Spaces:
Runtime error
Runtime error
jianuo
commited on
Commit
•
446c342
0
Parent(s):
submit
Browse filesThis view is limited to 50 files because it contains too many changes.
See raw diff
- .gitattributes +35 -0
- .gitignore +3 -0
- README.md +86 -0
- TTSs/__init__.py +0 -0
- TTSs/__pycache__/__init__.cpython-310.pyc +0 -0
- TTSs/__pycache__/elevenlabs_tts.cpython-310.pyc +0 -0
- TTSs/__pycache__/genshin_api_tts.cpython-310.pyc +0 -0
- TTSs/__pycache__/openai_tts.cpython-310.pyc +0 -0
- TTSs/__pycache__/reecho_ai.cpython-310.pyc +0 -0
- TTSs/__pycache__/tts_utils.cpython-310.pyc +0 -0
- TTSs/elevenlabs_tts.py +66 -0
- TTSs/genshin/.gitignore +188 -0
- TTSs/genshin/.gitmodules +0 -0
- TTSs/genshin/.pre-commit-config.yaml +25 -0
- TTSs/genshin/LICENSE +661 -0
- TTSs/genshin/README.md +42 -0
- TTSs/genshin/__init__.py +0 -0
- TTSs/genshin/attentions.py +464 -0
- TTSs/genshin/bert/bert-base-japanese-v3/.gitattributes +34 -0
- TTSs/genshin/bert/bert-base-japanese-v3/README.md +53 -0
- TTSs/genshin/bert/bert-base-japanese-v3/config.json +19 -0
- TTSs/genshin/bert/bert-base-japanese-v3/tokenizer_config.json +10 -0
- TTSs/genshin/bert/bert-base-japanese-v3/vocab.txt +0 -0
- TTSs/genshin/bert/bert-large-japanese-v2/.gitattributes +34 -0
- TTSs/genshin/bert/bert-large-japanese-v2/README.md +53 -0
- TTSs/genshin/bert/bert-large-japanese-v2/config.json +19 -0
- TTSs/genshin/bert/bert-large-japanese-v2/tokenizer_config.json +10 -0
- TTSs/genshin/bert/bert-large-japanese-v2/vocab.txt +0 -0
- TTSs/genshin/bert/bert_models.json +14 -0
- TTSs/genshin/bert/chinese-roberta-wwm-ext-large/.gitattributes +9 -0
- TTSs/genshin/bert/chinese-roberta-wwm-ext-large/README.md +57 -0
- TTSs/genshin/bert/chinese-roberta-wwm-ext-large/added_tokens.json +1 -0
- TTSs/genshin/bert/chinese-roberta-wwm-ext-large/config.json +28 -0
- TTSs/genshin/bert/chinese-roberta-wwm-ext-large/special_tokens_map.json +1 -0
- TTSs/genshin/bert/chinese-roberta-wwm-ext-large/tokenizer.json +0 -0
- TTSs/genshin/bert/chinese-roberta-wwm-ext-large/tokenizer_config.json +1 -0
- TTSs/genshin/bert/chinese-roberta-wwm-ext-large/vocab.txt +0 -0
- TTSs/genshin/bert/deberta-v2-large-japanese-char-wwm/.gitattributes +34 -0
- TTSs/genshin/bert/deberta-v2-large-japanese-char-wwm/README.md +89 -0
- TTSs/genshin/bert/deberta-v2-large-japanese-char-wwm/config.json +37 -0
- TTSs/genshin/bert/deberta-v2-large-japanese-char-wwm/special_tokens_map.json +7 -0
- TTSs/genshin/bert/deberta-v2-large-japanese-char-wwm/tokenizer_config.json +19 -0
- TTSs/genshin/bert/deberta-v2-large-japanese-char-wwm/vocab.txt +0 -0
- TTSs/genshin/bert/deberta-v2-large-japanese/.gitattributes +34 -0
- TTSs/genshin/bert/deberta-v2-large-japanese/README.md +111 -0
- TTSs/genshin/bert/deberta-v2-large-japanese/config.json +38 -0
- TTSs/genshin/bert/deberta-v2-large-japanese/special_tokens_map.json +9 -0
- TTSs/genshin/bert/deberta-v2-large-japanese/tokenizer.json +0 -0
- TTSs/genshin/bert/deberta-v2-large-japanese/tokenizer_config.json +15 -0
- TTSs/genshin/bert/deberta-v3-large/.gitattributes +27 -0
.gitattributes
ADDED
@@ -0,0 +1,35 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
*.7z filter=lfs diff=lfs merge=lfs -text
|
2 |
+
*.arrow filter=lfs diff=lfs merge=lfs -text
|
3 |
+
*.bin filter=lfs diff=lfs merge=lfs -text
|
4 |
+
*.bz2 filter=lfs diff=lfs merge=lfs -text
|
5 |
+
*.ckpt filter=lfs diff=lfs merge=lfs -text
|
6 |
+
*.ftz filter=lfs diff=lfs merge=lfs -text
|
7 |
+
*.gz filter=lfs diff=lfs merge=lfs -text
|
8 |
+
*.h5 filter=lfs diff=lfs merge=lfs -text
|
9 |
+
*.joblib filter=lfs diff=lfs merge=lfs -text
|
10 |
+
*.lfs.* filter=lfs diff=lfs merge=lfs -text
|
11 |
+
*.mlmodel filter=lfs diff=lfs merge=lfs -text
|
12 |
+
*.model filter=lfs diff=lfs merge=lfs -text
|
13 |
+
*.msgpack filter=lfs diff=lfs merge=lfs -text
|
14 |
+
*.npy filter=lfs diff=lfs merge=lfs -text
|
15 |
+
*.npz filter=lfs diff=lfs merge=lfs -text
|
16 |
+
*.onnx filter=lfs diff=lfs merge=lfs -text
|
17 |
+
*.ot filter=lfs diff=lfs merge=lfs -text
|
18 |
+
*.parquet filter=lfs diff=lfs merge=lfs -text
|
19 |
+
*.pb filter=lfs diff=lfs merge=lfs -text
|
20 |
+
*.pickle filter=lfs diff=lfs merge=lfs -text
|
21 |
+
*.pkl filter=lfs diff=lfs merge=lfs -text
|
22 |
+
*.pt filter=lfs diff=lfs merge=lfs -text
|
23 |
+
*.pth filter=lfs diff=lfs merge=lfs -text
|
24 |
+
*.rar filter=lfs diff=lfs merge=lfs -text
|
25 |
+
*.safetensors filter=lfs diff=lfs merge=lfs -text
|
26 |
+
*.tar.* filter=lfs diff=lfs merge=lfs -text
|
27 |
+
*.tar filter=lfs diff=lfs merge=lfs -text
|
28 |
+
*.tflite filter=lfs diff=lfs merge=lfs -text
|
29 |
+
*.tgz filter=lfs diff=lfs merge=lfs -text
|
30 |
+
*.wasm filter=lfs diff=lfs merge=lfs -text
|
31 |
+
*.xz filter=lfs diff=lfs merge=lfs -text
|
32 |
+
*.zip filter=lfs diff=lfs merge=lfs -text
|
33 |
+
*.zst filter=lfs diff=lfs merge=lfs -text
|
34 |
+
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
35 |
+
saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
.gitignore
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
load_env.py
|
2 |
+
/.idea
|
3 |
+
/__pycache__
|
README.md
ADDED
@@ -0,0 +1,86 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
title: TTS All In One
|
3 |
+
emoji: 👂
|
4 |
+
colorFrom: gray
|
5 |
+
colorTo: pink
|
6 |
+
sdk: gradio
|
7 |
+
sdk_version: 4.16.0
|
8 |
+
app_file: main.py
|
9 |
+
pinned: false
|
10 |
+
license: mit
|
11 |
+
---
|
12 |
+
|
13 |
+
# TTS 4合1 引擎
|
14 |
+
|
15 |
+
这是一个网页版的TTS引擎,支持以下4个TTS引擎:
|
16 |
+
|
17 |
+
- elevenlabs
|
18 |
+
- openai
|
19 |
+
- genshin(原神)
|
20 |
+
- REECHO 睿声
|
21 |
+
|
22 |
+
其中,原神TTS引擎原作者为[红血球AE3803](https://space.bilibili.com/6589795)
|
23 |
+
|
24 |
+
在线推理接口来自其[创空间](https://modelscope.cn/studios/erythrocyte/Bert-VITS2_Genshin_TTS/summary)
|
25 |
+
|
26 |
+
## 所需环境变量
|
27 |
+
|
28 |
+
### elevenlabs
|
29 |
+
|
30 |
+
`ELEVENLABS_API_KEY`:用于elevenlabs的TTS服务
|
31 |
+
|
32 |
+
申请地址:https://elevenlabs.io/
|
33 |
+
|
34 |
+
### openai
|
35 |
+
|
36 |
+
`OPENAI_API_KEY`:用于openai的TTS服务
|
37 |
+
|
38 |
+
申请地址:https://platform.openai.com/api-keys
|
39 |
+
|
40 |
+
`http_proxy`:用于openai的代理
|
41 |
+
|
42 |
+
`https_proxy`:用于openai的代理
|
43 |
+
|
44 |
+
### genshin
|
45 |
+
|
46 |
+
`GENSHIN_API_KEY`:用于genshin的TTS服务
|
47 |
+
|
48 |
+
试用API KEY:`db6704a35055631bc6ba29c26d995f62`
|
49 |
+
|
50 |
+
申请地址:https://tts.ai-hobbyist.org/#/apikey
|
51 |
+
|
52 |
+
### 本地genshin推理
|
53 |
+
|
54 |
+
默认关闭该服务,若启动需将环境变量‘USING_LOCAL’设置为`true`。
|
55 |
+
|
56 |
+
此外`./TTSs/genshin/config.yml`为该模型的配置,为方便在huggingface空间上运行,已修改为各模型在`cpu`
|
57 |
+
上运行,如有需要,可以手动改将参数`device`改为`cuda`
|
58 |
+
|
59 |
+
### REECHO 睿声
|
60 |
+
|
61 |
+
`REECHO_API_KEY`:用于REECHO的TTS服务
|
62 |
+
|
63 |
+
控制台在此:https://dash.reecho.ai/'
|
64 |
+
|
65 |
+
API申请地址:https://dash.reecho.ai/apiKey
|
66 |
+
|
67 |
+
## 安装方法
|
68 |
+
|
69 |
+
安装python(推荐3.10)
|
70 |
+
|
71 |
+
安装ffmpeg,需要将ffmpeg所在目录添加到环境变量中,即可在命令行中使用ffmpeg命令。
|
72 |
+
|
73 |
+
随后在命令行中输入以下命令:
|
74 |
+
|
75 |
+
```bash
|
76 |
+
pip install -r requirements.txt
|
77 |
+
python main.py
|
78 |
+
```
|
79 |
+
|
80 |
+
之后应该看到这样的提示:
|
81 |
+
|
82 |
+
![img.png](img/img.png)
|
83 |
+
|
84 |
+
点开网页后,就可以直接使用了
|
85 |
+
|
86 |
+
![img.png](img/img2.png)
|
TTSs/__init__.py
ADDED
File without changes
|
TTSs/__pycache__/__init__.cpython-310.pyc
ADDED
Binary file (116 Bytes). View file
|
|
TTSs/__pycache__/elevenlabs_tts.cpython-310.pyc
ADDED
Binary file (1.79 kB). View file
|
|
TTSs/__pycache__/genshin_api_tts.cpython-310.pyc
ADDED
Binary file (1.49 kB). View file
|
|
TTSs/__pycache__/openai_tts.cpython-310.pyc
ADDED
Binary file (747 Bytes). View file
|
|
TTSs/__pycache__/reecho_ai.cpython-310.pyc
ADDED
Binary file (2.16 kB). View file
|
|
TTSs/__pycache__/tts_utils.cpython-310.pyc
ADDED
Binary file (780 Bytes). View file
|
|
TTSs/elevenlabs_tts.py
ADDED
@@ -0,0 +1,66 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import io
|
2 |
+
import json
|
3 |
+
|
4 |
+
import requests
|
5 |
+
from elevenlabs import voices, generate, set_api_key, VoiceSettings, Voice
|
6 |
+
from pydub import AudioSegment
|
7 |
+
|
8 |
+
from .tts_utils import mix_background_music
|
9 |
+
|
10 |
+
|
11 |
+
def merge_audio(token, text, 背景音乐, 音色选择, stability, similarity_boost, style, use_speaker_boost, TTS_up, bg_up):
|
12 |
+
try:
|
13 |
+
set_api_key(token)
|
14 |
+
|
15 |
+
original_audio_bytes = generate(
|
16 |
+
text=text,
|
17 |
+
voice=Voice(voice_id=voices_eleven[speakers_eleven[音色选择]].voice_id,
|
18 |
+
settings=VoiceSettings(stability=stability, similarity_boost=similarity_boost, style=style,
|
19 |
+
use_speaker_boost=use_speaker_boost)),
|
20 |
+
model="eleven_multilingual_v2",
|
21 |
+
|
22 |
+
)
|
23 |
+
|
24 |
+
# 将字节串转换为AudioSegment对象
|
25 |
+
original_audio = AudioSegment.from_file(io.BytesIO(original_audio_bytes), format="mp3")
|
26 |
+
|
27 |
+
return None, *mix_background_music(original_audio, 背景音乐, TTS_up, bg_up)
|
28 |
+
except Exception as e:
|
29 |
+
return str(e), None, None
|
30 |
+
|
31 |
+
|
32 |
+
def get_eleven_spk():
|
33 |
+
try:
|
34 |
+
voices_eleven = voices()
|
35 |
+
speakers_eleven = {}
|
36 |
+
for i, v in enumerate(voices_eleven):
|
37 |
+
label = {}
|
38 |
+
label['口音'] = v.labels.get('accent')
|
39 |
+
|
40 |
+
if 'description ' in v.labels:
|
41 |
+
label['描述'] = v.labels['description ']
|
42 |
+
elif 'description' in v.labels:
|
43 |
+
label['描述'] = v.labels['description']
|
44 |
+
|
45 |
+
label['年龄'] = v.labels['age']
|
46 |
+
label['性别'] = v.labels['gender']
|
47 |
+
|
48 |
+
if 'use case' in v.labels:
|
49 |
+
label['用例'] = v.labels['use case']
|
50 |
+
elif 'usecase' in v.labels:
|
51 |
+
label['用例'] = v.labels['usecase']
|
52 |
+
else:
|
53 |
+
raise Exception('use case not found')
|
54 |
+
|
55 |
+
key = v.name + '-' + str(label)
|
56 |
+
speakers_eleven[key] = i
|
57 |
+
select_key = list(speakers_eleven.keys())
|
58 |
+
|
59 |
+
return voices_eleven, speakers_eleven, select_key
|
60 |
+
except Exception as e:
|
61 |
+
print(e)
|
62 |
+
return [], {'获取音色失败': '获取音色失败'}, ['获取音色失败']
|
63 |
+
|
64 |
+
|
65 |
+
|
66 |
+
voices_eleven, speakers_eleven, select_key = get_eleven_spk()
|
TTSs/genshin/.gitignore
ADDED
@@ -0,0 +1,188 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Byte-compiled / optimized / DLL files
|
2 |
+
__pycache__/
|
3 |
+
*.py[cod]
|
4 |
+
*$py.class
|
5 |
+
|
6 |
+
# C extensions
|
7 |
+
*.so
|
8 |
+
|
9 |
+
# Distribution / packaging
|
10 |
+
.Python
|
11 |
+
build/
|
12 |
+
develop-eggs/
|
13 |
+
dist/
|
14 |
+
downloads/
|
15 |
+
eggs/
|
16 |
+
.eggs/
|
17 |
+
lib/
|
18 |
+
lib64/
|
19 |
+
parts/
|
20 |
+
sdist/
|
21 |
+
var/
|
22 |
+
wheels/
|
23 |
+
share/python-wheels/
|
24 |
+
*.egg-info/
|
25 |
+
.installed.cfg
|
26 |
+
*.egg
|
27 |
+
MANIFEST
|
28 |
+
|
29 |
+
# PyInstaller
|
30 |
+
# Usually these files are written by a python script from a template
|
31 |
+
# before PyInstaller builds the exe, so as to inject date/other infos into it.
|
32 |
+
*.manifest
|
33 |
+
*.spec
|
34 |
+
|
35 |
+
# Installer logs
|
36 |
+
pip-log.txt
|
37 |
+
pip-delete-this-directory.txt
|
38 |
+
|
39 |
+
# Unit test / coverage reports
|
40 |
+
htmlcov/
|
41 |
+
.tox/
|
42 |
+
.nox/
|
43 |
+
.coverage
|
44 |
+
.coverage.*
|
45 |
+
.cache
|
46 |
+
nosetests.xml
|
47 |
+
coverage.xml
|
48 |
+
*.cover
|
49 |
+
*.py,cover
|
50 |
+
.hypothesis/
|
51 |
+
.pytest_cache/
|
52 |
+
cover/
|
53 |
+
|
54 |
+
# Translations
|
55 |
+
*.mo
|
56 |
+
*.pot
|
57 |
+
|
58 |
+
# Django stuff:
|
59 |
+
*.log
|
60 |
+
local_settings.py
|
61 |
+
db.sqlite3
|
62 |
+
db.sqlite3-journal
|
63 |
+
|
64 |
+
# Flask stuff:
|
65 |
+
instance/
|
66 |
+
.webassets-cache
|
67 |
+
|
68 |
+
# Scrapy stuff:
|
69 |
+
.scrapy
|
70 |
+
|
71 |
+
# Sphinx documentation
|
72 |
+
docs/_build/
|
73 |
+
|
74 |
+
# PyBuilder
|
75 |
+
.pybuilder/
|
76 |
+
target/
|
77 |
+
|
78 |
+
# Jupyter Notebook
|
79 |
+
.ipynb_checkpoints
|
80 |
+
|
81 |
+
# IPython
|
82 |
+
profile_default/
|
83 |
+
ipython_config.py
|
84 |
+
|
85 |
+
# pyenv
|
86 |
+
# For a library or package, you might want to ignore these files since the code is
|
87 |
+
# intended to run in multiple environments; otherwise, check them in:
|
88 |
+
# .python-version
|
89 |
+
|
90 |
+
# pipenv
|
91 |
+
# According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
|
92 |
+
# However, in case of collaboration, if having platform-specific dependencies or dependencies
|
93 |
+
# having no cross-platform support, pipenv may install dependencies that don't work, or not
|
94 |
+
# install all needed dependencies.
|
95 |
+
#Pipfile.lock
|
96 |
+
|
97 |
+
# poetry
|
98 |
+
# Similar to Pipfile.lock, it is generally recommended to include poetry.lock in version control.
|
99 |
+
# This is especially recommended for binary packages to ensure reproducibility, and is more
|
100 |
+
# commonly ignored for libraries.
|
101 |
+
# https://python-poetry.org/docs/basic-usage/#commit-your-poetrylock-file-to-version-control
|
102 |
+
#poetry.lock
|
103 |
+
|
104 |
+
# pdm
|
105 |
+
# Similar to Pipfile.lock, it is generally recommended to include pdm.lock in version control.
|
106 |
+
#pdm.lock
|
107 |
+
# pdm stores project-wide configurations in .pdm.toml, but it is recommended to not include it
|
108 |
+
# in version control.
|
109 |
+
# https://pdm.fming.dev/#use-with-ide
|
110 |
+
.pdm.toml
|
111 |
+
|
112 |
+
# PEP 582; used by e.g. github.com/David-OConnor/pyflow and github.com/pdm-project/pdm
|
113 |
+
__pypackages__/
|
114 |
+
|
115 |
+
# Celery stuff
|
116 |
+
celerybeat-schedule
|
117 |
+
celerybeat.pid
|
118 |
+
|
119 |
+
# SageMath parsed files
|
120 |
+
*.sage.py
|
121 |
+
|
122 |
+
# Environments
|
123 |
+
.env
|
124 |
+
.venv
|
125 |
+
env/
|
126 |
+
venv/
|
127 |
+
ENV/
|
128 |
+
env.bak/
|
129 |
+
venv.bak/
|
130 |
+
|
131 |
+
# Spyder project settings
|
132 |
+
.spyderproject
|
133 |
+
.spyproject
|
134 |
+
|
135 |
+
# Rope project settings
|
136 |
+
.ropeproject
|
137 |
+
|
138 |
+
# mkdocs documentation
|
139 |
+
/site
|
140 |
+
|
141 |
+
# mypy
|
142 |
+
.mypy_cache/
|
143 |
+
.dmypy.json
|
144 |
+
dmypy.json
|
145 |
+
|
146 |
+
# Pyre type checker
|
147 |
+
.pyre/
|
148 |
+
|
149 |
+
# pytype static type analyzer
|
150 |
+
.pytype/
|
151 |
+
|
152 |
+
# Cython debug symbols
|
153 |
+
cython_debug/
|
154 |
+
|
155 |
+
# PyCharm
|
156 |
+
# JetBrains specific template is maintained in a separate JetBrains.gitignore that can
|
157 |
+
# be found at https://github.com/github/gitignore/blob/main/Global/JetBrains.gitignore
|
158 |
+
# and can be added to the global gitignore or merged into this file. For a more nuclear
|
159 |
+
# option (not recommended) you can uncomment the following to ignore the entire idea folder.
|
160 |
+
#.idea/
|
161 |
+
|
162 |
+
.DS_Store
|
163 |
+
/models
|
164 |
+
/logs
|
165 |
+
|
166 |
+
filelists/sample.list
|
167 |
+
!/filelists/esd.list
|
168 |
+
data/*
|
169 |
+
/*.yml
|
170 |
+
!/default_config.yml
|
171 |
+
/Web/
|
172 |
+
/emotional/*/*.bin
|
173 |
+
/slm/*/*.bin
|
174 |
+
/bert/*/*.bin
|
175 |
+
/bert/*/*.h5
|
176 |
+
/bert/*/*.model
|
177 |
+
/bert/*/*.safetensors
|
178 |
+
/bert/*/*.msgpack
|
179 |
+
asr_transcript.py
|
180 |
+
extract_list.py
|
181 |
+
dataset
|
182 |
+
/Data
|
183 |
+
Model
|
184 |
+
raw/
|
185 |
+
logs/
|
186 |
+
Data/*
|
187 |
+
/onnx
|
188 |
+
/.vs
|
TTSs/genshin/.gitmodules
ADDED
File without changes
|
TTSs/genshin/.pre-commit-config.yaml
ADDED
@@ -0,0 +1,25 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
repos:
|
2 |
+
- repo: https://github.com/pre-commit/pre-commit-hooks
|
3 |
+
rev: v4.5.0
|
4 |
+
hooks:
|
5 |
+
- id: check-yaml
|
6 |
+
- id: end-of-file-fixer
|
7 |
+
- id: trailing-whitespace
|
8 |
+
|
9 |
+
- repo: https://github.com/astral-sh/ruff-pre-commit
|
10 |
+
rev: v0.1.8
|
11 |
+
hooks:
|
12 |
+
- id: ruff
|
13 |
+
args: [ --fix ]
|
14 |
+
|
15 |
+
- repo: https://github.com/psf/black
|
16 |
+
rev: 23.12.0
|
17 |
+
hooks:
|
18 |
+
- id: black
|
19 |
+
|
20 |
+
- repo: https://github.com/codespell-project/codespell
|
21 |
+
rev: v2.2.6
|
22 |
+
hooks:
|
23 |
+
- id: codespell
|
24 |
+
files: ^.*\.(py|md|rst|yml)$
|
25 |
+
args: [-L=fro]
|
TTSs/genshin/LICENSE
ADDED
@@ -0,0 +1,661 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
GNU AFFERO GENERAL PUBLIC LICENSE
|
2 |
+
Version 3, 19 November 2007
|
3 |
+
|
4 |
+
Copyright (C) 2007 Free Software Foundation, Inc. <https://fsf.org/>
|
5 |
+
Everyone is permitted to copy and distribute verbatim copies
|
6 |
+
of this license document, but changing it is not allowed.
|
7 |
+
|
8 |
+
Preamble
|
9 |
+
|
10 |
+
The GNU Affero General Public License is a free, copyleft license for
|
11 |
+
software and other kinds of works, specifically designed to ensure
|
12 |
+
cooperation with the community in the case of network server software.
|
13 |
+
|
14 |
+
The licenses for most software and other practical works are designed
|
15 |
+
to take away your freedom to share and change the works. By contrast,
|
16 |
+
our General Public Licenses are intended to guarantee your freedom to
|
17 |
+
share and change all versions of a program--to make sure it remains free
|
18 |
+
software for all its users.
|
19 |
+
|
20 |
+
When we speak of free software, we are referring to freedom, not
|
21 |
+
price. Our General Public Licenses are designed to make sure that you
|
22 |
+
have the freedom to distribute copies of free software (and charge for
|
23 |
+
them if you wish), that you receive source code or can get it if you
|
24 |
+
want it, that you can change the software or use pieces of it in new
|
25 |
+
free programs, and that you know you can do these things.
|
26 |
+
|
27 |
+
Developers that use our General Public Licenses protect your rights
|
28 |
+
with two steps: (1) assert copyright on the software, and (2) offer
|
29 |
+
you this License which gives you legal permission to copy, distribute
|
30 |
+
and/or modify the software.
|
31 |
+
|
32 |
+
A secondary benefit of defending all users' freedom is that
|
33 |
+
improvements made in alternate versions of the program, if they
|
34 |
+
receive widespread use, become available for other developers to
|
35 |
+
incorporate. Many developers of free software are heartened and
|
36 |
+
encouraged by the resulting cooperation. However, in the case of
|
37 |
+
software used on network servers, this result may fail to come about.
|
38 |
+
The GNU General Public License permits making a modified version and
|
39 |
+
letting the public access it on a server without ever releasing its
|
40 |
+
source code to the public.
|
41 |
+
|
42 |
+
The GNU Affero General Public License is designed specifically to
|
43 |
+
ensure that, in such cases, the modified source code becomes available
|
44 |
+
to the community. It requires the operator of a network server to
|
45 |
+
provide the source code of the modified version running there to the
|
46 |
+
users of that server. Therefore, public use of a modified version, on
|
47 |
+
a publicly accessible server, gives the public access to the source
|
48 |
+
code of the modified version.
|
49 |
+
|
50 |
+
An older license, called the Affero General Public License and
|
51 |
+
published by Affero, was designed to accomplish similar goals. This is
|
52 |
+
a different license, not a version of the Affero GPL, but Affero has
|
53 |
+
released a new version of the Affero GPL which permits relicensing under
|
54 |
+
this license.
|
55 |
+
|
56 |
+
The precise terms and conditions for copying, distribution and
|
57 |
+
modification follow.
|
58 |
+
|
59 |
+
TERMS AND CONDITIONS
|
60 |
+
|
61 |
+
0. Definitions.
|
62 |
+
|
63 |
+
"This License" refers to version 3 of the GNU Affero General Public License.
|
64 |
+
|
65 |
+
"Copyright" also means copyright-like laws that apply to other kinds of
|
66 |
+
works, such as semiconductor masks.
|
67 |
+
|
68 |
+
"The Program" refers to any copyrightable work licensed under this
|
69 |
+
License. Each licensee is addressed as "you". "Licensees" and
|
70 |
+
"recipients" may be individuals or organizations.
|
71 |
+
|
72 |
+
To "modify" a work means to copy from or adapt all or part of the work
|
73 |
+
in a fashion requiring copyright permission, other than the making of an
|
74 |
+
exact copy. The resulting work is called a "modified version" of the
|
75 |
+
earlier work or a work "based on" the earlier work.
|
76 |
+
|
77 |
+
A "covered work" means either the unmodified Program or a work based
|
78 |
+
on the Program.
|
79 |
+
|
80 |
+
To "propagate" a work means to do anything with it that, without
|
81 |
+
permission, would make you directly or secondarily liable for
|
82 |
+
infringement under applicable copyright law, except executing it on a
|
83 |
+
computer or modifying a private copy. Propagation includes copying,
|
84 |
+
distribution (with or without modification), making available to the
|
85 |
+
public, and in some countries other activities as well.
|
86 |
+
|
87 |
+
To "convey" a work means any kind of propagation that enables other
|
88 |
+
parties to make or receive copies. Mere interaction with a user through
|
89 |
+
a computer network, with no transfer of a copy, is not conveying.
|
90 |
+
|
91 |
+
An interactive user interface displays "Appropriate Legal Notices"
|
92 |
+
to the extent that it includes a convenient and prominently visible
|
93 |
+
feature that (1) displays an appropriate copyright notice, and (2)
|
94 |
+
tells the user that there is no warranty for the work (except to the
|
95 |
+
extent that warranties are provided), that licensees may convey the
|
96 |
+
work under this License, and how to view a copy of this License. If
|
97 |
+
the interface presents a list of user commands or options, such as a
|
98 |
+
menu, a prominent item in the list meets this criterion.
|
99 |
+
|
100 |
+
1. Source Code.
|
101 |
+
|
102 |
+
The "source code" for a work means the preferred form of the work
|
103 |
+
for making modifications to it. "Object code" means any non-source
|
104 |
+
form of a work.
|
105 |
+
|
106 |
+
A "Standard Interface" means an interface that either is an official
|
107 |
+
standard defined by a recognized standards body, or, in the case of
|
108 |
+
interfaces specified for a particular programming language, one that
|
109 |
+
is widely used among developers working in that language.
|
110 |
+
|
111 |
+
The "System Libraries" of an executable work include anything, other
|
112 |
+
than the work as a whole, that (a) is included in the normal form of
|
113 |
+
packaging a Major Component, but which is not part of that Major
|
114 |
+
Component, and (b) serves only to enable use of the work with that
|
115 |
+
Major Component, or to implement a Standard Interface for which an
|
116 |
+
implementation is available to the public in source code form. A
|
117 |
+
"Major Component", in this context, means a major essential component
|
118 |
+
(kernel, window system, and so on) of the specific operating system
|
119 |
+
(if any) on which the executable work runs, or a compiler used to
|
120 |
+
produce the work, or an object code interpreter used to run it.
|
121 |
+
|
122 |
+
The "Corresponding Source" for a work in object code form means all
|
123 |
+
the source code needed to generate, install, and (for an executable
|
124 |
+
work) run the object code and to modify the work, including scripts to
|
125 |
+
control those activities. However, it does not include the work's
|
126 |
+
System Libraries, or general-purpose tools or generally available free
|
127 |
+
programs which are used unmodified in performing those activities but
|
128 |
+
which are not part of the work. For example, Corresponding Source
|
129 |
+
includes interface definition files associated with source files for
|
130 |
+
the work, and the source code for shared libraries and dynamically
|
131 |
+
linked subprograms that the work is specifically designed to require,
|
132 |
+
such as by intimate data communication or control flow between those
|
133 |
+
subprograms and other parts of the work.
|
134 |
+
|
135 |
+
The Corresponding Source need not include anything that users
|
136 |
+
can regenerate automatically from other parts of the Corresponding
|
137 |
+
Source.
|
138 |
+
|
139 |
+
The Corresponding Source for a work in source code form is that
|
140 |
+
same work.
|
141 |
+
|
142 |
+
2. Basic Permissions.
|
143 |
+
|
144 |
+
All rights granted under this License are granted for the term of
|
145 |
+
copyright on the Program, and are irrevocable provided the stated
|
146 |
+
conditions are met. This License explicitly affirms your unlimited
|
147 |
+
permission to run the unmodified Program. The output from running a
|
148 |
+
covered work is covered by this License only if the output, given its
|
149 |
+
content, constitutes a covered work. This License acknowledges your
|
150 |
+
rights of fair use or other equivalent, as provided by copyright law.
|
151 |
+
|
152 |
+
You may make, run and propagate covered works that you do not
|
153 |
+
convey, without conditions so long as your license otherwise remains
|
154 |
+
in force. You may convey covered works to others for the sole purpose
|
155 |
+
of having them make modifications exclusively for you, or provide you
|
156 |
+
with facilities for running those works, provided that you comply with
|
157 |
+
the terms of this License in conveying all material for which you do
|
158 |
+
not control copyright. Those thus making or running the covered works
|
159 |
+
for you must do so exclusively on your behalf, under your direction
|
160 |
+
and control, on terms that prohibit them from making any copies of
|
161 |
+
your copyrighted material outside their relationship with you.
|
162 |
+
|
163 |
+
Conveying under any other circumstances is permitted solely under
|
164 |
+
the conditions stated below. Sublicensing is not allowed; section 10
|
165 |
+
makes it unnecessary.
|
166 |
+
|
167 |
+
3. Protecting Users' Legal Rights From Anti-Circumvention Law.
|
168 |
+
|
169 |
+
No covered work shall be deemed part of an effective technological
|
170 |
+
measure under any applicable law fulfilling obligations under article
|
171 |
+
11 of the WIPO copyright treaty adopted on 20 December 1996, or
|
172 |
+
similar laws prohibiting or restricting circumvention of such
|
173 |
+
measures.
|
174 |
+
|
175 |
+
When you convey a covered work, you waive any legal power to forbid
|
176 |
+
circumvention of technological measures to the extent such circumvention
|
177 |
+
is effected by exercising rights under this License with respect to
|
178 |
+
the covered work, and you disclaim any intention to limit operation or
|
179 |
+
modification of the work as a means of enforcing, against the work's
|
180 |
+
users, your or third parties' legal rights to forbid circumvention of
|
181 |
+
technological measures.
|
182 |
+
|
183 |
+
4. Conveying Verbatim Copies.
|
184 |
+
|
185 |
+
You may convey verbatim copies of the Program's source code as you
|
186 |
+
receive it, in any medium, provided that you conspicuously and
|
187 |
+
appropriately publish on each copy an appropriate copyright notice;
|
188 |
+
keep intact all notices stating that this License and any
|
189 |
+
non-permissive terms added in accord with section 7 apply to the code;
|
190 |
+
keep intact all notices of the absence of any warranty; and give all
|
191 |
+
recipients a copy of this License along with the Program.
|
192 |
+
|
193 |
+
You may charge any price or no price for each copy that you convey,
|
194 |
+
and you may offer support or warranty protection for a fee.
|
195 |
+
|
196 |
+
5. Conveying Modified Source Versions.
|
197 |
+
|
198 |
+
You may convey a work based on the Program, or the modifications to
|
199 |
+
produce it from the Program, in the form of source code under the
|
200 |
+
terms of section 4, provided that you also meet all of these conditions:
|
201 |
+
|
202 |
+
a) The work must carry prominent notices stating that you modified
|
203 |
+
it, and giving a relevant date.
|
204 |
+
|
205 |
+
b) The work must carry prominent notices stating that it is
|
206 |
+
released under this License and any conditions added under section
|
207 |
+
7. This requirement modifies the requirement in section 4 to
|
208 |
+
"keep intact all notices".
|
209 |
+
|
210 |
+
c) You must license the entire work, as a whole, under this
|
211 |
+
License to anyone who comes into possession of a copy. This
|
212 |
+
License will therefore apply, along with any applicable section 7
|
213 |
+
additional terms, to the whole of the work, and all its parts,
|
214 |
+
regardless of how they are packaged. This License gives no
|
215 |
+
permission to license the work in any other way, but it does not
|
216 |
+
invalidate such permission if you have separately received it.
|
217 |
+
|
218 |
+
d) If the work has interactive user interfaces, each must display
|
219 |
+
Appropriate Legal Notices; however, if the Program has interactive
|
220 |
+
interfaces that do not display Appropriate Legal Notices, your
|
221 |
+
work need not make them do so.
|
222 |
+
|
223 |
+
A compilation of a covered work with other separate and independent
|
224 |
+
works, which are not by their nature extensions of the covered work,
|
225 |
+
and which are not combined with it such as to form a larger program,
|
226 |
+
in or on a volume of a storage or distribution medium, is called an
|
227 |
+
"aggregate" if the compilation and its resulting copyright are not
|
228 |
+
used to limit the access or legal rights of the compilation's users
|
229 |
+
beyond what the individual works permit. Inclusion of a covered work
|
230 |
+
in an aggregate does not cause this License to apply to the other
|
231 |
+
parts of the aggregate.
|
232 |
+
|
233 |
+
6. Conveying Non-Source Forms.
|
234 |
+
|
235 |
+
You may convey a covered work in object code form under the terms
|
236 |
+
of sections 4 and 5, provided that you also convey the
|
237 |
+
machine-readable Corresponding Source under the terms of this License,
|
238 |
+
in one of these ways:
|
239 |
+
|
240 |
+
a) Convey the object code in, or embodied in, a physical product
|
241 |
+
(including a physical distribution medium), accompanied by the
|
242 |
+
Corresponding Source fixed on a durable physical medium
|
243 |
+
customarily used for software interchange.
|
244 |
+
|
245 |
+
b) Convey the object code in, or embodied in, a physical product
|
246 |
+
(including a physical distribution medium), accompanied by a
|
247 |
+
written offer, valid for at least three years and valid for as
|
248 |
+
long as you offer spare parts or customer support for that product
|
249 |
+
model, to give anyone who possesses the object code either (1) a
|
250 |
+
copy of the Corresponding Source for all the software in the
|
251 |
+
product that is covered by this License, on a durable physical
|
252 |
+
medium customarily used for software interchange, for a price no
|
253 |
+
more than your reasonable cost of physically performing this
|
254 |
+
conveying of source, or (2) access to copy the
|
255 |
+
Corresponding Source from a network server at no charge.
|
256 |
+
|
257 |
+
c) Convey individual copies of the object code with a copy of the
|
258 |
+
written offer to provide the Corresponding Source. This
|
259 |
+
alternative is allowed only occasionally and noncommercially, and
|
260 |
+
only if you received the object code with such an offer, in accord
|
261 |
+
with subsection 6b.
|
262 |
+
|
263 |
+
d) Convey the object code by offering access from a designated
|
264 |
+
place (gratis or for a charge), and offer equivalent access to the
|
265 |
+
Corresponding Source in the same way through the same place at no
|
266 |
+
further charge. You need not require recipients to copy the
|
267 |
+
Corresponding Source along with the object code. If the place to
|
268 |
+
copy the object code is a network server, the Corresponding Source
|
269 |
+
may be on a different server (operated by you or a third party)
|
270 |
+
that supports equivalent copying facilities, provided you maintain
|
271 |
+
clear directions next to the object code saying where to find the
|
272 |
+
Corresponding Source. Regardless of what server hosts the
|
273 |
+
Corresponding Source, you remain obligated to ensure that it is
|
274 |
+
available for as long as needed to satisfy these requirements.
|
275 |
+
|
276 |
+
e) Convey the object code using peer-to-peer transmission, provided
|
277 |
+
you inform other peers where the object code and Corresponding
|
278 |
+
Source of the work are being offered to the general public at no
|
279 |
+
charge under subsection 6d.
|
280 |
+
|
281 |
+
A separable portion of the object code, whose source code is excluded
|
282 |
+
from the Corresponding Source as a System Library, need not be
|
283 |
+
included in conveying the object code work.
|
284 |
+
|
285 |
+
A "User Product" is either (1) a "consumer product", which means any
|
286 |
+
tangible personal property which is normally used for personal, family,
|
287 |
+
or household purposes, or (2) anything designed or sold for incorporation
|
288 |
+
into a dwelling. In determining whether a product is a consumer product,
|
289 |
+
doubtful cases shall be resolved in favor of coverage. For a particular
|
290 |
+
product received by a particular user, "normally used" refers to a
|
291 |
+
typical or common use of that class of product, regardless of the status
|
292 |
+
of the particular user or of the way in which the particular user
|
293 |
+
actually uses, or expects or is expected to use, the product. A product
|
294 |
+
is a consumer product regardless of whether the product has substantial
|
295 |
+
commercial, industrial or non-consumer uses, unless such uses represent
|
296 |
+
the only significant mode of use of the product.
|
297 |
+
|
298 |
+
"Installation Information" for a User Product means any methods,
|
299 |
+
procedures, authorization keys, or other information required to install
|
300 |
+
and execute modified versions of a covered work in that User Product from
|
301 |
+
a modified version of its Corresponding Source. The information must
|
302 |
+
suffice to ensure that the continued functioning of the modified object
|
303 |
+
code is in no case prevented or interfered with solely because
|
304 |
+
modification has been made.
|
305 |
+
|
306 |
+
If you convey an object code work under this section in, or with, or
|
307 |
+
specifically for use in, a User Product, and the conveying occurs as
|
308 |
+
part of a transaction in which the right of possession and use of the
|
309 |
+
User Product is transferred to the recipient in perpetuity or for a
|
310 |
+
fixed term (regardless of how the transaction is characterized), the
|
311 |
+
Corresponding Source conveyed under this section must be accompanied
|
312 |
+
by the Installation Information. But this requirement does not apply
|
313 |
+
if neither you nor any third party retains the ability to install
|
314 |
+
modified object code on the User Product (for example, the work has
|
315 |
+
been installed in ROM).
|
316 |
+
|
317 |
+
The requirement to provide Installation Information does not include a
|
318 |
+
requirement to continue to provide support service, warranty, or updates
|
319 |
+
for a work that has been modified or installed by the recipient, or for
|
320 |
+
the User Product in which it has been modified or installed. Access to a
|
321 |
+
network may be denied when the modification itself materially and
|
322 |
+
adversely affects the operation of the network or violates the rules and
|
323 |
+
protocols for communication across the network.
|
324 |
+
|
325 |
+
Corresponding Source conveyed, and Installation Information provided,
|
326 |
+
in accord with this section must be in a format that is publicly
|
327 |
+
documented (and with an implementation available to the public in
|
328 |
+
source code form), and must require no special password or key for
|
329 |
+
unpacking, reading or copying.
|
330 |
+
|
331 |
+
7. Additional Terms.
|
332 |
+
|
333 |
+
"Additional permissions" are terms that supplement the terms of this
|
334 |
+
License by making exceptions from one or more of its conditions.
|
335 |
+
Additional permissions that are applicable to the entire Program shall
|
336 |
+
be treated as though they were included in this License, to the extent
|
337 |
+
that they are valid under applicable law. If additional permissions
|
338 |
+
apply only to part of the Program, that part may be used separately
|
339 |
+
under those permissions, but the entire Program remains governed by
|
340 |
+
this License without regard to the additional permissions.
|
341 |
+
|
342 |
+
When you convey a copy of a covered work, you may at your option
|
343 |
+
remove any additional permissions from that copy, or from any part of
|
344 |
+
it. (Additional permissions may be written to require their own
|
345 |
+
removal in certain cases when you modify the work.) You may place
|
346 |
+
additional permissions on material, added by you to a covered work,
|
347 |
+
for which you have or can give appropriate copyright permission.
|
348 |
+
|
349 |
+
Notwithstanding any other provision of this License, for material you
|
350 |
+
add to a covered work, you may (if authorized by the copyright holders of
|
351 |
+
that material) supplement the terms of this License with terms:
|
352 |
+
|
353 |
+
a) Disclaiming warranty or limiting liability differently from the
|
354 |
+
terms of sections 15 and 16 of this License; or
|
355 |
+
|
356 |
+
b) Requiring preservation of specified reasonable legal notices or
|
357 |
+
author attributions in that material or in the Appropriate Legal
|
358 |
+
Notices displayed by works containing it; or
|
359 |
+
|
360 |
+
c) Prohibiting misrepresentation of the origin of that material, or
|
361 |
+
requiring that modified versions of such material be marked in
|
362 |
+
reasonable ways as different from the original version; or
|
363 |
+
|
364 |
+
d) Limiting the use for publicity purposes of names of licensors or
|
365 |
+
authors of the material; or
|
366 |
+
|
367 |
+
e) Declining to grant rights under trademark law for use of some
|
368 |
+
trade names, trademarks, or service marks; or
|
369 |
+
|
370 |
+
f) Requiring indemnification of licensors and authors of that
|
371 |
+
material by anyone who conveys the material (or modified versions of
|
372 |
+
it) with contractual assumptions of liability to the recipient, for
|
373 |
+
any liability that these contractual assumptions directly impose on
|
374 |
+
those licensors and authors.
|
375 |
+
|
376 |
+
All other non-permissive additional terms are considered "further
|
377 |
+
restrictions" within the meaning of section 10. If the Program as you
|
378 |
+
received it, or any part of it, contains a notice stating that it is
|
379 |
+
governed by this License along with a term that is a further
|
380 |
+
restriction, you may remove that term. If a license document contains
|
381 |
+
a further restriction but permits relicensing or conveying under this
|
382 |
+
License, you may add to a covered work material governed by the terms
|
383 |
+
of that license document, provided that the further restriction does
|
384 |
+
not survive such relicensing or conveying.
|
385 |
+
|
386 |
+
If you add terms to a covered work in accord with this section, you
|
387 |
+
must place, in the relevant source files, a statement of the
|
388 |
+
additional terms that apply to those files, or a notice indicating
|
389 |
+
where to find the applicable terms.
|
390 |
+
|
391 |
+
Additional terms, permissive or non-permissive, may be stated in the
|
392 |
+
form of a separately written license, or stated as exceptions;
|
393 |
+
the above requirements apply either way.
|
394 |
+
|
395 |
+
8. Termination.
|
396 |
+
|
397 |
+
You may not propagate or modify a covered work except as expressly
|
398 |
+
provided under this License. Any attempt otherwise to propagate or
|
399 |
+
modify it is void, and will automatically terminate your rights under
|
400 |
+
this License (including any patent licenses granted under the third
|
401 |
+
paragraph of section 11).
|
402 |
+
|
403 |
+
However, if you cease all violation of this License, then your
|
404 |
+
license from a particular copyright holder is reinstated (a)
|
405 |
+
provisionally, unless and until the copyright holder explicitly and
|
406 |
+
finally terminates your license, and (b) permanently, if the copyright
|
407 |
+
holder fails to notify you of the violation by some reasonable means
|
408 |
+
prior to 60 days after the cessation.
|
409 |
+
|
410 |
+
Moreover, your license from a particular copyright holder is
|
411 |
+
reinstated permanently if the copyright holder notifies you of the
|
412 |
+
violation by some reasonable means, this is the first time you have
|
413 |
+
received notice of violation of this License (for any work) from that
|
414 |
+
copyright holder, and you cure the violation prior to 30 days after
|
415 |
+
your receipt of the notice.
|
416 |
+
|
417 |
+
Termination of your rights under this section does not terminate the
|
418 |
+
licenses of parties who have received copies or rights from you under
|
419 |
+
this License. If your rights have been terminated and not permanently
|
420 |
+
reinstated, you do not qualify to receive new licenses for the same
|
421 |
+
material under section 10.
|
422 |
+
|
423 |
+
9. Acceptance Not Required for Having Copies.
|
424 |
+
|
425 |
+
You are not required to accept this License in order to receive or
|
426 |
+
run a copy of the Program. Ancillary propagation of a covered work
|
427 |
+
occurring solely as a consequence of using peer-to-peer transmission
|
428 |
+
to receive a copy likewise does not require acceptance. However,
|
429 |
+
nothing other than this License grants you permission to propagate or
|
430 |
+
modify any covered work. These actions infringe copyright if you do
|
431 |
+
not accept this License. Therefore, by modifying or propagating a
|
432 |
+
covered work, you indicate your acceptance of this License to do so.
|
433 |
+
|
434 |
+
10. Automatic Licensing of Downstream Recipients.
|
435 |
+
|
436 |
+
Each time you convey a covered work, the recipient automatically
|
437 |
+
receives a license from the original licensors, to run, modify and
|
438 |
+
propagate that work, subject to this License. You are not responsible
|
439 |
+
for enforcing compliance by third parties with this License.
|
440 |
+
|
441 |
+
An "entity transaction" is a transaction transferring control of an
|
442 |
+
organization, or substantially all assets of one, or subdividing an
|
443 |
+
organization, or merging organizations. If propagation of a covered
|
444 |
+
work results from an entity transaction, each party to that
|
445 |
+
transaction who receives a copy of the work also receives whatever
|
446 |
+
licenses to the work the party's predecessor in interest had or could
|
447 |
+
give under the previous paragraph, plus a right to possession of the
|
448 |
+
Corresponding Source of the work from the predecessor in interest, if
|
449 |
+
the predecessor has it or can get it with reasonable efforts.
|
450 |
+
|
451 |
+
You may not impose any further restrictions on the exercise of the
|
452 |
+
rights granted or affirmed under this License. For example, you may
|
453 |
+
not impose a license fee, royalty, or other charge for exercise of
|
454 |
+
rights granted under this License, and you may not initiate litigation
|
455 |
+
(including a cross-claim or counterclaim in a lawsuit) alleging that
|
456 |
+
any patent claim is infringed by making, using, selling, offering for
|
457 |
+
sale, or importing the Program or any portion of it.
|
458 |
+
|
459 |
+
11. Patents.
|
460 |
+
|
461 |
+
A "contributor" is a copyright holder who authorizes use under this
|
462 |
+
License of the Program or a work on which the Program is based. The
|
463 |
+
work thus licensed is called the contributor's "contributor version".
|
464 |
+
|
465 |
+
A contributor's "essential patent claims" are all patent claims
|
466 |
+
owned or controlled by the contributor, whether already acquired or
|
467 |
+
hereafter acquired, that would be infringed by some manner, permitted
|
468 |
+
by this License, of making, using, or selling its contributor version,
|
469 |
+
but do not include claims that would be infringed only as a
|
470 |
+
consequence of further modification of the contributor version. For
|
471 |
+
purposes of this definition, "control" includes the right to grant
|
472 |
+
patent sublicenses in a manner consistent with the requirements of
|
473 |
+
this License.
|
474 |
+
|
475 |
+
Each contributor grants you a non-exclusive, worldwide, royalty-free
|
476 |
+
patent license under the contributor's essential patent claims, to
|
477 |
+
make, use, sell, offer for sale, import and otherwise run, modify and
|
478 |
+
propagate the contents of its contributor version.
|
479 |
+
|
480 |
+
In the following three paragraphs, a "patent license" is any express
|
481 |
+
agreement or commitment, however denominated, not to enforce a patent
|
482 |
+
(such as an express permission to practice a patent or covenant not to
|
483 |
+
sue for patent infringement). To "grant" such a patent license to a
|
484 |
+
party means to make such an agreement or commitment not to enforce a
|
485 |
+
patent against the party.
|
486 |
+
|
487 |
+
If you convey a covered work, knowingly relying on a patent license,
|
488 |
+
and the Corresponding Source of the work is not available for anyone
|
489 |
+
to copy, free of charge and under the terms of this License, through a
|
490 |
+
publicly available network server or other readily accessible means,
|
491 |
+
then you must either (1) cause the Corresponding Source to be so
|
492 |
+
available, or (2) arrange to deprive yourself of the benefit of the
|
493 |
+
patent license for this particular work, or (3) arrange, in a manner
|
494 |
+
consistent with the requirements of this License, to extend the patent
|
495 |
+
license to downstream recipients. "Knowingly relying" means you have
|
496 |
+
actual knowledge that, but for the patent license, your conveying the
|
497 |
+
covered work in a country, or your recipient's use of the covered work
|
498 |
+
in a country, would infringe one or more identifiable patents in that
|
499 |
+
country that you have reason to believe are valid.
|
500 |
+
|
501 |
+
If, pursuant to or in connection with a single transaction or
|
502 |
+
arrangement, you convey, or propagate by procuring conveyance of, a
|
503 |
+
covered work, and grant a patent license to some of the parties
|
504 |
+
receiving the covered work authorizing them to use, propagate, modify
|
505 |
+
or convey a specific copy of the covered work, then the patent license
|
506 |
+
you grant is automatically extended to all recipients of the covered
|
507 |
+
work and works based on it.
|
508 |
+
|
509 |
+
A patent license is "discriminatory" if it does not include within
|
510 |
+
the scope of its coverage, prohibits the exercise of, or is
|
511 |
+
conditioned on the non-exercise of one or more of the rights that are
|
512 |
+
specifically granted under this License. You may not convey a covered
|
513 |
+
work if you are a party to an arrangement with a third party that is
|
514 |
+
in the business of distributing software, under which you make payment
|
515 |
+
to the third party based on the extent of your activity of conveying
|
516 |
+
the work, and under which the third party grants, to any of the
|
517 |
+
parties who would receive the covered work from you, a discriminatory
|
518 |
+
patent license (a) in connection with copies of the covered work
|
519 |
+
conveyed by you (or copies made from those copies), or (b) primarily
|
520 |
+
for and in connection with specific products or compilations that
|
521 |
+
contain the covered work, unless you entered into that arrangement,
|
522 |
+
or that patent license was granted, prior to 28 March 2007.
|
523 |
+
|
524 |
+
Nothing in this License shall be construed as excluding or limiting
|
525 |
+
any implied license or other defenses to infringement that may
|
526 |
+
otherwise be available to you under applicable patent law.
|
527 |
+
|
528 |
+
12. No Surrender of Others' Freedom.
|
529 |
+
|
530 |
+
If conditions are imposed on you (whether by court order, agreement or
|
531 |
+
otherwise) that contradict the conditions of this License, they do not
|
532 |
+
excuse you from the conditions of this License. If you cannot convey a
|
533 |
+
covered work so as to satisfy simultaneously your obligations under this
|
534 |
+
License and any other pertinent obligations, then as a consequence you may
|
535 |
+
not convey it at all. For example, if you agree to terms that obligate you
|
536 |
+
to collect a royalty for further conveying from those to whom you convey
|
537 |
+
the Program, the only way you could satisfy both those terms and this
|
538 |
+
License would be to refrain entirely from conveying the Program.
|
539 |
+
|
540 |
+
13. Remote Network Interaction; Use with the GNU General Public License.
|
541 |
+
|
542 |
+
Notwithstanding any other provision of this License, if you modify the
|
543 |
+
Program, your modified version must prominently offer all users
|
544 |
+
interacting with it remotely through a computer network (if your version
|
545 |
+
supports such interaction) an opportunity to receive the Corresponding
|
546 |
+
Source of your version by providing access to the Corresponding Source
|
547 |
+
from a network server at no charge, through some standard or customary
|
548 |
+
means of facilitating copying of software. This Corresponding Source
|
549 |
+
shall include the Corresponding Source for any work covered by version 3
|
550 |
+
of the GNU General Public License that is incorporated pursuant to the
|
551 |
+
following paragraph.
|
552 |
+
|
553 |
+
Notwithstanding any other provision of this License, you have
|
554 |
+
permission to link or combine any covered work with a work licensed
|
555 |
+
under version 3 of the GNU General Public License into a single
|
556 |
+
combined work, and to convey the resulting work. The terms of this
|
557 |
+
License will continue to apply to the part which is the covered work,
|
558 |
+
but the work with which it is combined will remain governed by version
|
559 |
+
3 of the GNU General Public License.
|
560 |
+
|
561 |
+
14. Revised Versions of this License.
|
562 |
+
|
563 |
+
The Free Software Foundation may publish revised and/or new versions of
|
564 |
+
the GNU Affero General Public License from time to time. Such new versions
|
565 |
+
will be similar in spirit to the present version, but may differ in detail to
|
566 |
+
address new problems or concerns.
|
567 |
+
|
568 |
+
Each version is given a distinguishing version number. If the
|
569 |
+
Program specifies that a certain numbered version of the GNU Affero General
|
570 |
+
Public License "or any later version" applies to it, you have the
|
571 |
+
option of following the terms and conditions either of that numbered
|
572 |
+
version or of any later version published by the Free Software
|
573 |
+
Foundation. If the Program does not specify a version number of the
|
574 |
+
GNU Affero General Public License, you may choose any version ever published
|
575 |
+
by the Free Software Foundation.
|
576 |
+
|
577 |
+
If the Program specifies that a proxy can decide which future
|
578 |
+
versions of the GNU Affero General Public License can be used, that proxy's
|
579 |
+
public statement of acceptance of a version permanently authorizes you
|
580 |
+
to choose that version for the Program.
|
581 |
+
|
582 |
+
Later license versions may give you additional or different
|
583 |
+
permissions. However, no additional obligations are imposed on any
|
584 |
+
author or copyright holder as a result of your choosing to follow a
|
585 |
+
later version.
|
586 |
+
|
587 |
+
15. Disclaimer of Warranty.
|
588 |
+
|
589 |
+
THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY
|
590 |
+
APPLICABLE LAW. EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT
|
591 |
+
HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY
|
592 |
+
OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO,
|
593 |
+
THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
|
594 |
+
PURPOSE. THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM
|
595 |
+
IS WITH YOU. SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF
|
596 |
+
ALL NECESSARY SERVICING, REPAIR OR CORRECTION.
|
597 |
+
|
598 |
+
16. Limitation of Liability.
|
599 |
+
|
600 |
+
IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING
|
601 |
+
WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS
|
602 |
+
THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY
|
603 |
+
GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE
|
604 |
+
USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF
|
605 |
+
DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD
|
606 |
+
PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS),
|
607 |
+
EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF
|
608 |
+
SUCH DAMAGES.
|
609 |
+
|
610 |
+
17. Interpretation of Sections 15 and 16.
|
611 |
+
|
612 |
+
If the disclaimer of warranty and limitation of liability provided
|
613 |
+
above cannot be given local legal effect according to their terms,
|
614 |
+
reviewing courts shall apply local law that most closely approximates
|
615 |
+
an absolute waiver of all civil liability in connection with the
|
616 |
+
Program, unless a warranty or assumption of liability accompanies a
|
617 |
+
copy of the Program in return for a fee.
|
618 |
+
|
619 |
+
END OF TERMS AND CONDITIONS
|
620 |
+
|
621 |
+
How to Apply These Terms to Your New Programs
|
622 |
+
|
623 |
+
If you develop a new program, and you want it to be of the greatest
|
624 |
+
possible use to the public, the best way to achieve this is to make it
|
625 |
+
free software which everyone can redistribute and change under these terms.
|
626 |
+
|
627 |
+
To do so, attach the following notices to the program. It is safest
|
628 |
+
to attach them to the start of each source file to most effectively
|
629 |
+
state the exclusion of warranty; and each file should have at least
|
630 |
+
the "copyright" line and a pointer to where the full notice is found.
|
631 |
+
|
632 |
+
<one line to give the program's name and a brief idea of what it does.>
|
633 |
+
Copyright (C) <year> <name of author>
|
634 |
+
|
635 |
+
This program is free software: you can redistribute it and/or modify
|
636 |
+
it under the terms of the GNU Affero General Public License as published
|
637 |
+
by the Free Software Foundation, either version 3 of the License, or
|
638 |
+
(at your option) any later version.
|
639 |
+
|
640 |
+
This program is distributed in the hope that it will be useful,
|
641 |
+
but WITHOUT ANY WARRANTY; without even the implied warranty of
|
642 |
+
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
643 |
+
GNU Affero General Public License for more details.
|
644 |
+
|
645 |
+
You should have received a copy of the GNU Affero General Public License
|
646 |
+
along with this program. If not, see <https://www.gnu.org/licenses/>.
|
647 |
+
|
648 |
+
Also add information on how to contact you by electronic and paper mail.
|
649 |
+
|
650 |
+
If your software can interact with users remotely through a computer
|
651 |
+
network, you should also make sure that it provides a way for users to
|
652 |
+
get its source. For example, if your program is a web application, its
|
653 |
+
interface could display a "Source" link that leads users to an archive
|
654 |
+
of the code. There are many ways you could offer source, and different
|
655 |
+
solutions will be better for different programs; see section 13 for the
|
656 |
+
specific requirements.
|
657 |
+
|
658 |
+
You should also get your employer (if you work as a programmer) or school,
|
659 |
+
if any, to sign a "copyright disclaimer" for the program, if necessary.
|
660 |
+
For more information on this, and how to apply and follow the GNU AGPL, see
|
661 |
+
<https://www.gnu.org/licenses/>.
|
TTSs/genshin/README.md
ADDED
@@ -0,0 +1,42 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
<div align="center">
|
2 |
+
|
3 |
+
<img alt="LOGO" src="https://cdn.jsdelivr.net/gh/fishaudio/fish-diffusion@main/images/logo_512x512.png" width="256" height="256" />
|
4 |
+
|
5 |
+
# Bert-VITS2
|
6 |
+
|
7 |
+
VITS2 Backbone with multilingual bert
|
8 |
+
|
9 |
+
For quick guide, please refer to `webui_preprocess.py`.
|
10 |
+
|
11 |
+
简易教程请参见 `webui_preprocess.py`。
|
12 |
+
|
13 |
+
## 请注意,本项目核心思路来源于[anyvoiceai/MassTTS](https://github.com/anyvoiceai/MassTTS) 一个非常好的tts项目
|
14 |
+
## MassTTS的演示demo为[ai版峰哥锐评峰哥本人,并找回了在金三角失落的腰子](https://www.bilibili.com/video/BV1w24y1c7z9)
|
15 |
+
|
16 |
+
[//]: # (## 本项目与[PlayVoice/vits_chinese](https://github.com/PlayVoice/vits_chinese) 没有任何关系)
|
17 |
+
|
18 |
+
[//]: # ()
|
19 |
+
[//]: # (本仓库来源于之前朋友分享了ai峰哥的视频,本人被其中的效果惊艳,在自己尝试MassTTS以后发现fs在音质方面与vits有一定差距,并且training的pipeline比vits更复杂,因此按照其思路将bert)
|
20 |
+
|
21 |
+
## 成熟的旅行者/开拓者/舰长/博士/sensei/猎魔人/喵喵露/V应当参阅代码自己学习如何训练。
|
22 |
+
|
23 |
+
### 严禁将此项目用于一切违反《中华人民共和国宪法》,《中华人民共和国刑法》,《中华人民共和国治安管理处罚法》和《中华人民共和国民法典》之用途。
|
24 |
+
### 严禁用于任何政治相关用途。
|
25 |
+
#### Video:https://www.bilibili.com/video/BV1hp4y1K78E
|
26 |
+
#### Demo:https://www.bilibili.com/video/BV1TF411k78w
|
27 |
+
#### QQ Group:815818430
|
28 |
+
## References
|
29 |
+
+ [anyvoiceai/MassTTS](https://github.com/anyvoiceai/MassTTS)
|
30 |
+
+ [jaywalnut310/vits](https://github.com/jaywalnut310/vits)
|
31 |
+
+ [p0p4k/vits2_pytorch](https://github.com/p0p4k/vits2_pytorch)
|
32 |
+
+ [svc-develop-team/so-vits-svc](https://github.com/svc-develop-team/so-vits-svc)
|
33 |
+
+ [PaddlePaddle/PaddleSpeech](https://github.com/PaddlePaddle/PaddleSpeech)
|
34 |
+
+ [emotional-vits](https://github.com/innnky/emotional-vits)
|
35 |
+
+ [fish-speech](https://github.com/fishaudio/fish-speech)
|
36 |
+
+ [Bert-VITS2-UI](https://github.com/jiangyuxiaoxiao/Bert-VITS2-UI)
|
37 |
+
## 感谢所有贡献者作出的努力
|
38 |
+
<a href="https://github.com/fishaudio/Bert-VITS2/graphs/contributors" target="_blank">
|
39 |
+
<img src="https://contrib.rocks/image?repo=fishaudio/Bert-VITS2"/>
|
40 |
+
</a>
|
41 |
+
|
42 |
+
[//]: # (# 本项目所有代码引用均已写明,bert部分代码思路来源于[AI峰哥](https://www.bilibili.com/video/BV1w24y1c7z9),与[vits_chinese](https://github.com/PlayVoice/vits_chinese)无任何关系。欢迎各位查阅代码。同时,我们也对该开发者的[碰瓷,乃至开盒开发者的行为](https://www.bilibili.com/read/cv27101514/)表示强烈谴责。)
|
TTSs/genshin/__init__.py
ADDED
File without changes
|
TTSs/genshin/attentions.py
ADDED
@@ -0,0 +1,464 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import math
|
2 |
+
import torch
|
3 |
+
from torch import nn
|
4 |
+
from torch.nn import functional as F
|
5 |
+
|
6 |
+
import commons
|
7 |
+
import logging
|
8 |
+
|
9 |
+
logger = logging.getLogger(__name__)
|
10 |
+
|
11 |
+
|
12 |
+
class LayerNorm(nn.Module):
|
13 |
+
def __init__(self, channels, eps=1e-5):
|
14 |
+
super().__init__()
|
15 |
+
self.channels = channels
|
16 |
+
self.eps = eps
|
17 |
+
|
18 |
+
self.gamma = nn.Parameter(torch.ones(channels))
|
19 |
+
self.beta = nn.Parameter(torch.zeros(channels))
|
20 |
+
|
21 |
+
def forward(self, x):
|
22 |
+
x = x.transpose(1, -1)
|
23 |
+
x = F.layer_norm(x, (self.channels,), self.gamma, self.beta, self.eps)
|
24 |
+
return x.transpose(1, -1)
|
25 |
+
|
26 |
+
|
27 |
+
@torch.jit.script
|
28 |
+
def fused_add_tanh_sigmoid_multiply(input_a, input_b, n_channels):
|
29 |
+
n_channels_int = n_channels[0]
|
30 |
+
in_act = input_a + input_b
|
31 |
+
t_act = torch.tanh(in_act[:, :n_channels_int, :])
|
32 |
+
s_act = torch.sigmoid(in_act[:, n_channels_int:, :])
|
33 |
+
acts = t_act * s_act
|
34 |
+
return acts
|
35 |
+
|
36 |
+
|
37 |
+
class Encoder(nn.Module):
|
38 |
+
def __init__(
|
39 |
+
self,
|
40 |
+
hidden_channels,
|
41 |
+
filter_channels,
|
42 |
+
n_heads,
|
43 |
+
n_layers,
|
44 |
+
kernel_size=1,
|
45 |
+
p_dropout=0.0,
|
46 |
+
window_size=4,
|
47 |
+
isflow=True,
|
48 |
+
**kwargs
|
49 |
+
):
|
50 |
+
super().__init__()
|
51 |
+
self.hidden_channels = hidden_channels
|
52 |
+
self.filter_channels = filter_channels
|
53 |
+
self.n_heads = n_heads
|
54 |
+
self.n_layers = n_layers
|
55 |
+
self.kernel_size = kernel_size
|
56 |
+
self.p_dropout = p_dropout
|
57 |
+
self.window_size = window_size
|
58 |
+
# if isflow:
|
59 |
+
# cond_layer = torch.nn.Conv1d(256, 2*hidden_channels*n_layers, 1)
|
60 |
+
# self.cond_pre = torch.nn.Conv1d(hidden_channels, 2*hidden_channels, 1)
|
61 |
+
# self.cond_layer = weight_norm(cond_layer, name='weight')
|
62 |
+
# self.gin_channels = 256
|
63 |
+
self.cond_layer_idx = self.n_layers
|
64 |
+
if "gin_channels" in kwargs:
|
65 |
+
self.gin_channels = kwargs["gin_channels"]
|
66 |
+
if self.gin_channels != 0:
|
67 |
+
self.spk_emb_linear = nn.Linear(self.gin_channels, self.hidden_channels)
|
68 |
+
# vits2 says 3rd block, so idx is 2 by default
|
69 |
+
self.cond_layer_idx = (
|
70 |
+
kwargs["cond_layer_idx"] if "cond_layer_idx" in kwargs else 2
|
71 |
+
)
|
72 |
+
logging.debug(self.gin_channels, self.cond_layer_idx)
|
73 |
+
assert (
|
74 |
+
self.cond_layer_idx < self.n_layers
|
75 |
+
), "cond_layer_idx should be less than n_layers"
|
76 |
+
self.drop = nn.Dropout(p_dropout)
|
77 |
+
self.attn_layers = nn.ModuleList()
|
78 |
+
self.norm_layers_1 = nn.ModuleList()
|
79 |
+
self.ffn_layers = nn.ModuleList()
|
80 |
+
self.norm_layers_2 = nn.ModuleList()
|
81 |
+
for i in range(self.n_layers):
|
82 |
+
self.attn_layers.append(
|
83 |
+
MultiHeadAttention(
|
84 |
+
hidden_channels,
|
85 |
+
hidden_channels,
|
86 |
+
n_heads,
|
87 |
+
p_dropout=p_dropout,
|
88 |
+
window_size=window_size,
|
89 |
+
)
|
90 |
+
)
|
91 |
+
self.norm_layers_1.append(LayerNorm(hidden_channels))
|
92 |
+
self.ffn_layers.append(
|
93 |
+
FFN(
|
94 |
+
hidden_channels,
|
95 |
+
hidden_channels,
|
96 |
+
filter_channels,
|
97 |
+
kernel_size,
|
98 |
+
p_dropout=p_dropout,
|
99 |
+
)
|
100 |
+
)
|
101 |
+
self.norm_layers_2.append(LayerNorm(hidden_channels))
|
102 |
+
|
103 |
+
def forward(self, x, x_mask, g=None):
|
104 |
+
attn_mask = x_mask.unsqueeze(2) * x_mask.unsqueeze(-1)
|
105 |
+
x = x * x_mask
|
106 |
+
for i in range(self.n_layers):
|
107 |
+
if i == self.cond_layer_idx and g is not None:
|
108 |
+
g = self.spk_emb_linear(g.transpose(1, 2))
|
109 |
+
g = g.transpose(1, 2)
|
110 |
+
x = x + g
|
111 |
+
x = x * x_mask
|
112 |
+
y = self.attn_layers[i](x, x, attn_mask)
|
113 |
+
y = self.drop(y)
|
114 |
+
x = self.norm_layers_1[i](x + y)
|
115 |
+
|
116 |
+
y = self.ffn_layers[i](x, x_mask)
|
117 |
+
y = self.drop(y)
|
118 |
+
x = self.norm_layers_2[i](x + y)
|
119 |
+
x = x * x_mask
|
120 |
+
return x
|
121 |
+
|
122 |
+
|
123 |
+
class Decoder(nn.Module):
|
124 |
+
def __init__(
|
125 |
+
self,
|
126 |
+
hidden_channels,
|
127 |
+
filter_channels,
|
128 |
+
n_heads,
|
129 |
+
n_layers,
|
130 |
+
kernel_size=1,
|
131 |
+
p_dropout=0.0,
|
132 |
+
proximal_bias=False,
|
133 |
+
proximal_init=True,
|
134 |
+
**kwargs
|
135 |
+
):
|
136 |
+
super().__init__()
|
137 |
+
self.hidden_channels = hidden_channels
|
138 |
+
self.filter_channels = filter_channels
|
139 |
+
self.n_heads = n_heads
|
140 |
+
self.n_layers = n_layers
|
141 |
+
self.kernel_size = kernel_size
|
142 |
+
self.p_dropout = p_dropout
|
143 |
+
self.proximal_bias = proximal_bias
|
144 |
+
self.proximal_init = proximal_init
|
145 |
+
|
146 |
+
self.drop = nn.Dropout(p_dropout)
|
147 |
+
self.self_attn_layers = nn.ModuleList()
|
148 |
+
self.norm_layers_0 = nn.ModuleList()
|
149 |
+
self.encdec_attn_layers = nn.ModuleList()
|
150 |
+
self.norm_layers_1 = nn.ModuleList()
|
151 |
+
self.ffn_layers = nn.ModuleList()
|
152 |
+
self.norm_layers_2 = nn.ModuleList()
|
153 |
+
for i in range(self.n_layers):
|
154 |
+
self.self_attn_layers.append(
|
155 |
+
MultiHeadAttention(
|
156 |
+
hidden_channels,
|
157 |
+
hidden_channels,
|
158 |
+
n_heads,
|
159 |
+
p_dropout=p_dropout,
|
160 |
+
proximal_bias=proximal_bias,
|
161 |
+
proximal_init=proximal_init,
|
162 |
+
)
|
163 |
+
)
|
164 |
+
self.norm_layers_0.append(LayerNorm(hidden_channels))
|
165 |
+
self.encdec_attn_layers.append(
|
166 |
+
MultiHeadAttention(
|
167 |
+
hidden_channels, hidden_channels, n_heads, p_dropout=p_dropout
|
168 |
+
)
|
169 |
+
)
|
170 |
+
self.norm_layers_1.append(LayerNorm(hidden_channels))
|
171 |
+
self.ffn_layers.append(
|
172 |
+
FFN(
|
173 |
+
hidden_channels,
|
174 |
+
hidden_channels,
|
175 |
+
filter_channels,
|
176 |
+
kernel_size,
|
177 |
+
p_dropout=p_dropout,
|
178 |
+
causal=True,
|
179 |
+
)
|
180 |
+
)
|
181 |
+
self.norm_layers_2.append(LayerNorm(hidden_channels))
|
182 |
+
|
183 |
+
def forward(self, x, x_mask, h, h_mask):
|
184 |
+
"""
|
185 |
+
x: decoder input
|
186 |
+
h: encoder output
|
187 |
+
"""
|
188 |
+
self_attn_mask = commons.subsequent_mask(x_mask.size(2)).to(
|
189 |
+
device=x.device, dtype=x.dtype
|
190 |
+
)
|
191 |
+
encdec_attn_mask = h_mask.unsqueeze(2) * x_mask.unsqueeze(-1)
|
192 |
+
x = x * x_mask
|
193 |
+
for i in range(self.n_layers):
|
194 |
+
y = self.self_attn_layers[i](x, x, self_attn_mask)
|
195 |
+
y = self.drop(y)
|
196 |
+
x = self.norm_layers_0[i](x + y)
|
197 |
+
|
198 |
+
y = self.encdec_attn_layers[i](x, h, encdec_attn_mask)
|
199 |
+
y = self.drop(y)
|
200 |
+
x = self.norm_layers_1[i](x + y)
|
201 |
+
|
202 |
+
y = self.ffn_layers[i](x, x_mask)
|
203 |
+
y = self.drop(y)
|
204 |
+
x = self.norm_layers_2[i](x + y)
|
205 |
+
x = x * x_mask
|
206 |
+
return x
|
207 |
+
|
208 |
+
|
209 |
+
class MultiHeadAttention(nn.Module):
|
210 |
+
def __init__(
|
211 |
+
self,
|
212 |
+
channels,
|
213 |
+
out_channels,
|
214 |
+
n_heads,
|
215 |
+
p_dropout=0.0,
|
216 |
+
window_size=None,
|
217 |
+
heads_share=True,
|
218 |
+
block_length=None,
|
219 |
+
proximal_bias=False,
|
220 |
+
proximal_init=False,
|
221 |
+
):
|
222 |
+
super().__init__()
|
223 |
+
assert channels % n_heads == 0
|
224 |
+
|
225 |
+
self.channels = channels
|
226 |
+
self.out_channels = out_channels
|
227 |
+
self.n_heads = n_heads
|
228 |
+
self.p_dropout = p_dropout
|
229 |
+
self.window_size = window_size
|
230 |
+
self.heads_share = heads_share
|
231 |
+
self.block_length = block_length
|
232 |
+
self.proximal_bias = proximal_bias
|
233 |
+
self.proximal_init = proximal_init
|
234 |
+
self.attn = None
|
235 |
+
|
236 |
+
self.k_channels = channels // n_heads
|
237 |
+
self.conv_q = nn.Conv1d(channels, channels, 1)
|
238 |
+
self.conv_k = nn.Conv1d(channels, channels, 1)
|
239 |
+
self.conv_v = nn.Conv1d(channels, channels, 1)
|
240 |
+
self.conv_o = nn.Conv1d(channels, out_channels, 1)
|
241 |
+
self.drop = nn.Dropout(p_dropout)
|
242 |
+
|
243 |
+
if window_size is not None:
|
244 |
+
n_heads_rel = 1 if heads_share else n_heads
|
245 |
+
rel_stddev = self.k_channels**-0.5
|
246 |
+
self.emb_rel_k = nn.Parameter(
|
247 |
+
torch.randn(n_heads_rel, window_size * 2 + 1, self.k_channels)
|
248 |
+
* rel_stddev
|
249 |
+
)
|
250 |
+
self.emb_rel_v = nn.Parameter(
|
251 |
+
torch.randn(n_heads_rel, window_size * 2 + 1, self.k_channels)
|
252 |
+
* rel_stddev
|
253 |
+
)
|
254 |
+
|
255 |
+
nn.init.xavier_uniform_(self.conv_q.weight)
|
256 |
+
nn.init.xavier_uniform_(self.conv_k.weight)
|
257 |
+
nn.init.xavier_uniform_(self.conv_v.weight)
|
258 |
+
if proximal_init:
|
259 |
+
with torch.no_grad():
|
260 |
+
self.conv_k.weight.copy_(self.conv_q.weight)
|
261 |
+
self.conv_k.bias.copy_(self.conv_q.bias)
|
262 |
+
|
263 |
+
def forward(self, x, c, attn_mask=None):
|
264 |
+
q = self.conv_q(x)
|
265 |
+
k = self.conv_k(c)
|
266 |
+
v = self.conv_v(c)
|
267 |
+
|
268 |
+
x, self.attn = self.attention(q, k, v, mask=attn_mask)
|
269 |
+
|
270 |
+
x = self.conv_o(x)
|
271 |
+
return x
|
272 |
+
|
273 |
+
def attention(self, query, key, value, mask=None):
|
274 |
+
# reshape [b, d, t] -> [b, n_h, t, d_k]
|
275 |
+
b, d, t_s, t_t = (*key.size(), query.size(2))
|
276 |
+
query = query.view(b, self.n_heads, self.k_channels, t_t).transpose(2, 3)
|
277 |
+
key = key.view(b, self.n_heads, self.k_channels, t_s).transpose(2, 3)
|
278 |
+
value = value.view(b, self.n_heads, self.k_channels, t_s).transpose(2, 3)
|
279 |
+
|
280 |
+
scores = torch.matmul(query / math.sqrt(self.k_channels), key.transpose(-2, -1))
|
281 |
+
if self.window_size is not None:
|
282 |
+
assert (
|
283 |
+
t_s == t_t
|
284 |
+
), "Relative attention is only available for self-attention."
|
285 |
+
key_relative_embeddings = self._get_relative_embeddings(self.emb_rel_k, t_s)
|
286 |
+
rel_logits = self._matmul_with_relative_keys(
|
287 |
+
query / math.sqrt(self.k_channels), key_relative_embeddings
|
288 |
+
)
|
289 |
+
scores_local = self._relative_position_to_absolute_position(rel_logits)
|
290 |
+
scores = scores + scores_local
|
291 |
+
if self.proximal_bias:
|
292 |
+
assert t_s == t_t, "Proximal bias is only available for self-attention."
|
293 |
+
scores = scores + self._attention_bias_proximal(t_s).to(
|
294 |
+
device=scores.device, dtype=scores.dtype
|
295 |
+
)
|
296 |
+
if mask is not None:
|
297 |
+
scores = scores.masked_fill(mask == 0, -1e4)
|
298 |
+
if self.block_length is not None:
|
299 |
+
assert (
|
300 |
+
t_s == t_t
|
301 |
+
), "Local attention is only available for self-attention."
|
302 |
+
block_mask = (
|
303 |
+
torch.ones_like(scores)
|
304 |
+
.triu(-self.block_length)
|
305 |
+
.tril(self.block_length)
|
306 |
+
)
|
307 |
+
scores = scores.masked_fill(block_mask == 0, -1e4)
|
308 |
+
p_attn = F.softmax(scores, dim=-1) # [b, n_h, t_t, t_s]
|
309 |
+
p_attn = self.drop(p_attn)
|
310 |
+
output = torch.matmul(p_attn, value)
|
311 |
+
if self.window_size is not None:
|
312 |
+
relative_weights = self._absolute_position_to_relative_position(p_attn)
|
313 |
+
value_relative_embeddings = self._get_relative_embeddings(
|
314 |
+
self.emb_rel_v, t_s
|
315 |
+
)
|
316 |
+
output = output + self._matmul_with_relative_values(
|
317 |
+
relative_weights, value_relative_embeddings
|
318 |
+
)
|
319 |
+
output = (
|
320 |
+
output.transpose(2, 3).contiguous().view(b, d, t_t)
|
321 |
+
) # [b, n_h, t_t, d_k] -> [b, d, t_t]
|
322 |
+
return output, p_attn
|
323 |
+
|
324 |
+
def _matmul_with_relative_values(self, x, y):
|
325 |
+
"""
|
326 |
+
x: [b, h, l, m]
|
327 |
+
y: [h or 1, m, d]
|
328 |
+
ret: [b, h, l, d]
|
329 |
+
"""
|
330 |
+
ret = torch.matmul(x, y.unsqueeze(0))
|
331 |
+
return ret
|
332 |
+
|
333 |
+
def _matmul_with_relative_keys(self, x, y):
|
334 |
+
"""
|
335 |
+
x: [b, h, l, d]
|
336 |
+
y: [h or 1, m, d]
|
337 |
+
ret: [b, h, l, m]
|
338 |
+
"""
|
339 |
+
ret = torch.matmul(x, y.unsqueeze(0).transpose(-2, -1))
|
340 |
+
return ret
|
341 |
+
|
342 |
+
def _get_relative_embeddings(self, relative_embeddings, length):
|
343 |
+
2 * self.window_size + 1
|
344 |
+
# Pad first before slice to avoid using cond ops.
|
345 |
+
pad_length = max(length - (self.window_size + 1), 0)
|
346 |
+
slice_start_position = max((self.window_size + 1) - length, 0)
|
347 |
+
slice_end_position = slice_start_position + 2 * length - 1
|
348 |
+
if pad_length > 0:
|
349 |
+
padded_relative_embeddings = F.pad(
|
350 |
+
relative_embeddings,
|
351 |
+
commons.convert_pad_shape([[0, 0], [pad_length, pad_length], [0, 0]]),
|
352 |
+
)
|
353 |
+
else:
|
354 |
+
padded_relative_embeddings = relative_embeddings
|
355 |
+
used_relative_embeddings = padded_relative_embeddings[
|
356 |
+
:, slice_start_position:slice_end_position
|
357 |
+
]
|
358 |
+
return used_relative_embeddings
|
359 |
+
|
360 |
+
def _relative_position_to_absolute_position(self, x):
|
361 |
+
"""
|
362 |
+
x: [b, h, l, 2*l-1]
|
363 |
+
ret: [b, h, l, l]
|
364 |
+
"""
|
365 |
+
batch, heads, length, _ = x.size()
|
366 |
+
# Concat columns of pad to shift from relative to absolute indexing.
|
367 |
+
x = F.pad(x, commons.convert_pad_shape([[0, 0], [0, 0], [0, 0], [0, 1]]))
|
368 |
+
|
369 |
+
# Concat extra elements so to add up to shape (len+1, 2*len-1).
|
370 |
+
x_flat = x.view([batch, heads, length * 2 * length])
|
371 |
+
x_flat = F.pad(
|
372 |
+
x_flat, commons.convert_pad_shape([[0, 0], [0, 0], [0, length - 1]])
|
373 |
+
)
|
374 |
+
|
375 |
+
# Reshape and slice out the padded elements.
|
376 |
+
x_final = x_flat.view([batch, heads, length + 1, 2 * length - 1])[
|
377 |
+
:, :, :length, length - 1 :
|
378 |
+
]
|
379 |
+
return x_final
|
380 |
+
|
381 |
+
def _absolute_position_to_relative_position(self, x):
|
382 |
+
"""
|
383 |
+
x: [b, h, l, l]
|
384 |
+
ret: [b, h, l, 2*l-1]
|
385 |
+
"""
|
386 |
+
batch, heads, length, _ = x.size()
|
387 |
+
# pad along column
|
388 |
+
x = F.pad(
|
389 |
+
x, commons.convert_pad_shape([[0, 0], [0, 0], [0, 0], [0, length - 1]])
|
390 |
+
)
|
391 |
+
x_flat = x.view([batch, heads, length**2 + length * (length - 1)])
|
392 |
+
# add 0's in the beginning that will skew the elements after reshape
|
393 |
+
x_flat = F.pad(x_flat, commons.convert_pad_shape([[0, 0], [0, 0], [length, 0]]))
|
394 |
+
x_final = x_flat.view([batch, heads, length, 2 * length])[:, :, :, 1:]
|
395 |
+
return x_final
|
396 |
+
|
397 |
+
def _attention_bias_proximal(self, length):
|
398 |
+
"""Bias for self-attention to encourage attention to close positions.
|
399 |
+
Args:
|
400 |
+
length: an integer scalar.
|
401 |
+
Returns:
|
402 |
+
a Tensor with shape [1, 1, length, length]
|
403 |
+
"""
|
404 |
+
r = torch.arange(length, dtype=torch.float32)
|
405 |
+
diff = torch.unsqueeze(r, 0) - torch.unsqueeze(r, 1)
|
406 |
+
return torch.unsqueeze(torch.unsqueeze(-torch.log1p(torch.abs(diff)), 0), 0)
|
407 |
+
|
408 |
+
|
409 |
+
class FFN(nn.Module):
|
410 |
+
def __init__(
|
411 |
+
self,
|
412 |
+
in_channels,
|
413 |
+
out_channels,
|
414 |
+
filter_channels,
|
415 |
+
kernel_size,
|
416 |
+
p_dropout=0.0,
|
417 |
+
activation=None,
|
418 |
+
causal=False,
|
419 |
+
):
|
420 |
+
super().__init__()
|
421 |
+
self.in_channels = in_channels
|
422 |
+
self.out_channels = out_channels
|
423 |
+
self.filter_channels = filter_channels
|
424 |
+
self.kernel_size = kernel_size
|
425 |
+
self.p_dropout = p_dropout
|
426 |
+
self.activation = activation
|
427 |
+
self.causal = causal
|
428 |
+
|
429 |
+
if causal:
|
430 |
+
self.padding = self._causal_padding
|
431 |
+
else:
|
432 |
+
self.padding = self._same_padding
|
433 |
+
|
434 |
+
self.conv_1 = nn.Conv1d(in_channels, filter_channels, kernel_size)
|
435 |
+
self.conv_2 = nn.Conv1d(filter_channels, out_channels, kernel_size)
|
436 |
+
self.drop = nn.Dropout(p_dropout)
|
437 |
+
|
438 |
+
def forward(self, x, x_mask):
|
439 |
+
x = self.conv_1(self.padding(x * x_mask))
|
440 |
+
if self.activation == "gelu":
|
441 |
+
x = x * torch.sigmoid(1.702 * x)
|
442 |
+
else:
|
443 |
+
x = torch.relu(x)
|
444 |
+
x = self.drop(x)
|
445 |
+
x = self.conv_2(self.padding(x * x_mask))
|
446 |
+
return x * x_mask
|
447 |
+
|
448 |
+
def _causal_padding(self, x):
|
449 |
+
if self.kernel_size == 1:
|
450 |
+
return x
|
451 |
+
pad_l = self.kernel_size - 1
|
452 |
+
pad_r = 0
|
453 |
+
padding = [[0, 0], [0, 0], [pad_l, pad_r]]
|
454 |
+
x = F.pad(x, commons.convert_pad_shape(padding))
|
455 |
+
return x
|
456 |
+
|
457 |
+
def _same_padding(self, x):
|
458 |
+
if self.kernel_size == 1:
|
459 |
+
return x
|
460 |
+
pad_l = (self.kernel_size - 1) // 2
|
461 |
+
pad_r = self.kernel_size // 2
|
462 |
+
padding = [[0, 0], [0, 0], [pad_l, pad_r]]
|
463 |
+
x = F.pad(x, commons.convert_pad_shape(padding))
|
464 |
+
return x
|
TTSs/genshin/bert/bert-base-japanese-v3/.gitattributes
ADDED
@@ -0,0 +1,34 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
*.7z filter=lfs diff=lfs merge=lfs -text
|
2 |
+
*.arrow filter=lfs diff=lfs merge=lfs -text
|
3 |
+
*.bin filter=lfs diff=lfs merge=lfs -text
|
4 |
+
*.bz2 filter=lfs diff=lfs merge=lfs -text
|
5 |
+
*.ckpt filter=lfs diff=lfs merge=lfs -text
|
6 |
+
*.ftz filter=lfs diff=lfs merge=lfs -text
|
7 |
+
*.gz filter=lfs diff=lfs merge=lfs -text
|
8 |
+
*.h5 filter=lfs diff=lfs merge=lfs -text
|
9 |
+
*.joblib filter=lfs diff=lfs merge=lfs -text
|
10 |
+
*.lfs.* filter=lfs diff=lfs merge=lfs -text
|
11 |
+
*.mlmodel filter=lfs diff=lfs merge=lfs -text
|
12 |
+
*.model filter=lfs diff=lfs merge=lfs -text
|
13 |
+
*.msgpack filter=lfs diff=lfs merge=lfs -text
|
14 |
+
*.npy filter=lfs diff=lfs merge=lfs -text
|
15 |
+
*.npz filter=lfs diff=lfs merge=lfs -text
|
16 |
+
*.onnx filter=lfs diff=lfs merge=lfs -text
|
17 |
+
*.ot filter=lfs diff=lfs merge=lfs -text
|
18 |
+
*.parquet filter=lfs diff=lfs merge=lfs -text
|
19 |
+
*.pb filter=lfs diff=lfs merge=lfs -text
|
20 |
+
*.pickle filter=lfs diff=lfs merge=lfs -text
|
21 |
+
*.pkl filter=lfs diff=lfs merge=lfs -text
|
22 |
+
*.pt filter=lfs diff=lfs merge=lfs -text
|
23 |
+
*.pth filter=lfs diff=lfs merge=lfs -text
|
24 |
+
*.rar filter=lfs diff=lfs merge=lfs -text
|
25 |
+
*.safetensors filter=lfs diff=lfs merge=lfs -text
|
26 |
+
saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
27 |
+
*.tar.* filter=lfs diff=lfs merge=lfs -text
|
28 |
+
*.tflite filter=lfs diff=lfs merge=lfs -text
|
29 |
+
*.tgz filter=lfs diff=lfs merge=lfs -text
|
30 |
+
*.wasm filter=lfs diff=lfs merge=lfs -text
|
31 |
+
*.xz filter=lfs diff=lfs merge=lfs -text
|
32 |
+
*.zip filter=lfs diff=lfs merge=lfs -text
|
33 |
+
*.zst filter=lfs diff=lfs merge=lfs -text
|
34 |
+
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
TTSs/genshin/bert/bert-base-japanese-v3/README.md
ADDED
@@ -0,0 +1,53 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
license: apache-2.0
|
3 |
+
datasets:
|
4 |
+
- cc100
|
5 |
+
- wikipedia
|
6 |
+
language:
|
7 |
+
- ja
|
8 |
+
widget:
|
9 |
+
- text: 東北大学で[MASK]の研究をしています。
|
10 |
+
---
|
11 |
+
|
12 |
+
# BERT base Japanese (unidic-lite with whole word masking, CC-100 and jawiki-20230102)
|
13 |
+
|
14 |
+
This is a [BERT](https://github.com/google-research/bert) model pretrained on texts in the Japanese language.
|
15 |
+
|
16 |
+
This version of the model processes input texts with word-level tokenization based on the Unidic 2.1.2 dictionary (available in [unidic-lite](https://pypi.org/project/unidic-lite/) package), followed by the WordPiece subword tokenization.
|
17 |
+
Additionally, the model is trained with the whole word masking enabled for the masked language modeling (MLM) objective.
|
18 |
+
|
19 |
+
The codes for the pretraining are available at [cl-tohoku/bert-japanese](https://github.com/cl-tohoku/bert-japanese/).
|
20 |
+
|
21 |
+
## Model architecture
|
22 |
+
|
23 |
+
The model architecture is the same as the original BERT base model; 12 layers, 768 dimensions of hidden states, and 12 attention heads.
|
24 |
+
|
25 |
+
## Training Data
|
26 |
+
|
27 |
+
The model is trained on the Japanese portion of [CC-100 dataset](https://data.statmt.org/cc-100/) and the Japanese version of Wikipedia.
|
28 |
+
For Wikipedia, we generated a text corpus from the [Wikipedia Cirrussearch dump file](https://dumps.wikimedia.org/other/cirrussearch/) as of January 2, 2023.
|
29 |
+
The corpus files generated from CC-100 and Wikipedia are 74.3GB and 4.9GB in size and consist of approximately 392M and 34M sentences, respectively.
|
30 |
+
|
31 |
+
For the purpose of splitting texts into sentences, we used [fugashi](https://github.com/polm/fugashi) with [mecab-ipadic-NEologd](https://github.com/neologd/mecab-ipadic-neologd) dictionary (v0.0.7).
|
32 |
+
|
33 |
+
## Tokenization
|
34 |
+
|
35 |
+
The texts are first tokenized by MeCab with the Unidic 2.1.2 dictionary and then split into subwords by the WordPiece algorithm.
|
36 |
+
The vocabulary size is 32768.
|
37 |
+
|
38 |
+
We used [fugashi](https://github.com/polm/fugashi) and [unidic-lite](https://github.com/polm/unidic-lite) packages for the tokenization.
|
39 |
+
|
40 |
+
## Training
|
41 |
+
|
42 |
+
We trained the model first on the CC-100 corpus for 1M steps and then on the Wikipedia corpus for another 1M steps.
|
43 |
+
For training of the MLM (masked language modeling) objective, we introduced whole word masking in which all of the subword tokens corresponding to a single word (tokenized by MeCab) are masked at once.
|
44 |
+
|
45 |
+
For training of each model, we used a v3-8 instance of Cloud TPUs provided by [TPU Research Cloud](https://sites.research.google/trc/about/).
|
46 |
+
|
47 |
+
## Licenses
|
48 |
+
|
49 |
+
The pretrained models are distributed under the Apache License 2.0.
|
50 |
+
|
51 |
+
## Acknowledgments
|
52 |
+
|
53 |
+
This model is trained with Cloud TPUs provided by [TPU Research Cloud](https://sites.research.google/trc/about/) program.
|
TTSs/genshin/bert/bert-base-japanese-v3/config.json
ADDED
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"architectures": [
|
3 |
+
"BertForPreTraining"
|
4 |
+
],
|
5 |
+
"attention_probs_dropout_prob": 0.1,
|
6 |
+
"hidden_act": "gelu",
|
7 |
+
"hidden_dropout_prob": 0.1,
|
8 |
+
"hidden_size": 768,
|
9 |
+
"initializer_range": 0.02,
|
10 |
+
"intermediate_size": 3072,
|
11 |
+
"layer_norm_eps": 1e-12,
|
12 |
+
"max_position_embeddings": 512,
|
13 |
+
"model_type": "bert",
|
14 |
+
"num_attention_heads": 12,
|
15 |
+
"num_hidden_layers": 12,
|
16 |
+
"pad_token_id": 0,
|
17 |
+
"type_vocab_size": 2,
|
18 |
+
"vocab_size": 32768
|
19 |
+
}
|
TTSs/genshin/bert/bert-base-japanese-v3/tokenizer_config.json
ADDED
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"tokenizer_class": "BertJapaneseTokenizer",
|
3 |
+
"model_max_length": 512,
|
4 |
+
"do_lower_case": false,
|
5 |
+
"word_tokenizer_type": "mecab",
|
6 |
+
"subword_tokenizer_type": "wordpiece",
|
7 |
+
"mecab_kwargs": {
|
8 |
+
"mecab_dic": "unidic_lite"
|
9 |
+
}
|
10 |
+
}
|
TTSs/genshin/bert/bert-base-japanese-v3/vocab.txt
ADDED
The diff for this file is too large to render.
See raw diff
|
|
TTSs/genshin/bert/bert-large-japanese-v2/.gitattributes
ADDED
@@ -0,0 +1,34 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
*.7z filter=lfs diff=lfs merge=lfs -text
|
2 |
+
*.arrow filter=lfs diff=lfs merge=lfs -text
|
3 |
+
*.bin filter=lfs diff=lfs merge=lfs -text
|
4 |
+
*.bz2 filter=lfs diff=lfs merge=lfs -text
|
5 |
+
*.ckpt filter=lfs diff=lfs merge=lfs -text
|
6 |
+
*.ftz filter=lfs diff=lfs merge=lfs -text
|
7 |
+
*.gz filter=lfs diff=lfs merge=lfs -text
|
8 |
+
*.h5 filter=lfs diff=lfs merge=lfs -text
|
9 |
+
*.joblib filter=lfs diff=lfs merge=lfs -text
|
10 |
+
*.lfs.* filter=lfs diff=lfs merge=lfs -text
|
11 |
+
*.mlmodel filter=lfs diff=lfs merge=lfs -text
|
12 |
+
*.model filter=lfs diff=lfs merge=lfs -text
|
13 |
+
*.msgpack filter=lfs diff=lfs merge=lfs -text
|
14 |
+
*.npy filter=lfs diff=lfs merge=lfs -text
|
15 |
+
*.npz filter=lfs diff=lfs merge=lfs -text
|
16 |
+
*.onnx filter=lfs diff=lfs merge=lfs -text
|
17 |
+
*.ot filter=lfs diff=lfs merge=lfs -text
|
18 |
+
*.parquet filter=lfs diff=lfs merge=lfs -text
|
19 |
+
*.pb filter=lfs diff=lfs merge=lfs -text
|
20 |
+
*.pickle filter=lfs diff=lfs merge=lfs -text
|
21 |
+
*.pkl filter=lfs diff=lfs merge=lfs -text
|
22 |
+
*.pt filter=lfs diff=lfs merge=lfs -text
|
23 |
+
*.pth filter=lfs diff=lfs merge=lfs -text
|
24 |
+
*.rar filter=lfs diff=lfs merge=lfs -text
|
25 |
+
*.safetensors filter=lfs diff=lfs merge=lfs -text
|
26 |
+
saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
27 |
+
*.tar.* filter=lfs diff=lfs merge=lfs -text
|
28 |
+
*.tflite filter=lfs diff=lfs merge=lfs -text
|
29 |
+
*.tgz filter=lfs diff=lfs merge=lfs -text
|
30 |
+
*.wasm filter=lfs diff=lfs merge=lfs -text
|
31 |
+
*.xz filter=lfs diff=lfs merge=lfs -text
|
32 |
+
*.zip filter=lfs diff=lfs merge=lfs -text
|
33 |
+
*.zst filter=lfs diff=lfs merge=lfs -text
|
34 |
+
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
TTSs/genshin/bert/bert-large-japanese-v2/README.md
ADDED
@@ -0,0 +1,53 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
license: apache-2.0
|
3 |
+
datasets:
|
4 |
+
- cc100
|
5 |
+
- wikipedia
|
6 |
+
language:
|
7 |
+
- ja
|
8 |
+
widget:
|
9 |
+
- text: 東北大学で[MASK]の研究をしています。
|
10 |
+
---
|
11 |
+
|
12 |
+
# BERT large Japanese (unidic-lite with whole word masking, CC-100 and jawiki-20230102)
|
13 |
+
|
14 |
+
This is a [BERT](https://github.com/google-research/bert) model pretrained on texts in the Japanese language.
|
15 |
+
|
16 |
+
This version of the model processes input texts with word-level tokenization based on the Unidic 2.1.2 dictionary (available in [unidic-lite](https://pypi.org/project/unidic-lite/) package), followed by the WordPiece subword tokenization.
|
17 |
+
Additionally, the model is trained with the whole word masking enabled for the masked language modeling (MLM) objective.
|
18 |
+
|
19 |
+
The codes for the pretraining are available at [cl-tohoku/bert-japanese](https://github.com/cl-tohoku/bert-japanese/).
|
20 |
+
|
21 |
+
## Model architecture
|
22 |
+
|
23 |
+
The model architecture is the same as the original BERT large model; 24 layers, 1024 dimensions of hidden states, and 16 attention heads.
|
24 |
+
|
25 |
+
## Training Data
|
26 |
+
|
27 |
+
The model is trained on the Japanese portion of [CC-100 dataset](https://data.statmt.org/cc-100/) and the Japanese version of Wikipedia.
|
28 |
+
For Wikipedia, we generated a text corpus from the [Wikipedia Cirrussearch dump file](https://dumps.wikimedia.org/other/cirrussearch/) as of January 2, 2023.
|
29 |
+
The corpus files generated from CC-100 and Wikipedia are 74.3GB and 4.9GB in size and consist of approximately 392M and 34M sentences, respectively.
|
30 |
+
|
31 |
+
For the purpose of splitting texts into sentences, we used [fugashi](https://github.com/polm/fugashi) with [mecab-ipadic-NEologd](https://github.com/neologd/mecab-ipadic-neologd) dictionary (v0.0.7).
|
32 |
+
|
33 |
+
## Tokenization
|
34 |
+
|
35 |
+
The texts are first tokenized by MeCab with the Unidic 2.1.2 dictionary and then split into subwords by the WordPiece algorithm.
|
36 |
+
The vocabulary size is 32768.
|
37 |
+
|
38 |
+
We used [fugashi](https://github.com/polm/fugashi) and [unidic-lite](https://github.com/polm/unidic-lite) packages for the tokenization.
|
39 |
+
|
40 |
+
## Training
|
41 |
+
|
42 |
+
We trained the model first on the CC-100 corpus for 1M steps and then on the Wikipedia corpus for another 1M steps.
|
43 |
+
For training of the MLM (masked language modeling) objective, we introduced whole word masking in which all of the subword tokens corresponding to a single word (tokenized by MeCab) are masked at once.
|
44 |
+
|
45 |
+
For training of each model, we used a v3-8 instance of Cloud TPUs provided by [TPU Research Cloud](https://sites.research.google/trc/about/).
|
46 |
+
|
47 |
+
## Licenses
|
48 |
+
|
49 |
+
The pretrained models are distributed under the Apache License 2.0.
|
50 |
+
|
51 |
+
## Acknowledgments
|
52 |
+
|
53 |
+
This model is trained with Cloud TPUs provided by [TPU Research Cloud](https://sites.research.google/trc/about/) program.
|
TTSs/genshin/bert/bert-large-japanese-v2/config.json
ADDED
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"architectures": [
|
3 |
+
"BertForPreTraining"
|
4 |
+
],
|
5 |
+
"attention_probs_dropout_prob": 0.1,
|
6 |
+
"hidden_act": "gelu",
|
7 |
+
"hidden_dropout_prob": 0.1,
|
8 |
+
"hidden_size": 1024,
|
9 |
+
"initializer_range": 0.02,
|
10 |
+
"intermediate_size": 4096,
|
11 |
+
"layer_norm_eps": 1e-12,
|
12 |
+
"max_position_embeddings": 512,
|
13 |
+
"model_type": "bert",
|
14 |
+
"num_attention_heads": 16,
|
15 |
+
"num_hidden_layers": 24,
|
16 |
+
"pad_token_id": 0,
|
17 |
+
"type_vocab_size": 2,
|
18 |
+
"vocab_size": 32768
|
19 |
+
}
|
TTSs/genshin/bert/bert-large-japanese-v2/tokenizer_config.json
ADDED
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"tokenizer_class": "BertJapaneseTokenizer",
|
3 |
+
"model_max_length": 512,
|
4 |
+
"do_lower_case": false,
|
5 |
+
"word_tokenizer_type": "mecab",
|
6 |
+
"subword_tokenizer_type": "wordpiece",
|
7 |
+
"mecab_kwargs": {
|
8 |
+
"mecab_dic": "unidic_lite"
|
9 |
+
}
|
10 |
+
}
|
TTSs/genshin/bert/bert-large-japanese-v2/vocab.txt
ADDED
The diff for this file is too large to render.
See raw diff
|
|
TTSs/genshin/bert/bert_models.json
ADDED
@@ -0,0 +1,14 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"deberta-v2-large-japanese-char-wwm": {
|
3 |
+
"repo_id": "ku-nlp/deberta-v2-large-japanese-char-wwm",
|
4 |
+
"files": ["pytorch_model.bin"]
|
5 |
+
},
|
6 |
+
"chinese-roberta-wwm-ext-large": {
|
7 |
+
"repo_id": "hfl/chinese-roberta-wwm-ext-large",
|
8 |
+
"files": ["pytorch_model.bin"]
|
9 |
+
},
|
10 |
+
"deberta-v3-large": {
|
11 |
+
"repo_id": "microsoft/deberta-v3-large",
|
12 |
+
"files": ["spm.model", "pytorch_model.bin"]
|
13 |
+
}
|
14 |
+
}
|
TTSs/genshin/bert/chinese-roberta-wwm-ext-large/.gitattributes
ADDED
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
*.bin.* filter=lfs diff=lfs merge=lfs -text
|
2 |
+
*.lfs.* filter=lfs diff=lfs merge=lfs -text
|
3 |
+
*.bin filter=lfs diff=lfs merge=lfs -text
|
4 |
+
*.h5 filter=lfs diff=lfs merge=lfs -text
|
5 |
+
*.tflite filter=lfs diff=lfs merge=lfs -text
|
6 |
+
*.tar.gz filter=lfs diff=lfs merge=lfs -text
|
7 |
+
*.ot filter=lfs diff=lfs merge=lfs -text
|
8 |
+
*.onnx filter=lfs diff=lfs merge=lfs -text
|
9 |
+
*.msgpack filter=lfs diff=lfs merge=lfs -text
|
TTSs/genshin/bert/chinese-roberta-wwm-ext-large/README.md
ADDED
@@ -0,0 +1,57 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
language:
|
3 |
+
- zh
|
4 |
+
tags:
|
5 |
+
- bert
|
6 |
+
license: "apache-2.0"
|
7 |
+
---
|
8 |
+
|
9 |
+
# Please use 'Bert' related functions to load this model!
|
10 |
+
|
11 |
+
## Chinese BERT with Whole Word Masking
|
12 |
+
For further accelerating Chinese natural language processing, we provide **Chinese pre-trained BERT with Whole Word Masking**.
|
13 |
+
|
14 |
+
**[Pre-Training with Whole Word Masking for Chinese BERT](https://arxiv.org/abs/1906.08101)**
|
15 |
+
Yiming Cui, Wanxiang Che, Ting Liu, Bing Qin, Ziqing Yang, Shijin Wang, Guoping Hu
|
16 |
+
|
17 |
+
This repository is developed based on:https://github.com/google-research/bert
|
18 |
+
|
19 |
+
You may also interested in,
|
20 |
+
- Chinese BERT series: https://github.com/ymcui/Chinese-BERT-wwm
|
21 |
+
- Chinese MacBERT: https://github.com/ymcui/MacBERT
|
22 |
+
- Chinese ELECTRA: https://github.com/ymcui/Chinese-ELECTRA
|
23 |
+
- Chinese XLNet: https://github.com/ymcui/Chinese-XLNet
|
24 |
+
- Knowledge Distillation Toolkit - TextBrewer: https://github.com/airaria/TextBrewer
|
25 |
+
|
26 |
+
More resources by HFL: https://github.com/ymcui/HFL-Anthology
|
27 |
+
|
28 |
+
## Citation
|
29 |
+
If you find the technical report or resource is useful, please cite the following technical report in your paper.
|
30 |
+
- Primary: https://arxiv.org/abs/2004.13922
|
31 |
+
```
|
32 |
+
@inproceedings{cui-etal-2020-revisiting,
|
33 |
+
title = "Revisiting Pre-Trained Models for {C}hinese Natural Language Processing",
|
34 |
+
author = "Cui, Yiming and
|
35 |
+
Che, Wanxiang and
|
36 |
+
Liu, Ting and
|
37 |
+
Qin, Bing and
|
38 |
+
Wang, Shijin and
|
39 |
+
Hu, Guoping",
|
40 |
+
booktitle = "Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing: Findings",
|
41 |
+
month = nov,
|
42 |
+
year = "2020",
|
43 |
+
address = "Online",
|
44 |
+
publisher = "Association for Computational Linguistics",
|
45 |
+
url = "https://www.aclweb.org/anthology/2020.findings-emnlp.58",
|
46 |
+
pages = "657--668",
|
47 |
+
}
|
48 |
+
```
|
49 |
+
- Secondary: https://arxiv.org/abs/1906.08101
|
50 |
+
```
|
51 |
+
@article{chinese-bert-wwm,
|
52 |
+
title={Pre-Training with Whole Word Masking for Chinese BERT},
|
53 |
+
author={Cui, Yiming and Che, Wanxiang and Liu, Ting and Qin, Bing and Yang, Ziqing and Wang, Shijin and Hu, Guoping},
|
54 |
+
journal={arXiv preprint arXiv:1906.08101},
|
55 |
+
year={2019}
|
56 |
+
}
|
57 |
+
```
|
TTSs/genshin/bert/chinese-roberta-wwm-ext-large/added_tokens.json
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
{}
|
TTSs/genshin/bert/chinese-roberta-wwm-ext-large/config.json
ADDED
@@ -0,0 +1,28 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"architectures": [
|
3 |
+
"BertForMaskedLM"
|
4 |
+
],
|
5 |
+
"attention_probs_dropout_prob": 0.1,
|
6 |
+
"bos_token_id": 0,
|
7 |
+
"directionality": "bidi",
|
8 |
+
"eos_token_id": 2,
|
9 |
+
"hidden_act": "gelu",
|
10 |
+
"hidden_dropout_prob": 0.1,
|
11 |
+
"hidden_size": 1024,
|
12 |
+
"initializer_range": 0.02,
|
13 |
+
"intermediate_size": 4096,
|
14 |
+
"layer_norm_eps": 1e-12,
|
15 |
+
"max_position_embeddings": 512,
|
16 |
+
"model_type": "bert",
|
17 |
+
"num_attention_heads": 16,
|
18 |
+
"num_hidden_layers": 24,
|
19 |
+
"output_past": true,
|
20 |
+
"pad_token_id": 0,
|
21 |
+
"pooler_fc_size": 768,
|
22 |
+
"pooler_num_attention_heads": 12,
|
23 |
+
"pooler_num_fc_layers": 3,
|
24 |
+
"pooler_size_per_head": 128,
|
25 |
+
"pooler_type": "first_token_transform",
|
26 |
+
"type_vocab_size": 2,
|
27 |
+
"vocab_size": 21128
|
28 |
+
}
|
TTSs/genshin/bert/chinese-roberta-wwm-ext-large/special_tokens_map.json
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
{"unk_token": "[UNK]", "sep_token": "[SEP]", "pad_token": "[PAD]", "cls_token": "[CLS]", "mask_token": "[MASK]"}
|
TTSs/genshin/bert/chinese-roberta-wwm-ext-large/tokenizer.json
ADDED
The diff for this file is too large to render.
See raw diff
|
|
TTSs/genshin/bert/chinese-roberta-wwm-ext-large/tokenizer_config.json
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
{"init_inputs": []}
|
TTSs/genshin/bert/chinese-roberta-wwm-ext-large/vocab.txt
ADDED
The diff for this file is too large to render.
See raw diff
|
|
TTSs/genshin/bert/deberta-v2-large-japanese-char-wwm/.gitattributes
ADDED
@@ -0,0 +1,34 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
*.7z filter=lfs diff=lfs merge=lfs -text
|
2 |
+
*.arrow filter=lfs diff=lfs merge=lfs -text
|
3 |
+
*.bin filter=lfs diff=lfs merge=lfs -text
|
4 |
+
*.bz2 filter=lfs diff=lfs merge=lfs -text
|
5 |
+
*.ckpt filter=lfs diff=lfs merge=lfs -text
|
6 |
+
*.ftz filter=lfs diff=lfs merge=lfs -text
|
7 |
+
*.gz filter=lfs diff=lfs merge=lfs -text
|
8 |
+
*.h5 filter=lfs diff=lfs merge=lfs -text
|
9 |
+
*.joblib filter=lfs diff=lfs merge=lfs -text
|
10 |
+
*.lfs.* filter=lfs diff=lfs merge=lfs -text
|
11 |
+
*.mlmodel filter=lfs diff=lfs merge=lfs -text
|
12 |
+
*.model filter=lfs diff=lfs merge=lfs -text
|
13 |
+
*.msgpack filter=lfs diff=lfs merge=lfs -text
|
14 |
+
*.npy filter=lfs diff=lfs merge=lfs -text
|
15 |
+
*.npz filter=lfs diff=lfs merge=lfs -text
|
16 |
+
*.onnx filter=lfs diff=lfs merge=lfs -text
|
17 |
+
*.ot filter=lfs diff=lfs merge=lfs -text
|
18 |
+
*.parquet filter=lfs diff=lfs merge=lfs -text
|
19 |
+
*.pb filter=lfs diff=lfs merge=lfs -text
|
20 |
+
*.pickle filter=lfs diff=lfs merge=lfs -text
|
21 |
+
*.pkl filter=lfs diff=lfs merge=lfs -text
|
22 |
+
*.pt filter=lfs diff=lfs merge=lfs -text
|
23 |
+
*.pth filter=lfs diff=lfs merge=lfs -text
|
24 |
+
*.rar filter=lfs diff=lfs merge=lfs -text
|
25 |
+
*.safetensors filter=lfs diff=lfs merge=lfs -text
|
26 |
+
saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
27 |
+
*.tar.* filter=lfs diff=lfs merge=lfs -text
|
28 |
+
*.tflite filter=lfs diff=lfs merge=lfs -text
|
29 |
+
*.tgz filter=lfs diff=lfs merge=lfs -text
|
30 |
+
*.wasm filter=lfs diff=lfs merge=lfs -text
|
31 |
+
*.xz filter=lfs diff=lfs merge=lfs -text
|
32 |
+
*.zip filter=lfs diff=lfs merge=lfs -text
|
33 |
+
*.zst filter=lfs diff=lfs merge=lfs -text
|
34 |
+
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
TTSs/genshin/bert/deberta-v2-large-japanese-char-wwm/README.md
ADDED
@@ -0,0 +1,89 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
language: ja
|
3 |
+
license: cc-by-sa-4.0
|
4 |
+
library_name: transformers
|
5 |
+
tags:
|
6 |
+
- deberta
|
7 |
+
- deberta-v2
|
8 |
+
- fill-mask
|
9 |
+
- character
|
10 |
+
- wwm
|
11 |
+
datasets:
|
12 |
+
- wikipedia
|
13 |
+
- cc100
|
14 |
+
- oscar
|
15 |
+
metrics:
|
16 |
+
- accuracy
|
17 |
+
mask_token: "[MASK]"
|
18 |
+
widget:
|
19 |
+
- text: "京都大学で自然言語処理を[MASK][MASK]する。"
|
20 |
+
---
|
21 |
+
|
22 |
+
# Model Card for Japanese character-level DeBERTa V2 large
|
23 |
+
|
24 |
+
## Model description
|
25 |
+
|
26 |
+
This is a Japanese DeBERTa V2 large model pre-trained on Japanese Wikipedia, the Japanese portion of CC-100, and the Japanese portion of OSCAR.
|
27 |
+
This model is trained with character-level tokenization and whole word masking.
|
28 |
+
|
29 |
+
## How to use
|
30 |
+
|
31 |
+
You can use this model for masked language modeling as follows:
|
32 |
+
|
33 |
+
```python
|
34 |
+
from transformers import AutoTokenizer, AutoModelForMaskedLM
|
35 |
+
tokenizer = AutoTokenizer.from_pretrained('ku-nlp/deberta-v2-large-japanese-char-wwm')
|
36 |
+
model = AutoModelForMaskedLM.from_pretrained('ku-nlp/deberta-v2-large-japanese-char-wwm')
|
37 |
+
|
38 |
+
sentence = '京都大学で自然言語処理を[MASK][MASK]する。'
|
39 |
+
encoding = tokenizer(sentence, return_tensors='pt')
|
40 |
+
...
|
41 |
+
```
|
42 |
+
|
43 |
+
You can also fine-tune this model on downstream tasks.
|
44 |
+
|
45 |
+
## Tokenization
|
46 |
+
|
47 |
+
There is no need to tokenize texts in advance, and you can give raw texts to the tokenizer.
|
48 |
+
The texts are tokenized into character-level tokens by [sentencepiece](https://github.com/google/sentencepiece).
|
49 |
+
|
50 |
+
## Training data
|
51 |
+
|
52 |
+
We used the following corpora for pre-training:
|
53 |
+
|
54 |
+
- Japanese Wikipedia (as of 20221020, 3.2GB, 27M sentences, 1.3M documents)
|
55 |
+
- Japanese portion of CC-100 (85GB, 619M sentences, 66M documents)
|
56 |
+
- Japanese portion of OSCAR (54GB, 326M sentences, 25M documents)
|
57 |
+
|
58 |
+
Note that we filtered out documents annotated with "header", "footer", or "noisy" tags in OSCAR.
|
59 |
+
Also note that Japanese Wikipedia was duplicated 10 times to make the total size of the corpus comparable to that of CC-100 and OSCAR. As a result, the total size of the training data is 171GB.
|
60 |
+
|
61 |
+
## Training procedure
|
62 |
+
|
63 |
+
We first segmented texts in the corpora into words using [Juman++ 2.0.0-rc3](https://github.com/ku-nlp/jumanpp/releases/tag/v2.0.0-rc3) for whole word masking.
|
64 |
+
Then, we built a sentencepiece model with 22,012 tokens including all characters that appear in the training corpus.
|
65 |
+
|
66 |
+
We tokenized raw corpora into character-level subwords using the sentencepiece model and trained the Japanese DeBERTa model using [transformers](https://github.com/huggingface/transformers) library.
|
67 |
+
The training took 26 days using 16 NVIDIA A100-SXM4-40GB GPUs.
|
68 |
+
|
69 |
+
The following hyperparameters were used during pre-training:
|
70 |
+
|
71 |
+
- learning_rate: 1e-4
|
72 |
+
- per_device_train_batch_size: 26
|
73 |
+
- distributed_type: multi-GPU
|
74 |
+
- num_devices: 16
|
75 |
+
- gradient_accumulation_steps: 8
|
76 |
+
- total_train_batch_size: 3,328
|
77 |
+
- max_seq_length: 512
|
78 |
+
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-06
|
79 |
+
- lr_scheduler_type: linear schedule with warmup (lr = 0 at 300k steps)
|
80 |
+
- training_steps: 260,000
|
81 |
+
- warmup_steps: 10,000
|
82 |
+
|
83 |
+
The accuracy of the trained model on the masked language modeling task was 0.795.
|
84 |
+
The evaluation set consists of 5,000 randomly sampled documents from each of the training corpora.
|
85 |
+
|
86 |
+
## Acknowledgments
|
87 |
+
|
88 |
+
This work was supported by Joint Usage/Research Center for Interdisciplinary Large-scale Information Infrastructures (JHPCN) through General Collaboration Project no. jh221004, "Developing a Platform for Constructing and Sharing of Large-Scale Japanese Language Models".
|
89 |
+
For training models, we used the mdx: a platform for the data-driven future.
|
TTSs/genshin/bert/deberta-v2-large-japanese-char-wwm/config.json
ADDED
@@ -0,0 +1,37 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"architectures": [
|
3 |
+
"DebertaV2ForMaskedLM"
|
4 |
+
],
|
5 |
+
"attention_head_size": 64,
|
6 |
+
"attention_probs_dropout_prob": 0.1,
|
7 |
+
"conv_act": "gelu",
|
8 |
+
"conv_kernel_size": 3,
|
9 |
+
"hidden_act": "gelu",
|
10 |
+
"hidden_dropout_prob": 0.1,
|
11 |
+
"hidden_size": 1024,
|
12 |
+
"initializer_range": 0.02,
|
13 |
+
"intermediate_size": 4096,
|
14 |
+
"layer_norm_eps": 1e-07,
|
15 |
+
"max_position_embeddings": 512,
|
16 |
+
"max_relative_positions": -1,
|
17 |
+
"model_type": "deberta-v2",
|
18 |
+
"norm_rel_ebd": "layer_norm",
|
19 |
+
"num_attention_heads": 16,
|
20 |
+
"num_hidden_layers": 24,
|
21 |
+
"pad_token_id": 0,
|
22 |
+
"pooler_dropout": 0,
|
23 |
+
"pooler_hidden_act": "gelu",
|
24 |
+
"pooler_hidden_size": 1024,
|
25 |
+
"pos_att_type": [
|
26 |
+
"p2c",
|
27 |
+
"c2p"
|
28 |
+
],
|
29 |
+
"position_biased_input": false,
|
30 |
+
"position_buckets": 256,
|
31 |
+
"relative_attention": true,
|
32 |
+
"share_att_key": true,
|
33 |
+
"torch_dtype": "float16",
|
34 |
+
"transformers_version": "4.25.1",
|
35 |
+
"type_vocab_size": 0,
|
36 |
+
"vocab_size": 22012
|
37 |
+
}
|
TTSs/genshin/bert/deberta-v2-large-japanese-char-wwm/special_tokens_map.json
ADDED
@@ -0,0 +1,7 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"cls_token": "[CLS]",
|
3 |
+
"mask_token": "[MASK]",
|
4 |
+
"pad_token": "[PAD]",
|
5 |
+
"sep_token": "[SEP]",
|
6 |
+
"unk_token": "[UNK]"
|
7 |
+
}
|
TTSs/genshin/bert/deberta-v2-large-japanese-char-wwm/tokenizer_config.json
ADDED
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"cls_token": "[CLS]",
|
3 |
+
"do_lower_case": false,
|
4 |
+
"do_subword_tokenize": true,
|
5 |
+
"do_word_tokenize": true,
|
6 |
+
"jumanpp_kwargs": null,
|
7 |
+
"mask_token": "[MASK]",
|
8 |
+
"mecab_kwargs": null,
|
9 |
+
"model_max_length": 1000000000000000019884624838656,
|
10 |
+
"never_split": null,
|
11 |
+
"pad_token": "[PAD]",
|
12 |
+
"sep_token": "[SEP]",
|
13 |
+
"special_tokens_map_file": null,
|
14 |
+
"subword_tokenizer_type": "character",
|
15 |
+
"sudachi_kwargs": null,
|
16 |
+
"tokenizer_class": "BertJapaneseTokenizer",
|
17 |
+
"unk_token": "[UNK]",
|
18 |
+
"word_tokenizer_type": "basic"
|
19 |
+
}
|
TTSs/genshin/bert/deberta-v2-large-japanese-char-wwm/vocab.txt
ADDED
The diff for this file is too large to render.
See raw diff
|
|
TTSs/genshin/bert/deberta-v2-large-japanese/.gitattributes
ADDED
@@ -0,0 +1,34 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
*.7z filter=lfs diff=lfs merge=lfs -text
|
2 |
+
*.arrow filter=lfs diff=lfs merge=lfs -text
|
3 |
+
*.bin filter=lfs diff=lfs merge=lfs -text
|
4 |
+
*.bz2 filter=lfs diff=lfs merge=lfs -text
|
5 |
+
*.ckpt filter=lfs diff=lfs merge=lfs -text
|
6 |
+
*.ftz filter=lfs diff=lfs merge=lfs -text
|
7 |
+
*.gz filter=lfs diff=lfs merge=lfs -text
|
8 |
+
*.h5 filter=lfs diff=lfs merge=lfs -text
|
9 |
+
*.joblib filter=lfs diff=lfs merge=lfs -text
|
10 |
+
*.lfs.* filter=lfs diff=lfs merge=lfs -text
|
11 |
+
*.mlmodel filter=lfs diff=lfs merge=lfs -text
|
12 |
+
*.model filter=lfs diff=lfs merge=lfs -text
|
13 |
+
*.msgpack filter=lfs diff=lfs merge=lfs -text
|
14 |
+
*.npy filter=lfs diff=lfs merge=lfs -text
|
15 |
+
*.npz filter=lfs diff=lfs merge=lfs -text
|
16 |
+
*.onnx filter=lfs diff=lfs merge=lfs -text
|
17 |
+
*.ot filter=lfs diff=lfs merge=lfs -text
|
18 |
+
*.parquet filter=lfs diff=lfs merge=lfs -text
|
19 |
+
*.pb filter=lfs diff=lfs merge=lfs -text
|
20 |
+
*.pickle filter=lfs diff=lfs merge=lfs -text
|
21 |
+
*.pkl filter=lfs diff=lfs merge=lfs -text
|
22 |
+
*.pt filter=lfs diff=lfs merge=lfs -text
|
23 |
+
*.pth filter=lfs diff=lfs merge=lfs -text
|
24 |
+
*.rar filter=lfs diff=lfs merge=lfs -text
|
25 |
+
*.safetensors filter=lfs diff=lfs merge=lfs -text
|
26 |
+
saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
27 |
+
*.tar.* filter=lfs diff=lfs merge=lfs -text
|
28 |
+
*.tflite filter=lfs diff=lfs merge=lfs -text
|
29 |
+
*.tgz filter=lfs diff=lfs merge=lfs -text
|
30 |
+
*.wasm filter=lfs diff=lfs merge=lfs -text
|
31 |
+
*.xz filter=lfs diff=lfs merge=lfs -text
|
32 |
+
*.zip filter=lfs diff=lfs merge=lfs -text
|
33 |
+
*.zst filter=lfs diff=lfs merge=lfs -text
|
34 |
+
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
TTSs/genshin/bert/deberta-v2-large-japanese/README.md
ADDED
@@ -0,0 +1,111 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
language: ja
|
3 |
+
license: cc-by-sa-4.0
|
4 |
+
library_name: transformers
|
5 |
+
tags:
|
6 |
+
- deberta
|
7 |
+
- deberta-v2
|
8 |
+
- fill-mask
|
9 |
+
datasets:
|
10 |
+
- wikipedia
|
11 |
+
- cc100
|
12 |
+
- oscar
|
13 |
+
metrics:
|
14 |
+
- accuracy
|
15 |
+
mask_token: "[MASK]"
|
16 |
+
widget:
|
17 |
+
- text: "京都 大学 で 自然 言語 処理 を [MASK] する 。"
|
18 |
+
---
|
19 |
+
|
20 |
+
# Model Card for Japanese DeBERTa V2 large
|
21 |
+
|
22 |
+
## Model description
|
23 |
+
|
24 |
+
This is a Japanese DeBERTa V2 large model pre-trained on Japanese Wikipedia, the Japanese portion of CC-100, and the
|
25 |
+
Japanese portion of OSCAR.
|
26 |
+
|
27 |
+
## How to use
|
28 |
+
|
29 |
+
You can use this model for masked language modeling as follows:
|
30 |
+
|
31 |
+
```python
|
32 |
+
from transformers import AutoTokenizer, AutoModelForMaskedLM
|
33 |
+
|
34 |
+
tokenizer = AutoTokenizer.from_pretrained('ku-nlp/deberta-v2-large-japanese')
|
35 |
+
model = AutoModelForMaskedLM.from_pretrained('ku-nlp/deberta-v2-large-japanese')
|
36 |
+
|
37 |
+
sentence = '京都 大学 で 自然 言語 処理 を [MASK] する 。' # input should be segmented into words by Juman++ in advance
|
38 |
+
encoding = tokenizer(sentence, return_tensors='pt')
|
39 |
+
...
|
40 |
+
```
|
41 |
+
|
42 |
+
You can also fine-tune this model on downstream tasks.
|
43 |
+
|
44 |
+
## Tokenization
|
45 |
+
|
46 |
+
The input text should be segmented into words by [Juman++](https://github.com/ku-nlp/jumanpp) in
|
47 |
+
advance. [Juman++ 2.0.0-rc3](https://github.com/ku-nlp/jumanpp/releases/tag/v2.0.0-rc3) was used for pre-training. Each
|
48 |
+
word is tokenized into subwords by [sentencepiece](https://github.com/google/sentencepiece).
|
49 |
+
|
50 |
+
## Training data
|
51 |
+
|
52 |
+
We used the following corpora for pre-training:
|
53 |
+
|
54 |
+
- Japanese Wikipedia (as of 20221020, 3.2GB, 27M sentences, 1.3M documents)
|
55 |
+
- Japanese portion of CC-100 (85GB, 619M sentences, 66M documents)
|
56 |
+
- Japanese portion of OSCAR (54GB, 326M sentences, 25M documents)
|
57 |
+
|
58 |
+
Note that we filtered out documents annotated with "header", "footer", or "noisy" tags in OSCAR.
|
59 |
+
Also note that Japanese Wikipedia was duplicated 10 times to make the total size of the corpus comparable to that of
|
60 |
+
CC-100 and OSCAR. As a result, the total size of the training data is 171GB.
|
61 |
+
|
62 |
+
## Training procedure
|
63 |
+
|
64 |
+
We first segmented texts in the corpora into words using [Juman++](https://github.com/ku-nlp/jumanpp).
|
65 |
+
Then, we built a sentencepiece model with 32000 tokens including words ([JumanDIC](https://github.com/ku-nlp/JumanDIC))
|
66 |
+
and subwords induced by the unigram language model of [sentencepiece](https://github.com/google/sentencepiece).
|
67 |
+
|
68 |
+
We tokenized the segmented corpora into subwords using the sentencepiece model and trained the Japanese DeBERTa model
|
69 |
+
using [transformers](https://github.com/huggingface/transformers) library.
|
70 |
+
The training took 36 days using 8 NVIDIA A100-SXM4-40GB GPUs.
|
71 |
+
|
72 |
+
The following hyperparameters were used during pre-training:
|
73 |
+
|
74 |
+
- learning_rate: 1e-4
|
75 |
+
- per_device_train_batch_size: 18
|
76 |
+
- distributed_type: multi-GPU
|
77 |
+
- num_devices: 8
|
78 |
+
- gradient_accumulation_steps: 16
|
79 |
+
- total_train_batch_size: 2,304
|
80 |
+
- max_seq_length: 512
|
81 |
+
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-06
|
82 |
+
- lr_scheduler_type: linear schedule with warmup
|
83 |
+
- training_steps: 300,000
|
84 |
+
- warmup_steps: 10,000
|
85 |
+
|
86 |
+
The accuracy of the trained model on the masked language modeling task was 0.799.
|
87 |
+
The evaluation set consists of 5,000 randomly sampled documents from each of the training corpora.
|
88 |
+
|
89 |
+
## Fine-tuning on NLU tasks
|
90 |
+
|
91 |
+
We fine-tuned the following models and evaluated them on the dev set of JGLUE.
|
92 |
+
We tuned learning rate and training epochs for each model and task
|
93 |
+
following [the JGLUE paper](https://www.jstage.jst.go.jp/article/jnlp/30/1/30_63/_pdf/-char/ja).
|
94 |
+
|
95 |
+
| Model | MARC-ja/acc | JSTS/pearson | JSTS/spearman | JNLI/acc | JSQuAD/EM | JSQuAD/F1 | JComQA/acc |
|
96 |
+
|-------------------------------|-------------|--------------|---------------|----------|-----------|-----------|------------|
|
97 |
+
| Waseda RoBERTa base | 0.965 | 0.913 | 0.876 | 0.905 | 0.853 | 0.916 | 0.853 |
|
98 |
+
| Waseda RoBERTa large (seq512) | 0.969 | 0.925 | 0.890 | 0.928 | 0.910 | 0.955 | 0.900 |
|
99 |
+
| LUKE Japanese base* | 0.965 | 0.916 | 0.877 | 0.912 | - | - | 0.842 |
|
100 |
+
| LUKE Japanese large* | 0.965 | 0.932 | 0.902 | 0.927 | - | - | 0.893 |
|
101 |
+
| DeBERTaV2 base | 0.970 | 0.922 | 0.886 | 0.922 | 0.899 | 0.951 | 0.873 |
|
102 |
+
| DeBERTaV2 large | 0.968 | 0.925 | 0.892 | 0.924 | 0.912 | 0.959 | 0.890 |
|
103 |
+
|
104 |
+
*The scores of LUKE are from [the official repository](https://github.com/studio-ousia/luke).
|
105 |
+
|
106 |
+
## Acknowledgments
|
107 |
+
|
108 |
+
This work was supported by Joint Usage/Research Center for Interdisciplinary Large-scale Information Infrastructures (
|
109 |
+
JHPCN) through General Collaboration Project no. jh221004, "Developing a Platform for Constructing and Sharing of
|
110 |
+
Large-Scale Japanese Language Models".
|
111 |
+
For training models, we used the mdx: a platform for the data-driven future.
|
TTSs/genshin/bert/deberta-v2-large-japanese/config.json
ADDED
@@ -0,0 +1,38 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"_name_or_path": "configs/deberta_v2_large.json",
|
3 |
+
"architectures": [
|
4 |
+
"DebertaV2ForMaskedLM"
|
5 |
+
],
|
6 |
+
"attention_head_size": 64,
|
7 |
+
"attention_probs_dropout_prob": 0.1,
|
8 |
+
"conv_act": "gelu",
|
9 |
+
"conv_kernel_size": 3,
|
10 |
+
"hidden_act": "gelu",
|
11 |
+
"hidden_dropout_prob": 0.1,
|
12 |
+
"hidden_size": 1024,
|
13 |
+
"initializer_range": 0.02,
|
14 |
+
"intermediate_size": 4096,
|
15 |
+
"layer_norm_eps": 1e-07,
|
16 |
+
"max_position_embeddings": 512,
|
17 |
+
"max_relative_positions": -1,
|
18 |
+
"model_type": "deberta-v2",
|
19 |
+
"norm_rel_ebd": "layer_norm",
|
20 |
+
"num_attention_heads": 16,
|
21 |
+
"num_hidden_layers": 24,
|
22 |
+
"pad_token_id": 0,
|
23 |
+
"pooler_dropout": 0,
|
24 |
+
"pooler_hidden_act": "gelu",
|
25 |
+
"pooler_hidden_size": 1024,
|
26 |
+
"pos_att_type": [
|
27 |
+
"p2c",
|
28 |
+
"c2p"
|
29 |
+
],
|
30 |
+
"position_biased_input": false,
|
31 |
+
"position_buckets": 256,
|
32 |
+
"relative_attention": true,
|
33 |
+
"share_att_key": true,
|
34 |
+
"torch_dtype": "float32",
|
35 |
+
"transformers_version": "4.23.1",
|
36 |
+
"type_vocab_size": 0,
|
37 |
+
"vocab_size": 32000
|
38 |
+
}
|
TTSs/genshin/bert/deberta-v2-large-japanese/special_tokens_map.json
ADDED
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"bos_token": "[CLS]",
|
3 |
+
"cls_token": "[CLS]",
|
4 |
+
"eos_token": "[SEP]",
|
5 |
+
"mask_token": "[MASK]",
|
6 |
+
"pad_token": "[PAD]",
|
7 |
+
"sep_token": "[SEP]",
|
8 |
+
"unk_token": "[UNK]"
|
9 |
+
}
|
TTSs/genshin/bert/deberta-v2-large-japanese/tokenizer.json
ADDED
The diff for this file is too large to render.
See raw diff
|
|
TTSs/genshin/bert/deberta-v2-large-japanese/tokenizer_config.json
ADDED
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"bos_token": "[CLS]",
|
3 |
+
"cls_token": "[CLS]",
|
4 |
+
"do_lower_case": false,
|
5 |
+
"eos_token": "[SEP]",
|
6 |
+
"keep_accents": true,
|
7 |
+
"mask_token": "[MASK]",
|
8 |
+
"pad_token": "[PAD]",
|
9 |
+
"sep_token": "[SEP]",
|
10 |
+
"sp_model_kwargs": {},
|
11 |
+
"special_tokens_map_file": null,
|
12 |
+
"split_by_punct": false,
|
13 |
+
"tokenizer_class": "DebertaV2Tokenizer",
|
14 |
+
"unk_token": "[UNK]"
|
15 |
+
}
|
TTSs/genshin/bert/deberta-v3-large/.gitattributes
ADDED
@@ -0,0 +1,27 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
*.7z filter=lfs diff=lfs merge=lfs -text
|
2 |
+
*.arrow filter=lfs diff=lfs merge=lfs -text
|
3 |
+
*.bin filter=lfs diff=lfs merge=lfs -text
|
4 |
+
*.bin.* filter=lfs diff=lfs merge=lfs -text
|
5 |
+
*.bz2 filter=lfs diff=lfs merge=lfs -text
|
6 |
+
*.ftz filter=lfs diff=lfs merge=lfs -text
|
7 |
+
*.gz filter=lfs diff=lfs merge=lfs -text
|
8 |
+
*.h5 filter=lfs diff=lfs merge=lfs -text
|
9 |
+
*.joblib filter=lfs diff=lfs merge=lfs -text
|
10 |
+
*.lfs.* filter=lfs diff=lfs merge=lfs -text
|
11 |
+
*.model filter=lfs diff=lfs merge=lfs -text
|
12 |
+
*.msgpack filter=lfs diff=lfs merge=lfs -text
|
13 |
+
*.onnx filter=lfs diff=lfs merge=lfs -text
|
14 |
+
*.ot filter=lfs diff=lfs merge=lfs -text
|
15 |
+
*.parquet filter=lfs diff=lfs merge=lfs -text
|
16 |
+
*.pb filter=lfs diff=lfs merge=lfs -text
|
17 |
+
*.pt filter=lfs diff=lfs merge=lfs -text
|
18 |
+
*.pth filter=lfs diff=lfs merge=lfs -text
|
19 |
+
*.rar filter=lfs diff=lfs merge=lfs -text
|
20 |
+
saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
21 |
+
*.tar.* filter=lfs diff=lfs merge=lfs -text
|
22 |
+
*.tflite filter=lfs diff=lfs merge=lfs -text
|
23 |
+
*.tgz filter=lfs diff=lfs merge=lfs -text
|
24 |
+
*.xz filter=lfs diff=lfs merge=lfs -text
|
25 |
+
*.zip filter=lfs diff=lfs merge=lfs -text
|
26 |
+
*.zstandard filter=lfs diff=lfs merge=lfs -text
|
27 |
+
*tfevents* filter=lfs diff=lfs merge=lfs -text
|