update/config.json
| 1 | { |
| 2 | "_local_rope_theta_note": "local_rope_theta is DEPRECATED/unused: acoustic decoder now uses a learned slot-position embedding, not RoPE.", |
| 3 | "_reserved_tokens_note": "ids 13..42 are reserved/preserve slots (random-init), inserted right after emotion_4 (12); unk and phoneme vocab shifted +30.", |
| 4 | "architectures": [ |
| 5 | "VieNeuV3TurboForTTS" |
| 6 | ], |
| 7 | "attention_dropout": 0.0, |
| 8 | "audio_pad_token_id": 1024, |
| 9 | "audio_ref_slot_token_id": 7, |
| 10 | "audio_sample_rate": 48000, |
| 11 | "audio_tokenizer_pretrained_name_or_path": "OpenMOSS-Team/MOSS-Audio-Tokenizer-Nano", |
| 12 | "audio_vocab_size": 1024, |
| 13 | "bos_token_id": 1, |
| 14 | "default_style_token_id": 16, |
| 15 | "dtype": "float32", |
| 16 | "emotion_0_token_id": 8, |
| 17 | "emotion_1_token_id": 9, |
| 18 | "emotion_2_token_id": 10, |
| 19 | "emotion_3_token_id": 11, |
| 20 | "emotion_4_token_id": 12, |
| 21 | "emotion_5_token_id": 13, |
| 22 | "emotion_6_token_id": 14, |
| 23 | "emotion_7_token_id": 15, |
| 24 | "eos_token_id": 2, |
| 25 | "head_dim": 64, |
| 26 | "hidden_size": 768, |
| 27 | "intermediate_size": 3072, |
| 28 | "local_intermediate_size": 2048, |
| 29 | "local_num_attention_heads": 8, |
| 30 | "local_num_hidden_layers": 1, |
| 31 | "local_rope_theta": 10000.0, |
| 32 | "max_position_embeddings": 1024, |
| 33 | "model_type": "vieneu_v3", |
| 34 | "n_vq": 16, |
| 35 | "num_attention_heads": 12, |
| 36 | "num_hidden_layers": 12, |
| 37 | "num_key_value_heads": 4, |
| 38 | "num_reserved_tokens": 30, |
| 39 | "pad_token_id": 0, |
| 40 | "ref_drop_rate": 0.45, |
| 41 | "reserved_token_start": 13, |
| 42 | "rms_norm_eps": 1e-06, |
| 43 | "rope_theta": 10000.0, |
| 44 | "speaker_embedding_dim": 192, |
| 45 | "speaker_encoder_filename": "speaker_encoder.onnx", |
| 46 | "speech_generation_end_token_id": 6, |
| 47 | "speech_generation_start_token_id": 5, |
| 48 | "style_0_token_id": 16, |
| 49 | "style_1_token_id": 17, |
| 50 | "style_2_token_id": 18, |
| 51 | "style_3_token_id": 19, |
| 52 | "style_4_token_id": 20, |
| 53 | "style_5_token_id": 21, |
| 54 | "style_6_token_id": 22, |
| 55 | "style_7_token_id": 23, |
| 56 | "style_8_token_id": 24, |
| 57 | "style_9_token_id": 25, |
| 58 | "style_labels": { |
| 59 | "doc_truyen": 18, |
| 60 | "tin_tuc": 17, |
| 61 | "tu_nhien": 16 |
| 62 | }, |
| 63 | "text_prompt_end_token_id": 4, |
| 64 | "text_prompt_start_token_id": 3, |
| 65 | "text_vocab_size": 419, |
| 66 | "tie_word_embeddings": false, |
| 67 | "transformers_version": "4.56.0", |
| 68 | "unk_token_id": 43, |
| 69 | "use_speaker_embedding": true |
| 70 | } |
| 71 | |