txtai API 库

本教程系列将涵盖txtai的主要用例，这是一个 AI 驱动的语义搜索平台。该系列的每章都有相关代码，可也可以在colab 中使用。
colab 地址

txtai API 是由FastAPI支持的基于 Web 的服务。所有 txtai 功能，包括相似性搜索、提取 QA 和零样本标记都可以通过 API 获得。

本文安装了 txtai API 并展示了一个使用 txtai 支持的每种语言绑定的示例。

安装依赖

安装txtai和所有依赖项。由于本文使用了API，我们需要安装api extras包。

pip install txtai[api]

Python

我们将尝试的第一种方法是通过 Python 直接访问。我们将在此处的所有示例中使用零样本标记。有关零样本分类的更多详细信息，请参阅此文章。

import os
from IPython.core.display import display, HTML
from txtai.pipeline import Labels

def table(rows):
    html = """
    <style type='text/css'>
    @import url('https://fonts.googleapis.com/css?family=Oswald&display=swap');
    table {
      border-collapse: collapse;
      width: 900px;
    }
    th, td {
        border: 1px solid #9e9e9e;
        padding: 10px;
        font: 20px Oswald;
    }
    </style>
    """

    html += "<table><thead><tr><th>Text</th><th>Label</th></tr></thead>"
    for text, label in rows:
        html += "<tr><td>%s</td><td>%s</td></tr>" % (text, label)
    html += "</table>"

    display(HTML(html))

# Create labels model
labels = Labels()

将标签应用于文本

data = ["Wears a red suit and says ho ho",
        "Pulls a flying sleigh",
        "This is cut down and decorated",
        "Santa puts these under the tree",
        "Best way to spend the holidays"]

# List of labels
tags = [" Santa Clause", "歷 Reindeer", " Cookies", " Christmas Tree", " Gifts", " Family"]

# Render output to table
table([(text, tags[labels(text, tags)[0][0]]) for text in data])

文本	标签
穿着红色西装说ho ho	圣诞老人
拉着飞行的雪橇	歷 驯鹿
这是削减和装饰	圣诞树
圣诞老人把这些放在树下	 礼物
度过假期的最佳方式	家人

我们再次看到零样本标记的力量。该模型未针对此示例的任何特定数据进行训练。仍然对大型 NLP 模型中存储了多少知识感到惊讶。

启动 API 实例

现在我们将启动一个 API 实例来运行其余的示例。API 需要一个配置文件才能运行。下面的示例已简化为仅包含标签。有关更详细的配置示例，请参阅此链接。

API 实例在后台启动。

CONFIG=index.yml nohup uvicorn "txtai.api:app" &> api.log & sleep 90

JavaScript

txtai.js 可通过 NPM 获得，可以按如下方式安装。

npm install txtai

对于此示例，我们将克隆 txtai.js 项目以导入示例构建配置。

git clone https://github.com/neuml/txtai.js

创建标签.js

以下文件是标签示例的 JavaScript 版本。

import {Labels} from "txtai";
import {sprintf} from "sprintf-js";

const run = async () => {
    try {
        let labels = new Labels("http://localhost:8000");

        let data = ["Wears a red suit and says ho ho",
                    "Pulls a flying sleigh",
                    "This is cut down and decorated",
                    "Santa puts these under the tree",
                    "Best way to spend the holidays"];

        // List of labels
        let tags = [" Santa Clause", "歷 Reindeer", " Cookies", " Christmas Tree", " Gifts", " Family"];

        console.log(sprintf("%-40s %s", "Text", "Label"));
        console.log("-".repeat(75))

        for (let text of data) {
            let label = await labels.label(text, tags);
            label = tags[label[0].id];

            console.log(sprintf("%-40s %s", text, label));
        }
    }
    catch (e) {
        console.trace(e);
    }
};

run();

构建和运行标签示例

cd txtai.js/examples/node 
npm install
npm run build

node dist/labels.js

Text                                     Label
---------------------------------------------------------------------------
Wears a red suit and says ho ho           Santa Clause
Pulls a flying sleigh                    歷 Reindeer
This is cut down and decorated            Christmas Tree
Santa puts these under the tree           Gifts
Best way to spend the holidays            Family

JavaScript 程序显示的结果与通过 Python 本地运行时相同！

Java

txtai.java 与标准 Java 构建工具（Gradle、Maven、SBT）集成。下面展示了如何将 txtai 作为依赖添加到 Gradle。

implementation 'com.github.neuml:txtai.java:v2.0.0'

对于此示例，我们将克隆 txtai.java 项目以导入示例构建配置。

git clone https://github.com/neuml/txtai.java

创建 LabelsDemo.java

以下文件是标签示例的 Java 版本。

import java.util.Arrays;
import java.util.ArrayList;
import java.util.List;

import txtai.API.IndexResult;
import txtai.Labels;

public class LabelsDemo {
    public static void main(String[] args) {
        try {
            Labels labels = new Labels("http://localhost:8000");

            List <String> data = 
                Arrays.asList("Wears a red suit and says ho ho",
                              "Pulls a flying sleigh",
                              "This is cut down and decorated",
                              "Santa puts these under the tree",
                              "Best way to spend the holidays");

            // List of labels
            List<String> tags = Arrays.asList(" Santa Clause", "歷 Reindeer", " Cookies", " Christmas Tree", " Gifts", " Family");

            System.out.printf("%-40s %s%n", "Text", "Label");
            System.out.println(new String(new char[75]).replace("\0", "-"));

            for (String text: data) {
                List<IndexResult> label = labels.label(text, tags);
                System.out.printf("%-40s %s%n", text, tags.get(label.get(0).id));
            }
        }
        catch (Exception ex) {
            ex.printStackTrace();
        }
    }
}

cd txtai.java/examples
../gradlew -q --console=plain labels 2> /dev/null

Text                                     Label
---------------------------------------------------------------------------
Wears a red suit and says ho ho           Santa Clause
Pulls a flying sleigh                    歷 Reindeer
This is cut down and decorated            Christmas Tree
Santa puts these under the tree           Gifts
Best way to spend the holidays            Family

Java 程序显示的结果与通过 Python 本地运行时相同！

Rust

txtai.rs 可以通过 crates.io 获得，并且可以通过将以下内容添加到您的 cargo.toml 文件来安装

[dependencies]
txtai = { version = "2.0" }
tokio = { version = "0.2", features = ["full"] }

对于此示例，我们将克隆 txtai.rs 项目以导入示例构建配置。首先我们需要安装 Rust。

apt-get install rustc
git clone https://github.com/neuml/txtai.rs

创建balel.rs

以下文件是标签示例的 Rust 版本。

use std::error::Error;

use txtai::labels::Labels;

pub async fn labels() -> Result<(), Box<dyn Error>> {
    let labels = Labels::new("http://localhost:8000");

    let data = ["Wears a red suit and says ho ho",
                "Pulls a flying sleigh",
                "This is cut down and decorated",
                "Santa puts these under the tree",
                "Best way to spend the holidays"];

    println!("{:<40} {}", "Text", "Label");
    println!("{}", "-".repeat(75));

    for text in data.iter() {
        let tags = vec![" Santa Clause", "歷 Reindeer", " Cookies", " Christmas Tree", " Gifts", " Family"];
        let label = labels.label(text, &tags).await?[0].id;

        println!("{:<40} {}", text, tags[label]);
    }

    Ok(())
}

构建和运行标签示例

cd txtai.rs/examples/demo
cargo build

cargo run labels

Text                                     Label
--------------------------------------------------------------------------------
Wears a red suit and says ho ho           Santa Clause
Pulls a flying sleigh                    歷 Reindeer
This is cut down and decorated            Christmas Tree
Santa puts these under the tree           Gifts
Best way to spend the holidays            Family

Rust 程序显示的结果与通过 Python 本地运行时相同！

GO

txtai.go 可以通过添加以下导入语句来安装。使用模块时，会自动安装txtai.go。否则使用go get.

import "github.com/neuml/txtai.go"

对于此示例，我们将创建一个独立的标签流程。首先我们需要安装 Go。

apt install golang-go
go get "github.com/neuml/txtai.go"

创建label.go

以下文件是标签示例的 Go 版本。

package main

import (
    "fmt"
    "strings"
    "github.com/neuml/txtai.go"
)

func main() {
    labels := txtai.Labels("http://localhost:8000")

    data := []string{"Wears a red suit and says ho ho",
                   "Pulls a flying sleigh",
                   "This is cut down and decorated",
                   "Santa puts these under the tree",
                   "Best way to spend the holidays"}

    // List of labels
    tags := []string{" Santa Clause", "歷 Reindeer", " Cookies", " Christmas Tree", " Gifts", " Family"}

    fmt.Printf("%-40s %s\n", "Text", "Label")
    fmt.Println(strings.Repeat("-", 75))

    for _, text := range data {
        label := labels.Label(text, tags)
        fmt.Printf("%-40s %s\n", text, tags[label[0].Id])
    }
}

go run labels.go

Text                                     Label
--------------------------------------------------------------------------------
Wears a red suit and says ho ho           Santa Clause
Pulls a flying sleigh                    歷 Reindeer
This is cut down and decorated            Christmas Tree
Santa puts these under the tree           Gifts
Best way to spend the holidays            Family

参考

https://dev.to/neuml/tutorial-series-on-txtai-ibg