config.json

{
    "patch_size": [
        1,
        2,
        2
    ],
    "num_attention_heads": 12,
    "attention_head_dim": 128,
    "in_channels": 16,
    "out_channels": 16,
    "text_dim": 4096,
    "freq_dim": 256,
    "ffn_dim": 8960,
    "num_layers": 30,
    "cross_attn_norm": true,
    "qk_norm": "rms_norm_across_heads",
    "eps": 1e-06,
    "image_dim": null,
    "mel_dim": null,
    "added_kv_proj_dim": 1536,
    "context_pre_only": true,
    "rope_max_seq_len": 1024,
    "rope_max_seq_len_text": 2048,
    "mel_num_attention_heads": 20,
    "mel_in_channels": 8,
    "mlp_ratio": 2.5,
    "mel_out_channels": 8,
    "max_position": 32768,
    "rope_theta": 1000000.0,
    "speaker_embedding_dim": 512,
    "text_embedding_dim": 768,
    "ssl_encoder_depths": [
        12,
        12
    ],
    "ssl_names": [
        "mert",
        "m-hubert"
    ],
    "ssl_latent_dims": [
        1024,
        768
    ],
    "lyric_encoder_vocab_size": 6693,
    "lyric_hidden_size": 1024,
    "mel_patch_size": [
        16,
        1
    ],
    "max_height": 16,
    "max_width": 32768,
    "max_speech_token_num": 128,
    "_use_default_values": [
        "mel_dim"
    ],
    "_class_name": "WanTransformer3DModel",
    "_diffusers_version": "0.33.0.dev0"
}