A model is still running

A model is still running or downloading. Leaving will stop the current operation. Stay on this page to let it finish.

Skip to content
LocalAI.toolsBeta

Join the waitlist

Get an email when the new apps and model runs launch.

No account needed.

Nemotron Diarization 3

Estimate anonymous speaker turns from complete WAV recordings with the native eight-channel Nemotron3 diarization model.

Desktop inferencePrivate · Runs on your computer
About this model

Estimate anonymous speaker turns from complete WAV recordings with the native eight-channel Nemotron3 diarization model.

Created by
NVIDIA
Model released
Sep 23, 2026
Source repository created
Sep 1, 2026
Model weights · FP32
397.0 MB

Best for

  • Local anonymous speaker timelines
  • Speaker attribution experiments on authorized recordings

Keep in mind

  • Labels are anonymous within-recording channels, not identities. No transcript is produced.
  • Synthetic Kokoro/Qwen-cloned voice challenge merged two voices into one, DER46.3%; retained failures limit similar-voice reliability.
  • Validated mixed-source two-speaker/noisy/stereo/single/silence fixtures are not general natural-meeting, multilingual, overlap or eight-speaker accuracy certification.
  • Complete-recording offline mode only. Upstream streaming architecture does not grant live transport support.
  • WAV mono/stereo is downmixed and resampled locally; malformed/truncated input is rejected.
  • FP32 CPU/Metal only; FP16/BF16 single-fixture probes are not advertised. Other hardware/OS/Windows unverified.
  • OpenMDW1.1 license has no matching SPDX identifier in the checked list; SPDX LicenseRef retains exact version and official URL.
  • Accuracy limitation: the retained synthetic two-voice challenge merged speakers on CPU and Metal (DER about 46.3%, above 25%). Passing mixed-source fixtures do not certify cloned/synthetic speaker separation or general meeting accuracy.

Processed on your deviceYour input is processed locally in the desktop app. Model and runtime files download from external hosts.

Target platforms · verification required

  • macos · aarch64, x86_64
  • windows · x86_64

Recommended configuration

FP32 · mps (verified platform only)

Sources & technical details
Original model
https://huggingface.co/nvidia/Nemotron-3-Diarization
Release verification
Verified
Release evidence
Exact original checkpoint release stated in official model card; repository creation and conversion dates are distinct.
Source updated
Sep 24, 2026
Native model files
https://huggingface.co/nvidia/Nemotron-3-Diarization
Files updated
Sep 24, 2026
Artifact revision
f667ed73aee57d40cc39428eb768b4fd87a0a29e

Desktop app

Sample result

Nemotron 3 Diarization

Run Nemotron Diarization 3 on your computer

Download the desktop app to load this model, provide input and view results locally.

Public desktop downloads are coming soon.

Device requirements

macOS Apple SiliconCPU / FP32Tested8.0 GiBRAM estimate
macOS Apple SiliconMPS / FP32Tested8.0 GiBRAM estimate
macOS Apple SiliconMPS / FP16Unverified8.0 GiBRAM estimate
macOS Apple SiliconMPS / BF16Unverified8.0 GiBRAM estimate
Windows Intel / AMDCPU / FP32Unverified8.0 GiBRAM estimate
macOS Intel / AMDCPU / FP32Unverified8.0 GiBRAM estimate
If your device falls short

MemoryAdvisory only: close other apps, reduce image/tile/token/batch settings or choose a smaller verified model/precision. CPU can need more RAM than GPU.

GPUIf the selected GPU runtime fails, existing CPU fallback may recover when enabled, but CPU has its own memory/runtime needs.

PlatformUse a compatible OS/architecture/runtime or another model. These estimates do not add hardware detection or block runs.

Benchmark evidence

Saved desktop measurements apply to the recorded hardware and runtime. Fallback cases report the device that actually ran; they do not prove GPU support on the requested device.

Nemotron 3 Diarization

43 configurations tested
Passed38
Failed5
TestedNov 2, 2026
Runtimetransformers 5.22.1

Quality criteria

No explicit quality gate recorded. Review per-case evidence; full certification is pending.

PrecisionDeviceResultPlatformInference / loadTask metric
FP32mpsAcceleratedPassedVery fastmacos · aarch6473 ms / 0 ms0.003 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch6473 ms / 0 ms0.003 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch6475 ms / 0 ms0.003 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch6475 ms / 0 ms0.003 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch6477 ms / 0 ms0.003 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch6478 ms / 0 ms0.003 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch6486 ms / 0 ms0.003 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64235 ms / 595 ms0.008 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64273 ms / 595 ms0.009 RTF
BF16mpsAcceleratedPassedVery fastmacos · aarch64330 ms / 571 ms0.011 RTF
FP16mpsAcceleratedPassedVery fastmacos · aarch64350 ms / 606 ms0.011 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64362 ms / 586 ms0.012 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64240 ms / 0 ms0.012 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch6467 ms / 0 ms0.034 RTF
FP32cpuPassedVery fastmacos · aarch6480 ms / 0 ms0.004 RTF
FP32cpuPassedVery fastmacos · aarch64105 ms / 0 ms0.005 RTF
FP32cpuPassedVery fastmacos · aarch64189 ms / 0 ms0.005 RTF
FP32cpuPassedVery fastmacos · aarch64190 ms / 0 ms0.005 RTF
FP32cpuPassedVery fastmacos · aarch64191 ms / 0 ms0.005 RTF
FP32cpuPassedVery fastmacos · aarch64193 ms / 0 ms0.006 RTF
FP32cpuPassedVery fastmacos · aarch64196 ms / 0 ms0.006 RTF
FP32cpuPassedVery fastmacos · aarch64197 ms / 0 ms0.006 RTF
FP32cpuPassedVery fastmacos · aarch64222 ms / 0 ms0.006 RTF
FP32cpuPassedVery fastmacos · aarch64224 ms / 475 ms0.006 RTF
FP32cpuPassedVery fastmacos · aarch64226 ms / 459 ms0.007 RTF
FP32cpuPassedVery fastmacos · aarch64226 ms / 487 ms0.007 RTF
FP32cpuPassedVery fastmacos · aarch64283 ms / 0 ms0.008 RTF
FP32cpuPassedVery fastmacos · aarch64286 ms / 0 ms0.008 RTF
FP32cpuPassedVery fastmacos · aarch64287 ms / 0 ms0.008 RTF
FP32cpuPassedVery fastmacos · aarch64287 ms / 0 ms0.008 RTF
FP32cpuPassedVery fastmacos · aarch64291 ms / 0 ms0.008 RTF
FP32cpuPassedVery fastmacos · aarch64292 ms / 0 ms0.008 RTF
FP32cpuPassedVery fastmacos · aarch64306 ms / 0 ms0.009 RTF
FP32cpuPassedVery fastmacos · aarch64319 ms / 445 ms0.009 RTF
FP32cpuPassedVery fastmacos · aarch64323 ms / 452 ms0.009 RTF
FP32cpuPassedVery fastmacos · aarch6428 ms / 0 ms0.010 RTF
FP32cpuPassedVery fastmacos · aarch6428 ms / 0 ms0.010 RTF
FP32cpuPassedVery fastmacos · aarch64378 ms / 485 ms0.011 RTF
FP32cpuFailed—macos · aarch64313 ms / 464 ms0.009 RTF
FP32cpuFailed—macos · aarch64225 ms / 490 ms0.007 RTF
FP32mpsFailed—macos · aarch64275 ms / 772 ms0.010 RTF
FP16mpsFailed—macos · aarch64985 ms / 597 ms0.030 RTF
BF16mpsFailed—macos · aarch64888 ms / 600 ms0.029 RTF
Test setup & sample details
Quality gate
No explicit quality gate recorded. Review the per-case quality evidence; full certification is pending.
Tested
2026-11-02T22:19:05.223209+00:00
Environment
{ "os": "macos", "arch": "aarch64", "hardware": "Apple M3 Max", "osVersion": "27.2", "runtime": "Transformers5.22.1/Torch2.10.0", "interpreter": "App-ownedCPython3.13.18 private development venv" }
Runtime
transformers 5.22.1
Artifact revision
f667ed73aee57d40cc39428eb768b4fd87a0a29e
Fixture
{
  "file": "model-samples/desktop/nemotron-diarization/two-speakers.wav",
  "size": 2281320,
  "sha1": "dd1f5044b4dfa54815d41ae8c786458557b5344e",
  "sha256": "65ff2090f2ae357a70d7d2b65ba00634a37b60642c9de5857acabe97648f287a",
  "url": "https://assets.localai.tools/samples/v1/files/65ff2090f2ae357a70d7d2b65ba00634a37b60642c9de5857acabe97648f287a/two-speakers.wav"
}
Rerun when
Artifact/runtime revision; Audio decoding/resample/processor/inference/postprocessing; Fixture/reference/quality gate; Relevant OS/backend behavior
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "diarizationErrorRate": 0.05330764290301863,
      "maximumDiarizationErrorRate": 0.25,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02,
      "speakerMapping": {
        "0": "A",
        "1": "B"
      },
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 2,
      "passed": true,
      "silenceNoHallucination": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 452.3403749917634,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 417
      },
      "parameterDtype": "torch.float32",
      "acceleration": "cpu",
      "accelerationDetails": "Neural audio encoder/head and logits on selected device; WAV decode/resample, mel features and segment serialization on CPU",
      "attentionImplementation": "sdpa",
      "durationMs": 333.3511670061853,
      "inferenceMs": 322.55333400098607,
      "audioSeconds": 35.644375,
      "realTimeFactor": 0.009352139489223343,
      "audioSecondsPerSecond": 110.50691852371624,
      "activationPlacement": {
        "audioEncoder": "cpu"
      },
      "logitDevice": "cpu",
      "logitShape": [
        1,
        3565,
        8
      ],
      "threshold": 0.5,
      "mode": "offline",
      "streaming": false,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "totalMs": 785.6915419979487,
      "artifactRevision": "f667ed73aee57d40cc39428eb768b4fd87a0a29e",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 938622976,
      "rssScope": "Helper RSS sampled at100ms; includes runtime, not isolated tensor/VRAM minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "diarizationErrorRate": 0.057161207450224794,
      "maximumDiarizationErrorRate": 0.25,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02,
      "speakerMapping": {
        "0": "A",
        "1": "B"
      },
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 2,
      "passed": true,
      "silenceNoHallucination": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 417
      },
      "parameterDtype": "torch.float32",
      "acceleration": "cpu",
      "accelerationDetails": "Neural audio encoder/head and logits on selected device; WAV decode/resample, mel features and segment serialization on CPU",
      "attentionImplementation": "sdpa",
      "durationMs": 293.3331249951152,
      "inferenceMs": 285.6822080066195,
      "audioSeconds": 35.644375,
      "realTimeFactor": 0.008229436622050891,
      "audioSecondsPerSecond": 124.76932059827152,
      "activationPlacement": {
        "audioEncoder": "cpu"
      },
      "logitDevice": "cpu",
      "logitShape": [
        1,
        3565,
        8
      ],
      "threshold": 0.5,
      "mode": "offline",
      "streaming": false,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "totalMs": 293.3331249951152,
      "artifactRevision": "f667ed73aee57d40cc39428eb768b4fd87a0a29e",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 1012432896,
      "rssScope": "Helper RSS sampled at100ms; includes runtime, not isolated tensor/VRAM minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "diarizationErrorRate": 0.19845857418111754,
      "maximumDiarizationErrorRate": 0.25,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02,
      "speakerMapping": {
        "0": "A",
        "1": "B"
      },
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 2,
      "passed": true,
      "silenceNoHallucination": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 417
      },
      "parameterDtype": "torch.float32",
      "acceleration": "cpu",
      "accelerationDetails": "Neural audio encoder/head and logits on selected device; WAV decode/resample, mel features and segment serialization on CPU",
      "attentionImplementation": "sdpa",
      "durationMs": 294.5196670043515,
      "inferenceMs": 286.84495799825527,
      "audioSeconds": 35.644375,
      "realTimeFactor": 0.008262724960231496,
      "audioSecondsPerSecond": 124.26355773775464,
      "activationPlacement": {
        "audioEncoder": "cpu"
      },
      "logitDevice": "cpu",
      "logitShape": [
        1,
        3565,
        8
      ],
      "threshold": 0.5,
      "mode": "offline",
      "streaming": false,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "totalMs": 294.5196670043515,
      "artifactRevision": "f667ed73aee57d40cc39428eb768b4fd87a0a29e",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 1028374528,
      "rssScope": "Helper RSS sampled at100ms; includes runtime, not isolated tensor/VRAM minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "diarizationErrorRate": 0.16475095785440613,
      "maximumDiarizationErrorRate": 0.25,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02,
      "speakerMapping": {
        "0": "A"
      },
      "referenceSpeakers": 1,
      "hypothesisSpeakers": 1,
      "passed": true,
      "silenceNoHallucination": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 417
      },
      "parameterDtype": "torch.float32",
      "acceleration": "cpu",
      "accelerationDetails": "Neural audio encoder/head and logits on selected device; WAV decode/resample, mel features and segment serialization on CPU",
      "attentionImplementation": "sdpa",
      "durationMs": 110.15349999070168,
      "inferenceMs": 105.20841699326411,
      "audioSeconds": 23.387375,
      "realTimeFactor": 0.004709955691508845,
      "audioSecondsPerSecond": 222.29566481831353,
      "activationPlacement": {
        "audioEncoder": "cpu"
      },
      "logitDevice": "cpu",
      "logitShape": [
        1,
        2339,
        8
      ],
      "threshold": 0.5,
      "mode": "offline",
      "streaming": false,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "totalMs": 110.15349999070168,
      "artifactRevision": "f667ed73aee57d40cc39428eb768b4fd87a0a29e",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 1038319616,
      "rssScope": "Helper RSS sampled at100ms; includes runtime, not isolated tensor/VRAM minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "diarizationErrorRate": 0,
      "maximumDiarizationErrorRate": 0.25,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02,
      "speakerMapping": {},
      "referenceSpeakers": 0,
      "hypothesisSpeakers": 0,
      "passed": true,
      "silenceNoHallucination": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 417
      },
      "parameterDtype": "torch.float32",
      "acceleration": "cpu",
      "accelerationDetails": "Neural audio encoder/head and logits on selected device; WAV decode/resample, mel features and segment serialization on CPU",
      "attentionImplementation": "sdpa",
      "durationMs": 28.819208004279062,
      "inferenceMs": 27.672208001604304,
      "audioSeconds": 3,
      "realTimeFactor": 0.00960640266809302,
      "audioSecondsPerSecond": 108.41202118118201,
      "activationPlacement": {
        "audioEncoder": "cpu"
      },
      "logitDevice": "cpu",
      "logitShape": [
        1,
        301,
        8
      ],
      "threshold": 0.5,
      "mode": "offline",
      "streaming": false,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "totalMs": 28.819208004279062,
      "artifactRevision": "f667ed73aee57d40cc39428eb768b4fd87a0a29e",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 1038680064,
      "rssScope": "Helper RSS sampled at100ms; includes runtime, not isolated tensor/VRAM minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "diarizationErrorRate": 0.05330764290301863,
      "maximumDiarizationErrorRate": 0.25,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02,
      "speakerMapping": {
        "0": "A",
        "1": "B"
      },
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 2,
      "passed": true,
      "silenceNoHallucination": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 417
      },
      "parameterDtype": "torch.float32",
      "acceleration": "cpu",
      "accelerationDetails": "Neural audio encoder/head and logits on selected device; WAV decode/resample, mel features and segment serialization on CPU",
      "attentionImplementation": "sdpa",
      "durationMs": 290.6075420032721,
      "inferenceMs": 282.7185839996673,
      "audioSeconds": 35.644375,
      "realTimeFactor": 0.00815297061607258,
      "audioSecondsPerSecond": 126.07722667443022,
      "activationPlacement": {
        "audioEncoder": "cpu"
      },
      "logitDevice": "cpu",
      "logitShape": [
        1,
        3565,
        8
      ],
      "threshold": 0.5,
      "mode": "offline",
      "streaming": false,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "totalMs": 290.6075420032721,
      "artifactRevision": "f667ed73aee57d40cc39428eb768b4fd87a0a29e",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 1061732352,
      "rssScope": "Helper RSS sampled at100ms; includes runtime, not isolated tensor/VRAM minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "diarizationErrorRate": 0.05330764290301863,
      "maximumDiarizationErrorRate": 0.25,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02,
      "speakerMapping": {
        "0": "A",
        "1": "B"
      },
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 2,
      "passed": true,
      "silenceNoHallucination": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 417
      },
      "parameterDtype": "torch.float32",
      "acceleration": "cpu",
      "accelerationDetails": "Neural audio encoder/head and logits on selected device; WAV decode/resample, mel features and segment serialization on CPU",
      "attentionImplementation": "sdpa",
      "durationMs": 296.20345900184475,
      "inferenceMs": 287.3655000003055,
      "audioSeconds": 35.644375,
      "realTimeFactor": 0.008309963605810025,
      "audioSecondsPerSecond": 124.03846321135316,
      "activationPlacement": {
        "audioEncoder": "cpu"
      },
      "logitDevice": "cpu",
      "logitShape": [
        1,
        3565,
        8
      ],
      "threshold": 0.5,
      "mode": "offline",
      "streaming": false,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "totalMs": 296.20345900184475,
      "artifactRevision": "f667ed73aee57d40cc39428eb768b4fd87a0a29e",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 1081081856,
      "rssScope": "Helper RSS sampled at100ms; includes runtime, not isolated tensor/VRAM minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "diarizationErrorRate": 0.05330764290301863,
      "maximumDiarizationErrorRate": 0.25,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02,
      "speakerMapping": {
        "0": "A",
        "1": "B"
      },
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 2,
      "passed": true,
      "silenceNoHallucination": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 417
      },
      "parameterDtype": "torch.float32",
      "acceleration": "cpu",
      "accelerationDetails": "Neural audio encoder/head and logits on selected device; WAV decode/resample, mel features and segment serialization on CPU",
      "attentionImplementation": "sdpa",
      "durationMs": 298.73316700104624,
      "inferenceMs": 291.22245899634436,
      "audioSeconds": 35.644375,
      "realTimeFactor": 0.008380934355029264,
      "audioSecondsPerSecond": 122.39569407813919,
      "activationPlacement": {
        "audioEncoder": "cpu"
      },
      "logitDevice": "cpu",
      "logitShape": [
        1,
        3565,
        8
      ],
      "threshold": 0.5,
      "mode": "offline",
      "streaming": false,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "totalMs": 298.73316700104624,
      "artifactRevision": "f667ed73aee57d40cc39428eb768b4fd87a0a29e",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 1099481088,
      "rssScope": "Helper RSS sampled at100ms; includes runtime, not isolated tensor/VRAM minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "diarizationErrorRate": 0.05330764290301863,
      "maximumDiarizationErrorRate": 0.25,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02,
      "speakerMapping": {
        "0": "A",
        "1": "B"
      },
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 2,
      "passed": true,
      "silenceNoHallucination": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 417
      },
      "parameterDtype": "torch.float32",
      "acceleration": "cpu",
      "accelerationDetails": "Neural audio encoder/head and logits on selected device; WAV decode/resample, mel features and segment serialization on CPU",
      "attentionImplementation": "sdpa",
      "durationMs": 315.4148329922464,
      "inferenceMs": 306.3786669954425,
      "audioSeconds": 35.644375,
      "realTimeFactor": 0.00884893711819176,
      "audioSecondsPerSecond": 116.34091677972548,
      "activationPlacement": {
        "audioEncoder": "cpu"
      },
      "logitDevice": "cpu",
      "logitShape": [
        1,
        3565,
        8
      ],
      "threshold": 0.5,
      "mode": "offline",
      "streaming": false,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "totalMs": 315.4148329922464,
      "artifactRevision": "f667ed73aee57d40cc39428eb768b4fd87a0a29e",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 1119600640,
      "rssScope": "Helper RSS sampled at100ms; includes runtime, not isolated tensor/VRAM minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "diarizationErrorRate": 0.05330764290301863,
      "maximumDiarizationErrorRate": 0.25,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02,
      "speakerMapping": {
        "0": "A",
        "1": "B"
      },
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 2,
      "passed": true,
      "silenceNoHallucination": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 417
      },
      "parameterDtype": "torch.float32",
      "acceleration": "cpu",
      "accelerationDetails": "Neural audio encoder/head and logits on selected device; WAV decode/resample, mel features and segment serialization on CPU",
      "attentionImplementation": "sdpa",
      "durationMs": 300.44120900856797,
      "inferenceMs": 292.16625000117347,
      "audioSeconds": 35.644375,
      "realTimeFactor": 0.008428853332638543,
      "audioSecondsPerSecond": 122.00031660007558,
      "activationPlacement": {
        "audioEncoder": "cpu"
      },
      "logitDevice": "cpu",
      "logitShape": [
        1,
        3565,
        8
      ],
      "threshold": 0.5,
      "mode": "offline",
      "streaming": false,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "totalMs": 300.44120900856797,
      "artifactRevision": "f667ed73aee57d40cc39428eb768b4fd87a0a29e",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 1139802112,
      "rssScope": "Helper RSS sampled at100ms; includes runtime, not isolated tensor/VRAM minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "diarizationErrorRate": 0.05330764290301863,
      "maximumDiarizationErrorRate": 0.25,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02,
      "speakerMapping": {
        "0": "A",
        "1": "B"
      },
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 2,
      "passed": true,
      "silenceNoHallucination": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 484.6222089981893,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 417
      },
      "parameterDtype": "torch.float32",
      "acceleration": "cpu",
      "accelerationDetails": "Neural audio encoder/head and logits on selected device; WAV decode/resample, mel features and segment serialization on CPU",
      "attentionImplementation": "sdpa",
      "durationMs": 389.6052499912912,
      "inferenceMs": 378.07791600062046,
      "audioSeconds": 35.644375,
      "realTimeFactor": 0.01093034314646536,
      "audioSecondsPerSecond": 94.27785514968164,
      "activationPlacement": {
        "audioEncoder": "cpu"
      },
      "logitDevice": "cpu",
      "logitShape": [
        1,
        3565,
        8
      ],
      "threshold": 0.5,
      "mode": "offline",
      "streaming": false,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "totalMs": 874.2274589894805,
      "artifactRevision": "f667ed73aee57d40cc39428eb768b4fd87a0a29e",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 946274304,
      "rssScope": "Helper RSS sampled at100ms; includes runtime, not isolated tensor/VRAM minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "diarizationErrorRate": 0.05330764290301863,
      "maximumDiarizationErrorRate": 0.25,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02,
      "speakerMapping": {
        "0": "A",
        "1": "B"
      },
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 2,
      "passed": true,
      "silenceNoHallucination": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 445.3030409931671,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 417
      },
      "parameterDtype": "torch.float32",
      "acceleration": "cpu",
      "accelerationDetails": "Neural audio encoder/head and logits on selected device; WAV decode/resample, mel features and segment serialization on CPU",
      "attentionImplementation": "sdpa",
      "durationMs": 328.83916700666305,
      "inferenceMs": 318.5634580004262,
      "audioSeconds": 35.644375,
      "realTimeFactor": 0.009225555701472198,
      "audioSecondsPerSecond": 111.89097212760765,
      "activationPlacement": {
        "audioEncoder": "cpu"
      },
      "logitDevice": "cpu",
      "logitShape": [
        1,
        3565,
        8
      ],
      "threshold": 0.5,
      "mode": "offline",
      "streaming": false,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "totalMs": 774.1422079998301,
      "artifactRevision": "f667ed73aee57d40cc39428eb768b4fd87a0a29e",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 938590208,
      "rssScope": "Helper RSS sampled at100ms; includes runtime, not isolated tensor/VRAM minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "diarizationErrorRate": 0.05330764290301863,
      "maximumDiarizationErrorRate": 0.25,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02,
      "speakerMapping": {
        "0": "A",
        "1": "B"
      },
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 2,
      "passed": true,
      "silenceNoHallucination": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 458.8170000060927,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 417
      },
      "parameterDtype": "torch.float32",
      "acceleration": "cpu",
      "accelerationDetails": "Neural audio encoder/head and logits on selected device; WAV decode/resample, mel features and segment serialization on CPU",
      "attentionImplementation": "sdpa",
      "durationMs": 232.80066601000726,
      "inferenceMs": 226.284874996054,
      "audioSeconds": 35.644375,
      "realTimeFactor": 0.006531203479090523,
      "audioSecondsPerSecond": 157.51991820320103,
      "activationPlacement": {
        "audioEncoder": "cpu"
      },
      "logitDevice": "cpu",
      "logitShape": [
        1,
        3565,
        8
      ],
      "threshold": 0.5,
      "mode": "offline",
      "streaming": false,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 691.6176660161,
      "artifactRevision": "f667ed73aee57d40cc39428eb768b4fd87a0a29e",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 936460288,
      "rssScope": "Helper RSS sampled at100ms; includes runtime, not isolated tensor/VRAM minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "diarizationErrorRate": 0.057161207450224794,
      "maximumDiarizationErrorRate": 0.25,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02,
      "speakerMapping": {
        "0": "A",
        "1": "B"
      },
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 2,
      "passed": true,
      "silenceNoHallucination": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 417
      },
      "parameterDtype": "torch.float32",
      "acceleration": "cpu",
      "accelerationDetails": "Neural audio encoder/head and logits on selected device; WAV decode/resample, mel features and segment serialization on CPU",
      "attentionImplementation": "sdpa",
      "durationMs": 226.81433300022036,
      "inferenceMs": 221.74449999874923,
      "audioSeconds": 35.644375,
      "realTimeFactor": 0.006363257400367389,
      "audioSecondsPerSecond": 160.74524960123497,
      "activationPlacement": {
        "audioEncoder": "cpu"
      },
      "logitDevice": "cpu",
      "logitShape": [
        1,
        3565,
        8
      ],
      "threshold": 0.5,
      "mode": "offline",
      "streaming": false,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 226.81433300022036,
      "artifactRevision": "f667ed73aee57d40cc39428eb768b4fd87a0a29e",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 1012858880,
      "rssScope": "Helper RSS sampled at100ms; includes runtime, not isolated tensor/VRAM minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "diarizationErrorRate": 0.19845857418111754,
      "maximumDiarizationErrorRate": 0.25,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02,
      "speakerMapping": {
        "0": "A",
        "1": "B"
      },
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 2,
      "passed": true,
      "silenceNoHallucination": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 417
      },
      "parameterDtype": "torch.float32",
      "acceleration": "cpu",
      "accelerationDetails": "Neural audio encoder/head and logits on selected device; WAV decode/resample, mel features and segment serialization on CPU",
      "attentionImplementation": "sdpa",
      "durationMs": 202.24366598995402,
      "inferenceMs": 197.09108400275,
      "audioSeconds": 35.644375,
      "realTimeFactor": 0.005673929364449623,
      "audioSecondsPerSecond": 180.85229568021785,
      "activationPlacement": {
        "audioEncoder": "cpu"
      },
      "logitDevice": "cpu",
      "logitShape": [
        1,
        3565,
        8
      ],
      "threshold": 0.5,
      "mode": "offline",
      "streaming": false,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 202.24366598995402,
      "artifactRevision": "f667ed73aee57d40cc39428eb768b4fd87a0a29e",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 1032126464,
      "rssScope": "Helper RSS sampled at100ms; includes runtime, not isolated tensor/VRAM minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "diarizationErrorRate": 0.16475095785440613,
      "maximumDiarizationErrorRate": 0.25,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02,
      "speakerMapping": {
        "0": "A"
      },
      "referenceSpeakers": 1,
      "hypothesisSpeakers": 1,
      "passed": true,
      "silenceNoHallucination": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 417
      },
      "parameterDtype": "torch.float32",
      "acceleration": "cpu",
      "accelerationDetails": "Neural audio encoder/head and logits on selected device; WAV decode/resample, mel features and segment serialization on CPU",
      "attentionImplementation": "sdpa",
      "durationMs": 83.12554200529121,
      "inferenceMs": 79.90124999196269,
      "audioSeconds": 23.387375,
      "realTimeFactor": 0.003554291236416708,
      "audioSecondsPerSecond": 292.7034934040775,
      "activationPlacement": {
        "audioEncoder": "cpu"
      },
      "logitDevice": "cpu",
      "logitShape": [
        1,
        2339,
        8
      ],
      "threshold": 0.5,
      "mode": "offline",
      "streaming": false,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 83.12554200529121,
      "artifactRevision": "f667ed73aee57d40cc39428eb768b4fd87a0a29e",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 1041547264,
      "rssScope": "Helper RSS sampled at100ms; includes runtime, not isolated tensor/VRAM minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "diarizationErrorRate": 0,
      "maximumDiarizationErrorRate": 0.25,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02,
      "speakerMapping": {},
      "referenceSpeakers": 0,
      "hypothesisSpeakers": 0,
      "passed": true,
      "silenceNoHallucination": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 417
      },
      "parameterDtype": "torch.float32",
      "acceleration": "cpu",
      "accelerationDetails": "Neural audio encoder/head and logits on selected device; WAV decode/resample, mel features and segment serialization on CPU",
      "attentionImplementation": "sdpa",
      "durationMs": 29.688708993489854,
      "inferenceMs": 28.192041005240753,
      "audioSeconds": 3,
      "realTimeFactor": 0.009896236331163285,
      "audioSecondsPerSecond": 106.4130120782073,
      "activationPlacement": {
        "audioEncoder": "cpu"
      },
      "logitDevice": "cpu",
      "logitShape": [
        1,
        301,
        8
      ],
      "threshold": 0.5,
      "mode": "offline",
      "streaming": false,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 29.688708993489854,
      "artifactRevision": "f667ed73aee57d40cc39428eb768b4fd87a0a29e",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 1042382848,
      "rssScope": "Helper RSS sampled at100ms; includes runtime, not isolated tensor/VRAM minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "diarizationErrorRate": 0.05330764290301863,
      "maximumDiarizationErrorRate": 0.25,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02,
      "speakerMapping": {
        "0": "A",
        "1": "B"
      },
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 2,
      "passed": true,
      "silenceNoHallucination": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 417
      },
      "parameterDtype": "torch.float32",
      "acceleration": "cpu",
      "accelerationDetails": "Neural audio encoder/head and logits on selected device; WAV decode/resample, mel features and segment serialization on CPU",
      "attentionImplementation": "sdpa",
      "durationMs": 200.84616699023172,
      "inferenceMs": 196.06066700362135,
      "audioSeconds": 35.644375,
      "realTimeFactor": 0.0056347226453046725,
      "audioSecondsPerSecond": 181.80278351976446,
      "activationPlacement": {
        "audioEncoder": "cpu"
      },
      "logitDevice": "cpu",
      "logitShape": [
        1,
        3565,
        8
      ],
      "threshold": 0.5,
      "mode": "offline",
      "streaming": false,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 200.84616699023172,
      "artifactRevision": "f667ed73aee57d40cc39428eb768b4fd87a0a29e",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 1065517056,
      "rssScope": "Helper RSS sampled at100ms; includes runtime, not isolated tensor/VRAM minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "diarizationErrorRate": 0.05330764290301863,
      "maximumDiarizationErrorRate": 0.25,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02,
      "speakerMapping": {
        "0": "A",
        "1": "B"
      },
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 2,
      "passed": true,
      "silenceNoHallucination": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 417
      },
      "parameterDtype": "torch.float32",
      "acceleration": "cpu",
      "accelerationDetails": "Neural audio encoder/head and logits on selected device; WAV decode/resample, mel features and segment serialization on CPU",
      "attentionImplementation": "sdpa",
      "durationMs": 198.25699999637436,
      "inferenceMs": 193.1556250056019,
      "audioSeconds": 35.644375,
      "realTimeFactor": 0.00556208377889567,
      "audioSecondsPerSecond": 184.5370798751848,
      "activationPlacement": {
        "audioEncoder": "cpu"
      },
      "logitDevice": "cpu",
      "logitShape": [
        1,
        3565,
        8
      ],
      "threshold": 0.5,
      "mode": "offline",
      "streaming": false,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 198.25699999637436,
      "artifactRevision": "f667ed73aee57d40cc39428eb768b4fd87a0a29e",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 1084014592,
      "rssScope": "Helper RSS sampled at100ms; includes runtime, not isolated tensor/VRAM minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "diarizationErrorRate": 0.05330764290301863,
      "maximumDiarizationErrorRate": 0.25,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02,
      "speakerMapping": {
        "0": "A",
        "1": "B"
      },
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 2,
      "passed": true,
      "silenceNoHallucination": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 417
      },
      "parameterDtype": "torch.float32",
      "acceleration": "cpu",
      "accelerationDetails": "Neural audio encoder/head and logits on selected device; WAV decode/resample, mel features and segment serialization on CPU",
      "attentionImplementation": "sdpa",
      "durationMs": 195.13658300274983,
      "inferenceMs": 190.50375001097564,
      "audioSeconds": 35.644375,
      "realTimeFactor": 0.00547454073757079,
      "audioSecondsPerSecond": 187.10589685476737,
      "activationPlacement": {
        "audioEncoder": "cpu"
      },
      "logitDevice": "cpu",
      "logitShape": [
        1,
        3565,
        8
      ],
      "threshold": 0.5,
      "mode": "offline",
      "streaming": false,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 195.13658300274983,
      "artifactRevision": "f667ed73aee57d40cc39428eb768b4fd87a0a29e",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 1099907072,
      "rssScope": "Helper RSS sampled at100ms; includes runtime, not isolated tensor/VRAM minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "diarizationErrorRate": 0.05330764290301863,
      "maximumDiarizationErrorRate": 0.25,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02,
      "speakerMapping": {
        "0": "A",
        "1": "B"
      },
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 2,
      "passed": true,
      "silenceNoHallucination": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 417
      },
      "parameterDtype": "torch.float32",
      "acceleration": "cpu",
      "accelerationDetails": "Neural audio encoder/head and logits on selected device; WAV decode/resample, mel features and segment serialization on CPU",
      "attentionImplementation": "sdpa",
      "durationMs": 193.58562500565313,
      "inferenceMs": 188.89383299392648,
      "audioSeconds": 35.644375,
      "realTimeFactor": 0.005431028738914713,
      "audioSecondsPerSecond": 188.70057552988547,
      "activationPlacement": {
        "audioEncoder": "cpu"
      },
      "logitDevice": "cpu",
      "logitShape": [
        1,
        3565,
        8
      ],
      "threshold": 0.5,
      "mode": "offline",
      "streaming": false,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 193.58562500565313,
      "artifactRevision": "f667ed73aee57d40cc39428eb768b4fd87a0a29e",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 1122156544,
      "rssScope": "Helper RSS sampled at100ms; includes runtime, not isolated tensor/VRAM minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "diarizationErrorRate": 0.05330764290301863,
      "maximumDiarizationErrorRate": 0.25,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02,
      "speakerMapping": {
        "0": "A",
        "1": "B"
      },
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 2,
      "passed": true,
      "silenceNoHallucination": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 417
      },
      "parameterDtype": "torch.float32",
      "acceleration": "cpu",
      "accelerationDetails": "Neural audio encoder/head and logits on selected device; WAV decode/resample, mel features and segment serialization on CPU",
      "attentionImplementation": "sdpa",
      "durationMs": 194.86337499984074,
      "inferenceMs": 189.9699580098968,
      "audioSeconds": 35.644375,
      "realTimeFactor": 0.00546687590958856,
      "audioSecondsPerSecond": 187.63164119951557,
      "activationPlacement": {
        "audioEncoder": "cpu"
      },
      "logitDevice": "cpu",
      "logitShape": [
        1,
        3565,
        8
      ],
      "threshold": 0.5,
      "mode": "offline",
      "streaming": false,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 194.86337499984074,
      "artifactRevision": "f667ed73aee57d40cc39428eb768b4fd87a0a29e",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 1140621312,
      "rssScope": "Helper RSS sampled at100ms; includes runtime, not isolated tensor/VRAM minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "diarizationErrorRate": 0.05330764290301863,
      "maximumDiarizationErrorRate": 0.25,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02,
      "speakerMapping": {
        "0": "A",
        "1": "B"
      },
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 2,
      "passed": true,
      "silenceNoHallucination": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 475.28212500037625,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 417
      },
      "parameterDtype": "torch.float32",
      "acceleration": "cpu",
      "accelerationDetails": "Neural audio encoder/head and logits on selected device; WAV decode/resample, mel features and segment serialization on CPU",
      "attentionImplementation": "sdpa",
      "durationMs": 230.65087500435766,
      "inferenceMs": 224.18829100206494,
      "audioSeconds": 35.644375,
      "realTimeFactor": 0.006470891269782614,
      "audioSecondsPerSecond": 158.9930269804844,
      "activationPlacement": {
        "audioEncoder": "cpu"
      },
      "logitDevice": "cpu",
      "logitShape": [
        1,
        3565,
        8
      ],
      "threshold": 0.5,
      "mode": "offline",
      "streaming": false,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 705.9330000047339,
      "artifactRevision": "f667ed73aee57d40cc39428eb768b4fd87a0a29e",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 938459136,
      "rssScope": "Helper RSS sampled at100ms; includes runtime, not isolated tensor/VRAM minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "diarizationErrorRate": 0.05330764290301863,
      "maximumDiarizationErrorRate": 0.25,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02,
      "speakerMapping": {
        "0": "A",
        "1": "B"
      },
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 2,
      "passed": true,
      "silenceNoHallucination": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 486.9487909891177,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 417
      },
      "parameterDtype": "torch.float32",
      "acceleration": "cpu",
      "accelerationDetails": "Neural audio encoder/head and logits on selected device; WAV decode/resample, mel features and segment serialization on CPU",
      "attentionImplementation": "sdpa",
      "durationMs": 233.80649999307934,
      "inferenceMs": 226.29554200102575,
      "audioSeconds": 35.644375,
      "realTimeFactor": 0.006559422068505321,
      "audioSecondsPerSecond": 157.51249310884978,
      "activationPlacement": {
        "audioEncoder": "cpu"
      },
      "logitDevice": "cpu",
      "logitShape": [
        1,
        3565,
        8
      ],
      "threshold": 0.5,
      "mode": "offline",
      "streaming": false,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 720.755290982197,
      "artifactRevision": "f667ed73aee57d40cc39428eb768b4fd87a0a29e",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 935706624,
      "rssScope": "Helper RSS sampled at100ms; includes runtime, not isolated tensor/VRAM minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "diarizationErrorRate": 0.05330764290301863,
      "maximumDiarizationErrorRate": 0.25,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02,
      "speakerMapping": {
        "0": "A",
        "1": "B"
      },
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 2,
      "passed": true,
      "silenceNoHallucination": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 586.1367500037886,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 417
      },
      "parameterDtype": "torch.float32",
      "acceleration": "partial",
      "accelerationDetails": "Neural audio encoder/head and logits on selected device; WAV decode/resample, mel features and segment serialization on CPU",
      "attentionImplementation": "sdpa",
      "durationMs": 411.23674999107607,
      "inferenceMs": 362.28579201269895,
      "audioSeconds": 35.644375,
      "realTimeFactor": 0.011537213094382383,
      "audioSecondsPerSecond": 98.38744931722462,
      "activationPlacement": {
        "audioEncoder": "mps:0"
      },
      "logitDevice": "mps:0",
      "logitShape": [
        1,
        3565,
        8
      ],
      "threshold": 0.5,
      "mode": "offline",
      "streaming": false,
      "driverAllocatedMemoryBytes": 508100608,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 997.3734999948647,
      "artifactRevision": "f667ed73aee57d40cc39428eb768b4fd87a0a29e",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 627933184,
      "rssScope": "Helper RSS sampled at100ms; includes runtime, not isolated tensor/VRAM minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "diarizationErrorRate": 0.057161207450224794,
      "maximumDiarizationErrorRate": 0.25,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02,
      "speakerMapping": {
        "0": "A",
        "1": "B"
      },
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 2,
      "passed": true,
      "silenceNoHallucination": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 417
      },
      "parameterDtype": "torch.float32",
      "acceleration": "partial",
      "accelerationDetails": "Neural audio encoder/head and logits on selected device; WAV decode/resample, mel features and segment serialization on CPU",
      "attentionImplementation": "sdpa",
      "durationMs": 93.66254099586513,
      "inferenceMs": 76.53741700050887,
      "audioSeconds": 35.644375,
      "realTimeFactor": 0.002627694860573797,
      "audioSecondsPerSecond": 465.71175768530327,
      "activationPlacement": {
        "audioEncoder": "mps:0"
      },
      "logitDevice": "mps:0",
      "logitShape": [
        1,
        3565,
        8
      ],
      "threshold": 0.5,
      "mode": "offline",
      "streaming": false,
      "driverAllocatedMemoryBytes": 508084224,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 93.66254099586513,
      "artifactRevision": "f667ed73aee57d40cc39428eb768b4fd87a0a29e",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 729726976,
      "rssScope": "Helper RSS sampled at100ms; includes runtime, not isolated tensor/VRAM minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "diarizationErrorRate": 0.19845857418111754,
      "maximumDiarizationErrorRate": 0.25,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02,
      "speakerMapping": {
        "0": "A",
        "1": "B"
      },
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 2,
      "passed": true,
      "silenceNoHallucination": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 417
      },
      "parameterDtype": "torch.float32",
      "acceleration": "partial",
      "accelerationDetails": "Neural audio encoder/head and logits on selected device; WAV decode/resample, mel features and segment serialization on CPU",
      "attentionImplementation": "sdpa",
      "durationMs": 89.22920799523126,
      "inferenceMs": 72.77404199703597,
      "audioSeconds": 35.644375,
      "realTimeFactor": 0.0025033180689865166,
      "audioSecondsPerSecond": 489.79518001008887,
      "activationPlacement": {
        "audioEncoder": "mps:0"
      },
      "logitDevice": "mps:0",
      "logitShape": [
        1,
        3565,
        8
      ],
      "threshold": 0.5,
      "mode": "offline",
      "streaming": false,
      "driverAllocatedMemoryBytes": 508084224,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 89.22920799523126,
      "artifactRevision": "f667ed73aee57d40cc39428eb768b4fd87a0a29e",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 748060672,
      "rssScope": "Helper RSS sampled at100ms; includes runtime, not isolated tensor/VRAM minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "diarizationErrorRate": 0.16475095785440613,
      "maximumDiarizationErrorRate": 0.25,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02,
      "speakerMapping": {
        "0": "A"
      },
      "referenceSpeakers": 1,
      "hypothesisSpeakers": 1,
      "passed": true,
      "silenceNoHallucination": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 417
      },
      "parameterDtype": "torch.float32",
      "acceleration": "partial",
      "accelerationDetails": "Neural audio encoder/head and logits on selected device; WAV decode/resample, mel features and segment serialization on CPU",
      "attentionImplementation": "sdpa",
      "durationMs": 273.2904169970425,
      "inferenceMs": 239.94479200337082,
      "audioSeconds": 23.387375,
      "realTimeFactor": 0.011685382262739725,
      "audioSecondsPerSecond": 97.46981713890021,
      "activationPlacement": {
        "audioEncoder": "mps:0"
      },
      "logitDevice": "mps:0",
      "logitShape": [
        1,
        2339,
        8
      ],
      "threshold": 0.5,
      "mode": "offline",
      "streaming": false,
      "driverAllocatedMemoryBytes": 515063808,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 273.2904169970425,
      "artifactRevision": "f667ed73aee57d40cc39428eb768b4fd87a0a29e",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 779288576,
      "rssScope": "Helper RSS sampled at100ms; includes runtime, not isolated tensor/VRAM minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "diarizationErrorRate": 0,
      "maximumDiarizationErrorRate": 0.25,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02,
      "speakerMapping": {},
      "referenceSpeakers": 0,
      "hypothesisSpeakers": 0,
      "passed": true,
      "silenceNoHallucination": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 417
      },
      "parameterDtype": "torch.float32",
      "acceleration": "partial",
      "accelerationDetails": "Neural audio encoder/head and logits on selected device; WAV decode/resample, mel features and segment serialization on CPU",
      "attentionImplementation": "sdpa",
      "durationMs": 100.70616700977553,
      "inferenceMs": 67.24845900316723,
      "audioSeconds": 3,
      "realTimeFactor": 0.03356872233659184,
      "audioSecondsPerSecond": 44.61068765692769,
      "activationPlacement": {
        "audioEncoder": "mps:0"
      },
      "logitDevice": "mps:0",
      "logitShape": [
        1,
        301,
        8
      ],
      "threshold": 0.5,
      "mode": "offline",
      "streaming": false,
      "driverAllocatedMemoryBytes": 515293184,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 100.70616700977553,
      "artifactRevision": "f667ed73aee57d40cc39428eb768b4fd87a0a29e",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 797376512,
      "rssScope": "Helper RSS sampled at100ms; includes runtime, not isolated tensor/VRAM minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "diarizationErrorRate": 0.05330764290301863,
      "maximumDiarizationErrorRate": 0.25,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02,
      "speakerMapping": {
        "0": "A",
        "1": "B"
      },
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 2,
      "passed": true,
      "silenceNoHallucination": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 417
      },
      "parameterDtype": "torch.float32",
      "acceleration": "partial",
      "accelerationDetails": "Neural audio encoder/head and logits on selected device; WAV decode/resample, mel features and segment serialization on CPU",
      "attentionImplementation": "sdpa",
      "durationMs": 103.04120799992234,
      "inferenceMs": 85.80666700436268,
      "audioSeconds": 35.644375,
      "realTimeFactor": 0.0028908125896420496,
      "audioSecondsPerSecond": 415.40332755481256,
      "activationPlacement": {
        "audioEncoder": "mps:0"
      },
      "logitDevice": "mps:0",
      "logitShape": [
        1,
        3565,
        8
      ],
      "threshold": 0.5,
      "mode": "offline",
      "streaming": false,
      "driverAllocatedMemoryBytes": 515293184,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 103.04120799992234,
      "artifactRevision": "f667ed73aee57d40cc39428eb768b4fd87a0a29e",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 827211776,
      "rssScope": "Helper RSS sampled at100ms; includes runtime, not isolated tensor/VRAM minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "diarizationErrorRate": 0.05330764290301863,
      "maximumDiarizationErrorRate": 0.25,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02,
      "speakerMapping": {
        "0": "A",
        "1": "B"
      },
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 2,
      "passed": true,
      "silenceNoHallucination": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 417
      },
      "parameterDtype": "torch.float32",
      "acceleration": "partial",
      "accelerationDetails": "Neural audio encoder/head and logits on selected device; WAV decode/resample, mel features and segment serialization on CPU",
      "attentionImplementation": "sdpa",
      "durationMs": 90.10358399245888,
      "inferenceMs": 73.06041600531898,
      "audioSeconds": 35.644375,
      "realTimeFactor": 0.002527848615453599,
      "audioSecondsPerSecond": 487.8753359056291,
      "activationPlacement": {
        "audioEncoder": "mps:0"
      },
      "logitDevice": "mps:0",
      "logitShape": [
        1,
        3565,
        8
      ],
      "threshold": 0.5,
      "mode": "offline",
      "streaming": false,
      "driverAllocatedMemoryBytes": 515293184,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 90.10358399245888,
      "artifactRevision": "f667ed73aee57d40cc39428eb768b4fd87a0a29e",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 845529088,
      "rssScope": "Helper RSS sampled at100ms; includes runtime, not isolated tensor/VRAM minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "diarizationErrorRate": 0.05330764290301863,
      "maximumDiarizationErrorRate": 0.25,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02,
      "speakerMapping": {
        "0": "A",
        "1": "B"
      },
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 2,
      "passed": true,
      "silenceNoHallucination": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 417
      },
      "parameterDtype": "torch.float32",
      "acceleration": "partial",
      "accelerationDetails": "Neural audio encoder/head and logits on selected device; WAV decode/resample, mel features and segment serialization on CPU",
      "attentionImplementation": "sdpa",
      "durationMs": 91.94383400608785,
      "inferenceMs": 74.96429199818522,
      "audioSeconds": 35.644375,
      "realTimeFactor": 0.0025794766777671894,
      "audioSecondsPerSecond": 475.48471478744705,
      "activationPlacement": {
        "audioEncoder": "mps:0"
      },
      "logitDevice": "mps:0",
      "logitShape": [
        1,
        3565,
        8
      ],
      "threshold": 0.5,
      "mode": "offline",
      "streaming": false,
      "driverAllocatedMemoryBytes": 515293184,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 91.94383400608785,
      "artifactRevision": "f667ed73aee57d40cc39428eb768b4fd87a0a29e",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 863862784,
      "rssScope": "Helper RSS sampled at100ms; includes runtime, not isolated tensor/VRAM minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "diarizationErrorRate": 0.05330764290301863,
      "maximumDiarizationErrorRate": 0.25,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02,
      "speakerMapping": {
        "0": "A",
        "1": "B"
      },
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 2,
      "passed": true,
      "silenceNoHallucination": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 417
      },
      "parameterDtype": "torch.float32",
      "acceleration": "partial",
      "accelerationDetails": "Neural audio encoder/head and logits on selected device; WAV decode/resample, mel features and segment serialization on CPU",
      "attentionImplementation": "sdpa",
      "durationMs": 91.41262499906588,
      "inferenceMs": 75.18087500648107,
      "audioSeconds": 35.644375,
      "realTimeFactor": 0.0025645736528993953,
      "audioSecondsPerSecond": 474.1149261288489,
      "activationPlacement": {
        "audioEncoder": "mps:0"
      },
      "logitDevice": "mps:0",
      "logitShape": [
        1,
        3565,
        8
      ],
      "threshold": 0.5,
      "mode": "offline",
      "streaming": false,
      "driverAllocatedMemoryBytes": 515293184,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 91.41262499906588,
      "artifactRevision": "f667ed73aee57d40cc39428eb768b4fd87a0a29e",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 882196480,
      "rssScope": "Helper RSS sampled at100ms; includes runtime, not isolated tensor/VRAM minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "diarizationErrorRate": 0.05330764290301863,
      "maximumDiarizationErrorRate": 0.25,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02,
      "speakerMapping": {
        "0": "A",
        "1": "B"
      },
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 2,
      "passed": true,
      "silenceNoHallucination": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 417
      },
      "parameterDtype": "torch.float32",
      "acceleration": "partial",
      "accelerationDetails": "Neural audio encoder/head and logits on selected device; WAV decode/resample, mel features and segment serialization on CPU",
      "attentionImplementation": "sdpa",
      "durationMs": 97.98120800405741,
      "inferenceMs": 78.19529200787656,
      "audioSeconds": 35.644375,
      "realTimeFactor": 0.002748854707202957,
      "audioSecondsPerSecond": 455.83786548696,
      "activationPlacement": {
        "audioEncoder": "mps:0"
      },
      "logitDevice": "mps:0",
      "logitShape": [
        1,
        3565,
        8
      ],
      "threshold": 0.5,
      "mode": "offline",
      "streaming": false,
      "driverAllocatedMemoryBytes": 515293184,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 97.98120800405741,
      "artifactRevision": "f667ed73aee57d40cc39428eb768b4fd87a0a29e",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 900497408,
      "rssScope": "Helper RSS sampled at100ms; includes runtime, not isolated tensor/VRAM minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "diarizationErrorRate": 0.05330764290301863,
      "maximumDiarizationErrorRate": 0.25,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02,
      "speakerMapping": {
        "0": "A",
        "1": "B"
      },
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 2,
      "passed": true,
      "silenceNoHallucination": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 594.9126659979811,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 417
      },
      "parameterDtype": "torch.float32",
      "acceleration": "partial",
      "accelerationDetails": "Neural audio encoder/head and logits on selected device; WAV decode/resample, mel features and segment serialization on CPU",
      "attentionImplementation": "sdpa",
      "durationMs": 326.8607499921927,
      "inferenceMs": 272.8932500031078,
      "audioSeconds": 35.644375,
      "realTimeFactor": 0.009170051375348641,
      "audioSecondsPerSecond": 130.61655060941985,
      "activationPlacement": {
        "audioEncoder": "mps:0"
      },
      "logitDevice": "mps:0",
      "logitShape": [
        1,
        3565,
        8
      ],
      "threshold": 0.5,
      "mode": "offline",
      "streaming": false,
      "driverAllocatedMemoryBytes": 508100608,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 921.7734159901738,
      "artifactRevision": "f667ed73aee57d40cc39428eb768b4fd87a0a29e",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 629276672,
      "rssScope": "Helper RSS sampled at100ms; includes runtime, not isolated tensor/VRAM minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "diarizationErrorRate": 0.05330764290301863,
      "maximumDiarizationErrorRate": 0.25,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02,
      "speakerMapping": {
        "0": "A",
        "1": "B"
      },
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 2,
      "passed": true,
      "silenceNoHallucination": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 594.5489580044523,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 417
      },
      "parameterDtype": "torch.float32",
      "acceleration": "partial",
      "accelerationDetails": "Neural audio encoder/head and logits on selected device; WAV decode/resample, mel features and segment serialization on CPU",
      "attentionImplementation": "sdpa",
      "durationMs": 281.2325000122655,
      "inferenceMs": 235.48295799992047,
      "audioSeconds": 35.644375,
      "realTimeFactor": 0.007889954586446403,
      "audioSecondsPerSecond": 151.367110820869,
      "activationPlacement": {
        "audioEncoder": "mps:0"
      },
      "logitDevice": "mps:0",
      "logitShape": [
        1,
        3565,
        8
      ],
      "threshold": 0.5,
      "mode": "offline",
      "streaming": false,
      "driverAllocatedMemoryBytes": 508100608,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 875.7814580167178,
      "artifactRevision": "f667ed73aee57d40cc39428eb768b4fd87a0a29e",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 658833408,
      "rssScope": "Helper RSS sampled at100ms; includes runtime, not isolated tensor/VRAM minimum"
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "diarizationErrorRate": 0.05330764290301863,
      "maximumDiarizationErrorRate": 0.25,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02,
      "speakerMapping": {
        "0": "A",
        "1": "B"
      },
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 2,
      "passed": true,
      "silenceNoHallucination": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 606.4514169993345,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float16": 417
      },
      "parameterDtype": "torch.float16",
      "acceleration": "partial",
      "accelerationDetails": "Neural audio encoder/head and logits on selected device; WAV decode/resample, mel features and segment serialization on CPU",
      "attentionImplementation": "sdpa",
      "durationMs": 400.1084589981474,
      "inferenceMs": 349.58991600433365,
      "audioSeconds": 35.644375,
      "realTimeFactor": 0.011225009808648557,
      "audioSecondsPerSecond": 101.96053538214224,
      "activationPlacement": {
        "audioEncoder": "mps:0"
      },
      "logitDevice": "mps:0",
      "logitShape": [
        1,
        3565,
        8
      ],
      "threshold": 0.5,
      "mode": "offline",
      "streaming": false,
      "driverAllocatedMemoryBytes": 298401792,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 1006.5598759974819,
      "artifactRevision": "f667ed73aee57d40cc39428eb768b4fd87a0a29e",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 830472192,
      "rssScope": "Helper RSS sampled at100ms; includes runtime, not isolated tensor/VRAM minimum"
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "diarizationErrorRate": 0.05202312138728324,
      "maximumDiarizationErrorRate": 0.25,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02,
      "speakerMapping": {
        "0": "A",
        "1": "B"
      },
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 2,
      "passed": true,
      "silenceNoHallucination": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 571.1150420102058,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.bfloat16": 417
      },
      "parameterDtype": "torch.bfloat16",
      "acceleration": "partial",
      "accelerationDetails": "Neural audio encoder/head and logits on selected device; WAV decode/resample, mel features and segment serialization on CPU",
      "attentionImplementation": "sdpa",
      "durationMs": 379.2660000035539,
      "inferenceMs": 330.2432089985814,
      "audioSeconds": 35.644375,
      "realTimeFactor": 0.01064027634103709,
      "audioSecondsPerSecond": 107.93371075846441,
      "activationPlacement": {
        "audioEncoder": "mps:0"
      },
      "logitDevice": "mps:0",
      "logitShape": [
        1,
        3565,
        8
      ],
      "threshold": 0.5,
      "mode": "offline",
      "streaming": false,
      "driverAllocatedMemoryBytes": 298500096,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "device": "mps",
      "precision": "bf16",
      "cpuThreads": 4,
      "totalMs": 950.3810420137597,
      "artifactRevision": "f667ed73aee57d40cc39428eb768b4fd87a0a29e",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 857686016,
      "rssScope": "Helper RSS sampled at100ms; includes runtime, not isolated tensor/VRAM minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "fail",
    "quality": {
      "diarizationErrorRate": 0.4625668449197861,
      "maximumDiarizationErrorRate": 0.25,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02,
      "speakerMapping": {
        "0": "A"
      },
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 1,
      "passed": false,
      "silenceNoHallucination": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 463.93550001084805,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 417
      },
      "parameterDtype": "torch.float32",
      "acceleration": "cpu",
      "accelerationDetails": "Neural audio encoder/head and logits on selected device; WAV decode/resample, mel features and segment serialization on CPU",
      "attentionImplementation": "sdpa",
      "durationMs": 322.738874994684,
      "inferenceMs": 312.5376670068363,
      "audioSeconds": 34.439375,
      "realTimeFactor": 0.00937121753791072,
      "audioSecondsPerSecond": 110.19271798444278,
      "activationPlacement": {
        "audioEncoder": "cpu"
      },
      "logitDevice": "cpu",
      "logitShape": [
        1,
        3444,
        8
      ],
      "threshold": 0.5,
      "mode": "offline",
      "streaming": false,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "totalMs": 786.674375005532,
      "artifactRevision": "f667ed73aee57d40cc39428eb768b4fd87a0a29e",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 937525248,
      "rssScope": "Helper RSS sampled at100ms; includes runtime, not isolated tensor/VRAM minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "fail",
    "quality": {
      "diarizationErrorRate": 0.4625668449197861,
      "maximumDiarizationErrorRate": 0.25,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02,
      "speakerMapping": {
        "0": "A"
      },
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 1,
      "passed": false,
      "silenceNoHallucination": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 490.07054200046696,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 417
      },
      "parameterDtype": "torch.float32",
      "acceleration": "cpu",
      "accelerationDetails": "Neural audio encoder/head and logits on selected device; WAV decode/resample, mel features and segment serialization on CPU",
      "attentionImplementation": "sdpa",
      "durationMs": 232.00079100206494,
      "inferenceMs": 224.791542001185,
      "audioSeconds": 34.439375,
      "realTimeFactor": 0.006736498295978511,
      "audioSecondsPerSecond": 153.20583102641135,
      "activationPlacement": {
        "audioEncoder": "cpu"
      },
      "logitDevice": "cpu",
      "logitShape": [
        1,
        3444,
        8
      ],
      "threshold": 0.5,
      "mode": "offline",
      "streaming": false,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 722.0713330025319,
      "artifactRevision": "f667ed73aee57d40cc39428eb768b4fd87a0a29e",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 938672128,
      "rssScope": "Helper RSS sampled at100ms; includes runtime, not isolated tensor/VRAM minimum"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "fail",
    "quality": {
      "diarizationErrorRate": 0.4625668449197861,
      "maximumDiarizationErrorRate": 0.25,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02,
      "speakerMapping": {
        "0": "A"
      },
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 1,
      "passed": false,
      "silenceNoHallucination": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 771.5842920006253,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 417
      },
      "parameterDtype": "torch.float32",
      "acceleration": "partial",
      "accelerationDetails": "Neural audio encoder/head and logits on selected device; WAV decode/resample, mel features and segment serialization on CPU",
      "attentionImplementation": "sdpa",
      "durationMs": 328.1763749982929,
      "inferenceMs": 275.3020839882083,
      "audioSeconds": 34.439375,
      "realTimeFactor": 0.009529103678516027,
      "audioSecondsPerSecond": 125.09667381041365,
      "activationPlacement": {
        "audioEncoder": "mps:0"
      },
      "logitDevice": "mps:0",
      "logitShape": [
        1,
        3444,
        8
      ],
      "threshold": 0.5,
      "mode": "offline",
      "streaming": false,
      "driverAllocatedMemoryBytes": 499990528,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 1099.7606669989182,
      "artifactRevision": "f667ed73aee57d40cc39428eb768b4fd87a0a29e",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 625311744,
      "rssScope": "Helper RSS sampled at100ms; includes runtime, not isolated tensor/VRAM minimum"
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "fail",
    "quality": {
      "diarizationErrorRate": 0.4625668449197861,
      "maximumDiarizationErrorRate": 0.25,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02,
      "speakerMapping": {
        "0": "A"
      },
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 1,
      "passed": false,
      "silenceNoHallucination": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 596.6885829984676,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float16": 417
      },
      "parameterDtype": "torch.float16",
      "acceleration": "partial",
      "accelerationDetails": "Neural audio encoder/head and logits on selected device; WAV decode/resample, mel features and segment serialization on CPU",
      "attentionImplementation": "sdpa",
      "durationMs": 1036.7030000052182,
      "inferenceMs": 984.6470420015976,
      "audioSeconds": 34.439375,
      "realTimeFactor": 0.030102259405265577,
      "audioSecondsPerSecond": 34.97636567311611,
      "activationPlacement": {
        "audioEncoder": "mps:0"
      },
      "logitDevice": "mps:0",
      "logitShape": [
        1,
        3444,
        8
      ],
      "threshold": 0.5,
      "mode": "offline",
      "streaming": false,
      "driverAllocatedMemoryBytes": 298663936,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 1633.3915830036858,
      "artifactRevision": "f667ed73aee57d40cc39428eb768b4fd87a0a29e",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 812236800,
      "rssScope": "Helper RSS sampled at100ms; includes runtime, not isolated tensor/VRAM minimum"
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "fail",
    "quality": {
      "diarizationErrorRate": 0.4625668449197861,
      "maximumDiarizationErrorRate": 0.25,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02,
      "speakerMapping": {
        "0": "A"
      },
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 1,
      "passed": false,
      "silenceNoHallucination": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 600.2538750035455,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.bfloat16": 417
      },
      "parameterDtype": "torch.bfloat16",
      "acceleration": "partial",
      "accelerationDetails": "Neural audio encoder/head and logits on selected device; WAV decode/resample, mel features and segment serialization on CPU",
      "attentionImplementation": "sdpa",
      "durationMs": 1003.0719580099685,
      "inferenceMs": 888.4784580004634,
      "audioSeconds": 34.439375,
      "realTimeFactor": 0.02912573059209026,
      "audioSecondsPerSecond": 38.762194727271634,
      "activationPlacement": {
        "audioEncoder": "mps:0"
      },
      "logitDevice": "mps:0",
      "logitShape": [
        1,
        3444,
        8
      ],
      "threshold": 0.5,
      "mode": "offline",
      "streaming": false,
      "driverAllocatedMemoryBytes": 298762240,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "device": "mps",
      "precision": "bf16",
      "cpuThreads": 4,
      "totalMs": 1603.325833013514,
      "artifactRevision": "f667ed73aee57d40cc39428eb768b4fd87a0a29e",
      "fallback": false,
      "grade": "fast",
      "sampledPeakHelperRssBytes": 825049088,
      "rssScope": "Helper RSS sampled at100ms; includes runtime, not isolated tensor/VRAM minimum"
    }
  }
]

Something not quite right?

Tell us what happened with Nemotron Diarization 3. Loading, results, or the page itself—we’re here for all of it.

Attachments Optional

Up to 3 files · 50 MB each

Browser, OS, CPU/GPU, memory, screen, storage, and network details are included automatically. Model inputs, results, and browsing history are not collected.