config.json

{
    "model": {
        "extractor_mode": "layer_norm",
        "extractor_conv_bias": false,
        "extractor_conv_layer_config": [
            [
                512,
                10,
                5
            ],
            [
                512,
                3,
                2
            ],
            [
                512,
                3,
                2
            ],
            [
                512,
                3,
                2
            ],
            [
                512,
                3,
                2
            ],
            [
                512,
                2,
                2
            ],
            [
                512,
                2,
                2
            ]
        ],
        "encoder_embed_dim": 768,
        "encoder_projection_dropout": 0,
        "encoder_pos_conv_kernel": 95,
        "encoder_pos_conv_groups": 16,
        "encoder_pos_conv_depth": 5,
        "encoder_num_layers": 12,
        "encoder_num_heads": 12,
        "encoder_attention_dropout": 0.1,
        "encoder_ff_interm_features": 3072,
        "encoder_ff_interm_dropout": 0.0,
        "encoder_dropout": 0.1,
        "encoder_layer_norm_first": false,
        "encoder_layer_drop": 0.0,
        "encoder_qkv_bias": false,
        "codebook_size": 256,
        "codebook_decay": 0.9,
        "num_codebooks": 8,
        "ema_start_decay": 0.999,
        "ema_final_decay": 0.9999,
        "ema_final_step": 30000,
        "ema_exclude_layers": [
            "pos_conv_embed"
        ],
        "freeze_step": 200000,
        "ema_timescale": 20000,
        "ema_threshold": 1e-07
    }
}