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.

DETR Desktop

Separate common objects and background into a full-resolution panoptic segment-ID map, segment classes and scores, and detection boxes.

Desktop inferencePrivate · Runs on your computer
About this model

Separate common objects and background into a full-resolution panoptic segment-ID map, segment classes and scores, and detection boxes.

Created by
Meta
License
Apache-2.0
Model released
Unknown
Source repository created
Mar 2, 2022
Model weights · FP32
172.2 MB

Best for

  • Review common objects and their spatial locations in photos.
  • Separate objects and background regions into class-labeled pixel segments.

Keep in mind

  • Small controlled person/dog/blank fixture suite only; broad COCO accuracy unverified. Small, crowded, occluded, unusual or blurred objects may be missed or mislabeled.
  • Confidence is not a correctness guarantee. Dataset bias and unsafe errors make this unsuitable for safety-critical or high-impact decisions.
  • PNG/JPEG still images only;20MB encoded,20megapixels,16–10000pixels per edge. EXIF orientation applied and alpha composited onto white.
  • Original2020DETR family explicitly requested via queue; recent format updates do not make this a recent original checkpoint.
  • Windows, Intel Mac and other OS/hardware unverified. No incremental streaming: complete query/mask outputs arrive after one forward pass.
  • Panoptic CPU configurations fail the frozen5second usability gate; no CPU recovery is advertised for this checkpoint.

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/facebook/detr-resnet-50-panoptic
Release verification
Unknown
Release evidence
Original2020DETR family explicitly requested. Exact checkpoint release day not established by bounded official repository/card checks; do not infer from paper, conversion or repository creation dates.
Source updated
Apr 10, 2024
Native model files
https://huggingface.co/facebook/detr-resnet-50-panoptic
Files updated
Apr 10, 2024
Artifact revision
d53b52a799403a8867920f82c869e40732b47037

Desktop app

Sample result

DETR ResNet-50-panoptic Desktop

Run DETR 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 SiliconMPS / FP32Tested16.0 GiBRAM estimate
macOS Apple SiliconMPS / FP16Tested16.0 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.

DETR ResNet-50-panoptic Desktop

48 configurations tested
Passed42
Failed6
TestedNov 11, 2026
Runtimetransformers 5.22.1

Quality criteria

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

PrecisionDeviceResultPlatformInference / load
FP16mpsAcceleratedPassedFastmacos · aarch64279 ms / 1022 ms
FP32mpsAcceleratedPassedFastmacos · aarch64291 ms / 1138 ms
FP32mpsAcceleratedPassedFastmacos · aarch64291 ms / 0 ms
FP16mpsAcceleratedPassedFastmacos · aarch64323 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64336 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64338 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64344 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64345 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64347 ms / 0 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64351 ms / 0 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64352 ms / 0 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64352 ms / 0 ms
FP16mpsAcceleratedPassedModeratemacos · aarch64352 ms / 0 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64353 ms / 0 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64355 ms / 0 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64356 ms / 0 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64359 ms / 0 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64359 ms / 0 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64359 ms / 0 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64362 ms / 0 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64364 ms / 0 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64367 ms / 0 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64367 ms / 0 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64372 ms / 0 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64373 ms / 0 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64375 ms / 0 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64379 ms / 0 ms
FP16mpsAcceleratedPassedModeratemacos · aarch64379 ms / 0 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64400 ms / 1074 ms
FP16mpsAcceleratedPassedModeratemacos · aarch64405 ms / 1022 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64419 ms / 1138 ms
FP16mpsAcceleratedPassedModeratemacos · aarch64420 ms / 0 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64436 ms / 0 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64442 ms / 0 ms
FP16mpsAcceleratedPassedModeratemacos · aarch64473 ms / 0 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64516 ms / 1115 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64516 ms / 1105 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64523 ms / 1138 ms
FP16mpsAcceleratedPassedModeratemacos · aarch64526 ms / 1022 ms
FP16mpsAcceleratedPassedModeratemacos · aarch64538 ms / 1058 ms
FP16mpsAcceleratedPassedModeratemacos · aarch64553 ms / 1054 ms
FP16mpsAcceleratedPassedModeratemacos · aarch64554 ms / 1079 ms
FP32cpuFailed—macos · aarch6422104 ms / 24069 ms
FP32cpuFailed—macos · aarch6427017 ms / 24069 ms
FP32cpuFailed—macos · aarch6422057 ms / 24069 ms
FP32cpuFailed—macos · aarch646297 ms / 7230 ms
FP32cpuFailed—macos · aarch647666 ms / 7230 ms
FP32cpuFailed—macos · aarch646237 ms / 7230 ms
Test setup & sample details
Quality gate
No explicit quality gate recorded. Review the per-case quality evidence; full certification is pending.
Tested
2026-11-11T22:08:02.760596+00:00
Environment
{ "os": "macos", "arch": "aarch64", "hardware": "Apple M3 Max", "osVersion": "27.2", "runtime": "Transformers5.22.1/Torch2.10.0/Torchvision0.25.0/timm1.0.22", "interpreter": "App-owned CPython3.13.18 private managed environment", "transport": "Local immutable original snapshots; downloads excluded from speed measurements." }
Runtime
transformers 5.22.1
Artifact revision
d53b52a799403a8867920f82c869e40732b47037
Fixture
{
  "file": "model-samples/desktop/mobilenet-dog.jpg",
  "size": 83281,
  "sha1": "f1d042827c2f2154a6d8043b8e8f07c1be612941",
  "sha256": "d4cb12ea8a768582562e1f1c0c930ea6fc4757f5b66551b986b41498efd75ca1",
  "url": "https://assets.localai.tools/samples/v1/files/d4cb12ea8a768582562e1f1c0c930ea6fc4757f5b66551b986b41498efd75ca1/mobilenet-dog.jpg"
}
Rerun when
Artifact/runtime revision; Preprocessing/postprocessing/presentation; Fixtures or frozen gates; Relevant platform/backend behavior
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "fail",
    "quality": {
      "maximumIoU": 0.9729969809260902,
      "passed": true,
      "interiorPointPassed": [
        true,
        true,
        true
      ],
      "segmentMapComplete": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 24068.784375005635,
      "placement": {
        "cpu": 350
      },
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 25673.869374993956,
      "inferenceMs": 22103.98733301554,
      "durationMs": 22152.047582989326,
      "imagesPerSecond": 0.04524070634560823,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "maskDevice": "cpu",
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 13679165440,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "totalMs": 46220.83195799496,
      "grade": "slow",
      "fallback": false,
      "artifactRevision": "d53b52a799403a8867920f82c869e40732b47037",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "fail",
    "quality": {
      "maximumIoU": 0.8975541998315539,
      "passed": true,
      "interiorPointPassed": [
        true,
        true,
        true
      ],
      "segmentMapComplete": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 24068.784375005635,
      "placement": {
        "cpu": 350
      },
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 25673.869374993956,
      "inferenceMs": 27017.048250010703,
      "durationMs": 27104.527457995573,
      "imagesPerSecond": 0.037013665991420946,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "maskDevice": "cpu",
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        1199
      ],
      "peakHelperRssBytes": 16304635904,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "totalMs": 51173.31183300121,
      "grade": "slow",
      "fallback": false,
      "artifactRevision": "d53b52a799403a8867920f82c869e40732b47037",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "fail",
    "quality": {
      "maximumIoU": 0,
      "passed": true,
      "interiorPointPassed": [],
      "segmentMapComplete": true,
      "correction": "Blank panoptic may contain background stuff; frozen panoptic negative policy prohibits person/dog, not stuff classes. Regraded complete retained output without rerunning or changing measurement time."
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 24068.784375005635,
      "placement": {
        "cpu": 350
      },
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 25673.869374993956,
      "inferenceMs": 22056.67925000307,
      "durationMs": 22076.718541997252,
      "imagesPerSecond": 0.045337740494179816,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "maskDevice": "cpu",
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 16304635904,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "totalMs": 46145.50291700289,
      "grade": "slow",
      "fallback": false,
      "artifactRevision": "d53b52a799403a8867920f82c869e40732b47037",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "fail",
    "quality": {
      "maximumIoU": 0.9729969809260902,
      "passed": true,
      "interiorPointPassed": [
        true,
        true,
        true
      ],
      "segmentMapComplete": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 7230.432334006764,
      "placement": {
        "cpu": 350
      },
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 9787.328292004531,
      "inferenceMs": 6297.431624989258,
      "durationMs": 6365.730250021443,
      "imagesPerSecond": 0.15879489600678368,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "maskDevice": "cpu",
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 14459944960,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 13596.162584028207,
      "grade": "slow",
      "fallback": false,
      "artifactRevision": "d53b52a799403a8867920f82c869e40732b47037",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "fail",
    "quality": {
      "maximumIoU": 0.8975541998315539,
      "passed": true,
      "interiorPointPassed": [
        true,
        true,
        true
      ],
      "segmentMapComplete": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 7230.432334006764,
      "placement": {
        "cpu": 350
      },
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 9787.328292004531,
      "inferenceMs": 7666.031208005734,
      "durationMs": 7776.026583975181,
      "imagesPerSecond": 0.1304455947108182,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "maskDevice": "cpu",
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        1199
      ],
      "peakHelperRssBytes": 16361717760,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 15006.458917981945,
      "grade": "slow",
      "fallback": false,
      "artifactRevision": "d53b52a799403a8867920f82c869e40732b47037",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "fail",
    "quality": {
      "maximumIoU": 0,
      "passed": true,
      "interiorPointPassed": [],
      "segmentMapComplete": true,
      "correction": "Blank panoptic may contain background stuff; frozen panoptic negative policy prohibits person/dog, not stuff classes. Regraded complete retained output without rerunning or changing measurement time."
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 7230.432334006764,
      "placement": {
        "cpu": 350
      },
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 9787.328292004531,
      "inferenceMs": 6237.091083981795,
      "durationMs": 6253.498208010569,
      "imagesPerSecond": 0.16033115222065897,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "maskDevice": "cpu",
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 16361717760,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 13483.930542017333,
      "grade": "slow",
      "fallback": false,
      "artifactRevision": "d53b52a799403a8867920f82c869e40732b47037",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "maximumIoU": 0.9729969809260902,
      "passed": true,
      "interiorPointPassed": [
        true,
        true,
        true
      ],
      "segmentMapComplete": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1138.4890410117805,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 3527.9383340093773,
      "inferenceMs": 418.8126660010312,
      "durationMs": 520.1164160098415,
      "imagesPerSecond": 2.3877023814688974,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 699613184,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5016371200,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 1658.605457021622,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "d53b52a799403a8867920f82c869e40732b47037",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "maximumIoU": 0.8975541918516228,
      "passed": true,
      "interiorPointPassed": [
        true,
        true,
        true
      ],
      "segmentMapComplete": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1138.4890410117805,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 3527.9383340093773,
      "inferenceMs": 522.6321250083856,
      "durationMs": 598.8129579927772,
      "imagesPerSecond": 1.9133917571254446,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        1199
      ],
      "peakHelperRssBytes": 829308928,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 7447560192,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 1737.3019990045577,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "d53b52a799403a8867920f82c869e40732b47037",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "maximumIoU": 0,
      "passed": true,
      "interiorPointPassed": [],
      "segmentMapComplete": true,
      "correction": "Blank panoptic may contain background stuff; frozen panoptic negative policy prohibits person/dog, not stuff classes. Regraded complete retained output without rerunning or changing measurement time."
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1138.4890410117805,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 3527.9383340093773,
      "inferenceMs": 290.77950000646524,
      "durationMs": 322.3232080054004,
      "imagesPerSecond": 3.439031981201446,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 832749568,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 7449182208,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 1460.8122490171809,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "d53b52a799403a8867920f82c869e40732b47037",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "maximumIoU": 0.9718833839084553,
      "passed": true,
      "interiorPointPassed": [
        true,
        true,
        true
      ],
      "segmentMapComplete": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1022.4807500198949,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float16",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2596.3787919899914,
      "inferenceMs": 404.5948330021929,
      "durationMs": 490.8867499907501,
      "imagesPerSecond": 2.4716084300428522,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 768622592,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 4580163584,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 1513.367500010645,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "d53b52a799403a8867920f82c869e40732b47037",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "maximumIoU": 0.8970879844124954,
      "passed": true,
      "interiorPointPassed": [
        true,
        true,
        true
      ],
      "segmentMapComplete": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1022.4807500198949,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float16",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2596.3787919899914,
      "inferenceMs": 526.1461659974884,
      "durationMs": 601.389499992365,
      "imagesPerSecond": 1.9006125381606858,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        1199
      ],
      "peakHelperRssBytes": 895811584,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 7915225088,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 1623.87025001226,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "d53b52a799403a8867920f82c869e40732b47037",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "maximumIoU": 0,
      "passed": true,
      "interiorPointPassed": [],
      "segmentMapComplete": true,
      "correction": "Blank panoptic may contain background stuff; frozen panoptic negative policy prohibits person/dog, not stuff classes. Regraded complete retained output without rerunning or changing measurement time."
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1022.4807500198949,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float16",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2596.3787919899914,
      "inferenceMs": 279.2307080235332,
      "durationMs": 310.8791249978822,
      "imagesPerSecond": 3.581268002642894,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 898252800,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 7916847104,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 1333.359875017777,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "d53b52a799403a8867920f82c869e40732b47037",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9729969809260902
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1074.1282910166774,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2683.633375010686,
      "inferenceMs": 399.5679579966236,
      "durationMs": 515.00354098971,
      "imagesPerSecond": 2.5027031822417807,
      "visualizationMs": 33.50037499330938,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 700530688,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5016371200,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 1589.1318320063874,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "d53b52a799403a8867920f82c869e40732b47037",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9729969809260902
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2683.633375010686,
      "inferenceMs": 290.8374590042513,
      "durationMs": 320.6262500025332,
      "imagesPerSecond": 3.4383466401602094,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 701415424,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5016371200,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 320.6262500025332,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "d53b52a799403a8867920f82c869e40732b47037",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9729969809260902
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2683.633375010686,
      "inferenceMs": 362.21162500442006,
      "durationMs": 395.28920900193043,
      "imagesPerSecond": 2.760816967119145,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 704610304,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5016371200,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 395.28920900193043,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "d53b52a799403a8867920f82c869e40732b47037",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9729969809260902
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2683.633375010686,
      "inferenceMs": 354.8259999952279,
      "durationMs": 385.8476250024978,
      "imagesPerSecond": 2.818282763984176,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 712327168,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5016371200,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 385.8476250024978,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "d53b52a799403a8867920f82c869e40732b47037",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9729969809260902
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2683.633375010686,
      "inferenceMs": 364.05091697815806,
      "durationMs": 397.60574998217635,
      "imagesPerSecond": 2.746868510318838,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 713523200,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5016371200,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 397.60574998217635,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "d53b52a799403a8867920f82c869e40732b47037",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9729969809260902
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2683.633375010686,
      "inferenceMs": 350.7348330167588,
      "durationMs": 383.4850419953,
      "imagesPerSecond": 2.8511567881602966,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 713539584,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5016371200,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 383.4850419953,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "d53b52a799403a8867920f82c869e40732b47037",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9729969809260902
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2683.633375010686,
      "inferenceMs": 344.8120000248309,
      "durationMs": 377.2632500040345,
      "imagesPerSecond": 2.900131085716237,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 713539584,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5016371200,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 377.2632500040345,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "d53b52a799403a8867920f82c869e40732b47037",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9729969809260902
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2683.633375010686,
      "inferenceMs": 441.74595799995586,
      "durationMs": 475.9158749948256,
      "imagesPerSecond": 2.263744538891966,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 713572352,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5016371200,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 475.9158749948256,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "d53b52a799403a8867920f82c869e40732b47037",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9729969809260902
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2683.633375010686,
      "inferenceMs": 436.25337499543093,
      "durationMs": 468.72762497514486,
      "imagesPerSecond": 2.292245876631839,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 713588736,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5016371200,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 468.72762497514486,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "d53b52a799403a8867920f82c869e40732b47037",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9729969809260902
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2683.633375010686,
      "inferenceMs": 358.98349998751655,
      "durationMs": 392.4323329993058,
      "imagesPerSecond": 2.7856433513929595,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 713588736,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5016371200,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 392.4323329993058,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "d53b52a799403a8867920f82c869e40732b47037",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9729969809260902
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2683.633375010686,
      "inferenceMs": 336.16558302310295,
      "durationMs": 370.5452079884708,
      "imagesPerSecond": 2.9747245122689288,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 713605120,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5016371200,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 370.5452079884708,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "d53b52a799403a8867920f82c869e40732b47037",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9729969809260902
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2683.633375010686,
      "inferenceMs": 337.6058340072632,
      "durationMs": 372.0990419969894,
      "imagesPerSecond": 2.962034121657051,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 713605120,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5016371200,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 372.0990419969894,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "d53b52a799403a8867920f82c869e40732b47037",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9729969809260902
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2683.633375010686,
      "inferenceMs": 355.8237499964889,
      "durationMs": 389.9957079847809,
      "imagesPerSecond": 2.8103801390712886,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 713621504,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5016371200,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 389.9957079847809,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "d53b52a799403a8867920f82c869e40732b47037",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9729969809260902
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2683.633375010686,
      "inferenceMs": 343.6145420128014,
      "durationMs": 375.85404198034666,
      "imagesPerSecond": 2.9102377161987074,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 713621504,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5016371200,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 375.85404198034666,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "d53b52a799403a8867920f82c869e40732b47037",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9729969809260902
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2683.633375010686,
      "inferenceMs": 366.6062500269618,
      "durationMs": 400.64608299871907,
      "imagesPerSecond": 2.7277221812952064,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 713621504,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5016371200,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 400.64608299871907,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "d53b52a799403a8867920f82c869e40732b47037",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9729969809260902
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2683.633375010686,
      "inferenceMs": 352.0097089931369,
      "durationMs": 384.1876669903286,
      "imagesPerSecond": 2.8408307340735792,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 713654272,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5016371200,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 384.1876669903286,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "d53b52a799403a8867920f82c869e40732b47037",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9729969809260902
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2683.633375010686,
      "inferenceMs": 359.0848339954391,
      "durationMs": 391.8136249994859,
      "imagesPerSecond": 2.784857240761946,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 713654272,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5016371200,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 391.8136249994859,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "d53b52a799403a8867920f82c869e40732b47037",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9729969809260902
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2683.633375010686,
      "inferenceMs": 378.96649999311194,
      "durationMs": 412.69729199120775,
      "imagesPerSecond": 2.6387556684249818,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 713654272,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5016371200,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 412.69729199120775,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "d53b52a799403a8867920f82c869e40732b47037",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9729969809260902
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2683.633375010686,
      "inferenceMs": 347.4635839811526,
      "durationMs": 380.0893329898827,
      "imagesPerSecond": 2.8779994396599635,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 713670656,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5016371200,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 380.0893329898827,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "d53b52a799403a8867920f82c869e40732b47037",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9729969809260902
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2683.633375010686,
      "inferenceMs": 352.8566670138389,
      "durationMs": 386.526624992257,
      "imagesPerSecond": 2.8340119189551274,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 713687040,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5016371200,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 386.526624992257,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "d53b52a799403a8867920f82c869e40732b47037",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9729969809260902
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2683.633375010686,
      "inferenceMs": 374.92437500623055,
      "durationMs": 406.75470800488256,
      "imagesPerSecond": 2.667204552873848,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 713703424,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5016371200,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 406.75470800488256,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "d53b52a799403a8867920f82c869e40732b47037",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9729969809260902
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2683.633375010686,
      "inferenceMs": 366.5748749917839,
      "durationMs": 399.140791996615,
      "imagesPerSecond": 2.727955646230291,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 713703424,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5016371200,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 399.140791996615,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "d53b52a799403a8867920f82c869e40732b47037",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9729969809260902
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2683.633375010686,
      "inferenceMs": 358.9108329906594,
      "durationMs": 392.10699999239296,
      "imagesPerSecond": 2.7862073475670903,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 713703424,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5016371200,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 392.10699999239296,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "d53b52a799403a8867920f82c869e40732b47037",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9729969809260902
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2683.633375010686,
      "inferenceMs": 372.351666999748,
      "durationMs": 404.3790840078145,
      "imagesPerSecond": 2.685633202766558,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 713719808,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5016371200,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 404.3790840078145,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "d53b52a799403a8867920f82c869e40732b47037",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9729969809260902
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2683.633375010686,
      "inferenceMs": 352.05170800327323,
      "durationMs": 383.7524999980815,
      "imagesPerSecond": 2.8404918290886476,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 713719808,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5016371200,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 383.7524999980815,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "d53b52a799403a8867920f82c869e40732b47037",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9729969809260902
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2683.633375010686,
      "inferenceMs": 373.3421250071842,
      "durationMs": 409.884749999037,
      "imagesPerSecond": 2.6785083520397195,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 713736192,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5016371200,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 409.884749999037,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "d53b52a799403a8867920f82c869e40732b47037",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9729969809260902
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1105.4218749923166,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2802.54849998164,
      "inferenceMs": 516.2557089934126,
      "durationMs": 630.9501250216272,
      "imagesPerSecond": 1.9370245840957083,
      "visualizationMs": 32.099165982799605,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 700186624,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5016371200,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 1736.3720000139438,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "d53b52a799403a8867920f82c869e40732b47037",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9729969809260902
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1114.7751250246074,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2684.329208015697,
      "inferenceMs": 516.0337499983143,
      "durationMs": 633.6451660026796,
      "imagesPerSecond": 1.9378577467137887,
      "visualizationMs": 33.867041987832636,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 702955520,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5016371200,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 1748.420291027287,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "d53b52a799403a8867920f82c869e40732b47037",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9718833839084553
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1057.9079589806497,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float16",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2628.019708005013,
      "inferenceMs": 538.1260839931201,
      "durationMs": 659.3345000001136,
      "imagesPerSecond": 1.8583005539883566,
      "visualizationMs": 29.7512499964796,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 764805120,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 4580163584,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 1717.2424589807633,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "d53b52a799403a8867920f82c869e40732b47037",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9718833839084553
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float16",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2628.019708005013,
      "inferenceMs": 322.5272499839775,
      "durationMs": 353.5564160265494,
      "imagesPerSecond": 3.1005132126035178,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 765689856,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 4580163584,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 353.5564160265494,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "d53b52a799403a8867920f82c869e40732b47037",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9718833839084553
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float16",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2628.019708005013,
      "inferenceMs": 379.4354169804137,
      "durationMs": 413.9434159733355,
      "imagesPerSecond": 2.635494619764553,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 765755392,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 4580163584,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 413.9434159733355,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "d53b52a799403a8867920f82c869e40732b47037",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9718833839084553
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float16",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2628.019708005013,
      "inferenceMs": 473.0714579927735,
      "durationMs": 506.8032499984838,
      "imagesPerSecond": 2.1138455578000137,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 765804544,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 4580163584,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 506.8032499984838,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "d53b52a799403a8867920f82c869e40732b47037",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9718833839084553
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float16",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2628.019708005013,
      "inferenceMs": 419.7291669843253,
      "durationMs": 455.72874997742474,
      "imagesPerSecond": 2.3824887061931173,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 777404416,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 4580163584,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 455.72874997742474,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "d53b52a799403a8867920f82c869e40732b47037",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9718833839084553
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float16",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2628.019708005013,
      "inferenceMs": 352.30254201451316,
      "durationMs": 385.3774580056779,
      "imagesPerSecond": 2.8384694424339543,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 778633216,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 4580163584,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 385.3774580056779,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "d53b52a799403a8867920f82c869e40732b47037",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9718833839084553
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1079.4388329959475,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float16",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2674.1449169931,
      "inferenceMs": 553.9042910095304,
      "durationMs": 663.4479579806793,
      "imagesPerSecond": 1.8053660465013335,
      "visualizationMs": 29.592041013529524,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 767082496,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 4580163584,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 1742.8867909766268,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "d53b52a799403a8867920f82c869e40732b47037",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9718833839084553
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1054.032540996559,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float16",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2722.9063749837223,
      "inferenceMs": 552.5503749959171,
      "durationMs": 670.1685829902999,
      "imagesPerSecond": 1.8097897409035135,
      "visualizationMs": 32.273582997731864,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 778633216,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 4580163584,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 1724.2011239868589,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "d53b52a799403a8867920f82c869e40732b47037",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  }
]

DETR ResNet-101-panoptic Desktop

48 configurations tested
Passed42
Failed6
TestedNov 11, 2026
Runtimetransformers 5.22.1

Quality criteria

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

PrecisionDeviceResultPlatformInference / load
FP16mpsAcceleratedPassedFastmacos · aarch64308 ms / 0 ms
FP16mpsAcceleratedPassedFastmacos · aarch64309 ms / 0 ms
FP16mpsAcceleratedPassedFastmacos · aarch64310 ms / 0 ms
FP16mpsAcceleratedPassedFastmacos · aarch64310 ms / 0 ms
FP16mpsAcceleratedPassedFastmacos · aarch64312 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64321 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64323 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64323 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64323 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64323 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64324 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64324 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64324 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64324 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64324 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64324 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64324 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64325 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64325 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64325 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64325 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64325 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64325 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64326 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64326 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64327 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64327 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64328 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64328 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64330 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64337 ms / 2288 ms
FP16mpsAcceleratedPassedModeratemacos · aarch64424 ms / 1052 ms
FP16mpsAcceleratedPassedModeratemacos · aarch64424 ms / 1060 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64432 ms / 1106 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64436 ms / 1262 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64436 ms / 1146 ms
FP16mpsAcceleratedPassedModeratemacos · aarch64453 ms / 1070 ms
FP16mpsAcceleratedPassedModeratemacos · aarch64472 ms / 1781 ms
FP16mpsAcceleratedPassedModeratemacos · aarch64587 ms / 1781 ms
FP16mpsAcceleratedPassedModeratemacos · aarch64909 ms / 1781 ms
FP32mpsAcceleratedPassedModeratemacos · aarch641297 ms / 2288 ms
FP32mpsAcceleratedPassedSlowmacos · aarch641673 ms / 2288 ms
FP32cpuFailed—macos · aarch6421937 ms / 24579 ms
FP32cpuFailed—macos · aarch6427050 ms / 24579 ms
FP32cpuFailed—macos · aarch6422119 ms / 24579 ms
FP32cpuFailed—macos · aarch646415 ms / 7340 ms
FP32cpuFailed—macos · aarch647718 ms / 7340 ms
FP32cpuFailed—macos · aarch646464 ms / 7340 ms
Test setup & sample details
Quality gate
No explicit quality gate recorded. Review the per-case quality evidence; full certification is pending.
Tested
2026-11-11T22:05:43.542795+00:00
Environment
{ "os": "macos", "arch": "aarch64", "hardware": "Apple M3 Max", "osVersion": "27.2", "runtime": "Transformers5.22.1/Torch2.10.0/Torchvision0.25.0/timm1.0.22", "interpreter": "App-owned CPython3.13.18 private managed environment", "transport": "Local immutable original snapshots; downloads excluded from speed measurements." }
Runtime
transformers 5.22.1
Artifact revision
8ac2b64b1bb2b5927a2c38ec6c74dccc25693e9e
Fixture
{
  "file": "model-samples/desktop/mobilenet-dog.jpg",
  "size": 83281,
  "sha1": "f1d042827c2f2154a6d8043b8e8f07c1be612941",
  "sha256": "d4cb12ea8a768582562e1f1c0c930ea6fc4757f5b66551b986b41498efd75ca1",
  "url": "https://assets.localai.tools/samples/v1/files/d4cb12ea8a768582562e1f1c0c930ea6fc4757f5b66551b986b41498efd75ca1/mobilenet-dog.jpg"
}
Rerun when
Artifact/runtime revision; Preprocessing/postprocessing/presentation; Fixtures or frozen gates; Relevant platform/backend behavior
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "fail",
    "quality": {
      "maximumIoU": 0.98573333976876,
      "passed": true,
      "interiorPointPassed": [
        true,
        true,
        true
      ],
      "segmentMapComplete": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 24578.952124982607,
      "placement": {
        "cpu": 401
      },
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 26200.967290991684,
      "inferenceMs": 21937.065583013464,
      "durationMs": 21983.633083000313,
      "imagesPerSecond": 0.045584948279241616,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 13760348160,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "totalMs": 46562.58520798292,
      "grade": "slow",
      "fallback": false,
      "artifactRevision": "8ac2b64b1bb2b5927a2c38ec6c74dccc25693e9e",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "fail",
    "quality": {
      "maximumIoU": 0.9186241859465498,
      "passed": true,
      "interiorPointPassed": [
        true,
        true,
        true
      ],
      "segmentMapComplete": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 24578.952124982607,
      "placement": {
        "cpu": 401
      },
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 26200.967290991684,
      "inferenceMs": 27049.594000010984,
      "durationMs": 27133.541457995307,
      "imagesPerSecond": 0.03696913158842953,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        1199
      ],
      "peakHelperRssBytes": 16702128128,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "totalMs": 51712.49358297791,
      "grade": "slow",
      "fallback": false,
      "artifactRevision": "8ac2b64b1bb2b5927a2c38ec6c74dccc25693e9e",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "fail",
    "quality": {
      "maximumIoU": 0,
      "passed": true,
      "interiorPointPassed": [],
      "segmentMapComplete": true,
      "correction": "Blank panoptic may contain background stuff; frozen panoptic negative policy prohibits person/dog, not stuff classes. Regraded complete retained output without rerunning or changing measurement time."
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 24578.952124982607,
      "placement": {
        "cpu": 401
      },
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 26200.967290991684,
      "inferenceMs": 22118.85812500259,
      "durationMs": 22138.36291700136,
      "imagesPerSecond": 0.04521029043852972,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 16702128128,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "totalMs": 46717.315041983966,
      "grade": "slow",
      "fallback": false,
      "artifactRevision": "8ac2b64b1bb2b5927a2c38ec6c74dccc25693e9e",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "fail",
    "quality": {
      "maximumIoU": 0.98573333976876,
      "passed": true,
      "interiorPointPassed": [
        true,
        true,
        true
      ],
      "segmentMapComplete": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 7340.006959013408,
      "placement": {
        "cpu": 401
      },
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 9635.772334004287,
      "inferenceMs": 6414.762791013345,
      "durationMs": 6484.2757919977885,
      "imagesPerSecond": 0.1558904097593341,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 14547255296,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 13824.282751011197,
      "grade": "slow",
      "fallback": false,
      "artifactRevision": "8ac2b64b1bb2b5927a2c38ec6c74dccc25693e9e",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "fail",
    "quality": {
      "maximumIoU": 0.9186241859465498,
      "passed": true,
      "interiorPointPassed": [
        true,
        true,
        true
      ],
      "segmentMapComplete": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 7340.006959013408,
      "placement": {
        "cpu": 401
      },
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 9635.772334004287,
      "inferenceMs": 7718.2870830001775,
      "durationMs": 7831.146917014848,
      "imagesPerSecond": 0.12956242612464342,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        1199
      ],
      "peakHelperRssBytes": 16451387392,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 15171.153876028256,
      "grade": "slow",
      "fallback": false,
      "artifactRevision": "8ac2b64b1bb2b5927a2c38ec6c74dccc25693e9e",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "fail",
    "quality": {
      "maximumIoU": 0,
      "passed": true,
      "interiorPointPassed": [],
      "segmentMapComplete": true,
      "correction": "Blank panoptic may contain background stuff; frozen panoptic negative policy prohibits person/dog, not stuff classes. Regraded complete retained output without rerunning or changing measurement time."
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 7340.006959013408,
      "placement": {
        "cpu": 401
      },
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 9635.772334004287,
      "inferenceMs": 6464.179500006139,
      "durationMs": 6480.992458004039,
      "imagesPerSecond": 0.15469867444105634,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 16451387392,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 13820.999417017447,
      "grade": "slow",
      "fallback": false,
      "artifactRevision": "8ac2b64b1bb2b5927a2c38ec6c74dccc25693e9e",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "maximumIoU": 0.98573333976876,
      "passed": true,
      "interiorPointPassed": [
        true,
        true,
        true
      ],
      "segmentMapComplete": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 2288.115999981528,
      "placement": {
        "mps:0": 401
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 4552.934582985472,
      "inferenceMs": 1296.9478750019334,
      "durationMs": 1554.1407919954509,
      "imagesPerSecond": 0.771041010417253,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 698155008,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5091868672,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 3842.256791976979,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "8ac2b64b1bb2b5927a2c38ec6c74dccc25693e9e",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "maximumIoU": 0.918624408081841,
      "passed": true,
      "interiorPointPassed": [
        true,
        true,
        true
      ],
      "segmentMapComplete": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 2288.115999981528,
      "placement": {
        "mps:0": 401
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 4552.934582985472,
      "inferenceMs": 1673.0308329861145,
      "durationMs": 1760.5857079906855,
      "imagesPerSecond": 0.5977176154100798,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        1199
      ],
      "peakHelperRssBytes": 829997056,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 7514669056,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 4048.7017079722136,
      "grade": "slow",
      "fallback": false,
      "artifactRevision": "8ac2b64b1bb2b5927a2c38ec6c74dccc25693e9e",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "maximumIoU": 0,
      "passed": true,
      "interiorPointPassed": [],
      "segmentMapComplete": true,
      "correction": "Blank panoptic may contain background stuff; frozen panoptic negative policy prohibits person/dog, not stuff classes. Regraded complete retained output without rerunning or changing measurement time."
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 2288.115999981528,
      "placement": {
        "mps:0": 401
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 4552.934582985472,
      "inferenceMs": 336.9806249975227,
      "durationMs": 393.3273750008084,
      "imagesPerSecond": 2.9675296614081343,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 837681152,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 7516291072,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 2681.4433749823365,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "8ac2b64b1bb2b5927a2c38ec6c74dccc25693e9e",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "maximumIoU": 0.9861876700365128,
      "passed": true,
      "interiorPointPassed": [
        true,
        true,
        true
      ],
      "segmentMapComplete": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1781.4454159815796,
      "placement": {
        "mps:0": 401
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float16",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 3360.8141249860637,
      "inferenceMs": 587.0447500201408,
      "durationMs": 734.9598330038134,
      "imagesPerSecond": 1.7034476502271612,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 804421632,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 4596940800,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 2516.405248985393,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "8ac2b64b1bb2b5927a2c38ec6c74dccc25693e9e",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "maximumIoU": 0.9173972880765147,
      "passed": true,
      "interiorPointPassed": [
        true,
        true,
        true
      ],
      "segmentMapComplete": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1781.4454159815796,
      "placement": {
        "mps:0": 401
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float16",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 3360.8141249860637,
      "inferenceMs": 908.8852499844506,
      "durationMs": 993.3013339759782,
      "imagesPerSecond": 1.1002489038270873,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        1199
      ],
      "peakHelperRssBytes": 933429248,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 7932002304,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 2774.7467499575578,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "8ac2b64b1bb2b5927a2c38ec6c74dccc25693e9e",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "maximumIoU": 0,
      "passed": true,
      "interiorPointPassed": [],
      "segmentMapComplete": true,
      "correction": "Blank panoptic may contain background stuff; frozen panoptic negative policy prohibits person/dog, not stuff classes. Regraded complete retained output without rerunning or changing measurement time."
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1781.4454159815796,
      "placement": {
        "mps:0": 401
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float16",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 3360.8141249860637,
      "inferenceMs": 472.37279202090576,
      "durationMs": 524.7766659886111,
      "imagesPerSecond": 2.1169720544695196,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 936230912,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 7933624320,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 2306.2220819701906,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "8ac2b64b1bb2b5927a2c38ec6c74dccc25693e9e",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.98573333976876
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1261.899208999239,
      "placement": {
        "mps:0": 401
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2841.8894589995034,
      "inferenceMs": 436.1570840119384,
      "durationMs": 549.417041009292,
      "imagesPerSecond": 2.292751938823555,
      "visualizationMs": 32.73312500095926,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 700792832,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5091868672,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 1811.316250008531,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "8ac2b64b1bb2b5927a2c38ec6c74dccc25693e9e",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.98573333976876
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 401
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2841.8894589995034,
      "inferenceMs": 324.5371660159435,
      "durationMs": 354.25545898033306,
      "imagesPerSecond": 3.0813111862536973,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 701153280,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5091868672,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 354.25545898033306,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "8ac2b64b1bb2b5927a2c38ec6c74dccc25693e9e",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.98573333976876
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 401
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2841.8894589995034,
      "inferenceMs": 322.921791987028,
      "durationMs": 352.6161669869907,
      "imagesPerSecond": 3.0967250424529134,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 701251584,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5091868672,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 352.6161669869907,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "8ac2b64b1bb2b5927a2c38ec6c74dccc25693e9e",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.98573333976876
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 401
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2841.8894589995034,
      "inferenceMs": 324.60958301089704,
      "durationMs": 354.1258749901317,
      "imagesPerSecond": 3.080623778030701,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 701300736,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5091868672,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 354.1258749901317,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "8ac2b64b1bb2b5927a2c38ec6c74dccc25693e9e",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.98573333976876
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 401
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2841.8894589995034,
      "inferenceMs": 324.13241601898335,
      "durationMs": 354.8109160037711,
      "imagesPerSecond": 3.0851588751352574,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 705003520,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5091868672,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 354.8109160037711,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "8ac2b64b1bb2b5927a2c38ec6c74dccc25693e9e",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.98573333976876
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 401
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2841.8894589995034,
      "inferenceMs": 320.88366700918414,
      "durationMs": 350.68375000264496,
      "imagesPerSecond": 3.1163942039199477,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 709754880,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5091868672,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 350.68375000264496,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "8ac2b64b1bb2b5927a2c38ec6c74dccc25693e9e",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.98573333976876
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 401
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2841.8894589995034,
      "inferenceMs": 325.0275000173133,
      "durationMs": 355.2577919908799,
      "imagesPerSecond": 3.076662743757783,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 713965568,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5091868672,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 355.2577919908799,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "8ac2b64b1bb2b5927a2c38ec6c74dccc25693e9e",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.98573333976876
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 401
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2841.8894589995034,
      "inferenceMs": 324.0694169944618,
      "durationMs": 354.5257079822477,
      "imagesPerSecond": 3.0857586293528265,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 713965568,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5091868672,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 354.5257079822477,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "8ac2b64b1bb2b5927a2c38ec6c74dccc25693e9e",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.98573333976876
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 401
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2841.8894589995034,
      "inferenceMs": 324.0024999831803,
      "durationMs": 354.69649999868125,
      "imagesPerSecond": 3.0863959384631667,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 714047488,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5091868672,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 354.69649999868125,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "8ac2b64b1bb2b5927a2c38ec6c74dccc25693e9e",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.98573333976876
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 401
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2841.8894589995034,
      "inferenceMs": 327.5075000128709,
      "durationMs": 357.26537500158884,
      "imagesPerSecond": 3.0533651899901546,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 714047488,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5091868672,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 357.26537500158884,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "8ac2b64b1bb2b5927a2c38ec6c74dccc25693e9e",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.98573333976876
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 401
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2841.8894589995034,
      "inferenceMs": 323.37958397693,
      "durationMs": 353.9440410095267,
      "imagesPerSecond": 3.0923411666932576,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 714047488,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5091868672,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 353.9440410095267,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "8ac2b64b1bb2b5927a2c38ec6c74dccc25693e9e",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.98573333976876
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 401
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2841.8894589995034,
      "inferenceMs": 324.7194580035284,
      "durationMs": 355.02087499480695,
      "imagesPerSecond": 3.0795813904971903,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 714047488,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5091868672,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 355.02087499480695,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "8ac2b64b1bb2b5927a2c38ec6c74dccc25693e9e",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.98573333976876
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 401
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2841.8894589995034,
      "inferenceMs": 323.99350000196137,
      "durationMs": 354.89529100595973,
      "imagesPerSecond": 3.08648167322476,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 714047488,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5091868672,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 354.89529100595973,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "8ac2b64b1bb2b5927a2c38ec6c74dccc25693e9e",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.98573333976876
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 401
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2841.8894589995034,
      "inferenceMs": 325.18887499463744,
      "durationMs": 357.45345801115036,
      "imagesPerSecond": 3.075135949889093,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 714080256,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5091868672,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 357.45345801115036,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "8ac2b64b1bb2b5927a2c38ec6c74dccc25693e9e",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.98573333976876
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 401
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2841.8894589995034,
      "inferenceMs": 322.90716699208133,
      "durationMs": 353.7814999872353,
      "imagesPerSecond": 3.096865298206661,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 714080256,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5091868672,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 353.7814999872353,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "8ac2b64b1bb2b5927a2c38ec6c74dccc25693e9e",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.98573333976876
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 401
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2841.8894589995034,
      "inferenceMs": 325.6410419999156,
      "durationMs": 356.2467919837218,
      "imagesPerSecond": 3.0708659874643787,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 714080256,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5091868672,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 356.2467919837218,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "8ac2b64b1bb2b5927a2c38ec6c74dccc25693e9e",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.98573333976876
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 401
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2841.8894589995034,
      "inferenceMs": 329.7627919819206,
      "durationMs": 359.81124997488223,
      "imagesPerSecond": 3.032482815874586,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 714113024,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5091868672,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 359.81124997488223,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "8ac2b64b1bb2b5927a2c38ec6c74dccc25693e9e",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.98573333976876
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 401
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2841.8894589995034,
      "inferenceMs": 325.905125006102,
      "durationMs": 354.97308400226757,
      "imagesPerSecond": 3.068377645123797,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 714113024,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5091868672,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 354.97308400226757,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "8ac2b64b1bb2b5927a2c38ec6c74dccc25693e9e",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.98573333976876
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 401
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2841.8894589995034,
      "inferenceMs": 326.834624982439,
      "durationMs": 358.08841697871685,
      "imagesPerSecond": 3.0596513452444967,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 714129408,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5091868672,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 358.08841697871685,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "8ac2b64b1bb2b5927a2c38ec6c74dccc25693e9e",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.98573333976876
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 401
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2841.8894589995034,
      "inferenceMs": 328.2677500101272,
      "durationMs": 358.65287500200793,
      "imagesPerSecond": 3.0462937646757853,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 714162176,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5091868672,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 358.65287500200793,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "8ac2b64b1bb2b5927a2c38ec6c74dccc25693e9e",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.98573333976876
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 401
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2841.8894589995034,
      "inferenceMs": 323.74975000857376,
      "durationMs": 353.933833015617,
      "imagesPerSecond": 3.0888054738992614,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 714162176,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5091868672,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 353.933833015617,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "8ac2b64b1bb2b5927a2c38ec6c74dccc25693e9e",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.98573333976876
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 401
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2841.8894589995034,
      "inferenceMs": 324.7934590035584,
      "durationMs": 354.7967500053346,
      "imagesPerSecond": 3.0788797381201083,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 714178560,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5091868672,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 354.7967500053346,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "8ac2b64b1bb2b5927a2c38ec6c74dccc25693e9e",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.98573333976876
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 401
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2841.8894589995034,
      "inferenceMs": 323.32237501395866,
      "durationMs": 353.72758298763074,
      "imagesPerSecond": 3.092888328427092,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 714227712,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5091868672,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 353.72758298763074,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "8ac2b64b1bb2b5927a2c38ec6c74dccc25693e9e",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.98573333976876
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 401
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2841.8894589995034,
      "inferenceMs": 323.50999998743646,
      "durationMs": 354.15437500341795,
      "imagesPerSecond": 3.091094556702529,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 714227712,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5091868672,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 354.15437500341795,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "8ac2b64b1bb2b5927a2c38ec6c74dccc25693e9e",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.98573333976876
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 401
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2841.8894589995034,
      "inferenceMs": 327.2315000067465,
      "durationMs": 357.34862499521114,
      "imagesPerSecond": 3.055940519110731,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 714244096,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5091868672,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 357.34862499521114,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "8ac2b64b1bb2b5927a2c38ec6c74dccc25693e9e",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.98573333976876
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 401
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2841.8894589995034,
      "inferenceMs": 324.47095899260603,
      "durationMs": 356.62908299127594,
      "imagesPerSecond": 3.081939915685298,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 714244096,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5091868672,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 356.62908299127594,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "8ac2b64b1bb2b5927a2c38ec6c74dccc25693e9e",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.98573333976876
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1106.3692919851746,
      "placement": {
        "mps:0": 401
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2729.7882909770124,
      "inferenceMs": 431.8339999881573,
      "durationMs": 543.561457976466,
      "imagesPerSecond": 2.3157046458301664,
      "visualizationMs": 32.30466600507498,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 702693376,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5091868672,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 1649.9307499616407,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "8ac2b64b1bb2b5927a2c38ec6c74dccc25693e9e",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.98573333976876
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1145.6812499964144,
      "placement": {
        "mps:0": 401
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2726.9406669947784,
      "inferenceMs": 436.16091599687934,
      "durationMs": 550.3374999971129,
      "imagesPerSecond": 2.2927317953614046,
      "visualizationMs": 32.75266700075008,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 700760064,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5091868672,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 1696.0187499935273,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "8ac2b64b1bb2b5927a2c38ec6c74dccc25693e9e",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9861876700365128
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1059.6938749949913,
      "placement": {
        "mps:0": 401
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float16",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2588.971124991076,
      "inferenceMs": 424.27825002232566,
      "durationMs": 534.129292005673,
      "imagesPerSecond": 2.3569438215307517,
      "visualizationMs": 29.499958996893838,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 808960000,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 4596940800,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 1593.8231670006644,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "8ac2b64b1bb2b5927a2c38ec6c74dccc25693e9e",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9861876700365128
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 401
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float16",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2588.971124991076,
      "inferenceMs": 312.1850000170525,
      "durationMs": 341.5276250161696,
      "imagesPerSecond": 3.2032288545105527,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 809926656,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 4596940800,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 341.5276250161696,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "8ac2b64b1bb2b5927a2c38ec6c74dccc25693e9e",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9861876700365128
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 401
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float16",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2588.971124991076,
      "inferenceMs": 309.8278749966994,
      "durationMs": 340.63183297985233,
      "imagesPerSecond": 3.2275985497129756,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 809992192,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 4596940800,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 340.63183297985233,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "8ac2b64b1bb2b5927a2c38ec6c74dccc25693e9e",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9861876700365128
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 401
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float16",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2588.971124991076,
      "inferenceMs": 308.05316698388197,
      "durationMs": 337.5997919938527,
      "imagesPerSecond": 3.246192888685096,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 810057728,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 4596940800,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 337.5997919938527,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "8ac2b64b1bb2b5927a2c38ec6c74dccc25693e9e",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9861876700365128
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 401
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float16",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2588.971124991076,
      "inferenceMs": 308.77058298210613,
      "durationMs": 339.4285420072265,
      "imagesPerSecond": 3.238650490412657,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 810090496,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 4596940800,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 339.4285420072265,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "8ac2b64b1bb2b5927a2c38ec6c74dccc25693e9e",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9861876700365128
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 401
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float16",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2588.971124991076,
      "inferenceMs": 309.74437500117347,
      "durationMs": 338.5859169939067,
      "imagesPerSecond": 3.2284686364238624,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 810123264,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 4596940800,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 338.5859169939067,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "8ac2b64b1bb2b5927a2c38ec6c74dccc25693e9e",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9861876700365128
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1051.8829579814337,
      "placement": {
        "mps:0": 401
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float16",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2619.3784170027357,
      "inferenceMs": 424.2209169897251,
      "durationMs": 533.9903749991208,
      "imagesPerSecond": 2.3572623601306786,
      "visualizationMs": 30.43658399838023,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 806207488,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 4596940800,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 1585.8733329805546,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "8ac2b64b1bb2b5927a2c38ec6c74dccc25693e9e",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9861876700365128
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1069.839875010075,
      "placement": {
        "mps:0": 401
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float16",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2733.5032080009114,
      "inferenceMs": 452.98804200137965,
      "durationMs": 571.175958000822,
      "imagesPerSecond": 2.207563792593347,
      "visualizationMs": 31.500166980549693,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 805896192,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 4596940800,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 1641.015833010897,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "8ac2b64b1bb2b5927a2c38ec6c74dccc25693e9e",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  }
]

DETR ResNet-50-dc5-panoptic Desktop

48 configurations tested
Passed42
Failed6
TestedNov 11, 2026
Runtimetransformers 5.22.1

Quality criteria

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

PrecisionDeviceResultPlatformInference / load
FP16mpsAcceleratedPassedModeratemacos · aarch64467 ms / 0 ms
FP16mpsAcceleratedPassedModeratemacos · aarch64468 ms / 0 ms
FP16mpsAcceleratedPassedModeratemacos · aarch64469 ms / 0 ms
FP16mpsAcceleratedPassedModeratemacos · aarch64471 ms / 1239 ms
FP16mpsAcceleratedPassedModeratemacos · aarch64471 ms / 0 ms
FP16mpsAcceleratedPassedModeratemacos · aarch64472 ms / 0 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64486 ms / 0 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64486 ms / 0 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64487 ms / 0 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64487 ms / 0 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64488 ms / 0 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64488 ms / 0 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64488 ms / 0 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64488 ms / 0 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64488 ms / 0 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64489 ms / 0 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64489 ms / 0 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64489 ms / 0 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64489 ms / 0 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64489 ms / 0 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64489 ms / 0 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64490 ms / 0 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64490 ms / 0 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64490 ms / 0 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64490 ms / 0 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64490 ms / 0 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64491 ms / 0 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64491 ms / 0 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64491 ms / 0 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64493 ms / 0 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64497 ms / 0 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64554 ms / 1545 ms
FP16mpsAcceleratedPassedModeratemacos · aarch64574 ms / 1130 ms
FP16mpsAcceleratedPassedModeratemacos · aarch64579 ms / 1140 ms
FP16mpsAcceleratedPassedModeratemacos · aarch64591 ms / 1239 ms
FP16mpsAcceleratedPassedModeratemacos · aarch64594 ms / 1136 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64603 ms / 1174 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64610 ms / 1302 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64629 ms / 1180 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64790 ms / 1545 ms
FP16mpsAcceleratedPassedModeratemacos · aarch64828 ms / 1239 ms
FP32mpsAcceleratedPassedModeratemacos · aarch641064 ms / 1545 ms
FP32cpuFailed—macos · aarch6425100 ms / 27770 ms
FP32cpuFailed—macos · aarch6431284 ms / 27770 ms
FP32cpuFailed—macos · aarch6425404 ms / 27770 ms
FP32cpuFailed—macos · aarch647240 ms / 8303 ms
FP32cpuFailed—macos · aarch648917 ms / 8303 ms
FP32cpuFailed—macos · aarch647263 ms / 8303 ms
Test setup & sample details
Quality gate
No explicit quality gate recorded. Review the per-case quality evidence; full certification is pending.
Tested
2026-11-11T22:07:26.826194+00:00
Environment
{ "os": "macos", "arch": "aarch64", "hardware": "Apple M3 Max", "osVersion": "27.2", "runtime": "Transformers5.22.1/Torch2.10.0/Torchvision0.25.0/timm1.0.22", "interpreter": "App-owned CPython3.13.18 private managed environment", "transport": "Local immutable original snapshots; downloads excluded from speed measurements." }
Runtime
transformers 5.22.1
Artifact revision
9592dad3091c0c650ec967461a1c27a630b414e9
Fixture
{
  "file": "model-samples/desktop/mobilenet-dog.jpg",
  "size": 83281,
  "sha1": "f1d042827c2f2154a6d8043b8e8f07c1be612941",
  "sha256": "d4cb12ea8a768582562e1f1c0c930ea6fc4757f5b66551b986b41498efd75ca1",
  "url": "https://assets.localai.tools/samples/v1/files/d4cb12ea8a768582562e1f1c0c930ea6fc4757f5b66551b986b41498efd75ca1/mobilenet-dog.jpg"
}
Rerun when
Artifact/runtime revision; Preprocessing/postprocessing/presentation; Fixtures or frozen gates; Relevant platform/backend behavior
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "fail",
    "quality": {
      "maximumIoU": 0.9871514305631379,
      "passed": true,
      "interiorPointPassed": [
        true,
        true,
        true
      ],
      "segmentMapComplete": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 27770.200375001878,
      "placement": {
        "cpu": 350
      },
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 29321.745958994143,
      "inferenceMs": 25099.732917005895,
      "durationMs": 25148.07045800262,
      "imagesPerSecond": 0.0398410613892416,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 14233550848,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "totalMs": 52918.2708330045,
      "grade": "slow",
      "fallback": false,
      "artifactRevision": "9592dad3091c0c650ec967461a1c27a630b414e9",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "fail",
    "quality": {
      "maximumIoU": 0.9124726205107025,
      "passed": true,
      "interiorPointPassed": [
        true,
        true,
        true
      ],
      "segmentMapComplete": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 27770.200375001878,
      "placement": {
        "cpu": 350
      },
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 29321.745958994143,
      "inferenceMs": 31284.423332981532,
      "durationMs": 31366.943834000267,
      "imagesPerSecond": 0.03196478929326315,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        1199
      ],
      "peakHelperRssBytes": 15001812992,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "totalMs": 59137.144209002145,
      "grade": "slow",
      "fallback": false,
      "artifactRevision": "9592dad3091c0c650ec967461a1c27a630b414e9",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "fail",
    "quality": {
      "maximumIoU": 0,
      "passed": true,
      "interiorPointPassed": [],
      "segmentMapComplete": true,
      "correction": "Blank panoptic may contain background stuff; frozen panoptic negative policy prohibits person/dog, not stuff classes. Regraded complete retained output without rerunning or changing measurement time."
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 27770.200375001878,
      "placement": {
        "cpu": 350
      },
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 29321.745958994143,
      "inferenceMs": 25404.245250014355,
      "durationMs": 25424.471415986773,
      "imagesPerSecond": 0.039363499689070075,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 15001812992,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "totalMs": 53194.67179098865,
      "grade": "slow",
      "fallback": false,
      "artifactRevision": "9592dad3091c0c650ec967461a1c27a630b414e9",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "fail",
    "quality": {
      "maximumIoU": 0.9871514305631379,
      "passed": true,
      "interiorPointPassed": [
        true,
        true,
        true
      ],
      "segmentMapComplete": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 8302.607040997827,
      "placement": {
        "cpu": 350
      },
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 10673.602708993712,
      "inferenceMs": 7239.864249975653,
      "durationMs": 7306.980541994562,
      "imagesPerSecond": 0.13812413678935526,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 14759411712,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 15609.58758299239,
      "grade": "slow",
      "fallback": false,
      "artifactRevision": "9592dad3091c0c650ec967461a1c27a630b414e9",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "fail",
    "quality": {
      "maximumIoU": 0.9124726205107025,
      "passed": true,
      "interiorPointPassed": [
        true,
        true,
        true
      ],
      "segmentMapComplete": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 8302.607040997827,
      "placement": {
        "cpu": 350
      },
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 10673.602708993712,
      "inferenceMs": 8916.618957999162,
      "durationMs": 9028.54120801203,
      "imagesPerSecond": 0.11215013277010036,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        1199
      ],
      "peakHelperRssBytes": 17946574848,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 17331.148249009857,
      "grade": "slow",
      "fallback": false,
      "artifactRevision": "9592dad3091c0c650ec967461a1c27a630b414e9",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "fail",
    "quality": {
      "maximumIoU": 0,
      "passed": true,
      "interiorPointPassed": [],
      "segmentMapComplete": true,
      "correction": "Blank panoptic may contain background stuff; frozen panoptic negative policy prohibits person/dog, not stuff classes. Regraded complete retained output without rerunning or changing measurement time."
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 8302.607040997827,
      "placement": {
        "cpu": 350
      },
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 10673.602708993712,
      "inferenceMs": 7263.226000010036,
      "durationMs": 7280.02349997405,
      "imagesPerSecond": 0.1376798684219131,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 17946574848,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 15582.630540971877,
      "grade": "slow",
      "fallback": false,
      "artifactRevision": "9592dad3091c0c650ec967461a1c27a630b414e9",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "maximumIoU": 0.9871515165354218,
      "passed": true,
      "interiorPointPassed": [
        true,
        true,
        true
      ],
      "segmentMapComplete": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1545.2814169984777,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 3124.8497499909718,
      "inferenceMs": 789.5834170049056,
      "durationMs": 896.3972500059754,
      "imagesPerSecond": 1.2664906309624118,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 693354496,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 8022212608,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 2441.678667004453,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "9592dad3091c0c650ec967461a1c27a630b414e9",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "maximumIoU": 0.9124723990527835,
      "passed": true,
      "interiorPointPassed": [
        true,
        true,
        true
      ],
      "segmentMapComplete": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1545.2814169984777,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 3124.8497499909718,
      "inferenceMs": 1064.441499998793,
      "durationMs": 1141.4187080226839,
      "imagesPerSecond": 0.9394598012207659,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        1199
      ],
      "peakHelperRssBytes": 821559296,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 12792135680,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 2686.7001250211615,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "9592dad3091c0c650ec967461a1c27a630b414e9",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "maximumIoU": 0,
      "passed": true,
      "interiorPointPassed": [],
      "segmentMapComplete": true,
      "correction": "Blank panoptic may contain background stuff; frozen panoptic negative policy prohibits person/dog, not stuff classes. Regraded complete retained output without rerunning or changing measurement time."
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1545.2814169984777,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 3124.8497499909718,
      "inferenceMs": 553.9880419964902,
      "durationMs": 586.4050830132328,
      "imagesPerSecond": 1.80509311427761,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 828375040,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 12793757696,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 2131.6865000117105,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "9592dad3091c0c650ec967461a1c27a630b414e9",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "maximumIoU": 0.9869621694836871,
      "passed": true,
      "interiorPointPassed": [
        true,
        true,
        true
      ],
      "segmentMapComplete": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1239.3870830128435,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float16",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 3326.0197909839917,
      "inferenceMs": 590.5573329946492,
      "durationMs": 682.1962909889407,
      "imagesPerSecond": 1.6933156937178537,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 768966656,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 6254313472,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 1921.5833740017843,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "9592dad3091c0c650ec967461a1c27a630b414e9",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "maximumIoU": 0.9128492148118397,
      "passed": true,
      "interiorPointPassed": [
        true,
        true,
        true
      ],
      "segmentMapComplete": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1239.3870830128435,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float16",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 3326.0197909839917,
      "inferenceMs": 827.7409999864176,
      "durationMs": 901.6396249935497,
      "imagesPerSecond": 1.2081073669377367,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        1199
      ],
      "peakHelperRssBytes": 894517248,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 11936497664,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 2141.026708006393,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "9592dad3091c0c650ec967461a1c27a630b414e9",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "maximumIoU": 0,
      "passed": true,
      "interiorPointPassed": [],
      "segmentMapComplete": true,
      "correction": "Blank panoptic may contain background stuff; frozen panoptic negative policy prohibits person/dog, not stuff classes. Regraded complete retained output without rerunning or changing measurement time."
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1239.3870830128435,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float16",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 3326.0197909839917,
      "inferenceMs": 471.3823330239393,
      "durationMs": 504.8205000057351,
      "imagesPerSecond": 2.121420193211218,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 904396800,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 11938119680,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 1744.2075830185786,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "9592dad3091c0c650ec967461a1c27a630b414e9",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9871515165354218
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1302.2330410021823,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2852.5784999947064,
      "inferenceMs": 609.7801249998156,
      "durationMs": 727.3835410014726,
      "imagesPerSecond": 1.6399353783469122,
      "visualizationMs": 30.26904200669378,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 697450496,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 8022212608,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 2029.616582003655,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "9592dad3091c0c650ec967461a1c27a630b414e9",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9871515165354218
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2852.5784999947064,
      "inferenceMs": 493.0474159773439,
      "durationMs": 521.7812080227304,
      "imagesPerSecond": 2.0282024965443712,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 698056704,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 8022212608,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 521.7812080227304,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "9592dad3091c0c650ec967461a1c27a630b414e9",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9871515165354218
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2852.5784999947064,
      "inferenceMs": 490.6198749959003,
      "durationMs": 520.6952500157058,
      "imagesPerSecond": 2.0382378516735713,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 698318848,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 8022212608,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 520.6952500157058,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "9592dad3091c0c650ec967461a1c27a630b414e9",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9871515165354218
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2852.5784999947064,
      "inferenceMs": 496.96029198821634,
      "durationMs": 527.1603750006761,
      "imagesPerSecond": 2.012233202776111,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 702382080,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 8022212608,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 527.1603750006761,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "9592dad3091c0c650ec967461a1c27a630b414e9",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9871515165354218
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2852.5784999947064,
      "inferenceMs": 490.4961670108605,
      "durationMs": 526.7662089900114,
      "imagesPerSecond": 2.038751915420897,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 710066176,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 8022212608,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 526.7662089900114,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "9592dad3091c0c650ec967461a1c27a630b414e9",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9871515165354218
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2852.5784999947064,
      "inferenceMs": 489.40741701517254,
      "durationMs": 520.5585419898853,
      "imagesPerSecond": 2.043287382318111,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 711294976,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 8022212608,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 520.5585419898853,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "9592dad3091c0c650ec967461a1c27a630b414e9",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9871515165354218
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2852.5784999947064,
      "inferenceMs": 488.3321670058649,
      "durationMs": 519.0166249813046,
      "imagesPerSecond": 2.0477864608660727,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 711376896,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 8022212608,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 519.0166249813046,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "9592dad3091c0c650ec967461a1c27a630b414e9",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9871515165354218
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2852.5784999947064,
      "inferenceMs": 487.5729589839466,
      "durationMs": 519.2578330170363,
      "imagesPerSecond": 2.0509751034673873,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 711376896,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 8022212608,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 519.2578330170363,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "9592dad3091c0c650ec967461a1c27a630b414e9",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9871515165354218
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2852.5784999947064,
      "inferenceMs": 489.7541669779457,
      "durationMs": 519.950375019107,
      "imagesPerSecond": 2.041840718110789,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 711393280,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 8009629696,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 519.950375019107,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "9592dad3091c0c650ec967461a1c27a630b414e9",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9871515165354218
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2852.5784999947064,
      "inferenceMs": 490.7530840137042,
      "durationMs": 521.5948749973904,
      "imagesPerSecond": 2.037684596541578,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 711442432,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 8009629696,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 521.5948749973904,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "9592dad3091c0c650ec967461a1c27a630b414e9",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9871515165354218
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2852.5784999947064,
      "inferenceMs": 489.31529201217927,
      "durationMs": 518.7803749868181,
      "imagesPerSecond": 2.0436720787690192,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 711458816,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 8009629696,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 518.7803749868181,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "9592dad3091c0c650ec967461a1c27a630b414e9",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9871515165354218
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2852.5784999947064,
      "inferenceMs": 489.37895899871364,
      "durationMs": 519.3573330179788,
      "imagesPerSecond": 2.043406202109782,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 711491584,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 8009629696,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 519.3573330179788,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "9592dad3091c0c650ec967461a1c27a630b414e9",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9871515165354218
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2852.5784999947064,
      "inferenceMs": 489.7294160036836,
      "durationMs": 522.5744589988608,
      "imagesPerSecond": 2.0419439129473864,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 711540736,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 8009629696,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 522.5744589988608,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "9592dad3091c0c650ec967461a1c27a630b414e9",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9871515165354218
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2852.5784999947064,
      "inferenceMs": 488.6499169806484,
      "durationMs": 520.0844580249395,
      "imagesPerSecond": 2.0464548652314662,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 711622656,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 8009629696,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 520.0844580249395,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "9592dad3091c0c650ec967461a1c27a630b414e9",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9871515165354218
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2852.5784999947064,
      "inferenceMs": 487.9787500249222,
      "durationMs": 517.2950840205885,
      "imagesPerSecond": 2.049269563375306,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 711671808,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 8009629696,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 517.2950840205885,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "9592dad3091c0c650ec967461a1c27a630b414e9",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9871515165354218
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2852.5784999947064,
      "inferenceMs": 485.9831669891719,
      "durationMs": 515.6172919960227,
      "imagesPerSecond": 2.057684438321875,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 711688192,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 8009629696,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 515.6172919960227,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "9592dad3091c0c650ec967461a1c27a630b414e9",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9871515165354218
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2852.5784999947064,
      "inferenceMs": 489.65466700610705,
      "durationMs": 519.3923330225516,
      "imagesPerSecond": 2.0422556290830327,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 711737344,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 8009629696,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 519.3923330225516,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "9592dad3091c0c650ec967461a1c27a630b414e9",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9871515165354218
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2852.5784999947064,
      "inferenceMs": 489.3240000237711,
      "durationMs": 520.2072500251234,
      "imagesPerSecond": 2.0436357095736577,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 711737344,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 8009629696,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 520.2072500251234,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "9592dad3091c0c650ec967461a1c27a630b414e9",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9871515165354218
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2852.5784999947064,
      "inferenceMs": 487.1792909980286,
      "durationMs": 521.2744169984944,
      "imagesPerSecond": 2.0526324055183345,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 711770112,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 8009629696,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 521.2744169984944,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "9592dad3091c0c650ec967461a1c27a630b414e9",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9871515165354218
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2852.5784999947064,
      "inferenceMs": 488.2353749999311,
      "durationMs": 519.6634579915553,
      "imagesPerSecond": 2.0481924317756395,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 711786496,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 8009629696,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 519.6634579915553,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "9592dad3091c0c650ec967461a1c27a630b414e9",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9871515165354218
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2852.5784999947064,
      "inferenceMs": 491.3834589824546,
      "durationMs": 524.0089170110878,
      "imagesPerSecond": 2.035070537520283,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 711802880,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 8009629696,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 524.0089170110878,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "9592dad3091c0c650ec967461a1c27a630b414e9",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9871515165354218
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2852.5784999947064,
      "inferenceMs": 487.0696250000037,
      "durationMs": 519.5483340066858,
      "imagesPerSecond": 2.0530945652790242,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 711802880,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 8009629696,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 519.5483340066858,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "9592dad3091c0c650ec967461a1c27a630b414e9",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9871515165354218
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2852.5784999947064,
      "inferenceMs": 485.9273329784628,
      "durationMs": 516.9794170069508,
      "imagesPerSecond": 2.0579208703296423,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 711819264,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 8009629696,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 516.9794170069508,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "9592dad3091c0c650ec967461a1c27a630b414e9",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9871515165354218
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2852.5784999947064,
      "inferenceMs": 488.79083298379555,
      "durationMs": 518.5122920083813,
      "imagesPerSecond": 2.0458648823169563,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 711835648,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 8009629696,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 518.5122920083813,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "9592dad3091c0c650ec967461a1c27a630b414e9",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9871515165354218
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2852.5784999947064,
      "inferenceMs": 487.82183302682824,
      "durationMs": 517.4129160004668,
      "imagesPerSecond": 2.0499287491812694,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 711868416,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 8009629696,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 517.4129160004668,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "9592dad3091c0c650ec967461a1c27a630b414e9",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9871515165354218
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2852.5784999947064,
      "inferenceMs": 490.1799579965882,
      "durationMs": 521.4600000181235,
      "imagesPerSecond": 2.0400670890076666,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 711917568,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 8009629696,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 521.4600000181235,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "9592dad3091c0c650ec967461a1c27a630b414e9",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9871515165354218
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1173.6687909869943,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2756.6147500183433,
      "inferenceMs": 603.3314579981379,
      "durationMs": 716.8970420025289,
      "imagesPerSecond": 1.657463715414432,
      "visualizationMs": 30.360709002707154,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 698040320,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 8022212608,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 1890.5658329895232,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "9592dad3091c0c650ec967461a1c27a630b414e9",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9871515165354218
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1180.403500009561,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float32",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2751.966625015484,
      "inferenceMs": 629.3060419848189,
      "durationMs": 757.4333340162411,
      "imagesPerSecond": 1.5890519608647322,
      "visualizationMs": 35.461457999190316,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 707870720,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 8022212608,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 1937.836834025802,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "9592dad3091c0c650ec967461a1c27a630b414e9",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9869621694836871
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1140.2928339957725,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float16",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2726.4408339979127,
      "inferenceMs": 579.2861249938142,
      "durationMs": 693.7802920001559,
      "imagesPerSecond": 1.726262647859343,
      "visualizationMs": 33.381457993527874,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 764379136,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 6254313472,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 1834.0731259959284,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "9592dad3091c0c650ec967461a1c27a630b414e9",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9869621694836871
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float16",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2726.4408339979127,
      "inferenceMs": 468.96966599160805,
      "durationMs": 499.2262499872595,
      "imagesPerSecond": 2.1323340772703503,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 764968960,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 7328055296,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 499.2262499872595,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "9592dad3091c0c650ec967461a1c27a630b414e9",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9869621694836871
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float16",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2726.4408339979127,
      "inferenceMs": 471.9873750000261,
      "durationMs": 504.04333299957216,
      "imagesPerSecond": 2.1187007385524765,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 765149184,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 7328055296,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 504.04333299957216,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "9592dad3091c0c650ec967461a1c27a630b414e9",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9869621694836871
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float16",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2726.4408339979127,
      "inferenceMs": 466.92066598916426,
      "durationMs": 497.45037499815226,
      "imagesPerSecond": 2.1416914539036633,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 768802816,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 7328055296,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 497.45037499815226,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "9592dad3091c0c650ec967461a1c27a630b414e9",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9869621694836871
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float16",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2726.4408339979127,
      "inferenceMs": 471.4872079784982,
      "durationMs": 503.7172080192249,
      "imagesPerSecond": 2.1209483164718312,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 776601600,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 7328055296,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 503.7172080192249,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "9592dad3091c0c650ec967461a1c27a630b414e9",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9869621694836871
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float16",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2726.4408339979127,
      "inferenceMs": 467.57587502361275,
      "durationMs": 498.31395799992606,
      "imagesPerSecond": 2.138690324751036,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 777846784,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 7328055296,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 498.31395799992606,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "9592dad3091c0c650ec967461a1c27a630b414e9",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9869621694836871
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1135.7361249974929,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float16",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2686.0885830246843,
      "inferenceMs": 594.4993330049329,
      "durationMs": 720.7587910233997,
      "imagesPerSecond": 1.6820876735814647,
      "visualizationMs": 34.121249977033585,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 764821504,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 6254313472,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 1856.4949160208926,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "9592dad3091c0c650ec967461a1c27a630b414e9",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9869621694836871
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1129.567500000121,
      "placement": {
        "mps:0": 350
      },
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "parameterDtype": "torch.float16",
      "accelerationDetails": "Network parameters, inputs, logits and boxes on selected device; image decoding/resize/postprocessing/SVG on CPU.",
      "processInitWallMs": 2778.9037080074195,
      "inferenceMs": 574.4419999828096,
      "durationMs": 690.6068329990376,
      "imagesPerSecond": 1.74081978690612,
      "visualizationMs": 30.0415410019923,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 768720896,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 6254313472,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 1820.1743329991587,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "9592dad3091c0c650ec967461a1c27a630b414e9",
      "loadTimingScope": "Model construction/placement plus one warmup after imports; excludes process/runtime/artifact setup.",
      "totalTimingScope": "Model load plus preprocess/inference/postprocess duration. Warm repeats have zero model-load time."
    }
  }
]
Release validation still required
  • Public sample publication, complete execution-dependency backup, public asset verification and forced model/runtime-source recovery deferred under the existing upload exclusion. Local acceptance is tracked independently in the investigation log.

Something not quite right?

Tell us what happened with DETR 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.