langchain-react-chinese-parser
Version:
一个专门为中文大模型提供的 LangChain ReAct 输出解析器集合，支持通义千问、ChatGLM、百川、智谱GLM、文心一言、Minimax 等主流中文模型
745 lines (721 loc) • 26.3 kB
JavaScript
// langchain-react-chinese-parser - Chinese LLM ReAct Output Parsers for LangChain
"use strict";
var __defProp = Object.defineProperty;
var __getOwnPropDesc = Object.getOwnPropertyDescriptor;
var __getOwnPropNames = Object.getOwnPropertyNames;
var __hasOwnProp = Object.prototype.hasOwnProperty;
var __export = (target, all) => {
for (var name in all)
__defProp(target, name, { get: all[name], enumerable: true });
};
var __copyProps = (to, from, except, desc) => {
if (from && typeof from === "object" || typeof from === "function") {
for (let key of __getOwnPropNames(from))
if (!__hasOwnProp.call(to, key) && key !== except)
__defProp(to, key, { get: () => from[key], enumerable: !(desc = __getOwnPropDesc(from, key)) || desc.enumerable });
}
return to;
};
var __toCommonJS = (mod) => __copyProps(__defProp({}, "__esModule", { value: true }), mod);
// src/index.ts
var index_exports = {};
__export(index_exports, {
BaichuanReActOutputParser: () => BaichuanReActOutputParser,
BaseChineseReActParser: () => BaseChineseReActParser,
ChatGLMReActOutputParser: () => ChatGLMReActOutputParser,
ChineseReActParserFactory: () => ChineseReActParserFactory,
ERNIEReActOutputParser: () => ERNIEReActOutputParser,
QwenReActOutputParser: () => QwenReActOutputParser,
SUPPORTED_MODELS: () => SUPPORTED_MODELS,
VERSION: () => VERSION,
createAutoParser: () => createAutoParser,
createBaichuanParser: () => createBaichuanParser,
createChatGLMParser: () => createChatGLMParser,
createChineseReActParser: () => createChineseReActParser,
createERNIEParser: () => createERNIEParser,
createQwenParser: () => createQwenParser,
default: () => createChineseReActParser
});
module.exports = __toCommonJS(index_exports);
// src/base/BaseChineseReActParser.ts
var import_output_parsers = require("@langchain/core/output_parsers");
var BaseChineseReActParser = class extends import_output_parsers.BaseOutputParser {
constructor(options = {}) {
super();
this.options = {
debug: false,
relaxedMode: true,
maxRetries: 3,
...options
};
}
/**
* 解析模型输出
*/
async parse(text) {
const cleanText = this.preprocessText(text);
if (this.options.debug) {
console.log(`[${this.modelType}] \u89E3\u6790\u8F93\u5165:`, cleanText);
}
const finalAnswerMatch = this.extractFinalAnswer(cleanText);
if (finalAnswerMatch) {
if (this.options.debug) {
console.log(`[${this.modelType}] \u627E\u5230\u6700\u7EC8\u7B54\u6848:`, finalAnswerMatch);
}
return {
returnValues: { output: finalAnswerMatch.trim() },
log: cleanText
};
}
const actionMatch = this.extractAction(cleanText);
if (actionMatch) {
if (this.options.debug) {
console.log(`[${this.modelType}] \u627E\u5230\u52A8\u4F5C:`, actionMatch);
}
return {
tool: actionMatch.action.trim(),
toolInput: actionMatch.actionInput.trim(),
log: cleanText
};
}
const errorMessage = this.buildErrorMessage(cleanText);
throw new import_output_parsers.OutputParserException(errorMessage, cleanText);
}
/**
* 获取格式化指令
*/
getFormatInstructions() {
return this.buildFormatInstructions();
}
/**
* 获取解析器类型
*/
getType() {
return `${this.modelType}_react_output_parser`;
}
/**
* 预处理文本
*/
preprocessText(text) {
return text.trim().replace(/\r\n/g, "\n").replace(/[ \t]+/g, " ").replace(/\n\s+/g, "\n").replace(/\s*[:：]\s*/g, ": ");
}
/**
* 提取最终答案
*/
extractFinalAnswer(text) {
const allKeywords = [
...this.keywords.chinese.finalAnswer,
...this.keywords.english.finalAnswer,
...this.options.customKeywords?.chinese?.finalAnswer || [],
...this.options.customKeywords?.english?.finalAnswer || []
];
for (const keyword of allKeywords) {
const strictRegex = new RegExp(`^\\s*${this.escapeRegex(keyword)}\\s*[:\uFF1A]\\s*(.*)$`, "im");
const strictMatch = text.match(strictRegex);
if (strictMatch && strictMatch[1]) {
return strictMatch[1];
}
const multiLineRegex = new RegExp(
`(?:^|\\n)\\s*${this.escapeRegex(keyword)}\\s*[:\uFF1A]\\s*(.*)(?:\\n|$)`,
"im"
);
const multiLineMatch = text.match(multiLineRegex);
if (multiLineMatch && multiLineMatch[1]) {
return multiLineMatch[1];
}
if (this.options.relaxedMode) {
const relaxedPattern = `${this.escapeRegex(keyword)}[\\s]*[:\uFF1A][\\s]*([\\s\\S]*?)(?=\\n\\s*(?:\u601D\u8003|\u52A8\u4F5C|\u6700\u7EC8\u7B54\u6848|Thought|Action|Final Answer)|$)`;
const relaxedRegex = new RegExp(relaxedPattern, "i");
const relaxedMatch = text.match(relaxedRegex);
if (relaxedMatch && relaxedMatch[1]) {
return relaxedMatch[1].trim();
}
}
}
return null;
}
/**
* 提取动作和动作输入
*/
extractAction(text) {
const chineseAction = this.extractActionByLanguage(text, "chinese");
if (chineseAction) {
return chineseAction;
}
const englishAction = this.extractActionByLanguage(text, "english");
if (englishAction) {
return englishAction;
}
return null;
}
/**
* 按语言提取动作
*/
extractActionByLanguage(text, language) {
const keywords = this.keywords[language];
let action = "";
let actionInput = "";
let thought = "";
for (const thoughtKeyword of keywords.thought) {
const thoughtRegex = new RegExp(
`(?:^|\\n)\\s*${this.escapeRegex(thoughtKeyword)}\\s*[:\uFF1A]\\s*([^\\n]+)`,
"im"
);
const thoughtMatch = text.match(thoughtRegex);
if (thoughtMatch) {
thought = thoughtMatch[1].trim();
break;
}
}
for (const actionKeyword of keywords.action) {
const actionRegex = new RegExp(
`(?:^|\\n)\\s*${this.escapeRegex(actionKeyword)}\\s*[:\uFF1A]\\s*([^\\n]+)`,
"im"
);
const actionMatch = text.match(actionRegex);
if (actionMatch) {
action = actionMatch[1].trim();
break;
}
}
for (const inputKeyword of keywords.actionInput) {
const inputRegex = new RegExp(
`(?:^|\\n)\\s*${this.escapeRegex(inputKeyword)}\\s*[:\uFF1A]\\s*([^\\n]+)`,
"im"
);
const inputMatch = text.match(inputRegex);
if (inputMatch) {
actionInput = inputMatch[1].trim();
break;
}
}
if (action && !actionInput && this.options.relaxedMode) {
actionInput = this.extractActionInputFallback(text, action, language);
}
return action && actionInput ? { action, actionInput, thought } : null;
}
/**
* 兜底策略：提取动作输入
*/
extractActionInputFallback(text, action, language) {
const actionKeywords = this.keywords[language].action;
for (const actionKeyword of actionKeywords) {
const regexPattern = `(?:^|\\n)\\s*${this.escapeRegex(actionKeyword)}\\s*[:\uFF1A]\\s*${this.escapeRegex(action)}\\s*\\n([^\\n]+)`;
const regex = new RegExp(regexPattern, "im");
const match = text.match(regex);
if (match) {
return match[1].trim();
}
}
return "";
}
/**
* 构建格式化指令
*/
buildFormatInstructions() {
const chineseKeywords = this.keywords.chinese;
const englishKeywords = this.keywords.english;
return `\u8BF7\u6309\u7167\u4EE5\u4E0B\u683C\u5F0F\u56DE\u7B54\uFF1A
**\u4F7F\u7528\u5DE5\u5177\u65F6\uFF1A**
${chineseKeywords.thought[0]}: \u63CF\u8FF0\u4F60\u7684\u601D\u8003\u8FC7\u7A0B
${chineseKeywords.action[0]}: \u5DE5\u5177\u540D\u79F0
${chineseKeywords.actionInput[0]}: \u5DE5\u5177\u7684\u8F93\u5165\u53C2\u6570
**\u6709\u6700\u7EC8\u7B54\u6848\u65F6\uFF1A**
${chineseKeywords.thought[0]}: \u63CF\u8FF0\u4F60\u7684\u601D\u8003\u8FC7\u7A0B
${chineseKeywords.finalAnswer[0]}: \u4F60\u7684\u6700\u7EC8\u7B54\u6848
**\u6CE8\u610F\u4E8B\u9879\uFF1A**
- \u6BCF\u884C\u90FD\u8981\u4EE5\u5BF9\u5E94\u7684\u5173\u952E\u5B57\u5F00\u5934\uFF0C\u540E\u9762\u8DDF\u5192\u53F7
- \u652F\u6301\u4E2D\u82F1\u6587\u5173\u952E\u5B57\u6DF7\u7528
- \u82F1\u6587\u683C\u5F0F\uFF1A${englishKeywords.thought[0]}:, ${englishKeywords.action[0]}:, ${englishKeywords.actionInput[0]}:, ${englishKeywords.finalAnswer[0]}:
- \u4E2D\u6587\u683C\u5F0F\uFF1A${chineseKeywords.thought[0]}:, ${chineseKeywords.action[0]}:, ${chineseKeywords.actionInput[0]}:, ${chineseKeywords.finalAnswer[0]}:`;
}
/**
* 构建错误消息
*/
buildErrorMessage(text) {
const chineseKeywords = this.keywords.chinese;
return `\u65E0\u6CD5\u89E3\u6790 ${this.modelType} \u6A21\u578B\u8F93\u51FA\u3002\u8BF7\u786E\u4FDD\u8F93\u51FA\u5305\u542B\u6B63\u786E\u7684\u5173\u952E\u5B57\u683C\u5F0F\u3002
\u671F\u671B\u683C\u5F0F\uFF1A
${chineseKeywords.thought[0]}: [\u601D\u8003\u8FC7\u7A0B]
${chineseKeywords.action[0]}: [\u5DE5\u5177\u540D\u79F0]
${chineseKeywords.actionInput[0]}: [\u5DE5\u5177\u53C2\u6570]
\u6216\u8005\uFF1A
${chineseKeywords.thought[0]}: [\u601D\u8003\u8FC7\u7A0B]
${chineseKeywords.finalAnswer[0]}: [\u6700\u7EC8\u7B54\u6848]
\u5B9E\u9645\u8F93\u51FA: ${text}`;
}
/**
* 转义正则表达式特殊字符
*/
escapeRegex(str) {
return str.replace(/[.*+?^${}()|[\]\\]/g, "\\$&");
}
/**
* LangChain 兼容性
*/
_type() {
return this.getType();
}
};
// src/parsers/BaichuanReActOutputParser.ts
var BaichuanReActOutputParser = class extends BaseChineseReActParser {
constructor(options = {}) {
super(options);
this.lc_namespace = ["langchain-react-chinese-parser", "parsers", "baichuan"];
this.modelType = "baichuan";
this.keywords = {
chinese: {
thought: ["\u601D\u8003", "\u5206\u6790", "\u63A8\u7406"],
action: ["\u5DE5\u5177", "\u52A8\u4F5C", "\u64CD\u4F5C", "\u4F7F\u7528"],
actionInput: ["\u5DE5\u5177\u8F93\u5165", "\u53C2\u6570", "\u8F93\u5165", "\u5185\u5BB9"],
finalAnswer: ["\u6700\u7EC8\u7B54\u6848", "\u7B54\u6848", "\u7ED3\u679C"],
observation: ["\u89C2\u5BDF", "\u5DE5\u5177\u8FD4\u56DE", "\u8FD4\u56DE\u7ED3\u679C"]
},
english: {
thought: ["thought", "think", "reasoning"],
action: ["tool", "action", "use"],
actionInput: ["tool input", "parameter", "input"],
finalAnswer: ["final answer", "answer", "result"],
observation: ["observation", "tool result"]
}
};
}
/**
* 百川特殊格式处理
* 百川模型倾向于使用"工具"而非"动作"
*/
extractCustomAction(text) {
const useToolPattern = /使用([^工具\n]+)工具[，]?\s*输入\s*([^。\n]*)/;
const useToolMatch = text.match(useToolPattern);
if (useToolMatch) {
return {
action: useToolMatch[1].trim(),
actionInput: useToolMatch[2].trim(),
thought: "\u4F7F\u7528\u5DE5\u5177"
};
}
const callToolPattern = /调用工具([^进行\n]+)进行([^。\n]*)/;
const callToolMatch = text.match(callToolPattern);
if (callToolMatch) {
return {
action: callToolMatch[1].trim(),
actionInput: callToolMatch[2].trim(),
thought: "\u8C03\u7528\u5DE5\u5177"
};
}
const directToolPattern = /(?:^|\n)\s*([a-zA-Z_]+)\s*[:：]\s*([^\n]+)/im;
const directToolMatch = text.match(directToolPattern);
if (directToolMatch && this.isToolName(directToolMatch[1])) {
return {
action: directToolMatch[1].trim(),
actionInput: directToolMatch[2].trim(),
thought: "\u76F4\u63A5\u8C03\u7528\u5DE5\u5177"
};
}
return null;
}
/**
* 判断是否为工具名称
*/
isToolName(name) {
const commonTools = [
"search",
"calculator",
"weather",
"translate",
"file",
"web",
"database",
"api",
"code"
];
return commonTools.some((tool) => name.toLowerCase().includes(tool)) || /^[a-zA-Z_][a-zA-Z0-9_]*$/.test(name);
}
/**
* 重写格式化指令
*/
buildFormatInstructions() {
const base = super.buildFormatInstructions();
return `${base}
**\u767E\u5DDD\u7279\u6B8A\u683C\u5F0F\u652F\u6301\uFF1A**
- \u4F7F\u7528[\u5DE5\u5177\u540D]\u5DE5\u5177\uFF0C\u8F93\u5165[\u53C2\u6570]
- \u8C03\u7528\u5DE5\u5177[\u5DE5\u5177\u540D]\u8FDB\u884C[\u64CD\u4F5C]
- \u63A8\u8350\u4F7F\u7528"\u5DE5\u5177"\u800C\u975E"\u52A8\u4F5C"\u5173\u952E\u5B57
- \u652F\u6301\u76F4\u63A5\u5DE5\u5177\u540D\u8C03\u7528\uFF1Atool_name: parameter`;
}
};
// src/parsers/ChatGLMReActOutputParser.ts
var ChatGLMReActOutputParser = class extends BaseChineseReActParser {
constructor(options = {}) {
super(options);
this.lc_namespace = ["langchain-react-chinese-parser", "parsers", "chatglm"];
this.modelType = "chatglm";
this.keywords = {
chinese: {
thought: ["\u601D\u8003", "\u5206\u6790", "\u7406\u89E3", "\u8003\u8651"],
action: ["\u52A8\u4F5C", "\u884C\u52A8", "\u5DE5\u5177", "\u64CD\u4F5C"],
actionInput: ["\u52A8\u4F5C\u8F93\u5165", "\u5DE5\u5177\u8F93\u5165", "\u8F93\u5165", "\u53C2\u6570"],
finalAnswer: ["\u6700\u7EC8\u7B54\u6848", "\u7B54\u6848", "\u7ED3\u8BBA", "\u56DE\u7B54"],
observation: ["\u89C2\u5BDF", "\u89C2\u5BDF\u7ED3\u679C", "\u7ED3\u679C"]
},
english: {
thought: ["thought", "thinking", "analysis"],
action: ["action", "tool", "operation"],
actionInput: ["action input", "tool input", "input"],
finalAnswer: ["final answer", "answer", "conclusion"],
observation: ["observation", "result"]
}
};
}
/**
* ChatGLM 特殊格式处理
* ChatGLM 经常合并思考和动作，或使用冒号变体
*/
extractCustomAction(text) {
const toolQueryPattern = /使用工具[：:]([^，\n]+)[\s]*查询[：:]([^。\n]*)/;
const toolQueryMatch = text.match(toolQueryPattern);
if (toolQueryMatch) {
return {
action: toolQueryMatch[1].trim(),
actionInput: toolQueryMatch[2].trim(),
thought: "\u9700\u8981\u4F7F\u7528\u5DE5\u5177\u67E5\u8BE2"
};
}
const callToolPattern = /调用([^，\n]+)[，]?\s*参数为\s*([^。\n]*)/;
const callToolMatch = text.match(callToolPattern);
if (callToolMatch) {
return {
action: callToolMatch[1].trim(),
actionInput: callToolMatch[2].trim(),
thought: "\u8C03\u7528\u5DE5\u5177"
};
}
const colonOnlyPattern = /(?:^|\n)\s*(动作|工具|操作)\s*:\s*([^\n]+)\s*\n\s*(?:输入|参数)\s*:\s*([^\n]+)/im;
const colonOnlyMatch = text.match(colonOnlyPattern);
if (colonOnlyMatch) {
return {
action: colonOnlyMatch[2].trim(),
actionInput: colonOnlyMatch[3].trim(),
thought: "\u6267\u884C\u64CD\u4F5C"
};
}
return null;
}
/**
* ChatGLM 倾向于使用中文冒号，重写预处理逻辑
*/
preprocessText(text) {
return text.trim().replace(/\r\n/g, "\n").replace(/：/g, ":").replace(/\s+/g, " ").replace(/\n\s+/g, "\n");
}
/**
* 重写格式化指令
*/
buildFormatInstructions() {
const base = super.buildFormatInstructions();
return `${base}
**ChatGLM \u7279\u6B8A\u683C\u5F0F\u652F\u6301\uFF1A**
- \u4F7F\u7528\u5DE5\u5177\uFF1A[\u5DE5\u5177\u540D] \u67E5\u8BE2\uFF1A[\u53C2\u6570]
- \u8C03\u7528[\u5DE5\u5177\u540D]\uFF0C\u53C2\u6570\u4E3A[\u53C2\u6570]
- \u652F\u6301\u4E2D\u6587\u5192\u53F7\uFF1A\u548C\u82F1\u6587\u5192\u53F7:\u6DF7\u7528
- \u63A8\u8350\u4F7F\u7528\u4E2D\u6587\u5192\u53F7\uFF1A\u683C\u5F0F`;
}
};
// src/parsers/ERNIEReActOutputParser.ts
var ERNIEReActOutputParser = class extends BaseChineseReActParser {
constructor(options = {}) {
super(options);
this.lc_namespace = ["langchain-react-chinese-parser", "parsers", "ernie"];
this.modelType = "ernie";
this.keywords = {
chinese: {
thought: ["\u601D\u8003", "\u5206\u6790", "\u5224\u65AD", "\u8003\u8651"],
action: ["\u8C03\u7528\u5DE5\u5177", "\u4F7F\u7528\u5DE5\u5177", "\u6267\u884C", "\u64CD\u4F5C"],
actionInput: ["\u8F93\u5165", "\u53C2\u6570", "\u5185\u5BB9", "\u67E5\u8BE2"],
finalAnswer: ["\u6700\u7EC8\u7B54\u6848", "\u7B54\u6848", "\u7ED3\u8BBA", "\u56DE\u590D"],
observation: ["\u89C2\u5BDF", "\u5DE5\u5177\u7ED3\u679C", "\u8FD4\u56DE", "\u8F93\u51FA"]
},
english: {
thought: ["thought", "analysis", "thinking"],
action: ["call tool", "use tool", "action"],
actionInput: ["input", "parameter", "query"],
finalAnswer: ["final answer", "answer", "conclusion"],
observation: ["observation", "tool result", "output"]
}
};
}
/**
* ERNIE 特殊格式处理
* ERNIE 倾向于使用"调用工具"和较长的关键字
*/
extractCustomAction(text) {
const needCallPattern = /我需要调用([^来\n]+)来([^。\n]*)/;
const needCallMatch = text.match(needCallPattern);
if (needCallMatch) {
return {
action: needCallMatch[1].trim(),
actionInput: needCallMatch[2].trim(),
thought: "\u9700\u8981\u8C03\u7528\u5DE5\u5177"
};
}
const nowCallPattern = /现在调用([^，\n]+)[，]?\s*参数是\s*([^。\n]*)/;
const nowCallMatch = text.match(nowCallPattern);
if (nowCallMatch) {
return {
action: nowCallMatch[1].trim(),
actionInput: nowCallMatch[2].trim(),
thought: "\u8C03\u7528\u5DE5\u5177"
};
}
const mixedPattern = /(?:调用|call)\s*([a-zA-Z_\u4e00-\u9fff]+)\s*[:：]\s*([^\n]+)/i;
const mixedMatch = text.match(mixedPattern);
if (mixedMatch) {
return {
action: mixedMatch[1].trim(),
actionInput: mixedMatch[2].trim(),
thought: "\u6267\u884C\u64CD\u4F5C"
};
}
return null;
}
/**
* ERNIE 特殊的预处理
* 处理 ERNIE 经常出现的多余描述文字
*/
preprocessText(text) {
let cleaned = super.preprocessText(text);
cleaned = cleaned.replace(/好的[，,]?/g, "").replace(/我来帮您[^。]*。?/g, "").replace(/让我[^。]*。?/g, "").replace(/根据您的要求[，,]?/g, "");
return cleaned;
}
/**
* 重写格式化指令
*/
buildFormatInstructions() {
const base = super.buildFormatInstructions();
return `${base}
**\u6587\u5FC3\u4E00\u8A00\u7279\u6B8A\u683C\u5F0F\u652F\u6301\uFF1A**
- \u6211\u9700\u8981\u8C03\u7528[\u5DE5\u5177\u540D]\u6765[\u64CD\u4F5C\u63CF\u8FF0]
- \u73B0\u5728\u8C03\u7528[\u5DE5\u5177\u540D]\uFF0C\u53C2\u6570\u662F[\u53C2\u6570]
- \u63A8\u8350\u4F7F\u7528"\u8C03\u7528\u5DE5\u5177"\u5173\u952E\u5B57
- \u652F\u6301\u4E2D\u82F1\u6587\u6DF7\u5408\uFF1Acall [tool]: [parameter]
- \u4F1A\u81EA\u52A8\u8FC7\u6EE4\u5E38\u89C1\u7684\u5197\u4F59\u8868\u8FBE`;
}
};
// src/parsers/QwenReActOutputParser.ts
var QwenReActOutputParser = class extends BaseChineseReActParser {
constructor(options = {}) {
super(options);
this.lc_namespace = ["langchain-react-chinese-parser", "parsers", "qwen"];
this.modelType = "qwen";
this.keywords = {
chinese: {
thought: ["\u601D\u8003", "\u63A8\u7406", "\u5206\u6790", "\u60F3\u6CD5"],
action: ["\u52A8\u4F5C", "\u884C\u52A8", "\u64CD\u4F5C"],
actionInput: ["\u52A8\u4F5C\u8F93\u5165", "\u64CD\u4F5C\u8F93\u5165", "\u8F93\u5165", "\u53C2\u6570"],
finalAnswer: ["\u6700\u7EC8\u7B54\u6848", "\u7B54\u6848", "\u7ED3\u679C", "\u56DE\u7B54"],
observation: ["\u89C2\u5BDF", "\u7ED3\u679C", "\u8FD4\u56DE"]
},
english: {
thought: ["thought", "thinking", "think"],
action: ["action", "act"],
actionInput: ["action input", "action_input", "input"],
finalAnswer: ["final answer", "answer", "result"],
observation: ["observation", "obs"]
}
};
}
/**
* 通义千问特殊格式处理
* 处理一些通义千问模型特有的输出变体
*/
extractCustomAction(text) {
const willDoPattern = /我将(?:使用|调用|执行)([^，。\n]+)(?:工具|功能)[，。]?(?:来|去)?([^。\n]*)/;
const willDoMatch = text.match(willDoPattern);
if (willDoMatch) {
return {
action: willDoMatch[1].trim(),
actionInput: willDoMatch[2].trim() || "default",
thought: "\u9700\u8981\u4F7F\u7528\u5DE5\u5177"
};
}
const executePattern = /(?:^|\n)\s*执行\s*[:：]\s*([^\n]+)/im;
const executeMatch = text.match(executePattern);
if (executeMatch) {
const parts = executeMatch[1].split(/[，,]/);
if (parts.length >= 2) {
return {
action: parts[0].trim(),
actionInput: parts.slice(1).join(",").trim(),
thought: "\u6267\u884C\u5DE5\u5177"
};
}
}
return null;
}
/**
* 重写格式化指令，添加通义千问特殊说明
*/
buildFormatInstructions() {
const base = super.buildFormatInstructions();
return `${base}
**\u901A\u4E49\u5343\u95EE\u7279\u6B8A\u683C\u5F0F\u652F\u6301\uFF1A**
- \u4E5F\u652F\u6301\uFF1A\u6211\u5C06\u4F7F\u7528[\u5DE5\u5177\u540D]\u6765[\u53C2\u6570\u63CF\u8FF0]
- \u4E5F\u652F\u6301\uFF1A\u6267\u884C: [\u5DE5\u5177\u540D], [\u53C2\u6570]
- \u652F\u6301\u89C2\u5BDF\u5173\u952E\u5B57\uFF1A\u89C2\u5BDF:, \u7ED3\u679C:, \u8FD4\u56DE:`;
}
};
// src/factory.ts
var import_output_parsers2 = require("@langchain/core/output_parsers");
var ChineseReActParserFactory = class {
/**
* 创建指定类型的解析器
*/
static createParser(modelType, options = {}) {
switch (modelType) {
case "qwen":
return new QwenReActOutputParser(options);
case "chatglm":
return new ChatGLMReActOutputParser(options);
case "baichuan":
return new BaichuanReActOutputParser(options);
case "ernie":
return new ERNIEReActOutputParser(options);
case "glm":
return new ChatGLMReActOutputParser(options);
case "minimax":
return new QwenReActOutputParser(options);
case "auto":
return new UniversalChineseReActParser(options);
default:
throw new Error(`\u4E0D\u652F\u6301\u7684\u6A21\u578B\u7C7B\u578B: ${modelType}`);
}
}
/**
* 获取所有支持的模型类型
*/
static getSupportedModels() {
return ["qwen", "chatglm", "baichuan", "glm", "ernie", "minimax", "auto"];
}
/**
* 检查是否支持指定模型
*/
static isSupported(modelType) {
return this.getSupportedModels().includes(modelType);
}
};
var UniversalChineseReActParser = class extends import_output_parsers2.BaseOutputParser {
constructor(options = {}) {
super();
this.lc_namespace = ["langchain-react-chinese-parser", "parsers", "universal"];
this.options = options;
this.parsers = [
new QwenReActOutputParser(options),
new ChatGLMReActOutputParser(options),
new BaichuanReActOutputParser(options),
new ERNIEReActOutputParser(options)
];
}
async parse(text) {
const errors = [];
for (const parser of this.parsers) {
try {
const result = await parser.parse(text);
if (this.options.debug) {
console.log(`[Universal] \u6210\u529F\u4F7F\u7528 ${parser.getType()} \u89E3\u6790`);
}
return result;
} catch (error) {
errors.push(error);
if (this.options.debug) {
console.log(`[Universal] ${parser.getType()} \u89E3\u6790\u5931\u8D25:`, error);
}
}
}
throw new Error(
`\u6240\u6709\u89E3\u6790\u5668\u90FD\u65E0\u6CD5\u89E3\u6790\u6B64\u8F93\u51FA\u3002\u9519\u8BEF\u8BE6\u60C5:
${errors.map((e) => e.message).join("\n\n")}`
);
}
getFormatInstructions() {
return `\u8BF7\u6309\u7167\u4EE5\u4E0B\u4EFB\u4E00\u683C\u5F0F\u56DE\u7B54\uFF1A
**\u901A\u4E49\u5343\u95EE\u683C\u5F0F\uFF1A**
\u601D\u8003: \u63CF\u8FF0\u601D\u8003\u8FC7\u7A0B
\u52A8\u4F5C: \u5DE5\u5177\u540D\u79F0
\u52A8\u4F5C\u8F93\u5165: \u5DE5\u5177\u53C2\u6570
**ChatGLM\u683C\u5F0F\uFF1A**
\u601D\u8003\uFF1A\u63CF\u8FF0\u601D\u8003\u8FC7\u7A0B
\u5DE5\u5177\uFF1A\u5DE5\u5177\u540D\u79F0
\u5DE5\u5177\u8F93\u5165\uFF1A\u5DE5\u5177\u53C2\u6570
**\u767E\u5DDD\u683C\u5F0F\uFF1A**
\u601D\u8003: \u63CF\u8FF0\u601D\u8003\u8FC7\u7A0B
\u5DE5\u5177: \u5DE5\u5177\u540D\u79F0
\u5DE5\u5177\u8F93\u5165: \u5DE5\u5177\u53C2\u6570
**\u6587\u5FC3\u4E00\u8A00\u683C\u5F0F\uFF1A**
\u601D\u8003\uFF1A\u63CF\u8FF0\u601D\u8003\u8FC7\u7A0B
\u8C03\u7528\u5DE5\u5177\uFF1A\u5DE5\u5177\u540D\u79F0
\u8F93\u5165\uFF1A\u5DE5\u5177\u53C2\u6570
**\u6700\u7EC8\u7B54\u6848\u683C\u5F0F\uFF08\u4EFB\u4F55\u6A21\u578B\uFF09\uFF1A**
\u601D\u8003: \u63CF\u8FF0\u601D\u8003\u8FC7\u7A0B
\u6700\u7EC8\u7B54\u6848: \u4F60\u7684\u7B54\u6848
**\u6CE8\u610F\uFF1A**
- \u652F\u6301\u4E2D\u82F1\u6587\u5173\u952E\u5B57\u6DF7\u7528
- \u652F\u6301\u4E2D\u6587\u5192\u53F7\uFF1A\u548C\u82F1\u6587\u5192\u53F7:
- \u5DE5\u5177\u540D\u548C\u53C2\u6570\u8981\u6E05\u6670\u660E\u786E`;
}
getType() {
return "universal_chinese_react_parser";
}
};
function createChineseReActParser(modelType, options = {}) {
return ChineseReActParserFactory.createParser(modelType, options);
}
function createQwenParser(options = {}) {
return new QwenReActOutputParser(options);
}
function createChatGLMParser(options = {}) {
return new ChatGLMReActOutputParser(options);
}
function createBaichuanParser(options = {}) {
return new BaichuanReActOutputParser(options);
}
function createERNIEParser(options = {}) {
return new ERNIEReActOutputParser(options);
}
function createAutoParser(options = {}) {
return ChineseReActParserFactory.createParser("auto", options);
}
// src/index.ts
var VERSION = "1.0.0";
var SUPPORTED_MODELS = [
"qwen",
// 通义千问
"chatglm",
// ChatGLM
"baichuan",
// 百川
"glm",
// 智谱GLM
"ernie",
// 文心一言
"minimax",
// Minimax
"auto"
// 自动识别
];
// Annotate the CommonJS export names for ESM import in node:
0 && (module.exports = {
BaichuanReActOutputParser,
BaseChineseReActParser,
ChatGLMReActOutputParser,
ChineseReActParserFactory,
ERNIEReActOutputParser,
QwenReActOutputParser,
SUPPORTED_MODELS,
VERSION,
createAutoParser,
createBaichuanParser,
createChatGLMParser,
createChineseReActParser,
createERNIEParser,
createQwenParser
});
//# sourceMappingURL=index.cjs.map