Hugging Face's logo

AEmotionStudio
/
vibevoice-models

Text-to-Speech
Transformers
Safetensors
English
Chinese
speech-recognition
tts
asr
voice-cloning
long-form
multi-speaker
streaming
mirror
Model card Files
xet
 Community
AEmotionStudio commited on
Commit
7151d9b
·
verified ·
1 Parent(s): 72d8020

add: tts-large-fp8/config.json (FP8 LM-backbone shard, pre-quantized from aoi-ot/VibeVoice-Large)

Browse files
Files changed (1) hide show
  1. tts-large-fp8/config.json +177 -0
tts-large-fp8/config.json ADDED
@@ -0,0 +1,177 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
+ "VibeVoiceForConditionalGenerationInference"
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
+ "quantization_config": {
119
+ "include_input_output_embeddings": false,
120
+ "modules_to_not_convert": [
121
+ "diffusion_head",
122
+ "acoustic_tokenizer",
123
+ "semantic_tokenizer",
124
+ "acoustic_connector",
125
+ "semantic_connector"
126
+ ],
127
+ "quant_method": "torchao",
128
+ "quant_type": {
129
+ "default": {
130
+ "_data": {
131
+ "set_inductor_config": true,
132
+ "weight_dtype": {
133
+ "_data": "float8_e4m3fn",
134
+ "_type": "torch.dtype"
135
+ }
136
+ },
137
+ "_type": "Float8WeightOnlyConfig",
138
+ "_version": 2
139
+ }
140
+ },
141
+ "quant_type_kwargs": {},
142
+ "untie_embedding_weights": false
143
+ },
144
+ "semantic_tokenizer_config": {
145
+ "causal": true,
146
+ "channels": 1,
147
+ "conv_bias": true,
148
+ "conv_norm": "none",
149
+ "corpus_normalize": 0.0,
150
+ "disable_last_norm": true,
151
+ "dtype": "bfloat16",
152
+ "encoder_depths": "3-3-3-3-3-3-8",
153
+ "encoder_n_filters": 32,
154
+ "encoder_ratios": [
155
+ 8,
156
+ 5,
157
+ 5,
158
+ 4,
159
+ 2,
160
+ 2
161
+ ],
162
+ "fix_std": 0,
163
+ "layer_scale_init_value": 1e-06,
164
+ "layernorm": "RMSNorm",
165
+ "layernorm_elementwise_affine": true,
166
+ "layernorm_eps": 1e-05,
167
+ "mixer_layer": "depthwise_conv",
168
+ "model_type": "vibevoice_semantic_tokenizer",
169
+ "pad_mode": "constant",
170
+ "std_dist_type": "none",
171
+ "vae_dim": 128,
172
+ "weight_init_value": 0.01
173
+ },
174
+ "semantic_vae_dim": 128,
175
+ "tie_word_embeddings": false,
176
+ "transformers_version": "4.57.6"
177
+ }