Skip to content

Latest commit

 

History

History
145 lines (113 loc) · 3.09 KB

readme.md

File metadata and controls

145 lines (113 loc) · 3.09 KB

text2vec 服务

基于 text2vec 封装函数提供 API 服务

环境要求

  1. python 3.8+

环境变量

分组 配置项 说明
模型参数 MODEL 需要调用的模型名称 默认 paraphrase-multilingual-mpnet-base-v2 小模型 paraphrase-multilingual-MiniLM-L12-v2
模型参数 MODELS_PATH 所有模型的存储位置
模型参数 MODELS_TRAIN 是否支持个性话模型，默认 False

文件目录说明

├── README.md -- 项目说明
├── run.py -- 程序运行文件
├── run.sh -- 容器运行脚本
├── /requirements.txt -- 项目使用到的依赖包
├── /config.py -- 项目配置文件
├── /init_model.py -- 构建镜像中预加载模型脚本
├── /Dockerfile -- 项目镜像构建文件

编译

# 初始化打包
nohup docker build . -f ./Dockerfile -t wuhanchu/text2vec_service:latest &

# 持续更新
docker build . -f ./Dockerfile.continue -t wuhanchu/text2vec_service:1.3.9
docker push wuhanchu/text2vec_service:1.3.9

部署/运行

python run.py

or

docker-compose -f ./compose/ai_service.yml -f ./consumer/dataknown/ai_service.yml -p dataknown --env-file ./env/dataknown_test.env up -d text2vec_service

使用到的框架

torch
text2vec

API

计算句子与文档集之间的相似度值

  • Method: POST
  • Url: /semantic_search
  • Body:
{
  "sentences1": ["sentence1", "sentence2", "sentence3"],
  "sentences2": ["sentence1", "sentence2", "sentence3"]
}
  • Response:
{
  "result": [
    [
      { "corpus_id": 0, "score": 0.9476608037948608 },
      { "corpus_id": 1, "score": 0.9476608037948608 },
      { "corpus_id": 2, "score": 0.9476608037948608 }
    ],
    [
      { "corpus_id": 0, "score": 0.9476608037948608 },
      { "corpus_id": 1, "score": 0.9476608037948608 },
      { "corpus_id": 2, "score": 0.9476608037948608 }
    ],
    [
      { "corpus_id": 0, "score": 0.9476608037948608 },
      { "corpus_id": 1, "score": 0.9476608037948608 },
      { "corpus_id": 2, "score": 0.9476608037948608 }
    ]
  ]
}

计算句子之间的相似度值

  • Method: POST
  • Url: /cos_sim
  • Body:
{
  "sentences1": ["sentence1", "sentence2", "sentence3"],
  "sentences2": ["sentence1", "sentence2", "sentence3"]
}
  • Response:
{
  "result": [
    [0, 1, 1],
    [0, 1, 1],
    [0, 1, 1]
  ]
}

计算句子文本向量

  • Method: POST
  • Url: /computing_embeddings
  • Body:
{
  "sentences": ["sentence1", "sentence2", "sentence3"]
}
  • Response:
{
  "result": [
    [0, 1, 1],
    [0, 1, 1],
    [0, 1, 1]
  ]
}