core[minor]: Add XML output parser (#4258)

* core[minor]: Add XML output parser

* cr

* docs

* chore: lint files

* cr

* streaming & docs

* cr

* chore: lint files

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

@@ -0,0 +1,29 @@
+# XML output parser
+
+The `XMLOutputParser` takes language model output which contains XML and parses it into a JSON object.
+
+The output parser also supports streaming outputs.
+
+Currently, the XML parser does not contain support for self closing tags, or attributes on tags.
+
+## Usage
+
+import CodeBlock from "@theme/CodeBlock";
+
+import IntegrationInstallTooltip from "@mdx_components/integration_install_tooltip.mdx";
+
+<IntegrationInstallTooltip></IntegrationInstallTooltip>
+
+```bash npm2yarn
+npm install @langchain/core
+```
+
+import XMLExample from "@examples/prompts/xml_output_parser.ts";
+
+<CodeBlock language="typescript">{XMLExample}</CodeBlock>
+
+## Streaming
+
+import XMLStreamingExample from "@examples/prompts/xml_output_parser_streaming.ts";
+
+<CodeBlock language="typescript">{XMLStreamingExample}</CodeBlock>

examples/src/prompts/xml_output_parser.ts

@@ -0,0 +1,67 @@
+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>`;
+
+const parser = new XMLOutputParser();
+
+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"
+        },
+        {
+          "role": "User"
+        }
+      ]
+    },
+    {
+      "preferences": [
+        {
+          "theme": "Dark"
+        },
+        {
+          "notifications": [
+            {
+              "email": "true"
+            },
+            {
+              "sms": "false"
+            }
+          ]
+        }
+      ]
+    }
+  ]
+}
+*/

examples/src/prompts/xml_output_parser_streaming.ts

@@ -0,0 +1,118 @@
+import { XMLOutputParser } from "@langchain/core/output_parsers";
+import { FakeStreamingLLM } from "@langchain/core/utils/testing";
+
+const XML_EXAMPLE = `<?xml version="1.0" encoding="UTF-8"?>
+<userProfile>
+  <userID>12345</userID>
+  <roles>
+    <role>Admin</role>
+    <role>User</role>
+  </roles>
+</userProfile>`;
+
+const parser = new XMLOutputParser();
+
+// Define your LLM, in this example we'll use demo streaming LLM
+const streamingLLM = new FakeStreamingLLM({
+  responses: [XML_EXAMPLE],
+}).pipe(parser); // Pipe the parser to the LLM
+
+const stream = await streamingLLM.stream(XML_EXAMPLE);
+for await (const chunk of stream) {
+  console.log(JSON.stringify(chunk, null, 2));
+}
+/*
+{}
+{
+  "userProfile": ""
+}
+{
+  "userProfile": "\n"
+}
+{
+  "userProfile": [
+    {
+      "userID": ""
+    }
+  ]
+}
+{
+  "userProfile": [
+    {
+      "userID": "123"
+    }
+  ]
+}
+{
+  "userProfile": [
+    {
+      "userID": "12345"
+    },
+    {
+      "roles": ""
+    }
+  ]
+}
+{
+  "userProfile": [
+    {
+      "userID": "12345"
+    },
+    {
+      "roles": [
+        {
+          "role": "A"
+        }
+      ]
+    }
+  ]
+}
+{
+  "userProfile": [
+    {
+      "userID": "12345"
+    },
+    {
+      "roles": [
+        {
+          "role": "Admi"
+        }
+      ]
+    }
+  ]
+}
+{
+  "userProfile": [
+    {
+      "userID": "12345"
+    },
+    {
+      "roles": [
+        {
+          "role": "Admin"
+        },
+        {
+          "role": "U"
+        }
+      ]
+    }
+  ]
+}
+{
+  "userProfile": [
+    {
+      "userID": "12345"
+    },
+    {
+      "roles": [
+        {
+          "role": "Admin"
+        },
+        {
+          "role": "User"
+        }
+      ]
+    }
+  ]
+}
+*/

langchain-core/package.json

@@ -46,6 +46,7 @@
     "ml-distance": "^4.0.0",
     "p-queue": "^6.6.2",
     "p-retry": "4",
+    "sax": "^1.3.0",
     "uuid": "^9.0.0",
     "zod": "^3.22.4",
     "zod-to-json-schema": "^3.22.3"
@@ -55,6 +56,7 @@
     "@langchain/scripts": "~0.0",
     "@swc/core": "^1.3.90",
     "@swc/jest": "^0.2.29",
+    "@types/sax": "^1",
     "dpdm": "^3.12.0",
     "eslint": "^8.33.0",
     "eslint-config-airbnb-base": "^15.0.0",

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 +1,93 @@
+import { FakeStreamingLLM } from "../../utils/testing/index.js";
+import { XMLOutputParser } from "../xml.js";
+
+const XML_EXAMPLE = `<?xml version="1.0" encoding="UTF-8"?>
+<userProfile>
+  <userID>12345</userID>
+  <email>john.doe@example.com</email>
+  <roles>
+    <role>Admin</role>
+    <role>User</role>
+  </roles>
+  <preferences>
+    <theme>Dark</theme>
+    <notifications>
+      <email>true</email>
+    </notifications>
+  </preferences>
+</userProfile>`;
+
+const BACKTICK_WRAPPED_XML = `\`\`\`xml\n${XML_EXAMPLE}\n\`\`\``;
+
+const expectedResult = {
+  userProfile: [
+    {
+      userID: "12345",
+    },
+    {
+      email: "john.doe@example.com",
+    },
+    {
+      roles: [
+        {
+          role: "Admin",
+        },
+        {
+          role: "User",
+        },
+      ],
+    },
+    {
+      preferences: [
+        {
+          theme: "Dark",
+        },
+        {
+          notifications: [
+            {
+              email: "true",
+            },
+          ],
+        },
+      ],
+    },
+  ],
+};
+
+test("Can parse XML", async () => {
+  const parser = new XMLOutputParser();
+
+  const result = await parser.invoke(XML_EXAMPLE);
+  expect(result).toStrictEqual(expectedResult);
+});
+
+test("Can parse backtick wrapped XML", async () => {
+  const parser = new XMLOutputParser();
+
+  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({ tags });
+
+  const formatInstructions = parser.getFormatInstructions();
+
+  expect(formatInstructions).toContain("tag1, tag2, tag3");
+});
+
+test("Can parse streams", async () => {
+  const parser = new XMLOutputParser();
+  const streamingLlm = new FakeStreamingLLM({
+    responses: [XML_EXAMPLE],
+  }).pipe(parser);
+
+  const result = await streamingLlm.stream(XML_EXAMPLE);
+  let finalResult = {};
+  for await (const chunk of result) {
+    console.log(chunk);
+    finalResult = chunk;
+  }
+  expect(finalResult).toStrictEqual(expectedResult);
+});