{
  "schema_version": "direct-multimodal-probe/v1",
  "date": "2026-07-27",
  "promotion_quality_evidence": false,
  "identity": {
    "base_url": "http://127.0.0.1:30013/v1",
    "model": "qwen25-omni-3b",
    "checkpoint_revision": "f75b40e3da2003cdd6e1829b1f420ca70797c34e",
    "container": "vllm-qwen25-omni-3b",
    "image": "anvil-vllm:omni-small-audio-a65f93fb2"
  },
  "input": {
    "type": "audio_url",
    "source": "https://github.com/vllm-project/vllm/raw/main/examples/offline_inference/multimodal_asset/mary_had_lamb.mp3"
  },
  "result": {
    "http_status": 200,
    "expected_terms": [
      "Mary",
      "lamb"
    ],
    "response_text": "Hey first words, let us see. The first words spoken in the recording. There is a little lamb, its fleece was white as snow. The first word spoken in the recording. Mary had a little lamb. The first word is Mary. Mary.",
    "expected_terms_present": true,
    "pass": true
  },
  "caveat": "This direct probe establishes accepted audio input and basic semantic recognition only. The response was noisy and is not a transcription-quality result; dedicated Parakeet remains the STT endpoint."
}