Hugging Face's logo

heiertech
/
Prat-9B-NF4

Text-to-Speech
Safetensors
VibeVoice
Norwegian
Norwegian Bokmål
tts
speech-synthesis
norwegian
bokmal
bitsandbytes
4bit
quantized
4-bit precision
Model card Files
xet
Community
marksverdhai commited on
Commit
650660e
·
verified ·
1 Parent(s): bc3818e

Upload folder using huggingface_hub

Browse files
Files changed (6) hide show
  1. config.json +135 -0
  2. generation_config.json +4 -0
  3. model-00001-of-00002.safetensors +3 -0
  4. model-00002-of-00002.safetensors +3 -0
  5. model.safetensors.index.json +0 -0
  6. preprocessor_config.json +12 -0
config.json ADDED
@@ -0,0 +1,135 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "acostic_vae_dim": 64,
3
+ "acoustic_tokenizer_config": {
4
+ "causal": true,
5
+ "channels": 1,
6
+ "conv_bias": true,
7
+ "conv_norm": "none",
8
+ "corpus_normalize": 0.0,
9
+ "decoder_depths": null,
10
+ "decoder_n_filters": 32,
11
+ "decoder_ratios": [
12
+ 8,
13
+ 5,
14
+ 5,
15
+ 4,
16
+ 2,
17
+ 2
18
+ ],
19
+ "disable_last_norm": true,
20
+ "encoder_depths": "3-3-3-3-3-3-8",
21
+ "encoder_n_filters": 32,
22
+ "encoder_ratios": [
23
+ 8,
24
+ 5,
25
+ 5,
26
+ 4,
27
+ 2,
28
+ 2
29
+ ],
30
+ "fix_std": 0.5,
31
+ "layer_scale_init_value": 1e-06,
32
+ "layernorm": "RMSNorm",
33
+ "layernorm_elementwise_affine": true,
34
+ "layernorm_eps": 1e-05,
35
+ "mixer_layer": "depthwise_conv",
36
+ "model_type": "vibevoice_acoustic_tokenizer",
37
+ "pad_mode": "constant",
38
+ "std_dist_type": "gaussian",
39
+ "torch_dtype": "bfloat16",
40
+ "vae_dim": 64,
41
+ "weight_init_value": 0.01
42
+ },
43
+ "acoustic_vae_dim": 64,
44
+ "architectures": [
45
+ "VibeVoiceForConditionalGenerationInference"
46
+ ],
47
+ "decoder_config": {
48
+ "attention_dropout": 0.0,
49
+ "hidden_act": "silu",
50
+ "hidden_size": 3584,
51
+ "initializer_range": 0.02,
52
+ "intermediate_size": 18944,
53
+ "max_position_embeddings": 32768,
54
+ "max_window_layers": 28,
55
+ "model_type": "qwen2",
56
+ "num_attention_heads": 28,
57
+ "num_hidden_layers": 28,
58
+ "num_key_value_heads": 4,
59
+ "rms_norm_eps": 1e-06,
60
+ "rope_scaling": null,
61
+ "rope_theta": 1000000.0,
62
+ "sliding_window": null,
63
+ "torch_dtype": "bfloat16",
64
+ "use_cache": true,
65
+ "use_mrope": false,
66
+ "use_sliding_window": false,
67
+ "vocab_size": 152064
68
+ },
69
+ "diffusion_head_config": {
70
+ "ddpm_batch_mul": 4,
71
+ "ddpm_beta_schedule": "cosine",
72
+ "ddpm_num_inference_steps": 20,
73
+ "ddpm_num_steps": 1000,
74
+ "diffusion_type": "ddpm",
75
+ "head_ffn_ratio": 3.0,
76
+ "head_layers": 4,
77
+ "hidden_size": 3584,
78
+ "latent_size": 64,
79
+ "model_type": "vibevoice_diffusion_head",
80
+ "prediction_type": "v_prediction",
81
+ "rms_norm_eps": 1e-05,
82
+ "speech_vae_dim": 64,
83
+ "torch_dtype": "bfloat16"
84
+ },
85
+ "model_type": "vibevoice",
86
+ "quantization_config": {
87
+ "_load_in_4bit": true,
88
+ "_load_in_8bit": false,
89
+ "bnb_4bit_compute_dtype": "bfloat16",
90
+ "bnb_4bit_quant_storage": "uint8",
91
+ "bnb_4bit_quant_type": "nf4",
92
+ "bnb_4bit_use_double_quant": true,
93
+ "llm_int8_enable_fp32_cpu_offload": false,
94
+ "llm_int8_has_fp16_weight": false,
95
+ "llm_int8_skip_modules": null,
96
+ "llm_int8_threshold": 6.0,
97
+ "load_in_4bit": true,
98
+ "load_in_8bit": false,
99
+ "quant_method": "bitsandbytes"
100
+ },
101
+ "semantic_tokenizer_config": {
102
+ "causal": true,
103
+ "channels": 1,
104
+ "conv_bias": true,
105
+ "conv_norm": "none",
106
+ "corpus_normalize": 0.0,
107
+ "disable_last_norm": true,
108
+ "encoder_depths": "3-3-3-3-3-3-8",
109
+ "encoder_n_filters": 32,
110
+ "encoder_ratios": [
111
+ 8,
112
+ 5,
113
+ 5,
114
+ 4,
115
+ 2,
116
+ 2
117
+ ],
118
+ "fix_std": 0,
119
+ "layer_scale_init_value": 1e-06,
120
+ "layernorm": "RMSNorm",
121
+ "layernorm_elementwise_affine": true,
122
+ "layernorm_eps": 1e-05,
123
+ "mixer_layer": "depthwise_conv",
124
+ "model_type": "vibevoice_semantic_tokenizer",
125
+ "pad_mode": "constant",
126
+ "std_dist_type": "none",
127
+ "torch_dtype": "bfloat16",
128
+ "vae_dim": 128,
129
+ "weight_init_value": 0.01
130
+ },
131
+ "semantic_vae_dim": 128,
132
+ "tie_word_embeddings": false,
133
+ "torch_dtype": "bfloat16",
134
+ "transformers_version": "4.51.3"
135
+ }
generation_config.json ADDED
@@ -0,0 +1,4 @@
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "transformers_version": "4.51.3"
4
+ }
model-00001-of-00002.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bb5b251fc215bcd5ee9d74bc4383aece74ab8a4e38d402017bcf349c20cc02a6
3
+ size 4949701350
model-00002-of-00002.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:40b3e0fe570de93c87842f6bca65a4c4012e41920a6a5342b967206db781b603
3
+ size 1676476837
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
+ }