config.json

{
  "model_type": "speech-llm",
  "preprocessor": "identity",
  "hidden_size": 3072,
  "num_layers": 32,
  "num_key_value_heads": 8,
  "head_dim": 128,
  "eos_token_ids": [
    200020,
    199999
  ],
  "max_sequence_length": 448,
  "prompt_prefix_ids": [
    200021
  ],
  "prompt_suffix_ids": [
    3248,
    4981,
    290,
    11065,
    19925,
    1511,
    2201,
    13,
    200020,
    200019
  ],
  "prompt_text": "<|user|><|audio_1|>Transcribe the audio clip into text.<|end|><|assistant|>",
  "source_model": "microsoft/Phi-4-multimodal-instruct"
}