{
    "architectures": [
        "MiniMaxMusic3ForConditionalGeneration"
    ],
    "audio_cfg_token_id": 151654,
    "audio_code_offset": 151675,
    "audio_end_token_id": 151670,
    "audio_vocab_size": 1024,
    "condition_out_dim": 2048,
    "depth_intermediate_size": 6144,
    "depth_max_position_embeddings": 16,
    "depth_num_heads": 16,
    "depth_num_layers": 4,
    "dit_ff_inner_dim": 8192,
    "dit_fourier_dim": 256,
    "dit_head_dim": 64,
    "dit_in_channels": 128,
    "dit_num_heads": 32,
    "dit_num_layers": 36,
    "dit_rotary_dim": 32,
    "extra": {},
    "frame_rate": 25.0,
    "head_dim": 128,
    "hidden_size": 4096,
    "input_hop_length": 960,
    "input_sampling_rate": 24000,
    "intermediate_size": 12288,
    "max_position_embeddings": 10240,
    "model_path": "",
    "model_type": "minimax_music3",
    "num_attention_heads": 32,
    "num_codebooks": 8,
    "num_condition_layers": 8,
    "num_hidden_layers": 36,
    "num_key_value_heads": 8,
    "output_hop_length": 512,
    "output_sampling_rate": 44100,
    "quantization": {
        "group_size": 32,
        "bits": 4,
        "mode": "mxfp4"
    },
    "quantization_config": {
        "group_size": 32,
        "bits": 4,
        "mode": "mxfp4"
    },
    "rms_norm_eps": 1e-06,
    "rope_theta": 1000000.0,
    "sample_rate": 44100,
    "semantic_vocab_size": 16384,
    "tie_word_embeddings": false,
    "torch_dtype": "bfloat16",
    "vocab_size": 200000,
    "vocoder_hidden_dim": 1536,
    "vocoder_input_dim": 1024,
    "vocoder_upsampling_ratios": [
        8,
        8,
        4,
        2
    ]
}