如何使用提示来改善结果

先决条件

本指南假设你熟悉以下内容

• 基于 SQL 数据的问答

在本指南中，我们将介绍提示策略，以改进 SQL 查询生成。我们将主要关注在提示中获取相关数据库特定信息的方法。

设置

首先，安装所需的软件包并设置你的环境变量。此示例将使用 OpenAI 作为 LLM。

npm install @langchain/community @langchain/openai typeorm sqlite3

export OPENAI_API_KEY="your api key"
# Uncomment the below to use LangSmith. Not required.
# export LANGCHAIN_API_KEY="your api key"
# export LANGCHAIN_TRACING_V2=true

# Reduce tracing latency if you are not in a serverless environment
# export LANGCHAIN_CALLBACKS_BACKGROUND=true

以下示例将使用带有 Chinook 数据库的 SQLite 连接。按照以下 安装步骤 在此笔记本的同一目录中创建 Chinook.db

• 将 此 文件另存为 Chinook_Sqlite.sql
• 运行 sqlite3 Chinook.db
• 运行 .read Chinook_Sqlite.sql
• 测试 SELECT * FROM Artist LIMIT 10;

现在，Chinhook.db 在我们的目录中，我们可以使用 Typeorm 驱动的 SqlDatabase 类与它交互

import { SqlDatabase } from "langchain/sql_db";
import { DataSource } from "typeorm";

const datasource = new DataSource({
  type: "sqlite",
  database: "../../../../Chinook.db",
});
const db = await SqlDatabase.fromDataSourceParams({
  appDataSource: datasource,
});
console.log(db.allTables.map((t) => t.tableName));
/**
[
  'Album',       'Artist',
  'Customer',    'Employee',
  'Genre',       'Invoice',
  'InvoiceLine', 'MediaType',
  'Playlist',    'PlaylistTrack',
  'Track'
]
 */

API 参考

• SqlDatabase 来自 langchain/sql_db

特定于方言的提示

我们可以做到的最简单的事情之一是使我们的提示特定于我们使用的 SQL 方言。当使用内置的 createSqlQueryChain 和 SqlDatabase 时，这将为你处理以下任何方言

import { SQL_PROMPTS_MAP } from "langchain/chains/sql_db";

console.log({ SQL_PROMPTS_MAP: Object.keys(SQL_PROMPTS_MAP) });
/**
{
  SQL_PROMPTS_MAP: [ 'oracle', 'postgres', 'sqlite', 'mysql', 'mssql', 'sap hana' ]
}
 */

// For example, using our current DB we can see that we’ll get a SQLite-specific prompt:

console.log({
  sqlite: SQL_PROMPTS_MAP.sqlite,
});
/**
{
  sqlite: PromptTemplate {
    inputVariables: [ 'dialect', 'table_info', 'input', 'top_k' ],
    template: 'You are a SQLite expert. Given an input question, first create a syntactically correct SQLite query to run, then look at the results of the query and return the answer to the input question.\n' +
      'Unless the user specifies in the question a specific number of examples to obtain, query for at most {top_k} results using the LIMIT clause as per SQLite. You can order the results to return the most informative data in the database.\n' +
      'Never query for all columns from a table. You must query only the columns that are needed to answer the question. Wrap each column name in double quotes (") to denote them as delimited identifiers.\n' +
      'Pay attention to use only the column names you can see in the tables below. Be careful to not query for columns that do not exist. Also, pay attention to which column is in which table.\n' +
      '\n' +
      'Use the following format:\n' +
      '\n' +
      'Question: "Question here"\n' +
      'SQLQuery: "SQL Query to run"\n' +
      'SQLResult: "Result of the SQLQuery"\n' +
      'Answer: "Final answer here"\n' +
      '\n' +
      'Only use the following tables:\n' +
      '{table_info}\n' +
      '\n' +
      'Question: {input}',
  }
}
 */

API 参考

• SQL_PROMPTS_MAP 来自 langchain/chains/sql_db

表定义和示例行

基本上在任何 SQL 链中，我们都需要至少向模型提供部分数据库模式。没有它，它将无法编写有效的查询。我们的数据库附带了一些方便的方法，可以让我们获得相关的上下文。具体来说，我们可以获取表名、它们的模式以及每个表的示例行

import { db } from "../db.js";

const context = await db.getTableInfo();

console.log(context);

/**
 
CREATE TABLE Album (
  AlbumId INTEGER NOT NULL,
  Title NVARCHAR(160) NOT NULL,
  ArtistId INTEGER NOT NULL
)

SELECT * FROM "Album" LIMIT 3;
 AlbumId Title ArtistId
 1 For Those About To Rock We Salute You 1
 2 Balls to the Wall 2
 3 Restless and Wild 2

CREATE TABLE Artist (
  ArtistId INTEGER NOT NULL,
  Name NVARCHAR(120)
)

SELECT * FROM "Artist" LIMIT 3;
 ArtistId Name
 1 AC/DC
 2 Accept
 3 Aerosmith

CREATE TABLE Customer (
  CustomerId INTEGER NOT NULL,
  FirstName NVARCHAR(40) NOT NULL,
  LastName NVARCHAR(20) NOT NULL,
  Company NVARCHAR(80),
  Address NVARCHAR(70),
  City NVARCHAR(40),
  State NVARCHAR(40),
  Country NVARCHAR(40),
  PostalCode NVARCHAR(10),
  Phone NVARCHAR(24),
  Fax NVARCHAR(24),
  Email NVARCHAR(60) NOT NULL,
  SupportRepId INTEGER
)

SELECT * FROM "Customer" LIMIT 3;
 CustomerId FirstName LastName Company Address City State Country PostalCode Phone Fax Email SupportRepId
 1 Luís Gonçalves Embraer - Empresa Brasileira de Aeronáutica S.A. Av. Brigadeiro Faria Lima,
2170 São José dos Campos SP Brazil 12227-000 +55 (12) 3923-5555 +55 (12) 3923-5566 [email protected] 3
 2 Leonie Köhler null Theodor-Heuss-Straße 34 Stuttgart null Germany 70174 +49 0711 2842222 null [email protected] 5
 3 François Tremblay null 1498 rue Bélanger Montréal QC Canada H2G 1A7 +1 (514) 721-4711 null [email protected] 3

CREATE TABLE Employee (
  EmployeeId INTEGER NOT NULL,
  LastName NVARCHAR(20) NOT NULL,
  FirstName NVARCHAR(20) NOT NULL,
  Title NVARCHAR(30),
  ReportsTo INTEGER,
  BirthDate DATETIME,
  HireDate DATETIME,
  Address NVARCHAR(70),
  City NVARCHAR(40),
  State NVARCHAR(40),
  Country NVARCHAR(40),
  PostalCode NVARCHAR(10),
  Phone NVARCHAR(24),
  Fax NVARCHAR(24),
  Email NVARCHAR(60)
)

SELECT * FROM "Employee" LIMIT 3;
 EmployeeId LastName FirstName Title ReportsTo BirthDate HireDate Address City State Country PostalCode Phone Fax Email
 1 Adams Andrew General Manager null 1962-02-18 00:00:00 2002-08-14 00:00:00 11120 Jasper Ave NW Edmonton AB Canada T5K 2N1 +1 (780) 428-9482 +1 (780) 428-3457 [email protected]
 2 Edwards Nancy Sales Manager 1 1958-12-08 00:00:00 2002-05-01 00:00:00 825 8 Ave SW Calgary AB Canada T2P 2T3 +1 (403) 262-3443 +1 (403) 262-3322 [email protected]
 3 Peacock Jane Sales Support Agent 2 1973-08-29 00:00:00 2002-04-01 00:00:00 1111 6 Ave SW Calgary AB Canada T2P 5M5 +1 (403) 262-3443 +1 (403) 262-6712 [email protected]

CREATE TABLE Genre (
  GenreId INTEGER NOT NULL,
  Name NVARCHAR(120)
)

SELECT * FROM "Genre" LIMIT 3;
 GenreId Name
 1 Rock
 2 Jazz
 3 Metal

CREATE TABLE Invoice (
  InvoiceId INTEGER NOT NULL,
  CustomerId INTEGER NOT NULL,
  InvoiceDate DATETIME NOT NULL,
  BillingAddress NVARCHAR(70),
  BillingCity NVARCHAR(40),
  BillingState NVARCHAR(40),
  BillingCountry NVARCHAR(40),
  BillingPostalCode NVARCHAR(10),
  Total NUMERIC(10,2) NOT NULL
)

SELECT * FROM "Invoice" LIMIT 3;
 InvoiceId CustomerId InvoiceDate BillingAddress BillingCity BillingState BillingCountry BillingPostalCode Total
 1 2 2009-01-01 00:00:00 Theodor-Heuss-Straße 34 Stuttgart null Germany 70174 1.98
 2 4 2009-01-02 00:00:00 Ullevålsveien 14 Oslo null Norway 0171 3.96
 3 8 2009-01-03 00:00:00 Grétrystraat 63 Brussels null Belgium 1000 5.94

CREATE TABLE InvoiceLine (
  InvoiceLineId INTEGER NOT NULL,
  InvoiceId INTEGER NOT NULL,
  TrackId INTEGER NOT NULL,
  UnitPrice NUMERIC(10,2) NOT NULL,
  Quantity INTEGER NOT NULL
)

SELECT * FROM "InvoiceLine" LIMIT 3;
 InvoiceLineId InvoiceId TrackId UnitPrice Quantity
 1 1 2 0.99 1
 2 1 4 0.99 1
 3 2 6 0.99 1

CREATE TABLE MediaType (
  MediaTypeId INTEGER NOT NULL,
  Name NVARCHAR(120)
)

SELECT * FROM "MediaType" LIMIT 3;
 MediaTypeId Name
 1 MPEG audio file
 2 Protected AAC audio file
 3 Protected MPEG-4 video file

CREATE TABLE Playlist (
  PlaylistId INTEGER NOT NULL,
  Name NVARCHAR(120)
)

SELECT * FROM "Playlist" LIMIT 3;
 PlaylistId Name
 1 Music
 2 Movies
 3 TV Shows
CREATE TABLE PlaylistTrack (
  PlaylistId INTEGER NOT NULL,
  TrackId INTEGER NOT NULL
)

SELECT * FROM "PlaylistTrack" LIMIT 3;
 PlaylistId TrackId
 1 3402
 1 3389
 1 3390

CREATE TABLE Track (
  TrackId INTEGER NOT NULL,
  Name NVARCHAR(200) NOT NULL,
  AlbumId INTEGER,
  MediaTypeId INTEGER NOT NULL,
  GenreId INTEGER,
  Composer NVARCHAR(220),
  Milliseconds INTEGER NOT NULL,
  Bytes INTEGER,
  UnitPrice NUMERIC(10,2) NOT NULL
)

SELECT * FROM "Track" LIMIT 3;
 TrackId Name AlbumId MediaTypeId GenreId Composer Milliseconds Bytes UnitPrice
 1 For Those About To Rock (We Salute You) 1 1 1 Angus Young,
Malcolm Young,
Brian Johnson 343719 11170334 0.99
 2 Balls to the Wall 2 2 1 U. Dirkschneider,
W. Hoffmann,
H. Frank,
P. Baltes,
S. Kaufmann,
G. Hoffmann 342562 5510424 0.99
 3 Fast As a Shark 3 2 1 F. Baltes,
S. Kaufman,
U. Dirkscneider & W. Hoffman 230619 3990994 0.99
 
*/

API 参考

少量样本示例

在提示中包含将自然语言问题转换为针对我们数据库的有效 SQL 查询的示例通常会提高模型性能，特别是对于复杂的查询。

假设我们有以下示例

export const examples = [
  { input: "List all artists.", query: "SELECT * FROM Artist;" },
  {
    input: "Find all albums for the artist 'AC/DC'.",
    query:
      "SELECT * FROM Album WHERE ArtistId = (SELECT ArtistId FROM Artist WHERE Name = 'AC/DC');",
  },
  {
    input: "List all tracks in the 'Rock' genre.",
    query:
      "SELECT * FROM Track WHERE GenreId = (SELECT GenreId FROM Genre WHERE Name = 'Rock');",
  },
  {
    input: "Find the total duration of all tracks.",
    query: "SELECT SUM(Milliseconds) FROM Track;",
  },
  {
    input: "List all customers from Canada.",
    query: "SELECT * FROM Customer WHERE Country = 'Canada';",
  },
  {
    input: "How many tracks are there in the album with ID 5?",
    query: "SELECT COUNT(*) FROM Track WHERE AlbumId = 5;",
  },
  {
    input: "Find the total number of invoices.",
    query: "SELECT COUNT(*) FROM Invoice;",
  },
  {
    input: "List all tracks that are longer than 5 minutes.",
    query: "SELECT * FROM Track WHERE Milliseconds > 300000;",
  },
  {
    input: "Who are the top 5 customers by total purchase?",
    query:
      "SELECT CustomerId, SUM(Total) AS TotalPurchase FROM Invoice GROUP BY CustomerId ORDER BY TotalPurchase DESC LIMIT 5;",
  },
  {
    input: "Which albums are from the year 2000?",
    query: "SELECT * FROM Album WHERE strftime('%Y', ReleaseDate) = '2000';",
  },
  {
    input: "How many employees are there",
    query: 'SELECT COUNT(*) FROM "Employee"',
  },
];

API 参考

我们可以使用它们创建一个少量样本提示，如下所示

import { FewShotPromptTemplate, PromptTemplate } from "@langchain/core/prompts";
import { examples } from "./examples.js";

const examplePrompt = PromptTemplate.fromTemplate(
  `User input: {input}\nSQL Query: {query}`
);

const prompt = new FewShotPromptTemplate({
  examples: examples.slice(0, 5),
  examplePrompt,
  prefix: `You are a SQLite expert. Given an input question, create a syntactically correct SQLite query to run.
Unless otherwise specified, do not return more than {top_k} rows.

Here is the relevant table info: {table_info}

Below are a number of examples of questions and their corresponding SQL queries.`,
  suffix: "User input: {input}\nSQL query: ",
  inputVariables: ["input", "top_k", "table_info"],
});

console.log(
  await prompt.format({
    input: "How many artists are there?",
    top_k: "3",
    table_info: "foo",
  })
);

/**
You are a SQLite expert. Given an input question, create a syntactically correct SQLite query to run.
Unless otherwise specified, do not return more than 3 rows.

Here is the relevant table info: foo

Below are a number of examples of questions and their corresponding SQL queries.

User input: List all artists.
SQL Query: SELECT * FROM Artist;

User input: Find all albums for the artist 'AC/DC'.
SQL Query: SELECT * FROM Album WHERE ArtistId = (SELECT ArtistId FROM Artist WHERE Name = 'AC/DC');

User input: List all tracks in the 'Rock' genre.
SQL Query: SELECT * FROM Track WHERE GenreId = (SELECT GenreId FROM Genre WHERE Name = 'Rock');

User input: Find the total duration of all tracks.
SQL Query: SELECT SUM(Milliseconds) FROM Track;

User input: List all customers from Canada.
SQL Query: SELECT * FROM Customer WHERE Country = 'Canada';

User input: How many artists are there?
SQL query:
 */

API 参考

• FewShotPromptTemplate 来自 @langchain/core/prompts
• PromptTemplate 来自 @langchain/core/prompts

动态少量样本示例

如果我们有足够的示例，我们可能希望只在提示中包含最相关的示例，要么是因为它们不适合模型的上下文窗口，要么是因为示例的尾部会分散模型的注意力。具体来说，给定任何输入，我们希望包含与该输入最相关的示例。

我们可以使用 ExampleSelector 来做到这一点。在这种情况下，我们将使用 SemanticSimilarityExampleSelector，它将在我们选择的向量数据库中存储示例。在运行时，它将在输入和我们的示例之间执行相似度搜索，并返回语义上最相似的示例

import { MemoryVectorStore } from "langchain/vectorstores/memory";
import { SemanticSimilarityExampleSelector } from "@langchain/core/example_selectors";
import { ChatOpenAI, OpenAIEmbeddings } from "@langchain/openai";
import { FewShotPromptTemplate, PromptTemplate } from "@langchain/core/prompts";
import { createSqlQueryChain } from "langchain/chains/sql_db";
import { examples } from "./examples.js";
import { db } from "../db.js";

const exampleSelector = await SemanticSimilarityExampleSelector.fromExamples<
  typeof MemoryVectorStore
>(examples, new OpenAIEmbeddings(), MemoryVectorStore, {
  k: 5,
  inputKeys: ["input"],
});

console.log(
  await exampleSelector.selectExamples({ input: "how many artists are there?" })
);
/**
[
  { input: 'List all artists.', query: 'SELECT * FROM Artist;' },
  {
    input: 'How many employees are there',
    query: 'SELECT COUNT(*) FROM "Employee"'
  },
  {
    input: 'How many tracks are there in the album with ID 5?',
    query: 'SELECT COUNT(*) FROM Track WHERE AlbumId = 5;'
  },
  {
    input: 'Which albums are from the year 2000?',
    query: "SELECT * FROM Album WHERE strftime('%Y', ReleaseDate) = '2000';"
  },
  {
    input: "List all tracks in the 'Rock' genre.",
    query: "SELECT * FROM Track WHERE GenreId = (SELECT GenreId FROM Genre WHERE Name = 'Rock');"
  }
]
 */

// To use it, we can pass the ExampleSelector directly in to our FewShotPromptTemplate:

const examplePrompt = PromptTemplate.fromTemplate(
  `User input: {input}\nSQL Query: {query}`
);

const prompt = new FewShotPromptTemplate({
  exampleSelector,
  examplePrompt,
  prefix: `You are a SQLite expert. Given an input question, create a syntactically correct SQLite query to run.
Unless otherwise specified, do not return more than {top_k} rows.

Here is the relevant table info: {table_info}

Below are a number of examples of questions and their corresponding SQL queries.`,
  suffix: "User input: {input}\nSQL query: ",
  inputVariables: ["input", "top_k", "table_info"],
});

console.log(
  await prompt.format({
    input: "How many artists are there?",
    top_k: "3",
    table_info: "foo",
  })
);
/**
You are a SQLite expert. Given an input question, create a syntactically correct SQLite query to run.
Unless otherwise specified, do not return more than 3 rows.

Here is the relevant table info: foo

Below are a number of examples of questions and their corresponding SQL queries.

User input: List all artists.
SQL Query: SELECT * FROM Artist;

User input: How many employees are there
SQL Query: SELECT COUNT(*) FROM "Employee"

User input: How many tracks are there in the album with ID 5?
SQL Query: SELECT COUNT(*) FROM Track WHERE AlbumId = 5;

User input: Which albums are from the year 2000?
SQL Query: SELECT * FROM Album WHERE strftime('%Y', ReleaseDate) = '2000';

User input: List all tracks in the 'Rock' genre.
SQL Query: SELECT * FROM Track WHERE GenreId = (SELECT GenreId FROM Genre WHERE Name = 'Rock');

User input: How many artists are there?
SQL query:
 */

// Now we can use it in a chain:

const llm = new ChatOpenAI({
  temperature: 0,
});
const chain = await createSqlQueryChain({
  db,
  llm,
  prompt,
  dialect: "sqlite",
});

console.log(await chain.invoke({ question: "how many artists are there?" }));

/**
SELECT COUNT(*) FROM Artist;
 */

API 参考

• MemoryVectorStore 来自 langchain/vectorstores/memory
• SemanticSimilarityExampleSelector 来自 @langchain/core/example_selectors
• ChatOpenAI 来自 @langchain/openai
• OpenAIEmbeddings 来自 @langchain/openai
• FewShotPromptTemplate 来自 @langchain/core/prompts
• PromptTemplate 来自 @langchain/core/prompts
• createSqlQueryChain 来自 langchain/chains/sql_db

下一步

您现在已经了解了一些提高 SQL 生成效果的提示策略。

接下来，查看本节中的其他指南，例如 如何查询大型数据库。