config.json

{
  "format": "songgeneration-mlx-audiolm",
  "source": "tencent/SongGeneration",
  "official_code": "https://github.com/tencent-ailab/songgeneration",
  "variant": "v2-medium",
  "precision": "4bit",
  "runtime": {
    "hidden_size": 1536,
    "intermediate_size": 8960,
    "num_heads": 12,
    "num_layers": 28,
    "num_layers_sub": 12,
    "code_depth": 3,
    "code_size": 16384,
    "prompt_len": 10,
    "frame_rate": 25,
    "max_position_embeddings": 10000,
    "max_position_embeddings_sub": 10000,
    "rope_theta": 500000.0,
    "rope_theta_sub": 500000.0,
    "cfg_coef": 1.5
  },
  "components": {
    "audiolm": "converted to MLX safetensors",
    "qwen_tokenizer": "copied for lyric/style tokenization",
    "audio_decoder": "use official PyTorch Flow1dVAE/separate tokenizer bridge for now"
  },
  "skipped_keys": [
    "audiolm.out_norm.weight",
    "audiolm.out_norm.bias",
    "audiolm.transformer2.lm_head.weight"
  ],
  "quantization": {
    "bits": 4,
    "group_size": 64,
    "mode": "affine"
  },
  "weight_files": [
    "model-00001-of-00024.safetensors",
    "model-00002-of-00024.safetensors",
    "model-00003-of-00024.safetensors",
    "model-00004-of-00024.safetensors",
    "model-00005-of-00024.safetensors",
    "model-00006-of-00024.safetensors",
    "model-00007-of-00024.safetensors",
    "model-00008-of-00024.safetensors",
    "model-00009-of-00024.safetensors",
    "model-00010-of-00024.safetensors",
    "model-00011-of-00024.safetensors",
    "model-00012-of-00024.safetensors",
    "model-00013-of-00024.safetensors",
    "model-00014-of-00024.safetensors",
    "model-00015-of-00024.safetensors",
    "model-00016-of-00024.safetensors",
    "model-00017-of-00024.safetensors",
    "model-00018-of-00024.safetensors",
    "model-00019-of-00024.safetensors",
    "model-00020-of-00024.safetensors",
    "model-00021-of-00024.safetensors",
    "model-00022-of-00024.safetensors",
    "model-00023-of-00024.safetensors",
    "model-00024-of-00024.safetensors"
  ]
}