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.

VibeVoice ASR BitNet

Transcribe audio files locally with a compact BitNet model. Text only; no speaker labels or timestamps.

Desktop inferencePrivate · Runs on your computer
About this model

Transcribe audio files locally with a compact BitNet model. Text only; no speaker labels or timestamps.

Created by
Microsoft
License
MIT
Model released
Jul 24, 2026
Source repository created
Jul 24, 2026
Model weights · I2_S
1.7 GB

Best for

  • Local audio-file transcription

Keep in mind

  • May omit or hallucinate speech; review noisy, accented or overlapping recordings.
  • Text-only output: speaker/timestamp JSON prompt failed and is not exposed.
  • English fixtures tested; upstream multilingual claims are not local language-quality certification.
  • Complete PCM16/float32 WAV, mono/stereo, 8–192kHz, at most 120 seconds and 100MB. No microphone/live input.
  • Metal accelerates the language decoder; audio encoders remain on CPU. Other platforms are unverified.
  • Output capped at2048 tokens; long/dense speech can be truncated.

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

I2_S · metal (verified platform only)

Sources & technical details
Original model
microsoft/VibeVoice-ASR-BitNet
Release verification
Verified
Release evidence
Official checkpoint commit explicitly titled Initial model release; distinct from July23 technical report.
Source updated
Jul 24, 2026
Native model files
microsoft/VibeVoice-ASR-BitNet
Files updated
Jul 24, 2026
Artifact revision
66e78021ab8f5f06133d1ab421ba4d348bda97c9

Desktop app

Sample result

VibeVoice ASR BitNet 1.5B Desktop

Run VibeVoice ASR BitNet 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 / I2_STested7.5 GiBRAM estimate
macOS Apple SiliconMETAL / I2_STested7.5 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.

VibeVoice ASR BitNet 1.5B Desktop

54 configurations tested
Passed54
Failed0
TestedNov 4, 2026
RuntimeVibeASR.cpp c4334009c88060f86cdbbd684b62662f710b6c20

Quality criteria

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

PrecisionDeviceResultPlatformInference / loadTask metric
I2_SmetalAcceleratedPassedFastmacos · aarch645557 ms / 380 ms0.156 RTF
I2_SmetalAcceleratedPassedFastmacos · aarch641742 ms / 385 ms0.158 RTF
I2_SmetalAcceleratedPassedFastmacos · aarch645650 ms / 385 ms0.158 RTF
I2_SmetalAcceleratedPassedFastmacos · aarch645662 ms / 385 ms0.159 RTF
I2_SmetalAcceleratedPassedFastmacos · aarch641762 ms / 380 ms0.160 RTF
I2_SmetalAcceleratedPassedFastmacos · aarch641765 ms / 380 ms0.160 RTF
I2_SmetalAcceleratedPassedFastmacos · aarch641777 ms / 385 ms0.162 RTF
I2_SmetalAcceleratedPassedFastmacos · aarch641796 ms / 385 ms0.163 RTF
I2_SmetalAcceleratedPassedFastmacos · aarch641800 ms / 385 ms0.164 RTF
I2_SmetalAcceleratedPassedFastmacos · aarch641827 ms / 380 ms0.166 RTF
I2_SmetalAcceleratedPassedFastmacos · aarch641861 ms / 385 ms0.169 RTF
I2_SmetalAcceleratedPassedFastmacos · aarch641881 ms / 385 ms0.171 RTF
I2_SmetalAcceleratedPassedFastmacos · aarch64422 ms / 385 ms0.211 RTF
I2_SmetalAcceleratedPassedFastmacos · aarch64422 ms / 380 ms0.211 RTF
I2_SmetalAcceleratedPassedFastmacos · aarch64423 ms / 385 ms0.211 RTF
I2_SmetalAcceleratedPassedFastmacos · aarch64733 ms / 380 ms0.216 RTF
I2_SmetalAcceleratedPassedFastmacos · aarch64745 ms / 385 ms0.219 RTF
I2_SmetalAcceleratedPassedFastmacos · aarch64749 ms / 385 ms0.220 RTF
I2_ScpuPassedFastmacos · aarch647293 ms / 378 ms0.205 RTF
I2_ScpuPassedFastmacos · aarch647325 ms / 379 ms0.206 RTF
I2_ScpuPassedFastmacos · aarch647350 ms / 400 ms0.206 RTF
I2_ScpuPassedFastmacos · aarch642351 ms / 400 ms0.214 RTF
I2_ScpuPassedFastmacos · aarch642366 ms / 378 ms0.215 RTF
I2_ScpuPassedFastmacos · aarch642369 ms / 378 ms0.215 RTF
I2_ScpuPassedFastmacos · aarch642374 ms / 379 ms0.216 RTF
I2_ScpuPassedFastmacos · aarch642405 ms / 400 ms0.219 RTF
I2_ScpuPassedFastmacos · aarch642441 ms / 378 ms0.222 RTF
I2_ScpuPassedFastmacos · aarch642452 ms / 379 ms0.223 RTF
I2_ScpuPassedFastmacos · aarch642462 ms / 379 ms0.224 RTF
I2_ScpuPassedFastmacos · aarch642462 ms / 400 ms0.224 RTF
I2_ScpuPassedModeratemacos · aarch641041 ms / 400 ms0.306 RTF
I2_ScpuPassedModeratemacos · aarch641043 ms / 378 ms0.307 RTF
I2_ScpuPassedModeratemacos · aarch641052 ms / 379 ms0.309 RTF
I2_ScpuPassedModeratemacos · aarch64620 ms / 378 ms0.310 RTF
I2_ScpuPassedModeratemacos · aarch64626 ms / 400 ms0.313 RTF
I2_ScpuPassedModeratemacos · aarch64637 ms / 379 ms0.319 RTF
I2_ScpuPassedModeratemacos · aarch6425396 ms / 376 ms0.712 RTF
I2_ScpuPassedModeratemacos · aarch6426096 ms / 414 ms0.732 RTF
I2_ScpuPassedModeratemacos · aarch6426291 ms / 388 ms0.738 RTF
I2_ScpuPassedModeratemacos · aarch648262 ms / 376 ms0.751 RTF
I2_ScpuPassedModeratemacos · aarch648289 ms / 376 ms0.754 RTF
I2_ScpuPassedModeratemacos · aarch648291 ms / 376 ms0.754 RTF
I2_ScpuPassedModeratemacos · aarch648313 ms / 414 ms0.756 RTF
I2_ScpuPassedModeratemacos · aarch648328 ms / 414 ms0.757 RTF
I2_ScpuPassedModeratemacos · aarch648433 ms / 388 ms0.767 RTF
I2_ScpuPassedModeratemacos · aarch648471 ms / 388 ms0.770 RTF
I2_ScpuPassedModeratemacos · aarch648503 ms / 414 ms0.773 RTF
I2_ScpuPassedModeratemacos · aarch648611 ms / 388 ms0.783 RTF
I2_ScpuPassedModeratemacos · aarch643342 ms / 376 ms0.983 RTF
I2_ScpuPassedModeratemacos · aarch643439 ms / 414 ms1.011 RTF
I2_ScpuPassedModeratemacos · aarch643489 ms / 388 ms1.026 RTF
I2_ScpuPassedModeratemacos · aarch642281 ms / 376 ms1.140 RTF
I2_ScpuPassedModeratemacos · aarch642296 ms / 388 ms1.148 RTF
I2_ScpuPassedModeratemacos · aarch642340 ms / 414 ms1.170 RTF
Test setup & sample details
Quality gate
No explicit quality gate recorded. Review the per-case quality evidence; full certification is pending.
Tested
2026-11-04T10:24:20Z
Environment
{ "os": "macos", "arch": "aarch64", "hardware": "Apple M3 Max", "osVersion": "27.2" }
Runtime
VibeASR.cpp c4334009c88060f86cdbbd684b62662f710b6c20
Artifact revision
66e78021ab8f5f06133d1ab421ba4d348bda97c9
Fixture
{
  "file": "model-samples/desktop/vibevoice-bitnet/vibevoice-bitnet-jfk.wav",
  "url": "https://assets.localai.tools/samples/v1/files/c72e27f68334ae37e037ef819fa5aa1fc1552aa840b93e308df4ffe52317bedb/vibevoice-bitnet-jfk.wav",
  "size": 3880888,
  "sha1": "6fb9dfd0ec18bb61a6af995c0df0cd32ee2afe34",
  "sha256": "c72e27f68334ae37e037ef819fa5aa1fc1552aa840b93e308df4ffe52317bedb"
}
Rerun when
artifact revision; runtime binary/source; fixture checksum; quality gate; backend behavior
Per-case quality & failure details
[
  {
    "precision": "i2_s",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0.045454545454545456,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 375.72300000465475,
      "placement": "llm_load_tensors: offloading 0 repeating layers to GPU\nllm_load_tensors: offloaded 0/29 layers to GPU\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)",
      "provider": "ggml CPU",
      "acceleration": "cpu",
      "inferenceMs": 8289.247999966145,
      "audioSeconds": 11,
      "realTimeFactor": 0.7535679999969223,
      "outputTokens": 29,
      "totalMs": 8665.275416016812,
      "durationMs": 8289.552416012157,
      "grade": "acceptable",
      "rssBytes": 4525637632,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 1,
      "device": "cpu",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 375.72300000465475,
      "placement": "llm_load_tensors: offloading 0 repeating layers to GPU\nllm_load_tensors: offloaded 0/29 layers to GPU\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)",
      "provider": "ggml CPU",
      "acceleration": "cpu",
      "inferenceMs": 3341.6440000534058,
      "audioSeconds": 3.4000000953674316,
      "realTimeFactor": 0.982836443036123,
      "outputTokens": 14,
      "totalMs": 3717.7314999862574,
      "durationMs": 3342.0084999816027,
      "grade": "acceptable",
      "rssBytes": 4526030848,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 1,
      "device": "cpu",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 375.72300000465475,
      "placement": "llm_load_tensors: offloading 0 repeating layers to GPU\nllm_load_tensors: offloaded 0/29 layers to GPU\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)",
      "provider": "ggml CPU",
      "acceleration": "cpu",
      "inferenceMs": 25395.925999999046,
      "audioSeconds": 35.64437484741211,
      "realTimeFactor": 0.7124806118416991,
      "outputTokens": 95,
      "totalMs": 25771.976458025165,
      "durationMs": 25396.25345802051,
      "grade": "acceptable",
      "rssBytes": 9523232768,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 1,
      "device": "cpu",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 375.72300000465475,
      "placement": "llm_load_tensors: offloading 0 repeating layers to GPU\nllm_load_tensors: offloaded 0/29 layers to GPU\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)",
      "provider": "ggml CPU",
      "acceleration": "cpu",
      "inferenceMs": 2280.9280000329018,
      "audioSeconds": 2,
      "realTimeFactor": 1.140464000016451,
      "outputTokens": 4,
      "totalMs": 2657.0254170219414,
      "durationMs": 2281.3024170172866,
      "grade": "acceptable",
      "rssBytes": 9525133312,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 1,
      "device": "cpu",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0.045454545454545456,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 375.72300000465475,
      "placement": "llm_load_tensors: offloading 0 repeating layers to GPU\nllm_load_tensors: offloaded 0/29 layers to GPU\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)",
      "provider": "ggml CPU",
      "acceleration": "cpu",
      "inferenceMs": 8261.810000002384,
      "audioSeconds": 11,
      "realTimeFactor": 0.7510736363638532,
      "outputTokens": 29,
      "totalMs": 8637.96100000036,
      "durationMs": 8262.237999995705,
      "grade": "acceptable",
      "rssBytes": 9525444608,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 1,
      "device": "cpu",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0.045454545454545456,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 375.72300000465475,
      "placement": "llm_load_tensors: offloading 0 repeating layers to GPU\nllm_load_tensors: offloaded 0/29 layers to GPU\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)",
      "provider": "ggml CPU",
      "acceleration": "cpu",
      "inferenceMs": 8291.20600003004,
      "audioSeconds": 11,
      "realTimeFactor": 0.753746000002731,
      "outputTokens": 29,
      "totalMs": 8667.273167025996,
      "durationMs": 8291.550167021342,
      "grade": "acceptable",
      "rssBytes": 9525460992,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 1,
      "device": "cpu",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0.045454545454545456,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 388.34345899522305,
      "placement": "llm_load_tensors: offloading 0 repeating layers to GPU\nllm_load_tensors: offloaded 0/29 layers to GPU\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)",
      "provider": "ggml CPU",
      "acceleration": "cpu",
      "inferenceMs": 8610.614000022411,
      "audioSeconds": 11,
      "realTimeFactor": 0.7827830909111282,
      "outputTokens": 29,
      "totalMs": 8999.336084001698,
      "durationMs": 8610.992625006475,
      "grade": "acceptable",
      "rssBytes": 4527472640,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 1,
      "device": "cpu",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 388.34345899522305,
      "placement": "llm_load_tensors: offloading 0 repeating layers to GPU\nllm_load_tensors: offloaded 0/29 layers to GPU\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)",
      "provider": "ggml CPU",
      "acceleration": "cpu",
      "inferenceMs": 3489.310000002384,
      "audioSeconds": 3.4000000953674316,
      "realTimeFactor": 1.0262676182734933,
      "outputTokens": 14,
      "totalMs": 3878.013208974153,
      "durationMs": 3489.6697499789298,
      "grade": "acceptable",
      "rssBytes": 4527718400,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 1,
      "device": "cpu",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 388.34345899522305,
      "placement": "llm_load_tensors: offloading 0 repeating layers to GPU\nllm_load_tensors: offloaded 0/29 layers to GPU\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)",
      "provider": "ggml CPU",
      "acceleration": "cpu",
      "inferenceMs": 26291.324000000954,
      "audioSeconds": 35.64437484741211,
      "realTimeFactor": 0.7376009289698564,
      "outputTokens": 95,
      "totalMs": 26680.031875992427,
      "durationMs": 26291.688416997204,
      "grade": "acceptable",
      "rssBytes": 9526345728,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 1,
      "device": "cpu",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 388.34345899522305,
      "placement": "llm_load_tensors: offloading 0 repeating layers to GPU\nllm_load_tensors: offloaded 0/29 layers to GPU\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)",
      "provider": "ggml CPU",
      "acceleration": "cpu",
      "inferenceMs": 2296.3579999804497,
      "audioSeconds": 2,
      "realTimeFactor": 1.1481789999902248,
      "outputTokens": 4,
      "totalMs": 2685.065208992455,
      "durationMs": 2296.721749997232,
      "grade": "acceptable",
      "rssBytes": 9526935552,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 1,
      "device": "cpu",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0.045454545454545456,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 388.34345899522305,
      "placement": "llm_load_tensors: offloading 0 repeating layers to GPU\nllm_load_tensors: offloaded 0/29 layers to GPU\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)",
      "provider": "ggml CPU",
      "acceleration": "cpu",
      "inferenceMs": 8470.589999973774,
      "audioSeconds": 11,
      "realTimeFactor": 0.7700536363612521,
      "outputTokens": 29,
      "totalMs": 8859.273083973676,
      "durationMs": 8470.929624978453,
      "grade": "acceptable",
      "rssBytes": 9527721984,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 1,
      "device": "cpu",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0.045454545454545456,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 388.34345899522305,
      "placement": "llm_load_tensors: offloading 0 repeating layers to GPU\nllm_load_tensors: offloaded 0/29 layers to GPU\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)",
      "provider": "ggml CPU",
      "acceleration": "cpu",
      "inferenceMs": 8432.560000002384,
      "audioSeconds": 11,
      "realTimeFactor": 0.7665963636365803,
      "outputTokens": 29,
      "totalMs": 8821.244833990932,
      "durationMs": 8432.901374995708,
      "grade": "acceptable",
      "rssBytes": 9527721984,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 1,
      "device": "cpu",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0.045454545454545456,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 413.9662919915281,
      "placement": "llm_load_tensors: offloading 0 repeating layers to GPU\nllm_load_tensors: offloaded 0/29 layers to GPU\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)",
      "provider": "ggml CPU",
      "acceleration": "cpu",
      "inferenceMs": 8502.936000049114,
      "audioSeconds": 11,
      "realTimeFactor": 0.7729941818226468,
      "outputTokens": 29,
      "totalMs": 8917.268624994904,
      "durationMs": 8503.302333003376,
      "grade": "acceptable",
      "rssBytes": 4530798592,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 1,
      "device": "cpu",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 413.9662919915281,
      "placement": "llm_load_tensors: offloading 0 repeating layers to GPU\nllm_load_tensors: offloaded 0/29 layers to GPU\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)",
      "provider": "ggml CPU",
      "acceleration": "cpu",
      "inferenceMs": 3438.844000041485,
      "audioSeconds": 3.4000000953674316,
      "realTimeFactor": 1.0114246775248563,
      "outputTokens": 14,
      "totalMs": 3853.1666260096245,
      "durationMs": 3439.2003340180963,
      "grade": "acceptable",
      "rssBytes": 4530814976,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 1,
      "device": "cpu",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 413.9662919915281,
      "placement": "llm_load_tensors: offloading 0 repeating layers to GPU\nllm_load_tensors: offloaded 0/29 layers to GPU\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)",
      "provider": "ggml CPU",
      "acceleration": "cpu",
      "inferenceMs": 26096.01399999857,
      "audioSeconds": 35.64437484741211,
      "realTimeFactor": 0.7321215230092111,
      "outputTokens": 95,
      "totalMs": 26510.40716699208,
      "durationMs": 26096.440875000553,
      "grade": "acceptable",
      "rssBytes": 9528508416,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 1,
      "device": "cpu",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 413.9662919915281,
      "placement": "llm_load_tensors: offloading 0 repeating layers to GPU\nllm_load_tensors: offloaded 0/29 layers to GPU\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)",
      "provider": "ggml CPU",
      "acceleration": "cpu",
      "inferenceMs": 2339.548999965191,
      "audioSeconds": 2,
      "realTimeFactor": 1.1697744999825954,
      "outputTokens": 4,
      "totalMs": 2753.907624981366,
      "durationMs": 2339.941332989838,
      "grade": "acceptable",
      "rssBytes": 9528868864,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 1,
      "device": "cpu",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0.045454545454545456,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 413.9662919915281,
      "placement": "llm_load_tensors: offloading 0 repeating layers to GPU\nllm_load_tensors: offloaded 0/29 layers to GPU\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)",
      "provider": "ggml CPU",
      "acceleration": "cpu",
      "inferenceMs": 8312.985000014305,
      "audioSeconds": 11,
      "realTimeFactor": 0.7557259090922095,
      "outputTokens": 29,
      "totalMs": 8727.392708999105,
      "durationMs": 8313.426417007577,
      "grade": "acceptable",
      "rssBytes": 9528885248,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 1,
      "device": "cpu",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0.045454545454545456,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 413.9662919915281,
      "placement": "llm_load_tensors: offloading 0 repeating layers to GPU\nllm_load_tensors: offloaded 0/29 layers to GPU\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)",
      "provider": "ggml CPU",
      "acceleration": "cpu",
      "inferenceMs": 8327.512000024319,
      "audioSeconds": 11,
      "realTimeFactor": 0.7570465454567562,
      "outputTokens": 29,
      "totalMs": 8741.865541989682,
      "durationMs": 8327.899249998154,
      "grade": "acceptable",
      "rssBytes": 9528885248,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 1,
      "device": "cpu",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0.045454545454545456,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 379.1782499756664,
      "placement": "llm_load_tensors: offloading 0 repeating layers to GPU\nllm_load_tensors: offloaded 0/29 layers to GPU\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)",
      "provider": "ggml CPU",
      "acceleration": "cpu",
      "inferenceMs": 2461.878000020981,
      "audioSeconds": 11,
      "realTimeFactor": 0.22380709091099826,
      "outputTokens": 29,
      "totalMs": 2841.432666988112,
      "durationMs": 2462.2544170124456,
      "grade": "fast",
      "rssBytes": 4527669248,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 4,
      "device": "cpu",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 379.1782499756664,
      "placement": "llm_load_tensors: offloading 0 repeating layers to GPU\nllm_load_tensors: offloaded 0/29 layers to GPU\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)",
      "provider": "ggml CPU",
      "acceleration": "cpu",
      "inferenceMs": 1051.561999976635,
      "audioSeconds": 3.4000000953674316,
      "realTimeFactor": 0.3092829324944459,
      "outputTokens": 14,
      "totalMs": 1431.0996249550954,
      "durationMs": 1051.921374979429,
      "grade": "acceptable",
      "rssBytes": 4527734784,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 4,
      "device": "cpu",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 379.1782499756664,
      "placement": "llm_load_tensors: offloading 0 repeating layers to GPU\nllm_load_tensors: offloaded 0/29 layers to GPU\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)",
      "provider": "ggml CPU",
      "acceleration": "cpu",
      "inferenceMs": 7325.122000038624,
      "audioSeconds": 35.64437484741211,
      "realTimeFactor": 0.20550569427564108,
      "outputTokens": 95,
      "totalMs": 7704.637749964604,
      "durationMs": 7325.459499988938,
      "grade": "fast",
      "rssBytes": 9519710208,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 4,
      "device": "cpu",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 379.1782499756664,
      "placement": "llm_load_tensors: offloading 0 repeating layers to GPU\nllm_load_tensors: offloaded 0/29 layers to GPU\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)",
      "provider": "ggml CPU",
      "acceleration": "cpu",
      "inferenceMs": 637.3849999904633,
      "audioSeconds": 2,
      "realTimeFactor": 0.31869249999523164,
      "outputTokens": 4,
      "totalMs": 1016.9252909836359,
      "durationMs": 637.7470410079695,
      "grade": "acceptable",
      "rssBytes": 9521954816,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 4,
      "device": "cpu",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0.045454545454545456,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 379.1782499756664,
      "placement": "llm_load_tensors: offloading 0 repeating layers to GPU\nllm_load_tensors: offloaded 0/29 layers to GPU\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)",
      "provider": "ggml CPU",
      "acceleration": "cpu",
      "inferenceMs": 2451.6800000071526,
      "audioSeconds": 11,
      "realTimeFactor": 0.22288000000065022,
      "outputTokens": 29,
      "totalMs": 2831.313417002093,
      "durationMs": 2452.1351670264266,
      "grade": "fast",
      "rssBytes": 9522626560,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 4,
      "device": "cpu",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0.045454545454545456,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 379.1782499756664,
      "placement": "llm_load_tensors: offloading 0 repeating layers to GPU\nllm_load_tensors: offloaded 0/29 layers to GPU\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)",
      "provider": "ggml CPU",
      "acceleration": "cpu",
      "inferenceMs": 2374.440999984741,
      "audioSeconds": 11,
      "realTimeFactor": 0.21585827272588556,
      "outputTokens": 29,
      "totalMs": 2753.9729999552947,
      "durationMs": 2374.7947499796283,
      "grade": "fast",
      "rssBytes": 9522888704,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 4,
      "device": "cpu",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0.045454545454545456,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 378.1811660155654,
      "placement": "llm_load_tensors: offloading 0 repeating layers to GPU\nllm_load_tensors: offloaded 0/29 layers to GPU\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)",
      "provider": "ggml CPU",
      "acceleration": "cpu",
      "inferenceMs": 2440.6050000190735,
      "audioSeconds": 11,
      "realTimeFactor": 0.22187318181991575,
      "outputTokens": 29,
      "totalMs": 2819.1074580245186,
      "durationMs": 2440.926292008953,
      "grade": "fast",
      "rssBytes": 4527620096,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 4,
      "device": "cpu",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 378.1811660155654,
      "placement": "llm_load_tensors: offloading 0 repeating layers to GPU\nllm_load_tensors: offloaded 0/29 layers to GPU\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)",
      "provider": "ggml CPU",
      "acceleration": "cpu",
      "inferenceMs": 1043.1389999985695,
      "audioSeconds": 3.4000000953674316,
      "realTimeFactor": 0.3068055796292086,
      "outputTokens": 14,
      "totalMs": 1421.6597500198986,
      "durationMs": 1043.4785840043332,
      "grade": "acceptable",
      "rssBytes": 4527751168,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 4,
      "device": "cpu",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 378.1811660155654,
      "placement": "llm_load_tensors: offloading 0 repeating layers to GPU\nllm_load_tensors: offloaded 0/29 layers to GPU\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)",
      "provider": "ggml CPU",
      "acceleration": "cpu",
      "inferenceMs": 7293.463999986649,
      "audioSeconds": 35.64437484741211,
      "realTimeFactor": 0.20461753169213392,
      "outputTokens": 95,
      "totalMs": 7672.014791023685,
      "durationMs": 7293.833625008119,
      "grade": "fast",
      "rssBytes": 9521594368,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 4,
      "device": "cpu",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 378.1811660155654,
      "placement": "llm_load_tensors: offloading 0 repeating layers to GPU\nllm_load_tensors: offloaded 0/29 layers to GPU\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)",
      "provider": "ggml CPU",
      "acceleration": "cpu",
      "inferenceMs": 620.3100000023842,
      "audioSeconds": 2,
      "realTimeFactor": 0.3101550000011921,
      "outputTokens": 4,
      "totalMs": 998.8083750067744,
      "durationMs": 620.627208991209,
      "grade": "acceptable",
      "rssBytes": 9521627136,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 4,
      "device": "cpu",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0.045454545454545456,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 378.1811660155654,
      "placement": "llm_load_tensors: offloading 0 repeating layers to GPU\nllm_load_tensors: offloaded 0/29 layers to GPU\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)",
      "provider": "ggml CPU",
      "acceleration": "cpu",
      "inferenceMs": 2365.5520000457764,
      "audioSeconds": 11,
      "realTimeFactor": 0.2150501818223433,
      "outputTokens": 29,
      "totalMs": 2744.057333038654,
      "durationMs": 2365.876167023089,
      "grade": "fast",
      "rssBytes": 9523265536,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 4,
      "device": "cpu",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0.045454545454545456,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 378.1811660155654,
      "placement": "llm_load_tensors: offloading 0 repeating layers to GPU\nllm_load_tensors: offloaded 0/29 layers to GPU\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)",
      "provider": "ggml CPU",
      "acceleration": "cpu",
      "inferenceMs": 2368.9330000281334,
      "audioSeconds": 11,
      "realTimeFactor": 0.21535754545710303,
      "outputTokens": 29,
      "totalMs": 2747.497166012181,
      "durationMs": 2369.3159999966156,
      "grade": "fast",
      "rssBytes": 9524314112,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 4,
      "device": "cpu",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0.045454545454545456,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 400.05970900529064,
      "placement": "llm_load_tensors: offloading 0 repeating layers to GPU\nllm_load_tensors: offloaded 0/29 layers to GPU\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)",
      "provider": "ggml CPU",
      "acceleration": "cpu",
      "inferenceMs": 2462.3199999928474,
      "audioSeconds": 11,
      "realTimeFactor": 0.2238472727266225,
      "outputTokens": 29,
      "totalMs": 2862.7008339972235,
      "durationMs": 2462.641124991933,
      "grade": "fast",
      "rssBytes": 4527489024,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 4,
      "device": "cpu",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 400.05970900529064,
      "placement": "llm_load_tensors: offloading 0 repeating layers to GPU\nllm_load_tensors: offloaded 0/29 layers to GPU\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)",
      "provider": "ggml CPU",
      "acceleration": "cpu",
      "inferenceMs": 1040.5020000338554,
      "audioSeconds": 3.4000000953674316,
      "realTimeFactor": 0.30602999142604737,
      "outputTokens": 14,
      "totalMs": 1440.9186679986306,
      "durationMs": 1040.85895899334,
      "grade": "acceptable",
      "rssBytes": 4527603712,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 4,
      "device": "cpu",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 400.05970900529064,
      "placement": "llm_load_tensors: offloading 0 repeating layers to GPU\nllm_load_tensors: offloaded 0/29 layers to GPU\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)",
      "provider": "ggml CPU",
      "acceleration": "cpu",
      "inferenceMs": 7349.965999960899,
      "audioSeconds": 35.64437484741211,
      "realTimeFactor": 0.2062026906468393,
      "outputTokens": 95,
      "totalMs": 7750.369292014511,
      "durationMs": 7350.309583009221,
      "grade": "fast",
      "rssBytes": 9520922624,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 4,
      "device": "cpu",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 400.05970900529064,
      "placement": "llm_load_tensors: offloading 0 repeating layers to GPU\nllm_load_tensors: offloaded 0/29 layers to GPU\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)",
      "provider": "ggml CPU",
      "acceleration": "cpu",
      "inferenceMs": 626.1610000133514,
      "audioSeconds": 2,
      "realTimeFactor": 0.3130805000066757,
      "outputTokens": 4,
      "totalMs": 1026.5933760092594,
      "durationMs": 626.5336670039687,
      "grade": "acceptable",
      "rssBytes": 9521070080,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 4,
      "device": "cpu",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0.045454545454545456,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 400.05970900529064,
      "placement": "llm_load_tensors: offloading 0 repeating layers to GPU\nllm_load_tensors: offloaded 0/29 layers to GPU\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)",
      "provider": "ggml CPU",
      "acceleration": "cpu",
      "inferenceMs": 2351.28100001812,
      "audioSeconds": 11,
      "realTimeFactor": 0.21375281818346545,
      "outputTokens": 29,
      "totalMs": 2751.673959020991,
      "durationMs": 2351.6142500157002,
      "grade": "fast",
      "rssBytes": 9521086464,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 4,
      "device": "cpu",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0.045454545454545456,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 400.05970900529064,
      "placement": "llm_load_tensors: offloading 0 repeating layers to GPU\nllm_load_tensors: offloaded 0/29 layers to GPU\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)",
      "provider": "ggml CPU",
      "acceleration": "cpu",
      "inferenceMs": 2405.242999970913,
      "audioSeconds": 11,
      "realTimeFactor": 0.21865845454281027,
      "outputTokens": 29,
      "totalMs": 2805.656959011685,
      "durationMs": 2405.5972500063945,
      "grade": "fast",
      "rssBytes": 9522544640,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 4,
      "device": "cpu",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "metal",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0.045454545454545456,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 379.93374999496154,
      "placement": "llm_load_tensors: offloading 28 repeating layers to GPU\nllm_load_tensors: offloading non-repeating layers to GPU\nllm_load_tensors: offloaded 29/29 layers to GPU\nllm_load_tensors:      Metal buffer size =   758.07 MiB\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)\nllama_kv_cache_init:      Metal KV buffer size =   448.00 MiB\nllama_new_context_with_model:      Metal compute buffer size =   428.00 MiB",
      "provider": "Metal LM + CPU VAE",
      "acceleration": "partial",
      "inferenceMs": 1826.6420000195503,
      "audioSeconds": 11,
      "realTimeFactor": 0.16605836363814094,
      "outputTokens": 29,
      "totalMs": 2206.848500005435,
      "durationMs": 1826.9147500104737,
      "grade": "fast",
      "rssBytes": 4519067648,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 4,
      "device": "metal",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "metal",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 379.93374999496154,
      "placement": "llm_load_tensors: offloading 28 repeating layers to GPU\nllm_load_tensors: offloading non-repeating layers to GPU\nllm_load_tensors: offloaded 29/29 layers to GPU\nllm_load_tensors:      Metal buffer size =   758.07 MiB\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)\nllama_kv_cache_init:      Metal KV buffer size =   448.00 MiB\nllama_new_context_with_model:      Metal compute buffer size =   428.00 MiB",
      "provider": "Metal LM + CPU VAE",
      "acceleration": "partial",
      "inferenceMs": 733.4830000400543,
      "audioSeconds": 3.4000000953674316,
      "realTimeFactor": 0.21573028807835612,
      "outputTokens": 14,
      "totalMs": 1113.7357499974314,
      "durationMs": 733.8020000024699,
      "grade": "fast",
      "rssBytes": 4520050688,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 4,
      "device": "metal",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "metal",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 379.93374999496154,
      "placement": "llm_load_tensors: offloading 28 repeating layers to GPU\nllm_load_tensors: offloading non-repeating layers to GPU\nllm_load_tensors: offloaded 29/29 layers to GPU\nllm_load_tensors:      Metal buffer size =   758.07 MiB\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)\nllama_kv_cache_init:      Metal KV buffer size =   448.00 MiB\nllama_new_context_with_model:      Metal compute buffer size =   428.00 MiB",
      "provider": "Metal LM + CPU VAE",
      "acceleration": "partial",
      "inferenceMs": 5557.037000000477,
      "audioSeconds": 35.64437484741211,
      "realTimeFactor": 0.15590221525245618,
      "outputTokens": 96,
      "totalMs": 5937.341667013243,
      "durationMs": 5557.407917018281,
      "grade": "fast",
      "rssBytes": 9495855104,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 4,
      "device": "metal",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "metal",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 379.93374999496154,
      "placement": "llm_load_tensors: offloading 28 repeating layers to GPU\nllm_load_tensors: offloading non-repeating layers to GPU\nllm_load_tensors: offloaded 29/29 layers to GPU\nllm_load_tensors:      Metal buffer size =   758.07 MiB\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)\nllama_kv_cache_init:      Metal KV buffer size =   448.00 MiB\nllama_new_context_with_model:      Metal compute buffer size =   428.00 MiB",
      "provider": "Metal LM + CPU VAE",
      "acceleration": "partial",
      "inferenceMs": 422.3449999690056,
      "audioSeconds": 2,
      "realTimeFactor": 0.2111724999845028,
      "outputTokens": 4,
      "totalMs": 802.5926249683835,
      "durationMs": 422.6588749734219,
      "grade": "fast",
      "rssBytes": 9496068096,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 4,
      "device": "metal",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "metal",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0.045454545454545456,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 379.93374999496154,
      "placement": "llm_load_tensors: offloading 28 repeating layers to GPU\nllm_load_tensors: offloading non-repeating layers to GPU\nllm_load_tensors: offloaded 29/29 layers to GPU\nllm_load_tensors:      Metal buffer size =   758.07 MiB\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)\nllama_kv_cache_init:      Metal KV buffer size =   448.00 MiB\nllama_new_context_with_model:      Metal compute buffer size =   428.00 MiB",
      "provider": "Metal LM + CPU VAE",
      "acceleration": "partial",
      "inferenceMs": 1765.4950000047684,
      "audioSeconds": 11,
      "realTimeFactor": 0.16049954545497894,
      "outputTokens": 29,
      "totalMs": 2145.770208007889,
      "durationMs": 1765.8364580129273,
      "grade": "fast",
      "rssBytes": 9498214400,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 4,
      "device": "metal",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "metal",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0.045454545454545456,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 379.93374999496154,
      "placement": "llm_load_tensors: offloading 28 repeating layers to GPU\nllm_load_tensors: offloading non-repeating layers to GPU\nllm_load_tensors: offloaded 29/29 layers to GPU\nllm_load_tensors:      Metal buffer size =   758.07 MiB\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)\nllama_kv_cache_init:      Metal KV buffer size =   448.00 MiB\nllama_new_context_with_model:      Metal compute buffer size =   428.00 MiB",
      "provider": "Metal LM + CPU VAE",
      "acceleration": "partial",
      "inferenceMs": 1762.033999979496,
      "audioSeconds": 11,
      "realTimeFactor": 0.1601849090890451,
      "outputTokens": 29,
      "totalMs": 2142.2799580032006,
      "durationMs": 1762.346208008239,
      "grade": "fast",
      "rssBytes": 9498329088,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 4,
      "device": "metal",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "metal",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0.045454545454545456,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 384.98616599827074,
      "placement": "llm_load_tensors: offloading 28 repeating layers to GPU\nllm_load_tensors: offloading non-repeating layers to GPU\nllm_load_tensors: offloaded 29/29 layers to GPU\nllm_load_tensors:      Metal buffer size =   758.07 MiB\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)\nllama_kv_cache_init:      Metal KV buffer size =   448.00 MiB\nllama_new_context_with_model:      Metal compute buffer size =   428.00 MiB",
      "provider": "Metal LM + CPU VAE",
      "acceleration": "partial",
      "inferenceMs": 1861.285000026226,
      "audioSeconds": 11,
      "realTimeFactor": 0.16920772727511146,
      "outputTokens": 29,
      "totalMs": 2246.587665984407,
      "durationMs": 1861.6014999861363,
      "grade": "fast",
      "rssBytes": 4517871616,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 4,
      "device": "metal",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "metal",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 384.98616599827074,
      "placement": "llm_load_tensors: offloading 28 repeating layers to GPU\nllm_load_tensors: offloading non-repeating layers to GPU\nllm_load_tensors: offloaded 29/29 layers to GPU\nllm_load_tensors:      Metal buffer size =   758.07 MiB\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)\nllama_kv_cache_init:      Metal KV buffer size =   448.00 MiB\nllama_new_context_with_model:      Metal compute buffer size =   428.00 MiB",
      "provider": "Metal LM + CPU VAE",
      "acceleration": "partial",
      "inferenceMs": 748.5569999814034,
      "audioSeconds": 3.4000000953674316,
      "realTimeFactor": 0.2201638173485134,
      "outputTokens": 14,
      "totalMs": 1133.8532499794383,
      "durationMs": 748.8670839811675,
      "grade": "fast",
      "rssBytes": 4518871040,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 4,
      "device": "metal",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "metal",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 384.98616599827074,
      "placement": "llm_load_tensors: offloading 28 repeating layers to GPU\nllm_load_tensors: offloading non-repeating layers to GPU\nllm_load_tensors: offloaded 29/29 layers to GPU\nllm_load_tensors:      Metal buffer size =   758.07 MiB\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)\nllama_kv_cache_init:      Metal KV buffer size =   448.00 MiB\nllama_new_context_with_model:      Metal compute buffer size =   428.00 MiB",
      "provider": "Metal LM + CPU VAE",
      "acceleration": "partial",
      "inferenceMs": 5661.7419999837875,
      "audioSeconds": 35.64437484741211,
      "realTimeFactor": 0.15883970540150594,
      "outputTokens": 96,
      "totalMs": 6047.160582995275,
      "durationMs": 5662.174416997004,
      "grade": "fast",
      "rssBytes": 9495003136,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 4,
      "device": "metal",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "metal",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 384.98616599827074,
      "placement": "llm_load_tensors: offloading 28 repeating layers to GPU\nllm_load_tensors: offloading non-repeating layers to GPU\nllm_load_tensors: offloaded 29/29 layers to GPU\nllm_load_tensors:      Metal buffer size =   758.07 MiB\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)\nllama_kv_cache_init:      Metal KV buffer size =   448.00 MiB\nllama_new_context_with_model:      Metal compute buffer size =   428.00 MiB",
      "provider": "Metal LM + CPU VAE",
      "acceleration": "partial",
      "inferenceMs": 421.89800000190735,
      "audioSeconds": 2,
      "realTimeFactor": 0.21094900000095368,
      "outputTokens": 4,
      "totalMs": 807.2174580011051,
      "durationMs": 422.2312920028344,
      "grade": "fast",
      "rssBytes": 9495314432,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 4,
      "device": "metal",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "metal",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0.045454545454545456,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 384.98616599827074,
      "placement": "llm_load_tensors: offloading 28 repeating layers to GPU\nllm_load_tensors: offloading non-repeating layers to GPU\nllm_load_tensors: offloaded 29/29 layers to GPU\nllm_load_tensors:      Metal buffer size =   758.07 MiB\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)\nllama_kv_cache_init:      Metal KV buffer size =   448.00 MiB\nllama_new_context_with_model:      Metal compute buffer size =   428.00 MiB",
      "provider": "Metal LM + CPU VAE",
      "acceleration": "partial",
      "inferenceMs": 1741.6279999613762,
      "audioSeconds": 11,
      "realTimeFactor": 0.15832981817830694,
      "outputTokens": 29,
      "totalMs": 2126.9271240162198,
      "durationMs": 1741.940958017949,
      "grade": "fast",
      "rssBytes": 9495330816,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 4,
      "device": "metal",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "metal",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0.045454545454545456,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 384.98616599827074,
      "placement": "llm_load_tensors: offloading 28 repeating layers to GPU\nllm_load_tensors: offloading non-repeating layers to GPU\nllm_load_tensors: offloaded 29/29 layers to GPU\nllm_load_tensors:      Metal buffer size =   758.07 MiB\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)\nllama_kv_cache_init:      Metal KV buffer size =   448.00 MiB\nllama_new_context_with_model:      Metal compute buffer size =   428.00 MiB",
      "provider": "Metal LM + CPU VAE",
      "acceleration": "partial",
      "inferenceMs": 1799.555999994278,
      "audioSeconds": 11,
      "realTimeFactor": 0.1635959999994798,
      "outputTokens": 29,
      "totalMs": 2184.8887910018675,
      "durationMs": 1799.9026250035968,
      "grade": "fast",
      "rssBytes": 9495576576,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 4,
      "device": "metal",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "metal",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0.045454545454545456,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 384.9825829965994,
      "placement": "llm_load_tensors: offloading 28 repeating layers to GPU\nllm_load_tensors: offloading non-repeating layers to GPU\nllm_load_tensors: offloaded 29/29 layers to GPU\nllm_load_tensors:      Metal buffer size =   758.07 MiB\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)\nllama_kv_cache_init:      Metal KV buffer size =   448.00 MiB\nllama_new_context_with_model:      Metal compute buffer size =   428.00 MiB",
      "provider": "Metal LM + CPU VAE",
      "acceleration": "partial",
      "inferenceMs": 1880.701000034809,
      "audioSeconds": 11,
      "realTimeFactor": 0.17097281818498267,
      "outputTokens": 29,
      "totalMs": 2266.010583000025,
      "durationMs": 1881.0280000034254,
      "grade": "fast",
      "rssBytes": 4518199296,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 4,
      "device": "metal",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "metal",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 384.9825829965994,
      "placement": "llm_load_tensors: offloading 28 repeating layers to GPU\nllm_load_tensors: offloading non-repeating layers to GPU\nllm_load_tensors: offloaded 29/29 layers to GPU\nllm_load_tensors:      Metal buffer size =   758.07 MiB\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)\nllama_kv_cache_init:      Metal KV buffer size =   448.00 MiB\nllama_new_context_with_model:      Metal compute buffer size =   428.00 MiB",
      "provider": "Metal LM + CPU VAE",
      "acceleration": "partial",
      "inferenceMs": 744.6560000181198,
      "audioSeconds": 3.4000000953674316,
      "realTimeFactor": 0.21901646445031828,
      "outputTokens": 14,
      "totalMs": 1130.3046669927426,
      "durationMs": 745.3220839961432,
      "grade": "fast",
      "rssBytes": 4518903808,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 4,
      "device": "metal",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "metal",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 384.9825829965994,
      "placement": "llm_load_tensors: offloading 28 repeating layers to GPU\nllm_load_tensors: offloading non-repeating layers to GPU\nllm_load_tensors: offloaded 29/29 layers to GPU\nllm_load_tensors:      Metal buffer size =   758.07 MiB\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)\nllama_kv_cache_init:      Metal KV buffer size =   448.00 MiB\nllama_new_context_with_model:      Metal compute buffer size =   428.00 MiB",
      "provider": "Metal LM + CPU VAE",
      "acceleration": "partial",
      "inferenceMs": 5649.620000004768,
      "audioSeconds": 35.64437484741211,
      "realTimeFactor": 0.15849962369069148,
      "outputTokens": 96,
      "totalMs": 6034.94391599088,
      "durationMs": 5649.96133299428,
      "grade": "fast",
      "rssBytes": 9495920640,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 4,
      "device": "metal",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "metal",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 384.9825829965994,
      "placement": "llm_load_tensors: offloading 28 repeating layers to GPU\nllm_load_tensors: offloading non-repeating layers to GPU\nllm_load_tensors: offloaded 29/29 layers to GPU\nllm_load_tensors:      Metal buffer size =   758.07 MiB\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)\nllama_kv_cache_init:      Metal KV buffer size =   448.00 MiB\nllama_new_context_with_model:      Metal compute buffer size =   428.00 MiB",
      "provider": "Metal LM + CPU VAE",
      "acceleration": "partial",
      "inferenceMs": 422.92099994421005,
      "audioSeconds": 2,
      "realTimeFactor": 0.21146049997210503,
      "outputTokens": 4,
      "totalMs": 808.2556249864865,
      "durationMs": 423.27304198988713,
      "grade": "fast",
      "rssBytes": 9495920640,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 4,
      "device": "metal",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "metal",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0.045454545454545456,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 384.9825829965994,
      "placement": "llm_load_tensors: offloading 28 repeating layers to GPU\nllm_load_tensors: offloading non-repeating layers to GPU\nllm_load_tensors: offloaded 29/29 layers to GPU\nllm_load_tensors:      Metal buffer size =   758.07 MiB\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)\nllama_kv_cache_init:      Metal KV buffer size =   448.00 MiB\nllama_new_context_with_model:      Metal compute buffer size =   428.00 MiB",
      "provider": "Metal LM + CPU VAE",
      "acceleration": "partial",
      "inferenceMs": 1776.722000002861,
      "audioSeconds": 11,
      "realTimeFactor": 0.1615201818184419,
      "outputTokens": 29,
      "totalMs": 2161.9989999744575,
      "durationMs": 1777.0164169778582,
      "grade": "fast",
      "rssBytes": 9496559616,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 4,
      "device": "metal",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  },
  {
    "precision": "i2_s",
    "device": "metal",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0.045454545454545456,
      "streamParity": true,
      "scope": "text transcription; no speaker/timestamp claim"
    },
    "runtime": {
      "engine": "VibeASR.cpp",
      "version": "c4334009c88060f86cdbbd684b62662f710b6c20",
      "loadMs": 384.9825829965994,
      "placement": "llm_load_tensors: offloading 28 repeating layers to GPU\nllm_load_tensors: offloading non-repeating layers to GPU\nllm_load_tensors: offloaded 29/29 layers to GPU\nllm_load_tensors:      Metal buffer size =   758.07 MiB\nggml_metal_init: GPU name:   Apple M3 Max\nggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)\nggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)\nggml_metal_init: GPU family: MTLGPUFamilyMetal4  (5002)\nllama_kv_cache_init:      Metal KV buffer size =   448.00 MiB\nllama_new_context_with_model:      Metal compute buffer size =   428.00 MiB",
      "provider": "Metal LM + CPU VAE",
      "acceleration": "partial",
      "inferenceMs": 1796.1930000185966,
      "audioSeconds": 11,
      "realTimeFactor": 0.16329027272896332,
      "outputTokens": 29,
      "totalMs": 2181.4777080144268,
      "durationMs": 1796.4951250178274,
      "grade": "fast",
      "rssBytes": 9496559616,
      "memoryMeasurementScope": "Post-inference RSS, not peak or validated minimum",
      "cpuThreads": 4,
      "device": "metal",
      "precision": "i2_s",
      "artifactRevision": "66e78021ab8f5f06133d1ab421ba4d348bda97c9"
    }
  }
]

Something not quite right?

Tell us what happened with VibeVoice ASR BitNet. 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.