| { | |
| "return_dict": true, | |
| "output_hidden_states": false, | |
| "torchscript": false, | |
| "dtype": "float32", | |
| "tie_word_embeddings": true, | |
| "is_encoder_decoder": false, | |
| "is_decoder": false, | |
| "add_cross_attention": false, | |
| "architectures": [ | |
| "YuE2VAE" | |
| ], | |
| "bos_token_id": null, | |
| "pad_token_id": null, | |
| "eos_token_id": null, | |
| "decoder_start_token_id": null, | |
| "transformers_version": "4.57.6", | |
| "auto_map": { | |
| "AutoConfig": "modeling_vae.YuE2VAEConfig", | |
| "AutoModel": "modeling_vae.YuE2VAE" | |
| }, | |
| "model_type": "yue2_vae", | |
| "encoder_config": { | |
| "c_mults": [ | |
| 1, | |
| 2, | |
| 4, | |
| 8, | |
| 16, | |
| 32 | |
| ], | |
| "channels": 64, | |
| "in_channels": 2, | |
| "latent_dim": 128, | |
| "strides": [ | |
| 2, | |
| 2, | |
| 4, | |
| 4, | |
| 5, | |
| 6 | |
| ], | |
| "use_snake": true | |
| }, | |
| "decoder_config": { | |
| "c_mults": [ | |
| 1, | |
| 2, | |
| 4, | |
| 8, | |
| 16, | |
| 32 | |
| ], | |
| "channels": 64, | |
| "final_tanh": false, | |
| "latent_dim": 64, | |
| "out_channels": 2, | |
| "snake_type": "vanilla", | |
| "strides": [ | |
| 2, | |
| 2, | |
| 4, | |
| 4, | |
| 5, | |
| 6 | |
| ], | |
| "use_filter": false, | |
| "use_snake": true | |
| }, | |
| "sample_rate": 48000, | |
| "latent_dim": 64, | |
| "downsampling_ratio": 1920, | |
| "audio_channels": 2, | |
| "release_variant": "standard", | |
| "decode_core_frames": 1024, | |
| "decode_halo_frames": 16, | |
| "output_attentions": false | |
| } | |