{
  "gpu": {
    "available": true,
    "source": "nvidia-smi",
    "devices": [
      {
        "index": 0,
        "name": "NVIDIA GeForce RTX 5090",
        "capability": "sm_120",
        "memory_total": "32607 MiB",
        "memory_used": "30725 MiB"
      },
      {
        "index": 1,
        "name": "NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition",
        "capability": "sm_120",
        "memory_total": "97887 MiB",
        "memory_used": "90660 MiB"
      }
    ],
    "detail": "torch not importable; GPU context collected with nvidia-smi"
  },
  "reference_text": "The quick brown fox jumps over the lazy dog. Anvil serving routes local models where proven and cloud where not.",
  "sample": "C:\\Users\\operator\\AppData\\Local\\Temp\\claude\\C--Users-operator-ai-code-anvil-serving\\21eba8ab-885c-4f97-acef-864745e5c375\\scratchpad\\test_stt.wav",
  "candidates": [
    {
      "name": "parakeet.cpp",
      "base_url": "http://127.0.0.1:30010/v1",
      "model": "tdt_ctc-110m",
      "stream": false,
      "latency_ms": 148.36,
      "hypothesis": "The quick brown fox jumps over the lazy dog. Anvil serving routes local models where proven and cloud wear not.",
      "wer": 0.05,
      "error": null,
      "readiness": {
        "docker_state": "raw-endpoint",
        "ready": true,
        "detail": "raw endpoint health/container probe; no managed serve_name given",
        "health": {
          "ready": true,
          "url": "http://127.0.0.1:30010/health",
          "status": 200
        },
        "container": {
          "name": "parakeet-stt",
          "available": true,
          "status": "running",
          "image": "ghcr.io/mudler/parakeet.cpp-server:latest-cuda",
          "docker_name": "parakeet-stt"
        }
      }
    },
    {
      "name": "vllm-whisper-tiny",
      "base_url": "http://127.0.0.1:30015/v1",
      "model": "whisper-tiny",
      "stream": false,
      "latency_ms": 162.46,
      "hypothesis": " The quick brown fox jumps over the lazy dog. Andville serving roots local models where proven and cloud were not.",
      "wer": 0.15,
      "error": null,
      "readiness": {
        "docker_state": "raw-endpoint",
        "ready": true,
        "detail": "raw endpoint health/container probe; no managed serve_name given",
        "health": {
          "ready": true,
          "url": "http://127.0.0.1:30015/health",
          "status": 200
        },
        "container": {
          "name": "anvil-stt-vllm-whisper-tiny-eager-test",
          "available": true,
          "status": "running",
          "image": "vllm/vllm-openai:nightly",
          "docker_name": "anvil-stt-vllm-whisper-tiny-eager-test"
        }
      }
    }
  ]
}