audio_tokenizer/config.json

{
  "_class_name": "AceStepAudioTokenizer",
  "_diffusers_version": "0.39.0.dev0",
  "attention_bias": false,
  "attention_dropout": 0.0,
  "audio_acoustic_hidden_dim": 8,
  "fsq_dim": 32,
  "fsq_input_levels": [
    4,
    4,
    4
  ],
  "fsq_input_num_quantizers": 1,
  "head_dim": 8,
  "hidden_size": 32,
  "intermediate_size": 64,
  "layer_types": null,
  "num_attention_heads": 4,
  "num_attention_pooler_hidden_layers": 1,
  "num_key_value_heads": 2,
  "pool_window_size": 2,
  "rms_norm_eps": 1e-06,
  "rope_theta": 10000.0,
  "sliding_window": 16
}