Hugging Face's logo

mlx-community
/
VoxCPM2-bf16

Text-to-Speech
Safetensors
MLX
mlx-audio
voxcpm2
tts
voice-cloning
voice-design
Model card Files
xet
 Community
acul3 commited on
Commit
953e9c8
·
verified ·
1 Parent(s): fd40606

Upload config.json with huggingface_hub

Browse files
Files changed (1) hide show
  1. config.json +212 -0
config.json ADDED
@@ -0,0 +1,212 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architecture": "voxcpm2",
3
+ "lm_config": {
4
+ "bos_token_id": 1,
5
+ "eos_token_id": 2,
6
+ "hidden_size": 2048,
7
+ "intermediate_size": 6144,
8
+ "max_position_embeddings": 32768,
9
+ "num_attention_heads": 16,
10
+ "num_hidden_layers": 28,
11
+ "num_key_value_heads": 2,
12
+ "rms_norm_eps": 1e-05,
13
+ "rope_theta": 10000,
14
+ "kv_channels": 128,
15
+ "rope_scaling": {
16
+ "type": "longrope",
17
+ "long_factor": [
18
+ 0.9977997200264581,
19
+ 1.014658295992452,
20
+ 1.0349680404997148,
21
+ 1.059429246056193,
22
+ 1.0888815016813513,
23
+ 1.1243301355211495,
24
+ 1.166977103606075,
25
+ 1.2182568066927284,
26
+ 1.2798772354275727,
27
+ 1.3538666751582975,
28
+ 1.4426259039919596,
29
+ 1.5489853358570191,
30
+ 1.6762658237220625,
31
+ 1.8283407612492941,
32
+ 2.0096956085876183,
33
+ 2.225478927469756,
34
+ 2.481536379650452,
35
+ 2.784415934557119,
36
+ 3.1413289096347365,
37
+ 3.560047844772632,
38
+ 4.048719380066383,
39
+ 4.615569542115128,
40
+ 5.2684819496549835,
41
+ 6.014438591970396,
42
+ 6.858830049237097,
43
+ 7.804668263503327,
44
+ 8.851768731513417,
45
+ 9.99600492938444,
46
+ 11.228766118181639,
47
+ 12.536757560834843,
48
+ 13.902257701387796,
49
+ 15.303885189125953,
50
+ 16.717837610115794,
51
+ 18.119465097853947,
52
+ 19.484965238406907,
53
+ 20.792956681060105,
54
+ 22.02571786985731,
55
+ 23.16995406772833,
56
+ 24.217054535738416,
57
+ 25.16289275000465,
58
+ 26.007284207271347,
59
+ 26.753240849586767,
60
+ 27.40615325712662,
61
+ 27.973003419175363,
62
+ 28.461674954469114,
63
+ 28.880393889607006,
64
+ 29.237306864684626,
65
+ 29.540186419591297,
66
+ 29.79624387177199,
67
+ 30.01202719065413,
68
+ 30.193382037992453,
69
+ 30.34545697551969,
70
+ 30.47273746338473,
71
+ 30.579096895249787,
72
+ 30.66785612408345,
73
+ 30.741845563814174,
74
+ 30.80346599254902,
75
+ 30.85474569563567,
76
+ 30.897392663720595,
77
+ 30.932841297560394,
78
+ 30.962293553185553,
79
+ 30.986754758742034,
80
+ 31.007064503249293,
81
+ 31.02392307921529
82
+ ],
83
+ "short_factor": [
84
+ 0.9977997200264581,
85
+ 1.014658295992452,
86
+ 1.0349680404997148,
87
+ 1.059429246056193,
88
+ 1.0888815016813513,
89
+ 1.1243301355211495,
90
+ 1.166977103606075,
91
+ 1.2182568066927284,
92
+ 1.2798772354275727,
93
+ 1.3538666751582975,
94
+ 1.4426259039919596,
95
+ 1.5489853358570191,
96
+ 1.6762658237220625,
97
+ 1.8283407612492941,
98
+ 2.0096956085876183,
99
+ 2.225478927469756,
100
+ 2.481536379650452,
101
+ 2.784415934557119,
102
+ 3.1413289096347365,
103
+ 3.560047844772632,
104
+ 4.048719380066383,
105
+ 4.615569542115128,
106
+ 5.2684819496549835,
107
+ 6.014438591970396,
108
+ 6.858830049237097,
109
+ 7.804668263503327,
110
+ 8.851768731513417,
111
+ 9.99600492938444,
112
+ 11.228766118181639,
113
+ 12.536757560834843,
114
+ 13.902257701387796,
115
+ 15.303885189125953,
116
+ 16.717837610115794,
117
+ 18.119465097853947,
118
+ 19.484965238406907,
119
+ 20.792956681060105,
120
+ 22.02571786985731,
121
+ 23.16995406772833,
122
+ 24.217054535738416,
123
+ 25.16289275000465,
124
+ 26.007284207271347,
125
+ 26.753240849586767,
126
+ 27.40615325712662,
127
+ 27.973003419175363,
128
+ 28.461674954469114,
129
+ 28.880393889607006,
130
+ 29.237306864684626,
131
+ 29.540186419591297,
132
+ 29.79624387177199,
133
+ 30.01202719065413,
134
+ 30.193382037992453,
135
+ 30.34545697551969,
136
+ 30.47273746338473,
137
+ 30.579096895249787,
138
+ 30.66785612408345,
139
+ 30.741845563814174,
140
+ 30.80346599254902,
141
+ 30.85474569563567,
142
+ 30.897392663720595,
143
+ 30.932841297560394,
144
+ 30.962293553185553,
145
+ 30.986754758742034,
146
+ 31.007064503249293,
147
+ 31.02392307921529
148
+ ],
149
+ "original_max_position_embeddings": 32768
150
+ },
151
+ "vocab_size": 73448,
152
+ "use_mup": false,
153
+ "scale_emb": 12,
154
+ "dim_model_base": 256,
155
+ "scale_depth": 1.4
156
+ },
157
+ "patch_size": 4,
158
+ "feat_dim": 64,
159
+ "scalar_quantization_latent_dim": 512,
160
+ "scalar_quantization_scale": 9,
161
+ "residual_lm_num_layers": 8,
162
+ "residual_lm_no_rope": true,
163
+ "encoder_config": {
164
+ "hidden_dim": 1024,
165
+ "ffn_dim": 4096,
166
+ "num_heads": 16,
167
+ "num_layers": 12,
168
+ "kv_channels": 128
169
+ },
170
+ "dit_config": {
171
+ "hidden_dim": 1024,
172
+ "ffn_dim": 4096,
173
+ "num_heads": 16,
174
+ "num_layers": 12,
175
+ "kv_channels": 128,
176
+ "mean_mode": false,
177
+ "cfm_config": {
178
+ "sigma_min": 1e-06,
179
+ "solver": "euler",
180
+ "t_scheduler": "log-norm",
181
+ "inference_cfg_rate": 2.0
182
+ }
183
+ },
184
+ "audio_vae_config": {
185
+ "encoder_dim": 128,
186
+ "encoder_rates": [
187
+ 2,
188
+ 5,
189
+ 8,
190
+ 8
191
+ ],
192
+ "latent_dim": 64,
193
+ "decoder_dim": 2048,
194
+ "decoder_rates": [
195
+ 8,
196
+ 6,
197
+ 5,
198
+ 2,
199
+ 2,
200
+ 2
201
+ ],
202
+ "sr_bin_boundaries": [
203
+ 20000,
204
+ 30000,
205
+ 40000
206
+ ],
207
+ "sample_rate": 16000,
208
+ "out_sample_rate": 48000
209
+ },
210
+ "max_length": 8192,
211
+ "model_type": "voxcpm2"
212
+ }