config.json

{
  "_commit_hash": null,
  "_name_or_path": "jukebox-1b-lyrics",
  "architectures": [
    "JukeboxModel"
  ],
  "hop_fraction": [
    0.125,
    0.5,
    0.5
  ],
  "init_std": 0.2,
  "max_duration": 600.0,
  "max_nb_genres": 1,
  "metadata_conditioning": true,
  "min_duration": 17.84,
  "model_type": "jukebox",
  "nb_priors": 3,
  "prior_0": {
    "act_fn": "quick_gelu",
    "alignment_head": 2,
    "alignment_layer": 68,
    "attention_multiplier": 0.25,
    "attention_pattern": "enc_dec_with_lyrics",
    "attn_dropout": 0,
    "attn_res_scale": false,
    "blocks": 64,
    "conv_res_scale": false,
    "emb_dropout": 0,
    "encoder_config": null,
    "encoder_loss_fraction": 0.4,
    "hidden_size": 2048,
    "init_scale": 0.2,
    "is_encoder_decoder": true,
    "level": 0,
    "lyric_vocab_size": 79,
    "mask": true,
    "max_duration": 600.0,
    "max_nb_genres": 1,
    "merged_decoder": false,
    "metadata_conditioning": true,
    "metadata_dims": [
      604,
      7898
    ],
    "min_duration": 17.84,
    "mlp_multiplier": 1.0,
    "model_type": "jukebox_prior",
    "music_vocab_size": 2048,
    "n_ctx": 6144,
    "n_heads": 2,
    "nb_relevant_lyric_tokens": 384,
    "num_layers": 72,
    "res_conv_depth": null,
    "res_conv_width": null,
    "res_convolution_multiplier": null,
    "res_dilation_cycle": null,
    "res_dilation_growth_rate": null,
    "res_downs_t": [
      3,
      2,
      2
    ],
    "res_strides_t": [
      2,
      2,
      2
    ],
    "resid_dropout": 0,
    "sampling_rate": 44100,
    "spread": null,
    "timing_dims": 64,
    "zero_out": false
  },
  "prior_1": {
    "act_fn": "quick_gelu",
    "alignment_head": null,
    "alignment_layer": null,
    "attention_multiplier": 0.25,
    "attention_pattern": "RawColumnPreviousRowAttention",
    "attn_dropout": 0,
    "attn_res_scale": false,
    "blocks": 64,
    "conv_res_scale": true,
    "emb_dropout": 0,
    "embed_dim": 2048,
    "encoder_config": null,
    "encoder_loss_fraction": 0,
    "hidden_size": 1920,
    "init_scale": 1,
    "is_encoder_decoder": false,
    "level": 1,
    "lyric_vocab_size": 80,
    "mask": true,
    "max_duration": 600.0,
    "max_nb_genres": 1,
    "merged_decoder": false,
    "metadata_conditioning": true,
    "metadata_dims": [
      120,
      4111
    ],
    "min_duration": 17.84,
    "mlp_multiplier": 1.0,
    "model_type": "jukebox_prior",
    "music_vocab_size": 2048,
    "n_ctx": 8192,
    "n_heads": 1,
    "nb_relevant_lyric_tokens": 0,
    "num_layers": 72,
    "res_conv_depth": 16,
    "res_conv_width": 1024,
    "res_convolution_multiplier": 1,
    "res_dilation_cycle": 8,
    "res_dilation_growth_rate": 3,
    "res_downs_t": [
      3,
      2,
      2
    ],
    "res_strides_t": [
      2,
      2,
      2
    ],
    "resid_dropout": 0,
    "sampling_rate": 44100,
    "spread": null,
    "timing_dims": 64,
    "zero_out": false
  },
  "prior_2": {
    "act_fn": "quick_gelu",
    "alignment_head": null,
    "alignment_layer": null,
    "attention_multiplier": 0.25,
    "attention_pattern": "RawColumnPreviousRowAttention",
    "attn_dropout": 0,
    "attn_res_scale": false,
    "blocks": 64,
    "conv_res_scale": null,
    "emb_dropout": 0,
    "embed_dim": 2048,
    "encoder_config": null,
    "encoder_loss_fraction": 0,
    "hidden_size": 1920,
    "init_scale": 1,
    "is_encoder_decoder": false,
    "level": 2,
    "lyric_vocab_size": 80,
    "mask": true,
    "max_duration": 600.0,
    "max_nb_genres": 1,
    "merged_decoder": false,
    "metadata_conditioning": true,
    "metadata_dims": [
      120,
      4111
    ],
    "min_duration": 17.84,
    "mlp_multiplier": 1.0,
    "model_type": "jukebox_prior",
    "music_vocab_size": 2048,
    "n_ctx": 8192,
    "n_heads": 1,
    "nb_relevant_lyric_tokens": 0,
    "num_layers": 72,
    "res_conv_depth": 16,
    "res_conv_width": 1024,
    "res_convolution_multiplier": 1,
    "res_dilation_cycle": 8,
    "res_dilation_growth_rate": 3,
    "res_downs_t": [
      3,
      2,
      2
    ],
    "res_strides_t": [
      2,
      2,
      2
    ],
    "resid_dropout": 0,
    "sampling_rate": 44100,
    "spread": null,
    "timing_dims": 64,
    "zero_out": false
  },
  "sampling_rate": 44100,
  "timing_dims": 64,
  "torch_dtype": "float32",
  "transformers_version": null,
  "vqvae_config": {
    "act_fn": "relu",
    "commit": 0.02,
    "conv_input_shape": 1,
    "conv_res_scale": false,
    "embed_dim": 64,
    "hop_fraction": [
      0.125,
      0.5,
      0.5
    ],
    "init_scale": 0.2,
    "levels": 3,
    "lmu": 0.99,
    "model_type": "jukebox_vqvae",
    "multipliers": [
      2,
      1,
      1
    ],
    "nb_discrete_codes": 2048,
    "res_conv_depth": 4,
    "res_conv_width": 32,
    "res_convolution_multiplier": 1,
    "res_dilation_cycle": null,
    "res_dilation_growth_rate": 3,
    "res_downs_t": [
      3,
      2,
      2
    ],
    "res_strides_t": [
      2,
      2,
      2
    ],
    "sample_length": 1058304,
    "transformers_version": "4.25.0.dev0",
    "zero_out": false
  }
}