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.

Parakeet TDT 3 Desktop

Transcribe speech with punctuation and approximate word-level timestamps. Native multilingual FastConformer/TDT decoding retains complete token timings and durations.

Desktop inferencePrivate · Runs on your computer
About this model

Transcribe speech with punctuation and approximate word-level timestamps. Native multilingual FastConformer/TDT decoding retains complete token timings and durations.

Created by
NVIDIA
License
CC-BY-4.0
Model released
Aug 14, 2025
Source repository created
Aug 4, 2025
Model weights · FP16
2.5 GB

Best for

  • Local transcription with word timing for captions and audio navigation
  • Exploring multilingual speech recognition on a compatible desktop

Keep in mind

  • Transcripts may hallucinate, omit or substitute words, especially with noise, overlapping speech, accents, music and languages outside the25 supported European languages. Never rely on output for safety-critical decisions.
  • Quality evidence covers two short English recordings, a silence shift and exact silence. It does not certify all25 languages, broad meetings, long-form recordings or human word-boundary accuracy. No diarization or translation is returned.
  • Word times are envelopes of native TDT token spans on an80ms encoder grid; punctuation can have zero duration. Approximate timing is not forced alignment or human annotation. Confidence is not provided.
  • Input is finite mono16kHz Float32 PCM. The app decodes audio locally. No arbitrary30-second or file-size cap is imposed; full-recording attention, JSON buffers and token budgets can exhaust memory. Split long recordings when necessary.
  • maxNewTokens limits decoding work; exhausted budgets return an explicit error instead of silently truncating. Increasing the budget can increase resource use.
  • Original pinned2.51GB FP32 checkpoint is downloaded on request; FP16 execution casts the same weights and does not reduce transfer size. App-owned Python and framework packages are installed separately on demand.
  • Only Apple M3 Max/macOS27.2 CPU FP32 and Metal FP32/FP16 are verified. Intel Mac, other Mac hardware/OS and Windows CUDA/AMD/Intel GPU providers remain unverified.
  • Neural parameter, encoder activation and generated tensor placement on Metal is verified with implicit MPS fallback disabled. Mel features, token grouping and JSON remain on CPU: acceleration is partial.
  • Pinned Transformers5.22.1 Metal generation can retain one terminal duration beyond the trimmed token sequence due to its deferred stop check. Both complete native arrays are preserved; the official processor uses the aligned prefix.

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

FP16 · mps (verified platform only)

Sources & technical details
Original model
https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3
Release verification
Verified
Release evidence
Exact checkpoint Release Date stated by original provider in model card.
Source updated
Aug 5, 2026
Native model files
https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3
Files updated
Aug 5, 2026
Artifact revision
541d1f99c6b0c3cd0b11a95167540bb8edefd82b

Desktop app

Sample result

Parakeet TDT 0.6B v3 Desktop

Run Parakeet TDT 3 Desktop 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 / FP32Tested16.0 GiBRAM estimate
macOS Apple SiliconMPS / FP32Tested16.0 GiBRAM estimate
macOS Apple SiliconMPS / FP16Tested16.0 GiBRAM estimate
Windows Intel / AMDCPU / FP32Unverified16.0 GiBRAM estimate
macOS Intel / AMDCPU / FP32Unverified16.0 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.

Parakeet TDT 0.6B v3 Desktop

36 configurations tested
Passed36
Failed0
TestedNov 2, 2026
Runtimetransformers 5.22.1

Quality criteria

{ "scope": "JFK and one authored synthetic English sentence, leading-silence consistency and exact silence; no broad25-language/human-word-boundary/meeting certification", "maximumWordErrorRate": 0.25, "maximumLeadingSilenceWordShiftErrorSeconds": 0.24, "minimumCorrespondingWordFraction": 0.7, "maximumConsecutiveRepeatedWords": 6, "wordTimes": "Finite monotonic starts, nonnegative spans within audio duration+0.2s; punctuation may have zero native duration", "nativeEvidence": "Complete text, native token timestamps/IDs/durations, grouped words/chunks, actual parameter/encoder/generated tensor placement", "silence": "No returned text/words" }

PrecisionDeviceResultPlatformInference / loadTask metric
FP16mpsAcceleratedPassedVery fastmacos · aarch64144 ms / 0 ms0.013 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64145 ms / 0 ms0.013 RTF
FP16mpsAcceleratedPassedVery fastmacos · aarch64147 ms / 0 ms0.013 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64153 ms / 0 ms0.014 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64155 ms / 0 ms0.014 RTF
FP16mpsAcceleratedPassedVery fastmacos · aarch64178 ms / 0 ms0.016 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64207 ms / 0 ms0.018 RTF
FP16mpsAcceleratedPassedVery fastmacos · aarch64213 ms / 0 ms0.018 RTF
FP16mpsAcceleratedPassedVery fastmacos · aarch64323 ms / 2502 ms0.029 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64339 ms / 2681 ms0.031 RTF
FP16mpsAcceleratedPassedVery fastmacos · aarch64342 ms / 2412 ms0.031 RTF
FP16mpsAcceleratedPassedVery fastmacos · aarch64343 ms / 2496 ms0.031 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64346 ms / 2650 ms0.031 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64422 ms / 3041 ms0.038 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64135 ms / 0 ms0.040 RTF
FP16mpsAcceleratedPassedVery fastmacos · aarch64140 ms / 0 ms0.041 RTF
FP16mpsAcceleratedPassedVery fastmacos · aarch64128 ms / 0 ms0.064 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64131 ms / 0 ms0.066 RTF
FP32cpuPassedVery fastmacos · aarch64318 ms / 0 ms0.029 RTF
FP32cpuPassedVery fastmacos · aarch64320 ms / 0 ms0.029 RTF
FP32cpuPassedVery fastmacos · aarch64335 ms / 0 ms0.029 RTF
FP32cpuPassedVery fastmacos · aarch64323 ms / 0 ms0.029 RTF
FP32cpuPassedVery fastmacos · aarch64498 ms / 2201 ms0.045 RTF
FP32cpuPassedVery fastmacos · aarch64502 ms / 2148 ms0.046 RTF
FP32cpuPassedVery fastmacos · aarch64503 ms / 2216 ms0.046 RTF
FP32cpuPassedVery fastmacos · aarch64554 ms / 0 ms0.048 RTF
FP32cpuPassedVery fastmacos · aarch64559 ms / 0 ms0.051 RTF
FP32cpuPassedVery fastmacos · aarch64562 ms / 0 ms0.051 RTF
FP32cpuPassedVery fastmacos · aarch64587 ms / 0 ms0.053 RTF
FP32cpuPassedVery fastmacos · aarch64185 ms / 0 ms0.054 RTF
FP32cpuPassedVery fastmacos · aarch64724 ms / 2144 ms0.066 RTF
FP32cpuPassedVery fastmacos · aarch64734 ms / 2200 ms0.067 RTF
FP32cpuPassedVery fastmacos · aarch64749 ms / 2216 ms0.068 RTF
FP32cpuPassedVery fastmacos · aarch64140 ms / 0 ms0.070 RTF
FP32cpuPassedFastmacos · aarch64443 ms / 0 ms0.130 RTF
FP32cpuPassedFastmacos · aarch64441 ms / 0 ms0.221 RTF
Test setup & sample details
Quality gate
{ "scope": "JFK and one authored synthetic English sentence, leading-silence consistency and exact silence; no broad25-language/human-word-boundary/meeting certification", "maximumWordErrorRate": 0.25, "maximumLeadingSilenceWordShiftErrorSeconds": 0.24, "minimumCorrespondingWordFraction": 0.7, "maximumConsecutiveRepeatedWords": 6, "wordTimes": "Finite monotonic starts, nonnegative spans within audio duration+0.2s; punctuation may have zero native duration", "nativeEvidence": "Complete text, native token timestamps/IDs/durations, grouped words/chunks, actual parameter/encoder/generated tensor placement", "silence": "No returned text/words" }
Tested
2026-11-02T14:01:21.205753+00:00
Environment
{ "os": "macos", "arch": "aarch64", "hardware": "Apple M3 Max", "osVersion": "27.2", "runtimeProvisioning": "Explicit development-only app-owned isolated CPython3.13 probe; managed packaged provisioning tested separately" }
Runtime
transformers 5.22.1
Artifact revision
541d1f99c6b0c3cd0b11a95167540bb8edefd82b
Fixture
{
  "file": "model-samples/vosk-jfk.flac",
  "url": "https://assets.localai.tools/samples/v1/files/63a4b1e4c1dc655ac70961ffbf518acd249df237e5a0152faae9a4a836949715/vosk-jfk.flac",
  "size": 1152693,
  "sha1": "537ddf75dff344f8d31c1cd6ebee23ee8362d921",
  "sha256": "63a4b1e4c1dc655ac70961ffbf518acd249df237e5a0152faae9a4a836949715"
}
Rerun when
Artifact/runtime/preprocessing/decoding changes; Fixture/config/quality gates change; Relevant OS/device/backend behavior changes
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "fixtureId": "jfk",
      "wordErrorRate": 0,
      "finiteOrderedTimes": true,
      "wordReconstruction": true,
      "completeNativeTokenDurations": true,
      "wordCount": 22,
      "passed": true,
      "maximumConsecutiveRepeatedWords": 1
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 2215.9973329980858,
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "placement": {
        "cpu/torch.float32": 651
      },
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "accelerationDetails": "FastConformer encoder/TDT network and generated token/duration tensors on selected device; mel features, token-to-word grouping and JSON on CPU. No implicit MPS fallback.",
      "durationMs": 833.4062909998465,
      "inferenceMs": 748.9317500003381,
      "realTimeFactor": 0.06808470454548528,
      "activationPlacement": {
        "encoderOutput": "cpu"
      },
      "predictionPlacement": "cpu",
      "durationPlacement": "cpu",
      "outputTokens": 47,
      "maxNewTokens": 1024,
      "decoding": "greedy TDT",
      "sampleRate": 16000,
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "artifactRevision": "541d1f99c6b0c3cd0b11a95167540bb8edefd82b",
      "totalMs": 3049.4036239979323,
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "fixtureId": "synthetic",
      "wordErrorRate": 0,
      "finiteOrderedTimes": true,
      "wordReconstruction": true,
      "completeNativeTokenDurations": true,
      "wordCount": 9,
      "passed": true,
      "maximumConsecutiveRepeatedWords": 1
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "placement": {
        "cpu/torch.float32": 651
      },
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "accelerationDetails": "FastConformer encoder/TDT network and generated token/duration tensors on selected device; mel features, token-to-word grouping and JSON on CPU. No implicit MPS fallback.",
      "durationMs": 478.6062079947442,
      "inferenceMs": 442.97024999104906,
      "realTimeFactor": 0.1302853676444262,
      "activationPlacement": {
        "encoderOutput": "cpu"
      },
      "predictionPlacement": "cpu",
      "durationPlacement": "cpu",
      "outputTokens": 19,
      "maxNewTokens": 1024,
      "decoding": "greedy TDT",
      "sampleRate": 16000,
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "artifactRevision": "541d1f99c6b0c3cd0b11a95167540bb8edefd82b",
      "totalMs": 478.6062079947442,
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "fixtureId": "jfk-shifted",
      "wordErrorRate": 0,
      "finiteOrderedTimes": true,
      "wordReconstruction": true,
      "completeNativeTokenDurations": true,
      "wordCount": 22,
      "passed": true,
      "maximumConsecutiveRepeatedWords": 1,
      "correspondingWordFraction": 1,
      "medianLeadingSilenceShiftErrorSeconds": 0.02000000000000024
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "placement": {
        "cpu/torch.float32": 651
      },
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "accelerationDetails": "FastConformer encoder/TDT network and generated token/duration tensors on selected device; mel features, token-to-word grouping and JSON on CPU. No implicit MPS fallback.",
      "durationMs": 636.9834999932209,
      "inferenceMs": 554.0613750054035,
      "realTimeFactor": 0.048179250000469874,
      "activationPlacement": {
        "encoderOutput": "cpu"
      },
      "predictionPlacement": "cpu",
      "durationPlacement": "cpu",
      "outputTokens": 48,
      "maxNewTokens": 1024,
      "decoding": "greedy TDT",
      "sampleRate": 16000,
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "artifactRevision": "541d1f99c6b0c3cd0b11a95167540bb8edefd82b",
      "totalMs": 636.9834999932209,
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "fixtureId": "silence",
      "wordErrorRate": 0,
      "finiteOrderedTimes": true,
      "wordReconstruction": true,
      "completeNativeTokenDurations": true,
      "wordCount": 0,
      "passed": true,
      "maximumConsecutiveRepeatedWords": 0
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "placement": {
        "cpu/torch.float32": 651
      },
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "accelerationDetails": "FastConformer encoder/TDT network and generated token/duration tensors on selected device; mel features, token-to-word grouping and JSON on CPU. No implicit MPS fallback.",
      "durationMs": 443.56700000935234,
      "inferenceMs": 441.08945901098195,
      "realTimeFactor": 0.22054472950549098,
      "activationPlacement": {
        "encoderOutput": "cpu"
      },
      "predictionPlacement": "cpu",
      "durationPlacement": "cpu",
      "outputTokens": 20,
      "maxNewTokens": 1024,
      "decoding": "greedy TDT",
      "sampleRate": 16000,
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "artifactRevision": "541d1f99c6b0c3cd0b11a95167540bb8edefd82b",
      "totalMs": 443.56700000935234,
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "fixtureId": "jfk",
      "wordErrorRate": 0,
      "finiteOrderedTimes": true,
      "wordReconstruction": true,
      "completeNativeTokenDurations": true,
      "wordCount": 22,
      "passed": true,
      "maximumConsecutiveRepeatedWords": 1
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "placement": {
        "cpu/torch.float32": 651
      },
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "accelerationDetails": "FastConformer encoder/TDT network and generated token/duration tensors on selected device; mel features, token-to-word grouping and JSON on CPU. No implicit MPS fallback.",
      "durationMs": 675.0692909990903,
      "inferenceMs": 586.8734170071548,
      "realTimeFactor": 0.05335212881883225,
      "activationPlacement": {
        "encoderOutput": "cpu"
      },
      "predictionPlacement": "cpu",
      "durationPlacement": "cpu",
      "outputTokens": 47,
      "maxNewTokens": 1024,
      "decoding": "greedy TDT",
      "sampleRate": 16000,
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "artifactRevision": "541d1f99c6b0c3cd0b11a95167540bb8edefd82b",
      "totalMs": 675.0692909990903,
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "fixtureId": "jfk",
      "wordErrorRate": 0,
      "finiteOrderedTimes": true,
      "wordReconstruction": true,
      "completeNativeTokenDurations": true,
      "wordCount": 22,
      "passed": true,
      "maximumConsecutiveRepeatedWords": 1
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 2144.152833003318,
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "placement": {
        "cpu/torch.float32": 651
      },
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "accelerationDetails": "FastConformer encoder/TDT network and generated token/duration tensors on selected device; mel features, token-to-word grouping and JSON on CPU. No implicit MPS fallback.",
      "durationMs": 808.8051669910783,
      "inferenceMs": 724.2293329909444,
      "realTimeFactor": 0.06583903027190403,
      "activationPlacement": {
        "encoderOutput": "cpu"
      },
      "predictionPlacement": "cpu",
      "durationPlacement": "cpu",
      "outputTokens": 47,
      "maxNewTokens": 1024,
      "decoding": "greedy TDT",
      "sampleRate": 16000,
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "artifactRevision": "541d1f99c6b0c3cd0b11a95167540bb8edefd82b",
      "totalMs": 2952.9579999943962,
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "fixtureId": "jfk",
      "wordErrorRate": 0,
      "finiteOrderedTimes": true,
      "wordReconstruction": true,
      "completeNativeTokenDurations": true,
      "wordCount": 22,
      "passed": true,
      "maximumConsecutiveRepeatedWords": 1
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "placement": {
        "cpu/torch.float32": 651
      },
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "accelerationDetails": "FastConformer encoder/TDT network and generated token/duration tensors on selected device; mel features, token-to-word grouping and JSON on CPU. No implicit MPS fallback.",
      "durationMs": 644.0681660023984,
      "inferenceMs": 561.7249999922933,
      "realTimeFactor": 0.05106590909020848,
      "activationPlacement": {
        "encoderOutput": "cpu"
      },
      "predictionPlacement": "cpu",
      "durationPlacement": "cpu",
      "outputTokens": 47,
      "maxNewTokens": 1024,
      "decoding": "greedy TDT",
      "sampleRate": 16000,
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "artifactRevision": "541d1f99c6b0c3cd0b11a95167540bb8edefd82b",
      "totalMs": 644.0681660023984,
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "fixtureId": "jfk",
      "wordErrorRate": 0,
      "finiteOrderedTimes": true,
      "wordReconstruction": true,
      "completeNativeTokenDurations": true,
      "wordCount": 22,
      "passed": true,
      "maximumConsecutiveRepeatedWords": 1
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 2200.333375003538,
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "placement": {
        "cpu/torch.float32": 651
      },
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "accelerationDetails": "FastConformer encoder/TDT network and generated token/duration tensors on selected device; mel features, token-to-word grouping and JSON on CPU. No implicit MPS fallback.",
      "durationMs": 817.1962920023361,
      "inferenceMs": 734.2680000001565,
      "realTimeFactor": 0.06675163636365058,
      "activationPlacement": {
        "encoderOutput": "cpu"
      },
      "predictionPlacement": "cpu",
      "durationPlacement": "cpu",
      "outputTokens": 47,
      "maxNewTokens": 1024,
      "decoding": "greedy TDT",
      "sampleRate": 16000,
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "artifactRevision": "541d1f99c6b0c3cd0b11a95167540bb8edefd82b",
      "totalMs": 3017.529667005874,
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "fixtureId": "jfk",
      "wordErrorRate": 0,
      "finiteOrderedTimes": true,
      "wordReconstruction": true,
      "completeNativeTokenDurations": true,
      "wordCount": 22,
      "passed": true,
      "maximumConsecutiveRepeatedWords": 1
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "placement": {
        "cpu/torch.float32": 651
      },
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "accelerationDetails": "FastConformer encoder/TDT network and generated token/duration tensors on selected device; mel features, token-to-word grouping and JSON on CPU. No implicit MPS fallback.",
      "durationMs": 640.1361669995822,
      "inferenceMs": 559.3320830084849,
      "realTimeFactor": 0.05084837118258954,
      "activationPlacement": {
        "encoderOutput": "cpu"
      },
      "predictionPlacement": "cpu",
      "durationPlacement": "cpu",
      "outputTokens": 47,
      "maxNewTokens": 1024,
      "decoding": "greedy TDT",
      "sampleRate": 16000,
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "artifactRevision": "541d1f99c6b0c3cd0b11a95167540bb8edefd82b",
      "totalMs": 640.1361669995822,
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "fixtureId": "jfk",
      "wordErrorRate": 0,
      "finiteOrderedTimes": true,
      "wordReconstruction": true,
      "completeNativeTokenDurations": true,
      "wordCount": 22,
      "passed": true,
      "maximumConsecutiveRepeatedWords": 1
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 2147.6812080072705,
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "placement": {
        "cpu/torch.float32": 651
      },
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "accelerationDetails": "FastConformer encoder/TDT network and generated token/duration tensors on selected device; mel features, token-to-word grouping and JSON on CPU. No implicit MPS fallback.",
      "durationMs": 590.6112080119783,
      "inferenceMs": 502.3293339909287,
      "realTimeFactor": 0.04566630309008443,
      "activationPlacement": {
        "encoderOutput": "cpu"
      },
      "predictionPlacement": "cpu",
      "durationPlacement": "cpu",
      "outputTokens": 47,
      "maxNewTokens": 1024,
      "decoding": "greedy TDT",
      "sampleRate": 16000,
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "artifactRevision": "541d1f99c6b0c3cd0b11a95167540bb8edefd82b",
      "totalMs": 2738.292416019249,
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "fixtureId": "synthetic",
      "wordErrorRate": 0,
      "finiteOrderedTimes": true,
      "wordReconstruction": true,
      "completeNativeTokenDurations": true,
      "wordCount": 9,
      "passed": true,
      "maximumConsecutiveRepeatedWords": 1
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "placement": {
        "cpu/torch.float32": 651
      },
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "accelerationDetails": "FastConformer encoder/TDT network and generated token/duration tensors on selected device; mel features, token-to-word grouping and JSON on CPU. No implicit MPS fallback.",
      "durationMs": 220.98349999578204,
      "inferenceMs": 184.8671669868054,
      "realTimeFactor": 0.05437269617258982,
      "activationPlacement": {
        "encoderOutput": "cpu"
      },
      "predictionPlacement": "cpu",
      "durationPlacement": "cpu",
      "outputTokens": 19,
      "maxNewTokens": 1024,
      "decoding": "greedy TDT",
      "sampleRate": 16000,
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "artifactRevision": "541d1f99c6b0c3cd0b11a95167540bb8edefd82b",
      "totalMs": 220.98349999578204,
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "fixtureId": "jfk-shifted",
      "wordErrorRate": 0,
      "finiteOrderedTimes": true,
      "wordReconstruction": true,
      "completeNativeTokenDurations": true,
      "wordCount": 22,
      "passed": true,
      "maximumConsecutiveRepeatedWords": 1,
      "correspondingWordFraction": 1,
      "medianLeadingSilenceShiftErrorSeconds": 0.02000000000000024
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "placement": {
        "cpu/torch.float32": 651
      },
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "accelerationDetails": "FastConformer encoder/TDT network and generated token/duration tensors on selected device; mel features, token-to-word grouping and JSON on CPU. No implicit MPS fallback.",
      "durationMs": 422.36420800327323,
      "inferenceMs": 335.1857499947073,
      "realTimeFactor": 0.029146586956061503,
      "activationPlacement": {
        "encoderOutput": "cpu"
      },
      "predictionPlacement": "cpu",
      "durationPlacement": "cpu",
      "outputTokens": 48,
      "maxNewTokens": 1024,
      "decoding": "greedy TDT",
      "sampleRate": 16000,
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "artifactRevision": "541d1f99c6b0c3cd0b11a95167540bb8edefd82b",
      "totalMs": 422.36420800327323,
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "fixtureId": "silence",
      "wordErrorRate": 0,
      "finiteOrderedTimes": true,
      "wordReconstruction": true,
      "completeNativeTokenDurations": true,
      "wordCount": 0,
      "passed": true,
      "maximumConsecutiveRepeatedWords": 0
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "placement": {
        "cpu/torch.float32": 651
      },
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "accelerationDetails": "FastConformer encoder/TDT network and generated token/duration tensors on selected device; mel features, token-to-word grouping and JSON on CPU. No implicit MPS fallback.",
      "durationMs": 142.43958299630322,
      "inferenceMs": 139.98608299880289,
      "realTimeFactor": 0.06999304149940144,
      "activationPlacement": {
        "encoderOutput": "cpu"
      },
      "predictionPlacement": "cpu",
      "durationPlacement": "cpu",
      "outputTokens": 20,
      "maxNewTokens": 1024,
      "decoding": "greedy TDT",
      "sampleRate": 16000,
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "artifactRevision": "541d1f99c6b0c3cd0b11a95167540bb8edefd82b",
      "totalMs": 142.43958299630322,
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "fixtureId": "jfk",
      "wordErrorRate": 0,
      "finiteOrderedTimes": true,
      "wordReconstruction": true,
      "completeNativeTokenDurations": true,
      "wordCount": 22,
      "passed": true,
      "maximumConsecutiveRepeatedWords": 1
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "placement": {
        "cpu/torch.float32": 651
      },
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "accelerationDetails": "FastConformer encoder/TDT network and generated token/duration tensors on selected device; mel features, token-to-word grouping and JSON on CPU. No implicit MPS fallback.",
      "durationMs": 407.7389580052113,
      "inferenceMs": 323.29470801050775,
      "realTimeFactor": 0.02939042800095525,
      "activationPlacement": {
        "encoderOutput": "cpu"
      },
      "predictionPlacement": "cpu",
      "durationPlacement": "cpu",
      "outputTokens": 47,
      "maxNewTokens": 1024,
      "decoding": "greedy TDT",
      "sampleRate": 16000,
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "artifactRevision": "541d1f99c6b0c3cd0b11a95167540bb8edefd82b",
      "totalMs": 407.7389580052113,
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "fixtureId": "jfk",
      "wordErrorRate": 0,
      "finiteOrderedTimes": true,
      "wordReconstruction": true,
      "completeNativeTokenDurations": true,
      "wordCount": 22,
      "passed": true,
      "maximumConsecutiveRepeatedWords": 1
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 2200.934167005471,
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "placement": {
        "cpu/torch.float32": 651
      },
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "accelerationDetails": "FastConformer encoder/TDT network and generated token/duration tensors on selected device; mel features, token-to-word grouping and JSON on CPU. No implicit MPS fallback.",
      "durationMs": 585.9738330036635,
      "inferenceMs": 498.38545799138956,
      "realTimeFactor": 0.04530776890830814,
      "activationPlacement": {
        "encoderOutput": "cpu"
      },
      "predictionPlacement": "cpu",
      "durationPlacement": "cpu",
      "outputTokens": 47,
      "maxNewTokens": 1024,
      "decoding": "greedy TDT",
      "sampleRate": 16000,
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "artifactRevision": "541d1f99c6b0c3cd0b11a95167540bb8edefd82b",
      "totalMs": 2786.9080000091344,
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "fixtureId": "jfk",
      "wordErrorRate": 0,
      "finiteOrderedTimes": true,
      "wordReconstruction": true,
      "completeNativeTokenDurations": true,
      "wordCount": 22,
      "passed": true,
      "maximumConsecutiveRepeatedWords": 1
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "placement": {
        "cpu/torch.float32": 651
      },
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "accelerationDetails": "FastConformer encoder/TDT network and generated token/duration tensors on selected device; mel features, token-to-word grouping and JSON on CPU. No implicit MPS fallback.",
      "durationMs": 405.0460839935113,
      "inferenceMs": 320.23854200087953,
      "realTimeFactor": 0.029112594727352684,
      "activationPlacement": {
        "encoderOutput": "cpu"
      },
      "predictionPlacement": "cpu",
      "durationPlacement": "cpu",
      "outputTokens": 47,
      "maxNewTokens": 1024,
      "decoding": "greedy TDT",
      "sampleRate": 16000,
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "artifactRevision": "541d1f99c6b0c3cd0b11a95167540bb8edefd82b",
      "totalMs": 405.0460839935113,
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "fixtureId": "jfk",
      "wordErrorRate": 0,
      "finiteOrderedTimes": true,
      "wordReconstruction": true,
      "completeNativeTokenDurations": true,
      "wordCount": 22,
      "passed": true,
      "maximumConsecutiveRepeatedWords": 1
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 2215.987999996287,
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "placement": {
        "cpu/torch.float32": 651
      },
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "accelerationDetails": "FastConformer encoder/TDT network and generated token/duration tensors on selected device; mel features, token-to-word grouping and JSON on CPU. No implicit MPS fallback.",
      "durationMs": 588.3055409940425,
      "inferenceMs": 502.9750000103377,
      "realTimeFactor": 0.04572500000093979,
      "activationPlacement": {
        "encoderOutput": "cpu"
      },
      "predictionPlacement": "cpu",
      "durationPlacement": "cpu",
      "outputTokens": 47,
      "maxNewTokens": 1024,
      "decoding": "greedy TDT",
      "sampleRate": 16000,
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "artifactRevision": "541d1f99c6b0c3cd0b11a95167540bb8edefd82b",
      "totalMs": 2804.2935409903293,
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "fixtureId": "jfk",
      "wordErrorRate": 0,
      "finiteOrderedTimes": true,
      "wordReconstruction": true,
      "completeNativeTokenDurations": true,
      "wordCount": 22,
      "passed": true,
      "maximumConsecutiveRepeatedWords": 1
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "placement": {
        "cpu/torch.float32": 651
      },
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "accelerationDetails": "FastConformer encoder/TDT network and generated token/duration tensors on selected device; mel features, token-to-word grouping and JSON on CPU. No implicit MPS fallback.",
      "durationMs": 399.9606249999488,
      "inferenceMs": 318.1246669992106,
      "realTimeFactor": 0.02892042427265551,
      "activationPlacement": {
        "encoderOutput": "cpu"
      },
      "predictionPlacement": "cpu",
      "durationPlacement": "cpu",
      "outputTokens": 47,
      "maxNewTokens": 1024,
      "decoding": "greedy TDT",
      "sampleRate": 16000,
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "artifactRevision": "541d1f99c6b0c3cd0b11a95167540bb8edefd82b",
      "totalMs": 399.9606249999488,
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "fixtureId": "jfk",
      "wordErrorRate": 0,
      "finiteOrderedTimes": true,
      "wordReconstruction": true,
      "completeNativeTokenDurations": true,
      "wordCount": 22,
      "passed": true,
      "maximumConsecutiveRepeatedWords": 1
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 2650.0925839936826,
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "placement": {
        "mps:0/torch.float32": 651
      },
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "accelerationDetails": "FastConformer encoder/TDT network and generated token/duration tensors on selected device; mel features, token-to-word grouping and JSON on CPU. No implicit MPS fallback.",
      "durationMs": 468.85058301268145,
      "inferenceMs": 346.32883399899583,
      "realTimeFactor": 0.031484439454454165,
      "activationPlacement": {
        "encoderOutput": "mps:0"
      },
      "predictionPlacement": "mps:0",
      "durationPlacement": "mps:0",
      "outputTokens": 47,
      "maxNewTokens": 1024,
      "decoding": "greedy TDT",
      "sampleRate": 16000,
      "driverAllocatedMemoryBytes": 3333914624,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "artifactRevision": "541d1f99c6b0c3cd0b11a95167540bb8edefd82b",
      "totalMs": 3118.943167006364,
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "fixtureId": "synthetic",
      "wordErrorRate": 0,
      "finiteOrderedTimes": true,
      "wordReconstruction": true,
      "completeNativeTokenDurations": true,
      "wordCount": 9,
      "passed": true,
      "maximumConsecutiveRepeatedWords": 1
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "placement": {
        "mps:0/torch.float32": 651
      },
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "accelerationDetails": "FastConformer encoder/TDT network and generated token/duration tensors on selected device; mel features, token-to-word grouping and JSON on CPU. No implicit MPS fallback.",
      "durationMs": 191.38658400333952,
      "inferenceMs": 135.22016700881068,
      "realTimeFactor": 0.03977063735553255,
      "activationPlacement": {
        "encoderOutput": "mps:0"
      },
      "predictionPlacement": "mps:0",
      "durationPlacement": "mps:0",
      "outputTokens": 19,
      "maxNewTokens": 1024,
      "decoding": "greedy TDT",
      "sampleRate": 16000,
      "driverAllocatedMemoryBytes": 3334242304,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "artifactRevision": "541d1f99c6b0c3cd0b11a95167540bb8edefd82b",
      "totalMs": 191.38658400333952,
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "fixtureId": "jfk-shifted",
      "wordErrorRate": 0,
      "finiteOrderedTimes": true,
      "wordReconstruction": true,
      "completeNativeTokenDurations": true,
      "wordCount": 22,
      "passed": true,
      "maximumConsecutiveRepeatedWords": 1,
      "correspondingWordFraction": 1,
      "medianLeadingSilenceShiftErrorSeconds": 0.02000000000000024
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "placement": {
        "mps:0/torch.float32": 651
      },
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "accelerationDetails": "FastConformer encoder/TDT network and generated token/duration tensors on selected device; mel features, token-to-word grouping and JSON on CPU. No implicit MPS fallback.",
      "durationMs": 327.94829200429376,
      "inferenceMs": 207.25708300597034,
      "realTimeFactor": 0.01802235504399742,
      "activationPlacement": {
        "encoderOutput": "mps:0"
      },
      "predictionPlacement": "mps:0",
      "durationPlacement": "mps:0",
      "outputTokens": 48,
      "maxNewTokens": 1024,
      "decoding": "greedy TDT",
      "sampleRate": 16000,
      "driverAllocatedMemoryBytes": 3343040512,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "artifactRevision": "541d1f99c6b0c3cd0b11a95167540bb8edefd82b",
      "totalMs": 327.94829200429376,
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "fixtureId": "silence",
      "wordErrorRate": 0,
      "finiteOrderedTimes": true,
      "wordReconstruction": true,
      "completeNativeTokenDurations": true,
      "wordCount": 0,
      "passed": true,
      "maximumConsecutiveRepeatedWords": 0
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "placement": {
        "mps:0/torch.float32": 651
      },
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "accelerationDetails": "FastConformer encoder/TDT network and generated token/duration tensors on selected device; mel features, token-to-word grouping and JSON on CPU. No implicit MPS fallback.",
      "durationMs": 142.00245900428854,
      "inferenceMs": 131.05279099545442,
      "realTimeFactor": 0.06552639549772721,
      "activationPlacement": {
        "encoderOutput": "mps:0"
      },
      "predictionPlacement": "mps:0",
      "durationPlacement": "mps:0",
      "outputTokens": 20,
      "maxNewTokens": 1024,
      "decoding": "greedy TDT",
      "sampleRate": 16000,
      "driverAllocatedMemoryBytes": 3343351808,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "artifactRevision": "541d1f99c6b0c3cd0b11a95167540bb8edefd82b",
      "totalMs": 142.00245900428854,
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "fixtureId": "jfk",
      "wordErrorRate": 0,
      "finiteOrderedTimes": true,
      "wordReconstruction": true,
      "completeNativeTokenDurations": true,
      "wordCount": 22,
      "passed": true,
      "maximumConsecutiveRepeatedWords": 1
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "placement": {
        "mps:0/torch.float32": 651
      },
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "accelerationDetails": "FastConformer encoder/TDT network and generated token/duration tensors on selected device; mel features, token-to-word grouping and JSON on CPU. No implicit MPS fallback.",
      "durationMs": 260.24524999957066,
      "inferenceMs": 144.95229099702556,
      "realTimeFactor": 0.013177480999729596,
      "activationPlacement": {
        "encoderOutput": "mps:0"
      },
      "predictionPlacement": "mps:0",
      "durationPlacement": "mps:0",
      "outputTokens": 47,
      "maxNewTokens": 1024,
      "decoding": "greedy TDT",
      "sampleRate": 16000,
      "driverAllocatedMemoryBytes": 3343351808,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "artifactRevision": "541d1f99c6b0c3cd0b11a95167540bb8edefd82b",
      "totalMs": 260.24524999957066,
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "fixtureId": "jfk",
      "wordErrorRate": 0,
      "finiteOrderedTimes": true,
      "wordReconstruction": true,
      "completeNativeTokenDurations": true,
      "wordCount": 22,
      "passed": true,
      "maximumConsecutiveRepeatedWords": 1
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 2680.532874990604,
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "placement": {
        "mps:0/torch.float32": 651
      },
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "accelerationDetails": "FastConformer encoder/TDT network and generated token/duration tensors on selected device; mel features, token-to-word grouping and JSON on CPU. No implicit MPS fallback.",
      "durationMs": 462.2284170036437,
      "inferenceMs": 338.54466599586885,
      "realTimeFactor": 0.03077678781780626,
      "activationPlacement": {
        "encoderOutput": "mps:0"
      },
      "predictionPlacement": "mps:0",
      "durationPlacement": "mps:0",
      "outputTokens": 47,
      "maxNewTokens": 1024,
      "decoding": "greedy TDT",
      "sampleRate": 16000,
      "driverAllocatedMemoryBytes": 3333914624,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "artifactRevision": "541d1f99c6b0c3cd0b11a95167540bb8edefd82b",
      "totalMs": 3142.7612919942476,
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "fixtureId": "jfk",
      "wordErrorRate": 0,
      "finiteOrderedTimes": true,
      "wordReconstruction": true,
      "completeNativeTokenDurations": true,
      "wordCount": 22,
      "passed": true,
      "maximumConsecutiveRepeatedWords": 1
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "placement": {
        "mps:0/torch.float32": 651
      },
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "accelerationDetails": "FastConformer encoder/TDT network and generated token/duration tensors on selected device; mel features, token-to-word grouping and JSON on CPU. No implicit MPS fallback.",
      "durationMs": 275.93124999839347,
      "inferenceMs": 155.33183299703524,
      "realTimeFactor": 0.014121075727003203,
      "activationPlacement": {
        "encoderOutput": "mps:0"
      },
      "predictionPlacement": "mps:0",
      "durationPlacement": "mps:0",
      "outputTokens": 47,
      "maxNewTokens": 1024,
      "decoding": "greedy TDT",
      "sampleRate": 16000,
      "driverAllocatedMemoryBytes": 3333914624,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "artifactRevision": "541d1f99c6b0c3cd0b11a95167540bb8edefd82b",
      "totalMs": 275.93124999839347,
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "fixtureId": "jfk",
      "wordErrorRate": 0,
      "finiteOrderedTimes": true,
      "wordReconstruction": true,
      "completeNativeTokenDurations": true,
      "wordCount": 22,
      "passed": true,
      "maximumConsecutiveRepeatedWords": 1
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 3041.3269589917036,
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "placement": {
        "mps:0/torch.float32": 651
      },
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "accelerationDetails": "FastConformer encoder/TDT network and generated token/duration tensors on selected device; mel features, token-to-word grouping and JSON on CPU. No implicit MPS fallback.",
      "durationMs": 553.2127500046045,
      "inferenceMs": 422.3334579874063,
      "realTimeFactor": 0.03839395072612784,
      "activationPlacement": {
        "encoderOutput": "mps:0"
      },
      "predictionPlacement": "mps:0",
      "durationPlacement": "mps:0",
      "outputTokens": 47,
      "maxNewTokens": 1024,
      "decoding": "greedy TDT",
      "sampleRate": 16000,
      "driverAllocatedMemoryBytes": 3333914624,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "artifactRevision": "541d1f99c6b0c3cd0b11a95167540bb8edefd82b",
      "totalMs": 3594.539708996308,
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "fixtureId": "jfk",
      "wordErrorRate": 0,
      "finiteOrderedTimes": true,
      "wordReconstruction": true,
      "completeNativeTokenDurations": true,
      "wordCount": 22,
      "passed": true,
      "maximumConsecutiveRepeatedWords": 1
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "placement": {
        "mps:0/torch.float32": 651
      },
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "accelerationDetails": "FastConformer encoder/TDT network and generated token/duration tensors on selected device; mel features, token-to-word grouping and JSON on CPU. No implicit MPS fallback.",
      "durationMs": 269.0517919982085,
      "inferenceMs": 152.57791700423695,
      "realTimeFactor": 0.013870719727657905,
      "activationPlacement": {
        "encoderOutput": "mps:0"
      },
      "predictionPlacement": "mps:0",
      "durationPlacement": "mps:0",
      "outputTokens": 47,
      "maxNewTokens": 1024,
      "decoding": "greedy TDT",
      "sampleRate": 16000,
      "driverAllocatedMemoryBytes": 3333914624,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "artifactRevision": "541d1f99c6b0c3cd0b11a95167540bb8edefd82b",
      "totalMs": 269.0517919982085,
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "fixtureId": "jfk",
      "wordErrorRate": 0,
      "finiteOrderedTimes": true,
      "wordReconstruction": true,
      "completeNativeTokenDurations": true,
      "wordCount": 22,
      "passed": true,
      "maximumConsecutiveRepeatedWords": 1
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 2501.905249999254,
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "placement": {
        "mps:0/torch.float16": 651
      },
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "accelerationDetails": "FastConformer encoder/TDT network and generated token/duration tensors on selected device; mel features, token-to-word grouping and JSON on CPU. No implicit MPS fallback.",
      "durationMs": 445.6847909896169,
      "inferenceMs": 323.33162501163315,
      "realTimeFactor": 0.02939378409196665,
      "activationPlacement": {
        "encoderOutput": "mps:0"
      },
      "predictionPlacement": "mps:0",
      "durationPlacement": "mps:0",
      "outputTokens": 47,
      "maxNewTokens": 1024,
      "decoding": "greedy TDT",
      "sampleRate": 16000,
      "driverAllocatedMemoryBytes": 2356625408,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "artifactRevision": "541d1f99c6b0c3cd0b11a95167540bb8edefd82b",
      "totalMs": 2947.590040988871,
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "fixtureId": "synthetic",
      "wordErrorRate": 0,
      "finiteOrderedTimes": true,
      "wordReconstruction": true,
      "completeNativeTokenDurations": true,
      "wordCount": 9,
      "passed": true,
      "maximumConsecutiveRepeatedWords": 1
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "placement": {
        "mps:0/torch.float16": 651
      },
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "accelerationDetails": "FastConformer encoder/TDT network and generated token/duration tensors on selected device; mel features, token-to-word grouping and JSON on CPU. No implicit MPS fallback.",
      "durationMs": 195.53829199867323,
      "inferenceMs": 139.73525000619702,
      "realTimeFactor": 0.04109860294299912,
      "activationPlacement": {
        "encoderOutput": "mps:0"
      },
      "predictionPlacement": "mps:0",
      "durationPlacement": "mps:0",
      "outputTokens": 19,
      "maxNewTokens": 1024,
      "decoding": "greedy TDT",
      "sampleRate": 16000,
      "driverAllocatedMemoryBytes": 2356920320,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "artifactRevision": "541d1f99c6b0c3cd0b11a95167540bb8edefd82b",
      "totalMs": 195.53829199867323,
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "fixtureId": "jfk-shifted",
      "wordErrorRate": 0,
      "finiteOrderedTimes": true,
      "wordReconstruction": true,
      "completeNativeTokenDurations": true,
      "wordCount": 22,
      "passed": true,
      "maximumConsecutiveRepeatedWords": 1,
      "correspondingWordFraction": 1,
      "medianLeadingSilenceShiftErrorSeconds": 0.02000000000000024
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "placement": {
        "mps:0/torch.float16": 651
      },
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "accelerationDetails": "FastConformer encoder/TDT network and generated token/duration tensors on selected device; mel features, token-to-word grouping and JSON on CPU. No implicit MPS fallback.",
      "durationMs": 333.74375000130385,
      "inferenceMs": 212.7033750002738,
      "realTimeFactor": 0.018495945652197723,
      "activationPlacement": {
        "encoderOutput": "mps:0"
      },
      "predictionPlacement": "mps:0",
      "durationPlacement": "mps:0",
      "outputTokens": 48,
      "maxNewTokens": 1024,
      "decoding": "greedy TDT",
      "sampleRate": 16000,
      "driverAllocatedMemoryBytes": 2357313536,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "artifactRevision": "541d1f99c6b0c3cd0b11a95167540bb8edefd82b",
      "totalMs": 333.74375000130385,
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "fixtureId": "silence",
      "wordErrorRate": 0,
      "finiteOrderedTimes": true,
      "wordReconstruction": true,
      "completeNativeTokenDurations": true,
      "wordCount": 0,
      "passed": true,
      "maximumConsecutiveRepeatedWords": 0
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "placement": {
        "mps:0/torch.float16": 651
      },
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "accelerationDetails": "FastConformer encoder/TDT network and generated token/duration tensors on selected device; mel features, token-to-word grouping and JSON on CPU. No implicit MPS fallback.",
      "durationMs": 139.12925000477117,
      "inferenceMs": 128.05391701112967,
      "realTimeFactor": 0.06402695850556483,
      "activationPlacement": {
        "encoderOutput": "mps:0"
      },
      "predictionPlacement": "mps:0",
      "durationPlacement": "mps:0",
      "outputTokens": 20,
      "maxNewTokens": 1024,
      "decoding": "greedy TDT",
      "sampleRate": 16000,
      "driverAllocatedMemoryBytes": 2357608448,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "artifactRevision": "541d1f99c6b0c3cd0b11a95167540bb8edefd82b",
      "totalMs": 139.12925000477117,
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "fixtureId": "jfk",
      "wordErrorRate": 0,
      "finiteOrderedTimes": true,
      "wordReconstruction": true,
      "completeNativeTokenDurations": true,
      "wordCount": 22,
      "passed": true,
      "maximumConsecutiveRepeatedWords": 1
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "placement": {
        "mps:0/torch.float16": 651
      },
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "accelerationDetails": "FastConformer encoder/TDT network and generated token/duration tensors on selected device; mel features, token-to-word grouping and JSON on CPU. No implicit MPS fallback.",
      "durationMs": 264.3546659965068,
      "inferenceMs": 147.49941599438898,
      "realTimeFactor": 0.013409037817671726,
      "activationPlacement": {
        "encoderOutput": "mps:0"
      },
      "predictionPlacement": "mps:0",
      "durationPlacement": "mps:0",
      "outputTokens": 47,
      "maxNewTokens": 1024,
      "decoding": "greedy TDT",
      "sampleRate": 16000,
      "driverAllocatedMemoryBytes": 2357608448,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "artifactRevision": "541d1f99c6b0c3cd0b11a95167540bb8edefd82b",
      "totalMs": 264.3546659965068,
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "fixtureId": "jfk",
      "wordErrorRate": 0,
      "finiteOrderedTimes": true,
      "wordReconstruction": true,
      "completeNativeTokenDurations": true,
      "wordCount": 22,
      "passed": true,
      "maximumConsecutiveRepeatedWords": 1
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 2411.880582993035,
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "placement": {
        "mps:0/torch.float16": 651
      },
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "accelerationDetails": "FastConformer encoder/TDT network and generated token/duration tensors on selected device; mel features, token-to-word grouping and JSON on CPU. No implicit MPS fallback.",
      "durationMs": 466.5447079896694,
      "inferenceMs": 342.26600000692997,
      "realTimeFactor": 0.031115090909720908,
      "activationPlacement": {
        "encoderOutput": "mps:0"
      },
      "predictionPlacement": "mps:0",
      "durationPlacement": "mps:0",
      "outputTokens": 47,
      "maxNewTokens": 1024,
      "decoding": "greedy TDT",
      "sampleRate": 16000,
      "driverAllocatedMemoryBytes": 2356625408,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "artifactRevision": "541d1f99c6b0c3cd0b11a95167540bb8edefd82b",
      "totalMs": 2878.4252909827046,
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "fixtureId": "jfk",
      "wordErrorRate": 0,
      "finiteOrderedTimes": true,
      "wordReconstruction": true,
      "completeNativeTokenDurations": true,
      "wordCount": 22,
      "passed": true,
      "maximumConsecutiveRepeatedWords": 1
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "placement": {
        "mps:0/torch.float16": 651
      },
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "accelerationDetails": "FastConformer encoder/TDT network and generated token/duration tensors on selected device; mel features, token-to-word grouping and JSON on CPU. No implicit MPS fallback.",
      "durationMs": 263.8829170027748,
      "inferenceMs": 144.07579200633336,
      "realTimeFactor": 0.013097799273303033,
      "activationPlacement": {
        "encoderOutput": "mps:0"
      },
      "predictionPlacement": "mps:0",
      "durationPlacement": "mps:0",
      "outputTokens": 47,
      "maxNewTokens": 1024,
      "decoding": "greedy TDT",
      "sampleRate": 16000,
      "driverAllocatedMemoryBytes": 2356609024,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "artifactRevision": "541d1f99c6b0c3cd0b11a95167540bb8edefd82b",
      "totalMs": 263.8829170027748,
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "fixtureId": "jfk",
      "wordErrorRate": 0,
      "finiteOrderedTimes": true,
      "wordReconstruction": true,
      "completeNativeTokenDurations": true,
      "wordCount": 22,
      "passed": true,
      "maximumConsecutiveRepeatedWords": 1
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 2495.5064169917023,
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "placement": {
        "mps:0/torch.float16": 651
      },
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "accelerationDetails": "FastConformer encoder/TDT network and generated token/duration tensors on selected device; mel features, token-to-word grouping and JSON on CPU. No implicit MPS fallback.",
      "durationMs": 468.37000000232365,
      "inferenceMs": 343.0703330086544,
      "realTimeFactor": 0.031188212091695856,
      "activationPlacement": {
        "encoderOutput": "mps:0"
      },
      "predictionPlacement": "mps:0",
      "durationPlacement": "mps:0",
      "outputTokens": 47,
      "maxNewTokens": 1024,
      "decoding": "greedy TDT",
      "sampleRate": 16000,
      "driverAllocatedMemoryBytes": 2356625408,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "artifactRevision": "541d1f99c6b0c3cd0b11a95167540bb8edefd82b",
      "totalMs": 2963.876416994026,
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "fixtureId": "jfk",
      "wordErrorRate": 0,
      "finiteOrderedTimes": true,
      "wordReconstruction": true,
      "completeNativeTokenDurations": true,
      "wordCount": 22,
      "passed": true,
      "maximumConsecutiveRepeatedWords": 1
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "placement": {
        "mps:0/torch.float16": 651
      },
      "checkpointLoading": {
        "missing_keys": [],
        "unexpected_keys": [],
        "mismatched_keys": [],
        "error_msgs": []
      },
      "accelerationDetails": "FastConformer encoder/TDT network and generated token/duration tensors on selected device; mel features, token-to-word grouping and JSON on CPU. No implicit MPS fallback.",
      "durationMs": 296.9959579932038,
      "inferenceMs": 177.57945900666527,
      "realTimeFactor": 0.016143587182424115,
      "activationPlacement": {
        "encoderOutput": "mps:0"
      },
      "predictionPlacement": "mps:0",
      "durationPlacement": "mps:0",
      "outputTokens": 47,
      "maxNewTokens": 1024,
      "decoding": "greedy TDT",
      "sampleRate": 16000,
      "driverAllocatedMemoryBytes": 2356609024,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "artifactRevision": "541d1f99c6b0c3cd0b11a95167540bb8edefd82b",
      "totalMs": 296.9959579932038,
      "fallback": false,
      "grade": "fast"
    }
  }
]

Something not quite right?

Tell us what happened with Parakeet TDT 3 Desktop. 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.