{ "latent_size": 80, "model_size": 2048, "num_layers": 24, "num_heads": 16, "intermediate_size": 5888, "norm_eps": 1e-5, "max_seq_len": 640, "text_vocab_size": 256, "text_model_size": 1280, "text_num_layers": 14, "text_num_heads": 10, "text_intermediate_size": 3328, "text_max_seq_len": 768, "speaker_patch_size": 4, "speaker_model_size": 1280, "speaker_num_layers": 14, "speaker_num_heads": 10, "speaker_intermediate_size": 3328, "speaker_max_patched_seq_len": 640, "timestep_embed_size": 512, "adaln_rank": 256 }