metadata.json

{
  "model_id": "nvidia/parakeet-tdt-0.6b-v2",
  "sample_rate": 16000,
  "max_audio_seconds": 15.0,
  "max_audio_samples": 240000,
  "max_symbol_steps": 1,
  "vocab_size": 1024,
  "joint_extra_outputs": 5,
  "checkpoint": {
    "type": "file",
    "path": "/Users/eddyalmeida/Desktop/ralphtrain/conversion/mobius/models/stt/parakeet-tdt-v2-0.6b/coreml/models/p-rad-a40.nemo"
  },
  "coreml": {
    "compute_units": "CPU_ONLY",
    "compute_precision": "FLOAT16"
  },
  "components": {
    "preprocessor": {
      "inputs": {
        "audio_signal": [
          1,
          240000
        ],
        "audio_length": [
          1
        ]
      },
      "outputs": {
        "mel": [
          1,
          128,
          1501
        ],
        "mel_length": [
          1
        ]
      },
      "path": "parakeet_preprocessor.mlpackage"
    },
    "encoder": {
      "inputs": {
        "mel": [
          1,
          128,
          1501
        ],
        "mel_length": [
          1
        ]
      },
      "outputs": {
        "encoder": [
          1,
          1024,
          188
        ],
        "encoder_length": [
          1
        ]
      },
      "path": "parakeet_encoder.mlpackage"
    },
    "mel_encoder": {
      "inputs": {
        "audio_signal": [
          1,
          240000
        ],
        "audio_length": [
          1
        ]
      },
      "outputs": {
        "encoder": [
          1,
          1024,
          188
        ],
        "encoder_length": [
          1
        ]
      },
      "path": "parakeet_mel_encoder.mlpackage"
    },
    "decoder": {
      "inputs": {
        "targets": [
          1,
          1
        ],
        "target_length": [
          1
        ],
        "h_in": [
          2,
          1,
          640
        ],
        "c_in": [
          2,
          1,
          640
        ]
      },
      "outputs": {
        "decoder": [
          1,
          640,
          1
        ],
        "h_out": [
          2,
          1,
          640
        ],
        "c_out": [
          2,
          1,
          640
        ]
      },
      "path": "parakeet_decoder.mlpackage"
    },
    "joint": {
      "inputs": {
        "encoder": [
          1,
          1024,
          188
        ],
        "decoder": [
          1,
          640,
          1
        ]
      },
      "outputs": {
        "logits": [
          1,
          188,
          1,
          1030
        ]
      },
      "path": "parakeet_joint.mlpackage"
    },
    "joint_decision": {
      "inputs": {
        "encoder": [
          1,
          1024,
          188
        ],
        "decoder": [
          1,
          640,
          1
        ]
      },
      "outputs": {
        "token_id": [
          1,
          1024,
          640
        ],
        "token_prob": [
          1,
          1024,
          640
        ],
        "duration": [
          1,
          1024,
          640
        ]
      },
      "path": "parakeet_joint_decision.mlpackage"
    },
    "joint_decision_single_step": {
      "inputs": {
        "encoder_step": [
          1,
          1024,
          1
        ],
        "decoder_step": [
          1,
          640,
          1
        ]
      },
      "outputs": {
        "token_id": [
          1,
          1,
          1
        ],
        "token_prob": [
          1,
          1,
          1
        ],
        "duration": [
          1,
          1,
          1
        ]
      },
      "path": "parakeet_joint_decision_single_step.mlpackage"
    }
  }
}