core[minor]: Add XML output parser

docs/core_docs/docs/modules/model_io/output_parsers/types/xml.mdx

@@ -0,0 +1,20 @@
+# XML output parser
+
+The `XMLOutputParser` takes language model output which contains XML and parses it into a JSON object.
+
+The output parser currently does _not_ support streaming results.
+
+## Usage
+
+import CodeBlock from "@theme/CodeBlock";
+import XMLExample from "@examples/prompts/xml_output_parser.ts";
+
+import IntegrationInstallTooltip from "@mdx_components/integration_install_tooltip.mdx";
+
+<IntegrationInstallTooltip></IntegrationInstallTooltip>
+
+```bash npm2yarn
+npm install @langchain/core
+```
+
+<CodeBlock language="typescript">{XMLExample}</CodeBlock>

examples/src/prompts/xml_output_parser.ts

@@ -0,0 +1,64 @@
+import { XMLOutputParser } from "@langchain/core/output_parsers";
+
+const XML_EXAMPLE = `<?xml version="1.0" encoding="UTF-8"?>
+<userProfile>
+  <userID>12345</userID>
+  <name>John Doe</name>
+  <email>john.doe@example.com</email>
+  <roles>
+    <role>Admin</role>
+    <role>User</role>
+  </roles>
+  <preferences>
+    <theme>Dark</theme>
+    <notifications>
+      <email>true</email>
+      <sms>false</sms>
+    </notifications>
+  </preferences>
+</userProfile>`;
+
+type MySchema = {
+  userProfile: {
+    userID: number;
+    name: string;
+    email: string;
+    roles: { role: string[] };
+    preferences: {
+      theme: string;
+      notifications: {
+        email: boolean;
+        sms: boolean;
+      };
+    };
+  };
+};
+
+// Pass in a generic type for the schema
+const parser = new XMLOutputParser<MySchema>();
+
+const result = await parser.invoke(XML_EXAMPLE);
+
+console.log(JSON.stringify(result, null, 2));
+/*
+{
+  "userProfile": {
+    "userID": 12345,
+    "name": "John Doe",
+    "email": "john.doe@example.com",
+    "roles": {
+      "role": [
+        "Admin",
+        "User"
+      ]
+    },
+    "preferences": {
+      "theme": "Dark",
+      "notifications": {
+        "email": true,
+        "sms": false
+      }
+    }
+  }
+}
+*/

langchain-core/package.json

@@ -40,6 +40,7 @@
     "ansi-styles": "^5.0.0",
     "camelcase": "6",
     "decamelize": "1.2.0",
+    "fast-xml-parser": "^4.3.4",
     "js-tiktoken": "^1.0.8",
     "langsmith": "~0.0.48",
     "ml-distance": "^4.0.0",

langchain-core/src/output_parsers/index.ts

@@ -4,3 +4,4 @@ export * from "./list.js";
 export * from "./string.js";
 export * from "./transform.js";
 export * from "./json.js";
+export * from "./xml.js";

langchain-core/src/output_parsers/json.ts

@@ -53,7 +53,7 @@ export function parseJsonMarkdown(s: string, parser = parsePartialJson) {
   }
 }
 
-// Adapted from https://github.com/KillianLucas/open-interpreter/blob/main/interpreter/utils/parse_partial_json.py
+// Adapted from https://github.com/KillianLucas/open-interpreter/blob/main/interpreter/core/llm/utils/parse_partial_json.py
 // MIT License
 export function parsePartialJson(s: string) {
   // If the input is undefined, return null to indicate failure.

langchain-core/src/output_parsers/tests/xml.test.ts

@@ -0,0 +1,76 @@
+import { XMLOutputParser } from "../xml.js";
+
+const XML_EXAMPLE = `<?xml version="1.0" encoding="UTF-8"?>
+<userProfile>
+  <userID>12345</userID>
+  <name>John Doe</name>
+  <email>john.doe@example.com</email>
+  <roles>
+    <role>Admin</role>
+    <role>User</role>
+  </roles>
+  <preferences>
+    <theme>Dark</theme>
+    <notifications>
+      <email>true</email>
+      <sms>false</sms>
+    </notifications>
+  </preferences>
+</userProfile>`;
+
+const BACKTICK_WRAPPED_XML = `\`\`\`xml\n${XML_EXAMPLE}\n\`\`\``;
+
+type MySchema = {
+  userProfile: {
+    userID: number;
+    name: string;
+    email: string;
+    roles: { role: string[] };
+    preferences: {
+      theme: string;
+      notifications: {
+        email: boolean;
+        sms: boolean;
+      };
+    };
+  };
+};
+
+const expectedResult = {
+  userProfile: {
+    userID: 12345,
+    name: "John Doe",
+    email: "john.doe@example.com",
+    roles: { role: ["Admin", "User"] },
+    preferences: {
+      theme: "Dark",
+      notifications: {
+        email: true,
+        sms: false,
+      },
+    },
+  },
+};
+
+test("Can parse XML", async () => {
+  const parser = new XMLOutputParser<MySchema>();
+
+  const result = await parser.invoke(XML_EXAMPLE);
+  expect(result).toStrictEqual(expectedResult);
+});
+
+test("Can parse backtick wrapped XML", async () => {
+  const parser = new XMLOutputParser<MySchema>();
+
+  const result = await parser.invoke(BACKTICK_WRAPPED_XML);
+  expect(result).toStrictEqual(expectedResult);
+});
+
+test("Can format instructions with passed tags.", async () => {
+  const tags = ["tag1", "tag2", "tag3"];
+  const parser = new XMLOutputParser<MySchema>({ tags });
+
+  const formatInstructions = parser.getFormatInstructions();
+
+  expect(formatInstructions).toContain("tag1, tag2, tag3");
+});

langchain-core/src/output_parsers/xml.ts

@@ -0,0 +1,78 @@
+import { XMLParser } from "fast-xml-parser";
+import { BaseOutputParser } from "./base.js";
+
+export const XML_FORMAT_INSTRUCTIONS = `The output should be formatted as a XML file.
+1. Output should conform to the tags below. 
+2. If tags are not given, make them on your own.
+3. Remember to always open and close all the tags.
+
+As an example, for the tags ["foo", "bar", "baz"]:
+1. String "<foo>\n   <bar>\n      <baz></baz>\n   </bar>\n</foo>" is a well-formatted instance of the schema. 
+2. String "<foo>\n   <bar>\n   </foo>" is a badly-formatted instance.
+3. String "<foo>\n   <tag>\n   </tag>\n</foo>" is a badly-formatted instance.
+
+Here are the output tags:
+\`\`\
+{tags}
+\`\`\``;
+
+export interface XMLOutputParserFields {
+  /**
+   * Optional list of tags that the output should conform to.
+   * Only used in formatting of the prompt.
+   */
+  tags?: string[];
+}
+
+export class XMLOutputParser<
+  // eslint-disable-next-line @typescript-eslint/no-explicit-any
+  T extends Record<string, any> = Record<string, any>
+> extends BaseOutputParser<T> {
+  tags?: string[];
+
+  constructor(fields?: XMLOutputParserFields) {
+    super();
+    this.tags = fields?.tags;
+  }
+
+  static lc_name() {
+    return "XMLOutputParser";
+  }
+
+  lc_namespace = ["langchain_core", "output_parsers"];
+
+  lc_serializable = true;
+
+  async parse(text: string): Promise<T> {
+    return parseXMLMarkdown<T>(text);
+  }
+
+  getFormatInstructions(): string {
+    const withTags = !!(this.tags && this.tags.length > 0);
+    return withTags
+      ? XML_FORMAT_INSTRUCTIONS.replace("{tags}", this.tags?.join(", ") ?? "")
+      : XML_FORMAT_INSTRUCTIONS;
+  }
+}
+
+export function parseXMLMarkdown<
+  // eslint-disable-next-line @typescript-eslint/no-explicit-any
+  T extends Record<string, any> = Record<string, any>
+>(s: string) {
+  const parser = new XMLParser();
+  const newString = s.trim();
+  // Try to find XML string within triple backticks.
+  const match = /```(xml)?(.*)```/s.exec(newString);
+  let parsedResult: T;
+  if (!match) {
+    // If match found, use the content within the backticks
+    parsedResult = parser.parse(newString);
+  } else {
+    parsedResult = parser.parse(match[2]);
+  }
+
+  if (parsedResult && "?xml" in parsedResult) {
+    delete parsedResult["?xml"];
+  }
+  return parsedResult;
+}

yarn.lock

@@ -9127,6 +9127,7 @@ __metadata:
     eslint-plugin-jest: ^27.6.0
     eslint-plugin-no-instanceof: ^1.0.1
     eslint-plugin-prettier: ^4.2.1
+    fast-xml-parser: ^4.3.4
     jest: ^29.5.0
     jest-environment-node: ^29.6.4
     js-tiktoken: ^1.0.8
@@ -21101,6 +21102,17 @@ __metadata:
   languageName: node
   linkType: hard
 
+"fast-xml-parser@npm:^4.3.4":
+  version: 4.3.4
+  resolution: "fast-xml-parser@npm:4.3.4"
+  dependencies:
+    strnum: ^1.0.5
+  bin:
+    fxparser: src/cli/cli.js
+  checksum: ab88177343f6d3d971d53462db3011003a83eb8a8db704840127ddaaf27105ea90cdf7903a0f9b2e1279ccc4adfca8dfc0277b33bae6262406f10c16bd60ccf9
+  languageName: node
+  linkType: hard
+
 "fastq@npm:^1.6.0":
   version: 1.15.0
   resolution: "fastq@npm:1.15.0"