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

Locate common COCO objects in still images with complete class labels, confidence scores and pixel bounding boxes.

Desktop inferencePrivate · Runs on your computer
About this model

Locate common COCO objects in still images with complete class labels, confidence scores and pixel bounding boxes.

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

Best for

  • Review common objects and their spatial locations in photos.

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.

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
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
Files updated
Apr 10, 2024
Artifact revision
1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b

Desktop app

Sample result

DETR ResNet-50 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 SiliconCPU / FP32Tested8.0 GiBRAM estimate
macOS Apple SiliconMPS / FP32Tested8.0 GiBRAM estimate
macOS Apple SiliconMPS / FP16Tested8.0 GiBRAM estimate
Windows Intel / AMDCPU / FP32Unverified8.0 GiBRAM estimate
macOS Intel / AMDCPU / FP32Unverified8.0 GiBRAM estimate
If your device falls short

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

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

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

Benchmark evidence

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

DETR ResNet-50 Desktop

56 configurations tested
Passed56
Failed0
TestedNov 11, 2026
Runtimetransformers 5.22.1

Quality criteria

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

PrecisionDeviceResultPlatformInference / load
FP16mpsAcceleratedPassedVery fastmacos · aarch6458 ms / 0 ms
FP16mpsAcceleratedPassedVery fastmacos · aarch6458 ms / 0 ms
FP16mpsAcceleratedPassedVery fastmacos · aarch6459 ms / 0 ms
FP16mpsAcceleratedPassedVery fastmacos · aarch6459 ms / 593 ms
FP16mpsAcceleratedPassedVery fastmacos · aarch6459 ms / 0 ms
FP16mpsAcceleratedPassedVery fastmacos · aarch6460 ms / 0 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch6466 ms / 0 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch6466 ms / 0 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch6466 ms / 0 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch6466 ms / 0 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch6466 ms / 0 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch6467 ms / 0 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch6467 ms / 0 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch6467 ms / 0 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch6467 ms / 0 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch6467 ms / 0 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch6467 ms / 0 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch6467 ms / 0 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch6467 ms / 0 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch6468 ms / 0 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch6468 ms / 0 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch6468 ms / 0 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch6468 ms / 0 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch6468 ms / 0 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch6468 ms / 0 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch6468 ms / 0 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch6468 ms / 0 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch6468 ms / 0 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch6468 ms / 0 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch6468 ms / 0 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch6468 ms / 592 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch6469 ms / 0 ms
FP16mpsAcceleratedPassedFastmacos · aarch64137 ms / 566 ms
FP16mpsAcceleratedPassedFastmacos · aarch64139 ms / 579 ms
FP16mpsAcceleratedPassedFastmacos · aarch64140 ms / 593 ms
FP32mpsAcceleratedPassedFastmacos · aarch64145 ms / 595 ms
FP32mpsAcceleratedPassedFastmacos · aarch64146 ms / 605 ms
FP32mpsAcceleratedPassedFastmacos · aarch64149 ms / 592 ms
FP16mpsAcceleratedPassedFastmacos · aarch64152 ms / 605 ms
FP16mpsAcceleratedPassedFastmacos · aarch64156 ms / 593 ms
FP32mpsAcceleratedPassedFastmacos · aarch64164 ms / 627 ms
FP32mpsAcceleratedPassedFastmacos · aarch64167 ms / 592 ms
FP32cpuPassedFastmacos · aarch64175 ms / 0 ms
FP32cpuPassedFastmacos · aarch64175 ms / 424 ms
FP32cpuPassedFastmacos · aarch64175 ms / 424 ms
FP32cpuPassedFastmacos · aarch64175 ms / 0 ms
FP32cpuPassedFastmacos · aarch64176 ms / 0 ms
FP32cpuPassedFastmacos · aarch64176 ms / 422 ms
FP32cpuPassedFastmacos · aarch64176 ms / 0 ms
FP32cpuPassedFastmacos · aarch64177 ms / 0 ms
FP32cpuPassedFastmacos · aarch64177 ms / 424 ms
FP32cpuPassedFastmacos · aarch64178 ms / 871 ms
FP32cpuPassedFastmacos · aarch64214 ms / 424 ms
FP32cpuPassedFastmacos · aarch64297 ms / 1029 ms
FP32cpuPassedFastmacos · aarch64299 ms / 1029 ms
FP32cpuPassedModeratemacos · aarch64375 ms / 1029 ms
Test setup & sample details
Quality gate
No explicit quality gate recorded. Review the per-case quality evidence; full certification is pending.
Tested
2026-11-11T22:06:11.396274+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
1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b
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": "pass",
    "quality": {
      "maximumIoU": 0.9819021708040253,
      "passed": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1028.6159999959636,
      "placement": {
        "cpu": 318
      },
      "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": 2595.952000003308,
      "inferenceMs": 297.00908399536274,
      "durationMs": 310.2181249996647,
      "imagesPerSecond": 3.3669003875168113,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 1227505664,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "totalMs": 1338.8341249956284,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": "pass",
    "quality": {
      "maximumIoU": 0.9242218051831882,
      "passed": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1028.6159999959636,
      "placement": {
        "cpu": 318
      },
      "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": 2595.952000003308,
      "inferenceMs": 375.01670798519626,
      "durationMs": 391.2328750011511,
      "imagesPerSecond": 2.6665478596208967,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        1199
      ],
      "peakHelperRssBytes": 1571373056,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "totalMs": 1419.8488749971148,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": "pass",
    "quality": {
      "maximumIoU": 0,
      "passed": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1028.6159999959636,
      "placement": {
        "cpu": 318
      },
      "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": 2595.952000003308,
      "inferenceMs": 299.3827919999603,
      "durationMs": 311.9086250080727,
      "imagesPerSecond": 3.3402053381883503,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 1571454976,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "totalMs": 1340.5246250040364,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": "pass",
    "quality": {
      "maximumIoU": 0.9819021708040253,
      "passed": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 424.1762080055196,
      "placement": {
        "cpu": 318
      },
      "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": 2013.607500004582,
      "inferenceMs": 175.18041699077003,
      "durationMs": 187.88658297853544,
      "imagesPerSecond": 5.708400614508689,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 1223770112,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 612.062790984055,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": "pass",
    "quality": {
      "maximumIoU": 0.9242218051831882,
      "passed": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 424.1762080055196,
      "placement": {
        "cpu": 318
      },
      "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": 2013.607500004582,
      "inferenceMs": 214.22983301454224,
      "durationMs": 230.3869999886956,
      "imagesPerSecond": 4.667883954015492,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        1199
      ],
      "peakHelperRssBytes": 1576124416,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 654.5632079942152,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": "pass",
    "quality": {
      "maximumIoU": 0,
      "passed": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 424.1762080055196,
      "placement": {
        "cpu": 318
      },
      "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": 2013.607500004582,
      "inferenceMs": 175.13016701559536,
      "durationMs": 188.4745409770403,
      "imagesPerSecond": 5.710038521866709,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 1576173568,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 612.6507489825599,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": 0.9819020845205763,
      "passed": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 592.3205830040388,
      "placement": {
        "mps:0": 318
      },
      "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": 2175.134082994191,
      "inferenceMs": 148.5192919790279,
      "durationMs": 186.47545800195076,
      "imagesPerSecond": 6.733132017227821,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 609091584,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1452146688,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 778.7960410059895,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": 0.924221741562551,
      "passed": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 592.3205830040388,
      "placement": {
        "mps:0": 318
      },
      "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": 2175.134082994191,
      "inferenceMs": 166.5119160024915,
      "durationMs": 192.1215000038501,
      "imagesPerSecond": 6.005576201435561,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        1199
      ],
      "peakHelperRssBytes": 703414272,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1554251776,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 784.4420830078889,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 592.3205830040388,
      "placement": {
        "mps:0": 318
      },
      "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": 2175.134082994191,
      "inferenceMs": 68.19512500078417,
      "durationMs": 83.2826669793576,
      "imagesPerSecond": 14.663804780598335,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 703610880,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1554251776,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 675.6032499833964,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": 0.9824125836025239,
      "passed": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 592.8491250087973,
      "placement": {
        "mps:0": 318
      },
      "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": 2168.8085830246564,
      "inferenceMs": 139.60774999577552,
      "durationMs": 176.5342500002589,
      "imagesPerSecond": 7.162926127168869,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 676200448,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1351483392,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 769.3833750090562,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": 0.92427464464452,
      "passed": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 592.8491250087973,
      "placement": {
        "mps:0": 318
      },
      "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": 2168.8085830246564,
      "inferenceMs": 156.16945800138637,
      "durationMs": 183.42099999426864,
      "imagesPerSecond": 6.403300701671915,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        1199
      ],
      "peakHelperRssBytes": 766705664,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1453588480,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 776.2701250030659,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 592.8491250087973,
      "placement": {
        "mps:0": 318
      },
      "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": 2168.8085830246564,
      "inferenceMs": 58.989834011299536,
      "durationMs": 75.24141698377207,
      "imagesPerSecond": 16.95207346758171,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 766820352,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1453588480,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 668.0905419925693,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9819021708040253
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 870.7149580004625,
      "placement": {
        "cpu": 318
      },
      "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": 2472.971541021252,
      "inferenceMs": 177.66799998935312,
      "durationMs": 210.75195897719823,
      "imagesPerSecond": 5.628475584010208,
      "visualizationMs": 20.117665990255773,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "maskDevice": null,
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 1223245824,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 1081.4669169776607,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9819021708040253
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "cpu": 318
      },
      "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": 2472.971541021252,
      "inferenceMs": 176.087333005853,
      "durationMs": 189.00208300328813,
      "imagesPerSecond": 5.679000203647589,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "maskDevice": null,
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 1226637312,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 189.00208300328813,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9819021708040253
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "cpu": 318
      },
      "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": 2472.971541021252,
      "inferenceMs": 174.76920800982043,
      "durationMs": 187.5214589817915,
      "imagesPerSecond": 5.721831731043887,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "maskDevice": null,
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 1226637312,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 187.5214589817915,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9819021708040253
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "cpu": 318
      },
      "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": 2472.971541021252,
      "inferenceMs": 176.5149580023717,
      "durationMs": 189.5035000052303,
      "imagesPerSecond": 5.665242262282178,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "maskDevice": null,
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 1226637312,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 189.5035000052303,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9819021708040253
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "cpu": 318
      },
      "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": 2472.971541021252,
      "inferenceMs": 175.51749999984168,
      "durationMs": 188.38241702178493,
      "imagesPerSecond": 5.697437577454681,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "maskDevice": null,
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 1226637312,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 188.38241702178493,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9819021708040253
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "cpu": 318
      },
      "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": 2472.971541021252,
      "inferenceMs": 175.444292020984,
      "durationMs": 188.04966699099168,
      "imagesPerSecond": 5.699814958245522,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "maskDevice": null,
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 1226670080,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 188.04966699099168,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9819021708040253
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 421.8220839975402,
      "placement": {
        "cpu": 318
      },
      "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": 2001.9725000020117,
      "inferenceMs": 175.85058300755918,
      "durationMs": 208.58766700257547,
      "imagesPerSecond": 5.686645917784723,
      "visualizationMs": 20.015499991131946,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "maskDevice": null,
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 1223966720,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 630.4097510001156,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9819021708040253
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 423.73966600280255,
      "placement": {
        "cpu": 318
      },
      "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": 2037.2288749786094,
      "inferenceMs": 177.2489580034744,
      "durationMs": 210.39270900655538,
      "imagesPerSecond": 5.641782108419493,
      "visualizationMs": 20.264457998564467,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "maskDevice": null,
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 1230831616,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 634.1323750093579,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": 0.9819020845205763
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 605.3942079888657,
      "placement": {
        "mps:0": 318
      },
      "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": 2121.528458024841,
      "inferenceMs": 146.39049998368137,
      "durationMs": 204.0243340015877,
      "imagesPerSecond": 6.831044364979102,
      "visualizationMs": 20.151791017269716,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 612237312,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1452146688,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 809.4185419904534,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": 0.9819020845205763
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2121.528458024841,
      "inferenceMs": 68.07866599410772,
      "durationMs": 85.63829198828898,
      "imagesPerSecond": 14.688889469228892,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 617250816,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1452146688,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 85.63829198828898,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": 0.9819020845205763
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2121.528458024841,
      "inferenceMs": 66.69433301431127,
      "durationMs": 84.01883399346843,
      "imagesPerSecond": 14.993777654023768,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 617267200,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1452146688,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 84.01883399346843,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": 0.9819020845205763
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2121.528458024841,
      "inferenceMs": 67.63741699978709,
      "durationMs": 84.97720799641684,
      "imagesPerSecond": 14.784715980551827,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 617267200,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1452130304,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 84.97720799641684,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": 0.9819020845205763
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2121.528458024841,
      "inferenceMs": 67.85908399615437,
      "durationMs": 85.33562498632818,
      "imagesPerSecond": 14.736420551398407,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 617299968,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1452130304,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 85.33562498632818,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": 0.9819020845205763
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2121.528458024841,
      "inferenceMs": 66.12874998245388,
      "durationMs": 84.25374998478219,
      "imagesPerSecond": 15.1220157687138,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 617316352,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1452130304,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 84.25374998478219,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": 0.9819020845205763
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2121.528458024841,
      "inferenceMs": 66.87395798508078,
      "durationMs": 84.31758399819955,
      "imagesPerSecond": 14.953504026531444,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 617316352,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1452130304,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 84.31758399819955,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": 0.9819020845205763
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2121.528458024841,
      "inferenceMs": 67.69495800836012,
      "durationMs": 85.27741700527258,
      "imagesPerSecond": 14.772148907700085,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 617316352,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1452130304,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 85.27741700527258,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": 0.9819020845205763
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2121.528458024841,
      "inferenceMs": 67.69445797544904,
      "durationMs": 84.97029199497774,
      "imagesPerSecond": 14.77225802387949,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 617332736,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1452130304,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 84.97029199497774,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": 0.9819020845205763
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2121.528458024841,
      "inferenceMs": 66.02316698990762,
      "durationMs": 83.59312498942018,
      "imagesPerSecond": 15.146198608631742,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 617332736,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1452130304,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 83.59312498942018,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": 0.9819020845205763
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2121.528458024841,
      "inferenceMs": 66.97004198213108,
      "durationMs": 83.96945800632238,
      "imagesPerSecond": 14.93204977035582,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 617332736,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1452130304,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 83.96945800632238,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": 0.9819020845205763
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2121.528458024841,
      "inferenceMs": 68.04874999215826,
      "durationMs": 85.6047090201173,
      "imagesPerSecond": 14.69534708742243,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 617332736,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1452130304,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 85.6047090201173,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": 0.9819020845205763
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2121.528458024841,
      "inferenceMs": 67.72133300546557,
      "durationMs": 84.90095901652239,
      "imagesPerSecond": 14.766395692761884,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 617332736,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1452130304,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 84.90095901652239,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": 0.9819020845205763
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2121.528458024841,
      "inferenceMs": 67.28187500266358,
      "durationMs": 85.54533400456421,
      "imagesPerSecond": 14.862843818790894,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 617332736,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1452130304,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 85.54533400456421,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": 0.9819020845205763
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2121.528458024841,
      "inferenceMs": 66.36233301833272,
      "durationMs": 84.00908397743478,
      "imagesPerSecond": 15.06878909341159,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 617349120,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1452130304,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 84.00908397743478,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": 0.9819020845205763
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2121.528458024841,
      "inferenceMs": 65.93916600104421,
      "durationMs": 83.97062501171604,
      "imagesPerSecond": 15.165493600331008,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 617349120,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1452130304,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 83.97062501171604,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": 0.9819020845205763
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2121.528458024841,
      "inferenceMs": 67.6014169875998,
      "durationMs": 84.8025829764083,
      "imagesPerSecond": 14.792589335567197,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 617349120,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1452130304,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 84.8025829764083,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": 0.9819020845205763
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2121.528458024841,
      "inferenceMs": 67.42379200295545,
      "durationMs": 85.59304100344889,
      "imagesPerSecond": 14.831559754992215,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 617349120,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1452130304,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 85.59304100344889,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": 0.9819020845205763
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2121.528458024841,
      "inferenceMs": 67.53020800533704,
      "durationMs": 85.122250020504,
      "imagesPerSecond": 14.80818776570284,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 617349120,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1452130304,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 85.122250020504,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": 0.9819020845205763
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2121.528458024841,
      "inferenceMs": 66.50858401553705,
      "durationMs": 83.31037499010563,
      "imagesPerSecond": 15.035653138644333,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 617365504,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1452130304,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 83.31037499010563,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": 0.9819020845205763
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2121.528458024841,
      "inferenceMs": 66.97374998475425,
      "durationMs": 84.03883298160508,
      "imagesPerSecond": 14.931223057207305,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 617365504,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1452130304,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 84.03883298160508,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": 0.9819020845205763
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2121.528458024841,
      "inferenceMs": 65.95354198361747,
      "durationMs": 84.55299999332055,
      "imagesPerSecond": 15.16218795721987,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 617398272,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1452130304,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 84.55299999332055,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": 0.9819020845205763
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2121.528458024841,
      "inferenceMs": 66.98879098985344,
      "durationMs": 84.21966599416919,
      "imagesPerSecond": 14.927870547051171,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 617414656,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1452130304,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 84.21966599416919,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": 0.9819020845205763
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2121.528458024841,
      "inferenceMs": 69.25470801070333,
      "durationMs": 86.32483301335014,
      "imagesPerSecond": 14.439451536572067,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 617414656,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1452130304,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 86.32483301335014,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": 0.9819020845205763
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2121.528458024841,
      "inferenceMs": 67.53400000161491,
      "durationMs": 85.78925000620075,
      "imagesPerSecond": 14.80735629425308,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 617431040,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1452130304,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 85.78925000620075,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": 0.9819020845205763
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2121.528458024841,
      "inferenceMs": 67.75516699417494,
      "durationMs": 85.19675000570714,
      "imagesPerSecond": 14.759021995857115,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 617431040,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1452130304,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 85.19675000570714,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": 0.9819020845205763
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 627.0267499785405,
      "placement": {
        "mps:0": 318
      },
      "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": 2188.0661670002155,
      "inferenceMs": 164.26170800696127,
      "durationMs": 231.25154100125656,
      "imagesPerSecond": 6.087846109317339,
      "visualizationMs": 22.031750006135553,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 612499456,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1452146688,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 858.278290979797,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": 0.9819020845205763
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 594.8345420183614,
      "placement": {
        "mps:0": 318
      },
      "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": 2215.552542009391,
      "inferenceMs": 145.4035000060685,
      "durationMs": 204.32929100934416,
      "imagesPerSecond": 6.877413542027973,
      "visualizationMs": 20.59233299223706,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 610156544,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1452146688,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 799.1638330277056,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": 0.9824125836025239
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 579.0819160174578,
      "placement": {
        "mps:0": 318
      },
      "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": 2150.136582989944,
      "inferenceMs": 138.87475000228733,
      "durationMs": 209.01504199719056,
      "imagesPerSecond": 7.200733034504325,
      "visualizationMs": 21.225416974630207,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 678969344,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1351483392,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 788.0969580146484,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": 0.9824125836025239
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2150.136582989944,
      "inferenceMs": 59.66524998075329,
      "durationMs": 77.47775001917034,
      "imagesPerSecond": 16.760174478822737,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 679100416,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1351483392,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 77.47775001917034,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": 0.9824125836025239
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2150.136582989944,
      "inferenceMs": 59.36950002796948,
      "durationMs": 77.13874999899417,
      "imagesPerSecond": 16.84366551055494,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 684736512,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1351483392,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 77.13874999899417,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": 0.9824125836025239
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2150.136582989944,
      "inferenceMs": 58.40458400780335,
      "durationMs": 75.27929099160247,
      "imagesPerSecond": 17.121943713637126,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 684802048,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1351467008,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 75.27929099160247,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": 0.9824125836025239
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2150.136582989944,
      "inferenceMs": 58.556374977342784,
      "durationMs": 76.82487499550916,
      "imagesPerSecond": 17.077559879465387,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 684834816,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1351467008,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 76.82487499550916,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": 0.9824125836025239
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2150.136582989944,
      "inferenceMs": 57.92266601929441,
      "durationMs": 75.08025001152419,
      "imagesPerSecond": 17.264398701311393,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 684883968,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1351467008,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 75.08025001152419,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": 0.9824125836025239
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 565.8804159902502,
      "placement": {
        "mps:0": 318
      },
      "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": 2144.2422500113025,
      "inferenceMs": 137.46095899841748,
      "durationMs": 195.46679200720973,
      "imagesPerSecond": 7.274792837808679,
      "visualizationMs": 20.5025420000311,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 673234944,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1351483392,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 761.3472079974599,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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": 0.9824125836025239
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 604.9112080072518,
      "placement": {
        "mps:0": 318
      },
      "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": 2171.857708017342,
      "inferenceMs": 152.1764999779407,
      "durationMs": 215.47633298905566,
      "imagesPerSecond": 6.571316859994537,
      "visualizationMs": 21.174166002310812,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 676986880,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1351483392,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 820.3875409963075,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "1d5f47bd3bdd2c4bbfa585418ffe6da5028b4c0b",
      "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 Desktop

56 configurations tested
Passed56
Failed0
TestedNov 11, 2026
Runtimetransformers 5.22.1

Quality criteria

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

PrecisionDeviceResultPlatformInference / load
FP16mpsAcceleratedPassedVery fastmacos · aarch6486 ms / 0 ms
FP16mpsAcceleratedPassedVery fastmacos · aarch6486 ms / 0 ms
FP16mpsAcceleratedPassedVery fastmacos · aarch6487 ms / 0 ms
FP16mpsAcceleratedPassedVery fastmacos · aarch6487 ms / 0 ms
FP16mpsAcceleratedPassedVery fastmacos · aarch6487 ms / 0 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch6497 ms / 0 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch6498 ms / 0 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch6498 ms / 0 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch6498 ms / 0 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch6498 ms / 0 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch6498 ms / 0 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch6498 ms / 0 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch6498 ms / 0 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch6498 ms / 0 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch6499 ms / 0 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch6499 ms / 0 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch6499 ms / 0 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch6499 ms / 0 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch6499 ms / 0 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch6499 ms / 0 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch6499 ms / 0 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch6499 ms / 0 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch6499 ms / 0 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch6499 ms / 0 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch64100 ms / 0 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch64100 ms / 0 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch64100 ms / 0 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch64100 ms / 0 ms
FP32mpsAcceleratedPassedVery fastmacos · aarch64100 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64101 ms / 0 ms
FP16mpsAcceleratedPassedFastmacos · aarch64103 ms / 1688 ms
FP16mpsAcceleratedPassedFastmacos · aarch64167 ms / 708 ms
FP16mpsAcceleratedPassedFastmacos · aarch64167 ms / 704 ms
FP32mpsAcceleratedPassedFastmacos · aarch64178 ms / 732 ms
FP32mpsAcceleratedPassedFastmacos · aarch64182 ms / 731 ms
FP16mpsAcceleratedPassedFastmacos · aarch64185 ms / 724 ms
FP32mpsAcceleratedPassedFastmacos · aarch64200 ms / 741 ms
FP32mpsAcceleratedPassedFastmacos · aarch64273 ms / 1250 ms
FP16mpsAcceleratedPassedFastmacos · aarch64273 ms / 1688 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64542 ms / 1250 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64618 ms / 1250 ms
FP16mpsAcceleratedPassedModeratemacos · aarch64948 ms / 1688 ms
FP32cpuPassedFastmacos · aarch64261 ms / 0 ms
FP32cpuPassedFastmacos · aarch64261 ms / 0 ms
FP32cpuPassedFastmacos · aarch64263 ms / 559 ms
FP32cpuPassedFastmacos · aarch64263 ms / 1284 ms
FP32cpuPassedFastmacos · aarch64264 ms / 0 ms
FP32cpuPassedFastmacos · aarch64265 ms / 0 ms
FP32cpuPassedFastmacos · aarch64266 ms / 600 ms
FP32cpuPassedFastmacos · aarch64267 ms / 0 ms
FP32cpuPassedFastmacos · aarch64270 ms / 598 ms
FP32cpuPassedFastmacos · aarch64271 ms / 598 ms
FP32cpuPassedFastmacos · aarch64324 ms / 598 ms
FP32cpuPassedModeratemacos · aarch64402 ms / 1463 ms
FP32cpuPassedModeratemacos · aarch64419 ms / 1463 ms
FP32cpuPassedModeratemacos · aarch64507 ms / 1463 ms
Test setup & sample details
Quality gate
No explicit quality gate recorded. Review the per-case quality evidence; full certification is pending.
Tested
2026-11-11T22:04:34.393804+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
7d14702e444d98d0b1764824567fc2b45e1eb218
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": "pass",
    "quality": {
      "maximumIoU": 0.9848284045811584,
      "passed": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1463.0022500059567,
      "placement": {
        "cpu": 369
      },
      "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": 3241.3602090091445,
      "inferenceMs": 418.8559579779394,
      "durationMs": 432.31387497507967,
      "imagesPerSecond": 2.3874555941082463,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 1304199168,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "totalMs": 1895.3161249810364,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": "pass",
    "quality": {
      "maximumIoU": 0.923114406793343,
      "passed": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1463.0022500059567,
      "placement": {
        "cpu": 369
      },
      "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": 3241.3602090091445,
      "inferenceMs": 507.1581249940209,
      "durationMs": 523.8586670020595,
      "imagesPerSecond": 1.9717716245042696,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        1199
      ],
      "peakHelperRssBytes": 1645674496,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "totalMs": 1986.8609170080163,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": "pass",
    "quality": {
      "maximumIoU": 0,
      "passed": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1463.0022500059567,
      "placement": {
        "cpu": 369
      },
      "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": 3241.3602090091445,
      "inferenceMs": 401.82699999422766,
      "durationMs": 414.90391598199494,
      "imagesPerSecond": 2.488633168040886,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 1645740032,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "totalMs": 1877.9061659879517,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": "pass",
    "quality": {
      "maximumIoU": 0.9848284045811584,
      "passed": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 597.6294170250185,
      "placement": {
        "cpu": 369
      },
      "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": 2360.3878330031876,
      "inferenceMs": 270.8116669964511,
      "durationMs": 284.0398330008611,
      "imagesPerSecond": 3.6926030960590213,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 1305329664,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 881.6692500258796,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": "pass",
    "quality": {
      "maximumIoU": 0.923114406793343,
      "passed": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 597.6294170250185,
      "placement": {
        "cpu": 369
      },
      "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": 2360.3878330031876,
      "inferenceMs": 323.9017080049962,
      "durationMs": 340.2514159970451,
      "imagesPerSecond": 3.0873563654828735,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        1199
      ],
      "peakHelperRssBytes": 1646657536,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 937.8808330220636,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": "pass",
    "quality": {
      "maximumIoU": 0,
      "passed": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 597.6294170250185,
      "placement": {
        "cpu": 369
      },
      "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": 2360.3878330031876,
      "inferenceMs": 270.0269579945598,
      "durationMs": 282.8401250008028,
      "imagesPerSecond": 3.7033339464578456,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 1650737152,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 880.4695420258213,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": 0.9848284681556776,
      "passed": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1250.3104579809587,
      "placement": {
        "mps:0": 369
      },
      "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": 2957.486958010122,
      "inferenceMs": 618.3304580044933,
      "durationMs": 676.8838749849238,
      "imagesPerSecond": 1.617258194311258,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 614514688,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1527627776,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 1927.1943329658825,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": 0.9231146298647251,
      "passed": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1250.3104579809587,
      "placement": {
        "mps:0": 369
      },
      "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": 2957.486958010122,
      "inferenceMs": 542.23470800207,
      "durationMs": 576.0907499934547,
      "imagesPerSecond": 1.844219828134245,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        1199
      ],
      "peakHelperRssBytes": 711327744,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1629732864,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 1826.4012079744134,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1250.3104579809587,
      "placement": {
        "mps:0": 369
      },
      "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": 2957.486958010122,
      "inferenceMs": 272.69291700213216,
      "durationMs": 288.56812498997897,
      "imagesPerSecond": 3.667128618497198,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 717553664,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1629732864,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 1538.8785829709377,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": 0.9853646791872573,
      "passed": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1687.904417020036,
      "placement": {
        "mps:0": 369
      },
      "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": 3490.045915998053,
      "inferenceMs": 948.4292920096777,
      "durationMs": 992.3407500027679,
      "imagesPerSecond": 1.054374857909593,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 726597632,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1368244224,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 2680.245167022804,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": 0.9226396220306777,
      "passed": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1687.904417020036,
      "placement": {
        "mps:0": 369
      },
      "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": 3490.045915998053,
      "inferenceMs": 273.3020409941673,
      "durationMs": 301.8154999881517,
      "imagesPerSecond": 3.658955477838314,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        1199
      ],
      "peakHelperRssBytes": 817201152,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1470349312,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 1989.7199170081876,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1687.904417020036,
      "placement": {
        "mps:0": 369
      },
      "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": 3490.045915998053,
      "inferenceMs": 102.57195800659247,
      "durationMs": 118.1834579911083,
      "imagesPerSecond": 9.74925329918854,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 817905664,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1470349312,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 1806.0878750111442,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9848284045811584
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1284.2833749891724,
      "placement": {
        "cpu": 369
      },
      "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": 2892.1151669928804,
      "inferenceMs": 263.03391699912027,
      "durationMs": 296.19633400579914,
      "imagesPerSecond": 3.801791082339182,
      "visualizationMs": 20.036625006468967,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "maskDevice": null,
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 1313587200,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 1580.4797089949716,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9848284045811584
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "cpu": 369
      },
      "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": 2892.1151669928804,
      "inferenceMs": 261.2408749992028,
      "durationMs": 273.9606249961071,
      "imagesPerSecond": 3.8278848974267583,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "maskDevice": null,
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 1313767424,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 273.9606249961071,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9848284045811584
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "cpu": 369
      },
      "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": 2892.1151669928804,
      "inferenceMs": 261.0307919967454,
      "durationMs": 273.97158299572766,
      "imagesPerSecond": 3.8309656586892946,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "maskDevice": null,
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 1313783808,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 273.97158299572766,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9848284045811584
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "cpu": 369
      },
      "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": 2892.1151669928804,
      "inferenceMs": 264.37162500224076,
      "durationMs": 276.9771250023041,
      "imagesPerSecond": 3.7825541980593576,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "maskDevice": null,
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 1313800192,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 276.9771250023041,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9848284045811584
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "cpu": 369
      },
      "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": 2892.1151669928804,
      "inferenceMs": 265.48816700233147,
      "durationMs": 278.0779589957092,
      "imagesPerSecond": 3.7666462173857194,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "maskDevice": null,
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 1313800192,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 278.0779589957092,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9848284045811584
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "cpu": 369
      },
      "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": 2892.1151669928804,
      "inferenceMs": 266.75662500201724,
      "durationMs": 279.4781250122469,
      "imagesPerSecond": 3.7487353875182587,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "maskDevice": null,
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 1313800192,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 279.4781250122469,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9848284045811584
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 558.5563750064466,
      "placement": {
        "cpu": 369
      },
      "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": 2132.9805409768596,
      "inferenceMs": 262.58441698155366,
      "durationMs": 295.49433299689554,
      "imagesPerSecond": 3.8082991043228938,
      "visualizationMs": 20.14883400988765,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "maskDevice": null,
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 1309065216,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 854.0507080033422,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9848284045811584
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 599.8890829796437,
      "placement": {
        "cpu": 369
      },
      "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": 2174.459083005786,
      "inferenceMs": 265.68650000263005,
      "durationMs": 298.68058301508427,
      "imagesPerSecond": 3.7638344439408886,
      "visualizationMs": 20.20449997507967,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "maskDevice": null,
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 1303134208,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 898.569665994728,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": 0.9848284681556776
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 740.6499159988016,
      "placement": {
        "mps:0": 369
      },
      "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": 2325.738875020761,
      "inferenceMs": 199.92529199225828,
      "durationMs": 265.0477499992121,
      "imagesPerSecond": 5.001868398115048,
      "visualizationMs": 21.33883399073966,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 615137280,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1527644160,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 1005.6976659980137,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": 0.9848284681556776
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2325.738875020761,
      "inferenceMs": 98.90341700520366,
      "durationMs": 117.0667500118725,
      "imagesPerSecond": 10.110874126294204,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 616562688,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1527644160,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 117.0667500118725,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": 0.9848284681556776
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2325.738875020761,
      "inferenceMs": 99.12716702092439,
      "durationMs": 117.68079199828207,
      "imagesPerSecond": 10.088051843435752,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 616841216,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1527627776,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 117.68079199828207,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": 0.9848284681556776
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2325.738875020761,
      "inferenceMs": 98.84170899749734,
      "durationMs": 116.49870799737982,
      "imagesPerSecond": 10.117186460478136,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 616841216,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1527627776,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 116.49870799737982,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": 0.9848284681556776
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2325.738875020761,
      "inferenceMs": 99.58037501201034,
      "durationMs": 118.00062502152286,
      "imagesPerSecond": 10.042139325940383,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 616841216,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1527627776,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 118.00062502152286,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": 0.9848284681556776
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2325.738875020761,
      "inferenceMs": 97.60604202165268,
      "durationMs": 115.51333300303668,
      "imagesPerSecond": 10.245267396235189,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 616841216,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1527627776,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 115.51333300303668,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": 0.9848284681556776
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2325.738875020761,
      "inferenceMs": 98.38558398769237,
      "durationMs": 117.16166700352915,
      "imagesPerSecond": 10.164090707893708,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 616857600,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1527627776,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 117.16166700352915,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": 0.9848284681556776
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2325.738875020761,
      "inferenceMs": 98.77220800262876,
      "durationMs": 117.11133300559595,
      "imagesPerSecond": 10.12430541163346,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 616857600,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1527627776,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 117.11133300559595,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": 0.9848284681556776
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2325.738875020761,
      "inferenceMs": 97.86300000268966,
      "durationMs": 115.29470898676664,
      "imagesPerSecond": 10.218366491651759,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 616890368,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1527627776,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 115.29470898676664,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": 0.9848284681556776
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2325.738875020761,
      "inferenceMs": 97.96783400815912,
      "durationMs": 115.75245799031109,
      "imagesPerSecond": 10.207431960950737,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 616906752,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1527627776,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 115.75245799031109,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": 0.9848284681556776
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2325.738875020761,
      "inferenceMs": 97.54975000396371,
      "durationMs": 114.2800000088755,
      "imagesPerSecond": 10.251179525927716,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 616906752,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1527627776,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 114.2800000088755,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": 0.9848284681556776
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2325.738875020761,
      "inferenceMs": 99.91162500227802,
      "durationMs": 117.74300000979565,
      "imagesPerSecond": 10.008845316820738,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 616923136,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1527627776,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 117.74300000979565,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": 0.9848284681556776
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2325.738875020761,
      "inferenceMs": 99.18895800365135,
      "durationMs": 117.5500419922173,
      "imagesPerSecond": 10.081767367323165,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 616972288,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1527627776,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 117.5500419922173,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": 0.9848284681556776
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2325.738875020761,
      "inferenceMs": 99.19700000318699,
      "durationMs": 116.44179199356586,
      "imagesPerSecond": 10.080950028406829,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 616972288,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1527627776,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 116.44179199356586,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": 0.9848284681556776
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2325.738875020761,
      "inferenceMs": 97.3745420051273,
      "durationMs": 114.21649999101646,
      "imagesPerSecond": 10.269624682263919,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 616988672,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1527627776,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 114.21649999101646,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": 0.9848284681556776
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2325.738875020761,
      "inferenceMs": 98.82324998034164,
      "durationMs": 116.42249999567866,
      "imagesPerSecond": 10.119076231543938,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 617021440,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1527627776,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 116.42249999567866,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": 0.9848284681556776
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2325.738875020761,
      "inferenceMs": 99.35025000595488,
      "durationMs": 116.78650000249036,
      "imagesPerSecond": 10.065399935481409,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 617037824,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1527627776,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 116.78650000249036,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": 0.9848284681556776
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2325.738875020761,
      "inferenceMs": 98.74279200448655,
      "durationMs": 116.24020800809376,
      "imagesPerSecond": 10.127321495573705,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 617037824,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1527627776,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 116.24020800809376,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": 0.9848284681556776
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2325.738875020761,
      "inferenceMs": 99.5405420253519,
      "durationMs": 116.48004199378192,
      "imagesPerSecond": 10.046157873495513,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 617037824,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1527627776,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 116.48004199378192,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": 0.9848284681556776
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2325.738875020761,
      "inferenceMs": 98.02954099723138,
      "durationMs": 116.71608299366198,
      "imagesPerSecond": 10.201006653986504,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 617037824,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1527627776,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 116.71608299366198,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": 0.9848284681556776
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2325.738875020761,
      "inferenceMs": 99.73737501422875,
      "durationMs": 118.66066700895317,
      "imagesPerSecond": 10.02633165207464,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 617054208,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1527627776,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 118.66066700895317,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": 0.9848284681556776
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2325.738875020761,
      "inferenceMs": 99.33729202020913,
      "durationMs": 117.84337501740083,
      "imagesPerSecond": 10.066712909755587,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 617070592,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1527627776,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 117.84337501740083,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": 0.9848284681556776
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2325.738875020761,
      "inferenceMs": 97.90158399846405,
      "durationMs": 116.1320419923868,
      "imagesPerSecond": 10.214339330972303,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 617070592,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1527627776,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 116.1320419923868,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": 0.9848284681556776
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2325.738875020761,
      "inferenceMs": 97.67608399852179,
      "durationMs": 114.6793749940116,
      "imagesPerSecond": 10.23792067682744,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 617070592,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1527627776,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 114.6793749940116,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": 0.9848284681556776
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2325.738875020761,
      "inferenceMs": 100.85595797863789,
      "durationMs": 118.0830410157796,
      "imagesPerSecond": 9.915130648125004,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 617070592,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1527627776,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 118.0830410157796,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": 0.9848284681556776
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2325.738875020761,
      "inferenceMs": 99.58916701725684,
      "durationMs": 116.98133399477229,
      "imagesPerSecond": 10.041252778293845,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 617070592,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1527627776,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 116.98133399477229,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": 0.9848284681556776
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 730.6192500109319,
      "placement": {
        "mps:0": 369
      },
      "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": 2306.9574999972247,
      "inferenceMs": 181.8332920083776,
      "durationMs": 240.1525409950409,
      "imagesPerSecond": 5.499542954729803,
      "visualizationMs": 20.330250001279637,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 615464960,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1527644160,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 970.7717910059728,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": 0.9848284681556776
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 731.6252500168048,
      "placement": {
        "mps:0": 369
      },
      "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": 2315.8992500102613,
      "inferenceMs": 178.01354097900912,
      "durationMs": 236.2166660022922,
      "imagesPerSecond": 5.617550184667791,
      "visualizationMs": 20.373832987388596,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 612761600,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1527644160,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 967.841916019097,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": 0.9853646791872573
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 724.0111250139307,
      "placement": {
        "mps:0": 369
      },
      "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": 2330.1502499962226,
      "inferenceMs": 184.57708301139064,
      "durationMs": 244.93554097716697,
      "imagesPerSecond": 5.417790679562792,
      "visualizationMs": 20.446082984562963,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 723697664,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1368260608,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 968.9466659910977,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": 0.9853646791872573
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2330.1502499962226,
      "inferenceMs": 86.06754097854719,
      "durationMs": 104.1244590014685,
      "imagesPerSecond": 11.618782047569542,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 723697664,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1368260608,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 104.1244590014685,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": 0.9853646791872573
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2330.1502499962226,
      "inferenceMs": 87.0956250000745,
      "durationMs": 104.8297920206096,
      "imagesPerSecond": 11.481632975240084,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 723697664,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1368260608,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 104.8297920206096,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": 0.9853646791872573
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2330.1502499962226,
      "inferenceMs": 87.36970802419819,
      "durationMs": 105.404333997285,
      "imagesPerSecond": 11.445614534078985,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 726237184,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1368244224,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 105.404333997285,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": 0.9853646791872573
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2330.1502499962226,
      "inferenceMs": 86.15466600167565,
      "durationMs": 104.19333400204778,
      "imagesPerSecond": 11.607032403567679,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 727220224,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1368244224,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 104.19333400204778,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": 0.9853646791872573
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2330.1502499962226,
      "inferenceMs": 87.12520799599588,
      "durationMs": 104.05454199644737,
      "imagesPerSecond": 11.47773443531932,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 727220224,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1368244224,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 104.05454199644737,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": 0.9853646791872573
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 707.5283330050297,
      "placement": {
        "mps:0": 369
      },
      "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": 2297.1959999995306,
      "inferenceMs": 166.83224998996593,
      "durationMs": 227.13620902504772,
      "imagesPerSecond": 5.99404491673609,
      "visualizationMs": 20.427583978744224,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 723566592,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1368260608,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 934.6645420300774,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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": 0.9853646791872573
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 703.5427919763606,
      "placement": {
        "mps:0": 369
      },
      "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": 2244.7554169921204,
      "inferenceMs": 167.1148750174325,
      "durationMs": 226.91362499608658,
      "imagesPerSecond": 5.983907775389148,
      "visualizationMs": 20.40149999083951,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 723746816,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 1368260608,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 930.4564169724472,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "7d14702e444d98d0b1764824567fc2b45e1eb218",
      "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-dc5 Desktop

56 configurations tested
Passed56
Failed0
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 · aarch64193 ms / 0 ms
FP16mpsAcceleratedPassedFastmacos · aarch64194 ms / 0 ms
FP16mpsAcceleratedPassedFastmacos · aarch64194 ms / 0 ms
FP16mpsAcceleratedPassedFastmacos · aarch64194 ms / 0 ms
FP16mpsAcceleratedPassedFastmacos · aarch64195 ms / 0 ms
FP16mpsAcceleratedPassedFastmacos · aarch64196 ms / 1012 ms
FP32mpsAcceleratedPassedFastmacos · aarch64209 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64210 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64210 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64210 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64210 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64211 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64211 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64211 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64211 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64211 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64211 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64211 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64212 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64212 ms / 1076 ms
FP32mpsAcceleratedPassedFastmacos · aarch64212 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64212 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64212 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64212 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64212 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64213 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64213 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64213 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64213 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64213 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64213 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64213 ms / 0 ms
FP16mpsAcceleratedPassedFastmacos · aarch64279 ms / 844 ms
FP16mpsAcceleratedPassedFastmacos · aarch64279 ms / 842 ms
FP16mpsAcceleratedPassedFastmacos · aarch64281 ms / 846 ms
FP32mpsAcceleratedPassedFastmacos · aarch64294 ms / 868 ms
FP32mpsAcceleratedPassedFastmacos · aarch64298 ms / 873 ms
FP32mpsAcceleratedPassedFastmacos · aarch64317 ms / 900 ms
FP16mpsAcceleratedPassedModeratemacos · aarch64386 ms / 1012 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64410 ms / 1076 ms
FP16mpsAcceleratedPassedModeratemacos · aarch64446 ms / 1012 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64476 ms / 1076 ms
FP32cpuPassedModeratemacos · aarch64416 ms / 0 ms
FP32cpuPassedModeratemacos · aarch64417 ms / 771 ms
FP32cpuPassedModeratemacos · aarch64417 ms / 756 ms
FP32cpuPassedModeratemacos · aarch64417 ms / 0 ms
FP32cpuPassedModeratemacos · aarch64421 ms / 749 ms
FP32cpuPassedModeratemacos · aarch64421 ms / 0 ms
FP32cpuPassedModeratemacos · aarch64422 ms / 1410 ms
FP32cpuPassedModeratemacos · aarch64422 ms / 0 ms
FP32cpuPassedModeratemacos · aarch64424 ms / 0 ms
FP32cpuPassedModeratemacos · aarch64427 ms / 771 ms
FP32cpuPassedModeratemacos · aarch64542 ms / 771 ms
FP32cpuPassedModeratemacos · aarch64790 ms / 1900 ms
FP32cpuPassedModeratemacos · aarch64797 ms / 1900 ms
FP32cpuPassedModeratemacos · aarch641070 ms / 1900 ms
Test setup & sample details
Quality gate
No explicit quality gate recorded. Review the per-case quality evidence; full certification is pending.
Tested
2026-11-11T22:05:11.436419+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
96317ca979e231bd960cb3cac31328e0165a3e94
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": "pass",
    "quality": {
      "maximumIoU": 0.9907821288840902,
      "passed": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1900.492332992144,
      "placement": {
        "cpu": 369
      },
      "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": 3432.891416014172,
      "inferenceMs": 797.0087499998044,
      "durationMs": 810.4248750023544,
      "imagesPerSecond": 1.2546913694488868,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 1350139904,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "totalMs": 2710.9172079944983,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": "pass",
    "quality": {
      "maximumIoU": 0.9280410684502828,
      "passed": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1900.492332992144,
      "placement": {
        "cpu": 369
      },
      "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": 3432.891416014172,
      "inferenceMs": 1069.7403750091325,
      "durationMs": 1086.2444170052186,
      "imagesPerSecond": 0.9348062608102109,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        1199
      ],
      "peakHelperRssBytes": 1733918720,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "totalMs": 2986.7367499973625,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": "pass",
    "quality": {
      "maximumIoU": 0,
      "passed": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1900.492332992144,
      "placement": {
        "cpu": 369
      },
      "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": 3432.891416014172,
      "inferenceMs": 790.0595829996746,
      "durationMs": 802.659125009086,
      "imagesPerSecond": 1.265727321733419,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 1739538432,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "totalMs": 2703.15145800123,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": "pass",
    "quality": {
      "maximumIoU": 0.9907821288840902,
      "passed": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 771.2485000083689,
      "placement": {
        "cpu": 369
      },
      "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": 2367.861791019095,
      "inferenceMs": 426.61958301323466,
      "durationMs": 439.8847080010455,
      "imagesPerSecond": 2.344008666777441,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 1350123520,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 1211.1332080094144,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": "pass",
    "quality": {
      "maximumIoU": 0.9280410684502828,
      "passed": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 771.2485000083689,
      "placement": {
        "cpu": 369
      },
      "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": 2367.861791019095,
      "inferenceMs": 542.2359590011183,
      "durationMs": 559.484708996024,
      "imagesPerSecond": 1.8442155733126833,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        1199
      ],
      "peakHelperRssBytes": 1733984256,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 1330.7332090043928,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": "pass",
    "quality": {
      "maximumIoU": 0,
      "passed": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 771.2485000083689,
      "placement": {
        "cpu": 369
      },
      "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": 2367.861791019095,
      "inferenceMs": 416.84062499552965,
      "durationMs": 429.49112501810305,
      "imagesPerSecond": 2.398998418186146,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 1738653696,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 1200.739625026472,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": 0.9907821106971505,
      "passed": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1076.3009999936912,
      "placement": {
        "mps:0": 369
      },
      "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": 2687.771625001915,
      "inferenceMs": 410.1061249966733,
      "durationMs": 449.9364999937825,
      "imagesPerSecond": 2.4383932329908795,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 620298240,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3197583360,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 1526.2374999874737,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": 0.928041407047093,
      "passed": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1076.3009999936912,
      "placement": {
        "mps:0": 369
      },
      "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": 2687.771625001915,
      "inferenceMs": 476.07429200434126,
      "durationMs": 503.8151249755174,
      "imagesPerSecond": 2.1005124973034275,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        1199
      ],
      "peakHelperRssBytes": 712376320,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5646811136,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 1580.1161249692086,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1076.3009999936912,
      "placement": {
        "mps:0": 369
      },
      "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": 2687.771625001915,
      "inferenceMs": 211.90491699962877,
      "durationMs": 227.10774998995475,
      "imagesPerSecond": 4.719097669648467,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 713146368,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5646811136,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 1303.408749983646,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": 0.9903103209716845,
      "passed": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1012.2654580045491,
      "placement": {
        "mps:0": 369
      },
      "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": 2587.1532910095993,
      "inferenceMs": 385.90312501764856,
      "durationMs": 426.38737498782575,
      "imagesPerSecond": 2.5913239234698264,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 723369984,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3042394112,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 1438.652832992375,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": 0.9273376455657704,
      "passed": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1012.2654580045491,
      "placement": {
        "mps:0": 369
      },
      "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": 2587.1532910095993,
      "inferenceMs": 446.0834579949733,
      "durationMs": 472.8069579869043,
      "imagesPerSecond": 2.24173298085236,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        1199
      ],
      "peakHelperRssBytes": 803160064,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 4417880064,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 1485.0724159914535,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1012.2654580045491,
      "placement": {
        "mps:0": 369
      },
      "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": 2587.1532910095993,
      "inferenceMs": 195.51770901307464,
      "durationMs": 210.78029202180915,
      "imagesPerSecond": 5.114626214923213,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 803831808,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 4417880064,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 1223.0457500263583,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9907821288840902
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1410.1239999872632,
      "placement": {
        "cpu": 369
      },
      "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": 3003.037874994334,
      "inferenceMs": 422.13758299476467,
      "durationMs": 456.63587501621805,
      "imagesPerSecond": 2.368895924654977,
      "visualizationMs": 21.730874985223636,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "maskDevice": null,
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 1350336512,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 1866.7598750034813,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9907821288840902
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "cpu": 369
      },
      "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": 3003.037874994334,
      "inferenceMs": 421.2247079994995,
      "durationMs": 434.29966701660305,
      "imagesPerSecond": 2.3740297779521238,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "maskDevice": null,
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 1350631424,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 434.29966701660305,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9907821288840902
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "cpu": 369
      },
      "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": 3003.037874994334,
      "inferenceMs": 423.88854201999493,
      "durationMs": 436.7271660012193,
      "imagesPerSecond": 2.3591107115908545,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "maskDevice": null,
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 1350991872,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 436.7271660012193,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9907821288840902
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "cpu": 369
      },
      "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": 3003.037874994334,
      "inferenceMs": 422.33079197467305,
      "durationMs": 435.4960419877898,
      "imagesPerSecond": 2.367812196037957,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "maskDevice": null,
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 1356873728,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 435.4960419877898,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9907821288840902
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "cpu": 369
      },
      "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": 3003.037874994334,
      "inferenceMs": 415.68849998293445,
      "durationMs": 428.21645899675786,
      "imagesPerSecond": 2.4056474981652216,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "maskDevice": null,
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 1357807616,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 428.21645899675786,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9907821288840902
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "cpu": 369
      },
      "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": 3003.037874994334,
      "inferenceMs": 417.41991601884365,
      "durationMs": 429.8548339866102,
      "imagesPerSecond": 2.3956691131020613,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "maskDevice": null,
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 1357807616,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 429.8548339866102,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9907821288840902
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 755.9385840140749,
      "placement": {
        "cpu": 369
      },
      "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": 2381.8180829985067,
      "inferenceMs": 416.96462500840425,
      "durationMs": 450.9427080047317,
      "imagesPerSecond": 2.3982849863578815,
      "visualizationMs": 21.154959016712382,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "maskDevice": null,
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 1350778880,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 1206.8812920188066,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.9907821288840902
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 749.3160839949269,
      "placement": {
        "cpu": 369
      },
      "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": 2405.7257910026237,
      "inferenceMs": 420.81808400689624,
      "durationMs": 453.6941249971278,
      "imagesPerSecond": 2.3763237322843103,
      "visualizationMs": 20.080917020095512,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "maskDevice": null,
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 1346764800,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 1203.0102089920547,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": 0.9907821106971505
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 899.9603329866659,
      "placement": {
        "mps:0": 369
      },
      "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": 2502.653916977579,
      "inferenceMs": 316.9858749897685,
      "durationMs": 381.9026249984745,
      "imagesPerSecond": 3.154714701506108,
      "visualizationMs": 20.86925000185147,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 618414080,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3197583360,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 1281.8629579851404,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": 0.9907821106971505
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2502.653916977579,
      "inferenceMs": 211.61499997833744,
      "durationMs": 230.37075001047924,
      "imagesPerSecond": 4.725562933168102,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 623230976,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3197583360,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 230.37075001047924,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": 0.9907821106971505
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2502.653916977579,
      "inferenceMs": 212.80591600225307,
      "durationMs": 231.19808299816214,
      "imagesPerSecond": 4.699117481251849,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 623525888,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3197583360,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 231.19808299816214,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": 0.9907821106971505
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2502.653916977579,
      "inferenceMs": 212.7810830133967,
      "durationMs": 230.43916700407863,
      "imagesPerSecond": 4.699665899985291,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 623706112,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3197583360,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 230.43916700407863,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": 0.9907821106971505
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2502.653916977579,
      "inferenceMs": 210.98037497722544,
      "durationMs": 227.99395900801755,
      "imagesPerSecond": 4.7397773376217875,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 623755264,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3197583360,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 227.99395900801755,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": 0.9907821106971505
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2502.653916977579,
      "inferenceMs": 212.05770800588652,
      "durationMs": 229.94137500063516,
      "imagesPerSecond": 4.715697483499355,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 623788032,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3197583360,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 229.94137500063516,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": 0.9907821106971505
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2502.653916977579,
      "inferenceMs": 210.08754198555835,
      "durationMs": 227.8947919839993,
      "imagesPerSecond": 4.759920510035484,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 629276672,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3197583360,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 227.8947919839993,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": 0.9907821106971505
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2502.653916977579,
      "inferenceMs": 212.70000000367872,
      "durationMs": 230.60145898489282,
      "imagesPerSecond": 4.701457451728747,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 629276672,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3197583360,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 230.60145898489282,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": 0.9907821106971505
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2502.653916977579,
      "inferenceMs": 211.24920799047686,
      "durationMs": 228.40504100895487,
      "imagesPerSecond": 4.733745558208579,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 629325824,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3197583360,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 228.40504100895487,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": 0.9907821106971505
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2502.653916977579,
      "inferenceMs": 212.2103750007227,
      "durationMs": 230.80104199470952,
      "imagesPerSecond": 4.712304947374012,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 629358592,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3197583360,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 230.80104199470952,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": 0.9907821106971505
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2502.653916977579,
      "inferenceMs": 212.23433400155045,
      "durationMs": 230.6584999896586,
      "imagesPerSecond": 4.711772978224601,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 629374976,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3197583360,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 230.6584999896586,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": 0.9907821106971505
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2502.653916977579,
      "inferenceMs": 212.98091701464728,
      "durationMs": 231.54579100082628,
      "imagesPerSecond": 4.695256335717755,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 629391360,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3197583360,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 231.54579100082628,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": 0.9907821106971505
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2502.653916977579,
      "inferenceMs": 211.2157919909805,
      "durationMs": 229.07229198608547,
      "imagesPerSecond": 4.734494473986598,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 629407744,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3197583360,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 229.07229198608547,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": 0.9907821106971505
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2502.653916977579,
      "inferenceMs": 210.422792006284,
      "durationMs": 227.37712497473694,
      "imagesPerSecond": 4.752336904502894,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 629424128,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3197583360,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 227.37712497473694,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": 0.9907821106971505
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2502.653916977579,
      "inferenceMs": 213.02475000265986,
      "durationMs": 230.1521670015063,
      "imagesPerSecond": 4.694290217392645,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 629473280,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3197583360,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 230.1521670015063,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": 0.9907821106971505
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2502.653916977579,
      "inferenceMs": 211.24104200862348,
      "durationMs": 228.41716700349934,
      "imagesPerSecond": 4.733928551437353,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 629473280,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3197583360,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 228.41716700349934,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": 0.9907821106971505
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2502.653916977579,
      "inferenceMs": 211.32254201802425,
      "durationMs": 229.38100001192652,
      "imagesPerSecond": 4.732102834134502,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 629473280,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3197583360,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 229.38100001192652,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": 0.9907821106971505
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2502.653916977579,
      "inferenceMs": 209.4633340020664,
      "durationMs": 226.1886660126038,
      "imagesPerSecond": 4.774105237865329,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 629473280,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3197583360,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 226.1886660126038,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": 0.9907821106971505
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2502.653916977579,
      "inferenceMs": 209.7664590110071,
      "durationMs": 226.33958401274867,
      "imagesPerSecond": 4.767206371861037,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 629489664,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3185000448,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 226.33958401274867,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": 0.9907821106971505
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2502.653916977579,
      "inferenceMs": 213.34645798197016,
      "durationMs": 230.57375001371838,
      "imagesPerSecond": 4.6872116343478725,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 629489664,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3185000448,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 230.57375001371838,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": 0.9907821106971505
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2502.653916977579,
      "inferenceMs": 210.14720801031217,
      "durationMs": 227.40570901078172,
      "imagesPerSecond": 4.758569050086684,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 629489664,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3185000448,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 227.40570901078172,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": 0.9907821106971505
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2502.653916977579,
      "inferenceMs": 212.10420900024474,
      "durationMs": 230.40808297810145,
      "imagesPerSecond": 4.71466363026698,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 629522432,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3185000448,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 230.40808297810145,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": 0.9907821106971505
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2502.653916977579,
      "inferenceMs": 210.6805829971563,
      "durationMs": 228.99995898478664,
      "imagesPerSecond": 4.746521894775171,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 631963648,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3185000448,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 228.99995898478664,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": 0.9907821106971505
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2502.653916977579,
      "inferenceMs": 212.1601249964442,
      "durationMs": 229.34175000409596,
      "imagesPerSecond": 4.713421054105996,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 636190720,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3185000448,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 229.34175000409596,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": 0.9907821106971505
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2502.653916977579,
      "inferenceMs": 212.6123330090195,
      "durationMs": 229.59183299099095,
      "imagesPerSecond": 4.703396015872596,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 636190720,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3185000448,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 229.59183299099095,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": 0.9907821106971505
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2502.653916977579,
      "inferenceMs": 211.26391598954797,
      "durationMs": 229.8514169815462,
      "imagesPerSecond": 4.7334159992067635,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 636207104,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3185000448,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 229.8514169815462,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": 0.9907821106971505
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 868.1346249941271,
      "placement": {
        "mps:0": 369
      },
      "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": 2402.792375010904,
      "inferenceMs": 294.2640830005985,
      "durationMs": 353.78566599683836,
      "imagesPerSecond": 3.398308042908404,
      "visualizationMs": 20.19520799512975,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 616431616,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3197583360,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 1221.9202909909654,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": 0.9907821106971505
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 872.6285829907283,
      "placement": {
        "mps:0": 369
      },
      "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": 2609.9996250122786,
      "inferenceMs": 297.75979099213146,
      "durationMs": 357.0892910065595,
      "imagesPerSecond": 3.3584118146644784,
      "visualizationMs": 20.500624988926575,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 616595456,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3197583360,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 1229.7178739972878,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": 0.9903103209716845
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 845.5439580138773,
      "placement": {
        "mps:0": 369
      },
      "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": 2485.7344579941127,
      "inferenceMs": 281.49525000480935,
      "durationMs": 340.2762920013629,
      "imagesPerSecond": 3.5524578122824986,
      "visualizationMs": 20.200875005684793,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 724140032,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3042410496,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 1185.8202500152402,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": 0.9903103209716845
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2485.7344579941127,
      "inferenceMs": 195.3058330109343,
      "durationMs": 212.45829097460955,
      "imagesPerSecond": 5.120174777084177,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 724140032,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3042394112,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 212.45829097460955,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": 0.9903103209716845
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2485.7344579941127,
      "inferenceMs": 194.1790840064641,
      "durationMs": 213.24933398864232,
      "imagesPerSecond": 5.149885246995555,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 724140032,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3042394112,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 213.24933398864232,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": 0.9903103209716845
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2485.7344579941127,
      "inferenceMs": 193.64354200661182,
      "durationMs": 211.05962499859743,
      "imagesPerSecond": 5.164127807401166,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 724140032,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3042394112,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 211.05962499859743,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": 0.9903103209716845
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2485.7344579941127,
      "inferenceMs": 192.59812499512918,
      "durationMs": 210.18916700268164,
      "imagesPerSecond": 5.1921585426924075,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 724140032,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3042394112,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 210.18916700268164,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": 0.9903103209716845
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 369
      },
      "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": 2485.7344579941127,
      "inferenceMs": 193.77129198983312,
      "durationMs": 211.58154201111756,
      "imagesPerSecond": 5.160723189338431,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 724140032,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3042394112,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 211.58154201111756,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": 0.9903103209716845
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 843.7968340003863,
      "placement": {
        "mps:0": 369
      },
      "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": 2462.918500008527,
      "inferenceMs": 278.82625002530403,
      "durationMs": 339.29983401321806,
      "imagesPerSecond": 3.586462895474325,
      "visualizationMs": 20.388749981066212,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 724287488,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3042410496,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 1183.0966680136044,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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": 0.9903103209716845
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 842.4543339933734,
      "placement": {
        "mps:0": 369
      },
      "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": 2541.1532500002068,
      "inferenceMs": 279.239041992696,
      "durationMs": 339.1970829979982,
      "imagesPerSecond": 3.581161118673931,
      "visualizationMs": 20.330833998741582,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 725467136,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3042410496,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 1181.6514169913717,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "96317ca979e231bd960cb3cac31328e0165a3e94",
      "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 Desktop

56 configurations tested
Passed56
Failed0
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 · aarch64165 ms / 726 ms
FP16mpsAcceleratedPassedFastmacos · aarch64166 ms / 0 ms
FP16mpsAcceleratedPassedFastmacos · aarch64166 ms / 0 ms
FP16mpsAcceleratedPassedFastmacos · aarch64166 ms / 0 ms
FP16mpsAcceleratedPassedFastmacos · aarch64167 ms / 0 ms
FP16mpsAcceleratedPassedFastmacos · aarch64167 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64178 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64180 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64180 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64180 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64180 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64180 ms / 748 ms
FP32mpsAcceleratedPassedFastmacos · aarch64180 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64181 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64181 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64181 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64181 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64181 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64181 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64181 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64181 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64182 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64182 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64182 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64182 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64182 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64182 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64182 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64183 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64183 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64183 ms / 0 ms
FP32mpsAcceleratedPassedFastmacos · aarch64184 ms / 0 ms
FP16mpsAcceleratedPassedFastmacos · aarch64253 ms / 726 ms
FP16mpsAcceleratedPassedFastmacos · aarch64254 ms / 730 ms
FP16mpsAcceleratedPassedFastmacos · aarch64254 ms / 749 ms
FP16mpsAcceleratedPassedFastmacos · aarch64259 ms / 752 ms
FP32mpsAcceleratedPassedFastmacos · aarch64265 ms / 747 ms
FP32mpsAcceleratedPassedFastmacos · aarch64266 ms / 757 ms
FP32mpsAcceleratedPassedFastmacos · aarch64267 ms / 756 ms
FP32mpsAcceleratedPassedFastmacos · aarch64267 ms / 748 ms
FP16mpsAcceleratedPassedFastmacos · aarch64323 ms / 726 ms
FP32mpsAcceleratedPassedModeratemacos · aarch64362 ms / 748 ms
FP32cpuPassedFastmacos · aarch64329 ms / 618 ms
FP32cpuPassedFastmacos · aarch64330 ms / 618 ms
FP32cpuPassedFastmacos · aarch64331 ms / 619 ms
FP32cpuPassedFastmacos · aarch64332 ms / 0 ms
FP32cpuPassedFastmacos · aarch64332 ms / 0 ms
FP32cpuPassedFastmacos · aarch64333 ms / 0 ms
FP32cpuPassedFastmacos · aarch64334 ms / 0 ms
FP32cpuPassedFastmacos · aarch64334 ms / 916 ms
FP32cpuPassedFastmacos · aarch64335 ms / 0 ms
FP32cpuPassedFastmacos · aarch64335 ms / 628 ms
FP32cpuPassedModeratemacos · aarch64438 ms / 618 ms
FP32cpuPassedModeratemacos · aarch64691 ms / 1383 ms
FP32cpuPassedModeratemacos · aarch64698 ms / 1383 ms
FP32cpuPassedModeratemacos · aarch64951 ms / 1383 ms
Test setup & sample details
Quality gate
No explicit quality gate recorded. Review the per-case quality evidence; full certification is pending.
Tested
2026-11-11T22:06:46.906850+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
6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9
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": "pass",
    "quality": {
      "maximumIoU": 0.987423113518462,
      "passed": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1382.9001669946592,
      "placement": {
        "cpu": 318
      },
      "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": 2985.1368340023328,
      "inferenceMs": 698.3067909895908,
      "durationMs": 711.9730419944972,
      "imagesPerSecond": 1.4320353359056854,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 1271709696,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "totalMs": 2094.8732089891564,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": "pass",
    "quality": {
      "maximumIoU": 0.9518670570395931,
      "passed": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1382.9001669946592,
      "placement": {
        "cpu": 318
      },
      "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": 2985.1368340023328,
      "inferenceMs": 950.7704999996349,
      "durationMs": 967.3810829990543,
      "imagesPerSecond": 1.0517785312022028,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        1199
      ],
      "peakHelperRssBytes": 1657602048,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "totalMs": 2350.2812499937136,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": "pass",
    "quality": {
      "maximumIoU": 0,
      "passed": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 1382.9001669946592,
      "placement": {
        "cpu": 318
      },
      "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": 2985.1368340023328,
      "inferenceMs": 690.620833018329,
      "durationMs": 703.1618329929188,
      "imagesPerSecond": 1.4479725374479981,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 1662304256,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "totalMs": 2086.061999987578,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": "pass",
    "quality": {
      "maximumIoU": 0.987423113518462,
      "passed": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 618.3092079882044,
      "placement": {
        "cpu": 318
      },
      "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": 2154.6914170030504,
      "inferenceMs": 330.20191700779833,
      "durationMs": 342.875708010979,
      "imagesPerSecond": 3.028450013439453,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 1269366784,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 961.1849159991834,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": "pass",
    "quality": {
      "maximumIoU": 0.9518670570395931,
      "passed": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 618.3092079882044,
      "placement": {
        "cpu": 318
      },
      "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": 2154.6914170030504,
      "inferenceMs": 437.557541008573,
      "durationMs": 453.39037498342805,
      "imagesPerSecond": 2.2854137028355024,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        1199
      ],
      "peakHelperRssBytes": 1653374976,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 1071.6995829716325,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": "pass",
    "quality": {
      "maximumIoU": 0,
      "passed": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 618.3092079882044,
      "placement": {
        "cpu": 318
      },
      "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": 2154.6914170030504,
      "inferenceMs": 329.43091698689386,
      "durationMs": 341.8341250217054,
      "imagesPerSecond": 3.0355377969572426,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 1657782272,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 960.1433330099098,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": 0.9874232677559731,
      "passed": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 748.4869579784572,
      "placement": {
        "mps:0": 318
      },
      "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": 2278.196208004374,
      "inferenceMs": 267.4082909943536,
      "durationMs": 306.96120799984783,
      "imagesPerSecond": 3.739599831708716,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 618135552,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3122102272,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 1055.448165978305,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": 0.9518671477514653,
      "passed": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 748.4869579784572,
      "placement": {
        "mps:0": 318
      },
      "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": 2278.196208004374,
      "inferenceMs": 361.98095799773,
      "durationMs": 386.78904200787656,
      "imagesPerSecond": 2.7625762568600942,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        1199
      ],
      "peakHelperRssBytes": 709410816,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 5579702272,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 1135.2759999863338,
      "grade": "moderate",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 748.4869579784572,
      "placement": {
        "mps:0": 318
      },
      "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": 2278.196208004374,
      "inferenceMs": 180.36508400109597,
      "durationMs": 196.25925002037548,
      "imagesPerSecond": 5.544310338878692,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "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": 5579702272,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 944.7462079988327,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": 0.9861222925065938,
      "passed": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 725.8044170157518,
      "placement": {
        "mps:0": 318
      },
      "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": 2381.4895000250544,
      "inferenceMs": 253.29479199717753,
      "durationMs": 291.70520798652433,
      "imagesPerSecond": 3.947969052641015,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 676134912,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3025633280,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 1017.5096250022762,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": 0.9524794198547647,
      "passed": true
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 725.8044170157518,
      "placement": {
        "mps:0": 318
      },
      "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": 2381.4895000250544,
      "inferenceMs": 323.3365420019254,
      "durationMs": 347.59741701418534,
      "imagesPerSecond": 3.092752813550054,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        1199
      ],
      "peakHelperRssBytes": 763756544,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 4401102848,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 1073.4018340299372,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 725.8044170157518,
      "placement": {
        "mps:0": 318
      },
      "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": 2381.4895000250544,
      "inferenceMs": 164.67016699607484,
      "durationMs": 179.52979198889807,
      "imagesPerSecond": 6.072745405206497,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 763953152,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 4401102848,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 905.3342090046499,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.987423113518462
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 915.8514169976115,
      "placement": {
        "cpu": 318
      },
      "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": 2522.924625023734,
      "inferenceMs": 333.97687500109896,
      "durationMs": 367.1022920170799,
      "imagesPerSecond": 2.9942192853822873,
      "visualizationMs": 20.01233398914337,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "maskDevice": null,
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 1269071872,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 1282.9537090146914,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.987423113518462
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "cpu": 318
      },
      "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": 2522.924625023734,
      "inferenceMs": 331.50062497588806,
      "durationMs": 343.8968330156058,
      "imagesPerSecond": 3.0165855647262676,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "maskDevice": null,
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 1271562240,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 343.8968330156058,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.987423113518462
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "cpu": 318
      },
      "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": 2522.924625023734,
      "inferenceMs": 335.04179102601483,
      "durationMs": 347.4779170064721,
      "imagesPerSecond": 2.984702287250946,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "maskDevice": null,
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 1271627776,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 347.4779170064721,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.987423113518462
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "cpu": 318
      },
      "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": 2522.924625023734,
      "inferenceMs": 332.3254589922726,
      "durationMs": 344.8489579895977,
      "imagesPerSecond": 3.0090983791381825,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "maskDevice": null,
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 1276936192,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 344.8489579895977,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.987423113518462
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "cpu": 318
      },
      "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": 2522.924625023734,
      "inferenceMs": 333.65483302623034,
      "durationMs": 346.2996669986751,
      "imagesPerSecond": 2.9971092908502386,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "maskDevice": null,
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 1282752512,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 346.2996669986751,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.987423113518462
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "cpu": 318
      },
      "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": 2522.924625023734,
      "inferenceMs": 332.6109999907203,
      "durationMs": 345.4541670216713,
      "imagesPerSecond": 3.0065151183451526,
      "visualizationMs": 0,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "maskDevice": null,
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 1282883584,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 345.4541670216713,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.987423113518462
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 619.3343749910127,
      "placement": {
        "cpu": 318
      },
      "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": 2181.6802919784095,
      "inferenceMs": 330.77241701539606,
      "durationMs": 363.434958999278,
      "imagesPerSecond": 3.0232266917028157,
      "visualizationMs": 20.006499980809167,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "maskDevice": null,
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 1268465664,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 982.7693339902908,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": "pass",
    "quality": {
      "passed": true,
      "maximumIoU": 0.987423113518462
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 628.2796250015963,
      "placement": {
        "cpu": 318
      },
      "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": 2197.1459590131417,
      "inferenceMs": 335.44358401559293,
      "durationMs": 368.9801250002347,
      "imagesPerSecond": 2.981127222732975,
      "visualizationMs": 20.69495798787102,
      "predictionDevice": "cpu",
      "boxDevice": "cpu",
      "maskDevice": null,
      "inputDevice": "cpu",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 1269743616,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": null,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 997.259750001831,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": 0.9874232677559731
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 747.274000023026,
      "placement": {
        "mps:0": 318
      },
      "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": 2333.722374984063,
      "inferenceMs": 265.3809580078814,
      "durationMs": 324.5107920083683,
      "imagesPerSecond": 3.7681678727314774,
      "visualizationMs": 20.706625015009195,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 618627072,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3122102272,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 1071.7847920313943,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": 0.9874232677559731
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2333.722374984063,
      "inferenceMs": 182.86829098360613,
      "durationMs": 200.60462498804554,
      "imagesPerSecond": 5.468416610781628,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 620134400,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3122085888,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 200.60462498804554,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": 0.9874232677559731
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2333.722374984063,
      "inferenceMs": 182.45691698393784,
      "durationMs": 199.22704101190902,
      "imagesPerSecond": 5.4807459017189935,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 620396544,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3122085888,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 199.22704101190902,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": 0.9874232677559731
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2333.722374984063,
      "inferenceMs": 181.07258301461115,
      "durationMs": 198.2799580146093,
      "imagesPerSecond": 5.522647235442086,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 620445696,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3122085888,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 198.2799580146093,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": 0.9874232677559731
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2333.722374984063,
      "inferenceMs": 182.39387500216253,
      "durationMs": 199.759625014849,
      "imagesPerSecond": 5.482640247585856,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 620478464,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3122085888,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 199.759625014849,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": 0.9874232677559731
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2333.722374984063,
      "inferenceMs": 182.56833299528807,
      "durationMs": 200.09354199282825,
      "imagesPerSecond": 5.477401165873652,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 620511232,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3122085888,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 200.09354199282825,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": 0.9874232677559731
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2333.722374984063,
      "inferenceMs": 179.9482919741422,
      "durationMs": 197.02941601281054,
      "imagesPerSecond": 5.557151940868079,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 620527616,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3122085888,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 197.02941601281054,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": 0.9874232677559731
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2333.722374984063,
      "inferenceMs": 181.54366599628702,
      "durationMs": 198.98445799481124,
      "imagesPerSecond": 5.508316660414097,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 620527616,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3122085888,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 198.98445799481124,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": 0.9874232677559731
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2333.722374984063,
      "inferenceMs": 178.4233330108691,
      "durationMs": 196.37345799128525,
      "imagesPerSecond": 5.604648131638044,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 620527616,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3122085888,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 196.37345799128525,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": 0.9874232677559731
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2333.722374984063,
      "inferenceMs": 181.20245798490942,
      "durationMs": 199.27145799738355,
      "imagesPerSecond": 5.518688935683644,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 620527616,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3122085888,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 199.27145799738355,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": 0.9874232677559731
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2333.722374984063,
      "inferenceMs": 180.82604100345634,
      "durationMs": 199.4608329841867,
      "imagesPerSecond": 5.530176928337915,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 620544000,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3122085888,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 199.4608329841867,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": 0.9874232677559731
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2333.722374984063,
      "inferenceMs": 182.09016701439396,
      "durationMs": 199.77966701844707,
      "imagesPerSecond": 5.49178473717887,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 620544000,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3122085888,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 199.77966701844707,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": 0.9874232677559731
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2333.722374984063,
      "inferenceMs": 181.00862498977222,
      "durationMs": 198.13012500526384,
      "imagesPerSecond": 5.524598620958003,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 620560384,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3122085888,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 198.13012500526384,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": 0.9874232677559731
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2333.722374984063,
      "inferenceMs": 182.577624975238,
      "durationMs": 199.47687498643063,
      "imagesPerSecond": 5.477122402789632,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 620576768,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3122085888,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 199.47687498643063,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": 0.9874232677559731
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2333.722374984063,
      "inferenceMs": 180.45149999670684,
      "durationMs": 197.4681249994319,
      "imagesPerSecond": 5.541655237103873,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 620593152,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3122085888,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 197.4681249994319,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": 0.9874232677559731
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2333.722374984063,
      "inferenceMs": 180.3608339978382,
      "durationMs": 197.79054200625978,
      "imagesPerSecond": 5.544440984410096,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 620609536,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3122085888,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 197.79054200625978,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": 0.9874232677559731
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2333.722374984063,
      "inferenceMs": 179.70150001929142,
      "durationMs": 196.77850001608022,
      "imagesPerSecond": 5.564783821463079,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 620609536,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3122085888,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 196.77850001608022,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": 0.9874232677559731
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2333.722374984063,
      "inferenceMs": 181.1744160077069,
      "durationMs": 198.22829100303352,
      "imagesPerSecond": 5.51954311229827,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 620609536,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3122085888,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 198.22829100303352,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": 0.9874232677559731
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2333.722374984063,
      "inferenceMs": 180.71016599424183,
      "durationMs": 197.8495000221301,
      "imagesPerSecond": 5.5337229895072095,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 620609536,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3109502976,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 197.8495000221301,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": 0.9874232677559731
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2333.722374984063,
      "inferenceMs": 181.08324997592717,
      "durationMs": 198.79525000578724,
      "imagesPerSecond": 5.522321916206706,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 620609536,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3109502976,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 198.79525000578724,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": 0.9874232677559731
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2333.722374984063,
      "inferenceMs": 181.9165829801932,
      "durationMs": 199.471874977462,
      "imagesPerSecond": 5.497024974951725,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 620609536,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3109502976,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 199.471874977462,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": 0.9874232677559731
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2333.722374984063,
      "inferenceMs": 181.68525001965463,
      "durationMs": 199.390958994627,
      "imagesPerSecond": 5.504024129046361,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 620609536,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3109502976,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 199.390958994627,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": 0.9874232677559731
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2333.722374984063,
      "inferenceMs": 181.306541984668,
      "durationMs": 198.65133299026638,
      "imagesPerSecond": 5.51552078073699,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 620609536,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3109502976,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 198.65133299026638,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": 0.9874232677559731
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2333.722374984063,
      "inferenceMs": 182.03737499425188,
      "durationMs": 199.0806660032831,
      "imagesPerSecond": 5.493377390393465,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 620642304,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3109502976,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 199.0806660032831,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": 0.9874232677559731
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2333.722374984063,
      "inferenceMs": 179.60354199749418,
      "durationMs": 197.06204198882915,
      "imagesPerSecond": 5.567818924272395,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 620658688,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3109502976,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 197.06204198882915,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": 0.9874232677559731
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2333.722374984063,
      "inferenceMs": 184.14208298781887,
      "durationMs": 201.0599579953123,
      "imagesPerSecond": 5.430589161230194,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 620675072,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3109502976,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 201.0599579953123,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": 0.9874232677559731
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 757.2483340045437,
      "placement": {
        "mps:0": 318
      },
      "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": 2403.527166985441,
      "inferenceMs": 265.8328330144286,
      "durationMs": 324.1580000030808,
      "imagesPerSecond": 3.761762565821668,
      "visualizationMs": 20.262125006411225,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 617037824,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3122102272,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 1081.4063340076245,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": 0.9874232677559731
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 755.6686250027269,
      "placement": {
        "mps:0": 318
      },
      "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": 2341.8135840038303,
      "inferenceMs": 266.71687501948327,
      "durationMs": 325.4118750046473,
      "imagesPerSecond": 3.749294077950454,
      "visualizationMs": 20.50775001407601,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 614547456,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3122102272,
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 1081.0805000073742,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": 0.9861222925065938
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 748.879584018141,
      "placement": {
        "mps:0": 318
      },
      "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": 2372.526499995729,
      "inferenceMs": 254.04745902051218,
      "durationMs": 311.780040996382,
      "imagesPerSecond": 3.9362723951482566,
      "visualizationMs": 20.06458299001679,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 688832512,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3025633280,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 1060.659625014523,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": 0.9861222925065938
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2372.526499995729,
      "inferenceMs": 166.59037498175167,
      "durationMs": 184.287417010637,
      "imagesPerSecond": 6.002747758443668,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 693534720,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3025616896,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 184.287417010637,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": 0.9861222925065938
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2372.526499995729,
      "inferenceMs": 166.31695898831822,
      "durationMs": 184.2782919993624,
      "imagesPerSecond": 6.012615947783401,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 693714944,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3025616896,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 184.2782919993624,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": 0.9861222925065938
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2372.526499995729,
      "inferenceMs": 166.2589169864077,
      "durationMs": 183.20879200473428,
      "imagesPerSecond": 6.0147149886809,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 693846016,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3025616896,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 183.20879200473428,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": 0.9861222925065938
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2372.526499995729,
      "inferenceMs": 165.6315840082243,
      "durationMs": 184.5618750085123,
      "imagesPerSecond": 6.037495843488074,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 693927936,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3025616896,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 184.5618750085123,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": 0.9861222925065938
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "mps:0": 318
      },
      "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": 2372.526499995729,
      "inferenceMs": 166.91612501745112,
      "durationMs": 185.44720800127834,
      "imagesPerSecond": 5.9910329208484185,
      "visualizationMs": 0,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 693960704,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3025616896,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 185.44720800127834,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": 0.9861222925065938
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 752.2608330182265,
      "placement": {
        "mps:0": 318
      },
      "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": 2382.2316250007134,
      "inferenceMs": 258.60787500278093,
      "durationMs": 317.6519169937819,
      "imagesPerSecond": 3.866858269452338,
      "visualizationMs": 20.496750017628074,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 688275456,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3025633280,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 1069.9127500120085,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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": 0.9861222925065938
    },
    "runtime": {
      "engine": "transformers",
      "version": "5.22.1",
      "torchVersion": "2.10.0",
      "loadMs": 730.1427920174319,
      "placement": {
        "mps:0": 318
      },
      "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": 2264.4864580070134,
      "inferenceMs": 253.8266250048764,
      "durationMs": 312.567749992013,
      "imagesPerSecond": 3.9396970273736587,
      "visualizationMs": 20.231958013027906,
      "predictionDevice": "mps:0",
      "boxDevice": "mps:0",
      "maskDevice": null,
      "inputDevice": "mps:0",
      "processedShape": [
        1,
        3,
        800,
        970
      ],
      "peakHelperRssBytes": 675119104,
      "memoryScope": "Process-lifetime peak RSS on macOS, not isolated model memory or GPU minimum",
      "driverAllocatedMemoryBytes": 3025633280,
      "device": "mps",
      "precision": "fp16",
      "cpuThreads": 4,
      "totalMs": 1042.710542009445,
      "grade": "fast",
      "fallback": false,
      "artifactRevision": "6da4ef8c063ac1e8b8df9522e8113c5cc9ade5c9",
      "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.