Upload merged Norwegian fine-tuned VibeVoice-7B model
Browse files- README.md +62 -0
- config.json +151 -0
- model-00001-of-00004.safetensors +3 -0
- model-00002-of-00004.safetensors +3 -0
- model-00003-of-00004.safetensors +3 -0
- model-00004-of-00004.safetensors +3 -0
- model.safetensors.index.json +0 -0
- preprocessor_config.json +12 -0
README.md
ADDED
|
@@ -0,0 +1,62 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: other
|
| 3 |
+
license_name: vibevoice-community
|
| 4 |
+
license_link: https://huggingface.co/vibevoice/VibeVoice-7B/blob/main/LICENSE
|
| 5 |
+
base_model: vibevoice/VibeVoice-7B
|
| 6 |
+
tags:
|
| 7 |
+
- tts
|
| 8 |
+
- text-to-speech
|
| 9 |
+
- speech-synthesis
|
| 10 |
+
- norwegian
|
| 11 |
+
- bokmal
|
| 12 |
+
- nynorsk
|
| 13 |
+
language:
|
| 14 |
+
- "no"
|
| 15 |
+
- nb
|
| 16 |
+
- nn
|
| 17 |
+
---
|
| 18 |
+
|
| 19 |
+
# VibeVoice 7B Norwegian
|
| 20 |
+
|
| 21 |
+
This is a Norwegian fine-tuned version of [VibeVoice-7B](https://huggingface.co/vibevoice/VibeVoice-7B), a state-of-the-art text-to-speech model.
|
| 22 |
+
|
| 23 |
+
## Training Details
|
| 24 |
+
|
| 25 |
+
This model was trained using a progressive 3-stage fine-tuning approach:
|
| 26 |
+
|
| 27 |
+
1. **Stage 1**: Initial Norwegian (Bokmal) training on Mozilla Common Voice
|
| 28 |
+
2. **Stage 2**: Continued training on broader Norwegian data
|
| 29 |
+
3. **Stage 3**: Dialect-specific fine-tuning for Ostnorsk/Oslo dialect
|
| 30 |
+
|
| 31 |
+
The LoRA adapter was merged into the base model weights to create this standalone fine-tuned model.
|
| 32 |
+
|
| 33 |
+
### Training Configuration
|
| 34 |
+
- LoRA rank: 32
|
| 35 |
+
- LoRA alpha: 128
|
| 36 |
+
- Target modules: q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj
|
| 37 |
+
- Diffusion head: Fully trained
|
| 38 |
+
- Precision: bfloat16
|
| 39 |
+
|
| 40 |
+
## Usage
|
| 41 |
+
|
| 42 |
+
```python
|
| 43 |
+
from transformers import AutoProcessor, AutoModel
|
| 44 |
+
import torch
|
| 45 |
+
|
| 46 |
+
processor = AutoProcessor.from_pretrained("heiertech/vibevoice-7b-norwegian")
|
| 47 |
+
model = AutoModel.from_pretrained("heiertech/vibevoice-7b-norwegian", torch_dtype=torch.bfloat16)
|
| 48 |
+
|
| 49 |
+
# Generate speech
|
| 50 |
+
text = "Hei, dette er en test av den norske stemmen."
|
| 51 |
+
inputs = processor(text=text, return_tensors="pt")
|
| 52 |
+
outputs = model.generate(**inputs)
|
| 53 |
+
```
|
| 54 |
+
|
| 55 |
+
## License
|
| 56 |
+
|
| 57 |
+
This model inherits the license from the base VibeVoice-7B model. Please see the [original license](https://huggingface.co/vibevoice/VibeVoice-7B/blob/main/LICENSE) for details.
|
| 58 |
+
|
| 59 |
+
## Acknowledgments
|
| 60 |
+
|
| 61 |
+
- Base model: [vibevoice/VibeVoice-7B](https://huggingface.co/vibevoice/VibeVoice-7B)
|
| 62 |
+
- Training data: Mozilla Common Voice Norwegian
|
config.json
ADDED
|
@@ -0,0 +1,151 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"_attn_implementation_autoset": false,
|
| 3 |
+
"acostic_vae_dim": 64,
|
| 4 |
+
"acoustic_tokenizer_config": {
|
| 5 |
+
"causal": true,
|
| 6 |
+
"channels": 1,
|
| 7 |
+
"conv_bias": true,
|
| 8 |
+
"conv_norm": "none",
|
| 9 |
+
"corpus_normalize": 0.0,
|
| 10 |
+
"decoder_depths": null,
|
| 11 |
+
"decoder_n_filters": 32,
|
| 12 |
+
"decoder_ratios": [
|
| 13 |
+
8,
|
| 14 |
+
5,
|
| 15 |
+
5,
|
| 16 |
+
4,
|
| 17 |
+
2,
|
| 18 |
+
2
|
| 19 |
+
],
|
| 20 |
+
"disable_last_norm": true,
|
| 21 |
+
"dtype": "bfloat16",
|
| 22 |
+
"encoder_depths": "3-3-3-3-3-3-8",
|
| 23 |
+
"encoder_n_filters": 32,
|
| 24 |
+
"encoder_ratios": [
|
| 25 |
+
8,
|
| 26 |
+
5,
|
| 27 |
+
5,
|
| 28 |
+
4,
|
| 29 |
+
2,
|
| 30 |
+
2
|
| 31 |
+
],
|
| 32 |
+
"fix_std": 0.5,
|
| 33 |
+
"layer_scale_init_value": 1e-06,
|
| 34 |
+
"layernorm": "RMSNorm",
|
| 35 |
+
"layernorm_elementwise_affine": true,
|
| 36 |
+
"layernorm_eps": 1e-05,
|
| 37 |
+
"mixer_layer": "depthwise_conv",
|
| 38 |
+
"model_type": "vibevoice_acoustic_tokenizer",
|
| 39 |
+
"pad_mode": "constant",
|
| 40 |
+
"std_dist_type": "gaussian",
|
| 41 |
+
"vae_dim": 64,
|
| 42 |
+
"weight_init_value": 0.01
|
| 43 |
+
},
|
| 44 |
+
"acoustic_vae_dim": 64,
|
| 45 |
+
"architectures": [
|
| 46 |
+
"VibeVoiceForConditionalGeneration"
|
| 47 |
+
],
|
| 48 |
+
"decoder_config": {
|
| 49 |
+
"attention_dropout": 0.0,
|
| 50 |
+
"dtype": "bfloat16",
|
| 51 |
+
"hidden_act": "silu",
|
| 52 |
+
"hidden_size": 3584,
|
| 53 |
+
"initializer_range": 0.02,
|
| 54 |
+
"intermediate_size": 18944,
|
| 55 |
+
"layer_types": [
|
| 56 |
+
"full_attention",
|
| 57 |
+
"full_attention",
|
| 58 |
+
"full_attention",
|
| 59 |
+
"full_attention",
|
| 60 |
+
"full_attention",
|
| 61 |
+
"full_attention",
|
| 62 |
+
"full_attention",
|
| 63 |
+
"full_attention",
|
| 64 |
+
"full_attention",
|
| 65 |
+
"full_attention",
|
| 66 |
+
"full_attention",
|
| 67 |
+
"full_attention",
|
| 68 |
+
"full_attention",
|
| 69 |
+
"full_attention",
|
| 70 |
+
"full_attention",
|
| 71 |
+
"full_attention",
|
| 72 |
+
"full_attention",
|
| 73 |
+
"full_attention",
|
| 74 |
+
"full_attention",
|
| 75 |
+
"full_attention",
|
| 76 |
+
"full_attention",
|
| 77 |
+
"full_attention",
|
| 78 |
+
"full_attention",
|
| 79 |
+
"full_attention",
|
| 80 |
+
"full_attention",
|
| 81 |
+
"full_attention",
|
| 82 |
+
"full_attention",
|
| 83 |
+
"full_attention"
|
| 84 |
+
],
|
| 85 |
+
"max_position_embeddings": 32768,
|
| 86 |
+
"max_window_layers": 28,
|
| 87 |
+
"model_type": "qwen2",
|
| 88 |
+
"num_attention_heads": 28,
|
| 89 |
+
"num_hidden_layers": 28,
|
| 90 |
+
"num_key_value_heads": 4,
|
| 91 |
+
"rms_norm_eps": 1e-06,
|
| 92 |
+
"rope_scaling": null,
|
| 93 |
+
"rope_theta": 1000000.0,
|
| 94 |
+
"sliding_window": null,
|
| 95 |
+
"use_cache": true,
|
| 96 |
+
"use_mrope": false,
|
| 97 |
+
"use_sliding_window": false,
|
| 98 |
+
"vocab_size": 152064
|
| 99 |
+
},
|
| 100 |
+
"diffusion_head_config": {
|
| 101 |
+
"ddpm_batch_mul": 4,
|
| 102 |
+
"ddpm_beta_schedule": "cosine",
|
| 103 |
+
"ddpm_num_inference_steps": 20,
|
| 104 |
+
"ddpm_num_steps": 1000,
|
| 105 |
+
"diffusion_type": "ddpm",
|
| 106 |
+
"dtype": "bfloat16",
|
| 107 |
+
"head_ffn_ratio": 3.0,
|
| 108 |
+
"head_layers": 4,
|
| 109 |
+
"hidden_size": 3584,
|
| 110 |
+
"latent_size": 64,
|
| 111 |
+
"model_type": "vibevoice_diffusion_head",
|
| 112 |
+
"prediction_type": "v_prediction",
|
| 113 |
+
"rms_norm_eps": 1e-05,
|
| 114 |
+
"speech_vae_dim": 64
|
| 115 |
+
},
|
| 116 |
+
"dtype": "bfloat16",
|
| 117 |
+
"model_type": "vibevoice",
|
| 118 |
+
"semantic_tokenizer_config": {
|
| 119 |
+
"causal": true,
|
| 120 |
+
"channels": 1,
|
| 121 |
+
"conv_bias": true,
|
| 122 |
+
"conv_norm": "none",
|
| 123 |
+
"corpus_normalize": 0.0,
|
| 124 |
+
"disable_last_norm": true,
|
| 125 |
+
"dtype": "bfloat16",
|
| 126 |
+
"encoder_depths": "3-3-3-3-3-3-8",
|
| 127 |
+
"encoder_n_filters": 32,
|
| 128 |
+
"encoder_ratios": [
|
| 129 |
+
8,
|
| 130 |
+
5,
|
| 131 |
+
5,
|
| 132 |
+
4,
|
| 133 |
+
2,
|
| 134 |
+
2
|
| 135 |
+
],
|
| 136 |
+
"fix_std": 0,
|
| 137 |
+
"layer_scale_init_value": 1e-06,
|
| 138 |
+
"layernorm": "RMSNorm",
|
| 139 |
+
"layernorm_elementwise_affine": true,
|
| 140 |
+
"layernorm_eps": 1e-05,
|
| 141 |
+
"mixer_layer": "depthwise_conv",
|
| 142 |
+
"model_type": "vibevoice_semantic_tokenizer",
|
| 143 |
+
"pad_mode": "constant",
|
| 144 |
+
"std_dist_type": "none",
|
| 145 |
+
"vae_dim": 128,
|
| 146 |
+
"weight_init_value": 0.01
|
| 147 |
+
},
|
| 148 |
+
"semantic_vae_dim": 128,
|
| 149 |
+
"tie_word_embeddings": false,
|
| 150 |
+
"transformers_version": "4.57.3"
|
| 151 |
+
}
|
model-00001-of-00004.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:eb423717f01ba6aa890e122cf7f662f71a8a4ea0431f218aa62a1f7174b2b6fd
|
| 3 |
+
size 4877662532
|
model-00002-of-00004.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:71d015dcc94e9d00cc4b4bafc57b567768bd723acab834e395364c21533ffbff
|
| 3 |
+
size 4932752840
|
model-00003-of-00004.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:f8489291ef93f10bcbd2042df9bba572e1d60972b3c99655c99a796c5ff3753b
|
| 3 |
+
size 4982901128
|
model-00004-of-00004.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:12575fd8c03051efa93e5f14740b5bf80e43de2f4d7fce4b21ed3f24f9d35ea3
|
| 3 |
+
size 3893553730
|
model.safetensors.index.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
preprocessor_config.json
ADDED
|
@@ -0,0 +1,12 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"processor_class": "VibeVoiceProcessor",
|
| 3 |
+
"speech_tok_compress_ratio": 3200,
|
| 4 |
+
"db_normalize": true,
|
| 5 |
+
"audio_processor": {
|
| 6 |
+
"feature_extractor_type": "VibeVoiceTokenizerProcessor",
|
| 7 |
+
"sampling_rate": 24000,
|
| 8 |
+
"normalize_audio": true,
|
| 9 |
+
"target_dB_FS": -25,
|
| 10 |
+
"eps": 1e-06
|
| 11 |
+
}
|
| 12 |
+
}
|