File size: 3,650 Bytes
cdc2d96
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
---
library_name: transformers
base_model:
- k2-fsa/OmniVoice
---

This tiny model is intended for debugging. It is randomly initialized using the configuration adapted from [k2-fsa/OmniVoice](https://huggingface.co/k2-fsa/OmniVoice).

| File path | Size |
|------|------|
| model.safetensors | 5.5MB |
| audio_tokenizer/model.safetensors | 6.7MB |


### Example usage:

```python
from omnivoice import OmniVoice
import torch
import torchaudio

model_id = "tiny-random/omnivoice"
model = OmniVoice.from_pretrained(
    model_id,
    dtype=torch.bfloat16,
)
audio = model.generate(
    text="Hello, this is test example 1",
    instruct="low pitch, british accent",
)
torchaudio.save("/tmp/example1.wav", audio[0], 24000)

audio2 = model.generate(
    text="Hello, this is test example 2",
    ref_audio="/tmp/example1.wav",
    ref_text="Hello, this is test example 1",
)
torchaudio.save("/tmp/example2.wav", audio2[0], 24000)
```

### Codes to create this repo:

<details>
<summary>Click to expand</summary>

```python
import torch
import os

from transformers import (
    set_seed,
    AutoConfig,
    AutoTokenizer,
    HiggsAudioV2TokenizerModel,
    AutoFeatureExtractor,
)
from huggingface_hub import hf_hub_download
import json
from omnivoice import OmniVoice, OmniVoiceConfig

source_model_id = "k2-fsa/OmniVoice"
save_folder = "/tmp/tiny-random/omnivoice"

set_seed(42)
tokenizer = AutoTokenizer.from_pretrained(source_model_id, trust_remote_code=True)
tokenizer.save_pretrained(save_folder)

with open(
    hf_hub_download(source_model_id, filename="audio_tokenizer/config.json", repo_type="model"),
    "r",
    encoding="utf-8",
) as f:
    config_dict = json.load(f)
config_dict["acoustic_model_config"].update(
    {
        "decoder_hidden_size": 32,
        "encoder_hidden_size": 4,
        "hidden_size": 4,
        "codebook_dim": 8,
    }
)
config_dict["semantic_model_config"].update(
    {
        "conv_dim": [8] * 7,
        "hidden_size": 16 * 4,
        "intermediate_size": 64,
        "num_attention_heads": 4,
        "num_hidden_layers": 2,
    }
)
os.makedirs(os.path.join(save_folder, "audio_tokenizer"), exist_ok=True)
with open(os.path.join(save_folder, "audio_tokenizer/config.json"), "w", encoding="utf-8") as f:
    json.dump(config_dict, f, ensure_ascii=False, indent=2)
audio_tokenizer = HiggsAudioV2TokenizerModel(
    AutoConfig.from_pretrained(os.path.join(save_folder, "audio_tokenizer"))
)
audio_tokenizer.save_pretrained(os.path.join(save_folder, "audio_tokenizer"))
print(audio_tokenizer)
set_seed(42)
with torch.no_grad():
    for name, p in sorted(audio_tokenizer.named_parameters()):
        torch.nn.init.normal_(p, 0, 0.2)
        print(name, p.shape)

feature_extractor = AutoFeatureExtractor.from_pretrained(source_model_id, subfolder="audio_tokenizer")
feature_extractor.save_pretrained(os.path.join(save_folder, "audio_tokenizer"))

with open(
    hf_hub_download(source_model_id, filename="config.json", repo_type="model"),
    "r",
    encoding="utf-8",
) as f:
    config_dict = json.load(f)
config_dict["llm_config"].update(
    {
        "hidden_size": 8,
        "head_dim": 32,
        "intermediate_size": 32,
        "num_attention_heads": 8,
        "num_key_value_heads": 4,
        "num_hidden_layers": 4,
        "max_window_layers": 2,
        "layer_types": ["full_attention"] * 4,
    }
)
config = OmniVoiceConfig.from_dict(config_dict)
model = OmniVoice(config).eval()
set_seed(42)
with torch.no_grad():
    for name, p in sorted(model.named_parameters()):
        torch.nn.init.normal_(p, 0, 0.2)
        print(name, p.shape)
model.save_pretrained(save_folder)
```

</details>