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.

SQuadGen Desktop

Generate a quad-mesh layout conditioned on a closed triangle mesh, retaining complete neural fields and an orthographic preview.

Desktop inferencePrivate · Runs on your computer
About this model

Generate a quad-mesh layout conditioned on a closed triangle mesh, retaining complete neural fields and an orthographic preview.

Created by
Microsoft
License
MIT
Model released
Unknown
Source repository created
Aug 27, 2026
Model weights · FP32
4.1 GB

Best for

  • Offline quad-layout experiments for clean closed smooth meshes

Keep in mind

  • Only two independently authored smooth meshes have measured quality. Complex CAD, scans, sharp features, holes and broad topology quality are unqualified.
  • Generated layouts can distort surfaces or contain degenerate/nonquad faces; the adapter rejects invalid geometry after a fixed extraction sequence. No engineering, fabrication or safety-critical accuracy guarantee.
  • Inputs must be finite closed consistently oriented manifold triangle meshes,4..10000vertices and4..5000triangles; extreme scales or poor tessellation may fail.
  • The conditioned geometry encoder and diffusion execute on strict MPS; surface sampling, PCA, JSON/SVG and native QCDF2 extraction use CPU. This is not end-to-end GPU execution.
  • CPU4 completes the sphere quality fixture but233seconds exceeds the frozen180second practical limit. CPU is not advertised as recovery; GPU failures return a clear error.
  • Only current Apple M3 Max/macOS27.2 is measured. Older macOS, Intel Mac, other Apple hardware and Windows GPU providers are unverified.
  • The original checkpoint also contains geometry-AE UDF functionality used internally for conditioning; general reconstruction is not exposed or quality-qualified.
  • Ordered native field chunks stream as each decoder batch finishes. Intermediate diffusion latents are not task results and are not streamed; the complete quad mesh is returned in the final result.
  • Regular and genuine field-streaming public APIs, complete final parity and scoped lifecycle pass. Packaged production app recovery, final source-bound contracts and backup/recovery remain pending; this is not release acceptance.

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/microsoft/SQuadGen
Release verification
Unknown
Release evidence
Bounded official review did not establish the exact checkpoint public release day. HF repository creation, paper and code dates are distinct and are not substituted.
Source updated
Aug 27, 2026
Native model files
microsoft/SQuadGen
Files updated
Aug 27, 2026
Artifact revision
67f8b71358cc5817cd7e759bd0c7503168e89291

Desktop app

Sample result

SQuadGen Base Desktop

Run SQuadGen Desktop on your computer

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

Public desktop downloads are coming soon.

Device requirements

macOS Apple SiliconMPS / FP32Tested32.0 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.

GPUNo CPU profile passes the practical speed gate. GPU setup/inference failure returns a clear error; no unverified CPU fallback is selected.

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.

SQuadGen Base Desktop

3 configurations tested
Passed3
Failed0
TestedNov 16, 2026
Runtimepytorch 2.10.0

Quality criteria

{ "minimumQuadFaceFraction": 0.95, "maximumAnalyticNormalizedSurfaceError": 0.08, "minimumSurfaceNormalConsistency": 0.9, "nonemptyFiniteNondegenerateDistinctInRangeFaces": true, "maximumRecommendedGenerationSeconds": 180 }

PrecisionDeviceResultPlatformInference / load
FP32mpsAcceleratedPassedFastmacos · aarch6491190 ms / 5196 ms
FP32mpsAcceleratedPassedFastmacos · aarch6495626 ms / 5196 ms
FP32mpsAcceleratedPassedFastmacos · aarch6496638 ms / 8587 ms
Test setup & sample details
Quality gate
{ "minimumQuadFaceFraction": 0.95, "maximumAnalyticNormalizedSurfaceError": 0.08, "minimumSurfaceNormalConsistency": 0.9, "nonemptyFiniteNondegenerateDistinctInRangeFaces": true, "maximumRecommendedGenerationSeconds": 180 }
Tested
2026-11-16T19:51:31.545703+00:00
Environment
{ "hardware": "Apple M3 Max", "os": "macOS27.2", "memoryBytes": 103079215104, "cpuThreads": 4, "backgroundUploads": "Existing persistent mirror active; no overlapping neural benchmark. Other user native builds may be active. Timings are retained current-host observations, no Windows claim." }
Runtime
pytorch 2.10.0
Artifact revision
67f8b71358cc5817cd7e759bd0c7503168e89291
Fixture
{
  "catalog": "research/squadgen/fixture-catalog.json",
  "source": "CC0 independent procedural meshes"
}
Rerun when
Original checkpoint/source/config revision; Torch/runtime package or neural implementation; Fixture checksum/quality gate; MPS placement/backend behavior; Input sampling, extraction or visualization behavior
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "testedAt": "2026-11-16T18:53:41.089708+00:00",
      "fixture": "sphere",
      "vertices": 2082,
      "polygons": 2080,
      "polygonDegrees": {
        "4": 2080
      },
      "finite": true,
      "distinctInRangeFaceVertices": true,
      "nonDegenerate": true,
      "quadFraction": 1,
      "maxAnalyticNormalizedSurfaceError": 0.03428546471997784,
      "surfaceNormalConsistency": 0.9944125196539589,
      "passed": true,
      "limits": "Analytic original shape includes finite tessellation error. No broad CAD/scan claim."
    },
    "runtime": {
      "acceleration": "gpu",
      "extractionMs": 1431.2587500025984,
      "functionEvaluations": 224,
      "inferenceMs": 91190.3790840006,
      "neuralInferenceMs": 79865.93520799943,
      "provider": "mps",
      "visualizationMs": 10.812084001372568,
      "loadMs": 5195.945000006759,
      "totalMs": 96386.32408400736,
      "grade": "good",
      "meshesPerSecond": 0.010966069118748192,
      "precision": "fp32",
      "device": "mps",
      "actualDevice": "mps",
      "actualPrecision": "fp32",
      "cpuThreads": 4,
      "engine": "pytorch",
      "version": "2.10.0",
      "neuralExecuted": true,
      "fallback": false,
      "artifactRevision": "67f8b71358cc5817cd7e759bd0c7503168e89291"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "testedAt": "2026-11-16T18:53:49.658734+00:00",
      "fixture": "ellipsoid",
      "vertices": 7682,
      "polygons": 7680,
      "polygonDegrees": {
        "4": 7680
      },
      "finite": true,
      "distinctInRangeFaceVertices": true,
      "nonDegenerate": true,
      "quadFraction": 1,
      "maxAnalyticNormalizedSurfaceError": 0.02300389110853862,
      "surfaceNormalConsistency": 0.9986982877848158,
      "passed": true,
      "limits": "Analytic original shape includes finite tessellation error. No broad CAD/scan claim."
    },
    "runtime": {
      "acceleration": "gpu",
      "extractionMs": 11240.728666001814,
      "functionEvaluations": 176,
      "inferenceMs": 95625.65383299807,
      "neuralInferenceMs": 83581.4205409988,
      "provider": "mps",
      "visualizationMs": 36.92945899820188,
      "loadMs": 5195.945000006759,
      "totalMs": 100821.59883300483,
      "grade": "good",
      "meshesPerSecond": 0.010457444837411659,
      "precision": "fp32",
      "device": "mps",
      "actualDevice": "mps",
      "actualPrecision": "fp32",
      "cpuThreads": 4,
      "engine": "pytorch",
      "version": "2.10.0",
      "neuralExecuted": true,
      "fallback": false,
      "artifactRevision": "67f8b71358cc5817cd7e759bd0c7503168e89291"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "testedAt": "2026-11-16T18:53:41.089708+00:00",
      "fixture": "sphere",
      "vertices": 2082,
      "polygons": 2080,
      "polygonDegrees": {
        "4": 2080
      },
      "finite": true,
      "distinctInRangeFaceVertices": true,
      "nonDegenerate": true,
      "quadFraction": 1,
      "maxAnalyticNormalizedSurfaceError": 0.03428546471997784,
      "surfaceNormalConsistency": 0.9944125196539589,
      "passed": true,
      "limits": "Analytic original shape includes finite tessellation error. No broad CAD/scan claim.",
      "sourceCaseId": "sphere-mps",
      "reusedUnchangedTaskQuality": true,
      "parityEvidence": "research/squadgen/native-public-streaming-packed.json.chunks.json"
    },
    "runtime": {
      "acceleration": "gpu",
      "artifactRevision": "67f8b71358cc5817cd7e759bd0c7503168e89291",
      "cpuThreads": 4,
      "device": "mps",
      "durationMs": 96860.831459,
      "engine": "pytorch",
      "extractionMs": 1054.5549590024166,
      "fallback": false,
      "fallbackReason": null,
      "family": {
        "catalogId": "squadgen-native",
        "defaultModelId": "squadgen-base-native",
        "displayName": "SQuadGen Desktop",
        "id": "squadgen",
        "variantId": "base",
        "variantName": "Base pipeline"
      },
      "functionEvaluations": 224,
      "implicitMpsFallback": false,
      "inferenceMs": 96637.83245800005,
      "loadMs": 8586.773041,
      "modelId": "squadgen-base-native",
      "modelQuantization": "fp32",
      "neuralInferenceMs": 94667.47837499861,
      "placement": {
        "bufferDevices": [],
        "cpuStages": [
          "surface sampling",
          "PCA",
          "quad extraction/refinement",
          "JSON/SVG"
        ],
        "parameterDevices": [
          "mps"
        ]
      },
      "precision": "fp32",
      "provider": "mps",
      "requestedDevice": "mps",
      "requestedPrecision": "fp32",
      "reused": false,
      "runtimePrecision": "fp32",
      "selectionReason": "Explicit device selection; actual placement is reported after inference",
      "streamChunks": 5,
      "threadControl": {
        "torch": 4
      },
      "version": "2.10.0",
      "visualizationMs": 9.973416999855544,
      "grade": "good",
      "totalMs": 105224.60549900006,
      "meshesPerSecond": 0.01034791421294152,
      "actualDevice": "mps",
      "actualPrecision": "fp32",
      "neuralExecuted": true
    }
  }
]
Release validation still required
  • Packaged production app recovery and final source-bound contracts pending
  • Complete closure B2 coverage and cold model/runtime recovery pending

Something not quite right?

Tell us what happened with SQuadGen 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.