ttft_s": 5.014,
       "prompt_tps": 764.5,
       "decode_tps": 58.69
      }
     },
     "16384": {
      "loculus": {
       "prompt_tokens": 13149,
       "ttft_s": 6.893,
       "prompt_tps": 1907.7,
       "decode_tps": 60.77
      },
      "llama.cpp": {
       "prompt_tokens": 13149,
       "ttft_s": 15.553,
       "prompt_tps": 845.4,
       "decode_tps": 56.35
      }
     }
    },
    "agent": {
     "loculus": {
      "turns": 8,
      "first_ttft_s": 3.512,
      "followup_ttft_median_s": 0.662,
      "total_s": 24.47,
      "final_context": 9388
     },
     "llama.cpp": {
      "turns": 8,
      "first_ttft_s": 8.058,
      "followup_ttft_median_s": 1.382,
      "total_s": 35.52,
      "final_context": 9388
     }
    },
    "concurrent": {
     "1": {
      "loculus": {
       "output_tps": 47.1,
       "ttft_median_s": 0.726,
       "ttft_max_s": 0.726
      },
      "llama.cpp": {
       "output_tps": 33.7,
       "ttft_median_s": 1.693,
       "ttft_max_s": 1.693
      }
     },
     "2": {
      "loculus": {
       "output_tps": 47.0,
       "ttft_median_s": 3.45,
       "ttft_max_s": 3.45
      },
      "llama.cpp": {
       "output_tps": 38.4,
       "ttft_median_s": 3.918,
       "ttft_max_s": 3.918
      }
     },
     "4": {
      "loculus": {
       "output_tps": 45.3,
       "ttft_median_s": 6.366,
       "ttft_max_s": 9.297
      },
      "llama.cpp": {
       "output_tps": 39.7,
       "ttft_median_s": 8.62,
       "ttft_max_s": 8.621
      }
     }
    },
    "server_peak_gib": {
     "loculus": 23.63,
     "llama.cpp": 24.26
    },
    "errors": {}
   }
  },
  {
   "name": "Qwen3.8 Flash-Next 177B-A3B",
   "family": "qwen4exp",
   "quant": "UD-Q2_K_XL",
   "file_gib": 73.45,
   "active_gib": 4.131,
   "engine": {
    "depths": [
     0,
     4096,
     16384,
     32768
    ],
    "pp_tokens": 512,
    "tg_tokens": 128,
    "pp": {
     "loculus": [
      610.54,
      532.97,
      443.01,
      357.32
     ],
     "llama.cpp": [
      378.07,
      286.94,
      233.25,
      194.32
     ]
    },
    "tg": {
     "loculus": [
      34.86,
      30.93,
      29.84,
      27.29
     ],
     "llama.cpp": [
      32.81,
      24.2,
      17.91,
      16.86
     ]
    },
    "peak_gib": {
     "loculus": [
      48.38,
      48.59,
      48.98,
      49.36
     ],
     "llama.cpp": [
      74.24,
      74.59,
      75.41,
      76.5
     ]
    },
    "errors": {},
    "tg_weight_gbps": {
     "loculus": [
      154.6,
      137.2,
      132.4,
      121.0
     ],
     "llama.cpp": [
      145.5,
      107.3,
      79.4,
      74.8
     ]
    }
   },
   "serve": {
    "prompt": {
     "1024": {
      "loculus": {
       "prompt_tokens": 1068,
       "ttft_s": 3.227,
       "prompt_tps": 331.0,
       "decode_tps": 34.25
      },
      "llama.cpp": {
       "prompt_tokens": 1068,
       "ttft_s": 5.802,
       "prompt_tps": 184.1,
       "decode_tps": 28.56
      }
     },
     "4096": {
      "loculus": {
       "prompt_tokens": 3943,
       "ttft_s": 9.744,
       "prompt_tps": 404.7,
       "decode_tps": 32.72
      },
      "llama.cpp": {
       "prompt_tokens": 3943,
       "ttft_s": 17.647,
       "prompt_tps": 223.4,
       "decode_tps": 25.54
      }
     },
     "16384": {
      "loculus": {
       "prompt_tokens": 13623,
       "ttft_s": 28.564,
       "prompt_tps": 476.9,
       "decode_tps": 29.8
      },
      "llama.cpp": {
       "prompt_tokens": 13623,
       "ttft_s": 63.65,
       "prompt_tps": 214.0,
       "decode_tps": 21.89
      }
     }
    },
    "agent": {
     "loculus": {
      "turns": 8,
      "first_ttft_s": 12.373,
      "followup_ttft_median_s": 1.575,
      "total_s": 55.56,
      "final_context": 9712
     },
     "llama.cpp": {
      "turns": 8,
      "first_ttft_s": 27.9,
      "followup_ttft_median_s": 4.052,
      "total_s": 101.18,
      "final_context": 9712
     }
    },
    "concurrent": {
     "1": {
      "loculus": {
       "output_tps": 19.6,
       "ttft_median_s": 2.747,
       "ttft_max_s": 2.747
      },
      "llama.cpp": {
       "output_tps": 15.9,
       "ttft_median_s": 3.673,
       "ttft_max_s": 3.673
      }
     },
     "2": {
      "loculus": {
       "output_tps": 20.3,
       "ttft_median_s": 8.77,
       "ttft_max_s": 8.77
      },
      "llama.cpp": {
       "output_tps": 18.2,
       "ttft_median_s": 8.136,
       "ttft_max_s": 8.136
      }
     },
     "4": {
      "loculus": {
       "output_tps": 19.7,
       "ttft_median_s": 15.674,
       "ttft_max_s": 22.104
      },
      "llama.cpp": {
       "output_tps": 19.4,
       "ttft_median_s": 17.945,
       "ttft_max_s": 17.947
      }
     }
    },
    "server_peak_gib": {
     "loculus": 48.92,
     "llama.cpp": 78.41
    },
    "errors": {}
   }
  },
  {
   "name": "Qwen3 30B-A3B",
   "family": "qwen3moe",
   "quant": "UD-IQ1_S",
   "file_gib": 8.42,
   "active_gib": 1.167,
   "engine": {
    "depths": [
     0,
     4096,
     16384,
     32768
    ],
    "pp_tokens": 512,
    "tg_tokens": 128,
    "pp": {
     "loculus": [
      1598.87,
      1150.38,
      590.09,
      364.1
     ],
     "llama.cpp": [
      1175.58,
      765.7,
      342.66,
      183.32
     ]
    },
    "tg": {
     "loculus": [
      101.89,
      75.13,
      54.41,
      47.47
     ],
     "llama.cpp": [
      117.72,
      92.89,
      60.65,
      41.89
     ]
    },
    "peak_gib": {
     "loculus": [
      8.58,
      8.96,
      10.04,
      11.54
     ],
     "llama.cpp": [
      8.84,
      9.6,
      11.86,
      14.88
     ]
    },
    "errors": {},
    "tg_weight_gbps": {
     "loculus": [
      127.7,
      94.1,
      68.2,
      59.5
     ],
     "llama.cpp": [
      147.5,
      116.4,
      76.0,
      52.5
     ]
    }
   },
   "serve": {
    "prompt": {
     "1024": {
      "loculus": {
       "prompt_tokens": 968,
       "ttft_s": 0.719,
       "prompt_tps": 1346.6,
       "decode_tps": 90.7
      },
      "llama.cpp": {
       "prompt_tokens": 968,
       "ttft_s": 0.891,
       "prompt_tps": 1086.1,
       "decode_tps": 106.91
      }
     },
     "4096": {
      "loculus": {
       "prompt_tokens": 3655,
       "ttft_s": 2.839,
       "prompt_tps": 1287.2,
       "decode_tps": 76.56
      },
      "llama.cpp": {
       "prompt_tokens": 3655,
       "ttft_s": 3.982,
       "prompt_tps": 917.9,
       "decode_tps": 91.02
      }
     },
     "16384": {
      "loculus": {
       "prompt_tokens": 12598,
       "ttft_s": 12.294,
       "prompt_tps": 1024.8,
       "decode_tps": 58.01
      },
      "llama.cpp": {
       "prompt_tokens": 12598,
       "ttft_s": 19.929,
       "prompt_tps": 632.2,
       "decode_tps": 62.71
      }
     }
    },
    "agent": {
     "loculus": {
      "turns": 8,
      "first_ttft_s": 5.028,
      "followup_ttft_median_s": 0.593,
      "total_s": 10.25,
      "final_context": 9062
     },
     "llama.cpp": {
      "turns": 8,
      "first_ttft_s": 7.477,
      "followup_ttft_median_s": 0.839,
      "total_s": 14.4,
      "final_context": 9062
     }
    },
    "concurrent": {
     "1": {
      "loculus": {
       "output_tps": 62.0,
       "ttft_median_s": 0.659,
       "ttft_max_s": 0.659
      },
      "llama.cpp": {
       "output_tps": 62.7,
       "ttft_median_s": 0.836,
       "ttft_max_s": 0.836
      }
     },
     "2": {
      "loculus": {
       "output_tps": 67.4,
       "ttft_median_s": 1.296,
       "ttft_max_s": 1.296
      },
      "llama.cpp": {
       "output_tps": 73.4,
       "ttft_median_s": 1.819,
       "ttft_max_s": 1.819
      }
     },
     "4": {
      "loculus": {
       "output_tps": 73.5,
       "ttft_median_s": 2.569,
       "ttft_max_s": 3.353
      },
      "llama.cpp": {
       "output_tps": 71.1,
       "ttft_median_s": 4.42,
       "ttft_max_s": 4.421
      }
     }
    },
    "server_peak_gib": {
     "loculus": 10.5,
     "llama.cpp": 13.34
    },
    "errors": {}
   }
  },
  {
   "name": "Llama 3.2 3B",
   "family": "llama",
   "quant": "Q4_K_M",
   "file_gib": 1.88,
   "active_gib": 1.873,
   "engine": {
    "depths": [
     0,
     4096,
     16384,
     32768
    ],
    "pp_tokens": 512,
    "tg_tokens": 128,
    "pp": {
     "loculus": [
      3276.03,
      2387.44,
      1224.02,
      786.28
     ],
     "llama.cpp": [
      2409.21,
      1624.88,
      640.85,
      277.04
     ]
    },
    "tg": {
     "loculus": [
      99.92,
      81.01,
      62.76,
      48.83
     ],
     "llama.cpp": [
      99.13,
      80.47,
      53.0,
      36.7
     ]
    },
    "peak_gib": {
     "loculus": [
      2.01,
      2.45,
      3.7,
      5.46
     ],
     "llama.cpp": [
      2.57,
      3.47,
      6.09,
      9.6
     ]
    },
    "errors": {},
    "tg_weight_gbps": {
     "loculus": [
      201.0,
      162.9,
      126.2,
      98.2
     ],
     "llama.cpp": [
      199.4,
      161.8,
      106.6,
      73.8
     ]
    }
   },
   "serve": {
    "prompt": {
     "1024": {
      "loculus": {
       "prompt_tokens": 973,
       "ttft_s": 0.368,
       "prompt_tps": 2647.3,
       "decode_tps": 91.1
      },
      "llama.cpp": {
       "prompt_tokens": 973,
       "ttft_s": 0.465,
       "prompt_tps": 2094.3,
       "decode_tps": 93.31
      }
     },
     "4096": {
      "loculus": {
       "prompt_tokens": 3571,
       "ttft_s": 1.393,
       "prompt_tps": 2563.2,
       "decode_tps": 80.53
      },
      "llama.cpp": {
       "prompt_tokens": 3571,
       "ttft_s": 1.783,
       "prompt_tps": 2002.7,
       "decode_tps": 80.45
      }
     },
     "16384": {
      "loculus": {
       "prompt_tokens": 12514,
       "ttft_s": 6.258,
       "prompt_tps": 1999.6,
       "decode_tps": 66.97
      },
      "llama.cpp": {
       "prompt_tokens": 12514,
       "ttft_s": 10.102,
       "prompt_tps": 1238.8,
       "decode_tps": 58.81
      }
     }
    },
    "agent": {
     "loculus": {
      "turns": 8,
      "first_ttft_s": 2.402,
      "followup_ttft_median_s": 0.325,
      "total_s": 5.88,
      "final_context": 8873
     },
     "llama.cpp": {
      "turns": 8,
      "first_ttft_s": 4.611,
      "followup_ttft_median_s": 0.402,
      "total_s": 8.57,
      "final_context": 8873
     }
    },
    "concurrent": {
     "1": {
      "loculus": {
       "output_tps": 76.2,
       "ttft_median_s": 0.325,
       "ttft_max_s": 0.325
      },
      "llama.cpp": {
       "output_tps": 73.2,
       "ttft_median_s": 0.39,
       "ttft_max_s": 0.39
      }
     },
     "2": {
      "loculus": {
       "output_tps": 111.6,
       "ttft_median_s": 0.67,
       "ttft_max_s": 0.67
      },
      "llama.cpp": {
       "output_tps": 107.9,
       "ttft_median_s": 0.866,
       "ttft_max_s": 0.866
      }
     },
     "4": {
      "loculus": {
       "output_tps": 131.0,
       "ttft_median_s": 1.297,
       "ttft_max_s": 1.687
      },
      "llama.cpp": {
       "output_tps": 132.3,
       "ttft_median_s": 1.947,
       "ttft_max_s": 1.948
      }
     }
    },
    "server_peak_gib": {
     "loculus": 4.24,
     "llama.cpp": 7.81
    },
    "errors": {}
   }
  },
  {
   "name": "Mistral 7B v0.3",
   "family": "llama",
   "quant": "Q4_K_M",
   "file_gib": 4.07,
   "active_gib": 4.001,
   "engine": {
    "depths": [
     0,
     4096,
     16384,
     32768
    ],
    "pp_tokens": 512,
    "tg_tokens": 128,
    "pp": {
     "loculus": [
      1283.54,
      980.8,
      656.93,
      449.52
     ],
     "llama.cpp": [
      1039.54,
      742.63,
      324.53,
      132.47
     ]
    },
    "tg": {
     "loculus": [
      46.95,
      41.63,
      35.39,
      29.75
     ],
     "llama.cpp": [
      48.09,
      42.93,
      32.77,
      25.11
     ]
    },
    "peak_gib": {
     "loculus": [
      4.15,
      4.65,
      6.1,
      8.1
     ],
     "llama.cpp": [
      4.29,
      5.3,
      8.31,
      12.35
     ]
    },
    "errors": {},
    "tg_weight_gbps": {
     "loculus": [
      201.7,
      178.8,
      152.0,
      127.8
     ],
     "llama.cpp": [
      206.6,
      184.4,
      140.8,
      107.9
     ]
    }
   },
   "serve": {
    "prompt": {
     "1024": {
      "loculus": {
       "prompt_tokens": 1175,
       "ttft_s": 1.055,
       "prompt_tps": 1113.3,
       "decode_tps": 44.65
      },
      "llama.cpp": {
       "prompt_tokens": 1176,
       "ttft_s": 1.248,
       "prompt_tps": 942.0,
       "decode_tps": 45.9
      }
     },
     "4096": {
      "loculus": {
       "prompt_tokens": 4401,
       "ttft_s": 4.148,
       "prompt_tps": 1060.9,
       "decode_tps": 41.03
      },
      "llama.cpp": {
       "prompt_tokens": 4402,
       "ttft_s": 5.4,
       "prompt_tps": 815.2,
       "decode_tps": 42.03
      }
     },
     "16384": {
      "loculus": {
       "prompt_tokens": 15896,
       "ttft_s": 17.557,
       "prompt_tps": 905.4,
       "decode_tps": 35.36
      },
      "llama.cpp": {
       "prompt_tokens": 15897,
       "ttft_s": 29.891,
       "prompt_tps": 531.8,
       "decode_tps": 32.9
      }
     }
    },
    "agent": {
     "llama.cpp": {
      "turns": 8,
      "first_ttft_s": 10.39,
      "followup_ttft_median_s": 1.178,
      "total_s": 25.99,
      "final_context": 11147
     }
    },
    "concurrent": {
     "1": {
      "loculus": {
       "output_tps": 33.5,
       "ttft_median_s": 1.002,
       "ttft_max_s": 1.002
      },
      "llama.cpp": {
       "output_tps": 33.4,
       "ttft_median_s": 1.136,
       "ttft_max_s": 1.136
      }
     },
     "2": {
      "loculus": {
       "output_tps": 46.7,
       "ttft_median_s": 1.918,
       "ttft_max_s": 1.918
      },
      "llama.cpp": {
       "output_tps": 50.0,
       "ttft_median_s": 2.271,
       "ttft_max_s": 2.271
      }
     },
     "4": {
      "loculus": {
       "output_tps": 56.0,
       "ttft_median_s": 3.347,
       "ttft_max_s": 4.433
      },
      "llama.cpp": {
       "output_tps": 58.6,
       "ttft_median_s": 4.751,
       "ttft_max_s": 5.477
      }
     }
    },
    "server_peak_gib": {
     "loculus": 7.03,
     "llama.cpp": 11.02
    },
    "errors": {}
   }
  },
  {
   "name": "Phi-4 mini",
   "family": "phi3",
   "quant": "Q4_K_M",
   "file_gib": 2.32,
   "active_gib": 2.313,
   "engine": {
    "depths": [
     0,
     4096,
     16384,
     32768
    ],
    "pp_tokens": 512,
    "tg_tokens": 128,
    "pp": {
     "loculus": [
      2823.09,
      2089.88,
      1079.48,
      688.54
     ],
     "llama.cpp": [
      2089.44,
      1397.98,
      561.18,
      239.01
     ]
    },
    "tg": {
     "loculus": [
      79.5,
      65.8,
      51.93,
      40.67
     ],
     "llama.cpp": [
      81.53,
      67.0,
      44.37,
      31.85
     ]
    },
    "peak_gib": {
     "loculus": [
      2.46,
      2.97,
      4.4,
      6.4
     ],
     "llama.cpp": [
      3.33,
      4.34,
      7.37,
      11.4
     ]
    },
    "errors": {},
    "tg_weight_gbps": {
     "loculus": [
      197.4,
      163.4,
      129.0,
      101.0
     ],
     "llama.cpp": [
      202.5,
      166.4,
      110.2,
      79.1
     ]
    }
   },
   "serve": {
    "prompt": {
     "1024": {
      "loculus": {
       "prompt_tokens": 939,
       "ttft_s": 0.37,
       "prompt_tps": 2536.9,
       "decode_tps": 74.07
      },
      "llama.cpp": {
       "prompt_tokens": 939,
       "ttft_s": 0.507,
       "prompt_tps": 1852.2,
       "decode_tps": 75.67
      }
     },
     "4096": {
      "loculus": {
       "prompt_tokens": 3549,
       "ttft_s": 1.467,
       "prompt_tps": 2419.5,
       "decode_tps": 65.83
      },
      "llama.cpp": {
       "prompt_tokens": 3549,
       "ttft_s": 1.976,
       "prompt_tps": 1796.0,
       "decode_tps": 65.72
      }
     },
     "16384": {
      "loculus": {
       "prompt_tokens": 12486,
       "ttft_s": 7.059,
       "prompt_tps": 1768.7,
       "decode_tps": 54.12
      },
      "llama.cpp": {
       "prompt_tokens": 12486,
       "ttft_s": 11.644,
       "prompt_tps": 1072.3,
       "decode_tps": 48.15
      }
     }
    },
    "agent": {
     "loculus": {
      "turns": 1,
      "first_ttft_s": 2.548,
      "followup_ttft_median_s": null,
      "total_s": 3.05,
      "final_context": 5605
     },
     "llama.cpp": {
      "turns": 8,
      "first_ttft_s": 4.094,
      "followup_ttft_median_s": 0.464,
      "total_s": 8.45,
      "final_context": 8745
     }
    },
    "concurrent": {
     "1": {
      "loculus": {
       "output_tps": 63.7,
       "ttft_median_s": 0.314,
       "ttft_max_s": 0.314
      },
      "llama.cpp": {
       "output_tps": 60.4,
       "ttft_median_s": 0.428,
       "ttft_max_s": 0.428
      }
     },
     "2": {
      "loculus": {
       "output_tps": 96.4,
       "ttft_median_s": 0.663,
       "ttft_max_s": 0.663
      },
      "llama.cpp": {
       "output_tps": 93.4,
       "ttft_median_s": 0.915,
       "ttft_max_s": 0.915
      }
     },
     "4": {
      "loculus": {
       "output_tps": 114.7,
       "ttft_median_s": 1.311,
       "ttft_max_s": 1.695
      },
      "llama.cpp": {
       "output_tps": 112.8,
       "ttft_median_s": 2.159,
       "ttft_max_s": 2.159
      }
     }
    },
    "server_peak_gib": {
     "loculus": 4.9,
     "llama.cpp": 9.17
    },
    "errors": {}
   }
  },
  {
   "name": "GLM-4 9B",
   "family": "glm4",
   "quant": "Q4_K_M",
   "file_gib": 5.74,
   "active_gib": 5.409,
   "engine": {
    "depths": [
     0,
     4096,
     16384,
     32768
    ],
    "pp_tokens": 512,
    "tg_tokens": 128,
    "pp": {
     "loculus": [
      1130.43,
      877.18,
      564.33,
      381.0
     ],
     "llama.cpp": [
      925.66,
      726.1,
      422.42,
      280.77
     ]
    },
    "tg": {
     "loculus": [
      34.26,
      31.01,
      27.28,
      23.36
     ],
     "llama.cpp": [
      35.47,
      34.55,
      31.8,
      29.05
     ]
    },
    "peak_gib": {
     "loculus": [
      5.53,
      5.69,
      6.13,
      6.76
     ],
     "llama.cpp": [
      6.15,
      6.47,
      7.42,
      8.69
     ]
    },
    "errors": {},
    "tg_weight_gbps": {
     "loculus": [
      199.0,
      180.1,
      158.4,
      135.7
     ],
     "llama.cpp": [
      206.0,
      200.7,
      184.7,
      168.7
     ]
    }
   },
   "serve": {
    "prompt": {
     "1024": {
      "loculus": {
       "prompt_tokens": 944,
       "ttft_s": 0.949,
       "prompt_tps": 994.5,
       "decode_tps": 33.42
      },
      "llama.cpp": {
       "prompt_tokens": 944,
       "ttft_s": 1.151,
       "prompt_tps": 820.1,
       "decode_tps": 34.7
      }
     },
     "4096": {
      "loculus": {
       "prompt_tokens": 3547,
       "ttft_s": 3.73,
       "prompt_tps": 951.0,
       "decode_tps": 31.79
      },
      "llama.cpp": {
       "prompt_tokens": 3547,
       "ttft_s": 4.652,
       "prompt_tps": 762.5,
       "decode_tps": 33.31
      }
     },
     "16384": {
      "loculus": {
       "prompt_tokens": 12496,
       "ttft_s": 14.826,
       "prompt_tps": 842.8,
       "decode_tps": 28.45
      },
      "llama.cpp": {
       "prompt_tokens": 12496,
       "ttft_s": 20.482,
       "prompt_tps": 610.1,
       "decode_tps": 31.38
      }
     }
    },
    "agent": {
     "loculus": {
      "turns": 8,
      "first_ttft_s": 6.224,
      "followup_ttft_median_s": 0.704,
      "total_s": 17.81,
      "final_context": 8804
     },
     "llama.cpp": {
      "turns": 8,
      "first_ttft_s": 8.027,
      "followup_ttft_median_s": 0.883,
      "total_s": 20.52,
      "final_context": 8804
     }
    },
    "concurrent": {
     "1": {
      "loculus": {
       "output_tps": 27.3,
       "ttft_median_s": 0.839,
       "ttft_max_s": 0.839
      },
      "llama.cpp": {
       "output_tps": 27.7,
       "ttft_median_s": 0.976,
       "ttft_max_s": 0.976
      }
     },
     "2": {
      "loculus": {
       "output_tps": 40.8,
       "ttft_median_s": 1.849,
       "ttft_max_s": 1.849
      },
      "llama.cpp": {
       "output_tps": 42.3,
       "ttft_median_s": 2.107,
       "ttft_max_s": 2.107
      }
     },
     "4": {
      "loculus": {
       "output_tps": 52.0,
       "ttft_median_s": 4.049,
       "ttft_max_s": 4.166
      },
      "llama.cpp": {
       "output_tps": 54.3,
       "ttft_median_s": 5.07,
       "ttft_max_s": 5.071
      }
     }
    },
    "server_peak_gib": {
     "loculus": 6.33,
     "llama.cpp": 7.97
    },
    "errors": {}
   }
  }
 ],
 "overview": {
  "pp": {
   "geomean_ratio": 1.49,
   "min_ratio": 0.859,
   "max_ratio": 3.393,
   "ahead": 44,
   "of": 48
  },
  "tg": {
   "geomean_ratio": 1.03,
   "min_ratio": 0.804,
   "max_ratio": 1.666,
   "ahead": 26,
   "of": 48
  },
  "peak_gib": {
   "geomean_ratio": 0.829,
   "min_ratio": 0.561,
   "max_ratio": 1.01,
   "ahead": 44,
   "of": 48
  }
 }
}