config.json

{
  "architectures": [
    "StepAudio2ForCausalLM"
  ],
  "auto_map": {
    "AutoConfig": "configuration_step_audio_2.StepAudio2Config",
    "AutoModelForCausalLM": "modeling_step_audio_2.StepAudio2ForCausalLM"
  },
  "model_type": "step_audio_2",
  "text_config": {
    "hidden_size": 5120,
    "intermediate_size": 27648,
    "num_attention_heads": 40,
    "num_attention_groups": 8,
    "num_key_value_heads": 8,
    "num_hidden_layers": 64,
    "max_seq_len": 65536,
    "vocab_size": 158720,
    "rms_norm_eps": 1e-05,
    "eos_token_id": 151643,
    "pad_token_id": 151643,
    "rope_theta": 1000000.0,
    "max_position_embeddings": 65536,
    "rope_scaling": null,
    "torch_dtype": "bfloat16"
  },
  "audio_encoder_config": {
    "n_mels": 128,
    "n_audio_ctx": 1500,
    "n_audio_state": 1280, 
    "n_audio_head": 20,
    "n_audio_layer": 32,
    "n_codebook_size": 4096,
    "llm_dim": 5120,
    "kernel_size": 3,
    "adapter_stride": 2
  }
}