{
  "schema": 1,
  "runs": [
    {
      "schema": 1,
      "started_at": "2026-10-07T16:23:32+00:00",
      "finished_at": "2026-10-07T16:33:28+00:00",
      "comment": "Historical native image 68787abb versus current Nano backend; same GPU, checkpoint cache, API payloads, 10 steps, concurrency 1",
      "git": {
        "revision": "d9e163bcccc3bbcaf6927b411e3b7854937bb8a4",
        "dirty": true,
        "status": [
          " M Taskfile.yml",
          "?? benchmarks/"
        ],
        "diff_sha256": "4d2ec160b07ba1242b5e8a6185e91107707b3ab618086f3210a70bca0f561855"
      },
      "hardware": {
        "index": 0,
        "name": "NVIDIA GeForce RTX 5070 Ti",
        "driver": "610.88",
        "total_mib": 16303,
        "used_mib": 1889,
        "utilization_percent": 1
      },
      "settings": {
        "model_id": "openbmb/VoxCPM2",
        "gpu_index": 0,
        "calls": 10,
        "warmup_calls": 2,
        "sample_interval_seconds": 0.1,
        "inference_timesteps": 10,
        "cfg_value": 2.0,
        "output_format": "wav",
        "concurrency": 1,
        "nano_gpu_memory_utilization": 0.49,
        "nano_enforce_eager": false
      },
      "scenarios": [
        {
          "code": "PS",
          "label": "plain_short_english",
          "description": "Short English generation without voice control or reference audio."
        },
        {
          "code": "DM",
          "label": "designed_medium_english",
          "description": "Medium English generation with a fixed voice-design instruction."
        },
        {
          "code": "MN",
          "label": "designed_norwegian",
          "description": "Norwegian generation with the same fixed voice-design instruction."
        },
        {
          "code": "CR",
          "label": "reference_clone",
          "description": "English zero-shot clone with the same reference audio mounted into both containers."
        }
      ],
      "backends": [
        {
          "backend": "native",
          "image": {
            "tag": "voxcpmtts:benchmark-native",
            "id": "sha256:68787abb3ecca6fa04d579e43f93d0ce30d3e74d89497450896e1d04fe7d1413",
            "created": "2026-10-07T10:06:02.193219714Z",
            "size_bytes": 4832268446,
            "repo_digests": [
              "voxcpmtts@sha256:68787abb3ecca6fa04d579e43f93d0ce30d3e74d89497450896e1d04fe7d1413"
            ],
            "labels": {}
          },
          "runtime_versions": {
            "python": "3.13.16",
            "torch": "2.10.0+cu128",
            "transformers": "5.3.0",
            "huggingface-hub": null,
            "nano-vllm-voxcpm": null,
            "flash-attn": null,
            "triton": "3.6.0"
          },
          "api_ready_seconds": 6.056,
          "status_before_load": {
            "msg": "pong",
            "type": "VoxCPMTTS",
            "version": "0.1-snapshot",
            "build_id": "stable",
            "runtime": "GPU x1 (visible=0) [0:NVIDIA GeForce RTX 5070 Ti]",
            "device": "auto",
            "model_id": "openbmb/VoxCPM2",
            "load_denoiser": false,
            "load_asr": false,
            "asr_model_id": "iic/SenseVoiceSmall",
            "optimize": false,
            "languages": [
              "Arabic",
              "Burmese",
              "Chinese",
              "Danish",
              "Dutch",
              "English",
              "Finnish",
              "French",
              "German",
              "Greek",
              "Hebrew",
              "Hindi",
              "Indonesian",
              "Italian",
              "Japanese",
              "Khmer",
              "Korean",
              "Lao",
              "Malay",
              "Norwegian",
              "Polish",
              "Portuguese",
              "Russian",
              "Spanish",
              "Swahili",
              "Swedish",
              "Tagalog",
              "Thai",
              "Turkish",
              "Vietnamese"
            ],
            "loaded_model_devices": [],
            "output_formats": {
              "wav": {
                "label": "WAV",
                "extension": "wav",
                "media_type": "audio/wav"
              },
              "mp3": {
                "label": "MP3",
                "extension": "mp3",
                "media_type": "audio/mpeg"
              },
              "flac": {
                "label": "FLAC",
                "extension": "flac",
                "media_type": "audio/flac"
              },
              "ogg": {
                "label": "OGG Vorbis",
                "extension": "ogg",
                "media_type": "audio/ogg"
              }
            },
            "stream_formats": {
              "wav": {
                "label": "Full WAV response",
                "extension": "wav",
                "media_type": "audio/wav"
              },
              "mp3": {
                "label": "Full MP3 response",
                "extension": "mp3",
                "media_type": "audio/mpeg"
              }
            }
          },
          "status_after": {
            "msg": "pong",
            "type": "VoxCPMTTS",
            "version": "0.1-snapshot",
            "build_id": "stable",
            "runtime": "GPU x1 (visible=0) [0:NVIDIA GeForce RTX 5070 Ti]",
            "device": "auto",
            "model_id": "openbmb/VoxCPM2",
            "load_denoiser": false,
            "load_asr": false,
            "asr_model_id": "iic/SenseVoiceSmall",
            "optimize": false,
            "languages": [
              "Arabic",
              "Burmese",
              "Chinese",
              "Danish",
              "Dutch",
              "English",
              "Finnish",
              "French",
              "German",
              "Greek",
              "Hebrew",
              "Hindi",
              "Indonesian",
              "Italian",
              "Japanese",
              "Khmer",
              "Korean",
              "Lao",
              "Malay",
              "Norwegian",
              "Polish",
              "Portuguese",
              "Russian",
              "Spanish",
              "Swahili",
              "Swedish",
              "Tagalog",
              "Thai",
              "Turkish",
              "Vietnamese"
            ],
            "loaded_model_devices": [
              "cuda:0"
            ],
            "output_formats": {
              "wav": {
                "label": "WAV",
                "extension": "wav",
                "media_type": "audio/wav"
              },
              "mp3": {
                "label": "MP3",
                "extension": "mp3",
                "media_type": "audio/mpeg"
              },
              "flac": {
                "label": "FLAC",
                "extension": "flac",
                "media_type": "audio/flac"
              },
              "ogg": {
                "label": "OGG Vorbis",
                "extension": "ogg",
                "media_type": "audio/ogg"
              }
            },
            "stream_formats": {
              "wav": {
                "label": "Full WAV response",
                "extension": "wav",
                "media_type": "audio/wav"
              },
              "mp3": {
                "label": "Full MP3 response",
                "extension": "mp3",
                "media_type": "audio/mpeg"
              }
            }
          },
          "gpu_baseline": {
            "index": 0,
            "name": "NVIDIA GeForce RTX 5070 Ti",
            "driver": "610.88",
            "total_mib": 16303,
            "used_mib": 1890,
            "utilization_percent": 2
          },
          "cold_request": {
            "wall_seconds": 19.3369,
            "audio_seconds": 3.84,
            "rtf": 5.0356,
            "bytes": 368684,
            "sample_rate": 48000,
            "channels": 1,
            "response_duration": "3.840",
            "response_model": "openbmb/VoxCPM2",
            "gpu": {
              "samples": 80,
              "peak_mib": 7661,
              "peak_delta_mib": 5771,
              "mean_utilization_percent": 11.6,
              "peak_utilization_percent": 46
            }
          },
          "gpu_final": {
            "index": 0,
            "name": "NVIDIA GeForce RTX 5070 Ti",
            "driver": "610.88",
            "total_mib": 16303,
            "used_mib": 8349,
            "utilization_percent": 1
          },
          "gpu_final_delta_mib": 6459,
          "container_memory_before_load": {
            "current_mib": 475.6,
            "peak_mib": 478.2
          },
          "container_memory_after_load": {
            "current_mib": 3506.5,
            "peak_mib": 10866.8
          },
          "container_memory_final": {
            "current_mib": 3590.1,
            "peak_mib": 10866.8
          },
          "scenarios": [
            {
              "code": "PS",
              "label": "plain_short_english",
              "description": "Short English generation without voice control or reference audio.",
              "payload": {
                "language": "English",
                "device": "cuda:0",
                "output_format": "wav",
                "cfg_value": 2.0,
                "inference_timesteps": 10,
                "normalize": false,
                "denoise": false,
                "text": "The morning coffee negotiated hard, but eventually agreed to help."
              },
              "warmups": [
                {
                  "wall_seconds": 4.6025,
                  "audio_seconds": 3.68,
                  "rtf": 1.2507,
                  "bytes": 353324,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "3.680",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 18,
                    "peak_mib": 7854,
                    "peak_delta_mib": 5964,
                    "mean_utilization_percent": 22.2,
                    "peak_utilization_percent": 37
                  }
                },
                {
                  "wall_seconds": 4.539,
                  "audio_seconds": 4.0,
                  "rtf": 1.1348,
                  "bytes": 384044,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "4.000",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 19,
                    "peak_mib": 7863,
                    "peak_delta_mib": 5973,
                    "mean_utilization_percent": 26.4,
                    "peak_utilization_percent": 41
                  }
                }
              ],
              "measurements": [
                {
                  "wall_seconds": 4.167,
                  "audio_seconds": 3.52,
                  "rtf": 1.1838,
                  "bytes": 337964,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "3.520",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 17,
                    "peak_mib": 7863,
                    "peak_delta_mib": 5973,
                    "mean_utilization_percent": 37.3,
                    "peak_utilization_percent": 47
                  }
                },
                {
                  "wall_seconds": 4.0584,
                  "audio_seconds": 3.52,
                  "rtf": 1.1529,
                  "bytes": 337964,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "3.520",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 17,
                    "peak_mib": 7855,
                    "peak_delta_mib": 5965,
                    "mean_utilization_percent": 26.8,
                    "peak_utilization_percent": 44
                  }
                },
                {
                  "wall_seconds": 4.6097,
                  "audio_seconds": 4.0,
                  "rtf": 1.1524,
                  "bytes": 384044,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "4.000",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 19,
                    "peak_mib": 7862,
                    "peak_delta_mib": 5972,
                    "mean_utilization_percent": 34.0,
                    "peak_utilization_percent": 47
                  }
                },
                {
                  "wall_seconds": 3.9071,
                  "audio_seconds": 3.36,
                  "rtf": 1.1628,
                  "bytes": 322604,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "3.360",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 16,
                    "peak_mib": 7855,
                    "peak_delta_mib": 5965,
                    "mean_utilization_percent": 29.1,
                    "peak_utilization_percent": 41
                  }
                },
                {
                  "wall_seconds": 4.4934,
                  "audio_seconds": 4.0,
                  "rtf": 1.1234,
                  "bytes": 384044,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "4.000",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 18,
                    "peak_mib": 7861,
                    "peak_delta_mib": 5971,
                    "mean_utilization_percent": 34.0,
                    "peak_utilization_percent": 45
                  }
                },
                {
                  "wall_seconds": 3.6599,
                  "audio_seconds": 3.36,
                  "rtf": 1.0893,
                  "bytes": 322604,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "3.360",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 15,
                    "peak_mib": 7862,
                    "peak_delta_mib": 5972,
                    "mean_utilization_percent": 28.2,
                    "peak_utilization_percent": 47
                  }
                },
                {
                  "wall_seconds": 4.5232,
                  "audio_seconds": 4.0,
                  "rtf": 1.1308,
                  "bytes": 384044,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "4.000",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 18,
                    "peak_mib": 7855,
                    "peak_delta_mib": 5965,
                    "mean_utilization_percent": 36.3,
                    "peak_utilization_percent": 52
                  }
                },
                {
                  "wall_seconds": 5.7037,
                  "audio_seconds": 4.8,
                  "rtf": 1.1883,
                  "bytes": 460844,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "4.800",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 23,
                    "peak_mib": 7862,
                    "peak_delta_mib": 5972,
                    "mean_utilization_percent": 27.4,
                    "peak_utilization_percent": 43
                  }
                },
                {
                  "wall_seconds": 4.2667,
                  "audio_seconds": 3.52,
                  "rtf": 1.2121,
                  "bytes": 337964,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "3.520",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 17,
                    "peak_mib": 7856,
                    "peak_delta_mib": 5966,
                    "mean_utilization_percent": 25.7,
                    "peak_utilization_percent": 37
                  }
                },
                {
                  "wall_seconds": 4.981,
                  "audio_seconds": 4.32,
                  "rtf": 1.153,
                  "bytes": 414764,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "4.320",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 19,
                    "peak_mib": 7863,
                    "peak_delta_mib": 5973,
                    "mean_utilization_percent": 31.8,
                    "peak_utilization_percent": 48
                  }
                }
              ],
              "summary": {
                "calls": 10,
                "wall_median_seconds": 4.3801,
                "wall_p95_seconds": 5.3785,
                "wall_mean_seconds": 4.437,
                "rtf_median": 1.153,
                "rtf_p95": 1.2014,
                "rtf_mean": 1.1549,
                "audio_mean_seconds": 3.84,
                "gpu_peak_mib": 7863,
                "gpu_peak_delta_mib": 5973,
                "gpu_mean_utilization_percent": 31.1
              },
              "container_memory_after": {
                "current_mib": 3506.3,
                "peak_mib": 10866.8
              }
            },
            {
              "code": "DM",
              "label": "designed_medium_english",
              "description": "Medium English generation with a fixed voice-design instruction.",
              "payload": {
                "language": "English",
                "device": "cuda:0",
                "output_format": "wav",
                "cfg_value": 2.0,
                "inference_timesteps": 10,
                "normalize": false,
                "denoise": false,
                "text": "Local speech generation keeps private text on your own machine while producing clear, natural audio for applications, accessibility tools, and creative work.",
                "control": "warm adult female narrator, clear and friendly"
              },
              "warmups": [
                {
                  "wall_seconds": 10.1395,
                  "audio_seconds": 8.8,
                  "rtf": 1.1522,
                  "bytes": 844844,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "8.800",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 38,
                    "peak_mib": 7864,
                    "peak_delta_mib": 5974,
                    "mean_utilization_percent": 32.7,
                    "peak_utilization_percent": 46
                  }
                },
                {
                  "wall_seconds": 9.8613,
                  "audio_seconds": 8.64,
                  "rtf": 1.1414,
                  "bytes": 829484,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "8.640",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 37,
                    "peak_mib": 7864,
                    "peak_delta_mib": 5974,
                    "mean_utilization_percent": 32.5,
                    "peak_utilization_percent": 45
                  }
                }
              ],
              "measurements": [
                {
                  "wall_seconds": 11.2412,
                  "audio_seconds": 9.44,
                  "rtf": 1.1908,
                  "bytes": 906284,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "9.440",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 43,
                    "peak_mib": 7866,
                    "peak_delta_mib": 5976,
                    "mean_utilization_percent": 28.8,
                    "peak_utilization_percent": 45
                  }
                },
                {
                  "wall_seconds": 11.8399,
                  "audio_seconds": 9.92,
                  "rtf": 1.1935,
                  "bytes": 952364,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "9.920",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 46,
                    "peak_mib": 7865,
                    "peak_delta_mib": 5975,
                    "mean_utilization_percent": 31.7,
                    "peak_utilization_percent": 56
                  }
                },
                {
                  "wall_seconds": 9.8484,
                  "audio_seconds": 8.48,
                  "rtf": 1.1614,
                  "bytes": 814124,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "8.480",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 37,
                    "peak_mib": 8002,
                    "peak_delta_mib": 6112,
                    "mean_utilization_percent": 29.5,
                    "peak_utilization_percent": 45
                  }
                },
                {
                  "wall_seconds": 11.6134,
                  "audio_seconds": 10.4,
                  "rtf": 1.1167,
                  "bytes": 998444,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "10.400",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 44,
                    "peak_mib": 8061,
                    "peak_delta_mib": 6171,
                    "mean_utilization_percent": 32.3,
                    "peak_utilization_percent": 47
                  }
                },
                {
                  "wall_seconds": 11.4614,
                  "audio_seconds": 9.28,
                  "rtf": 1.2351,
                  "bytes": 890924,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "9.280",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 43,
                    "peak_mib": 8170,
                    "peak_delta_mib": 6280,
                    "mean_utilization_percent": 29.9,
                    "peak_utilization_percent": 48
                  }
                },
                {
                  "wall_seconds": 10.2654,
                  "audio_seconds": 9.12,
                  "rtf": 1.1256,
                  "bytes": 875564,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "9.120",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 40,
                    "peak_mib": 8095,
                    "peak_delta_mib": 6205,
                    "mean_utilization_percent": 31.5,
                    "peak_utilization_percent": 47
                  }
                },
                {
                  "wall_seconds": 10.9945,
                  "audio_seconds": 8.64,
                  "rtf": 1.2725,
                  "bytes": 829484,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "8.640",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 43,
                    "peak_mib": 8093,
                    "peak_delta_mib": 6203,
                    "mean_utilization_percent": 29.0,
                    "peak_utilization_percent": 47
                  }
                },
                {
                  "wall_seconds": 10.1818,
                  "audio_seconds": 8.64,
                  "rtf": 1.1784,
                  "bytes": 829484,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "8.640",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 38,
                    "peak_mib": 8086,
                    "peak_delta_mib": 6196,
                    "mean_utilization_percent": 30.4,
                    "peak_utilization_percent": 44
                  }
                },
                {
                  "wall_seconds": 12.4848,
                  "audio_seconds": 9.76,
                  "rtf": 1.2792,
                  "bytes": 937004,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "9.760",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 44,
                    "peak_mib": 8370,
                    "peak_delta_mib": 6480,
                    "mean_utilization_percent": 26.7,
                    "peak_utilization_percent": 43
                  }
                },
                {
                  "wall_seconds": 10.7598,
                  "audio_seconds": 8.64,
                  "rtf": 1.2453,
                  "bytes": 829484,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "8.640",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 39,
                    "peak_mib": 8410,
                    "peak_delta_mib": 6520,
                    "mean_utilization_percent": 28.8,
                    "peak_utilization_percent": 42
                  }
                }
              ],
              "summary": {
                "calls": 10,
                "wall_median_seconds": 11.1179,
                "wall_p95_seconds": 12.1946,
                "wall_mean_seconds": 11.0691,
                "rtf_median": 1.1922,
                "rtf_p95": 1.2762,
                "rtf_mean": 1.1999,
                "audio_mean_seconds": 9.232,
                "gpu_peak_mib": 8410,
                "gpu_peak_delta_mib": 6520,
                "gpu_mean_utilization_percent": 29.9
              },
              "container_memory_after": {
                "current_mib": 3506.9,
                "peak_mib": 10866.8
              }
            },
            {
              "code": "MN",
              "label": "designed_norwegian",
              "description": "Norwegian generation with the same fixed voice-design instruction.",
              "payload": {
                "language": "Norwegian",
                "device": "cuda:0",
                "output_format": "wav",
                "cfg_value": 2.0,
                "inference_timesteps": 10,
                "normalize": false,
                "denoise": false,
                "text": "Denne testen måler hvor raskt modellen lager tydelig norsk tale med de samme innstillingene.",
                "control": "warm adult female narrator, clear and friendly"
              },
              "warmups": [
                {
                  "wall_seconds": 7.3226,
                  "audio_seconds": 5.44,
                  "rtf": 1.3461,
                  "bytes": 522284,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "5.440",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 26,
                    "peak_mib": 8274,
                    "peak_delta_mib": 6384,
                    "mean_utilization_percent": 27.6,
                    "peak_utilization_percent": 45
                  }
                },
                {
                  "wall_seconds": 7.3394,
                  "audio_seconds": 5.6,
                  "rtf": 1.3106,
                  "bytes": 537644,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "5.600",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 29,
                    "peak_mib": 8149,
                    "peak_delta_mib": 6259,
                    "mean_utilization_percent": 28.7,
                    "peak_utilization_percent": 44
                  }
                }
              ],
              "measurements": [
                {
                  "wall_seconds": 5.6976,
                  "audio_seconds": 5.12,
                  "rtf": 1.1128,
                  "bytes": 491564,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "5.120",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 21,
                    "peak_mib": 8149,
                    "peak_delta_mib": 6259,
                    "mean_utilization_percent": 29.2,
                    "peak_utilization_percent": 45
                  }
                },
                {
                  "wall_seconds": 4.3653,
                  "audio_seconds": 4.0,
                  "rtf": 1.0913,
                  "bytes": 384044,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "4.000",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 18,
                    "peak_mib": 8368,
                    "peak_delta_mib": 6478,
                    "mean_utilization_percent": 32.3,
                    "peak_utilization_percent": 46
                  }
                },
                {
                  "wall_seconds": 7.2195,
                  "audio_seconds": 6.08,
                  "rtf": 1.1874,
                  "bytes": 583724,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "6.080",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 28,
                    "peak_mib": 8417,
                    "peak_delta_mib": 6527,
                    "mean_utilization_percent": 31.6,
                    "peak_utilization_percent": 46
                  }
                },
                {
                  "wall_seconds": 7.2642,
                  "audio_seconds": 6.56,
                  "rtf": 1.1073,
                  "bytes": 629804,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "6.560",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 28,
                    "peak_mib": 8420,
                    "peak_delta_mib": 6530,
                    "mean_utilization_percent": 29.2,
                    "peak_utilization_percent": 48
                  }
                },
                {
                  "wall_seconds": 5.7981,
                  "audio_seconds": 4.96,
                  "rtf": 1.169,
                  "bytes": 476204,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "4.960",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 23,
                    "peak_mib": 8362,
                    "peak_delta_mib": 6472,
                    "mean_utilization_percent": 29.3,
                    "peak_utilization_percent": 45
                  }
                },
                {
                  "wall_seconds": 5.7691,
                  "audio_seconds": 4.64,
                  "rtf": 1.2433,
                  "bytes": 445484,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "4.640",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 23,
                    "peak_mib": 8168,
                    "peak_delta_mib": 6278,
                    "mean_utilization_percent": 28.5,
                    "peak_utilization_percent": 46
                  }
                },
                {
                  "wall_seconds": 7.3345,
                  "audio_seconds": 6.24,
                  "rtf": 1.1754,
                  "bytes": 599084,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "6.240",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 29,
                    "peak_mib": 8168,
                    "peak_delta_mib": 6278,
                    "mean_utilization_percent": 32.7,
                    "peak_utilization_percent": 46
                  }
                },
                {
                  "wall_seconds": 7.7826,
                  "audio_seconds": 6.24,
                  "rtf": 1.2472,
                  "bytes": 599084,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "6.240",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 30,
                    "peak_mib": 8168,
                    "peak_delta_mib": 6278,
                    "mean_utilization_percent": 26.7,
                    "peak_utilization_percent": 46
                  }
                },
                {
                  "wall_seconds": 5.3838,
                  "audio_seconds": 4.64,
                  "rtf": 1.1603,
                  "bytes": 445484,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "4.640",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 22,
                    "peak_mib": 8160,
                    "peak_delta_mib": 6270,
                    "mean_utilization_percent": 29.6,
                    "peak_utilization_percent": 44
                  }
                },
                {
                  "wall_seconds": 5.549,
                  "audio_seconds": 5.12,
                  "rtf": 1.0838,
                  "bytes": 491564,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "5.120",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 22,
                    "peak_mib": 8309,
                    "peak_delta_mib": 6419,
                    "mean_utilization_percent": 27.4,
                    "peak_utilization_percent": 45
                  }
                }
              ],
              "summary": {
                "calls": 10,
                "wall_median_seconds": 5.7836,
                "wall_p95_seconds": 7.581,
                "wall_mean_seconds": 6.2164,
                "rtf_median": 1.1646,
                "rtf_p95": 1.2454,
                "rtf_mean": 1.1578,
                "audio_mean_seconds": 5.36,
                "gpu_peak_mib": 8420,
                "gpu_peak_delta_mib": 6530,
                "gpu_mean_utilization_percent": 29.6
              },
              "container_memory_after": {
                "current_mib": 3508.3,
                "peak_mib": 10866.8
              }
            },
            {
              "code": "CR",
              "label": "reference_clone",
              "description": "English zero-shot clone with the same reference audio mounted into both containers.",
              "payload": {
                "language": "English",
                "device": "cuda:0",
                "output_format": "wav",
                "cfg_value": 2.0,
                "inference_timesteps": 10,
                "normalize": false,
                "denoise": false,
                "text": "This sample checks how quickly the model can carry a reference voice into a new sentence.",
                "ref_audio": "/benchmark/original_clone.mp3"
              },
              "warmups": [
                {
                  "wall_seconds": 17.6292,
                  "audio_seconds": 5.6,
                  "rtf": 3.1481,
                  "bytes": 537644,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "5.600",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 73,
                    "peak_mib": 8293,
                    "peak_delta_mib": 6403,
                    "mean_utilization_percent": 11.1,
                    "peak_utilization_percent": 44
                  }
                },
                {
                  "wall_seconds": 5.811,
                  "audio_seconds": 4.96,
                  "rtf": 1.1716,
                  "bytes": 476204,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "4.960",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 23,
                    "peak_mib": 8425,
                    "peak_delta_mib": 6535,
                    "mean_utilization_percent": 31.4,
                    "peak_utilization_percent": 44
                  }
                }
              ],
              "measurements": [
                {
                  "wall_seconds": 5.6347,
                  "audio_seconds": 4.8,
                  "rtf": 1.1739,
                  "bytes": 460844,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "4.800",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 22,
                    "peak_mib": 8347,
                    "peak_delta_mib": 6457,
                    "mean_utilization_percent": 30.9,
                    "peak_utilization_percent": 44
                  }
                },
                {
                  "wall_seconds": 5.2156,
                  "audio_seconds": 4.64,
                  "rtf": 1.1241,
                  "bytes": 445484,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "4.640",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 21,
                    "peak_mib": 8340,
                    "peak_delta_mib": 6450,
                    "mean_utilization_percent": 30.0,
                    "peak_utilization_percent": 45
                  }
                },
                {
                  "wall_seconds": 4.4747,
                  "audio_seconds": 4.0,
                  "rtf": 1.1187,
                  "bytes": 384044,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "4.000",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 18,
                    "peak_mib": 8347,
                    "peak_delta_mib": 6457,
                    "mean_utilization_percent": 32.1,
                    "peak_utilization_percent": 47
                  }
                },
                {
                  "wall_seconds": 4.7618,
                  "audio_seconds": 4.32,
                  "rtf": 1.1023,
                  "bytes": 414764,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "4.320",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 19,
                    "peak_mib": 8349,
                    "peak_delta_mib": 6459,
                    "mean_utilization_percent": 28.3,
                    "peak_utilization_percent": 45
                  }
                },
                {
                  "wall_seconds": 6.0473,
                  "audio_seconds": 4.96,
                  "rtf": 1.2192,
                  "bytes": 476204,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "4.960",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 23,
                    "peak_mib": 8348,
                    "peak_delta_mib": 6458,
                    "mean_utilization_percent": 26.4,
                    "peak_utilization_percent": 44
                  }
                },
                {
                  "wall_seconds": 5.5075,
                  "audio_seconds": 4.64,
                  "rtf": 1.187,
                  "bytes": 445484,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "4.640",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 22,
                    "peak_mib": 8354,
                    "peak_delta_mib": 6464,
                    "mean_utilization_percent": 32.0,
                    "peak_utilization_percent": 45
                  }
                },
                {
                  "wall_seconds": 5.3604,
                  "audio_seconds": 4.48,
                  "rtf": 1.1965,
                  "bytes": 430124,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "4.480",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 22,
                    "peak_mib": 8362,
                    "peak_delta_mib": 6472,
                    "mean_utilization_percent": 31.4,
                    "peak_utilization_percent": 45
                  }
                },
                {
                  "wall_seconds": 5.2353,
                  "audio_seconds": 4.48,
                  "rtf": 1.1686,
                  "bytes": 430124,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "4.480",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 20,
                    "peak_mib": 8355,
                    "peak_delta_mib": 6465,
                    "mean_utilization_percent": 32.6,
                    "peak_utilization_percent": 45
                  }
                },
                {
                  "wall_seconds": 5.7813,
                  "audio_seconds": 4.8,
                  "rtf": 1.2044,
                  "bytes": 460844,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "4.800",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 22,
                    "peak_mib": 8363,
                    "peak_delta_mib": 6473,
                    "mean_utilization_percent": 30.5,
                    "peak_utilization_percent": 44
                  }
                },
                {
                  "wall_seconds": 5.4564,
                  "audio_seconds": 4.8,
                  "rtf": 1.1367,
                  "bytes": 460844,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "4.800",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 22,
                    "peak_mib": 8349,
                    "peak_delta_mib": 6459,
                    "mean_utilization_percent": 29.6,
                    "peak_utilization_percent": 45
                  }
                }
              ],
              "summary": {
                "calls": 10,
                "wall_median_seconds": 5.4084,
                "wall_p95_seconds": 5.9276,
                "wall_mean_seconds": 5.3475,
                "rtf_median": 1.1713,
                "rtf_p95": 1.2125,
                "rtf_mean": 1.1631,
                "audio_mean_seconds": 4.592,
                "gpu_peak_mib": 8363,
                "gpu_peak_delta_mib": 6473,
                "gpu_mean_utilization_percent": 30.4
              },
              "container_memory_after": {
                "current_mib": 3589.2,
                "peak_mib": 10866.8
              }
            }
          ]
        },
        {
          "backend": "nano",
          "image": {
            "tag": "voxcpmtts:benchmark-nano",
            "id": "sha256:3a4e1ccda4af76e389521374637905c6462974480dac20c476da9414255923f1",
            "created": "2026-10-07T16:21:51.274308704Z",
            "size_bytes": 4847177171,
            "repo_digests": [
              "voxcpmtts@sha256:3a4e1ccda4af76e389521374637905c6462974480dac20c476da9414255923f1"
            ],
            "labels": {
              "org.opencontainers.image.created": "2026-10-07T14:52:34Z",
              "org.opencontainers.image.revision": "ac1d0b3b2bcb",
              "org.opencontainers.image.source": "https://github.com/hangry-labs/VoxCPMTTS"
            }
          },
          "runtime_versions": {
            "python": "3.13.16",
            "torch": "2.8.0+cu128",
            "transformers": "5.3.0",
            "huggingface-hub": null,
            "nano-vllm-voxcpm": "2.0.4",
            "flash-attn": null,
            "triton": "3.4.0"
          },
          "api_ready_seconds": 4.037,
          "status_before_load": {
            "msg": "pong",
            "type": "VoxCPMTTS",
            "version": "0.1-snapshot",
            "build_id": "2026-10-07T14:52:34Z@ac1d0b3b2bcb",
            "build_date": "2026-10-07T14:52:34Z",
            "revision": "ac1d0b3b2bcb",
            "runtime": "GPU x1 (visible=0) [0:NVIDIA GeForce RTX 5070 Ti]",
            "device": "auto",
            "backend": "nano",
            "model_id": "openbmb/VoxCPM2",
            "load_denoiser": false,
            "load_asr": false,
            "asr_model_id": "iic/SenseVoiceSmall",
            "optimize": false,
            "languages": [
              "Arabic",
              "Burmese",
              "Chinese",
              "Danish",
              "Dutch",
              "English",
              "Finnish",
              "French",
              "German",
              "Greek",
              "Hebrew",
              "Hindi",
              "Indonesian",
              "Italian",
              "Japanese",
              "Khmer",
              "Korean",
              "Lao",
              "Malay",
              "Norwegian",
              "Polish",
              "Portuguese",
              "Russian",
              "Spanish",
              "Swahili",
              "Swedish",
              "Tagalog",
              "Thai",
              "Turkish",
              "Vietnamese"
            ],
            "loaded_model_devices": [],
            "output_formats": {
              "wav": {
                "label": "WAV",
                "extension": "wav",
                "media_type": "audio/wav"
              },
              "mp3": {
                "label": "MP3",
                "extension": "mp3",
                "media_type": "audio/mpeg"
              },
              "flac": {
                "label": "FLAC",
                "extension": "flac",
                "media_type": "audio/flac"
              },
              "ogg": {
                "label": "OGG Vorbis",
                "extension": "ogg",
                "media_type": "audio/ogg"
              }
            },
            "stream_formats": {
              "wav": {
                "label": "Full WAV response",
                "extension": "wav",
                "media_type": "audio/wav"
              },
              "mp3": {
                "label": "Full MP3 response",
                "extension": "mp3",
                "media_type": "audio/mpeg"
              }
            }
          },
          "status_after": {
            "msg": "pong",
            "type": "VoxCPMTTS",
            "version": "0.1-snapshot",
            "build_id": "2026-10-07T14:52:34Z@ac1d0b3b2bcb",
            "build_date": "2026-10-07T14:52:34Z",
            "revision": "ac1d0b3b2bcb",
            "runtime": "GPU x1 (visible=0) [0:NVIDIA GeForce RTX 5070 Ti]",
            "device": "auto",
            "backend": "nano",
            "model_id": "openbmb/VoxCPM2",
            "load_denoiser": false,
            "load_asr": false,
            "asr_model_id": "iic/SenseVoiceSmall",
            "optimize": false,
            "languages": [
              "Arabic",
              "Burmese",
              "Chinese",
              "Danish",
              "Dutch",
              "English",
              "Finnish",
              "French",
              "German",
              "Greek",
              "Hebrew",
              "Hindi",
              "Indonesian",
              "Italian",
              "Japanese",
              "Khmer",
              "Korean",
              "Lao",
              "Malay",
              "Norwegian",
              "Polish",
              "Portuguese",
              "Russian",
              "Spanish",
              "Swahili",
              "Swedish",
              "Tagalog",
              "Thai",
              "Turkish",
              "Vietnamese"
            ],
            "loaded_model_devices": [
              "cuda:0"
            ],
            "output_formats": {
              "wav": {
                "label": "WAV",
                "extension": "wav",
                "media_type": "audio/wav"
              },
              "mp3": {
                "label": "MP3",
                "extension": "mp3",
                "media_type": "audio/mpeg"
              },
              "flac": {
                "label": "FLAC",
                "extension": "flac",
                "media_type": "audio/flac"
              },
              "ogg": {
                "label": "OGG Vorbis",
                "extension": "ogg",
                "media_type": "audio/ogg"
              }
            },
            "stream_formats": {
              "wav": {
                "label": "Full WAV response",
                "extension": "wav",
                "media_type": "audio/wav"
              },
              "mp3": {
                "label": "Full MP3 response",
                "extension": "mp3",
                "media_type": "audio/mpeg"
              }
            }
          },
          "gpu_baseline": {
            "index": 0,
            "name": "NVIDIA GeForce RTX 5070 Ti",
            "driver": "610.88",
            "total_mib": 16303,
            "used_mib": 2063,
            "utilization_percent": 8
          },
          "cold_request": {
            "wall_seconds": 22.4175,
            "audio_seconds": 4.16,
            "rtf": 5.3888,
            "bytes": 399404,
            "sample_rate": 48000,
            "channels": 1,
            "response_duration": "4.160",
            "response_model": "openbmb/VoxCPM2",
            "gpu": {
              "samples": 95,
              "peak_mib": 10414,
              "peak_delta_mib": 8351,
              "mean_utilization_percent": 9.2,
              "peak_utilization_percent": 100
            }
          },
          "gpu_final": {
            "index": 0,
            "name": "NVIDIA GeForce RTX 5070 Ti",
            "driver": "610.88",
            "total_mib": 16303,
            "used_mib": 9564,
            "utilization_percent": 2
          },
          "gpu_final_delta_mib": 7501,
          "container_memory_before_load": {
            "current_mib": 369.8,
            "peak_mib": 372.7
          },
          "container_memory_after_load": {
            "current_mib": 2672.4,
            "peak_mib": 2701.8
          },
          "container_memory_final": {
            "current_mib": 3067.5,
            "peak_mib": 3324.3
          },
          "scenarios": [
            {
              "code": "PS",
              "label": "plain_short_english",
              "description": "Short English generation without voice control or reference audio.",
              "payload": {
                "language": "English",
                "device": "cuda:0",
                "output_format": "wav",
                "cfg_value": 2.0,
                "inference_timesteps": 10,
                "normalize": false,
                "denoise": false,
                "text": "The morning coffee negotiated hard, but eventually agreed to help."
              },
              "warmups": [
                {
                  "wall_seconds": 1.0834,
                  "audio_seconds": 3.84,
                  "rtf": 0.2821,
                  "bytes": 368684,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "3.840",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 6,
                    "peak_mib": 9502,
                    "peak_delta_mib": 7439,
                    "mean_utilization_percent": 25.7,
                    "peak_utilization_percent": 77
                  }
                },
                {
                  "wall_seconds": 1.0014,
                  "audio_seconds": 4.0,
                  "rtf": 0.2504,
                  "bytes": 384044,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "4.000",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 6,
                    "peak_mib": 9502,
                    "peak_delta_mib": 7439,
                    "mean_utilization_percent": 65.3,
                    "peak_utilization_percent": 82
                  }
                }
              ],
              "measurements": [
                {
                  "wall_seconds": 1.1387,
                  "audio_seconds": 4.0,
                  "rtf": 0.2847,
                  "bytes": 384044,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "4.000",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 6,
                    "peak_mib": 9494,
                    "peak_delta_mib": 7431,
                    "mean_utilization_percent": 33.7,
                    "peak_utilization_percent": 81
                  }
                },
                {
                  "wall_seconds": 1.2282,
                  "audio_seconds": 4.8,
                  "rtf": 0.2559,
                  "bytes": 460844,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "4.800",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 6,
                    "peak_mib": 9494,
                    "peak_delta_mib": 7431,
                    "mean_utilization_percent": 32.5,
                    "peak_utilization_percent": 85
                  }
                },
                {
                  "wall_seconds": 1.0627,
                  "audio_seconds": 3.84,
                  "rtf": 0.2767,
                  "bytes": 368684,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "3.840",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 6,
                    "peak_mib": 9494,
                    "peak_delta_mib": 7431,
                    "mean_utilization_percent": 63.8,
                    "peak_utilization_percent": 81
                  }
                },
                {
                  "wall_seconds": 1.344,
                  "audio_seconds": 4.8,
                  "rtf": 0.28,
                  "bytes": 460844,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "4.800",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 7,
                    "peak_mib": 9494,
                    "peak_delta_mib": 7431,
                    "mean_utilization_percent": 40.0,
                    "peak_utilization_percent": 78
                  }
                },
                {
                  "wall_seconds": 1.1481,
                  "audio_seconds": 4.16,
                  "rtf": 0.276,
                  "bytes": 399404,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "4.160",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 6,
                    "peak_mib": 9494,
                    "peak_delta_mib": 7431,
                    "mean_utilization_percent": 49.7,
                    "peak_utilization_percent": 84
                  }
                },
                {
                  "wall_seconds": 1.3264,
                  "audio_seconds": 4.64,
                  "rtf": 0.2859,
                  "bytes": 445484,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "4.640",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 6,
                    "peak_mib": 9494,
                    "peak_delta_mib": 7431,
                    "mean_utilization_percent": 50.0,
                    "peak_utilization_percent": 84
                  }
                },
                {
                  "wall_seconds": 0.944,
                  "audio_seconds": 3.68,
                  "rtf": 0.2565,
                  "bytes": 353324,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "3.680",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 5,
                    "peak_mib": 9502,
                    "peak_delta_mib": 7439,
                    "mean_utilization_percent": 30.0,
                    "peak_utilization_percent": 75
                  }
                },
                {
                  "wall_seconds": 1.0478,
                  "audio_seconds": 3.52,
                  "rtf": 0.2977,
                  "bytes": 337964,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "3.520",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 6,
                    "peak_mib": 9502,
                    "peak_delta_mib": 7439,
                    "mean_utilization_percent": 28.0,
                    "peak_utilization_percent": 84
                  }
                },
                {
                  "wall_seconds": 1.0808,
                  "audio_seconds": 4.16,
                  "rtf": 0.2598,
                  "bytes": 399404,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "4.160",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 6,
                    "peak_mib": 9502,
                    "peak_delta_mib": 7439,
                    "mean_utilization_percent": 32.8,
                    "peak_utilization_percent": 82
                  }
                },
                {
                  "wall_seconds": 1.0694,
                  "audio_seconds": 4.0,
                  "rtf": 0.2674,
                  "bytes": 384044,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "4.000",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 6,
                    "peak_mib": 9494,
                    "peak_delta_mib": 7431,
                    "mean_utilization_percent": 40.0,
                    "peak_utilization_percent": 78
                  }
                }
              ],
              "summary": {
                "calls": 10,
                "wall_median_seconds": 1.1098,
                "wall_p95_seconds": 1.3361,
                "wall_mean_seconds": 1.139,
                "rtf_median": 0.2763,
                "rtf_p95": 0.2924,
                "rtf_mean": 0.2741,
                "audio_mean_seconds": 4.16,
                "gpu_peak_mib": 9502,
                "gpu_peak_delta_mib": 7439,
                "gpu_mean_utilization_percent": 40.0
              },
              "container_memory_after": {
                "current_mib": 2676.1,
                "peak_mib": 2701.8
              }
            },
            {
              "code": "DM",
              "label": "designed_medium_english",
              "description": "Medium English generation with a fixed voice-design instruction.",
              "payload": {
                "language": "English",
                "device": "cuda:0",
                "output_format": "wav",
                "cfg_value": 2.0,
                "inference_timesteps": 10,
                "normalize": false,
                "denoise": false,
                "text": "Local speech generation keeps private text on your own machine while producing clear, natural audio for applications, accessibility tools, and creative work.",
                "control": "warm adult female narrator, clear and friendly"
              },
              "warmups": [
                {
                  "wall_seconds": 2.6789,
                  "audio_seconds": 9.92,
                  "rtf": 0.27,
                  "bytes": 952364,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "9.920",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 12,
                    "peak_mib": 9494,
                    "peak_delta_mib": 7431,
                    "mean_utilization_percent": 48.8,
                    "peak_utilization_percent": 84
                  }
                },
                {
                  "wall_seconds": 2.8458,
                  "audio_seconds": 8.96,
                  "rtf": 0.3176,
                  "bytes": 860204,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "8.960",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 10,
                    "peak_mib": 9502,
                    "peak_delta_mib": 7439,
                    "mean_utilization_percent": 55.0,
                    "peak_utilization_percent": 86
                  }
                }
              ],
              "measurements": [
                {
                  "wall_seconds": 2.9523,
                  "audio_seconds": 9.28,
                  "rtf": 0.3181,
                  "bytes": 890924,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "9.280",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 12,
                    "peak_mib": 9502,
                    "peak_delta_mib": 7439,
                    "mean_utilization_percent": 48.5,
                    "peak_utilization_percent": 83
                  }
                },
                {
                  "wall_seconds": 2.7632,
                  "audio_seconds": 9.6,
                  "rtf": 0.2878,
                  "bytes": 921644,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "9.600",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 12,
                    "peak_mib": 9474,
                    "peak_delta_mib": 7411,
                    "mean_utilization_percent": 51.8,
                    "peak_utilization_percent": 82
                  }
                },
                {
                  "wall_seconds": 2.5298,
                  "audio_seconds": 8.96,
                  "rtf": 0.2823,
                  "bytes": 860204,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "8.960",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 12,
                    "peak_mib": 9474,
                    "peak_delta_mib": 7411,
                    "mean_utilization_percent": 41.2,
                    "peak_utilization_percent": 81
                  }
                },
                {
                  "wall_seconds": 2.5078,
                  "audio_seconds": 9.28,
                  "rtf": 0.2702,
                  "bytes": 890924,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "9.280",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 11,
                    "peak_mib": 9474,
                    "peak_delta_mib": 7411,
                    "mean_utilization_percent": 58.7,
                    "peak_utilization_percent": 83
                  }
                },
                {
                  "wall_seconds": 2.7617,
                  "audio_seconds": 9.76,
                  "rtf": 0.283,
                  "bytes": 937004,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "9.760",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 12,
                    "peak_mib": 9482,
                    "peak_delta_mib": 7419,
                    "mean_utilization_percent": 34.9,
                    "peak_utilization_percent": 78
                  }
                },
                {
                  "wall_seconds": 3.106,
                  "audio_seconds": 10.88,
                  "rtf": 0.2855,
                  "bytes": 1044524,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "10.880",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 13,
                    "peak_mib": 9475,
                    "peak_delta_mib": 7412,
                    "mean_utilization_percent": 48.0,
                    "peak_utilization_percent": 80
                  }
                },
                {
                  "wall_seconds": 2.8919,
                  "audio_seconds": 10.72,
                  "rtf": 0.2698,
                  "bytes": 1029164,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "10.720",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 12,
                    "peak_mib": 9702,
                    "peak_delta_mib": 7639,
                    "mean_utilization_percent": 49.9,
                    "peak_utilization_percent": 82
                  }
                },
                {
                  "wall_seconds": 2.7315,
                  "audio_seconds": 9.92,
                  "rtf": 0.2753,
                  "bytes": 952364,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "9.920",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 12,
                    "peak_mib": 9705,
                    "peak_delta_mib": 7642,
                    "mean_utilization_percent": 37.6,
                    "peak_utilization_percent": 84
                  }
                },
                {
                  "wall_seconds": 2.4368,
                  "audio_seconds": 8.96,
                  "rtf": 0.272,
                  "bytes": 860204,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "8.960",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 11,
                    "peak_mib": 9705,
                    "peak_delta_mib": 7642,
                    "mean_utilization_percent": 58.3,
                    "peak_utilization_percent": 85
                  }
                },
                {
                  "wall_seconds": 2.2976,
                  "audio_seconds": 9.12,
                  "rtf": 0.2519,
                  "bytes": 875564,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "9.120",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 11,
                    "peak_mib": 9668,
                    "peak_delta_mib": 7605,
                    "mean_utilization_percent": 52.0,
                    "peak_utilization_percent": 84
                  }
                }
              ],
              "summary": {
                "calls": 10,
                "wall_median_seconds": 2.7466,
                "wall_p95_seconds": 3.0368,
                "wall_mean_seconds": 2.6979,
                "rtf_median": 0.2788,
                "rtf_p95": 0.3045,
                "rtf_mean": 0.2796,
                "audio_mean_seconds": 9.648,
                "gpu_peak_mib": 9705,
                "gpu_peak_delta_mib": 7642,
                "gpu_mean_utilization_percent": 48.1
              },
              "container_memory_after": {
                "current_mib": 2692.4,
                "peak_mib": 2701.8
              }
            },
            {
              "code": "MN",
              "label": "designed_norwegian",
              "description": "Norwegian generation with the same fixed voice-design instruction.",
              "payload": {
                "language": "Norwegian",
                "device": "cuda:0",
                "output_format": "wav",
                "cfg_value": 2.0,
                "inference_timesteps": 10,
                "normalize": false,
                "denoise": false,
                "text": "Denne testen måler hvor raskt modellen lager tydelig norsk tale med de samme innstillingene.",
                "control": "warm adult female narrator, clear and friendly"
              },
              "warmups": [
                {
                  "wall_seconds": 1.4953,
                  "audio_seconds": 5.44,
                  "rtf": 0.2749,
                  "bytes": 522284,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "5.440",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 7,
                    "peak_mib": 9548,
                    "peak_delta_mib": 7485,
                    "mean_utilization_percent": 44.3,
                    "peak_utilization_percent": 78
                  }
                },
                {
                  "wall_seconds": 1.4264,
                  "audio_seconds": 4.64,
                  "rtf": 0.3074,
                  "bytes": 445484,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "4.640",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 7,
                    "peak_mib": 9548,
                    "peak_delta_mib": 7485,
                    "mean_utilization_percent": 46.6,
                    "peak_utilization_percent": 81
                  }
                }
              ],
              "measurements": [
                {
                  "wall_seconds": 1.5716,
                  "audio_seconds": 5.92,
                  "rtf": 0.2655,
                  "bytes": 568364,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "5.920",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 8,
                    "peak_mib": 9515,
                    "peak_delta_mib": 7452,
                    "mean_utilization_percent": 30.6,
                    "peak_utilization_percent": 85
                  }
                },
                {
                  "wall_seconds": 1.5992,
                  "audio_seconds": 5.92,
                  "rtf": 0.2701,
                  "bytes": 568364,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "5.920",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 8,
                    "peak_mib": 9523,
                    "peak_delta_mib": 7460,
                    "mean_utilization_percent": 51.0,
                    "peak_utilization_percent": 82
                  }
                },
                {
                  "wall_seconds": 2.0514,
                  "audio_seconds": 7.36,
                  "rtf": 0.2787,
                  "bytes": 706604,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "7.360",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 10,
                    "peak_mib": 9516,
                    "peak_delta_mib": 7453,
                    "mean_utilization_percent": 44.4,
                    "peak_utilization_percent": 77
                  }
                },
                {
                  "wall_seconds": 2.163,
                  "audio_seconds": 5.6,
                  "rtf": 0.3862,
                  "bytes": 537644,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "5.600",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 8,
                    "peak_mib": 9516,
                    "peak_delta_mib": 7453,
                    "mean_utilization_percent": 28.1,
                    "peak_utilization_percent": 73
                  }
                },
                {
                  "wall_seconds": 1.4104,
                  "audio_seconds": 5.12,
                  "rtf": 0.2755,
                  "bytes": 491564,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "5.120",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 7,
                    "peak_mib": 9516,
                    "peak_delta_mib": 7453,
                    "mean_utilization_percent": 47.6,
                    "peak_utilization_percent": 84
                  }
                },
                {
                  "wall_seconds": 1.3455,
                  "audio_seconds": 4.64,
                  "rtf": 0.29,
                  "bytes": 445484,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "4.640",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 7,
                    "peak_mib": 9516,
                    "peak_delta_mib": 7453,
                    "mean_utilization_percent": 44.3,
                    "peak_utilization_percent": 79
                  }
                },
                {
                  "wall_seconds": 1.2764,
                  "audio_seconds": 5.12,
                  "rtf": 0.2493,
                  "bytes": 491564,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "5.120",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 6,
                    "peak_mib": 9626,
                    "peak_delta_mib": 7563,
                    "mean_utilization_percent": 51.3,
                    "peak_utilization_percent": 79
                  }
                },
                {
                  "wall_seconds": 1.3508,
                  "audio_seconds": 4.48,
                  "rtf": 0.3015,
                  "bytes": 430124,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "4.480",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 6,
                    "peak_mib": 9954,
                    "peak_delta_mib": 7891,
                    "mean_utilization_percent": 57.0,
                    "peak_utilization_percent": 83
                  }
                },
                {
                  "wall_seconds": 1.4682,
                  "audio_seconds": 5.12,
                  "rtf": 0.2868,
                  "bytes": 491564,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "5.120",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 7,
                    "peak_mib": 9966,
                    "peak_delta_mib": 7903,
                    "mean_utilization_percent": 44.9,
                    "peak_utilization_percent": 76
                  }
                },
                {
                  "wall_seconds": 1.5586,
                  "audio_seconds": 5.92,
                  "rtf": 0.2633,
                  "bytes": 568364,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "5.920",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 7,
                    "peak_mib": 9939,
                    "peak_delta_mib": 7876,
                    "mean_utilization_percent": 37.1,
                    "peak_utilization_percent": 84
                  }
                }
              ],
              "summary": {
                "calls": 10,
                "wall_median_seconds": 1.5134,
                "wall_p95_seconds": 2.1128,
                "wall_mean_seconds": 1.5795,
                "rtf_median": 0.2771,
                "rtf_p95": 0.3481,
                "rtf_mean": 0.2867,
                "audio_mean_seconds": 5.52,
                "gpu_peak_mib": 9966,
                "gpu_peak_delta_mib": 7903,
                "gpu_mean_utilization_percent": 43.6
              },
              "container_memory_after": {
                "current_mib": 2697.6,
                "peak_mib": 2701.8
              }
            },
            {
              "code": "CR",
              "label": "reference_clone",
              "description": "English zero-shot clone with the same reference audio mounted into both containers.",
              "payload": {
                "language": "English",
                "device": "cuda:0",
                "output_format": "wav",
                "cfg_value": 2.0,
                "inference_timesteps": 10,
                "normalize": false,
                "denoise": false,
                "text": "This sample checks how quickly the model can carry a reference voice into a new sentence.",
                "ref_audio": "/benchmark/original_clone.mp3"
              },
              "warmups": [
                {
                  "wall_seconds": 11.6984,
                  "audio_seconds": 4.64,
                  "rtf": 2.5212,
                  "bytes": 445484,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "4.640",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 51,
                    "peak_mib": 9939,
                    "peak_delta_mib": 7876,
                    "mean_utilization_percent": 9.9,
                    "peak_utilization_percent": 83
                  }
                },
                {
                  "wall_seconds": 1.2175,
                  "audio_seconds": 4.32,
                  "rtf": 0.2818,
                  "bytes": 414764,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "4.320",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 6,
                    "peak_mib": 9625,
                    "peak_delta_mib": 7562,
                    "mean_utilization_percent": 54.0,
                    "peak_utilization_percent": 81
                  }
                }
              ],
              "measurements": [
                {
                  "wall_seconds": 1.236,
                  "audio_seconds": 4.48,
                  "rtf": 0.2759,
                  "bytes": 430124,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "4.480",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 6,
                    "peak_mib": 9587,
                    "peak_delta_mib": 7524,
                    "mean_utilization_percent": 53.3,
                    "peak_utilization_percent": 83
                  }
                },
                {
                  "wall_seconds": 1.3923,
                  "audio_seconds": 4.96,
                  "rtf": 0.2807,
                  "bytes": 476204,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "4.960",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 7,
                    "peak_mib": 10009,
                    "peak_delta_mib": 7946,
                    "mean_utilization_percent": 35.0,
                    "peak_utilization_percent": 77
                  }
                },
                {
                  "wall_seconds": 1.1741,
                  "audio_seconds": 4.48,
                  "rtf": 0.2621,
                  "bytes": 430124,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "4.480",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 6,
                    "peak_mib": 9585,
                    "peak_delta_mib": 7522,
                    "mean_utilization_percent": 45.7,
                    "peak_utilization_percent": 77
                  }
                },
                {
                  "wall_seconds": 1.303,
                  "audio_seconds": 4.64,
                  "rtf": 0.2808,
                  "bytes": 445484,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "4.640",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 6,
                    "peak_mib": 9639,
                    "peak_delta_mib": 7576,
                    "mean_utilization_percent": 53.3,
                    "peak_utilization_percent": 79
                  }
                },
                {
                  "wall_seconds": 1.2046,
                  "audio_seconds": 4.32,
                  "rtf": 0.2788,
                  "bytes": 414764,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "4.320",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 6,
                    "peak_mib": 9601,
                    "peak_delta_mib": 7538,
                    "mean_utilization_percent": 53.0,
                    "peak_utilization_percent": 80
                  }
                },
                {
                  "wall_seconds": 1.2173,
                  "audio_seconds": 4.32,
                  "rtf": 0.2818,
                  "bytes": 414764,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "4.320",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 6,
                    "peak_mib": 9639,
                    "peak_delta_mib": 7576,
                    "mean_utilization_percent": 55.3,
                    "peak_utilization_percent": 80
                  }
                },
                {
                  "wall_seconds": 1.2762,
                  "audio_seconds": 4.32,
                  "rtf": 0.2954,
                  "bytes": 414764,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "4.320",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 6,
                    "peak_mib": 9599,
                    "peak_delta_mib": 7536,
                    "mean_utilization_percent": 54.5,
                    "peak_utilization_percent": 80
                  }
                },
                {
                  "wall_seconds": 1.2672,
                  "audio_seconds": 4.32,
                  "rtf": 0.2933,
                  "bytes": 414764,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "4.320",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 6,
                    "peak_mib": 9711,
                    "peak_delta_mib": 7648,
                    "mean_utilization_percent": 57.3,
                    "peak_utilization_percent": 85
                  }
                },
                {
                  "wall_seconds": 1.3511,
                  "audio_seconds": 4.32,
                  "rtf": 0.3127,
                  "bytes": 414764,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "4.320",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 6,
                    "peak_mib": 9988,
                    "peak_delta_mib": 7925,
                    "mean_utilization_percent": 71.3,
                    "peak_utilization_percent": 83
                  }
                },
                {
                  "wall_seconds": 1.1608,
                  "audio_seconds": 4.16,
                  "rtf": 0.279,
                  "bytes": 399404,
                  "sample_rate": 48000,
                  "channels": 1,
                  "response_duration": "4.160",
                  "response_model": "openbmb/VoxCPM2",
                  "gpu": {
                    "samples": 6,
                    "peak_mib": 9946,
                    "peak_delta_mib": 7883,
                    "mean_utilization_percent": 53.3,
                    "peak_utilization_percent": 81
                  }
                }
              ],
              "summary": {
                "calls": 10,
                "wall_median_seconds": 1.2516,
                "wall_p95_seconds": 1.3738,
                "wall_mean_seconds": 1.2583,
                "rtf_median": 0.2807,
                "rtf_p95": 0.3049,
                "rtf_mean": 0.2841,
                "audio_mean_seconds": 4.432,
                "gpu_peak_mib": 10009,
                "gpu_peak_delta_mib": 7946,
                "gpu_mean_utilization_percent": 53.2
              },
              "container_memory_after": {
                "current_mib": 3067.1,
                "peak_mib": 3069.7
              }
            }
          ]
        }
      ]
    }
  ],
  "latest": {
    "schema": 1,
    "started_at": "2026-10-07T16:23:32+00:00",
    "finished_at": "2026-10-07T16:33:28+00:00",
    "comment": "Historical native image 68787abb versus current Nano backend; same GPU, checkpoint cache, API payloads, 10 steps, concurrency 1",
    "git": {
      "revision": "d9e163bcccc3bbcaf6927b411e3b7854937bb8a4",
      "dirty": true,
      "status": [
        " M Taskfile.yml",
        "?? benchmarks/"
      ],
      "diff_sha256": "4d2ec160b07ba1242b5e8a6185e91107707b3ab618086f3210a70bca0f561855"
    },
    "hardware": {
      "index": 0,
      "name": "NVIDIA GeForce RTX 5070 Ti",
      "driver": "610.88",
      "total_mib": 16303,
      "used_mib": 1889,
      "utilization_percent": 1
    },
    "settings": {
      "model_id": "openbmb/VoxCPM2",
      "gpu_index": 0,
      "calls": 10,
      "warmup_calls": 2,
      "sample_interval_seconds": 0.1,
      "inference_timesteps": 10,
      "cfg_value": 2.0,
      "output_format": "wav",
      "concurrency": 1,
      "nano_gpu_memory_utilization": 0.49,
      "nano_enforce_eager": false
    },
    "scenarios": [
      {
        "code": "PS",
        "label": "plain_short_english",
        "description": "Short English generation without voice control or reference audio."
      },
      {
        "code": "DM",
        "label": "designed_medium_english",
        "description": "Medium English generation with a fixed voice-design instruction."
      },
      {
        "code": "MN",
        "label": "designed_norwegian",
        "description": "Norwegian generation with the same fixed voice-design instruction."
      },
      {
        "code": "CR",
        "label": "reference_clone",
        "description": "English zero-shot clone with the same reference audio mounted into both containers."
      }
    ],
    "backends": [
      {
        "backend": "native",
        "image": {
          "tag": "voxcpmtts:benchmark-native",
          "id": "sha256:68787abb3ecca6fa04d579e43f93d0ce30d3e74d89497450896e1d04fe7d1413",
          "created": "2026-10-07T10:06:02.193219714Z",
          "size_bytes": 4832268446,
          "repo_digests": [
            "voxcpmtts@sha256:68787abb3ecca6fa04d579e43f93d0ce30d3e74d89497450896e1d04fe7d1413"
          ],
          "labels": {}
        },
        "runtime_versions": {
          "python": "3.13.16",
          "torch": "2.10.0+cu128",
          "transformers": "5.3.0",
          "huggingface-hub": null,
          "nano-vllm-voxcpm": null,
          "flash-attn": null,
          "triton": "3.6.0"
        },
        "api_ready_seconds": 6.056,
        "status_before_load": {
          "msg": "pong",
          "type": "VoxCPMTTS",
          "version": "0.1-snapshot",
          "build_id": "stable",
          "runtime": "GPU x1 (visible=0) [0:NVIDIA GeForce RTX 5070 Ti]",
          "device": "auto",
          "model_id": "openbmb/VoxCPM2",
          "load_denoiser": false,
          "load_asr": false,
          "asr_model_id": "iic/SenseVoiceSmall",
          "optimize": false,
          "languages": [
            "Arabic",
            "Burmese",
            "Chinese",
            "Danish",
            "Dutch",
            "English",
            "Finnish",
            "French",
            "German",
            "Greek",
            "Hebrew",
            "Hindi",
            "Indonesian",
            "Italian",
            "Japanese",
            "Khmer",
            "Korean",
            "Lao",
            "Malay",
            "Norwegian",
            "Polish",
            "Portuguese",
            "Russian",
            "Spanish",
            "Swahili",
            "Swedish",
            "Tagalog",
            "Thai",
            "Turkish",
            "Vietnamese"
          ],
          "loaded_model_devices": [],
          "output_formats": {
            "wav": {
              "label": "WAV",
              "extension": "wav",
              "media_type": "audio/wav"
            },
            "mp3": {
              "label": "MP3",
              "extension": "mp3",
              "media_type": "audio/mpeg"
            },
            "flac": {
              "label": "FLAC",
              "extension": "flac",
              "media_type": "audio/flac"
            },
            "ogg": {
              "label": "OGG Vorbis",
              "extension": "ogg",
              "media_type": "audio/ogg"
            }
          },
          "stream_formats": {
            "wav": {
              "label": "Full WAV response",
              "extension": "wav",
              "media_type": "audio/wav"
            },
            "mp3": {
              "label": "Full MP3 response",
              "extension": "mp3",
              "media_type": "audio/mpeg"
            }
          }
        },
        "status_after": {
          "msg": "pong",
          "type": "VoxCPMTTS",
          "version": "0.1-snapshot",
          "build_id": "stable",
          "runtime": "GPU x1 (visible=0) [0:NVIDIA GeForce RTX 5070 Ti]",
          "device": "auto",
          "model_id": "openbmb/VoxCPM2",
          "load_denoiser": false,
          "load_asr": false,
          "asr_model_id": "iic/SenseVoiceSmall",
          "optimize": false,
          "languages": [
            "Arabic",
            "Burmese",
            "Chinese",
            "Danish",
            "Dutch",
            "English",
            "Finnish",
            "French",
            "German",
            "Greek",
            "Hebrew",
            "Hindi",
            "Indonesian",
            "Italian",
            "Japanese",
            "Khmer",
            "Korean",
            "Lao",
            "Malay",
            "Norwegian",
            "Polish",
            "Portuguese",
            "Russian",
            "Spanish",
            "Swahili",
            "Swedish",
            "Tagalog",
            "Thai",
            "Turkish",
            "Vietnamese"
          ],
          "loaded_model_devices": [
            "cuda:0"
          ],
          "output_formats": {
            "wav": {
              "label": "WAV",
              "extension": "wav",
              "media_type": "audio/wav"
            },
            "mp3": {
              "label": "MP3",
              "extension": "mp3",
              "media_type": "audio/mpeg"
            },
            "flac": {
              "label": "FLAC",
              "extension": "flac",
              "media_type": "audio/flac"
            },
            "ogg": {
              "label": "OGG Vorbis",
              "extension": "ogg",
              "media_type": "audio/ogg"
            }
          },
          "stream_formats": {
            "wav": {
              "label": "Full WAV response",
              "extension": "wav",
              "media_type": "audio/wav"
            },
            "mp3": {
              "label": "Full MP3 response",
              "extension": "mp3",
              "media_type": "audio/mpeg"
            }
          }
        },
        "gpu_baseline": {
          "index": 0,
          "name": "NVIDIA GeForce RTX 5070 Ti",
          "driver": "610.88",
          "total_mib": 16303,
          "used_mib": 1890,
          "utilization_percent": 2
        },
        "cold_request": {
          "wall_seconds": 19.3369,
          "audio_seconds": 3.84,
          "rtf": 5.0356,
          "bytes": 368684,
          "sample_rate": 48000,
          "channels": 1,
          "response_duration": "3.840",
          "response_model": "openbmb/VoxCPM2",
          "gpu": {
            "samples": 80,
            "peak_mib": 7661,
            "peak_delta_mib": 5771,
            "mean_utilization_percent": 11.6,
            "peak_utilization_percent": 46
          }
        },
        "gpu_final": {
          "index": 0,
          "name": "NVIDIA GeForce RTX 5070 Ti",
          "driver": "610.88",
          "total_mib": 16303,
          "used_mib": 8349,
          "utilization_percent": 1
        },
        "gpu_final_delta_mib": 6459,
        "container_memory_before_load": {
          "current_mib": 475.6,
          "peak_mib": 478.2
        },
        "container_memory_after_load": {
          "current_mib": 3506.5,
          "peak_mib": 10866.8
        },
        "container_memory_final": {
          "current_mib": 3590.1,
          "peak_mib": 10866.8
        },
        "scenarios": [
          {
            "code": "PS",
            "label": "plain_short_english",
            "description": "Short English generation without voice control or reference audio.",
            "payload": {
              "language": "English",
              "device": "cuda:0",
              "output_format": "wav",
              "cfg_value": 2.0,
              "inference_timesteps": 10,
              "normalize": false,
              "denoise": false,
              "text": "The morning coffee negotiated hard, but eventually agreed to help."
            },
            "warmups": [
              {
                "wall_seconds": 4.6025,
                "audio_seconds": 3.68,
                "rtf": 1.2507,
                "bytes": 353324,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "3.680",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 18,
                  "peak_mib": 7854,
                  "peak_delta_mib": 5964,
                  "mean_utilization_percent": 22.2,
                  "peak_utilization_percent": 37
                }
              },
              {
                "wall_seconds": 4.539,
                "audio_seconds": 4.0,
                "rtf": 1.1348,
                "bytes": 384044,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "4.000",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 19,
                  "peak_mib": 7863,
                  "peak_delta_mib": 5973,
                  "mean_utilization_percent": 26.4,
                  "peak_utilization_percent": 41
                }
              }
            ],
            "measurements": [
              {
                "wall_seconds": 4.167,
                "audio_seconds": 3.52,
                "rtf": 1.1838,
                "bytes": 337964,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "3.520",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 17,
                  "peak_mib": 7863,
                  "peak_delta_mib": 5973,
                  "mean_utilization_percent": 37.3,
                  "peak_utilization_percent": 47
                }
              },
              {
                "wall_seconds": 4.0584,
                "audio_seconds": 3.52,
                "rtf": 1.1529,
                "bytes": 337964,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "3.520",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 17,
                  "peak_mib": 7855,
                  "peak_delta_mib": 5965,
                  "mean_utilization_percent": 26.8,
                  "peak_utilization_percent": 44
                }
              },
              {
                "wall_seconds": 4.6097,
                "audio_seconds": 4.0,
                "rtf": 1.1524,
                "bytes": 384044,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "4.000",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 19,
                  "peak_mib": 7862,
                  "peak_delta_mib": 5972,
                  "mean_utilization_percent": 34.0,
                  "peak_utilization_percent": 47
                }
              },
              {
                "wall_seconds": 3.9071,
                "audio_seconds": 3.36,
                "rtf": 1.1628,
                "bytes": 322604,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "3.360",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 16,
                  "peak_mib": 7855,
                  "peak_delta_mib": 5965,
                  "mean_utilization_percent": 29.1,
                  "peak_utilization_percent": 41
                }
              },
              {
                "wall_seconds": 4.4934,
                "audio_seconds": 4.0,
                "rtf": 1.1234,
                "bytes": 384044,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "4.000",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 18,
                  "peak_mib": 7861,
                  "peak_delta_mib": 5971,
                  "mean_utilization_percent": 34.0,
                  "peak_utilization_percent": 45
                }
              },
              {
                "wall_seconds": 3.6599,
                "audio_seconds": 3.36,
                "rtf": 1.0893,
                "bytes": 322604,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "3.360",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 15,
                  "peak_mib": 7862,
                  "peak_delta_mib": 5972,
                  "mean_utilization_percent": 28.2,
                  "peak_utilization_percent": 47
                }
              },
              {
                "wall_seconds": 4.5232,
                "audio_seconds": 4.0,
                "rtf": 1.1308,
                "bytes": 384044,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "4.000",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 18,
                  "peak_mib": 7855,
                  "peak_delta_mib": 5965,
                  "mean_utilization_percent": 36.3,
                  "peak_utilization_percent": 52
                }
              },
              {
                "wall_seconds": 5.7037,
                "audio_seconds": 4.8,
                "rtf": 1.1883,
                "bytes": 460844,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "4.800",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 23,
                  "peak_mib": 7862,
                  "peak_delta_mib": 5972,
                  "mean_utilization_percent": 27.4,
                  "peak_utilization_percent": 43
                }
              },
              {
                "wall_seconds": 4.2667,
                "audio_seconds": 3.52,
                "rtf": 1.2121,
                "bytes": 337964,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "3.520",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 17,
                  "peak_mib": 7856,
                  "peak_delta_mib": 5966,
                  "mean_utilization_percent": 25.7,
                  "peak_utilization_percent": 37
                }
              },
              {
                "wall_seconds": 4.981,
                "audio_seconds": 4.32,
                "rtf": 1.153,
                "bytes": 414764,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "4.320",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 19,
                  "peak_mib": 7863,
                  "peak_delta_mib": 5973,
                  "mean_utilization_percent": 31.8,
                  "peak_utilization_percent": 48
                }
              }
            ],
            "summary": {
              "calls": 10,
              "wall_median_seconds": 4.3801,
              "wall_p95_seconds": 5.3785,
              "wall_mean_seconds": 4.437,
              "rtf_median": 1.153,
              "rtf_p95": 1.2014,
              "rtf_mean": 1.1549,
              "audio_mean_seconds": 3.84,
              "gpu_peak_mib": 7863,
              "gpu_peak_delta_mib": 5973,
              "gpu_mean_utilization_percent": 31.1
            },
            "container_memory_after": {
              "current_mib": 3506.3,
              "peak_mib": 10866.8
            }
          },
          {
            "code": "DM",
            "label": "designed_medium_english",
            "description": "Medium English generation with a fixed voice-design instruction.",
            "payload": {
              "language": "English",
              "device": "cuda:0",
              "output_format": "wav",
              "cfg_value": 2.0,
              "inference_timesteps": 10,
              "normalize": false,
              "denoise": false,
              "text": "Local speech generation keeps private text on your own machine while producing clear, natural audio for applications, accessibility tools, and creative work.",
              "control": "warm adult female narrator, clear and friendly"
            },
            "warmups": [
              {
                "wall_seconds": 10.1395,
                "audio_seconds": 8.8,
                "rtf": 1.1522,
                "bytes": 844844,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "8.800",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 38,
                  "peak_mib": 7864,
                  "peak_delta_mib": 5974,
                  "mean_utilization_percent": 32.7,
                  "peak_utilization_percent": 46
                }
              },
              {
                "wall_seconds": 9.8613,
                "audio_seconds": 8.64,
                "rtf": 1.1414,
                "bytes": 829484,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "8.640",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 37,
                  "peak_mib": 7864,
                  "peak_delta_mib": 5974,
                  "mean_utilization_percent": 32.5,
                  "peak_utilization_percent": 45
                }
              }
            ],
            "measurements": [
              {
                "wall_seconds": 11.2412,
                "audio_seconds": 9.44,
                "rtf": 1.1908,
                "bytes": 906284,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "9.440",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 43,
                  "peak_mib": 7866,
                  "peak_delta_mib": 5976,
                  "mean_utilization_percent": 28.8,
                  "peak_utilization_percent": 45
                }
              },
              {
                "wall_seconds": 11.8399,
                "audio_seconds": 9.92,
                "rtf": 1.1935,
                "bytes": 952364,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "9.920",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 46,
                  "peak_mib": 7865,
                  "peak_delta_mib": 5975,
                  "mean_utilization_percent": 31.7,
                  "peak_utilization_percent": 56
                }
              },
              {
                "wall_seconds": 9.8484,
                "audio_seconds": 8.48,
                "rtf": 1.1614,
                "bytes": 814124,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "8.480",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 37,
                  "peak_mib": 8002,
                  "peak_delta_mib": 6112,
                  "mean_utilization_percent": 29.5,
                  "peak_utilization_percent": 45
                }
              },
              {
                "wall_seconds": 11.6134,
                "audio_seconds": 10.4,
                "rtf": 1.1167,
                "bytes": 998444,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "10.400",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 44,
                  "peak_mib": 8061,
                  "peak_delta_mib": 6171,
                  "mean_utilization_percent": 32.3,
                  "peak_utilization_percent": 47
                }
              },
              {
                "wall_seconds": 11.4614,
                "audio_seconds": 9.28,
                "rtf": 1.2351,
                "bytes": 890924,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "9.280",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 43,
                  "peak_mib": 8170,
                  "peak_delta_mib": 6280,
                  "mean_utilization_percent": 29.9,
                  "peak_utilization_percent": 48
                }
              },
              {
                "wall_seconds": 10.2654,
                "audio_seconds": 9.12,
                "rtf": 1.1256,
                "bytes": 875564,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "9.120",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 40,
                  "peak_mib": 8095,
                  "peak_delta_mib": 6205,
                  "mean_utilization_percent": 31.5,
                  "peak_utilization_percent": 47
                }
              },
              {
                "wall_seconds": 10.9945,
                "audio_seconds": 8.64,
                "rtf": 1.2725,
                "bytes": 829484,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "8.640",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 43,
                  "peak_mib": 8093,
                  "peak_delta_mib": 6203,
                  "mean_utilization_percent": 29.0,
                  "peak_utilization_percent": 47
                }
              },
              {
                "wall_seconds": 10.1818,
                "audio_seconds": 8.64,
                "rtf": 1.1784,
                "bytes": 829484,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "8.640",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 38,
                  "peak_mib": 8086,
                  "peak_delta_mib": 6196,
                  "mean_utilization_percent": 30.4,
                  "peak_utilization_percent": 44
                }
              },
              {
                "wall_seconds": 12.4848,
                "audio_seconds": 9.76,
                "rtf": 1.2792,
                "bytes": 937004,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "9.760",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 44,
                  "peak_mib": 8370,
                  "peak_delta_mib": 6480,
                  "mean_utilization_percent": 26.7,
                  "peak_utilization_percent": 43
                }
              },
              {
                "wall_seconds": 10.7598,
                "audio_seconds": 8.64,
                "rtf": 1.2453,
                "bytes": 829484,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "8.640",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 39,
                  "peak_mib": 8410,
                  "peak_delta_mib": 6520,
                  "mean_utilization_percent": 28.8,
                  "peak_utilization_percent": 42
                }
              }
            ],
            "summary": {
              "calls": 10,
              "wall_median_seconds": 11.1179,
              "wall_p95_seconds": 12.1946,
              "wall_mean_seconds": 11.0691,
              "rtf_median": 1.1922,
              "rtf_p95": 1.2762,
              "rtf_mean": 1.1999,
              "audio_mean_seconds": 9.232,
              "gpu_peak_mib": 8410,
              "gpu_peak_delta_mib": 6520,
              "gpu_mean_utilization_percent": 29.9
            },
            "container_memory_after": {
              "current_mib": 3506.9,
              "peak_mib": 10866.8
            }
          },
          {
            "code": "MN",
            "label": "designed_norwegian",
            "description": "Norwegian generation with the same fixed voice-design instruction.",
            "payload": {
              "language": "Norwegian",
              "device": "cuda:0",
              "output_format": "wav",
              "cfg_value": 2.0,
              "inference_timesteps": 10,
              "normalize": false,
              "denoise": false,
              "text": "Denne testen måler hvor raskt modellen lager tydelig norsk tale med de samme innstillingene.",
              "control": "warm adult female narrator, clear and friendly"
            },
            "warmups": [
              {
                "wall_seconds": 7.3226,
                "audio_seconds": 5.44,
                "rtf": 1.3461,
                "bytes": 522284,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "5.440",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 26,
                  "peak_mib": 8274,
                  "peak_delta_mib": 6384,
                  "mean_utilization_percent": 27.6,
                  "peak_utilization_percent": 45
                }
              },
              {
                "wall_seconds": 7.3394,
                "audio_seconds": 5.6,
                "rtf": 1.3106,
                "bytes": 537644,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "5.600",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 29,
                  "peak_mib": 8149,
                  "peak_delta_mib": 6259,
                  "mean_utilization_percent": 28.7,
                  "peak_utilization_percent": 44
                }
              }
            ],
            "measurements": [
              {
                "wall_seconds": 5.6976,
                "audio_seconds": 5.12,
                "rtf": 1.1128,
                "bytes": 491564,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "5.120",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 21,
                  "peak_mib": 8149,
                  "peak_delta_mib": 6259,
                  "mean_utilization_percent": 29.2,
                  "peak_utilization_percent": 45
                }
              },
              {
                "wall_seconds": 4.3653,
                "audio_seconds": 4.0,
                "rtf": 1.0913,
                "bytes": 384044,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "4.000",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 18,
                  "peak_mib": 8368,
                  "peak_delta_mib": 6478,
                  "mean_utilization_percent": 32.3,
                  "peak_utilization_percent": 46
                }
              },
              {
                "wall_seconds": 7.2195,
                "audio_seconds": 6.08,
                "rtf": 1.1874,
                "bytes": 583724,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "6.080",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 28,
                  "peak_mib": 8417,
                  "peak_delta_mib": 6527,
                  "mean_utilization_percent": 31.6,
                  "peak_utilization_percent": 46
                }
              },
              {
                "wall_seconds": 7.2642,
                "audio_seconds": 6.56,
                "rtf": 1.1073,
                "bytes": 629804,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "6.560",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 28,
                  "peak_mib": 8420,
                  "peak_delta_mib": 6530,
                  "mean_utilization_percent": 29.2,
                  "peak_utilization_percent": 48
                }
              },
              {
                "wall_seconds": 5.7981,
                "audio_seconds": 4.96,
                "rtf": 1.169,
                "bytes": 476204,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "4.960",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 23,
                  "peak_mib": 8362,
                  "peak_delta_mib": 6472,
                  "mean_utilization_percent": 29.3,
                  "peak_utilization_percent": 45
                }
              },
              {
                "wall_seconds": 5.7691,
                "audio_seconds": 4.64,
                "rtf": 1.2433,
                "bytes": 445484,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "4.640",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 23,
                  "peak_mib": 8168,
                  "peak_delta_mib": 6278,
                  "mean_utilization_percent": 28.5,
                  "peak_utilization_percent": 46
                }
              },
              {
                "wall_seconds": 7.3345,
                "audio_seconds": 6.24,
                "rtf": 1.1754,
                "bytes": 599084,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "6.240",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 29,
                  "peak_mib": 8168,
                  "peak_delta_mib": 6278,
                  "mean_utilization_percent": 32.7,
                  "peak_utilization_percent": 46
                }
              },
              {
                "wall_seconds": 7.7826,
                "audio_seconds": 6.24,
                "rtf": 1.2472,
                "bytes": 599084,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "6.240",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 30,
                  "peak_mib": 8168,
                  "peak_delta_mib": 6278,
                  "mean_utilization_percent": 26.7,
                  "peak_utilization_percent": 46
                }
              },
              {
                "wall_seconds": 5.3838,
                "audio_seconds": 4.64,
                "rtf": 1.1603,
                "bytes": 445484,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "4.640",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 22,
                  "peak_mib": 8160,
                  "peak_delta_mib": 6270,
                  "mean_utilization_percent": 29.6,
                  "peak_utilization_percent": 44
                }
              },
              {
                "wall_seconds": 5.549,
                "audio_seconds": 5.12,
                "rtf": 1.0838,
                "bytes": 491564,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "5.120",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 22,
                  "peak_mib": 8309,
                  "peak_delta_mib": 6419,
                  "mean_utilization_percent": 27.4,
                  "peak_utilization_percent": 45
                }
              }
            ],
            "summary": {
              "calls": 10,
              "wall_median_seconds": 5.7836,
              "wall_p95_seconds": 7.581,
              "wall_mean_seconds": 6.2164,
              "rtf_median": 1.1646,
              "rtf_p95": 1.2454,
              "rtf_mean": 1.1578,
              "audio_mean_seconds": 5.36,
              "gpu_peak_mib": 8420,
              "gpu_peak_delta_mib": 6530,
              "gpu_mean_utilization_percent": 29.6
            },
            "container_memory_after": {
              "current_mib": 3508.3,
              "peak_mib": 10866.8
            }
          },
          {
            "code": "CR",
            "label": "reference_clone",
            "description": "English zero-shot clone with the same reference audio mounted into both containers.",
            "payload": {
              "language": "English",
              "device": "cuda:0",
              "output_format": "wav",
              "cfg_value": 2.0,
              "inference_timesteps": 10,
              "normalize": false,
              "denoise": false,
              "text": "This sample checks how quickly the model can carry a reference voice into a new sentence.",
              "ref_audio": "/benchmark/original_clone.mp3"
            },
            "warmups": [
              {
                "wall_seconds": 17.6292,
                "audio_seconds": 5.6,
                "rtf": 3.1481,
                "bytes": 537644,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "5.600",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 73,
                  "peak_mib": 8293,
                  "peak_delta_mib": 6403,
                  "mean_utilization_percent": 11.1,
                  "peak_utilization_percent": 44
                }
              },
              {
                "wall_seconds": 5.811,
                "audio_seconds": 4.96,
                "rtf": 1.1716,
                "bytes": 476204,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "4.960",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 23,
                  "peak_mib": 8425,
                  "peak_delta_mib": 6535,
                  "mean_utilization_percent": 31.4,
                  "peak_utilization_percent": 44
                }
              }
            ],
            "measurements": [
              {
                "wall_seconds": 5.6347,
                "audio_seconds": 4.8,
                "rtf": 1.1739,
                "bytes": 460844,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "4.800",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 22,
                  "peak_mib": 8347,
                  "peak_delta_mib": 6457,
                  "mean_utilization_percent": 30.9,
                  "peak_utilization_percent": 44
                }
              },
              {
                "wall_seconds": 5.2156,
                "audio_seconds": 4.64,
                "rtf": 1.1241,
                "bytes": 445484,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "4.640",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 21,
                  "peak_mib": 8340,
                  "peak_delta_mib": 6450,
                  "mean_utilization_percent": 30.0,
                  "peak_utilization_percent": 45
                }
              },
              {
                "wall_seconds": 4.4747,
                "audio_seconds": 4.0,
                "rtf": 1.1187,
                "bytes": 384044,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "4.000",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 18,
                  "peak_mib": 8347,
                  "peak_delta_mib": 6457,
                  "mean_utilization_percent": 32.1,
                  "peak_utilization_percent": 47
                }
              },
              {
                "wall_seconds": 4.7618,
                "audio_seconds": 4.32,
                "rtf": 1.1023,
                "bytes": 414764,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "4.320",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 19,
                  "peak_mib": 8349,
                  "peak_delta_mib": 6459,
                  "mean_utilization_percent": 28.3,
                  "peak_utilization_percent": 45
                }
              },
              {
                "wall_seconds": 6.0473,
                "audio_seconds": 4.96,
                "rtf": 1.2192,
                "bytes": 476204,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "4.960",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 23,
                  "peak_mib": 8348,
                  "peak_delta_mib": 6458,
                  "mean_utilization_percent": 26.4,
                  "peak_utilization_percent": 44
                }
              },
              {
                "wall_seconds": 5.5075,
                "audio_seconds": 4.64,
                "rtf": 1.187,
                "bytes": 445484,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "4.640",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 22,
                  "peak_mib": 8354,
                  "peak_delta_mib": 6464,
                  "mean_utilization_percent": 32.0,
                  "peak_utilization_percent": 45
                }
              },
              {
                "wall_seconds": 5.3604,
                "audio_seconds": 4.48,
                "rtf": 1.1965,
                "bytes": 430124,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "4.480",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 22,
                  "peak_mib": 8362,
                  "peak_delta_mib": 6472,
                  "mean_utilization_percent": 31.4,
                  "peak_utilization_percent": 45
                }
              },
              {
                "wall_seconds": 5.2353,
                "audio_seconds": 4.48,
                "rtf": 1.1686,
                "bytes": 430124,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "4.480",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 20,
                  "peak_mib": 8355,
                  "peak_delta_mib": 6465,
                  "mean_utilization_percent": 32.6,
                  "peak_utilization_percent": 45
                }
              },
              {
                "wall_seconds": 5.7813,
                "audio_seconds": 4.8,
                "rtf": 1.2044,
                "bytes": 460844,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "4.800",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 22,
                  "peak_mib": 8363,
                  "peak_delta_mib": 6473,
                  "mean_utilization_percent": 30.5,
                  "peak_utilization_percent": 44
                }
              },
              {
                "wall_seconds": 5.4564,
                "audio_seconds": 4.8,
                "rtf": 1.1367,
                "bytes": 460844,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "4.800",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 22,
                  "peak_mib": 8349,
                  "peak_delta_mib": 6459,
                  "mean_utilization_percent": 29.6,
                  "peak_utilization_percent": 45
                }
              }
            ],
            "summary": {
              "calls": 10,
              "wall_median_seconds": 5.4084,
              "wall_p95_seconds": 5.9276,
              "wall_mean_seconds": 5.3475,
              "rtf_median": 1.1713,
              "rtf_p95": 1.2125,
              "rtf_mean": 1.1631,
              "audio_mean_seconds": 4.592,
              "gpu_peak_mib": 8363,
              "gpu_peak_delta_mib": 6473,
              "gpu_mean_utilization_percent": 30.4
            },
            "container_memory_after": {
              "current_mib": 3589.2,
              "peak_mib": 10866.8
            }
          }
        ]
      },
      {
        "backend": "nano",
        "image": {
          "tag": "voxcpmtts:benchmark-nano",
          "id": "sha256:3a4e1ccda4af76e389521374637905c6462974480dac20c476da9414255923f1",
          "created": "2026-10-07T16:21:51.274308704Z",
          "size_bytes": 4847177171,
          "repo_digests": [
            "voxcpmtts@sha256:3a4e1ccda4af76e389521374637905c6462974480dac20c476da9414255923f1"
          ],
          "labels": {
            "org.opencontainers.image.created": "2026-10-07T14:52:34Z",
            "org.opencontainers.image.revision": "ac1d0b3b2bcb",
            "org.opencontainers.image.source": "https://github.com/hangry-labs/VoxCPMTTS"
          }
        },
        "runtime_versions": {
          "python": "3.13.16",
          "torch": "2.8.0+cu128",
          "transformers": "5.3.0",
          "huggingface-hub": null,
          "nano-vllm-voxcpm": "2.0.4",
          "flash-attn": null,
          "triton": "3.4.0"
        },
        "api_ready_seconds": 4.037,
        "status_before_load": {
          "msg": "pong",
          "type": "VoxCPMTTS",
          "version": "0.1-snapshot",
          "build_id": "2026-10-07T14:52:34Z@ac1d0b3b2bcb",
          "build_date": "2026-10-07T14:52:34Z",
          "revision": "ac1d0b3b2bcb",
          "runtime": "GPU x1 (visible=0) [0:NVIDIA GeForce RTX 5070 Ti]",
          "device": "auto",
          "backend": "nano",
          "model_id": "openbmb/VoxCPM2",
          "load_denoiser": false,
          "load_asr": false,
          "asr_model_id": "iic/SenseVoiceSmall",
          "optimize": false,
          "languages": [
            "Arabic",
            "Burmese",
            "Chinese",
            "Danish",
            "Dutch",
            "English",
            "Finnish",
            "French",
            "German",
            "Greek",
            "Hebrew",
            "Hindi",
            "Indonesian",
            "Italian",
            "Japanese",
            "Khmer",
            "Korean",
            "Lao",
            "Malay",
            "Norwegian",
            "Polish",
            "Portuguese",
            "Russian",
            "Spanish",
            "Swahili",
            "Swedish",
            "Tagalog",
            "Thai",
            "Turkish",
            "Vietnamese"
          ],
          "loaded_model_devices": [],
          "output_formats": {
            "wav": {
              "label": "WAV",
              "extension": "wav",
              "media_type": "audio/wav"
            },
            "mp3": {
              "label": "MP3",
              "extension": "mp3",
              "media_type": "audio/mpeg"
            },
            "flac": {
              "label": "FLAC",
              "extension": "flac",
              "media_type": "audio/flac"
            },
            "ogg": {
              "label": "OGG Vorbis",
              "extension": "ogg",
              "media_type": "audio/ogg"
            }
          },
          "stream_formats": {
            "wav": {
              "label": "Full WAV response",
              "extension": "wav",
              "media_type": "audio/wav"
            },
            "mp3": {
              "label": "Full MP3 response",
              "extension": "mp3",
              "media_type": "audio/mpeg"
            }
          }
        },
        "status_after": {
          "msg": "pong",
          "type": "VoxCPMTTS",
          "version": "0.1-snapshot",
          "build_id": "2026-10-07T14:52:34Z@ac1d0b3b2bcb",
          "build_date": "2026-10-07T14:52:34Z",
          "revision": "ac1d0b3b2bcb",
          "runtime": "GPU x1 (visible=0) [0:NVIDIA GeForce RTX 5070 Ti]",
          "device": "auto",
          "backend": "nano",
          "model_id": "openbmb/VoxCPM2",
          "load_denoiser": false,
          "load_asr": false,
          "asr_model_id": "iic/SenseVoiceSmall",
          "optimize": false,
          "languages": [
            "Arabic",
            "Burmese",
            "Chinese",
            "Danish",
            "Dutch",
            "English",
            "Finnish",
            "French",
            "German",
            "Greek",
            "Hebrew",
            "Hindi",
            "Indonesian",
            "Italian",
            "Japanese",
            "Khmer",
            "Korean",
            "Lao",
            "Malay",
            "Norwegian",
            "Polish",
            "Portuguese",
            "Russian",
            "Spanish",
            "Swahili",
            "Swedish",
            "Tagalog",
            "Thai",
            "Turkish",
            "Vietnamese"
          ],
          "loaded_model_devices": [
            "cuda:0"
          ],
          "output_formats": {
            "wav": {
              "label": "WAV",
              "extension": "wav",
              "media_type": "audio/wav"
            },
            "mp3": {
              "label": "MP3",
              "extension": "mp3",
              "media_type": "audio/mpeg"
            },
            "flac": {
              "label": "FLAC",
              "extension": "flac",
              "media_type": "audio/flac"
            },
            "ogg": {
              "label": "OGG Vorbis",
              "extension": "ogg",
              "media_type": "audio/ogg"
            }
          },
          "stream_formats": {
            "wav": {
              "label": "Full WAV response",
              "extension": "wav",
              "media_type": "audio/wav"
            },
            "mp3": {
              "label": "Full MP3 response",
              "extension": "mp3",
              "media_type": "audio/mpeg"
            }
          }
        },
        "gpu_baseline": {
          "index": 0,
          "name": "NVIDIA GeForce RTX 5070 Ti",
          "driver": "610.88",
          "total_mib": 16303,
          "used_mib": 2063,
          "utilization_percent": 8
        },
        "cold_request": {
          "wall_seconds": 22.4175,
          "audio_seconds": 4.16,
          "rtf": 5.3888,
          "bytes": 399404,
          "sample_rate": 48000,
          "channels": 1,
          "response_duration": "4.160",
          "response_model": "openbmb/VoxCPM2",
          "gpu": {
            "samples": 95,
            "peak_mib": 10414,
            "peak_delta_mib": 8351,
            "mean_utilization_percent": 9.2,
            "peak_utilization_percent": 100
          }
        },
        "gpu_final": {
          "index": 0,
          "name": "NVIDIA GeForce RTX 5070 Ti",
          "driver": "610.88",
          "total_mib": 16303,
          "used_mib": 9564,
          "utilization_percent": 2
        },
        "gpu_final_delta_mib": 7501,
        "container_memory_before_load": {
          "current_mib": 369.8,
          "peak_mib": 372.7
        },
        "container_memory_after_load": {
          "current_mib": 2672.4,
          "peak_mib": 2701.8
        },
        "container_memory_final": {
          "current_mib": 3067.5,
          "peak_mib": 3324.3
        },
        "scenarios": [
          {
            "code": "PS",
            "label": "plain_short_english",
            "description": "Short English generation without voice control or reference audio.",
            "payload": {
              "language": "English",
              "device": "cuda:0",
              "output_format": "wav",
              "cfg_value": 2.0,
              "inference_timesteps": 10,
              "normalize": false,
              "denoise": false,
              "text": "The morning coffee negotiated hard, but eventually agreed to help."
            },
            "warmups": [
              {
                "wall_seconds": 1.0834,
                "audio_seconds": 3.84,
                "rtf": 0.2821,
                "bytes": 368684,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "3.840",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 6,
                  "peak_mib": 9502,
                  "peak_delta_mib": 7439,
                  "mean_utilization_percent": 25.7,
                  "peak_utilization_percent": 77
                }
              },
              {
                "wall_seconds": 1.0014,
                "audio_seconds": 4.0,
                "rtf": 0.2504,
                "bytes": 384044,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "4.000",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 6,
                  "peak_mib": 9502,
                  "peak_delta_mib": 7439,
                  "mean_utilization_percent": 65.3,
                  "peak_utilization_percent": 82
                }
              }
            ],
            "measurements": [
              {
                "wall_seconds": 1.1387,
                "audio_seconds": 4.0,
                "rtf": 0.2847,
                "bytes": 384044,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "4.000",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 6,
                  "peak_mib": 9494,
                  "peak_delta_mib": 7431,
                  "mean_utilization_percent": 33.7,
                  "peak_utilization_percent": 81
                }
              },
              {
                "wall_seconds": 1.2282,
                "audio_seconds": 4.8,
                "rtf": 0.2559,
                "bytes": 460844,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "4.800",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 6,
                  "peak_mib": 9494,
                  "peak_delta_mib": 7431,
                  "mean_utilization_percent": 32.5,
                  "peak_utilization_percent": 85
                }
              },
              {
                "wall_seconds": 1.0627,
                "audio_seconds": 3.84,
                "rtf": 0.2767,
                "bytes": 368684,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "3.840",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 6,
                  "peak_mib": 9494,
                  "peak_delta_mib": 7431,
                  "mean_utilization_percent": 63.8,
                  "peak_utilization_percent": 81
                }
              },
              {
                "wall_seconds": 1.344,
                "audio_seconds": 4.8,
                "rtf": 0.28,
                "bytes": 460844,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "4.800",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 7,
                  "peak_mib": 9494,
                  "peak_delta_mib": 7431,
                  "mean_utilization_percent": 40.0,
                  "peak_utilization_percent": 78
                }
              },
              {
                "wall_seconds": 1.1481,
                "audio_seconds": 4.16,
                "rtf": 0.276,
                "bytes": 399404,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "4.160",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 6,
                  "peak_mib": 9494,
                  "peak_delta_mib": 7431,
                  "mean_utilization_percent": 49.7,
                  "peak_utilization_percent": 84
                }
              },
              {
                "wall_seconds": 1.3264,
                "audio_seconds": 4.64,
                "rtf": 0.2859,
                "bytes": 445484,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "4.640",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 6,
                  "peak_mib": 9494,
                  "peak_delta_mib": 7431,
                  "mean_utilization_percent": 50.0,
                  "peak_utilization_percent": 84
                }
              },
              {
                "wall_seconds": 0.944,
                "audio_seconds": 3.68,
                "rtf": 0.2565,
                "bytes": 353324,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "3.680",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 5,
                  "peak_mib": 9502,
                  "peak_delta_mib": 7439,
                  "mean_utilization_percent": 30.0,
                  "peak_utilization_percent": 75
                }
              },
              {
                "wall_seconds": 1.0478,
                "audio_seconds": 3.52,
                "rtf": 0.2977,
                "bytes": 337964,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "3.520",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 6,
                  "peak_mib": 9502,
                  "peak_delta_mib": 7439,
                  "mean_utilization_percent": 28.0,
                  "peak_utilization_percent": 84
                }
              },
              {
                "wall_seconds": 1.0808,
                "audio_seconds": 4.16,
                "rtf": 0.2598,
                "bytes": 399404,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "4.160",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 6,
                  "peak_mib": 9502,
                  "peak_delta_mib": 7439,
                  "mean_utilization_percent": 32.8,
                  "peak_utilization_percent": 82
                }
              },
              {
                "wall_seconds": 1.0694,
                "audio_seconds": 4.0,
                "rtf": 0.2674,
                "bytes": 384044,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "4.000",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 6,
                  "peak_mib": 9494,
                  "peak_delta_mib": 7431,
                  "mean_utilization_percent": 40.0,
                  "peak_utilization_percent": 78
                }
              }
            ],
            "summary": {
              "calls": 10,
              "wall_median_seconds": 1.1098,
              "wall_p95_seconds": 1.3361,
              "wall_mean_seconds": 1.139,
              "rtf_median": 0.2763,
              "rtf_p95": 0.2924,
              "rtf_mean": 0.2741,
              "audio_mean_seconds": 4.16,
              "gpu_peak_mib": 9502,
              "gpu_peak_delta_mib": 7439,
              "gpu_mean_utilization_percent": 40.0
            },
            "container_memory_after": {
              "current_mib": 2676.1,
              "peak_mib": 2701.8
            }
          },
          {
            "code": "DM",
            "label": "designed_medium_english",
            "description": "Medium English generation with a fixed voice-design instruction.",
            "payload": {
              "language": "English",
              "device": "cuda:0",
              "output_format": "wav",
              "cfg_value": 2.0,
              "inference_timesteps": 10,
              "normalize": false,
              "denoise": false,
              "text": "Local speech generation keeps private text on your own machine while producing clear, natural audio for applications, accessibility tools, and creative work.",
              "control": "warm adult female narrator, clear and friendly"
            },
            "warmups": [
              {
                "wall_seconds": 2.6789,
                "audio_seconds": 9.92,
                "rtf": 0.27,
                "bytes": 952364,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "9.920",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 12,
                  "peak_mib": 9494,
                  "peak_delta_mib": 7431,
                  "mean_utilization_percent": 48.8,
                  "peak_utilization_percent": 84
                }
              },
              {
                "wall_seconds": 2.8458,
                "audio_seconds": 8.96,
                "rtf": 0.3176,
                "bytes": 860204,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "8.960",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 10,
                  "peak_mib": 9502,
                  "peak_delta_mib": 7439,
                  "mean_utilization_percent": 55.0,
                  "peak_utilization_percent": 86
                }
              }
            ],
            "measurements": [
              {
                "wall_seconds": 2.9523,
                "audio_seconds": 9.28,
                "rtf": 0.3181,
                "bytes": 890924,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "9.280",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 12,
                  "peak_mib": 9502,
                  "peak_delta_mib": 7439,
                  "mean_utilization_percent": 48.5,
                  "peak_utilization_percent": 83
                }
              },
              {
                "wall_seconds": 2.7632,
                "audio_seconds": 9.6,
                "rtf": 0.2878,
                "bytes": 921644,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "9.600",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 12,
                  "peak_mib": 9474,
                  "peak_delta_mib": 7411,
                  "mean_utilization_percent": 51.8,
                  "peak_utilization_percent": 82
                }
              },
              {
                "wall_seconds": 2.5298,
                "audio_seconds": 8.96,
                "rtf": 0.2823,
                "bytes": 860204,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "8.960",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 12,
                  "peak_mib": 9474,
                  "peak_delta_mib": 7411,
                  "mean_utilization_percent": 41.2,
                  "peak_utilization_percent": 81
                }
              },
              {
                "wall_seconds": 2.5078,
                "audio_seconds": 9.28,
                "rtf": 0.2702,
                "bytes": 890924,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "9.280",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 11,
                  "peak_mib": 9474,
                  "peak_delta_mib": 7411,
                  "mean_utilization_percent": 58.7,
                  "peak_utilization_percent": 83
                }
              },
              {
                "wall_seconds": 2.7617,
                "audio_seconds": 9.76,
                "rtf": 0.283,
                "bytes": 937004,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "9.760",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 12,
                  "peak_mib": 9482,
                  "peak_delta_mib": 7419,
                  "mean_utilization_percent": 34.9,
                  "peak_utilization_percent": 78
                }
              },
              {
                "wall_seconds": 3.106,
                "audio_seconds": 10.88,
                "rtf": 0.2855,
                "bytes": 1044524,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "10.880",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 13,
                  "peak_mib": 9475,
                  "peak_delta_mib": 7412,
                  "mean_utilization_percent": 48.0,
                  "peak_utilization_percent": 80
                }
              },
              {
                "wall_seconds": 2.8919,
                "audio_seconds": 10.72,
                "rtf": 0.2698,
                "bytes": 1029164,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "10.720",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 12,
                  "peak_mib": 9702,
                  "peak_delta_mib": 7639,
                  "mean_utilization_percent": 49.9,
                  "peak_utilization_percent": 82
                }
              },
              {
                "wall_seconds": 2.7315,
                "audio_seconds": 9.92,
                "rtf": 0.2753,
                "bytes": 952364,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "9.920",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 12,
                  "peak_mib": 9705,
                  "peak_delta_mib": 7642,
                  "mean_utilization_percent": 37.6,
                  "peak_utilization_percent": 84
                }
              },
              {
                "wall_seconds": 2.4368,
                "audio_seconds": 8.96,
                "rtf": 0.272,
                "bytes": 860204,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "8.960",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 11,
                  "peak_mib": 9705,
                  "peak_delta_mib": 7642,
                  "mean_utilization_percent": 58.3,
                  "peak_utilization_percent": 85
                }
              },
              {
                "wall_seconds": 2.2976,
                "audio_seconds": 9.12,
                "rtf": 0.2519,
                "bytes": 875564,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "9.120",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 11,
                  "peak_mib": 9668,
                  "peak_delta_mib": 7605,
                  "mean_utilization_percent": 52.0,
                  "peak_utilization_percent": 84
                }
              }
            ],
            "summary": {
              "calls": 10,
              "wall_median_seconds": 2.7466,
              "wall_p95_seconds": 3.0368,
              "wall_mean_seconds": 2.6979,
              "rtf_median": 0.2788,
              "rtf_p95": 0.3045,
              "rtf_mean": 0.2796,
              "audio_mean_seconds": 9.648,
              "gpu_peak_mib": 9705,
              "gpu_peak_delta_mib": 7642,
              "gpu_mean_utilization_percent": 48.1
            },
            "container_memory_after": {
              "current_mib": 2692.4,
              "peak_mib": 2701.8
            }
          },
          {
            "code": "MN",
            "label": "designed_norwegian",
            "description": "Norwegian generation with the same fixed voice-design instruction.",
            "payload": {
              "language": "Norwegian",
              "device": "cuda:0",
              "output_format": "wav",
              "cfg_value": 2.0,
              "inference_timesteps": 10,
              "normalize": false,
              "denoise": false,
              "text": "Denne testen måler hvor raskt modellen lager tydelig norsk tale med de samme innstillingene.",
              "control": "warm adult female narrator, clear and friendly"
            },
            "warmups": [
              {
                "wall_seconds": 1.4953,
                "audio_seconds": 5.44,
                "rtf": 0.2749,
                "bytes": 522284,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "5.440",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 7,
                  "peak_mib": 9548,
                  "peak_delta_mib": 7485,
                  "mean_utilization_percent": 44.3,
                  "peak_utilization_percent": 78
                }
              },
              {
                "wall_seconds": 1.4264,
                "audio_seconds": 4.64,
                "rtf": 0.3074,
                "bytes": 445484,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "4.640",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 7,
                  "peak_mib": 9548,
                  "peak_delta_mib": 7485,
                  "mean_utilization_percent": 46.6,
                  "peak_utilization_percent": 81
                }
              }
            ],
            "measurements": [
              {
                "wall_seconds": 1.5716,
                "audio_seconds": 5.92,
                "rtf": 0.2655,
                "bytes": 568364,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "5.920",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 8,
                  "peak_mib": 9515,
                  "peak_delta_mib": 7452,
                  "mean_utilization_percent": 30.6,
                  "peak_utilization_percent": 85
                }
              },
              {
                "wall_seconds": 1.5992,
                "audio_seconds": 5.92,
                "rtf": 0.2701,
                "bytes": 568364,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "5.920",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 8,
                  "peak_mib": 9523,
                  "peak_delta_mib": 7460,
                  "mean_utilization_percent": 51.0,
                  "peak_utilization_percent": 82
                }
              },
              {
                "wall_seconds": 2.0514,
                "audio_seconds": 7.36,
                "rtf": 0.2787,
                "bytes": 706604,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "7.360",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 10,
                  "peak_mib": 9516,
                  "peak_delta_mib": 7453,
                  "mean_utilization_percent": 44.4,
                  "peak_utilization_percent": 77
                }
              },
              {
                "wall_seconds": 2.163,
                "audio_seconds": 5.6,
                "rtf": 0.3862,
                "bytes": 537644,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "5.600",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 8,
                  "peak_mib": 9516,
                  "peak_delta_mib": 7453,
                  "mean_utilization_percent": 28.1,
                  "peak_utilization_percent": 73
                }
              },
              {
                "wall_seconds": 1.4104,
                "audio_seconds": 5.12,
                "rtf": 0.2755,
                "bytes": 491564,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "5.120",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 7,
                  "peak_mib": 9516,
                  "peak_delta_mib": 7453,
                  "mean_utilization_percent": 47.6,
                  "peak_utilization_percent": 84
                }
              },
              {
                "wall_seconds": 1.3455,
                "audio_seconds": 4.64,
                "rtf": 0.29,
                "bytes": 445484,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "4.640",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 7,
                  "peak_mib": 9516,
                  "peak_delta_mib": 7453,
                  "mean_utilization_percent": 44.3,
                  "peak_utilization_percent": 79
                }
              },
              {
                "wall_seconds": 1.2764,
                "audio_seconds": 5.12,
                "rtf": 0.2493,
                "bytes": 491564,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "5.120",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 6,
                  "peak_mib": 9626,
                  "peak_delta_mib": 7563,
                  "mean_utilization_percent": 51.3,
                  "peak_utilization_percent": 79
                }
              },
              {
                "wall_seconds": 1.3508,
                "audio_seconds": 4.48,
                "rtf": 0.3015,
                "bytes": 430124,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "4.480",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 6,
                  "peak_mib": 9954,
                  "peak_delta_mib": 7891,
                  "mean_utilization_percent": 57.0,
                  "peak_utilization_percent": 83
                }
              },
              {
                "wall_seconds": 1.4682,
                "audio_seconds": 5.12,
                "rtf": 0.2868,
                "bytes": 491564,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "5.120",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 7,
                  "peak_mib": 9966,
                  "peak_delta_mib": 7903,
                  "mean_utilization_percent": 44.9,
                  "peak_utilization_percent": 76
                }
              },
              {
                "wall_seconds": 1.5586,
                "audio_seconds": 5.92,
                "rtf": 0.2633,
                "bytes": 568364,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "5.920",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 7,
                  "peak_mib": 9939,
                  "peak_delta_mib": 7876,
                  "mean_utilization_percent": 37.1,
                  "peak_utilization_percent": 84
                }
              }
            ],
            "summary": {
              "calls": 10,
              "wall_median_seconds": 1.5134,
              "wall_p95_seconds": 2.1128,
              "wall_mean_seconds": 1.5795,
              "rtf_median": 0.2771,
              "rtf_p95": 0.3481,
              "rtf_mean": 0.2867,
              "audio_mean_seconds": 5.52,
              "gpu_peak_mib": 9966,
              "gpu_peak_delta_mib": 7903,
              "gpu_mean_utilization_percent": 43.6
            },
            "container_memory_after": {
              "current_mib": 2697.6,
              "peak_mib": 2701.8
            }
          },
          {
            "code": "CR",
            "label": "reference_clone",
            "description": "English zero-shot clone with the same reference audio mounted into both containers.",
            "payload": {
              "language": "English",
              "device": "cuda:0",
              "output_format": "wav",
              "cfg_value": 2.0,
              "inference_timesteps": 10,
              "normalize": false,
              "denoise": false,
              "text": "This sample checks how quickly the model can carry a reference voice into a new sentence.",
              "ref_audio": "/benchmark/original_clone.mp3"
            },
            "warmups": [
              {
                "wall_seconds": 11.6984,
                "audio_seconds": 4.64,
                "rtf": 2.5212,
                "bytes": 445484,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "4.640",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 51,
                  "peak_mib": 9939,
                  "peak_delta_mib": 7876,
                  "mean_utilization_percent": 9.9,
                  "peak_utilization_percent": 83
                }
              },
              {
                "wall_seconds": 1.2175,
                "audio_seconds": 4.32,
                "rtf": 0.2818,
                "bytes": 414764,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "4.320",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 6,
                  "peak_mib": 9625,
                  "peak_delta_mib": 7562,
                  "mean_utilization_percent": 54.0,
                  "peak_utilization_percent": 81
                }
              }
            ],
            "measurements": [
              {
                "wall_seconds": 1.236,
                "audio_seconds": 4.48,
                "rtf": 0.2759,
                "bytes": 430124,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "4.480",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 6,
                  "peak_mib": 9587,
                  "peak_delta_mib": 7524,
                  "mean_utilization_percent": 53.3,
                  "peak_utilization_percent": 83
                }
              },
              {
                "wall_seconds": 1.3923,
                "audio_seconds": 4.96,
                "rtf": 0.2807,
                "bytes": 476204,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "4.960",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 7,
                  "peak_mib": 10009,
                  "peak_delta_mib": 7946,
                  "mean_utilization_percent": 35.0,
                  "peak_utilization_percent": 77
                }
              },
              {
                "wall_seconds": 1.1741,
                "audio_seconds": 4.48,
                "rtf": 0.2621,
                "bytes": 430124,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "4.480",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 6,
                  "peak_mib": 9585,
                  "peak_delta_mib": 7522,
                  "mean_utilization_percent": 45.7,
                  "peak_utilization_percent": 77
                }
              },
              {
                "wall_seconds": 1.303,
                "audio_seconds": 4.64,
                "rtf": 0.2808,
                "bytes": 445484,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "4.640",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 6,
                  "peak_mib": 9639,
                  "peak_delta_mib": 7576,
                  "mean_utilization_percent": 53.3,
                  "peak_utilization_percent": 79
                }
              },
              {
                "wall_seconds": 1.2046,
                "audio_seconds": 4.32,
                "rtf": 0.2788,
                "bytes": 414764,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "4.320",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 6,
                  "peak_mib": 9601,
                  "peak_delta_mib": 7538,
                  "mean_utilization_percent": 53.0,
                  "peak_utilization_percent": 80
                }
              },
              {
                "wall_seconds": 1.2173,
                "audio_seconds": 4.32,
                "rtf": 0.2818,
                "bytes": 414764,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "4.320",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 6,
                  "peak_mib": 9639,
                  "peak_delta_mib": 7576,
                  "mean_utilization_percent": 55.3,
                  "peak_utilization_percent": 80
                }
              },
              {
                "wall_seconds": 1.2762,
                "audio_seconds": 4.32,
                "rtf": 0.2954,
                "bytes": 414764,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "4.320",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 6,
                  "peak_mib": 9599,
                  "peak_delta_mib": 7536,
                  "mean_utilization_percent": 54.5,
                  "peak_utilization_percent": 80
                }
              },
              {
                "wall_seconds": 1.2672,
                "audio_seconds": 4.32,
                "rtf": 0.2933,
                "bytes": 414764,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "4.320",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 6,
                  "peak_mib": 9711,
                  "peak_delta_mib": 7648,
                  "mean_utilization_percent": 57.3,
                  "peak_utilization_percent": 85
                }
              },
              {
                "wall_seconds": 1.3511,
                "audio_seconds": 4.32,
                "rtf": 0.3127,
                "bytes": 414764,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "4.320",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 6,
                  "peak_mib": 9988,
                  "peak_delta_mib": 7925,
                  "mean_utilization_percent": 71.3,
                  "peak_utilization_percent": 83
                }
              },
              {
                "wall_seconds": 1.1608,
                "audio_seconds": 4.16,
                "rtf": 0.279,
                "bytes": 399404,
                "sample_rate": 48000,
                "channels": 1,
                "response_duration": "4.160",
                "response_model": "openbmb/VoxCPM2",
                "gpu": {
                  "samples": 6,
                  "peak_mib": 9946,
                  "peak_delta_mib": 7883,
                  "mean_utilization_percent": 53.3,
                  "peak_utilization_percent": 81
                }
              }
            ],
            "summary": {
              "calls": 10,
              "wall_median_seconds": 1.2516,
              "wall_p95_seconds": 1.3738,
              "wall_mean_seconds": 1.2583,
              "rtf_median": 0.2807,
              "rtf_p95": 0.3049,
              "rtf_mean": 0.2841,
              "audio_mean_seconds": 4.432,
              "gpu_peak_mib": 10009,
              "gpu_peak_delta_mib": 7946,
              "gpu_mean_utilization_percent": 53.2
            },
            "container_memory_after": {
              "current_mib": 3067.1,
              "peak_mib": 3069.7
            }
          }
        ]
      }
    ]
  }
}