File size: 40,273 Bytes
e706de2 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 |
# Output Parsers: Structured Output Extraction
**Part 2: Composition - Lesson 2**
> LLMs return text. You need data.
## Overview
You've learned to create great prompts. LLMs return unstructured text, and in some cases you might need structured data:
```javascript
// LLM returns this:
"The sentiment is positive with a confidence of 0.92"
// You need this:
{
sentiment: "positive",
confidence: 0.92
}
```
**Output parsers** transform LLM text into structured data you can use in your applications.
## Why This Matters
### The Problem: Parsing Chaos
Without parsers, your code is full of brittle string manipulation:
```javascript
const response = await llm.invoke("Classify: I love this product!");
// Fragile parsing code everywhere
if (response.includes("positive")) {
sentiment = "positive";
} else if (response.includes("negative")) {
sentiment = "negative";
}
// What if format changes?
// What if LLM adds extra text?
// How do you handle errors?
```
Problems:
- Brittle regex and string matching
- No validation of output format
- Hard to test parsing logic
- Inconsistent error handling
- Parser code duplicated everywhere
### The Solution: Output Parsers
```javascript
const parser = new JsonOutputParser();
const prompt = new PromptTemplate({
template: `Classify the sentiment. Respond in JSON:
{{"sentiment": "positive/negative/neutral", "confidence": 0.0-1.0}}
Text: {text}`,
inputVariables: ["text"]
});
const chain = prompt.pipe(llm).pipe(parser);
const result = await chain.invoke({ text: "I love this!" });
// { sentiment: "positive", confidence: 0.95 }
```
Benefits:
- ✅ Reliable structured extraction
- ✅ Format validation
- ✅ Error handling built-in
- ✅ Reusable parsing logic
- ✅ Type-safe outputs
## Learning Objectives
By the end of this lesson, you will:
- ✅ Build a BaseOutputParser abstraction
- ✅ Create a StringOutputParser for text cleanup
- ✅ Implement JsonOutputParser for JSON extraction
- ✅ Build ListOutputParser for arrays
- ✅ Create StructuredOutputParser with schemas
- ✅ Use parsers in chains with prompts
- ✅ Handle parsing errors gracefully
## Core Concepts
### What is an Output Parser?
An output parser **transforms LLM text output into structured data**.
**Flow:**
```
LLM Output (text) → Parser → Structured Data
↓ ↓ ↓
"positive: 0.95" parse() {sentiment: "positive", confidence: 0.95}
```
### The Parser Hierarchy
```
BaseOutputParser (abstract)
├── StringOutputParser (clean text)
├── JsonOutputParser (extract JSON)
├── ListOutputParser (extract lists)
├── RegexOutputParser (regex patterns)
└── StructuredOutputParser (schema validation)
```
Each parser handles a specific output format.
### Key Operations
1. **Parse**: Extract structured data from text
2. **Get Format Instructions**: Tell LLM how to format response
3. **Validate**: Check output matches expected structure
4. **Handle Errors**: Gracefully handle malformed outputs
## Implementation Guide
### Step 1: Base Output Parser
**Location:** `src/output-parsers/base-parser.js`
This is the abstract base class all parsers inherit from.
**What it does:**
- Defines the interface for all parsers
- Extends Runnable (so parsers work in chains)
- Provides format instruction generation
- Handles parsing errors
**Implementation:**
```javascript
import { Runnable } from '../core/runnable.js';
/**
* Base class for all output parsers
* Transforms LLM text output into structured data
*/
export class BaseOutputParser extends Runnable {
constructor() {
super();
this.name = this.constructor.name;
}
/**
* Parse the LLM output into structured data
* @abstract
* @param {string} text - Raw LLM output
* @returns {Promise<any>} Parsed data
*/
async parse(text) {
throw new Error(`${this.name} must implement parse()`);
}
/**
* Get instructions for the LLM on how to format output
* @returns {string} Format instructions
*/
getFormatInstructions() {
return '';
}
/**
* Runnable interface: parse the output
*/
async _call(input, config) {
// Input can be a string or a Message
const text = typeof input === 'string'
? input
: input.content;
return await this.parse(text);
}
/**
* Parse with error handling
*/
async parseWithPrompt(text, prompt) {
try {
return await this.parse(text);
} catch (error) {
throw new OutputParserException(
`Failed to parse output from prompt: ${error.message}`,
text,
error
);
}
}
}
/**
* Exception thrown when parsing fails
*/
export class OutputParserException extends Error {
constructor(message, llmOutput, originalError) {
super(message);
this.name = 'OutputParserException';
this.llmOutput = llmOutput;
this.originalError = originalError;
}
}
```
**Key insights:**
- Extends `Runnable` so parsers can be piped in chains
- `_call` extracts text from strings or Messages
- `getFormatInstructions()` helps prompt the LLM
- Error handling wraps parse failures with context
---
### Step 2: String Output Parser
**Location:** `src/output-parsers/string-parser.js`
The simplest parser - cleans up text output.
**What it does:**
- Strips leading/trailing whitespace
- Optionally removes markdown code blocks
- Returns clean string
**Use when:**
- You just need clean text
- No structure needed
- Want to remove formatting artifacts
**Implementation:**
```javascript
import { BaseOutputParser } from './base-parser.js';
/**
* Parser that returns cleaned string output
* Strips whitespace and optionally removes markdown
*
* Example:
* const parser = new StringOutputParser();
* const result = await parser.parse(" Hello World ");
* // "Hello World"
*/
export class StringOutputParser extends BaseOutputParser {
constructor(options = {}) {
super();
this.stripMarkdown = options.stripMarkdown ?? true;
}
/**
* Parse: clean the text
*/
async parse(text) {
let cleaned = text.trim();
if (this.stripMarkdown) {
cleaned = this._stripMarkdownCodeBlocks(cleaned);
}
return cleaned;
}
/**
* Remove markdown code blocks (```code```)
*/
_stripMarkdownCodeBlocks(text) {
// Remove ```language\ncode\n```
return text.replace(/```[\w]*\n([\s\S]*?)\n```/g, '$1').trim();
}
getFormatInstructions() {
return 'Respond with plain text. No markdown formatting.';
}
}
```
**Usage:**
```javascript
const parser = new StringOutputParser();
// Handles various formats
await parser.parse(" Hello "); // "Hello"
await parser.parse("```\ncode\n```"); // "code"
await parser.parse(" \n Text \n "); // "Text"
```
---
### Step 3: JSON Output Parser
**Location:** `src/output-parsers/json-parser.js`
Extracts and validates JSON from LLM output.
**What it does:**
- Finds JSON in text (handles markdown, extra text)
- Parses and validates JSON
- Optionally validates against a schema
**Use when:**
- Need structured objects
- Want type-safe data
- Need validation
**Implementation:**
```javascript
import { BaseOutputParser, OutputParserException } from './base-parser.js';
/**
* Parser that extracts JSON from LLM output
* Handles markdown code blocks and extra text
*
* Example:
* const parser = new JsonOutputParser();
* const result = await parser.parse('```json\n{"name": "Alice"}\n```');
* // { name: "Alice" }
*/
export class JsonOutputParser extends BaseOutputParser {
constructor(options = {}) {
super();
this.schema = options.schema;
}
/**
* Parse JSON from text
*/
async parse(text) {
try {
// Try to extract JSON from the text
const jsonText = this._extractJson(text);
const parsed = JSON.parse(jsonText);
// Validate against schema if provided
if (this.schema) {
this._validateSchema(parsed);
}
return parsed;
} catch (error) {
throw new OutputParserException(
`Failed to parse JSON: ${error.message}`,
text,
error
);
}
}
/**
* Extract JSON from text (handles markdown, extra text)
*/
_extractJson(text) {
// Try direct parse first
try {
JSON.parse(text.trim());
return text.trim();
} catch {
// Not direct JSON, try to find it
}
// Look for JSON in markdown code blocks
const markdownMatch = text.match(/```(?:json)?\s*\n?([\s\S]*?)\n?```/);
if (markdownMatch) {
return markdownMatch[1].trim();
}
// Look for JSON object/array patterns
const jsonObjectMatch = text.match(/\{[\s\S]*\}/);
if (jsonObjectMatch) {
return jsonObjectMatch[0];
}
const jsonArrayMatch = text.match(/\[[\s\S]*\]/);
if (jsonArrayMatch) {
return jsonArrayMatch[0];
}
// Give up, return original
return text.trim();
}
/**
* Validate parsed JSON against schema
*/
_validateSchema(parsed) {
if (!this.schema) return;
for (const [key, type] of Object.entries(this.schema)) {
if (!(key in parsed)) {
throw new Error(`Missing required field: ${key}`);
}
const actualType = typeof parsed[key];
if (actualType !== type) {
throw new Error(
`Field ${key} should be ${type}, got ${actualType}`
);
}
}
}
getFormatInstructions() {
let instructions = 'Respond with valid JSON.';
if (this.schema) {
const schemaDesc = Object.entries(this.schema)
.map(([key, type]) => `"${key}": ${type}`)
.join(', ');
instructions += ` Schema: { ${schemaDesc} }`;
}
return instructions;
}
}
```
**Usage:**
```javascript
const parser = new JsonOutputParser({
schema: {
name: 'string',
age: 'number',
active: 'boolean'
}
});
// Handles various JSON formats
await parser.parse('{"name": "Alice", "age": 30, "active": true}');
await parser.parse('```json\n{"name": "Bob", "age": 25, "active": false}\n```');
await parser.parse('Sure! Here\'s the data: {"name": "Charlie", "age": 35, "active": true}');
```
---
### Step 4: List Output Parser
**Location:** `src/output-parsers/list-parser.js`
Extracts lists/arrays from text.
**What it does:**
- Parses numbered lists, bullet points, comma-separated
- Returns array of items
- Cleans each item
**Use when:**
- Need arrays of strings
- LLM outputs lists
- Want simple arrays
**Implementation:**
```javascript
import { BaseOutputParser } from './base-parser.js';
/**
* Parser that extracts lists from text
* Handles: numbered lists, bullets, comma-separated
*
* Example:
* const parser = new ListOutputParser();
* const result = await parser.parse("1. Apple\n2. Banana\n3. Orange");
* // ["Apple", "Banana", "Orange"]
*/
export class ListOutputParser extends BaseOutputParser {
constructor(options = {}) {
super();
this.separator = options.separator;
}
/**
* Parse list from text
*/
async parse(text) {
const cleaned = text.trim();
// If separator specified, use it
if (this.separator) {
return cleaned
.split(this.separator)
.map(item => item.trim())
.filter(item => item.length > 0);
}
// Try to detect format
if (this._isNumberedList(cleaned)) {
return this._parseNumberedList(cleaned);
}
if (this._isBulletList(cleaned)) {
return this._parseBulletList(cleaned);
}
// Try comma-separated
if (cleaned.includes(',')) {
return cleaned
.split(',')
.map(item => item.trim())
.filter(item => item.length > 0);
}
// Try newline-separated
return cleaned
.split('\n')
.map(item => item.trim())
.filter(item => item.length > 0);
}
/**
* Check if text is numbered list (1. Item\n2. Item)
*/
_isNumberedList(text) {
return /^\d+\./.test(text);
}
/**
* Check if text is bullet list (- Item\n- Item or * Item)
*/
_isBulletList(text) {
return /^[-*•]/.test(text);
}
/**
* Parse numbered list
*/
_parseNumberedList(text) {
return text
.split('\n')
.map(line => line.replace(/^\d+\.\s*/, '').trim())
.filter(item => item.length > 0);
}
/**
* Parse bullet list
*/
_parseBulletList(text) {
return text
.split('\n')
.map(line => line.replace(/^[-*•]\s*/, '').trim())
.filter(item => item.length > 0);
}
getFormatInstructions() {
if (this.separator) {
return `Respond with items separated by "${this.separator}".`;
}
return 'Respond with a numbered list (1. Item) or bullet list (- Item).';
}
}
```
**Usage:**
```javascript
const parser = new ListOutputParser();
// Handles various list formats
await parser.parse("1. Apple\n2. Banana\n3. Orange");
// ["Apple", "Banana", "Orange"]
await parser.parse("- Red\n- Green\n- Blue");
// ["Red", "Green", "Blue"]
await parser.parse("cat, dog, bird");
// ["cat", "dog", "bird"]
// Custom separator
const csvParser = new ListOutputParser({ separator: ',' });
await csvParser.parse("apple,banana,orange");
// ["apple", "banana", "orange"]
```
---
### Step 5: Regex Output Parser
**Location:** `src/output-parsers/regex-parser.js`
Uses regex patterns to extract structured data.
**What it does:**
- Applies regex to extract groups
- Maps groups to field names
- Returns structured object
**Use when:**
- Output has predictable patterns
- Need custom extraction logic
- Regex is simplest solution
**Implementation:**
```javascript
import { BaseOutputParser, OutputParserException } from './base-parser.js';
/**
* Parser that uses regex to extract structured data
*
* Example:
* const parser = new RegexOutputParser({
* regex: /Sentiment: (\w+), Confidence: ([\d.]+)/,
* outputKeys: ["sentiment", "confidence"]
* });
*
* const result = await parser.parse("Sentiment: positive, Confidence: 0.92");
* // { sentiment: "positive", confidence: "0.92" }
*/
export class RegexOutputParser extends BaseOutputParser {
constructor(options = {}) {
super();
this.regex = options.regex;
this.outputKeys = options.outputKeys || [];
this.dotAll = options.dotAll ?? false;
if (this.dotAll) {
// Add 's' flag for dotAll if not present
const flags = this.regex.flags.includes('s')
? this.regex.flags
: this.regex.flags + 's';
this.regex = new RegExp(this.regex.source, flags);
}
}
/**
* Parse using regex
*/
async parse(text) {
const match = text.match(this.regex);
if (!match) {
throw new OutputParserException(
`Text does not match regex pattern: ${this.regex}`,
text
);
}
// If no output keys, return the groups as array
if (this.outputKeys.length === 0) {
return match.slice(1); // Exclude full match
}
// Map groups to keys
const result = {};
for (let i = 0; i < this.outputKeys.length; i++) {
result[this.outputKeys[i]] = match[i + 1]; // +1 to skip full match
}
return result;
}
getFormatInstructions() {
if (this.outputKeys.length > 0) {
return `Format your response to match: ${this.outputKeys.join(', ')}`;
}
return 'Follow the specified format exactly.';
}
}
```
**Usage:**
```javascript
const parser = new RegexOutputParser({
regex: /Sentiment: (\w+), Confidence: ([\d.]+)/,
outputKeys: ["sentiment", "confidence"]
});
const result = await parser.parse("Sentiment: positive, Confidence: 0.92");
// { sentiment: "positive", confidence: "0.92" }
```
---
# Output Parsers: Advanced Patterns & Integration
## Advanced Parser: Structured Output Parser
### Step 6: Structured Output Parser
**Location:** `src/output-parsers/structured-parser.js`
The most powerful parser - validates against a full schema with types and descriptions.
**What it does:**
- Defines expected schema with types
- Generates format instructions for LLM
- Validates all fields and types
- Provides detailed error messages
**Use when:**
- Need complex structured data
- Want strong type validation
- Need to generate format instructions automatically
**Implementation:**
```javascript
import { BaseOutputParser, OutputParserException } from './base-parser.js';
/**
* Parser with full schema validation
*
* Example:
* const parser = new StructuredOutputParser({
* responseSchemas: [
* {
* name: "sentiment",
* type: "string",
* description: "The sentiment (positive/negative/neutral)",
* enum: ["positive", "negative", "neutral"]
* },
* {
* name: "confidence",
* type: "number",
* description: "Confidence score between 0 and 1"
* }
* ]
* });
*/
export class StructuredOutputParser extends BaseOutputParser {
constructor(options = {}) {
super();
this.responseSchemas = options.responseSchemas || [];
}
/**
* Parse and validate against schema
*/
async parse(text) {
try {
// Extract JSON
const jsonText = this._extractJson(text);
const parsed = JSON.parse(jsonText);
// Validate against schema
this._validateAgainstSchema(parsed);
return parsed;
} catch (error) {
throw new OutputParserException(
`Failed to parse structured output: ${error.message}`,
text,
error
);
}
}
/**
* Extract JSON from text (same as JsonOutputParser)
*/
_extractJson(text) {
try {
JSON.parse(text.trim());
return text.trim();
} catch {}
const markdownMatch = text.match(/```(?:json)?\s*\n?([\s\S]*?)\n?```/);
if (markdownMatch) return markdownMatch[1].trim();
const jsonMatch = text.match(/\{[\s\S]*\}/);
if (jsonMatch) return jsonMatch[0];
return text.trim();
}
/**
* Validate parsed data against schema
*/
_validateAgainstSchema(parsed) {
for (const schema of this.responseSchemas) {
const { name, type, enum: enumValues, required = true } = schema;
// Check required fields
if (required && !(name in parsed)) {
throw new Error(`Missing required field: ${name}`);
}
if (name in parsed) {
const value = parsed[name];
// Check type
if (!this._checkType(value, type)) {
throw new Error(
`Field ${name} should be ${type}, got ${typeof value}`
);
}
// Check enum values
if (enumValues && !enumValues.includes(value)) {
throw new Error(
`Field ${name} must be one of: ${enumValues.join(', ')}`
);
}
}
}
}
/**
* Check if value matches expected type
*/
_checkType(value, type) {
switch (type) {
case 'string':
return typeof value === 'string';
case 'number':
return typeof value === 'number' && !isNaN(value);
case 'boolean':
return typeof value === 'boolean';
case 'array':
return Array.isArray(value);
case 'object':
return typeof value === 'object' && value !== null && !Array.isArray(value);
default:
return true;
}
}
/**
* Generate format instructions for LLM
*/
getFormatInstructions() {
const schemaDescriptions = this.responseSchemas.map(schema => {
let desc = `"${schema.name}": ${schema.type}`;
if (schema.description) {
desc += ` // ${schema.description}`;
}
if (schema.enum) {
desc += ` (one of: ${schema.enum.join(', ')})`;
}
return desc;
});
return `Respond with valid JSON matching this schema:
{
${schemaDescriptions.map(d => ' ' + d).join(',\n')}
}`;
}
/**
* Static helper to create from simple schema
*/
static fromNamesAndDescriptions(schemas) {
const responseSchemas = Object.entries(schemas).map(([name, description]) => ({
name,
description,
type: 'string' // Default type
}));
return new StructuredOutputParser({ responseSchemas });
}
}
```
**Usage:**
```javascript
const parser = new StructuredOutputParser({
responseSchemas: [
{
name: "sentiment",
type: "string",
description: "The sentiment of the text",
enum: ["positive", "negative", "neutral"],
required: true
},
{
name: "confidence",
type: "number",
description: "Confidence score from 0 to 1",
required: true
},
{
name: "keywords",
type: "array",
description: "Key themes in the text",
required: false
}
]
});
// Get format instructions to add to prompt
const instructions = parser.getFormatInstructions();
console.log(instructions);
// Parse and validate
const result = await parser.parse(`{
"sentiment": "positive",
"confidence": 0.92,
"keywords": ["great", "love", "excellent"]
}`);
```
---
## Real-World Examples
### Example 1: Email Classification with Structured Parser
```javascript
import { StructuredOutputParser } from './output-parsers/structured-parser.js';
import { PromptTemplate } from './prompts/prompt-template.js';
import { LlamaCppLLM } from './llm/llama-cpp-llm.js';
// Define the output structure
const parser = new StructuredOutputParser({
responseSchemas: [
{
name: "category",
type: "string",
description: "Email category",
enum: ["spam", "invoice", "meeting", "urgent", "personal", "other"]
},
{
name: "confidence",
type: "number",
description: "Confidence score (0-1)"
},
{
name: "reason",
type: "string",
description: "Brief explanation for classification"
},
{
name: "actionRequired",
type: "boolean",
description: "Does email require action?"
}
]
});
// Build prompt with format instructions
const prompt = new PromptTemplate({
template: `Classify this email.
Email:
From: {from}
Subject: {subject}
Body: {body}
{format_instructions}`,
inputVariables: ["from", "subject", "body"],
partialVariables: {
format_instructions: parser.getFormatInstructions()
}
});
// Create chain
const llm = new LlamaCppLLM({ modelPath: './model.gguf' });
const chain = prompt.pipe(llm).pipe(parser);
// Use it
const result = await chain.invoke({
from: "billing@company.com",
subject: "Invoice #12345",
body: "Payment due by March 15th"
});
console.log(result);
// {
// category: "invoice",
// confidence: 0.98,
// reason: "Email contains invoice number and payment deadline",
// actionRequired: true
// }
```
---
### Example 2: Content Extraction with JSON Parser
```javascript
import { JsonOutputParser } from './output-parsers/json-parser.js';
import { ChatPromptTemplate } from './prompts/chat-prompt-template.js';
const parser = new JsonOutputParser({
schema: {
title: 'string',
summary: 'string',
tags: 'object', // Will be array
author: 'string'
}
});
const prompt = ChatPromptTemplate.fromMessages([
["system", "Extract article metadata. Respond with JSON."],
["human", "Article: {article}"]
]);
const chain = prompt.pipe(llm).pipe(parser);
const result = await chain.invoke({
article: "Title: AI Revolution\nBy: John Doe\n\nAI is transforming..."
});
// {
// title: "AI Revolution",
// summary: "Article discusses AI's transformative impact",
// tags: ["AI", "technology", "future"],
// author: "John Doe"
// }
```
---
### Example 3: List Extraction for Recommendations
```javascript
import { ListOutputParser } from './output-parsers/list-parser.js';
import { PromptTemplate } from './prompts/prompt-template.js';
const parser = new ListOutputParser();
const prompt = new PromptTemplate({
template: `Recommend 5 {category} for someone interested in {interest}.
{format_instructions}
List:`,
inputVariables: ["category", "interest"],
partialVariables: {
format_instructions: parser.getFormatInstructions()
}
});
const chain = prompt.pipe(llm).pipe(parser);
const books = await chain.invoke({
category: "books",
interest: "machine learning"
});
console.log(books);
// [
// "Pattern Recognition and Machine Learning",
// "Deep Learning by Goodfellow",
// "Hands-On Machine Learning",
// "The Hundred-Page Machine Learning Book",
// "Machine Learning Yearning"
// ]
```
---
### Example 4: Sentiment Analysis with Retry
```javascript
import { JsonOutputParser } from './output-parsers/json-parser.js';
import { PromptTemplate } from './prompts/prompt-template.js';
const parser = new JsonOutputParser();
// If parsing fails, retry with clearer instructions
async function robustSentimentAnalysis(text) {
const prompt = new PromptTemplate({
template: `Analyze sentiment of: "{text}"
Respond with ONLY valid JSON:
{{"sentiment": "positive/negative/neutral", "score": 0.0-1.0}}`
});
const chain = prompt.pipe(llm).pipe(parser);
try {
return await chain.invoke({ text });
} catch (error) {
console.log('Parse failed, retrying with stricter prompt...');
// Retry with more explicit prompt
const strictPrompt = new PromptTemplate({
template: `Analyze: "{text}"
IMPORTANT: Respond with ONLY this JSON structure, nothing else:
{{"sentiment": "positive", "score": 0.9}}
Your response:`
});
const retryChain = strictPrompt.pipe(llm).pipe(parser);
return await retryChain.invoke({ text });
}
}
```
---
## Advanced Patterns
### Pattern 1: Fallback Parsing
```javascript
class FallbackOutputParser extends BaseOutputParser {
constructor(parsers) {
super();
this.parsers = parsers;
}
async parse(text) {
const errors = [];
for (const parser of this.parsers) {
try {
return await parser.parse(text);
} catch (error) {
errors.push({ parser: parser.name, error });
}
}
throw new OutputParserException(
`All parsers failed. Errors: ${JSON.stringify(errors)}`,
text
);
}
}
// Usage
const parser = new FallbackOutputParser([
new JsonOutputParser(), // Try JSON first
new RegexOutputParser({...}), // Try regex second
new StringOutputParser() // Fallback to string
]);
```
---
### Pattern 2: Transform After Parse
```javascript
class TransformOutputParser extends BaseOutputParser {
constructor(parser, transform) {
super();
this.parser = parser;
this.transform = transform;
}
async parse(text) {
const parsed = await this.parser.parse(text);
return this.transform(parsed);
}
}
// Usage: parse JSON then transform values
const parser = new TransformOutputParser(
new JsonOutputParser(),
(data) => ({
...data,
confidence: parseFloat(data.confidence),
timestamp: new Date().toISOString()
})
);
```
---
### Pattern 3: Conditional Parsing
```javascript
class ConditionalOutputParser extends BaseOutputParser {
constructor(condition, trueParser, falseParser) {
super();
this.condition = condition;
this.trueParser = trueParser;
this.falseParser = falseParser;
}
async parse(text) {
const useTrue = this.condition(text);
const parser = useTrue ? this.trueParser : this.falseParser;
return await parser.parse(text);
}
}
// Usage: different parsers based on content
const parser = new ConditionalOutputParser(
(text) => text.includes('{'), // Has JSON?
new JsonOutputParser(),
new ListOutputParser()
);
```
---
### Pattern 4: Validated Output
```javascript
class ValidatedOutputParser extends BaseOutputParser {
constructor(parser, validator) {
super();
this.parser = parser;
this.validator = validator;
}
async parse(text) {
const parsed = await this.parser.parse(text);
const isValid = this.validator(parsed);
if (!isValid) {
throw new OutputParserException(
'Parsed output failed validation',
text
);
}
return parsed;
}
}
// Usage: ensure confidence is in range
const parser = new ValidatedOutputParser(
new JsonOutputParser(),
(data) => data.confidence >= 0 && data.confidence <= 1
);
```
---
## Integration with Full Chain
### Complete Example: Sentiment Analysis API
```javascript
import { PromptTemplate } from './prompts/prompt-template.js';
import { LlamaCppLLM } from './llm/llama-cpp-llm.js';
import { StructuredOutputParser } from './output-parsers/structured-parser.js';
import { ConsoleCallback } from './utils/callbacks.js';
// Define output structure
const parser = new StructuredOutputParser({
responseSchemas: [
{
name: "sentiment",
type: "string",
enum: ["positive", "negative", "neutral"]
},
{
name: "confidence",
type: "number"
},
{
name: "emotions",
type: "array",
description: "List of detected emotions"
}
]
});
// Build prompt
const prompt = new PromptTemplate({
template: `Analyze the sentiment of this text:
"{text}"
{format_instructions}`,
inputVariables: ["text"],
partialVariables: {
format_instructions: parser.getFormatInstructions()
}
});
// Create LLM
const llm = new LlamaCppLLM({
modelPath: './model.gguf',
temperature: 0.1 // Low temp for consistent classification
});
// Build chain with logging
const chain = prompt.pipe(llm).pipe(parser);
const logger = new ConsoleCallback();
// Analyze sentiment
async function analyzeSentiment(text) {
try {
const result = await chain.invoke(
{ text },
{ callbacks: [logger] }
);
return {
success: true,
data: result
};
} catch (error) {
return {
success: false,
error: error.message,
rawOutput: error.llmOutput
};
}
}
// Use it
const result = await analyzeSentiment("I absolutely love this product! It's amazing!");
console.log(result);
// {
// success: true,
// data: {
// sentiment: "positive",
// confidence: 0.95,
// emotions: ["joy", "excitement", "satisfaction"]
// }
// }
```
---
## Error Handling
### Pattern: Graceful Degradation
```javascript
async function parseWithFallback(text, primaryParser, fallbackValue) {
try {
return await primaryParser.parse(text);
} catch (error) {
console.warn('Primary parser failed:', error.message);
console.warn('Using fallback value:', fallbackValue);
return fallbackValue;
}
}
// Usage
const result = await parseWithFallback(
llmOutput,
new JsonOutputParser(),
{ error: true, message: "Failed to parse", raw: llmOutput }
);
```
---
### Pattern: Retry with Fix Instructions
```javascript
async function parseWithRetry(text, parser, llm, maxRetries = 2) {
for (let attempt = 0; attempt < maxRetries; attempt++) {
try {
return await parser.parse(text);
} catch (error) {
if (attempt === maxRetries - 1) throw error;
// Ask LLM to fix the output
const fixPrompt = `The following output is malformed:
${text}
Error: ${error.message}
Please provide the output in correct format:
${parser.getFormatInstructions()}`;
text = await llm.invoke(fixPrompt);
}
}
}
```
---
## Testing Parsers
### Unit Tests
```javascript
import { describe, it, expect } from 'your-test-framework';
import { JsonOutputParser } from './output-parsers/json-parser.js';
describe('JsonOutputParser', () => {
it('should parse plain JSON', async () => {
const parser = new JsonOutputParser();
const result = await parser.parse('{"name": "Alice", "age": 30}');
expect(result.name).toBe('Alice');
expect(result.age).toBe(30);
});
it('should extract JSON from markdown', async () => {
const parser = new JsonOutputParser();
const text = '```json\n{"key": "value"}\n```';
const result = await parser.parse(text);
expect(result.key).toBe('value');
});
it('should validate against schema', async () => {
const parser = new JsonOutputParser({
schema: { name: 'string', age: 'number' }
});
await expect(
parser.parse('{"name": "Bob", "age": "invalid"}')
).rejects.toThrow();
});
it('should throw on invalid JSON', async () => {
const parser = new JsonOutputParser();
await expect(parser.parse('not json')).rejects.toThrow();
});
});
```
---
## Best Practices
### ✅ DO:
**1. Include format instructions in prompts**
```javascript
const prompt = new PromptTemplate({
template: `{task}
{format_instructions}`,
partialVariables: {
format_instructions: parser.getFormatInstructions()
}
});
```
**2. Use schema validation for complex outputs**
```javascript
const parser = new StructuredOutputParser({
responseSchemas: [
{ name: "field1", type: "string", required: true },
{ name: "field2", type: "number", required: true }
]
});
```
**3. Handle parsing errors gracefully**
```javascript
try {
const parsed = await parser.parse(text);
} catch (error) {
console.error('Parsing failed:', error.message);
// Fallback or retry logic
}
```
**4. Test parsers independently**
```javascript
// Test without LLM
const result = await parser.parse(mockLLMOutput);
expect(result).toMatchSchema();
```
**5. Use low temperature for structured outputs**
```javascript
const llm = new LlamaCppLLM({
temperature: 0.1 // More consistent formatting
});
```
---
### ❌ DON'T:
**1. Don't assume perfect LLM formatting**
```javascript
// Bad
const data = JSON.parse(llmOutput); // Will fail often
// Good
const data = await jsonParser.parse(llmOutput); // Handles variations
```
**2. Don't skip validation**
```javascript
// Bad
const result = await parser.parse(text);
// Use result.field without checking
// Good
const result = await parser.parse(text);
if (result.field && typeof result.field === 'string') {
// Use result.field
}
```
**3. Don't use parsers for simple text**
```javascript
// Bad
const parser = new JsonOutputParser();
const result = await parser.parse(simpleText);
// Good
const parser = new StringOutputParser();
const result = await parser.parse(simpleText);
```
---
## Exercises
Practice using output parsers in real-world scenarios from simple to complex:
### Exercise 21: Product Review Analyzer
Extract clean summaries and sentiment from product reviews using StringOutputParser.
**Starter code**: [`exercises/21-review-analyzer.js`](exercises/21-review-analyzer.js)
### Exercise 22: Contact Information Extractor
Parse structured contact details and skills from unstructured text using JSON and List parsers.
**Starter code**: [`exercises/22-contact-extractor.js`](exercises/22-contact-extractor.js)
### Exercise 23: Article Metadata Extractor
Extract complex metadata with schema validation using StructuredOutputParser.
**Starter code**: [`exercises/23-article-metadata.js`](exercises/23-article-metadata.js)
### Exercise 24: Multi-Parser Content Pipeline
Build production-ready pipelines with multiple parsers, fallback strategies, and content routing.
**Starter code**: [`exercises/24-multi-parser-pipeline.js`](exercises/24-multi-parser-pipeline.js)
---
## Summary
You've built a complete output parsing system!
### Key Takeaways
1. **BaseOutputParser**: Foundation for all parsers
2. **StringOutputParser**: Clean text output
3. **JsonOutputParser**: Extract and validate JSON
4. **ListOutputParser**: Parse lists/arrays
5. **RegexOutputParser**: Pattern-based extraction
6. **StructuredOutputParser**: Full schema validation
### What You Built
A parsing system that:
- ✅ Extracts structured data reliably
- ✅ Validates output formats
- ✅ Handles errors gracefully
- ✅ Generates format instructions
- ✅ Works in chains with prompts
- ✅ Is testable in isolation
### Next Steps
Now you can combine prompts + LLMs + parsers into complete chains.
➡️ **Next: [LLM Chains](./03-llm-chain.md)**
Learn how to build complete prompt → LLM → parser pipelines.
---
**Built with ❤️ for learners who want to understand AI frameworks deeply**
[← Previous: Prompts](./01-prompts.md) | [Tutorial Index](../README.md) | [Next: LLM Chains →](./03-llm-chain.md) |