{ "model": { "latent_dim": 32, "latent_patch_size": 1, "model_dim": 1280, "num_layers": 12, "num_heads": 20, "mlp_ratio": 2.875, "text_mlp_ratio": 2.6, "speaker_mlp_ratio": 2.6, "dropout": 0.0, "text_vocab_size": 99574, "text_tokenizer_repo": "llm-jp/llm-jp-3-150m", "text_add_bos": true, "text_dim": 512, "text_layers": 10, "text_heads": 8, "use_caption_condition": true, "use_speaker_condition": true, "caption_vocab_size": 99574, "caption_tokenizer_repo": "llm-jp/llm-jp-3-150m", "caption_add_bos": true, "caption_dim": 512, "caption_layers": 10, "caption_heads": 8, "caption_mlp_ratio": 2.6, "speaker_dim": 768, "speaker_layers": 8, "speaker_heads": 12, "speaker_patch_size": 1, "timestep_embed_dim": 512, "adaln_rank": 192, "norm_eps": 1e-05, "use_duration_predictor": true, "duration_aux_dim": 14, "duration_hidden_dim": 1024, "duration_layers": 3, "duration_dropout": 0.1, "duration_attention_heads": 8, "duration_architecture": "token_sum_dual_adarn_zero_no_aux", "duration_token_init_frames": 9.0, "duration_speaker_fusion": "adarn_zero", "duration_caption_fusion": "adarn_zero", "duration_caption_pooling": "masked_mean" }, "inference": { "max_text_len": 256, "max_caption_len": 512 }, "export": { "seq_len": 128, "text_len": 256, "caption_len": 512, "ref_len": 200, "sample_rate": 48000, "hop_length": 1920, "latent_dim": 32, "num_layers": 12, "has_caption": true, "has_speaker_kv_scale": true, "lora_merged": false, "dacvae_snake_coreml_safe": true, "context_kv_cache": true } }