config.json

{
  "activation_function": "gelu_new",
  "alignment_head": [
    2,
    null,
    null
  ],
  "alignment_layer": [
    68,
    null,
    null
  ],
  "architectures": [
    "JukeboxModel"
  ],
  "attn_dropout": 0.0,
  "attn_init_scale": 1.0,
  "attn_order": [
    12,
    2,
    2
  ],
  "blocks": 16,
  "bos_token_id": 50256,
  "c_res": 1,
  "cond_c_res": [
    0,
    1,
    1
  ],
  "cond_depth": [
    3,
    16,
    16
  ],
  "cond_dilation_cycle": [
    null,
    8,
    8
  ],
  "cond_dilation_growth_rate": [
    1,
    3,
    3
  ],
  "cond_m_conv": 1,
  "cond_res_scale": false,
  "cond_width": [
    128,
    128,
    64
  ],
  "cond_zero_out": false,
  "copy_input": false,
  "depth": [
    2,
    2,
    2
  ],
  "downs_t": [
    3,
    2,
    2
  ],
  "emb_dropout": 0.1,
  "eos_token_id": 50256,
  "fp16_params": true,
  "hop_length": 256,
  "init_scale": [
    0.7,
    1,
    1
  ],
  "initializer_range": 0.02,
  "l_bins": 128,
  "labels": true,
  "layer_norm_epsilon": 1e-05,
  "m_attn": 0.25,
  "max_bow_genre_size": 1,
  "max_duration": 600.0,
  "merged_decoder": [
    true,
    false,
    false
  ],
  "min_duration": 1,
  "mlp_init_scale": 0.02,
  "model_type": "jukebox",
  "multispec_loss_hop_length": [
    240,
    120,
    50
  ],
  "multispec_loss_n_fft": [
    2048,
    1024,
    512
  ],
  "multispec_loss_window_size": [
    1200,
    600,
    240
  ],
  "multispectral": 1.0,
  "n_ctx": [256,256,256],
  "n_embd": 768,
  "n_head": 12,
  "n_heads": [
    2,
    1,
    1
  ],
  "n_inner": null,
  "n_layer": 12,
  "n_positions": 1024,
  "n_tokens": [
    512,
    0,
    0
  ],
  "n_vocab": 79,
  "name": "AudioSamples",
  "nb_priors": 3,
  "pos_init": false,
  "prime_attn_dropout": 0.0,
  "prime_attn_order": [
    2,
    0,
    0
  ],
  "prime_blocks": 32,
  "prime_c_res": 1,
  "prime_cond_c_res": [
    0,
    1,
    1
  ],
  "prime_depth": [
    18,
    3,
    3
  ],
  "prime_emb_dropout": 0.0,
  "prime_heads": 4,
  "prime_init_scale": [
    0.1,
    0.4,
    0.4
  ],
  "prime_loss_fraction": [
    0.4,
    0.0,
    0.0
  ],
  "prime_m_attn": 0.25,
  "prime_m_mlp": 1.0,
  "prime_pos_init": false,
  "prime_res_scale": false,
  "prime_resid_dropout": 0.0,
  "prime_spread": null,
  "prime_width": [
    128,
    128,
    128
  ],
  "prime_zero_out": false,
  "priors_width": [
    128,
    64,
    32
  ],
  "reorder_and_upcast_attn": false,
  "res_scale": false,
  "resid_dropout": 0.0,
  "sample_hop_length": 30000,
  "sample_length": 44032,
  "sample_length_in_seconds": 1,
  "scale_attn_by_inverse_layer_idx": false,
  "scale_attn_weights": true,
  "single_enc_dec": [
    true,
    false,
    false
  ],
  "spectral": 0.0,
  "spread": null,
  "sr": 44100,
  "strides_t": [
    2,
    2,
    2
  ],
  "summary_activation": null,
  "summary_first_dropout": 0.1,
  "summary_proj_to_labels": true,
  "summary_type": "cls_index",
  "summary_use_proj": true,
  "t_bins": 64,
  "torch_dtype": "float32",
  "transformers_version": "4.19.0.dev0",
  "use_cache": true,
  "use_nonrelative_specloss": true,
  "use_tokens": [
    true,
    false,
    false
  ],
  "vocab_size": 50257,
  "vq_vae_codebook_dimension": 128,
  "vq_vae_commit": 0.02,
  "vq_vae_conv_block_depth": 4,
  "vq_vae_conv_block_width": 64,
  "vq_vae_depth": 4,
  "vq_vae_dilation_cycle": null,
  "vq_vae_dilation_growth_rate": 3,
  "vq_vae_downs_t": [
    3,
    2,
    2
  ],
  "vq_vae_emmbedding_width": 128,
  "vq_vae_levels": 3,
  "vq_vae_lmu": 0.99,
  "vq_vae_m_conv": 1,
  "vq_vae_multipliers": [
    2,
    1,
    1
  ],
  "vq_vae_reverse_decoder_dilation": 1,
  "vq_vae_strides_t": [
    2,
    2,
    2
  ],
  "vq_vae_width": 64,
  "vqvae_z_shapes": [
    [
      344
    ],
    [
      1376
    ],
    [
      5504
    ]
  ],
  "width": [
    128,
    64,
    32
  ],
  "y_bins": [
    [120,4111],[120,4111],[120,4111]
  ],
  "zero_out": false
}