repronets-k8s
所属分类:微服务
开发工具:JavaScript
文件大小:784KB
下载次数:0
上传日期:2022-08-30 21:29:04
上 传 者:
sh-1993
说明: 可复制命名实体音译模型中心
(Reproducible Named Entity Transliteration Model Hub)
文件列表:
.dockerignore (96, 2022-08-31)
LICENSE (11411, 2022-08-31)
Makefile (1863, 2022-08-31)
data (0, 2022-08-31)
data\wd_korean_16_fixed (156064, 2022-08-31)
data\wd_korean_20_fixed (196277, 2022-08-31)
data\wd_korean_64_fixed (628195, 2022-08-31)
docs (0, 2022-08-31)
docs\repronets.gif (427260, 2022-08-31)
models (0, 2022-08-31)
models\new_ksvc (0, 2022-08-31)
models\new_ksvc\predictor.py (3428, 2022-08-31)
models\new_ksvc\register.py (601, 2022-08-31)
models\new_ksvc\requirements.txt (26, 2022-08-31)
models\new_ksvc\trf-jpn.yaml (459, 2022-08-31)
models\new_ksvc\trf-kor.yaml (459, 2022-08-31)
models\new_ksvc\wsgi.py (54, 2022-08-31)
models\phonetisaurus (0, 2022-08-31)
models\phonetisaurus\Dockerfile (2383, 2022-08-31)
models\phonetisaurus\components (0, 2022-08-31)
models\phonetisaurus\components\ksvc_ara.yaml (297, 2022-08-31)
models\phonetisaurus\components\ksvc_chi.yaml (297, 2022-08-31)
models\phonetisaurus\components\ksvc_heb.yaml (297, 2022-08-31)
models\phonetisaurus\components\ksvc_jpn.yaml (297, 2022-08-31)
models\phonetisaurus\components\ksvc_kor.yaml (297, 2022-08-31)
models\phonetisaurus\components\ksvc_rus.yaml (297, 2022-08-31)
models\phonetisaurus\docker-entrypoint.sh (127, 2022-08-31)
models\phonetisaurus\predictor.py (3411, 2022-08-31)
models\phonetisaurus\requirements.txt (50, 2022-08-31)
models\phonetisaurus\wsgi.py (53, 2022-08-31)
models\transformer (0, 2022-08-31)
models\transformer\Dockerfile (901, 2022-08-31)
models\transformer\components (0, 2022-08-31)
models\transformer\components\ksvc_ara.yaml (297, 2022-08-31)
models\transformer\components\ksvc_chi.yaml (297, 2022-08-31)
models\transformer\components\ksvc_heb.yaml (297, 2022-08-31)
models\transformer\components\ksvc_jpn.yaml (297, 2022-08-31)
... ...
# Reproducible Named Entity Transliteration models (ReproNETs)
`repronets-k8s` is a hub for named entity transliteration (NET) models.
It hosts trained NET models and provides simple interface to make them accesssible. All models run as serverless services on [Knative](https://knative.dev/docs/). Our platform is highly scalable and economical thanks to Knative's `autoscaling (even to zero)` feature
Available models:
* [Phonetisaurus](https://github.com/AdolfVonKleist/Phonetisaurus)
* [Transformer](https://arxiv.org/abs/1706.03762)
The languages we currently support:
* English to Arabic
* English to Chinese
* English to Hebrew
* English to Japanese
* English to Korean
* English to Russian
## GUI
GUI provides a tool to easily and visually compare outputs across the `Transformer` and the `Phonetisaurus` with different beam sizes. It shows predicted sequence(s) along with the probability of each sequence.
Link: https://boxkey.me
## Web API
ReproNETs provide a public API that provides transliteration as a service via simple HTTP request.
The query parameters are:
* input: Any English word to transliterate. It must be less than or 37 characters.
* language: The target language. It must be either `ara`, `chi`, `heb`, `jpn`, `kor`, or `rus`.
* beam: The size of beam search. It must be an integer from 1 to 5.
* model: The name of a transliteration model. It must be either `phonetisaurus`, `transformer` or `all`.
### GET
This is mainly used for a real-time prediction.
#### Request
``` bash
curl --request GET \
--url 'https://www.boxkey.me/repronets/api/v1/predict?input=rocky&language=jpn&beam=2&model=all'
```
#### Response
```javascript
{
"data": {
"phonetisaurus": {
"data": {
"No.1": {
"prob": 0.0000018659,
"tokens": "ロッキー"
},
"No.2": {
"prob": 1.32e-8,
"tokens": "ロッキ"
}
},
"status": 200,
"message": "Successfully made predictions"
},
"transformer": {
"data": {
"No.1": {
"prob": 0.7512609939,
"tokens": "ロッキー"
},
"No.2": {
"prob": 0.0077490897,
"tokens": "ロキー"
}
},
"status": 200,
"message": "Successfully made predictions"
}
},
"status": 200
}
```
## Contribution
Thank you for your interest in contributing to the `repronets-k8s` codebase.
Models can be uploaded under the `./moedls` directory ideally as a docker container.
If you need help with containerizing your model, let us know at [`Issues`](https://github.com/box-key/ReproNETs/issues).
## License
Copyright 2020- Kei Nemoto, All Rights Reserved
Licensed under the Apache License 2.0: http://www.apache.org/licenses/
近期下载者:
相关文件:
收藏者: