config.json

{
  "model_type": "audiogen",
  "nQ": 4,
  "card": 2048,
  "dim": 1536,
  "numHeads": 24,
  "hiddenScale": 4,
  "numLayers": 48,
  "causal": true,
  "crossAttention": true,
  "dropout": 0.0,
  "activation": "gelu",
  "norm": "layer_norm",
  "normFirst": true,
  "biasFF": false,
  "biasAttn": false,
  "layerScale": null,
  "context": 4096,
  "maxPeriod": 10000,
  "positionalEmbedding": "sin",
  "positionalScale": 1.0,
  "xPos": false,
  "weight": 1.0,
  "conditionProvider": "t5",
  "twoStepCFG": false,
  "kvRepeat": 1,
  "qkLayerNorm": false,
  "emptyLikeInit": false,
  "emptyLikeInitDetokenized": false,
  "zeroHypothesisRate": 0.0,
  "quantize": false,
  "weightsPerStep": null,
  "normalize": true,
  "frameRate": 50.0,
  "sampleRate": 16000,
  "duration": 10.0,
  "numSamples": 1,
  "specialToken": 2048,
  "tokenizer": "t5-large",
  "t5_model_name": "t5-large",
  "clsToken": 2048,
  "padToken": 2048,
  "encodec": {
    "model_type": "encodec",
    "audio_channels": 1,
    "num_filters": 32,
    "kernel_size": 7,
    "num_residual_layers": 1,
    "dilation_growth_rate": 2,
    "codebook_size": 2048,
    "codebook_dim": 128,
    "hidden_size": 128,
    "num_lstm_layers": 2,
    "residual_kernel_size": 3,
    "use_causal_conv": true,
    "normalize": false,
    "pad_mode": "reflect",
    "norm_type": "time_group_norm",
    "last_kernel_size": 7,
    "trim_right_ratio": 1.0,
    "compress": 2,
    "upsampling_ratios": [8, 5, 4, 2],
    "target_bandwidths": [1.5, 3.0, 6.0, 12.0, 24.0],
    "sampling_rate": 16000,
    "chunk_length_s": null,
    "overlap": null,
    "use_conv_shortcut": false
  },
  "t5": {
    "model_name": "t5-large",
    "d_model": 1024,
    "d_kv": 64,
    "d_ff": 4096,
    "num_layers": 24,
    "num_heads": 16,
    "relative_attention_num_buckets": 32,
    "relative_attention_max_distance": 128,
    "dropout_rate": 0.0,
    "layer_norm_epsilon": 1e-06,
    "feed_forward_proj": "relu",
    "vocab_size": 32128,
    "tie_word_embeddings": true
  }
}