@langchain/classic 1.0.27 → 1.0.28

package/dist/output_parsers/openai_tools.cjs.map

@@ -1 +1 @@
-{"version":3,"file":"openai_tools.cjs","names":["BaseLLMOutputParser"],"sources":["../../src/output_parsers/openai_tools.ts"],"sourcesContent":["import { BaseLLMOutputParser } from \"@langchain/core/output_parsers\";\nimport type { ChatGeneration } from \"@langchain/core/outputs\";\n\n/**\n * @deprecated Import from \"@langchain/core/output_parsers/openai_tools\"\n */\nexport type ParsedToolCall = {\n  id?: string;\n\n  type: string;\n\n  // eslint-disable-next-line @typescript-eslint/no-explicit-any\n  args: Record<string, any>;\n\n  /** @deprecated Use `type` instead. Will be removed in 0.2.0. */\n  name: string;\n\n  /** @deprecated Use `args` instead. Will be removed in 0.2.0. */\n  arguments: Record<string, any>; // eslint-disable-line @typescript-eslint/no-explicit-any\n};\n\n/**\n * @deprecated Import from \"@langchain/core/output_parsers/openai_tools\"\n */\nexport type JsonOutputToolsParserParams = {\n  /** Whether to return the tool call id. */\n  returnId?: boolean;\n};\n\n/**\n * @deprecated Import from \"@langchain/core/output_parsers/openai_tools\"\n */\nexport class JsonOutputToolsParser extends BaseLLMOutputParser<\n  ParsedToolCall[]\n> {\n  static lc_name() {\n    return \"JsonOutputToolsParser\";\n  }\n\n  returnId = false;\n\n  lc_namespace = [\"langchain\", \"output_parsers\", \"openai_tools\"];\n\n  lc_serializable = true;\n\n  constructor(fields?: JsonOutputToolsParserParams) {\n    super(fields);\n    this.returnId = fields?.returnId ?? this.returnId;\n  }\n\n  /**\n   * Parses the output and returns a JSON object. If `argsOnly` is true,\n   * only the arguments of the function call are returned.\n   * @param generations The output of the LLM to parse.\n   * @returns A JSON object representation of the function call or its arguments.\n   */\n  async parseResult(generations: ChatGeneration[]): Promise<ParsedToolCall[]> {\n    const toolCalls = generations[0].message.additional_kwargs.tool_calls;\n    if (!toolCalls) {\n      throw new Error(\n        `No tools_call in message ${JSON.stringify(generations)}`\n      );\n    }\n    const clonedToolCalls = JSON.parse(JSON.stringify(toolCalls));\n    const parsedToolCalls = [];\n    for (const toolCall of clonedToolCalls) {\n      if (toolCall.function !== undefined) {\n        // @ts-expect-error name and arguemnts are defined by Object.defineProperty\n        const parsedToolCall: ParsedToolCall = {\n          type: toolCall.function.name,\n          args: JSON.parse(toolCall.function.arguments),\n        };\n\n        if (this.returnId) {\n          parsedToolCall.id = toolCall.id;\n        }\n\n        // backward-compatibility with previous\n        // versions of Langchain JS, which uses `name` and `arguments`\n        Object.defineProperty(parsedToolCall, \"name\", {\n          get() {\n            return this.type;\n          },\n        });\n\n        Object.defineProperty(parsedToolCall, \"arguments\", {\n          get() {\n            return this.args;\n          },\n        });\n\n        parsedToolCalls.push(parsedToolCall);\n      }\n    }\n    return parsedToolCalls;\n  }\n}\n\nexport type JsonOutputKeyToolsParserParams = {\n  keyName: string;\n  returnSingle?: boolean;\n  /** Whether to return the tool call id. */\n  returnId?: boolean;\n};\n\n/**\n * @deprecated Import from \"@langchain/core/output_parsers/openai_tools\"\n */\n// eslint-disable-next-line @typescript-eslint/no-explicit-any\nexport class JsonOutputKeyToolsParser extends BaseLLMOutputParser<any> {\n  static lc_name() {\n    return \"JsonOutputKeyToolsParser\";\n  }\n\n  lc_namespace = [\"langchain\", \"output_parsers\", \"openai_tools\"];\n\n  lc_serializable = true;\n\n  returnId = false;\n\n  /** The type of tool calls to return. */\n  keyName: string;\n\n  /** Whether to return only the first tool call. */\n  returnSingle = false;\n\n  initialParser: JsonOutputToolsParser;\n\n  constructor(params: JsonOutputKeyToolsParserParams) {\n    super(params);\n    this.keyName = params.keyName;\n    this.returnSingle = params.returnSingle ?? this.returnSingle;\n    this.initialParser = new JsonOutputToolsParser(params);\n  }\n\n  // eslint-disable-next-line @typescript-eslint/no-explicit-any\n  async parseResult(generations: ChatGeneration[]): Promise<any> {\n    const results = await this.initialParser.parseResult(generations);\n    const matchingResults = results.filter(\n      (result) => result.type === this.keyName\n    );\n    // eslint-disable-next-line @typescript-eslint/no-explicit-any\n    let returnedValues: ParsedToolCall[] | Record<string, any>[] =\n      matchingResults;\n    if (!this.returnId) {\n      returnedValues = matchingResults.map((result) => result.args);\n    }\n    if (this.returnSingle) {\n      return returnedValues[0];\n    }\n    return returnedValues;\n  }\n}\n"],"mappings":";;;;;;AAgCA,IAAa,wBAAb,cAA2CA,+BAAAA,oBAEzC;CACA,OAAO,UAAU;AACf,SAAO;;CAGT,WAAW;CAEX,eAAe;EAAC;EAAa;EAAkB;EAAe;CAE9D,kBAAkB;CAElB,YAAY,QAAsC;AAChD,QAAM,OAAO;AACb,OAAK,WAAW,QAAQ,YAAY,KAAK;;;;;;;;CAS3C,MAAM,YAAY,aAA0D;EAC1E,MAAM,YAAY,YAAY,GAAG,QAAQ,kBAAkB;AAC3D,MAAI,CAAC,UACH,OAAM,IAAI,MACR,4BAA4B,KAAK,UAAU,YAAY,GACxD;EAEH,MAAM,kBAAkB,KAAK,MAAM,KAAK,UAAU,UAAU,CAAC;EAC7D,MAAM,kBAAkB,EAAE;AAC1B,OAAK,MAAM,YAAY,gBACrB,KAAI,SAAS,aAAa,KAAA,GAAW;GAEnC,MAAM,iBAAiC;IACrC,MAAM,SAAS,SAAS;IACxB,MAAM,KAAK,MAAM,SAAS,SAAS,UAAU;IAC9C;AAED,OAAI,KAAK,SACP,gBAAe,KAAK,SAAS;AAK/B,UAAO,eAAe,gBAAgB,QAAQ,EAC5C,MAAM;AACJ,WAAO,KAAK;MAEf,CAAC;AAEF,UAAO,eAAe,gBAAgB,aAAa,EACjD,MAAM;AACJ,WAAO,KAAK;MAEf,CAAC;AAEF,mBAAgB,KAAK,eAAe;;AAGxC,SAAO;;;;;;AAeX,IAAa,2BAAb,cAA8CA,+BAAAA,oBAAyB;CACrE,OAAO,UAAU;AACf,SAAO;;CAGT,eAAe;EAAC;EAAa;EAAkB;EAAe;CAE9D,kBAAkB;CAElB,WAAW;;CAGX;;CAGA,eAAe;CAEf;CAEA,YAAY,QAAwC;AAClD,QAAM,OAAO;AACb,OAAK,UAAU,OAAO;AACtB,OAAK,eAAe,OAAO,gBAAgB,KAAK;AAChD,OAAK,gBAAgB,IAAI,sBAAsB,OAAO;;CAIxD,MAAM,YAAY,aAA6C;EAE7D,MAAM,mBADU,MAAM,KAAK,cAAc,YAAY,YAAY,EACjC,QAC7B,WAAW,OAAO,SAAS,KAAK,QAClC;EAED,IAAI,iBACF;AACF,MAAI,CAAC,KAAK,SACR,kBAAiB,gBAAgB,KAAK,WAAW,OAAO,KAAK;AAE/D,MAAI,KAAK,aACP,QAAO,eAAe;AAExB,SAAO"}
+{"version":3,"file":"openai_tools.cjs","names":["BaseLLMOutputParser"],"sources":["../../src/output_parsers/openai_tools.ts"],"sourcesContent":["import { BaseLLMOutputParser } from \"@langchain/core/output_parsers\";\nimport type { ChatGeneration } from \"@langchain/core/outputs\";\n\n/**\n * @deprecated Import from \"@langchain/core/output_parsers/openai_tools\"\n */\nexport type ParsedToolCall = {\n  id?: string;\n\n  type: string;\n\n  // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n  args: Record<string, any>;\n\n  /** @deprecated Use `type` instead. Will be removed in 0.2.0. */\n  name: string;\n\n  /** @deprecated Use `args` instead. Will be removed in 0.2.0. */\n  arguments: Record<string, any>; // oxlint-disable-line @typescript-eslint/no-explicit-any\n};\n\n/**\n * @deprecated Import from \"@langchain/core/output_parsers/openai_tools\"\n */\nexport type JsonOutputToolsParserParams = {\n  /** Whether to return the tool call id. */\n  returnId?: boolean;\n};\n\n/**\n * @deprecated Import from \"@langchain/core/output_parsers/openai_tools\"\n */\nexport class JsonOutputToolsParser extends BaseLLMOutputParser<\n  ParsedToolCall[]\n> {\n  static lc_name() {\n    return \"JsonOutputToolsParser\";\n  }\n\n  returnId = false;\n\n  lc_namespace = [\"langchain\", \"output_parsers\", \"openai_tools\"];\n\n  lc_serializable = true;\n\n  constructor(fields?: JsonOutputToolsParserParams) {\n    super(fields);\n    this.returnId = fields?.returnId ?? this.returnId;\n  }\n\n  /**\n   * Parses the output and returns a JSON object. If `argsOnly` is true,\n   * only the arguments of the function call are returned.\n   * @param generations The output of the LLM to parse.\n   * @returns A JSON object representation of the function call or its arguments.\n   */\n  async parseResult(generations: ChatGeneration[]): Promise<ParsedToolCall[]> {\n    const toolCalls = generations[0].message.additional_kwargs.tool_calls;\n    if (!toolCalls) {\n      throw new Error(\n        `No tools_call in message ${JSON.stringify(generations)}`\n      );\n    }\n    const clonedToolCalls = JSON.parse(JSON.stringify(toolCalls));\n    const parsedToolCalls = [];\n    for (const toolCall of clonedToolCalls) {\n      if (toolCall.function !== undefined) {\n        // @ts-expect-error name and arguemnts are defined by Object.defineProperty\n        const parsedToolCall: ParsedToolCall = {\n          type: toolCall.function.name,\n          args: JSON.parse(toolCall.function.arguments),\n        };\n\n        if (this.returnId) {\n          parsedToolCall.id = toolCall.id;\n        }\n\n        // backward-compatibility with previous\n        // versions of Langchain JS, which uses `name` and `arguments`\n        Object.defineProperty(parsedToolCall, \"name\", {\n          get() {\n            return this.type;\n          },\n        });\n\n        Object.defineProperty(parsedToolCall, \"arguments\", {\n          get() {\n            return this.args;\n          },\n        });\n\n        parsedToolCalls.push(parsedToolCall);\n      }\n    }\n    return parsedToolCalls;\n  }\n}\n\nexport type JsonOutputKeyToolsParserParams = {\n  keyName: string;\n  returnSingle?: boolean;\n  /** Whether to return the tool call id. */\n  returnId?: boolean;\n};\n\n/**\n * @deprecated Import from \"@langchain/core/output_parsers/openai_tools\"\n */\n// oxlint-disable-next-line @typescript-eslint/no-explicit-any\nexport class JsonOutputKeyToolsParser extends BaseLLMOutputParser<any> {\n  static lc_name() {\n    return \"JsonOutputKeyToolsParser\";\n  }\n\n  lc_namespace = [\"langchain\", \"output_parsers\", \"openai_tools\"];\n\n  lc_serializable = true;\n\n  returnId = false;\n\n  /** The type of tool calls to return. */\n  keyName: string;\n\n  /** Whether to return only the first tool call. */\n  returnSingle = false;\n\n  initialParser: JsonOutputToolsParser;\n\n  constructor(params: JsonOutputKeyToolsParserParams) {\n    super(params);\n    this.keyName = params.keyName;\n    this.returnSingle = params.returnSingle ?? this.returnSingle;\n    this.initialParser = new JsonOutputToolsParser(params);\n  }\n\n  // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n  async parseResult(generations: ChatGeneration[]): Promise<any> {\n    const results = await this.initialParser.parseResult(generations);\n    const matchingResults = results.filter(\n      (result) => result.type === this.keyName\n    );\n    // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n    let returnedValues: ParsedToolCall[] | Record<string, any>[] =\n      matchingResults;\n    if (!this.returnId) {\n      returnedValues = matchingResults.map((result) => result.args);\n    }\n    if (this.returnSingle) {\n      return returnedValues[0];\n    }\n    return returnedValues;\n  }\n}\n"],"mappings":";;;;;;AAgCA,IAAa,wBAAb,cAA2CA,+BAAAA,oBAEzC;CACA,OAAO,UAAU;AACf,SAAO;;CAGT,WAAW;CAEX,eAAe;EAAC;EAAa;EAAkB;EAAe;CAE9D,kBAAkB;CAElB,YAAY,QAAsC;AAChD,QAAM,OAAO;AACb,OAAK,WAAW,QAAQ,YAAY,KAAK;;;;;;;;CAS3C,MAAM,YAAY,aAA0D;EAC1E,MAAM,YAAY,YAAY,GAAG,QAAQ,kBAAkB;AAC3D,MAAI,CAAC,UACH,OAAM,IAAI,MACR,4BAA4B,KAAK,UAAU,YAAY,GACxD;EAEH,MAAM,kBAAkB,KAAK,MAAM,KAAK,UAAU,UAAU,CAAC;EAC7D,MAAM,kBAAkB,EAAE;AAC1B,OAAK,MAAM,YAAY,gBACrB,KAAI,SAAS,aAAa,KAAA,GAAW;GAEnC,MAAM,iBAAiC;IACrC,MAAM,SAAS,SAAS;IACxB,MAAM,KAAK,MAAM,SAAS,SAAS,UAAU;IAC9C;AAED,OAAI,KAAK,SACP,gBAAe,KAAK,SAAS;AAK/B,UAAO,eAAe,gBAAgB,QAAQ,EAC5C,MAAM;AACJ,WAAO,KAAK;MAEf,CAAC;AAEF,UAAO,eAAe,gBAAgB,aAAa,EACjD,MAAM;AACJ,WAAO,KAAK;MAEf,CAAC;AAEF,mBAAgB,KAAK,eAAe;;AAGxC,SAAO;;;;;;AAeX,IAAa,2BAAb,cAA8CA,+BAAAA,oBAAyB;CACrE,OAAO,UAAU;AACf,SAAO;;CAGT,eAAe;EAAC;EAAa;EAAkB;EAAe;CAE9D,kBAAkB;CAElB,WAAW;;CAGX;;CAGA,eAAe;CAEf;CAEA,YAAY,QAAwC;AAClD,QAAM,OAAO;AACb,OAAK,UAAU,OAAO;AACtB,OAAK,eAAe,OAAO,gBAAgB,KAAK;AAChD,OAAK,gBAAgB,IAAI,sBAAsB,OAAO;;CAIxD,MAAM,YAAY,aAA6C;EAE7D,MAAM,mBADU,MAAM,KAAK,cAAc,YAAY,YAAY,EACjC,QAC7B,WAAW,OAAO,SAAS,KAAK,QAClC;EAED,IAAI,iBACF;AACF,MAAI,CAAC,KAAK,SACR,kBAAiB,gBAAgB,KAAK,WAAW,OAAO,KAAK;AAE/D,MAAI,KAAK,aACP,QAAO,eAAe;AAExB,SAAO"}

package/dist/output_parsers/openai_tools.js.map

@@ -1 +1 @@
-{"version":3,"file":"openai_tools.js","names":[],"sources":["../../src/output_parsers/openai_tools.ts"],"sourcesContent":["import { BaseLLMOutputParser } from \"@langchain/core/output_parsers\";\nimport type { ChatGeneration } from \"@langchain/core/outputs\";\n\n/**\n * @deprecated Import from \"@langchain/core/output_parsers/openai_tools\"\n */\nexport type ParsedToolCall = {\n  id?: string;\n\n  type: string;\n\n  // eslint-disable-next-line @typescript-eslint/no-explicit-any\n  args: Record<string, any>;\n\n  /** @deprecated Use `type` instead. Will be removed in 0.2.0. */\n  name: string;\n\n  /** @deprecated Use `args` instead. Will be removed in 0.2.0. */\n  arguments: Record<string, any>; // eslint-disable-line @typescript-eslint/no-explicit-any\n};\n\n/**\n * @deprecated Import from \"@langchain/core/output_parsers/openai_tools\"\n */\nexport type JsonOutputToolsParserParams = {\n  /** Whether to return the tool call id. */\n  returnId?: boolean;\n};\n\n/**\n * @deprecated Import from \"@langchain/core/output_parsers/openai_tools\"\n */\nexport class JsonOutputToolsParser extends BaseLLMOutputParser<\n  ParsedToolCall[]\n> {\n  static lc_name() {\n    return \"JsonOutputToolsParser\";\n  }\n\n  returnId = false;\n\n  lc_namespace = [\"langchain\", \"output_parsers\", \"openai_tools\"];\n\n  lc_serializable = true;\n\n  constructor(fields?: JsonOutputToolsParserParams) {\n    super(fields);\n    this.returnId = fields?.returnId ?? this.returnId;\n  }\n\n  /**\n   * Parses the output and returns a JSON object. If `argsOnly` is true,\n   * only the arguments of the function call are returned.\n   * @param generations The output of the LLM to parse.\n   * @returns A JSON object representation of the function call or its arguments.\n   */\n  async parseResult(generations: ChatGeneration[]): Promise<ParsedToolCall[]> {\n    const toolCalls = generations[0].message.additional_kwargs.tool_calls;\n    if (!toolCalls) {\n      throw new Error(\n        `No tools_call in message ${JSON.stringify(generations)}`\n      );\n    }\n    const clonedToolCalls = JSON.parse(JSON.stringify(toolCalls));\n    const parsedToolCalls = [];\n    for (const toolCall of clonedToolCalls) {\n      if (toolCall.function !== undefined) {\n        // @ts-expect-error name and arguemnts are defined by Object.defineProperty\n        const parsedToolCall: ParsedToolCall = {\n          type: toolCall.function.name,\n          args: JSON.parse(toolCall.function.arguments),\n        };\n\n        if (this.returnId) {\n          parsedToolCall.id = toolCall.id;\n        }\n\n        // backward-compatibility with previous\n        // versions of Langchain JS, which uses `name` and `arguments`\n        Object.defineProperty(parsedToolCall, \"name\", {\n          get() {\n            return this.type;\n          },\n        });\n\n        Object.defineProperty(parsedToolCall, \"arguments\", {\n          get() {\n            return this.args;\n          },\n        });\n\n        parsedToolCalls.push(parsedToolCall);\n      }\n    }\n    return parsedToolCalls;\n  }\n}\n\nexport type JsonOutputKeyToolsParserParams = {\n  keyName: string;\n  returnSingle?: boolean;\n  /** Whether to return the tool call id. */\n  returnId?: boolean;\n};\n\n/**\n * @deprecated Import from \"@langchain/core/output_parsers/openai_tools\"\n */\n// eslint-disable-next-line @typescript-eslint/no-explicit-any\nexport class JsonOutputKeyToolsParser extends BaseLLMOutputParser<any> {\n  static lc_name() {\n    return \"JsonOutputKeyToolsParser\";\n  }\n\n  lc_namespace = [\"langchain\", \"output_parsers\", \"openai_tools\"];\n\n  lc_serializable = true;\n\n  returnId = false;\n\n  /** The type of tool calls to return. */\n  keyName: string;\n\n  /** Whether to return only the first tool call. */\n  returnSingle = false;\n\n  initialParser: JsonOutputToolsParser;\n\n  constructor(params: JsonOutputKeyToolsParserParams) {\n    super(params);\n    this.keyName = params.keyName;\n    this.returnSingle = params.returnSingle ?? this.returnSingle;\n    this.initialParser = new JsonOutputToolsParser(params);\n  }\n\n  // eslint-disable-next-line @typescript-eslint/no-explicit-any\n  async parseResult(generations: ChatGeneration[]): Promise<any> {\n    const results = await this.initialParser.parseResult(generations);\n    const matchingResults = results.filter(\n      (result) => result.type === this.keyName\n    );\n    // eslint-disable-next-line @typescript-eslint/no-explicit-any\n    let returnedValues: ParsedToolCall[] | Record<string, any>[] =\n      matchingResults;\n    if (!this.returnId) {\n      returnedValues = matchingResults.map((result) => result.args);\n    }\n    if (this.returnSingle) {\n      return returnedValues[0];\n    }\n    return returnedValues;\n  }\n}\n"],"mappings":";;;;;AAgCA,IAAa,wBAAb,cAA2C,oBAEzC;CACA,OAAO,UAAU;AACf,SAAO;;CAGT,WAAW;CAEX,eAAe;EAAC;EAAa;EAAkB;EAAe;CAE9D,kBAAkB;CAElB,YAAY,QAAsC;AAChD,QAAM,OAAO;AACb,OAAK,WAAW,QAAQ,YAAY,KAAK;;;;;;;;CAS3C,MAAM,YAAY,aAA0D;EAC1E,MAAM,YAAY,YAAY,GAAG,QAAQ,kBAAkB;AAC3D,MAAI,CAAC,UACH,OAAM,IAAI,MACR,4BAA4B,KAAK,UAAU,YAAY,GACxD;EAEH,MAAM,kBAAkB,KAAK,MAAM,KAAK,UAAU,UAAU,CAAC;EAC7D,MAAM,kBAAkB,EAAE;AAC1B,OAAK,MAAM,YAAY,gBACrB,KAAI,SAAS,aAAa,KAAA,GAAW;GAEnC,MAAM,iBAAiC;IACrC,MAAM,SAAS,SAAS;IACxB,MAAM,KAAK,MAAM,SAAS,SAAS,UAAU;IAC9C;AAED,OAAI,KAAK,SACP,gBAAe,KAAK,SAAS;AAK/B,UAAO,eAAe,gBAAgB,QAAQ,EAC5C,MAAM;AACJ,WAAO,KAAK;MAEf,CAAC;AAEF,UAAO,eAAe,gBAAgB,aAAa,EACjD,MAAM;AACJ,WAAO,KAAK;MAEf,CAAC;AAEF,mBAAgB,KAAK,eAAe;;AAGxC,SAAO;;;;;;AAeX,IAAa,2BAAb,cAA8C,oBAAyB;CACrE,OAAO,UAAU;AACf,SAAO;;CAGT,eAAe;EAAC;EAAa;EAAkB;EAAe;CAE9D,kBAAkB;CAElB,WAAW;;CAGX;;CAGA,eAAe;CAEf;CAEA,YAAY,QAAwC;AAClD,QAAM,OAAO;AACb,OAAK,UAAU,OAAO;AACtB,OAAK,eAAe,OAAO,gBAAgB,KAAK;AAChD,OAAK,gBAAgB,IAAI,sBAAsB,OAAO;;CAIxD,MAAM,YAAY,aAA6C;EAE7D,MAAM,mBADU,MAAM,KAAK,cAAc,YAAY,YAAY,EACjC,QAC7B,WAAW,OAAO,SAAS,KAAK,QAClC;EAED,IAAI,iBACF;AACF,MAAI,CAAC,KAAK,SACR,kBAAiB,gBAAgB,KAAK,WAAW,OAAO,KAAK;AAE/D,MAAI,KAAK,aACP,QAAO,eAAe;AAExB,SAAO"}
+{"version":3,"file":"openai_tools.js","names":[],"sources":["../../src/output_parsers/openai_tools.ts"],"sourcesContent":["import { BaseLLMOutputParser } from \"@langchain/core/output_parsers\";\nimport type { ChatGeneration } from \"@langchain/core/outputs\";\n\n/**\n * @deprecated Import from \"@langchain/core/output_parsers/openai_tools\"\n */\nexport type ParsedToolCall = {\n  id?: string;\n\n  type: string;\n\n  // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n  args: Record<string, any>;\n\n  /** @deprecated Use `type` instead. Will be removed in 0.2.0. */\n  name: string;\n\n  /** @deprecated Use `args` instead. Will be removed in 0.2.0. */\n  arguments: Record<string, any>; // oxlint-disable-line @typescript-eslint/no-explicit-any\n};\n\n/**\n * @deprecated Import from \"@langchain/core/output_parsers/openai_tools\"\n */\nexport type JsonOutputToolsParserParams = {\n  /** Whether to return the tool call id. */\n  returnId?: boolean;\n};\n\n/**\n * @deprecated Import from \"@langchain/core/output_parsers/openai_tools\"\n */\nexport class JsonOutputToolsParser extends BaseLLMOutputParser<\n  ParsedToolCall[]\n> {\n  static lc_name() {\n    return \"JsonOutputToolsParser\";\n  }\n\n  returnId = false;\n\n  lc_namespace = [\"langchain\", \"output_parsers\", \"openai_tools\"];\n\n  lc_serializable = true;\n\n  constructor(fields?: JsonOutputToolsParserParams) {\n    super(fields);\n    this.returnId = fields?.returnId ?? this.returnId;\n  }\n\n  /**\n   * Parses the output and returns a JSON object. If `argsOnly` is true,\n   * only the arguments of the function call are returned.\n   * @param generations The output of the LLM to parse.\n   * @returns A JSON object representation of the function call or its arguments.\n   */\n  async parseResult(generations: ChatGeneration[]): Promise<ParsedToolCall[]> {\n    const toolCalls = generations[0].message.additional_kwargs.tool_calls;\n    if (!toolCalls) {\n      throw new Error(\n        `No tools_call in message ${JSON.stringify(generations)}`\n      );\n    }\n    const clonedToolCalls = JSON.parse(JSON.stringify(toolCalls));\n    const parsedToolCalls = [];\n    for (const toolCall of clonedToolCalls) {\n      if (toolCall.function !== undefined) {\n        // @ts-expect-error name and arguemnts are defined by Object.defineProperty\n        const parsedToolCall: ParsedToolCall = {\n          type: toolCall.function.name,\n          args: JSON.parse(toolCall.function.arguments),\n        };\n\n        if (this.returnId) {\n          parsedToolCall.id = toolCall.id;\n        }\n\n        // backward-compatibility with previous\n        // versions of Langchain JS, which uses `name` and `arguments`\n        Object.defineProperty(parsedToolCall, \"name\", {\n          get() {\n            return this.type;\n          },\n        });\n\n        Object.defineProperty(parsedToolCall, \"arguments\", {\n          get() {\n            return this.args;\n          },\n        });\n\n        parsedToolCalls.push(parsedToolCall);\n      }\n    }\n    return parsedToolCalls;\n  }\n}\n\nexport type JsonOutputKeyToolsParserParams = {\n  keyName: string;\n  returnSingle?: boolean;\n  /** Whether to return the tool call id. */\n  returnId?: boolean;\n};\n\n/**\n * @deprecated Import from \"@langchain/core/output_parsers/openai_tools\"\n */\n// oxlint-disable-next-line @typescript-eslint/no-explicit-any\nexport class JsonOutputKeyToolsParser extends BaseLLMOutputParser<any> {\n  static lc_name() {\n    return \"JsonOutputKeyToolsParser\";\n  }\n\n  lc_namespace = [\"langchain\", \"output_parsers\", \"openai_tools\"];\n\n  lc_serializable = true;\n\n  returnId = false;\n\n  /** The type of tool calls to return. */\n  keyName: string;\n\n  /** Whether to return only the first tool call. */\n  returnSingle = false;\n\n  initialParser: JsonOutputToolsParser;\n\n  constructor(params: JsonOutputKeyToolsParserParams) {\n    super(params);\n    this.keyName = params.keyName;\n    this.returnSingle = params.returnSingle ?? this.returnSingle;\n    this.initialParser = new JsonOutputToolsParser(params);\n  }\n\n  // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n  async parseResult(generations: ChatGeneration[]): Promise<any> {\n    const results = await this.initialParser.parseResult(generations);\n    const matchingResults = results.filter(\n      (result) => result.type === this.keyName\n    );\n    // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n    let returnedValues: ParsedToolCall[] | Record<string, any>[] =\n      matchingResults;\n    if (!this.returnId) {\n      returnedValues = matchingResults.map((result) => result.args);\n    }\n    if (this.returnSingle) {\n      return returnedValues[0];\n    }\n    return returnedValues;\n  }\n}\n"],"mappings":";;;;;AAgCA,IAAa,wBAAb,cAA2C,oBAEzC;CACA,OAAO,UAAU;AACf,SAAO;;CAGT,WAAW;CAEX,eAAe;EAAC;EAAa;EAAkB;EAAe;CAE9D,kBAAkB;CAElB,YAAY,QAAsC;AAChD,QAAM,OAAO;AACb,OAAK,WAAW,QAAQ,YAAY,KAAK;;;;;;;;CAS3C,MAAM,YAAY,aAA0D;EAC1E,MAAM,YAAY,YAAY,GAAG,QAAQ,kBAAkB;AAC3D,MAAI,CAAC,UACH,OAAM,IAAI,MACR,4BAA4B,KAAK,UAAU,YAAY,GACxD;EAEH,MAAM,kBAAkB,KAAK,MAAM,KAAK,UAAU,UAAU,CAAC;EAC7D,MAAM,kBAAkB,EAAE;AAC1B,OAAK,MAAM,YAAY,gBACrB,KAAI,SAAS,aAAa,KAAA,GAAW;GAEnC,MAAM,iBAAiC;IACrC,MAAM,SAAS,SAAS;IACxB,MAAM,KAAK,MAAM,SAAS,SAAS,UAAU;IAC9C;AAED,OAAI,KAAK,SACP,gBAAe,KAAK,SAAS;AAK/B,UAAO,eAAe,gBAAgB,QAAQ,EAC5C,MAAM;AACJ,WAAO,KAAK;MAEf,CAAC;AAEF,UAAO,eAAe,gBAAgB,aAAa,EACjD,MAAM;AACJ,WAAO,KAAK;MAEf,CAAC;AAEF,mBAAgB,KAAK,eAAe;;AAGxC,SAAO;;;;;;AAeX,IAAa,2BAAb,cAA8C,oBAAyB;CACrE,OAAO,UAAU;AACf,SAAO;;CAGT,eAAe;EAAC;EAAa;EAAkB;EAAe;CAE9D,kBAAkB;CAElB,WAAW;;CAGX;;CAGA,eAAe;CAEf;CAEA,YAAY,QAAwC;AAClD,QAAM,OAAO;AACb,OAAK,UAAU,OAAO;AACtB,OAAK,eAAe,OAAO,gBAAgB,KAAK;AAChD,OAAK,gBAAgB,IAAI,sBAAsB,OAAO;;CAIxD,MAAM,YAAY,aAA6C;EAE7D,MAAM,mBADU,MAAM,KAAK,cAAc,YAAY,YAAY,EACjC,QAC7B,WAAW,OAAO,SAAS,KAAK,QAClC;EAED,IAAI,iBACF;AACF,MAAI,CAAC,KAAK,SACR,kBAAiB,gBAAgB,KAAK,WAAW,OAAO,KAAK;AAE/D,MAAI,KAAK,aACP,QAAO,eAAe;AAExB,SAAO"}

package/dist/output_parsers/regex.cjs.map

@@ -1 +1 @@
-{"version":3,"file":"regex.cjs","names":["BaseOutputParser","OutputParserException"],"sources":["../../src/output_parsers/regex.ts"],"sourcesContent":["import {\n  BaseOutputParser,\n  OutputParserException,\n} from \"@langchain/core/output_parsers\";\nimport type { SerializedFields } from \"../load/map_keys.js\";\n\nexport interface RegExpFields {\n  pattern: string;\n  flags?: string;\n}\n\n/**\n * Interface for the fields required to create a RegexParser instance.\n */\nexport interface RegexParserFields {\n  regex: string | RegExp | RegExpFields;\n  outputKeys: string[];\n  defaultOutputKey?: string;\n}\n\n/**\n * Class to parse the output of an LLM call into a dictionary.\n * @augments BaseOutputParser\n */\nexport class RegexParser extends BaseOutputParser<Record<string, string>> {\n  static lc_name() {\n    return \"RegexParser\";\n  }\n\n  lc_namespace = [\"langchain\", \"output_parsers\", \"regex\"];\n\n  lc_serializable = true;\n\n  get lc_attributes(): SerializedFields | undefined {\n    return {\n      regex: this.lc_kwargs.regex,\n    };\n  }\n\n  regex: string | RegExp;\n\n  outputKeys: string[];\n\n  defaultOutputKey?: string;\n\n  constructor(fields: RegexParserFields);\n\n  constructor(\n    regex: string | RegExp,\n    outputKeys: string[],\n    defaultOutputKey?: string\n  );\n\n  constructor(\n    fields: string | RegExp | RegexParserFields,\n    outputKeys?: string[],\n    defaultOutputKey?: string\n  ) {\n    // eslint-disable-next-line no-instanceof/no-instanceof\n    if (typeof fields === \"string\" || fields instanceof RegExp) {\n      // eslint-disable-next-line no-param-reassign\n      fields = { regex: fields, outputKeys: outputKeys!, defaultOutputKey };\n    }\n    // eslint-disable-next-line no-instanceof/no-instanceof\n    if (fields.regex instanceof RegExp) {\n      fields.regex = {\n        pattern: fields.regex.source,\n        flags: fields.regex.flags,\n      };\n    }\n    super(fields);\n    this.regex =\n      typeof fields.regex === \"string\"\n        ? new RegExp(fields.regex)\n        : \"pattern\" in fields.regex\n          ? new RegExp(fields.regex.pattern, fields.regex.flags)\n          : fields.regex;\n    this.outputKeys = fields.outputKeys;\n    this.defaultOutputKey = fields.defaultOutputKey;\n  }\n\n  _type() {\n    return \"regex_parser\";\n  }\n\n  /**\n   * Parses the given text using the regex pattern and returns a dictionary\n   * with the parsed output. If the regex pattern does not match the text\n   * and no defaultOutputKey is provided, throws an OutputParserException.\n   * @param text The text to be parsed.\n   * @returns A dictionary with the parsed output.\n   */\n  async parse(text: string): Promise<Record<string, string>> {\n    const match = text.match(this.regex);\n    if (match) {\n      return this.outputKeys.reduce(\n        (acc, key, index) => {\n          acc[key] = match[index + 1];\n          return acc;\n        },\n        {} as Record<string, string>\n      );\n    }\n\n    if (this.defaultOutputKey === undefined) {\n      throw new OutputParserException(`Could not parse output: ${text}`, text);\n    }\n\n    return this.outputKeys.reduce(\n      (acc, key) => {\n        acc[key] = key === this.defaultOutputKey ? text : \"\";\n        return acc;\n      },\n      {} as Record<string, string>\n    );\n  }\n\n  /**\n   * Returns a string with instructions on how the LLM output should be\n   * formatted to match the regex pattern.\n   * @returns A string with formatting instructions.\n   */\n  getFormatInstructions(): string {\n    return `Your response should match the following regex: ${this.regex}`;\n  }\n}\n"],"mappings":";;;;;;;AAwBA,IAAa,cAAb,cAAiCA,+BAAAA,iBAAyC;CACxE,OAAO,UAAU;AACf,SAAO;;CAGT,eAAe;EAAC;EAAa;EAAkB;EAAQ;CAEvD,kBAAkB;CAElB,IAAI,gBAA8C;AAChD,SAAO,EACL,OAAO,KAAK,UAAU,OACvB;;CAGH;CAEA;CAEA;CAUA,YACE,QACA,YACA,kBACA;AAEA,MAAI,OAAO,WAAW,YAAY,kBAAkB,OAElD,UAAS;GAAE,OAAO;GAAoB;GAAa;GAAkB;AAGvE,MAAI,OAAO,iBAAiB,OAC1B,QAAO,QAAQ;GACb,SAAS,OAAO,MAAM;GACtB,OAAO,OAAO,MAAM;GACrB;AAEH,QAAM,OAAO;AACb,OAAK,QACH,OAAO,OAAO,UAAU,WACpB,IAAI,OAAO,OAAO,MAAM,GACxB,aAAa,OAAO,QAClB,IAAI,OAAO,OAAO,MAAM,SAAS,OAAO,MAAM,MAAM,GACpD,OAAO;AACf,OAAK,aAAa,OAAO;AACzB,OAAK,mBAAmB,OAAO;;CAGjC,QAAQ;AACN,SAAO;;;;;;;;;CAUT,MAAM,MAAM,MAA+C;EACzD,MAAM,QAAQ,KAAK,MAAM,KAAK,MAAM;AACpC,MAAI,MACF,QAAO,KAAK,WAAW,QACpB,KAAK,KAAK,UAAU;AACnB,OAAI,OAAO,MAAM,QAAQ;AACzB,UAAO;KAET,EAAE,CACH;AAGH,MAAI,KAAK,qBAAqB,KAAA,EAC5B,OAAM,IAAIC,+BAAAA,sBAAsB,2BAA2B,QAAQ,KAAK;AAG1E,SAAO,KAAK,WAAW,QACpB,KAAK,QAAQ;AACZ,OAAI,OAAO,QAAQ,KAAK,mBAAmB,OAAO;AAClD,UAAO;KAET,EAAE,CACH;;;;;;;CAQH,wBAAgC;AAC9B,SAAO,mDAAmD,KAAK"}
+{"version":3,"file":"regex.cjs","names":["BaseOutputParser","OutputParserException"],"sources":["../../src/output_parsers/regex.ts"],"sourcesContent":["import {\n  BaseOutputParser,\n  OutputParserException,\n} from \"@langchain/core/output_parsers\";\nimport type { SerializedFields } from \"../load/map_keys.js\";\n\nexport interface RegExpFields {\n  pattern: string;\n  flags?: string;\n}\n\n/**\n * Interface for the fields required to create a RegexParser instance.\n */\nexport interface RegexParserFields {\n  regex: string | RegExp | RegExpFields;\n  outputKeys: string[];\n  defaultOutputKey?: string;\n}\n\n/**\n * Class to parse the output of an LLM call into a dictionary.\n * @augments BaseOutputParser\n */\nexport class RegexParser extends BaseOutputParser<Record<string, string>> {\n  static lc_name() {\n    return \"RegexParser\";\n  }\n\n  lc_namespace = [\"langchain\", \"output_parsers\", \"regex\"];\n\n  lc_serializable = true;\n\n  get lc_attributes(): SerializedFields | undefined {\n    return {\n      regex: this.lc_kwargs.regex,\n    };\n  }\n\n  regex: string | RegExp;\n\n  outputKeys: string[];\n\n  defaultOutputKey?: string;\n\n  constructor(fields: RegexParserFields);\n\n  constructor(\n    regex: string | RegExp,\n    outputKeys: string[],\n    defaultOutputKey?: string\n  );\n\n  constructor(\n    fields: string | RegExp | RegexParserFields,\n    outputKeys?: string[],\n    defaultOutputKey?: string\n  ) {\n    // oxlint-disable-next-line no-instanceof/no-instanceof\n    if (typeof fields === \"string\" || fields instanceof RegExp) {\n      // oxlint-disable-next-line no-param-reassign\n      fields = { regex: fields, outputKeys: outputKeys!, defaultOutputKey };\n    }\n    // oxlint-disable-next-line no-instanceof/no-instanceof\n    if (fields.regex instanceof RegExp) {\n      fields.regex = {\n        pattern: fields.regex.source,\n        flags: fields.regex.flags,\n      };\n    }\n    super(fields);\n    this.regex =\n      typeof fields.regex === \"string\"\n        ? new RegExp(fields.regex)\n        : \"pattern\" in fields.regex\n          ? new RegExp(fields.regex.pattern, fields.regex.flags)\n          : fields.regex;\n    this.outputKeys = fields.outputKeys;\n    this.defaultOutputKey = fields.defaultOutputKey;\n  }\n\n  _type() {\n    return \"regex_parser\";\n  }\n\n  /**\n   * Parses the given text using the regex pattern and returns a dictionary\n   * with the parsed output. If the regex pattern does not match the text\n   * and no defaultOutputKey is provided, throws an OutputParserException.\n   * @param text The text to be parsed.\n   * @returns A dictionary with the parsed output.\n   */\n  async parse(text: string): Promise<Record<string, string>> {\n    const match = text.match(this.regex);\n    if (match) {\n      return this.outputKeys.reduce(\n        (acc, key, index) => {\n          acc[key] = match[index + 1];\n          return acc;\n        },\n        {} as Record<string, string>\n      );\n    }\n\n    if (this.defaultOutputKey === undefined) {\n      throw new OutputParserException(`Could not parse output: ${text}`, text);\n    }\n\n    return this.outputKeys.reduce(\n      (acc, key) => {\n        acc[key] = key === this.defaultOutputKey ? text : \"\";\n        return acc;\n      },\n      {} as Record<string, string>\n    );\n  }\n\n  /**\n   * Returns a string with instructions on how the LLM output should be\n   * formatted to match the regex pattern.\n   * @returns A string with formatting instructions.\n   */\n  getFormatInstructions(): string {\n    return `Your response should match the following regex: ${this.regex}`;\n  }\n}\n"],"mappings":";;;;;;;AAwBA,IAAa,cAAb,cAAiCA,+BAAAA,iBAAyC;CACxE,OAAO,UAAU;AACf,SAAO;;CAGT,eAAe;EAAC;EAAa;EAAkB;EAAQ;CAEvD,kBAAkB;CAElB,IAAI,gBAA8C;AAChD,SAAO,EACL,OAAO,KAAK,UAAU,OACvB;;CAGH;CAEA;CAEA;CAUA,YACE,QACA,YACA,kBACA;AAEA,MAAI,OAAO,WAAW,YAAY,kBAAkB,OAElD,UAAS;GAAE,OAAO;GAAoB;GAAa;GAAkB;AAGvE,MAAI,OAAO,iBAAiB,OAC1B,QAAO,QAAQ;GACb,SAAS,OAAO,MAAM;GACtB,OAAO,OAAO,MAAM;GACrB;AAEH,QAAM,OAAO;AACb,OAAK,QACH,OAAO,OAAO,UAAU,WACpB,IAAI,OAAO,OAAO,MAAM,GACxB,aAAa,OAAO,QAClB,IAAI,OAAO,OAAO,MAAM,SAAS,OAAO,MAAM,MAAM,GACpD,OAAO;AACf,OAAK,aAAa,OAAO;AACzB,OAAK,mBAAmB,OAAO;;CAGjC,QAAQ;AACN,SAAO;;;;;;;;;CAUT,MAAM,MAAM,MAA+C;EACzD,MAAM,QAAQ,KAAK,MAAM,KAAK,MAAM;AACpC,MAAI,MACF,QAAO,KAAK,WAAW,QACpB,KAAK,KAAK,UAAU;AACnB,OAAI,OAAO,MAAM,QAAQ;AACzB,UAAO;KAET,EAAE,CACH;AAGH,MAAI,KAAK,qBAAqB,KAAA,EAC5B,OAAM,IAAIC,+BAAAA,sBAAsB,2BAA2B,QAAQ,KAAK;AAG1E,SAAO,KAAK,WAAW,QACpB,KAAK,QAAQ;AACZ,OAAI,OAAO,QAAQ,KAAK,mBAAmB,OAAO;AAClD,UAAO;KAET,EAAE,CACH;;;;;;;CAQH,wBAAgC;AAC9B,SAAO,mDAAmD,KAAK"}

package/dist/output_parsers/regex.js.map

@@ -1 +1 @@
-{"version":3,"file":"regex.js","names":[],"sources":["../../src/output_parsers/regex.ts"],"sourcesContent":["import {\n  BaseOutputParser,\n  OutputParserException,\n} from \"@langchain/core/output_parsers\";\nimport type { SerializedFields } from \"../load/map_keys.js\";\n\nexport interface RegExpFields {\n  pattern: string;\n  flags?: string;\n}\n\n/**\n * Interface for the fields required to create a RegexParser instance.\n */\nexport interface RegexParserFields {\n  regex: string | RegExp | RegExpFields;\n  outputKeys: string[];\n  defaultOutputKey?: string;\n}\n\n/**\n * Class to parse the output of an LLM call into a dictionary.\n * @augments BaseOutputParser\n */\nexport class RegexParser extends BaseOutputParser<Record<string, string>> {\n  static lc_name() {\n    return \"RegexParser\";\n  }\n\n  lc_namespace = [\"langchain\", \"output_parsers\", \"regex\"];\n\n  lc_serializable = true;\n\n  get lc_attributes(): SerializedFields | undefined {\n    return {\n      regex: this.lc_kwargs.regex,\n    };\n  }\n\n  regex: string | RegExp;\n\n  outputKeys: string[];\n\n  defaultOutputKey?: string;\n\n  constructor(fields: RegexParserFields);\n\n  constructor(\n    regex: string | RegExp,\n    outputKeys: string[],\n    defaultOutputKey?: string\n  );\n\n  constructor(\n    fields: string | RegExp | RegexParserFields,\n    outputKeys?: string[],\n    defaultOutputKey?: string\n  ) {\n    // eslint-disable-next-line no-instanceof/no-instanceof\n    if (typeof fields === \"string\" || fields instanceof RegExp) {\n      // eslint-disable-next-line no-param-reassign\n      fields = { regex: fields, outputKeys: outputKeys!, defaultOutputKey };\n    }\n    // eslint-disable-next-line no-instanceof/no-instanceof\n    if (fields.regex instanceof RegExp) {\n      fields.regex = {\n        pattern: fields.regex.source,\n        flags: fields.regex.flags,\n      };\n    }\n    super(fields);\n    this.regex =\n      typeof fields.regex === \"string\"\n        ? new RegExp(fields.regex)\n        : \"pattern\" in fields.regex\n          ? new RegExp(fields.regex.pattern, fields.regex.flags)\n          : fields.regex;\n    this.outputKeys = fields.outputKeys;\n    this.defaultOutputKey = fields.defaultOutputKey;\n  }\n\n  _type() {\n    return \"regex_parser\";\n  }\n\n  /**\n   * Parses the given text using the regex pattern and returns a dictionary\n   * with the parsed output. If the regex pattern does not match the text\n   * and no defaultOutputKey is provided, throws an OutputParserException.\n   * @param text The text to be parsed.\n   * @returns A dictionary with the parsed output.\n   */\n  async parse(text: string): Promise<Record<string, string>> {\n    const match = text.match(this.regex);\n    if (match) {\n      return this.outputKeys.reduce(\n        (acc, key, index) => {\n          acc[key] = match[index + 1];\n          return acc;\n        },\n        {} as Record<string, string>\n      );\n    }\n\n    if (this.defaultOutputKey === undefined) {\n      throw new OutputParserException(`Could not parse output: ${text}`, text);\n    }\n\n    return this.outputKeys.reduce(\n      (acc, key) => {\n        acc[key] = key === this.defaultOutputKey ? text : \"\";\n        return acc;\n      },\n      {} as Record<string, string>\n    );\n  }\n\n  /**\n   * Returns a string with instructions on how the LLM output should be\n   * formatted to match the regex pattern.\n   * @returns A string with formatting instructions.\n   */\n  getFormatInstructions(): string {\n    return `Your response should match the following regex: ${this.regex}`;\n  }\n}\n"],"mappings":";;;;;;AAwBA,IAAa,cAAb,cAAiC,iBAAyC;CACxE,OAAO,UAAU;AACf,SAAO;;CAGT,eAAe;EAAC;EAAa;EAAkB;EAAQ;CAEvD,kBAAkB;CAElB,IAAI,gBAA8C;AAChD,SAAO,EACL,OAAO,KAAK,UAAU,OACvB;;CAGH;CAEA;CAEA;CAUA,YACE,QACA,YACA,kBACA;AAEA,MAAI,OAAO,WAAW,YAAY,kBAAkB,OAElD,UAAS;GAAE,OAAO;GAAoB;GAAa;GAAkB;AAGvE,MAAI,OAAO,iBAAiB,OAC1B,QAAO,QAAQ;GACb,SAAS,OAAO,MAAM;GACtB,OAAO,OAAO,MAAM;GACrB;AAEH,QAAM,OAAO;AACb,OAAK,QACH,OAAO,OAAO,UAAU,WACpB,IAAI,OAAO,OAAO,MAAM,GACxB,aAAa,OAAO,QAClB,IAAI,OAAO,OAAO,MAAM,SAAS,OAAO,MAAM,MAAM,GACpD,OAAO;AACf,OAAK,aAAa,OAAO;AACzB,OAAK,mBAAmB,OAAO;;CAGjC,QAAQ;AACN,SAAO;;;;;;;;;CAUT,MAAM,MAAM,MAA+C;EACzD,MAAM,QAAQ,KAAK,MAAM,KAAK,MAAM;AACpC,MAAI,MACF,QAAO,KAAK,WAAW,QACpB,KAAK,KAAK,UAAU;AACnB,OAAI,OAAO,MAAM,QAAQ;AACzB,UAAO;KAET,EAAE,CACH;AAGH,MAAI,KAAK,qBAAqB,KAAA,EAC5B,OAAM,IAAI,sBAAsB,2BAA2B,QAAQ,KAAK;AAG1E,SAAO,KAAK,WAAW,QACpB,KAAK,QAAQ;AACZ,OAAI,OAAO,QAAQ,KAAK,mBAAmB,OAAO;AAClD,UAAO;KAET,EAAE,CACH;;;;;;;CAQH,wBAAgC;AAC9B,SAAO,mDAAmD,KAAK"}
+{"version":3,"file":"regex.js","names":[],"sources":["../../src/output_parsers/regex.ts"],"sourcesContent":["import {\n  BaseOutputParser,\n  OutputParserException,\n} from \"@langchain/core/output_parsers\";\nimport type { SerializedFields } from \"../load/map_keys.js\";\n\nexport interface RegExpFields {\n  pattern: string;\n  flags?: string;\n}\n\n/**\n * Interface for the fields required to create a RegexParser instance.\n */\nexport interface RegexParserFields {\n  regex: string | RegExp | RegExpFields;\n  outputKeys: string[];\n  defaultOutputKey?: string;\n}\n\n/**\n * Class to parse the output of an LLM call into a dictionary.\n * @augments BaseOutputParser\n */\nexport class RegexParser extends BaseOutputParser<Record<string, string>> {\n  static lc_name() {\n    return \"RegexParser\";\n  }\n\n  lc_namespace = [\"langchain\", \"output_parsers\", \"regex\"];\n\n  lc_serializable = true;\n\n  get lc_attributes(): SerializedFields | undefined {\n    return {\n      regex: this.lc_kwargs.regex,\n    };\n  }\n\n  regex: string | RegExp;\n\n  outputKeys: string[];\n\n  defaultOutputKey?: string;\n\n  constructor(fields: RegexParserFields);\n\n  constructor(\n    regex: string | RegExp,\n    outputKeys: string[],\n    defaultOutputKey?: string\n  );\n\n  constructor(\n    fields: string | RegExp | RegexParserFields,\n    outputKeys?: string[],\n    defaultOutputKey?: string\n  ) {\n    // oxlint-disable-next-line no-instanceof/no-instanceof\n    if (typeof fields === \"string\" || fields instanceof RegExp) {\n      // oxlint-disable-next-line no-param-reassign\n      fields = { regex: fields, outputKeys: outputKeys!, defaultOutputKey };\n    }\n    // oxlint-disable-next-line no-instanceof/no-instanceof\n    if (fields.regex instanceof RegExp) {\n      fields.regex = {\n        pattern: fields.regex.source,\n        flags: fields.regex.flags,\n      };\n    }\n    super(fields);\n    this.regex =\n      typeof fields.regex === \"string\"\n        ? new RegExp(fields.regex)\n        : \"pattern\" in fields.regex\n          ? new RegExp(fields.regex.pattern, fields.regex.flags)\n          : fields.regex;\n    this.outputKeys = fields.outputKeys;\n    this.defaultOutputKey = fields.defaultOutputKey;\n  }\n\n  _type() {\n    return \"regex_parser\";\n  }\n\n  /**\n   * Parses the given text using the regex pattern and returns a dictionary\n   * with the parsed output. If the regex pattern does not match the text\n   * and no defaultOutputKey is provided, throws an OutputParserException.\n   * @param text The text to be parsed.\n   * @returns A dictionary with the parsed output.\n   */\n  async parse(text: string): Promise<Record<string, string>> {\n    const match = text.match(this.regex);\n    if (match) {\n      return this.outputKeys.reduce(\n        (acc, key, index) => {\n          acc[key] = match[index + 1];\n          return acc;\n        },\n        {} as Record<string, string>\n      );\n    }\n\n    if (this.defaultOutputKey === undefined) {\n      throw new OutputParserException(`Could not parse output: ${text}`, text);\n    }\n\n    return this.outputKeys.reduce(\n      (acc, key) => {\n        acc[key] = key === this.defaultOutputKey ? text : \"\";\n        return acc;\n      },\n      {} as Record<string, string>\n    );\n  }\n\n  /**\n   * Returns a string with instructions on how the LLM output should be\n   * formatted to match the regex pattern.\n   * @returns A string with formatting instructions.\n   */\n  getFormatInstructions(): string {\n    return `Your response should match the following regex: ${this.regex}`;\n  }\n}\n"],"mappings":";;;;;;AAwBA,IAAa,cAAb,cAAiC,iBAAyC;CACxE,OAAO,UAAU;AACf,SAAO;;CAGT,eAAe;EAAC;EAAa;EAAkB;EAAQ;CAEvD,kBAAkB;CAElB,IAAI,gBAA8C;AAChD,SAAO,EACL,OAAO,KAAK,UAAU,OACvB;;CAGH;CAEA;CAEA;CAUA,YACE,QACA,YACA,kBACA;AAEA,MAAI,OAAO,WAAW,YAAY,kBAAkB,OAElD,UAAS;GAAE,OAAO;GAAoB;GAAa;GAAkB;AAGvE,MAAI,OAAO,iBAAiB,OAC1B,QAAO,QAAQ;GACb,SAAS,OAAO,MAAM;GACtB,OAAO,OAAO,MAAM;GACrB;AAEH,QAAM,OAAO;AACb,OAAK,QACH,OAAO,OAAO,UAAU,WACpB,IAAI,OAAO,OAAO,MAAM,GACxB,aAAa,OAAO,QAClB,IAAI,OAAO,OAAO,MAAM,SAAS,OAAO,MAAM,MAAM,GACpD,OAAO;AACf,OAAK,aAAa,OAAO;AACzB,OAAK,mBAAmB,OAAO;;CAGjC,QAAQ;AACN,SAAO;;;;;;;;;CAUT,MAAM,MAAM,MAA+C;EACzD,MAAM,QAAQ,KAAK,MAAM,KAAK,MAAM;AACpC,MAAI,MACF,QAAO,KAAK,WAAW,QACpB,KAAK,KAAK,UAAU;AACnB,OAAI,OAAO,MAAM,QAAQ;AACzB,UAAO;KAET,EAAE,CACH;AAGH,MAAI,KAAK,qBAAqB,KAAA,EAC5B,OAAM,IAAI,sBAAsB,2BAA2B,QAAQ,KAAK;AAG1E,SAAO,KAAK,WAAW,QACpB,KAAK,QAAQ;AACZ,OAAI,OAAO,QAAQ,KAAK,mBAAmB,OAAO;AAClD,UAAO;KAET,EAAE,CACH;;;;;;;CAQH,wBAAgC;AAC9B,SAAO,mDAAmD,KAAK"}

package/dist/output_parsers/structured.d.cts

@@ -1,6 +1,6 @@
 import { BaseOutputParser, FormatInstructionsOptions } from "@langchain/core/output_parsers";
 import { InferInteropZodOutput, InteropZodType } from "@langchain/core/utils/types";
-import * as _langchain_core_load_serializable0 from "@langchain/core/load/serializable";
+import * as _$_langchain_core_load_serializable0 from "@langchain/core/load/serializable";
 import { z } from "zod/v3";
 
 //#region src/output_parsers/structured.d.ts
@@ -14,7 +14,7 @@ declare class StructuredOutputParser<T extends InteropZodType> extends BaseOutpu
   schema: T;
   static lc_name(): string;
   lc_namespace: string[];
-  toJSON(): _langchain_core_load_serializable0.SerializedNotImplemented;
+  toJSON(): _$_langchain_core_load_serializable0.SerializedNotImplemented;
   constructor(schema: T);
   /**
    * Creates a new StructuredOutputParser from a Zod schema.

package/dist/output_parsers/structured.d.cts.map

@@ -1 +1 @@
-{"version":3,"file":"structured.d.cts","names":[],"sources":["../../src/output_parsers/structured.ts"],"mappings":";;;;;;KAqBY,uCAAA;EACV,kBAAA;AAAA;AAAA,UAGe,qCAAA,SAA8C,yBAAA;EAC7D,kBAAA;AAAA;AAAA,cAGW,sBAAA,WACD,cAAA,UACF,gBAAA,CAAiB,qBAAA,CAAsB,CAAA;EAW5B,MAAA,EAAQ,CAAA;EAAA,OAVpB,OAAA,CAAA;EAIP,YAAA;EAEA,MAAA,CAAA,GAI4B,kCAAA,CAJtB,wBAAA;EAIN,WAAA,CAAmB,MAAA,EAAQ,CAAA;EAhBT;AAGpB;;;;EAHoB,OAyBX,aAAA,WAAwB,cAAA,CAAA,CAAgB,MAAA,EAAQ,CAAA,GAAC,sBAAA,CAAA,CAAA;EApB/B;;;;;;EAAA,OA8BlB,wBAAA;IAAA,CAAsC,GAAA;EAAA,EAAA,CAC3C,OAAA,EAAS,CAAA,GAAC,sBAAA,CAAA,CAAA,CAAA,SAAA;IAAA;;;;;;EA2Ce;;;;;;EAvB3B,qBAAA,CAAA;EAnDyB;;;;;EA0EnB,KAAA,CAAM,IAAA,WAAe,OAAA,CAAQ,qBAAA,CAAsB,CAAA;AAAA;;;;;cAuB9C,kCAAA,WACD,cAAA,UACF,sBAAA,CAAuB,CAAA;EAAA,OACxB,OAAA,CAAA;EAIP,qBAAA,CACE,OAAA,GAAU,qCAAA;EAAA,QAcJ,oBAAA;EAAA,OAmED,aAAA,WAAwB,cAAA,CAAA,CAAgB,MAAA,EAAQ,CAAA,GAAC,kCAAA,CAAA,CAAA;EAAA,OAIjD,wBAAA;IAAA,CAAsC,GAAA;EAAA,EAAA,CAC3C,OAAA,EAAS,CAAA,GAAC,kCAAA,CAAA,CAAA,CAAA,SAAA;IAAA;;;;;;;UAeG,sCAAA,WACL,cAAA;EAEV,WAAA,EAAa,CAAA;AAAA;;;;;uBAOO,gCAAA,WACV,cAAA,uBAEF,gBAAA,CAAiB,CAAA;EAAA,QACjB,qBAAA;EAER,WAAA,CAAA;IAAc;EAAA,GAAe,sCAAA,CAAuC,CAAA;EApJzC;;;;;AAuB7B;EAvB6B,SAiKlB,eAAA,CAAgB,KAAA,EAAO,qBAAA,CAAsB,CAAA,IAAK,OAAA,CAAQ,CAAA;EAE7D,KAAA,CAAM,IAAA,WAAe,OAAA,CAAQ,CAAA;EAcnC,qBAAA,CAAA;AAAA"}
+{"version":3,"file":"structured.d.cts","names":[],"sources":["../../src/output_parsers/structured.ts"],"mappings":";;;;;;KAqBY,uCAAA;EACV,kBAAA;AAAA;AAAA,UAGe,qCAAA,SAA8C,yBAAA;EAC7D,kBAAA;AAAA;AAAA,cAGW,sBAAA,WACD,cAAA,UACF,gBAAA,CAAiB,qBAAA,CAAsB,CAAA;EAW5B,MAAA,EAAQ,CAAA;EAAA,OAVpB,OAAA,CAAA;EAIP,YAAA;EAEA,MAAA,CAAA,GAI4B,oCAAA,CAJtB,wBAAA;EAIN,WAAA,CAAmB,MAAA,EAAQ,CAAA;EAhBT;AAGpB;;;;EAHoB,OAyBX,aAAA,WAAwB,cAAA,CAAA,CAAgB,MAAA,EAAQ,CAAA,GAAC,sBAAA,CAAA,CAAA;EApB/B;;;;;;EAAA,OA8BlB,wBAAA;IAAA,CAAsC,GAAA;EAAA,EAAA,CAC3C,OAAA,EAAS,CAAA,GAAC,sBAAA,CAAA,CAAA,CAAA,SAAA;IAAA;;;;;;EA2Ce;;;;;;EAvB3B,qBAAA,CAAA;EAnDyB;;;;;EA0EnB,KAAA,CAAM,IAAA,WAAe,OAAA,CAAQ,qBAAA,CAAsB,CAAA;AAAA;;;;;cAuB9C,kCAAA,WACD,cAAA,UACF,sBAAA,CAAuB,CAAA;EAAA,OACxB,OAAA,CAAA;EAIP,qBAAA,CACE,OAAA,GAAU,qCAAA;EAAA,QAcJ,oBAAA;EAAA,OAmED,aAAA,WAAwB,cAAA,CAAA,CAAgB,MAAA,EAAQ,CAAA,GAAC,kCAAA,CAAA,CAAA;EAAA,OAIjD,wBAAA;IAAA,CAAsC,GAAA;EAAA,EAAA,CAC3C,OAAA,EAAS,CAAA,GAAC,kCAAA,CAAA,CAAA,CAAA,SAAA;IAAA;;;;;;;UAeG,sCAAA,WACL,cAAA;EAEV,WAAA,EAAa,CAAA;AAAA;;;;;uBAOO,gCAAA,WACV,cAAA,uBAEF,gBAAA,CAAiB,CAAA;EAAA,QACjB,qBAAA;EAER,WAAA,CAAA;IAAc;EAAA,GAAe,sCAAA,CAAuC,CAAA;EApJzC;;;;;AAuB7B;EAvB6B,SAiKlB,eAAA,CAAgB,KAAA,EAAO,qBAAA,CAAsB,CAAA,IAAK,OAAA,CAAQ,CAAA;EAE7D,KAAA,CAAM,IAAA,WAAe,OAAA,CAAQ,CAAA;EAcnC,qBAAA,CAAA;AAAA"}

package/dist/output_parsers/structured.d.ts

@@ -1,4 +1,4 @@
-import * as _langchain_core_load_serializable0 from "@langchain/core/load/serializable";
+import * as _$_langchain_core_load_serializable0 from "@langchain/core/load/serializable";
 import { BaseOutputParser, FormatInstructionsOptions } from "@langchain/core/output_parsers";
 import { InferInteropZodOutput, InteropZodType } from "@langchain/core/utils/types";
 import { z } from "zod/v3";
@@ -14,7 +14,7 @@ declare class StructuredOutputParser<T extends InteropZodType> extends BaseOutpu
   schema: T;
   static lc_name(): string;
   lc_namespace: string[];
-  toJSON(): _langchain_core_load_serializable0.SerializedNotImplemented;
+  toJSON(): _$_langchain_core_load_serializable0.SerializedNotImplemented;
   constructor(schema: T);
   /**
    * Creates a new StructuredOutputParser from a Zod schema.

package/dist/output_parsers/structured.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"structured.d.ts","names":[],"sources":["../../src/output_parsers/structured.ts"],"mappings":";;;;;;KAqBY,uCAAA;EACV,kBAAA;AAAA;AAAA,UAGe,qCAAA,SAA8C,yBAAA;EAC7D,kBAAA;AAAA;AAAA,cAGW,sBAAA,WACD,cAAA,UACF,gBAAA,CAAiB,qBAAA,CAAsB,CAAA;EAW5B,MAAA,EAAQ,CAAA;EAAA,OAVpB,OAAA,CAAA;EAIP,YAAA;EAEA,MAAA,CAAA,GAI4B,kCAAA,CAJtB,wBAAA;EAIN,WAAA,CAAmB,MAAA,EAAQ,CAAA;EAhBT;AAGpB;;;;EAHoB,OAyBX,aAAA,WAAwB,cAAA,CAAA,CAAgB,MAAA,EAAQ,CAAA,GAAC,sBAAA,CAAA,CAAA;EApB/B;;;;;;EAAA,OA8BlB,wBAAA;IAAA,CAAsC,GAAA;EAAA,EAAA,CAC3C,OAAA,EAAS,CAAA,GAAC,sBAAA,CAAA,CAAA,CAAA,SAAA;IAAA;;;;;;EA2Ce;;;;;;EAvB3B,qBAAA,CAAA;EAnDyB;;;;;EA0EnB,KAAA,CAAM,IAAA,WAAe,OAAA,CAAQ,qBAAA,CAAsB,CAAA;AAAA;;;;;cAuB9C,kCAAA,WACD,cAAA,UACF,sBAAA,CAAuB,CAAA;EAAA,OACxB,OAAA,CAAA;EAIP,qBAAA,CACE,OAAA,GAAU,qCAAA;EAAA,QAcJ,oBAAA;EAAA,OAmED,aAAA,WAAwB,cAAA,CAAA,CAAgB,MAAA,EAAQ,CAAA,GAAC,kCAAA,CAAA,CAAA;EAAA,OAIjD,wBAAA;IAAA,CAAsC,GAAA;EAAA,EAAA,CAC3C,OAAA,EAAS,CAAA,GAAC,kCAAA,CAAA,CAAA,CAAA,SAAA;IAAA;;;;;;;UAeG,sCAAA,WACL,cAAA;EAEV,WAAA,EAAa,CAAA;AAAA;;;;;uBAOO,gCAAA,WACV,cAAA,uBAEF,gBAAA,CAAiB,CAAA;EAAA,QACjB,qBAAA;EAER,WAAA,CAAA;IAAc;EAAA,GAAe,sCAAA,CAAuC,CAAA;EApJzC;;;;;AAuB7B;EAvB6B,SAiKlB,eAAA,CAAgB,KAAA,EAAO,qBAAA,CAAsB,CAAA,IAAK,OAAA,CAAQ,CAAA;EAE7D,KAAA,CAAM,IAAA,WAAe,OAAA,CAAQ,CAAA;EAcnC,qBAAA,CAAA;AAAA"}
+{"version":3,"file":"structured.d.ts","names":[],"sources":["../../src/output_parsers/structured.ts"],"mappings":";;;;;;KAqBY,uCAAA;EACV,kBAAA;AAAA;AAAA,UAGe,qCAAA,SAA8C,yBAAA;EAC7D,kBAAA;AAAA;AAAA,cAGW,sBAAA,WACD,cAAA,UACF,gBAAA,CAAiB,qBAAA,CAAsB,CAAA;EAW5B,MAAA,EAAQ,CAAA;EAAA,OAVpB,OAAA,CAAA;EAIP,YAAA;EAEA,MAAA,CAAA,GAI4B,oCAAA,CAJtB,wBAAA;EAIN,WAAA,CAAmB,MAAA,EAAQ,CAAA;EAhBT;AAGpB;;;;EAHoB,OAyBX,aAAA,WAAwB,cAAA,CAAA,CAAgB,MAAA,EAAQ,CAAA,GAAC,sBAAA,CAAA,CAAA;EApB/B;;;;;;EAAA,OA8BlB,wBAAA;IAAA,CAAsC,GAAA;EAAA,EAAA,CAC3C,OAAA,EAAS,CAAA,GAAC,sBAAA,CAAA,CAAA,CAAA,SAAA;IAAA;;;;;;EA2Ce;;;;;;EAvB3B,qBAAA,CAAA;EAnDyB;;;;;EA0EnB,KAAA,CAAM,IAAA,WAAe,OAAA,CAAQ,qBAAA,CAAsB,CAAA;AAAA;;;;;cAuB9C,kCAAA,WACD,cAAA,UACF,sBAAA,CAAuB,CAAA;EAAA,OACxB,OAAA,CAAA;EAIP,qBAAA,CACE,OAAA,GAAU,qCAAA;EAAA,QAcJ,oBAAA;EAAA,OAmED,aAAA,WAAwB,cAAA,CAAA,CAAgB,MAAA,EAAQ,CAAA,GAAC,kCAAA,CAAA,CAAA;EAAA,OAIjD,wBAAA;IAAA,CAAsC,GAAA;EAAA,EAAA,CAC3C,OAAA,EAAS,CAAA,GAAC,kCAAA,CAAA,CAAA,CAAA,SAAA;IAAA;;;;;;;UAeG,sCAAA,WACL,cAAA;EAEV,WAAA,EAAa,CAAA;AAAA;;;;;uBAOO,gCAAA,WACV,cAAA,uBAEF,gBAAA,CAAiB,CAAA;EAAA,QACjB,qBAAA;EAER,WAAA,CAAA;IAAc;EAAA,GAAe,sCAAA,CAAuC,CAAA;EApJzC;;;;;AAuB7B;EAvB6B,SAiKlB,eAAA,CAAgB,KAAA,EAAO,qBAAA,CAAsB,CAAA,IAAK,OAAA,CAAQ,CAAA;EAE7D,KAAA,CAAM,IAAA,WAAe,OAAA,CAAQ,CAAA;EAcnC,qBAAA,CAAA;AAAA"}

package/dist/retrievers/document_compressors/index.cjs.map

@@ -1 +1 @@
-{"version":3,"file":"index.cjs","names":[],"sources":["../../../src/retrievers/document_compressors/index.ts"],"sourcesContent":["import type { DocumentInterface } from \"@langchain/core/documents\";\nimport { BaseDocumentTransformer } from \"@langchain/core/documents\";\nimport { Callbacks } from \"@langchain/core/callbacks/manager\";\n\n/**\n * Base Document Compression class. All compressors should extend this class.\n */\nexport abstract class BaseDocumentCompressor {\n  /**\n   * Abstract method that must be implemented by any class that extends\n   * `BaseDocumentCompressor`. This method takes an array of `Document`\n   * objects and a query string as parameters and returns a Promise that\n   * resolves with an array of compressed `Document` objects.\n   * @param documents An array of `Document` objects to be compressed.\n   * @param query A query string.\n   * @returns A Promise that resolves with an array of compressed `Document` objects.\n   */\n  abstract compressDocuments(\n    documents: DocumentInterface[],\n    query: string,\n    callbacks?: Callbacks\n  ): Promise<DocumentInterface[]>;\n\n  // eslint-disable-next-line @typescript-eslint/no-explicit-any\n  static isBaseDocumentCompressor(x: any): x is BaseDocumentCompressor {\n    return x?.compressDocuments !== undefined;\n  }\n}\n\n/**\n * Document compressor that uses a pipeline of Transformers.\n * @example\n * ```typescript\n * const compressorPipeline = new DocumentCompressorPipeline({\n *   transformers: [\n *     new RecursiveCharacterTextSplitter({\n *       chunkSize: 200,\n *       chunkOverlap: 0,\n *     }),\n *     new EmbeddingsFilter({\n *       embeddings: new OpenAIEmbeddings(),\n *       similarityThreshold: 0.8,\n *       k: 5,\n *     }),\n *   ],\n * });\n * const retriever = new ContextualCompressionRetriever({\n *   baseCompressor: compressorPipeline,\n *   baseRetriever: new TavilySearchAPIRetriever({\n *     includeRawContent: true,\n *   }),\n * });\n * const retrievedDocs = await retriever.invoke(\n *   \"What did the speaker say about Justice Breyer in the 2022 State of the Union?\",\n * );\n * console.log({ retrievedDocs });\n * ```\n */\nexport class DocumentCompressorPipeline extends BaseDocumentCompressor {\n  transformers: (BaseDocumentTransformer | BaseDocumentCompressor)[];\n\n  constructor(fields: {\n    transformers: (BaseDocumentTransformer | BaseDocumentCompressor)[];\n  }) {\n    super();\n    this.transformers = fields.transformers;\n  }\n\n  async compressDocuments(\n    documents: DocumentInterface[],\n    query: string,\n    callbacks?: Callbacks\n  ): Promise<DocumentInterface[]> {\n    let transformedDocuments = documents;\n    for (const transformer of this.transformers) {\n      if (BaseDocumentCompressor.isBaseDocumentCompressor(transformer)) {\n        transformedDocuments = await transformer.compressDocuments(\n          transformedDocuments,\n          query,\n          callbacks\n        );\n      } else {\n        transformedDocuments =\n          await transformer.transformDocuments(transformedDocuments);\n      }\n    }\n    return transformedDocuments;\n  }\n}\n"],"mappings":";;;;;;;;;AAOA,IAAsB,yBAAtB,MAA6C;CAiB3C,OAAO,yBAAyB,GAAqC;AACnE,SAAO,GAAG,sBAAsB,KAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAiCpC,IAAa,6BAAb,cAAgD,uBAAuB;CACrE;CAEA,YAAY,QAET;AACD,SAAO;AACP,OAAK,eAAe,OAAO;;CAG7B,MAAM,kBACJ,WACA,OACA,WAC8B;EAC9B,IAAI,uBAAuB;AAC3B,OAAK,MAAM,eAAe,KAAK,aAC7B,KAAI,uBAAuB,yBAAyB,YAAY,CAC9D,wBAAuB,MAAM,YAAY,kBACvC,sBACA,OACA,UACD;MAED,wBACE,MAAM,YAAY,mBAAmB,qBAAqB;AAGhE,SAAO"}
+{"version":3,"file":"index.cjs","names":[],"sources":["../../../src/retrievers/document_compressors/index.ts"],"sourcesContent":["import type { DocumentInterface } from \"@langchain/core/documents\";\nimport { BaseDocumentTransformer } from \"@langchain/core/documents\";\nimport { Callbacks } from \"@langchain/core/callbacks/manager\";\n\n/**\n * Base Document Compression class. All compressors should extend this class.\n */\nexport abstract class BaseDocumentCompressor {\n  /**\n   * Abstract method that must be implemented by any class that extends\n   * `BaseDocumentCompressor`. This method takes an array of `Document`\n   * objects and a query string as parameters and returns a Promise that\n   * resolves with an array of compressed `Document` objects.\n   * @param documents An array of `Document` objects to be compressed.\n   * @param query A query string.\n   * @returns A Promise that resolves with an array of compressed `Document` objects.\n   */\n  abstract compressDocuments(\n    documents: DocumentInterface[],\n    query: string,\n    callbacks?: Callbacks\n  ): Promise<DocumentInterface[]>;\n\n  // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n  static isBaseDocumentCompressor(x: any): x is BaseDocumentCompressor {\n    return x?.compressDocuments !== undefined;\n  }\n}\n\n/**\n * Document compressor that uses a pipeline of Transformers.\n * @example\n * ```typescript\n * const compressorPipeline = new DocumentCompressorPipeline({\n *   transformers: [\n *     new RecursiveCharacterTextSplitter({\n *       chunkSize: 200,\n *       chunkOverlap: 0,\n *     }),\n *     new EmbeddingsFilter({\n *       embeddings: new OpenAIEmbeddings(),\n *       similarityThreshold: 0.8,\n *       k: 5,\n *     }),\n *   ],\n * });\n * const retriever = new ContextualCompressionRetriever({\n *   baseCompressor: compressorPipeline,\n *   baseRetriever: new TavilySearchAPIRetriever({\n *     includeRawContent: true,\n *   }),\n * });\n * const retrievedDocs = await retriever.invoke(\n *   \"What did the speaker say about Justice Breyer in the 2022 State of the Union?\",\n * );\n * console.log({ retrievedDocs });\n * ```\n */\nexport class DocumentCompressorPipeline extends BaseDocumentCompressor {\n  transformers: (BaseDocumentTransformer | BaseDocumentCompressor)[];\n\n  constructor(fields: {\n    transformers: (BaseDocumentTransformer | BaseDocumentCompressor)[];\n  }) {\n    super();\n    this.transformers = fields.transformers;\n  }\n\n  async compressDocuments(\n    documents: DocumentInterface[],\n    query: string,\n    callbacks?: Callbacks\n  ): Promise<DocumentInterface[]> {\n    let transformedDocuments = documents;\n    for (const transformer of this.transformers) {\n      if (BaseDocumentCompressor.isBaseDocumentCompressor(transformer)) {\n        transformedDocuments = await transformer.compressDocuments(\n          transformedDocuments,\n          query,\n          callbacks\n        );\n      } else {\n        transformedDocuments =\n          await transformer.transformDocuments(transformedDocuments);\n      }\n    }\n    return transformedDocuments;\n  }\n}\n"],"mappings":";;;;;;;;;AAOA,IAAsB,yBAAtB,MAA6C;CAiB3C,OAAO,yBAAyB,GAAqC;AACnE,SAAO,GAAG,sBAAsB,KAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAiCpC,IAAa,6BAAb,cAAgD,uBAAuB;CACrE;CAEA,YAAY,QAET;AACD,SAAO;AACP,OAAK,eAAe,OAAO;;CAG7B,MAAM,kBACJ,WACA,OACA,WAC8B;EAC9B,IAAI,uBAAuB;AAC3B,OAAK,MAAM,eAAe,KAAK,aAC7B,KAAI,uBAAuB,yBAAyB,YAAY,CAC9D,wBAAuB,MAAM,YAAY,kBACvC,sBACA,OACA,UACD;MAED,wBACE,MAAM,YAAY,mBAAmB,qBAAqB;AAGhE,SAAO"}

package/dist/retrievers/document_compressors/index.js.map

@@ -1 +1 @@
-{"version":3,"file":"index.js","names":[],"sources":["../../../src/retrievers/document_compressors/index.ts"],"sourcesContent":["import type { DocumentInterface } from \"@langchain/core/documents\";\nimport { BaseDocumentTransformer } from \"@langchain/core/documents\";\nimport { Callbacks } from \"@langchain/core/callbacks/manager\";\n\n/**\n * Base Document Compression class. All compressors should extend this class.\n */\nexport abstract class BaseDocumentCompressor {\n  /**\n   * Abstract method that must be implemented by any class that extends\n   * `BaseDocumentCompressor`. This method takes an array of `Document`\n   * objects and a query string as parameters and returns a Promise that\n   * resolves with an array of compressed `Document` objects.\n   * @param documents An array of `Document` objects to be compressed.\n   * @param query A query string.\n   * @returns A Promise that resolves with an array of compressed `Document` objects.\n   */\n  abstract compressDocuments(\n    documents: DocumentInterface[],\n    query: string,\n    callbacks?: Callbacks\n  ): Promise<DocumentInterface[]>;\n\n  // eslint-disable-next-line @typescript-eslint/no-explicit-any\n  static isBaseDocumentCompressor(x: any): x is BaseDocumentCompressor {\n    return x?.compressDocuments !== undefined;\n  }\n}\n\n/**\n * Document compressor that uses a pipeline of Transformers.\n * @example\n * ```typescript\n * const compressorPipeline = new DocumentCompressorPipeline({\n *   transformers: [\n *     new RecursiveCharacterTextSplitter({\n *       chunkSize: 200,\n *       chunkOverlap: 0,\n *     }),\n *     new EmbeddingsFilter({\n *       embeddings: new OpenAIEmbeddings(),\n *       similarityThreshold: 0.8,\n *       k: 5,\n *     }),\n *   ],\n * });\n * const retriever = new ContextualCompressionRetriever({\n *   baseCompressor: compressorPipeline,\n *   baseRetriever: new TavilySearchAPIRetriever({\n *     includeRawContent: true,\n *   }),\n * });\n * const retrievedDocs = await retriever.invoke(\n *   \"What did the speaker say about Justice Breyer in the 2022 State of the Union?\",\n * );\n * console.log({ retrievedDocs });\n * ```\n */\nexport class DocumentCompressorPipeline extends BaseDocumentCompressor {\n  transformers: (BaseDocumentTransformer | BaseDocumentCompressor)[];\n\n  constructor(fields: {\n    transformers: (BaseDocumentTransformer | BaseDocumentCompressor)[];\n  }) {\n    super();\n    this.transformers = fields.transformers;\n  }\n\n  async compressDocuments(\n    documents: DocumentInterface[],\n    query: string,\n    callbacks?: Callbacks\n  ): Promise<DocumentInterface[]> {\n    let transformedDocuments = documents;\n    for (const transformer of this.transformers) {\n      if (BaseDocumentCompressor.isBaseDocumentCompressor(transformer)) {\n        transformedDocuments = await transformer.compressDocuments(\n          transformedDocuments,\n          query,\n          callbacks\n        );\n      } else {\n        transformedDocuments =\n          await transformer.transformDocuments(transformedDocuments);\n      }\n    }\n    return transformedDocuments;\n  }\n}\n"],"mappings":";;;;;;;;;AAOA,IAAsB,yBAAtB,MAA6C;CAiB3C,OAAO,yBAAyB,GAAqC;AACnE,SAAO,GAAG,sBAAsB,KAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAiCpC,IAAa,6BAAb,cAAgD,uBAAuB;CACrE;CAEA,YAAY,QAET;AACD,SAAO;AACP,OAAK,eAAe,OAAO;;CAG7B,MAAM,kBACJ,WACA,OACA,WAC8B;EAC9B,IAAI,uBAAuB;AAC3B,OAAK,MAAM,eAAe,KAAK,aAC7B,KAAI,uBAAuB,yBAAyB,YAAY,CAC9D,wBAAuB,MAAM,YAAY,kBACvC,sBACA,OACA,UACD;MAED,wBACE,MAAM,YAAY,mBAAmB,qBAAqB;AAGhE,SAAO"}
+{"version":3,"file":"index.js","names":[],"sources":["../../../src/retrievers/document_compressors/index.ts"],"sourcesContent":["import type { DocumentInterface } from \"@langchain/core/documents\";\nimport { BaseDocumentTransformer } from \"@langchain/core/documents\";\nimport { Callbacks } from \"@langchain/core/callbacks/manager\";\n\n/**\n * Base Document Compression class. All compressors should extend this class.\n */\nexport abstract class BaseDocumentCompressor {\n  /**\n   * Abstract method that must be implemented by any class that extends\n   * `BaseDocumentCompressor`. This method takes an array of `Document`\n   * objects and a query string as parameters and returns a Promise that\n   * resolves with an array of compressed `Document` objects.\n   * @param documents An array of `Document` objects to be compressed.\n   * @param query A query string.\n   * @returns A Promise that resolves with an array of compressed `Document` objects.\n   */\n  abstract compressDocuments(\n    documents: DocumentInterface[],\n    query: string,\n    callbacks?: Callbacks\n  ): Promise<DocumentInterface[]>;\n\n  // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n  static isBaseDocumentCompressor(x: any): x is BaseDocumentCompressor {\n    return x?.compressDocuments !== undefined;\n  }\n}\n\n/**\n * Document compressor that uses a pipeline of Transformers.\n * @example\n * ```typescript\n * const compressorPipeline = new DocumentCompressorPipeline({\n *   transformers: [\n *     new RecursiveCharacterTextSplitter({\n *       chunkSize: 200,\n *       chunkOverlap: 0,\n *     }),\n *     new EmbeddingsFilter({\n *       embeddings: new OpenAIEmbeddings(),\n *       similarityThreshold: 0.8,\n *       k: 5,\n *     }),\n *   ],\n * });\n * const retriever = new ContextualCompressionRetriever({\n *   baseCompressor: compressorPipeline,\n *   baseRetriever: new TavilySearchAPIRetriever({\n *     includeRawContent: true,\n *   }),\n * });\n * const retrievedDocs = await retriever.invoke(\n *   \"What did the speaker say about Justice Breyer in the 2022 State of the Union?\",\n * );\n * console.log({ retrievedDocs });\n * ```\n */\nexport class DocumentCompressorPipeline extends BaseDocumentCompressor {\n  transformers: (BaseDocumentTransformer | BaseDocumentCompressor)[];\n\n  constructor(fields: {\n    transformers: (BaseDocumentTransformer | BaseDocumentCompressor)[];\n  }) {\n    super();\n    this.transformers = fields.transformers;\n  }\n\n  async compressDocuments(\n    documents: DocumentInterface[],\n    query: string,\n    callbacks?: Callbacks\n  ): Promise<DocumentInterface[]> {\n    let transformedDocuments = documents;\n    for (const transformer of this.transformers) {\n      if (BaseDocumentCompressor.isBaseDocumentCompressor(transformer)) {\n        transformedDocuments = await transformer.compressDocuments(\n          transformedDocuments,\n          query,\n          callbacks\n        );\n      } else {\n        transformedDocuments =\n          await transformer.transformDocuments(transformedDocuments);\n      }\n    }\n    return transformedDocuments;\n  }\n}\n"],"mappings":";;;;;;;;;AAOA,IAAsB,yBAAtB,MAA6C;CAiB3C,OAAO,yBAAyB,GAAqC;AACnE,SAAO,GAAG,sBAAsB,KAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAiCpC,IAAa,6BAAb,cAAgD,uBAAuB;CACrE;CAEA,YAAY,QAET;AACD,SAAO;AACP,OAAK,eAAe,OAAO;;CAG7B,MAAM,kBACJ,WACA,OACA,WAC8B;EAC9B,IAAI,uBAAuB;AAC3B,OAAK,MAAM,eAAe,KAAK,aAC7B,KAAI,uBAAuB,yBAAyB,YAAY,CAC9D,wBAAuB,MAAM,YAAY,kBACvC,sBACA,OACA,UACD;MAED,wBACE,MAAM,YAAY,mBAAmB,qBAAqB;AAGhE,SAAO"}

package/dist/retrievers/matryoshka_retriever.cjs.map

@@ -1 +1 @@
-{"version":3,"file":"matryoshka_retriever.cjs","names":["VectorStoreRetriever"],"sources":["../../src/retrievers/matryoshka_retriever.ts"],"sourcesContent":["import { DocumentInterface } from \"@langchain/core/documents\";\nimport { Embeddings } from \"@langchain/core/embeddings\";\nimport {\n  cosineSimilarity,\n  euclideanDistance,\n  innerProduct,\n} from \"@langchain/core/utils/math\";\nimport {\n  VectorStore,\n  VectorStoreRetriever,\n  VectorStoreRetrieverInput,\n} from \"@langchain/core/vectorstores\";\n\n/**\n * Type for options when adding a document to the VectorStore.\n */\n// eslint-disable-next-line @typescript-eslint/no-explicit-any\ntype AddDocumentOptions = Record<string, any>;\n\nexport interface MatryoshkaRetrieverFields {\n  /**\n   * The number of documents to retrieve from the small store.\n   * @default 50\n   */\n  smallK?: number;\n  /**\n   * The number of documents to retrieve from the large store.\n   * @default 8\n   */\n  largeK?: number;\n  /**\n   * The metadata key to store the larger embeddings.\n   * @default \"lc_large_embedding\"\n   */\n  largeEmbeddingKey?: string;\n  /**\n   * The embedding model to use when generating the large\n   * embeddings.\n   */\n  largeEmbeddingModel: Embeddings;\n  /**\n   * The type of search to perform using the large embeddings.\n   * @default \"cosine\"\n   */\n  searchType?: \"cosine\" | \"innerProduct\" | \"euclidean\";\n}\n\n/**\n * A retriever that uses two sets of embeddings to perform adaptive retrieval. Based\n * off of the \"Matryoshka embeddings: faster OpenAI vector search using Adaptive Retrieval\"\n * blog post {@link https://supabase.com/blog/matryoshka-embeddings}.\n *\n *\n * This class performs \"Adaptive Retrieval\" for searching text embeddings efficiently using the\n * Matryoshka Representation Learning (MRL) technique. It retrieves documents similar to a query\n * embedding in two steps:\n *\n * First-pass: Uses a lower dimensional sub-vector from the MRL embedding for an initial, fast,\n * but less accurate search.\n *\n * Second-pass: Re-ranks the top results from the first pass using the full, high-dimensional\n * embedding for higher accuracy.\n *\n *\n * This code implements MRL embeddings for efficient vector search by combining faster,\n * lower-dimensional initial search with accurate, high-dimensional re-ranking.\n */\nexport class MatryoshkaRetriever<\n  Store extends VectorStore = VectorStore,\n> extends VectorStoreRetriever<Store> {\n  smallK = 50;\n\n  largeK = 8;\n\n  largeEmbeddingKey = \"lc_large_embedding\";\n\n  largeEmbeddingModel: Embeddings;\n\n  searchType: \"cosine\" | \"innerProduct\" | \"euclidean\" = \"cosine\";\n\n  constructor(\n    fields: MatryoshkaRetrieverFields & VectorStoreRetrieverInput<Store>\n  ) {\n    super(fields);\n    this.smallK = fields.smallK ?? this.smallK;\n    this.largeK = fields.largeK ?? this.largeK;\n    this.largeEmbeddingKey = fields.largeEmbeddingKey ?? this.largeEmbeddingKey;\n    this.largeEmbeddingModel = fields.largeEmbeddingModel;\n    this.searchType = fields.searchType ?? this.searchType;\n  }\n\n  /**\n   * Ranks documents based on their similarity to a query embedding using larger embeddings.\n   *\n   * This method takes a query embedding and a list of documents (smallResults) as input. Each document\n   * in the smallResults array has previously been associated with a large embedding stored in its metadata.\n   * Depending on the `searchType` (cosine, innerProduct, or euclidean), it calculates the similarity scores\n   * between the query embedding and each document's large embedding. It then ranks the documents based on\n   * these similarity scores, from the most similar to the least similar.\n   *\n   * The method returns a promise that resolves to an array of the top `largeK` documents, where `largeK`\n   * is a class property defining the number of documents to return. This subset of documents is determined\n   * by sorting the entire list of documents based on their similarity scores and then selecting the top\n   * `largeK` documents.\n   *\n   * @param {number[]} embeddedQuery The embedding of the query, represented as an array of numbers.\n   * @param {DocumentInterface[]} smallResults An array of documents, each with metadata that includes a large embedding for similarity comparison.\n   * @returns {Promise<DocumentInterface[]>} A promise that resolves to an array of the top `largeK` ranked documents based on their similarity to the query embedding.\n   */\n  private _rankByLargeEmbeddings(\n    embeddedQuery: number[],\n    smallResults: DocumentInterface[]\n  ): DocumentInterface[] {\n    const largeEmbeddings: Array<number[]> = smallResults.map((doc) =>\n      JSON.parse(doc.metadata[this.largeEmbeddingKey])\n    );\n    let func: () => Array<number[]>;\n\n    switch (this.searchType) {\n      case \"cosine\":\n        func = () => cosineSimilarity([embeddedQuery], largeEmbeddings);\n        break;\n      case \"innerProduct\":\n        func = () => innerProduct([embeddedQuery], largeEmbeddings);\n        break;\n      case \"euclidean\":\n        func = () => euclideanDistance([embeddedQuery], largeEmbeddings);\n        break;\n      default:\n        throw new Error(`Unknown search type: ${this.searchType}`);\n    }\n\n    // Calculate the similarity scores between the query embedding and the large embeddings\n    const [similarityScores] = func();\n\n    // Create an array of indices from 0 to N-1, where N is the number of documents\n    let indices = Array.from(\n      { length: smallResults.length },\n      (_, index) => index\n    );\n\n    indices = indices\n      .map((v, i) => [similarityScores[i], v])\n      .sort(([a], [b]) => b - a)\n      .slice(0, this.largeK)\n      .map(([, i]) => i);\n\n    return indices.map((i) => smallResults[i]);\n  }\n\n  async _getRelevantDocuments(query: string): Promise<DocumentInterface[]> {\n    const [embeddedQuery, smallResults] = await Promise.all([\n      this.largeEmbeddingModel.embedQuery(query),\n      this.vectorStore.similaritySearch(query, this.smallK, this.filter),\n    ]);\n\n    return this._rankByLargeEmbeddings(embeddedQuery, smallResults);\n  }\n\n  /**\n   * Override the default `addDocuments` method to embed the documents twice,\n   * once using the larger embeddings model, and then again using the default\n   * embedding model linked to the vector store.\n   *\n   * @param {DocumentInterface[]} documents - An array of documents to add to the vector store.\n   * @param {AddDocumentOptions} options - An optional object containing additional options for adding documents.\n   * @returns {Promise<string[] | void>} A promise that resolves to an array of the document IDs that were added to the vector store.\n   */\n  override addDocuments = async (\n    documents: DocumentInterface[],\n    options?: AddDocumentOptions\n  ): Promise<string[] | void> => {\n    // Insure documents metadata does not contain the large embedding key\n    if (documents.some((doc) => this.largeEmbeddingKey in doc.metadata)) {\n      throw new Error(\n        `All documents must not contain the large embedding key: ${this.largeEmbeddingKey} in their metadata.`\n      );\n    }\n\n    const allDocPageContent = documents.map((doc) => doc.pageContent);\n    const allDocLargeEmbeddings =\n      await this.largeEmbeddingModel.embedDocuments(allDocPageContent);\n\n    const newDocuments: Array<DocumentInterface> = documents.map(\n      (doc, idx) => ({\n        ...doc,\n        metadata: {\n          ...doc.metadata,\n          [this.largeEmbeddingKey]: JSON.stringify(allDocLargeEmbeddings[idx]),\n        },\n      })\n    );\n\n    return this.vectorStore.addDocuments(newDocuments, options);\n  };\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;AAmEA,IAAa,sBAAb,cAEUA,6BAAAA,qBAA4B;CACpC,SAAS;CAET,SAAS;CAET,oBAAoB;CAEpB;CAEA,aAAsD;CAEtD,YACE,QACA;AACA,QAAM,OAAO;AACb,OAAK,SAAS,OAAO,UAAU,KAAK;AACpC,OAAK,SAAS,OAAO,UAAU,KAAK;AACpC,OAAK,oBAAoB,OAAO,qBAAqB,KAAK;AAC1D,OAAK,sBAAsB,OAAO;AAClC,OAAK,aAAa,OAAO,cAAc,KAAK;;;;;;;;;;;;;;;;;;;;CAqB9C,uBACE,eACA,cACqB;EACrB,MAAM,kBAAmC,aAAa,KAAK,QACzD,KAAK,MAAM,IAAI,SAAS,KAAK,mBAAmB,CACjD;EACD,IAAI;AAEJ,UAAQ,KAAK,YAAb;GACE,KAAK;AACH,kBAAA,GAAA,2BAAA,kBAA8B,CAAC,cAAc,EAAE,gBAAgB;AAC/D;GACF,KAAK;AACH,kBAAA,GAAA,2BAAA,cAA0B,CAAC,cAAc,EAAE,gBAAgB;AAC3D;GACF,KAAK;AACH,kBAAA,GAAA,2BAAA,mBAA+B,CAAC,cAAc,EAAE,gBAAgB;AAChE;GACF,QACE,OAAM,IAAI,MAAM,wBAAwB,KAAK,aAAa;;EAI9D,MAAM,CAAC,oBAAoB,MAAM;EAGjC,IAAI,UAAU,MAAM,KAClB,EAAE,QAAQ,aAAa,QAAQ,GAC9B,GAAG,UAAU,MACf;AAED,YAAU,QACP,KAAK,GAAG,MAAM,CAAC,iBAAiB,IAAI,EAAE,CAAC,CACvC,MAAM,CAAC,IAAI,CAAC,OAAO,IAAI,EAAE,CACzB,MAAM,GAAG,KAAK,OAAO,CACrB,KAAK,GAAG,OAAO,EAAE;AAEpB,SAAO,QAAQ,KAAK,MAAM,aAAa,GAAG;;CAG5C,MAAM,sBAAsB,OAA6C;EACvE,MAAM,CAAC,eAAe,gBAAgB,MAAM,QAAQ,IAAI,CACtD,KAAK,oBAAoB,WAAW,MAAM,EAC1C,KAAK,YAAY,iBAAiB,OAAO,KAAK,QAAQ,KAAK,OAAO,CACnE,CAAC;AAEF,SAAO,KAAK,uBAAuB,eAAe,aAAa;;;;;;;;;;;CAYjE,eAAwB,OACtB,WACA,YAC6B;AAE7B,MAAI,UAAU,MAAM,QAAQ,KAAK,qBAAqB,IAAI,SAAS,CACjE,OAAM,IAAI,MACR,2DAA2D,KAAK,kBAAkB,qBACnF;EAGH,MAAM,oBAAoB,UAAU,KAAK,QAAQ,IAAI,YAAY;EACjE,MAAM,wBACJ,MAAM,KAAK,oBAAoB,eAAe,kBAAkB;EAElE,MAAM,eAAyC,UAAU,KACtD,KAAK,SAAS;GACb,GAAG;GACH,UAAU;IACR,GAAG,IAAI;KACN,KAAK,oBAAoB,KAAK,UAAU,sBAAsB,KAAK;IACrE;GACF,EACF;AAED,SAAO,KAAK,YAAY,aAAa,cAAc,QAAQ"}
+{"version":3,"file":"matryoshka_retriever.cjs","names":["VectorStoreRetriever"],"sources":["../../src/retrievers/matryoshka_retriever.ts"],"sourcesContent":["import { DocumentInterface } from \"@langchain/core/documents\";\nimport { Embeddings } from \"@langchain/core/embeddings\";\nimport {\n  cosineSimilarity,\n  euclideanDistance,\n  innerProduct,\n} from \"@langchain/core/utils/math\";\nimport {\n  VectorStore,\n  VectorStoreRetriever,\n  VectorStoreRetrieverInput,\n} from \"@langchain/core/vectorstores\";\n\n/**\n * Type for options when adding a document to the VectorStore.\n */\n// oxlint-disable-next-line @typescript-eslint/no-explicit-any\ntype AddDocumentOptions = Record<string, any>;\n\nexport interface MatryoshkaRetrieverFields {\n  /**\n   * The number of documents to retrieve from the small store.\n   * @default 50\n   */\n  smallK?: number;\n  /**\n   * The number of documents to retrieve from the large store.\n   * @default 8\n   */\n  largeK?: number;\n  /**\n   * The metadata key to store the larger embeddings.\n   * @default \"lc_large_embedding\"\n   */\n  largeEmbeddingKey?: string;\n  /**\n   * The embedding model to use when generating the large\n   * embeddings.\n   */\n  largeEmbeddingModel: Embeddings;\n  /**\n   * The type of search to perform using the large embeddings.\n   * @default \"cosine\"\n   */\n  searchType?: \"cosine\" | \"innerProduct\" | \"euclidean\";\n}\n\n/**\n * A retriever that uses two sets of embeddings to perform adaptive retrieval. Based\n * off of the \"Matryoshka embeddings: faster OpenAI vector search using Adaptive Retrieval\"\n * blog post {@link https://supabase.com/blog/matryoshka-embeddings}.\n *\n *\n * This class performs \"Adaptive Retrieval\" for searching text embeddings efficiently using the\n * Matryoshka Representation Learning (MRL) technique. It retrieves documents similar to a query\n * embedding in two steps:\n *\n * First-pass: Uses a lower dimensional sub-vector from the MRL embedding for an initial, fast,\n * but less accurate search.\n *\n * Second-pass: Re-ranks the top results from the first pass using the full, high-dimensional\n * embedding for higher accuracy.\n *\n *\n * This code implements MRL embeddings for efficient vector search by combining faster,\n * lower-dimensional initial search with accurate, high-dimensional re-ranking.\n */\nexport class MatryoshkaRetriever<\n  Store extends VectorStore = VectorStore,\n> extends VectorStoreRetriever<Store> {\n  smallK = 50;\n\n  largeK = 8;\n\n  largeEmbeddingKey = \"lc_large_embedding\";\n\n  largeEmbeddingModel: Embeddings;\n\n  searchType: \"cosine\" | \"innerProduct\" | \"euclidean\" = \"cosine\";\n\n  constructor(\n    fields: MatryoshkaRetrieverFields & VectorStoreRetrieverInput<Store>\n  ) {\n    super(fields);\n    this.smallK = fields.smallK ?? this.smallK;\n    this.largeK = fields.largeK ?? this.largeK;\n    this.largeEmbeddingKey = fields.largeEmbeddingKey ?? this.largeEmbeddingKey;\n    this.largeEmbeddingModel = fields.largeEmbeddingModel;\n    this.searchType = fields.searchType ?? this.searchType;\n  }\n\n  /**\n   * Ranks documents based on their similarity to a query embedding using larger embeddings.\n   *\n   * This method takes a query embedding and a list of documents (smallResults) as input. Each document\n   * in the smallResults array has previously been associated with a large embedding stored in its metadata.\n   * Depending on the `searchType` (cosine, innerProduct, or euclidean), it calculates the similarity scores\n   * between the query embedding and each document's large embedding. It then ranks the documents based on\n   * these similarity scores, from the most similar to the least similar.\n   *\n   * The method returns a promise that resolves to an array of the top `largeK` documents, where `largeK`\n   * is a class property defining the number of documents to return. This subset of documents is determined\n   * by sorting the entire list of documents based on their similarity scores and then selecting the top\n   * `largeK` documents.\n   *\n   * @param {number[]} embeddedQuery The embedding of the query, represented as an array of numbers.\n   * @param {DocumentInterface[]} smallResults An array of documents, each with metadata that includes a large embedding for similarity comparison.\n   * @returns {Promise<DocumentInterface[]>} A promise that resolves to an array of the top `largeK` ranked documents based on their similarity to the query embedding.\n   */\n  private _rankByLargeEmbeddings(\n    embeddedQuery: number[],\n    smallResults: DocumentInterface[]\n  ): DocumentInterface[] {\n    const largeEmbeddings: Array<number[]> = smallResults.map((doc) =>\n      JSON.parse(doc.metadata[this.largeEmbeddingKey])\n    );\n    let func: () => Array<number[]>;\n\n    switch (this.searchType) {\n      case \"cosine\":\n        func = () => cosineSimilarity([embeddedQuery], largeEmbeddings);\n        break;\n      case \"innerProduct\":\n        func = () => innerProduct([embeddedQuery], largeEmbeddings);\n        break;\n      case \"euclidean\":\n        func = () => euclideanDistance([embeddedQuery], largeEmbeddings);\n        break;\n      default:\n        throw new Error(`Unknown search type: ${this.searchType}`);\n    }\n\n    // Calculate the similarity scores between the query embedding and the large embeddings\n    const [similarityScores] = func();\n\n    // Create an array of indices from 0 to N-1, where N is the number of documents\n    let indices = Array.from(\n      { length: smallResults.length },\n      (_, index) => index\n    );\n\n    indices = indices\n      .map((v, i) => [similarityScores[i], v])\n      .sort(([a], [b]) => b - a)\n      .slice(0, this.largeK)\n      .map(([, i]) => i);\n\n    return indices.map((i) => smallResults[i]);\n  }\n\n  async _getRelevantDocuments(query: string): Promise<DocumentInterface[]> {\n    const [embeddedQuery, smallResults] = await Promise.all([\n      this.largeEmbeddingModel.embedQuery(query),\n      this.vectorStore.similaritySearch(query, this.smallK, this.filter),\n    ]);\n\n    return this._rankByLargeEmbeddings(embeddedQuery, smallResults);\n  }\n\n  /**\n   * Override the default `addDocuments` method to embed the documents twice,\n   * once using the larger embeddings model, and then again using the default\n   * embedding model linked to the vector store.\n   *\n   * @param {DocumentInterface[]} documents - An array of documents to add to the vector store.\n   * @param {AddDocumentOptions} options - An optional object containing additional options for adding documents.\n   * @returns {Promise<string[] | void>} A promise that resolves to an array of the document IDs that were added to the vector store.\n   */\n  override addDocuments = async (\n    documents: DocumentInterface[],\n    options?: AddDocumentOptions\n  ): Promise<string[] | void> => {\n    // Insure documents metadata does not contain the large embedding key\n    if (documents.some((doc) => this.largeEmbeddingKey in doc.metadata)) {\n      throw new Error(\n        `All documents must not contain the large embedding key: ${this.largeEmbeddingKey} in their metadata.`\n      );\n    }\n\n    const allDocPageContent = documents.map((doc) => doc.pageContent);\n    const allDocLargeEmbeddings =\n      await this.largeEmbeddingModel.embedDocuments(allDocPageContent);\n\n    const newDocuments: Array<DocumentInterface> = documents.map(\n      (doc, idx) => ({\n        ...doc,\n        metadata: {\n          ...doc.metadata,\n          [this.largeEmbeddingKey]: JSON.stringify(allDocLargeEmbeddings[idx]),\n        },\n      })\n    );\n\n    return this.vectorStore.addDocuments(newDocuments, options);\n  };\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;AAmEA,IAAa,sBAAb,cAEUA,6BAAAA,qBAA4B;CACpC,SAAS;CAET,SAAS;CAET,oBAAoB;CAEpB;CAEA,aAAsD;CAEtD,YACE,QACA;AACA,QAAM,OAAO;AACb,OAAK,SAAS,OAAO,UAAU,KAAK;AACpC,OAAK,SAAS,OAAO,UAAU,KAAK;AACpC,OAAK,oBAAoB,OAAO,qBAAqB,KAAK;AAC1D,OAAK,sBAAsB,OAAO;AAClC,OAAK,aAAa,OAAO,cAAc,KAAK;;;;;;;;;;;;;;;;;;;;CAqB9C,uBACE,eACA,cACqB;EACrB,MAAM,kBAAmC,aAAa,KAAK,QACzD,KAAK,MAAM,IAAI,SAAS,KAAK,mBAAmB,CACjD;EACD,IAAI;AAEJ,UAAQ,KAAK,YAAb;GACE,KAAK;AACH,kBAAA,GAAA,2BAAA,kBAA8B,CAAC,cAAc,EAAE,gBAAgB;AAC/D;GACF,KAAK;AACH,kBAAA,GAAA,2BAAA,cAA0B,CAAC,cAAc,EAAE,gBAAgB;AAC3D;GACF,KAAK;AACH,kBAAA,GAAA,2BAAA,mBAA+B,CAAC,cAAc,EAAE,gBAAgB;AAChE;GACF,QACE,OAAM,IAAI,MAAM,wBAAwB,KAAK,aAAa;;EAI9D,MAAM,CAAC,oBAAoB,MAAM;EAGjC,IAAI,UAAU,MAAM,KAClB,EAAE,QAAQ,aAAa,QAAQ,GAC9B,GAAG,UAAU,MACf;AAED,YAAU,QACP,KAAK,GAAG,MAAM,CAAC,iBAAiB,IAAI,EAAE,CAAC,CACvC,MAAM,CAAC,IAAI,CAAC,OAAO,IAAI,EAAE,CACzB,MAAM,GAAG,KAAK,OAAO,CACrB,KAAK,GAAG,OAAO,EAAE;AAEpB,SAAO,QAAQ,KAAK,MAAM,aAAa,GAAG;;CAG5C,MAAM,sBAAsB,OAA6C;EACvE,MAAM,CAAC,eAAe,gBAAgB,MAAM,QAAQ,IAAI,CACtD,KAAK,oBAAoB,WAAW,MAAM,EAC1C,KAAK,YAAY,iBAAiB,OAAO,KAAK,QAAQ,KAAK,OAAO,CACnE,CAAC;AAEF,SAAO,KAAK,uBAAuB,eAAe,aAAa;;;;;;;;;;;CAYjE,eAAwB,OACtB,WACA,YAC6B;AAE7B,MAAI,UAAU,MAAM,QAAQ,KAAK,qBAAqB,IAAI,SAAS,CACjE,OAAM,IAAI,MACR,2DAA2D,KAAK,kBAAkB,qBACnF;EAGH,MAAM,oBAAoB,UAAU,KAAK,QAAQ,IAAI,YAAY;EACjE,MAAM,wBACJ,MAAM,KAAK,oBAAoB,eAAe,kBAAkB;EAElE,MAAM,eAAyC,UAAU,KACtD,KAAK,SAAS;GACb,GAAG;GACH,UAAU;IACR,GAAG,IAAI;KACN,KAAK,oBAAoB,KAAK,UAAU,sBAAsB,KAAK;IACrE;GACF,EACF;AAED,SAAO,KAAK,YAAY,aAAa,cAAc,QAAQ"}

package/dist/retrievers/matryoshka_retriever.d.cts

@@ -90,7 +90,7 @@ declare class MatryoshkaRetriever<Store extends VectorStore = VectorStore> exten
    * @param {AddDocumentOptions} options - An optional object containing additional options for adding documents.
    * @returns {Promise<string[] | void>} A promise that resolves to an array of the document IDs that were added to the vector store.
    */
-  addDocuments: (documents: DocumentInterface<Record<string, any>>[], options?: AddDocumentOptions | undefined) => Promise<void | string[]>;
+  addDocuments: (documents: DocumentInterface[], options?: AddDocumentOptions) => Promise<string[] | void>;
 }
 //#endregion
 export { MatryoshkaRetriever, MatryoshkaRetrieverFields };

package/dist/retrievers/matryoshka_retriever.d.cts.map

@@ -1 +1 @@
-{"version":3,"file":"matryoshka_retriever.d.cts","names":[],"sources":["../../src/retrievers/matryoshka_retriever.ts"],"mappings":";;;;;;;AAWsC;KAMjC,kBAAA,GAAqB,MAAA;AAAA,UAET,yBAAA;EAFS;;AAE1B;;EAKE,MAAA;EAe+B;;;;EAV/B,MAAA;EAUqB;;;;EALrB,iBAAA;EAiC8B;;;;EA5B9B,mBAAA,EAAqB,UAAA;EAqCA;;;;EAhCrB,UAAA;AAAA;;;;;;;;;;;;;;;;;;;;;cAuBW,mBAAA,eACG,WAAA,GAAc,WAAA,UACpB,oBAAA,CAAqB,KAAA;EAC7B,MAAA;EAEA,MAAA;EAEA,iBAAA;EAEA,mBAAA,EAAqB,UAAA;EAErB,UAAA;EAEA,WAAA,CACE,MAAA,EAAQ,yBAAA,GAA4B,yBAAA,CAA0B,KAAA;EAqEZ;;;;;;;;;;;;;;;;;;EAAA,QAzC5C,sBAAA;EAyCF,qBAAA,CAAsB,KAAA,WAAgB,OAAA,CAAQ,iBAAA;;;;;;;;;;EAkB3C,YAAA,GAAY,SAAA,EAAA,iBAAA,CAAA,MAAA,kBAAA,OAAA,GAAA,kBAAA,iBAAA,OAAA;AAAA"}
+{"version":3,"file":"matryoshka_retriever.d.cts","names":[],"sources":["../../src/retrievers/matryoshka_retriever.ts"],"mappings":";;;;;;;AAWsC;KAMjC,kBAAA,GAAqB,MAAA;AAAA,UAET,yBAAA;EAFS;;AAE1B;;EAKE,MAAA;EAe+B;;;;EAV/B,MAAA;EAUqB;;;;EALrB,iBAAA;EAiC8B;;;;EA5B9B,mBAAA,EAAqB,UAAA;EAqCA;;;;EAhCrB,UAAA;AAAA;;;;;;;;;;;;;;;;;;;;;cAuBW,mBAAA,eACG,WAAA,GAAc,WAAA,UACpB,oBAAA,CAAqB,KAAA;EAC7B,MAAA;EAEA,MAAA;EAEA,iBAAA;EAEA,mBAAA,EAAqB,UAAA;EAErB,UAAA;EAEA,WAAA,CACE,MAAA,EAAQ,yBAAA,GAA4B,yBAAA,CAA0B,KAAA;EAuFvD;;;;;;;;;;;;;;;;;;EAAA,QA3DD,sBAAA;EAyCF,qBAAA,CAAsB,KAAA,WAAgB,OAAA,CAAQ,iBAAA;;;;;;;;;;EAkB3C,YAAA,GAAY,SAAA,EACR,iBAAA,IAAmB,OAAA,GACpB,kBAAA,KACT,OAAA;AAAA"}

package/dist/retrievers/matryoshka_retriever.d.ts

@@ -90,7 +90,7 @@ declare class MatryoshkaRetriever<Store extends VectorStore = VectorStore> exten
    * @param {AddDocumentOptions} options - An optional object containing additional options for adding documents.
    * @returns {Promise<string[] | void>} A promise that resolves to an array of the document IDs that were added to the vector store.
    */
-  addDocuments: (documents: DocumentInterface<Record<string, any>>[], options?: AddDocumentOptions | undefined) => Promise<void | string[]>;
+  addDocuments: (documents: DocumentInterface[], options?: AddDocumentOptions) => Promise<string[] | void>;
 }
 //#endregion
 export { MatryoshkaRetriever, MatryoshkaRetrieverFields };

package/dist/retrievers/matryoshka_retriever.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"matryoshka_retriever.d.ts","names":[],"sources":["../../src/retrievers/matryoshka_retriever.ts"],"mappings":";;;;;;;AAWsC;KAMjC,kBAAA,GAAqB,MAAA;AAAA,UAET,yBAAA;EAFS;;AAE1B;;EAKE,MAAA;EAe+B;;;;EAV/B,MAAA;EAUqB;;;;EALrB,iBAAA;EAiC8B;;;;EA5B9B,mBAAA,EAAqB,UAAA;EAqCA;;;;EAhCrB,UAAA;AAAA;;;;;;;;;;;;;;;;;;;;;cAuBW,mBAAA,eACG,WAAA,GAAc,WAAA,UACpB,oBAAA,CAAqB,KAAA;EAC7B,MAAA;EAEA,MAAA;EAEA,iBAAA;EAEA,mBAAA,EAAqB,UAAA;EAErB,UAAA;EAEA,WAAA,CACE,MAAA,EAAQ,yBAAA,GAA4B,yBAAA,CAA0B,KAAA;EAqEZ;;;;;;;;;;;;;;;;;;EAAA,QAzC5C,sBAAA;EAyCF,qBAAA,CAAsB,KAAA,WAAgB,OAAA,CAAQ,iBAAA;;;;;;;;;;EAkB3C,YAAA,GAAY,SAAA,EAAA,iBAAA,CAAA,MAAA,kBAAA,OAAA,GAAA,kBAAA,iBAAA,OAAA;AAAA"}
+{"version":3,"file":"matryoshka_retriever.d.ts","names":[],"sources":["../../src/retrievers/matryoshka_retriever.ts"],"mappings":";;;;;;;AAWsC;KAMjC,kBAAA,GAAqB,MAAA;AAAA,UAET,yBAAA;EAFS;;AAE1B;;EAKE,MAAA;EAe+B;;;;EAV/B,MAAA;EAUqB;;;;EALrB,iBAAA;EAiC8B;;;;EA5B9B,mBAAA,EAAqB,UAAA;EAqCA;;;;EAhCrB,UAAA;AAAA;;;;;;;;;;;;;;;;;;;;;cAuBW,mBAAA,eACG,WAAA,GAAc,WAAA,UACpB,oBAAA,CAAqB,KAAA;EAC7B,MAAA;EAEA,MAAA;EAEA,iBAAA;EAEA,mBAAA,EAAqB,UAAA;EAErB,UAAA;EAEA,WAAA,CACE,MAAA,EAAQ,yBAAA,GAA4B,yBAAA,CAA0B,KAAA;EAuFvD;;;;;;;;;;;;;;;;;;EAAA,QA3DD,sBAAA;EAyCF,qBAAA,CAAsB,KAAA,WAAgB,OAAA,CAAQ,iBAAA;;;;;;;;;;EAkB3C,YAAA,GAAY,SAAA,EACR,iBAAA,IAAmB,OAAA,GACpB,kBAAA,KACT,OAAA;AAAA"}

package/dist/retrievers/matryoshka_retriever.js.map

@@ -1 +1 @@
-{"version":3,"file":"matryoshka_retriever.js","names":[],"sources":["../../src/retrievers/matryoshka_retriever.ts"],"sourcesContent":["import { DocumentInterface } from \"@langchain/core/documents\";\nimport { Embeddings } from \"@langchain/core/embeddings\";\nimport {\n  cosineSimilarity,\n  euclideanDistance,\n  innerProduct,\n} from \"@langchain/core/utils/math\";\nimport {\n  VectorStore,\n  VectorStoreRetriever,\n  VectorStoreRetrieverInput,\n} from \"@langchain/core/vectorstores\";\n\n/**\n * Type for options when adding a document to the VectorStore.\n */\n// eslint-disable-next-line @typescript-eslint/no-explicit-any\ntype AddDocumentOptions = Record<string, any>;\n\nexport interface MatryoshkaRetrieverFields {\n  /**\n   * The number of documents to retrieve from the small store.\n   * @default 50\n   */\n  smallK?: number;\n  /**\n   * The number of documents to retrieve from the large store.\n   * @default 8\n   */\n  largeK?: number;\n  /**\n   * The metadata key to store the larger embeddings.\n   * @default \"lc_large_embedding\"\n   */\n  largeEmbeddingKey?: string;\n  /**\n   * The embedding model to use when generating the large\n   * embeddings.\n   */\n  largeEmbeddingModel: Embeddings;\n  /**\n   * The type of search to perform using the large embeddings.\n   * @default \"cosine\"\n   */\n  searchType?: \"cosine\" | \"innerProduct\" | \"euclidean\";\n}\n\n/**\n * A retriever that uses two sets of embeddings to perform adaptive retrieval. Based\n * off of the \"Matryoshka embeddings: faster OpenAI vector search using Adaptive Retrieval\"\n * blog post {@link https://supabase.com/blog/matryoshka-embeddings}.\n *\n *\n * This class performs \"Adaptive Retrieval\" for searching text embeddings efficiently using the\n * Matryoshka Representation Learning (MRL) technique. It retrieves documents similar to a query\n * embedding in two steps:\n *\n * First-pass: Uses a lower dimensional sub-vector from the MRL embedding for an initial, fast,\n * but less accurate search.\n *\n * Second-pass: Re-ranks the top results from the first pass using the full, high-dimensional\n * embedding for higher accuracy.\n *\n *\n * This code implements MRL embeddings for efficient vector search by combining faster,\n * lower-dimensional initial search with accurate, high-dimensional re-ranking.\n */\nexport class MatryoshkaRetriever<\n  Store extends VectorStore = VectorStore,\n> extends VectorStoreRetriever<Store> {\n  smallK = 50;\n\n  largeK = 8;\n\n  largeEmbeddingKey = \"lc_large_embedding\";\n\n  largeEmbeddingModel: Embeddings;\n\n  searchType: \"cosine\" | \"innerProduct\" | \"euclidean\" = \"cosine\";\n\n  constructor(\n    fields: MatryoshkaRetrieverFields & VectorStoreRetrieverInput<Store>\n  ) {\n    super(fields);\n    this.smallK = fields.smallK ?? this.smallK;\n    this.largeK = fields.largeK ?? this.largeK;\n    this.largeEmbeddingKey = fields.largeEmbeddingKey ?? this.largeEmbeddingKey;\n    this.largeEmbeddingModel = fields.largeEmbeddingModel;\n    this.searchType = fields.searchType ?? this.searchType;\n  }\n\n  /**\n   * Ranks documents based on their similarity to a query embedding using larger embeddings.\n   *\n   * This method takes a query embedding and a list of documents (smallResults) as input. Each document\n   * in the smallResults array has previously been associated with a large embedding stored in its metadata.\n   * Depending on the `searchType` (cosine, innerProduct, or euclidean), it calculates the similarity scores\n   * between the query embedding and each document's large embedding. It then ranks the documents based on\n   * these similarity scores, from the most similar to the least similar.\n   *\n   * The method returns a promise that resolves to an array of the top `largeK` documents, where `largeK`\n   * is a class property defining the number of documents to return. This subset of documents is determined\n   * by sorting the entire list of documents based on their similarity scores and then selecting the top\n   * `largeK` documents.\n   *\n   * @param {number[]} embeddedQuery The embedding of the query, represented as an array of numbers.\n   * @param {DocumentInterface[]} smallResults An array of documents, each with metadata that includes a large embedding for similarity comparison.\n   * @returns {Promise<DocumentInterface[]>} A promise that resolves to an array of the top `largeK` ranked documents based on their similarity to the query embedding.\n   */\n  private _rankByLargeEmbeddings(\n    embeddedQuery: number[],\n    smallResults: DocumentInterface[]\n  ): DocumentInterface[] {\n    const largeEmbeddings: Array<number[]> = smallResults.map((doc) =>\n      JSON.parse(doc.metadata[this.largeEmbeddingKey])\n    );\n    let func: () => Array<number[]>;\n\n    switch (this.searchType) {\n      case \"cosine\":\n        func = () => cosineSimilarity([embeddedQuery], largeEmbeddings);\n        break;\n      case \"innerProduct\":\n        func = () => innerProduct([embeddedQuery], largeEmbeddings);\n        break;\n      case \"euclidean\":\n        func = () => euclideanDistance([embeddedQuery], largeEmbeddings);\n        break;\n      default:\n        throw new Error(`Unknown search type: ${this.searchType}`);\n    }\n\n    // Calculate the similarity scores between the query embedding and the large embeddings\n    const [similarityScores] = func();\n\n    // Create an array of indices from 0 to N-1, where N is the number of documents\n    let indices = Array.from(\n      { length: smallResults.length },\n      (_, index) => index\n    );\n\n    indices = indices\n      .map((v, i) => [similarityScores[i], v])\n      .sort(([a], [b]) => b - a)\n      .slice(0, this.largeK)\n      .map(([, i]) => i);\n\n    return indices.map((i) => smallResults[i]);\n  }\n\n  async _getRelevantDocuments(query: string): Promise<DocumentInterface[]> {\n    const [embeddedQuery, smallResults] = await Promise.all([\n      this.largeEmbeddingModel.embedQuery(query),\n      this.vectorStore.similaritySearch(query, this.smallK, this.filter),\n    ]);\n\n    return this._rankByLargeEmbeddings(embeddedQuery, smallResults);\n  }\n\n  /**\n   * Override the default `addDocuments` method to embed the documents twice,\n   * once using the larger embeddings model, and then again using the default\n   * embedding model linked to the vector store.\n   *\n   * @param {DocumentInterface[]} documents - An array of documents to add to the vector store.\n   * @param {AddDocumentOptions} options - An optional object containing additional options for adding documents.\n   * @returns {Promise<string[] | void>} A promise that resolves to an array of the document IDs that were added to the vector store.\n   */\n  override addDocuments = async (\n    documents: DocumentInterface[],\n    options?: AddDocumentOptions\n  ): Promise<string[] | void> => {\n    // Insure documents metadata does not contain the large embedding key\n    if (documents.some((doc) => this.largeEmbeddingKey in doc.metadata)) {\n      throw new Error(\n        `All documents must not contain the large embedding key: ${this.largeEmbeddingKey} in their metadata.`\n      );\n    }\n\n    const allDocPageContent = documents.map((doc) => doc.pageContent);\n    const allDocLargeEmbeddings =\n      await this.largeEmbeddingModel.embedDocuments(allDocPageContent);\n\n    const newDocuments: Array<DocumentInterface> = documents.map(\n      (doc, idx) => ({\n        ...doc,\n        metadata: {\n          ...doc.metadata,\n          [this.largeEmbeddingKey]: JSON.stringify(allDocLargeEmbeddings[idx]),\n        },\n      })\n    );\n\n    return this.vectorStore.addDocuments(newDocuments, options);\n  };\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;AAmEA,IAAa,sBAAb,cAEU,qBAA4B;CACpC,SAAS;CAET,SAAS;CAET,oBAAoB;CAEpB;CAEA,aAAsD;CAEtD,YACE,QACA;AACA,QAAM,OAAO;AACb,OAAK,SAAS,OAAO,UAAU,KAAK;AACpC,OAAK,SAAS,OAAO,UAAU,KAAK;AACpC,OAAK,oBAAoB,OAAO,qBAAqB,KAAK;AAC1D,OAAK,sBAAsB,OAAO;AAClC,OAAK,aAAa,OAAO,cAAc,KAAK;;;;;;;;;;;;;;;;;;;;CAqB9C,uBACE,eACA,cACqB;EACrB,MAAM,kBAAmC,aAAa,KAAK,QACzD,KAAK,MAAM,IAAI,SAAS,KAAK,mBAAmB,CACjD;EACD,IAAI;AAEJ,UAAQ,KAAK,YAAb;GACE,KAAK;AACH,iBAAa,iBAAiB,CAAC,cAAc,EAAE,gBAAgB;AAC/D;GACF,KAAK;AACH,iBAAa,aAAa,CAAC,cAAc,EAAE,gBAAgB;AAC3D;GACF,KAAK;AACH,iBAAa,kBAAkB,CAAC,cAAc,EAAE,gBAAgB;AAChE;GACF,QACE,OAAM,IAAI,MAAM,wBAAwB,KAAK,aAAa;;EAI9D,MAAM,CAAC,oBAAoB,MAAM;EAGjC,IAAI,UAAU,MAAM,KAClB,EAAE,QAAQ,aAAa,QAAQ,GAC9B,GAAG,UAAU,MACf;AAED,YAAU,QACP,KAAK,GAAG,MAAM,CAAC,iBAAiB,IAAI,EAAE,CAAC,CACvC,MAAM,CAAC,IAAI,CAAC,OAAO,IAAI,EAAE,CACzB,MAAM,GAAG,KAAK,OAAO,CACrB,KAAK,GAAG,OAAO,EAAE;AAEpB,SAAO,QAAQ,KAAK,MAAM,aAAa,GAAG;;CAG5C,MAAM,sBAAsB,OAA6C;EACvE,MAAM,CAAC,eAAe,gBAAgB,MAAM,QAAQ,IAAI,CACtD,KAAK,oBAAoB,WAAW,MAAM,EAC1C,KAAK,YAAY,iBAAiB,OAAO,KAAK,QAAQ,KAAK,OAAO,CACnE,CAAC;AAEF,SAAO,KAAK,uBAAuB,eAAe,aAAa;;;;;;;;;;;CAYjE,eAAwB,OACtB,WACA,YAC6B;AAE7B,MAAI,UAAU,MAAM,QAAQ,KAAK,qBAAqB,IAAI,SAAS,CACjE,OAAM,IAAI,MACR,2DAA2D,KAAK,kBAAkB,qBACnF;EAGH,MAAM,oBAAoB,UAAU,KAAK,QAAQ,IAAI,YAAY;EACjE,MAAM,wBACJ,MAAM,KAAK,oBAAoB,eAAe,kBAAkB;EAElE,MAAM,eAAyC,UAAU,KACtD,KAAK,SAAS;GACb,GAAG;GACH,UAAU;IACR,GAAG,IAAI;KACN,KAAK,oBAAoB,KAAK,UAAU,sBAAsB,KAAK;IACrE;GACF,EACF;AAED,SAAO,KAAK,YAAY,aAAa,cAAc,QAAQ"}
+{"version":3,"file":"matryoshka_retriever.js","names":[],"sources":["../../src/retrievers/matryoshka_retriever.ts"],"sourcesContent":["import { DocumentInterface } from \"@langchain/core/documents\";\nimport { Embeddings } from \"@langchain/core/embeddings\";\nimport {\n  cosineSimilarity,\n  euclideanDistance,\n  innerProduct,\n} from \"@langchain/core/utils/math\";\nimport {\n  VectorStore,\n  VectorStoreRetriever,\n  VectorStoreRetrieverInput,\n} from \"@langchain/core/vectorstores\";\n\n/**\n * Type for options when adding a document to the VectorStore.\n */\n// oxlint-disable-next-line @typescript-eslint/no-explicit-any\ntype AddDocumentOptions = Record<string, any>;\n\nexport interface MatryoshkaRetrieverFields {\n  /**\n   * The number of documents to retrieve from the small store.\n   * @default 50\n   */\n  smallK?: number;\n  /**\n   * The number of documents to retrieve from the large store.\n   * @default 8\n   */\n  largeK?: number;\n  /**\n   * The metadata key to store the larger embeddings.\n   * @default \"lc_large_embedding\"\n   */\n  largeEmbeddingKey?: string;\n  /**\n   * The embedding model to use when generating the large\n   * embeddings.\n   */\n  largeEmbeddingModel: Embeddings;\n  /**\n   * The type of search to perform using the large embeddings.\n   * @default \"cosine\"\n   */\n  searchType?: \"cosine\" | \"innerProduct\" | \"euclidean\";\n}\n\n/**\n * A retriever that uses two sets of embeddings to perform adaptive retrieval. Based\n * off of the \"Matryoshka embeddings: faster OpenAI vector search using Adaptive Retrieval\"\n * blog post {@link https://supabase.com/blog/matryoshka-embeddings}.\n *\n *\n * This class performs \"Adaptive Retrieval\" for searching text embeddings efficiently using the\n * Matryoshka Representation Learning (MRL) technique. It retrieves documents similar to a query\n * embedding in two steps:\n *\n * First-pass: Uses a lower dimensional sub-vector from the MRL embedding for an initial, fast,\n * but less accurate search.\n *\n * Second-pass: Re-ranks the top results from the first pass using the full, high-dimensional\n * embedding for higher accuracy.\n *\n *\n * This code implements MRL embeddings for efficient vector search by combining faster,\n * lower-dimensional initial search with accurate, high-dimensional re-ranking.\n */\nexport class MatryoshkaRetriever<\n  Store extends VectorStore = VectorStore,\n> extends VectorStoreRetriever<Store> {\n  smallK = 50;\n\n  largeK = 8;\n\n  largeEmbeddingKey = \"lc_large_embedding\";\n\n  largeEmbeddingModel: Embeddings;\n\n  searchType: \"cosine\" | \"innerProduct\" | \"euclidean\" = \"cosine\";\n\n  constructor(\n    fields: MatryoshkaRetrieverFields & VectorStoreRetrieverInput<Store>\n  ) {\n    super(fields);\n    this.smallK = fields.smallK ?? this.smallK;\n    this.largeK = fields.largeK ?? this.largeK;\n    this.largeEmbeddingKey = fields.largeEmbeddingKey ?? this.largeEmbeddingKey;\n    this.largeEmbeddingModel = fields.largeEmbeddingModel;\n    this.searchType = fields.searchType ?? this.searchType;\n  }\n\n  /**\n   * Ranks documents based on their similarity to a query embedding using larger embeddings.\n   *\n   * This method takes a query embedding and a list of documents (smallResults) as input. Each document\n   * in the smallResults array has previously been associated with a large embedding stored in its metadata.\n   * Depending on the `searchType` (cosine, innerProduct, or euclidean), it calculates the similarity scores\n   * between the query embedding and each document's large embedding. It then ranks the documents based on\n   * these similarity scores, from the most similar to the least similar.\n   *\n   * The method returns a promise that resolves to an array of the top `largeK` documents, where `largeK`\n   * is a class property defining the number of documents to return. This subset of documents is determined\n   * by sorting the entire list of documents based on their similarity scores and then selecting the top\n   * `largeK` documents.\n   *\n   * @param {number[]} embeddedQuery The embedding of the query, represented as an array of numbers.\n   * @param {DocumentInterface[]} smallResults An array of documents, each with metadata that includes a large embedding for similarity comparison.\n   * @returns {Promise<DocumentInterface[]>} A promise that resolves to an array of the top `largeK` ranked documents based on their similarity to the query embedding.\n   */\n  private _rankByLargeEmbeddings(\n    embeddedQuery: number[],\n    smallResults: DocumentInterface[]\n  ): DocumentInterface[] {\n    const largeEmbeddings: Array<number[]> = smallResults.map((doc) =>\n      JSON.parse(doc.metadata[this.largeEmbeddingKey])\n    );\n    let func: () => Array<number[]>;\n\n    switch (this.searchType) {\n      case \"cosine\":\n        func = () => cosineSimilarity([embeddedQuery], largeEmbeddings);\n        break;\n      case \"innerProduct\":\n        func = () => innerProduct([embeddedQuery], largeEmbeddings);\n        break;\n      case \"euclidean\":\n        func = () => euclideanDistance([embeddedQuery], largeEmbeddings);\n        break;\n      default:\n        throw new Error(`Unknown search type: ${this.searchType}`);\n    }\n\n    // Calculate the similarity scores between the query embedding and the large embeddings\n    const [similarityScores] = func();\n\n    // Create an array of indices from 0 to N-1, where N is the number of documents\n    let indices = Array.from(\n      { length: smallResults.length },\n      (_, index) => index\n    );\n\n    indices = indices\n      .map((v, i) => [similarityScores[i], v])\n      .sort(([a], [b]) => b - a)\n      .slice(0, this.largeK)\n      .map(([, i]) => i);\n\n    return indices.map((i) => smallResults[i]);\n  }\n\n  async _getRelevantDocuments(query: string): Promise<DocumentInterface[]> {\n    const [embeddedQuery, smallResults] = await Promise.all([\n      this.largeEmbeddingModel.embedQuery(query),\n      this.vectorStore.similaritySearch(query, this.smallK, this.filter),\n    ]);\n\n    return this._rankByLargeEmbeddings(embeddedQuery, smallResults);\n  }\n\n  /**\n   * Override the default `addDocuments` method to embed the documents twice,\n   * once using the larger embeddings model, and then again using the default\n   * embedding model linked to the vector store.\n   *\n   * @param {DocumentInterface[]} documents - An array of documents to add to the vector store.\n   * @param {AddDocumentOptions} options - An optional object containing additional options for adding documents.\n   * @returns {Promise<string[] | void>} A promise that resolves to an array of the document IDs that were added to the vector store.\n   */\n  override addDocuments = async (\n    documents: DocumentInterface[],\n    options?: AddDocumentOptions\n  ): Promise<string[] | void> => {\n    // Insure documents metadata does not contain the large embedding key\n    if (documents.some((doc) => this.largeEmbeddingKey in doc.metadata)) {\n      throw new Error(\n        `All documents must not contain the large embedding key: ${this.largeEmbeddingKey} in their metadata.`\n      );\n    }\n\n    const allDocPageContent = documents.map((doc) => doc.pageContent);\n    const allDocLargeEmbeddings =\n      await this.largeEmbeddingModel.embedDocuments(allDocPageContent);\n\n    const newDocuments: Array<DocumentInterface> = documents.map(\n      (doc, idx) => ({\n        ...doc,\n        metadata: {\n          ...doc.metadata,\n          [this.largeEmbeddingKey]: JSON.stringify(allDocLargeEmbeddings[idx]),\n        },\n      })\n    );\n\n    return this.vectorStore.addDocuments(newDocuments, options);\n  };\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;AAmEA,IAAa,sBAAb,cAEU,qBAA4B;CACpC,SAAS;CAET,SAAS;CAET,oBAAoB;CAEpB;CAEA,aAAsD;CAEtD,YACE,QACA;AACA,QAAM,OAAO;AACb,OAAK,SAAS,OAAO,UAAU,KAAK;AACpC,OAAK,SAAS,OAAO,UAAU,KAAK;AACpC,OAAK,oBAAoB,OAAO,qBAAqB,KAAK;AAC1D,OAAK,sBAAsB,OAAO;AAClC,OAAK,aAAa,OAAO,cAAc,KAAK;;;;;;;;;;;;;;;;;;;;CAqB9C,uBACE,eACA,cACqB;EACrB,MAAM,kBAAmC,aAAa,KAAK,QACzD,KAAK,MAAM,IAAI,SAAS,KAAK,mBAAmB,CACjD;EACD,IAAI;AAEJ,UAAQ,KAAK,YAAb;GACE,KAAK;AACH,iBAAa,iBAAiB,CAAC,cAAc,EAAE,gBAAgB;AAC/D;GACF,KAAK;AACH,iBAAa,aAAa,CAAC,cAAc,EAAE,gBAAgB;AAC3D;GACF,KAAK;AACH,iBAAa,kBAAkB,CAAC,cAAc,EAAE,gBAAgB;AAChE;GACF,QACE,OAAM,IAAI,MAAM,wBAAwB,KAAK,aAAa;;EAI9D,MAAM,CAAC,oBAAoB,MAAM;EAGjC,IAAI,UAAU,MAAM,KAClB,EAAE,QAAQ,aAAa,QAAQ,GAC9B,GAAG,UAAU,MACf;AAED,YAAU,QACP,KAAK,GAAG,MAAM,CAAC,iBAAiB,IAAI,EAAE,CAAC,CACvC,MAAM,CAAC,IAAI,CAAC,OAAO,IAAI,EAAE,CACzB,MAAM,GAAG,KAAK,OAAO,CACrB,KAAK,GAAG,OAAO,EAAE;AAEpB,SAAO,QAAQ,KAAK,MAAM,aAAa,GAAG;;CAG5C,MAAM,sBAAsB,OAA6C;EACvE,MAAM,CAAC,eAAe,gBAAgB,MAAM,QAAQ,IAAI,CACtD,KAAK,oBAAoB,WAAW,MAAM,EAC1C,KAAK,YAAY,iBAAiB,OAAO,KAAK,QAAQ,KAAK,OAAO,CACnE,CAAC;AAEF,SAAO,KAAK,uBAAuB,eAAe,aAAa;;;;;;;;;;;CAYjE,eAAwB,OACtB,WACA,YAC6B;AAE7B,MAAI,UAAU,MAAM,QAAQ,KAAK,qBAAqB,IAAI,SAAS,CACjE,OAAM,IAAI,MACR,2DAA2D,KAAK,kBAAkB,qBACnF;EAGH,MAAM,oBAAoB,UAAU,KAAK,QAAQ,IAAI,YAAY;EACjE,MAAM,wBACJ,MAAM,KAAK,oBAAoB,eAAe,kBAAkB;EAElE,MAAM,eAAyC,UAAU,KACtD,KAAK,SAAS;GACb,GAAG;GACH,UAAU;IACR,GAAG,IAAI;KACN,KAAK,oBAAoB,KAAK,UAAU,sBAAsB,KAAK;IACrE;GACF,EACF;AAED,SAAO,KAAK,YAAY,aAAa,cAAc,QAAQ"}

package/dist/retrievers/multi_query.cjs.map

@@ -1 +1 @@
-{"version":3,"file":"multi_query.cjs","names":["BaseOutputParser","PromptTemplate","BaseRetriever","LLMChain"],"sources":["../../src/retrievers/multi_query.ts"],"sourcesContent":["import type { BaseLanguageModelInterface } from \"@langchain/core/language_models/base\";\nimport {\n  BaseRetriever,\n  type BaseRetrieverInput,\n  type BaseRetrieverInterface,\n} from \"@langchain/core/retrievers\";\nimport { Document } from \"@langchain/core/documents\";\nimport { BaseOutputParser } from \"@langchain/core/output_parsers\";\nimport { PromptTemplate, BasePromptTemplate } from \"@langchain/core/prompts\";\nimport { CallbackManagerForRetrieverRun } from \"@langchain/core/callbacks/manager\";\nimport { LLMChain } from \"../chains/llm_chain.js\";\nimport type { BaseDocumentCompressor } from \"./document_compressors/index.js\";\n\ninterface LineList {\n  lines: string[];\n}\n\n// eslint-disable-next-line @typescript-eslint/no-explicit-any\nexport type MultiDocs = Document<Record<string, any>>[];\n\nclass LineListOutputParser extends BaseOutputParser<LineList> {\n  static lc_name() {\n    return \"LineListOutputParser\";\n  }\n\n  lc_namespace = [\"langchain\", \"retrievers\", \"multiquery\"];\n\n  async parse(text: string): Promise<LineList> {\n    const startKeyIndex = text.indexOf(\"<questions>\");\n    const endKeyIndex = text.indexOf(\"</questions>\");\n    const questionsStartIndex =\n      startKeyIndex === -1 ? 0 : startKeyIndex + \"<questions>\".length;\n    const questionsEndIndex = endKeyIndex === -1 ? text.length : endKeyIndex;\n    const lines = text\n      .slice(questionsStartIndex, questionsEndIndex)\n      .trim()\n      .split(\"\\n\")\n      .filter((line) => line.trim() !== \"\");\n    return { lines };\n  }\n\n  getFormatInstructions(): string {\n    throw new Error(\"Not implemented.\");\n  }\n}\n\n// Create template\nconst DEFAULT_QUERY_PROMPT = /* #__PURE__ */ new PromptTemplate({\n  inputVariables: [\"question\", \"queryCount\"],\n  template: `You are an AI language model assistant. Your task is\nto generate {queryCount} different versions of the given user\nquestion to retrieve relevant documents from a vector database.\nBy generating multiple perspectives on the user question,\nyour goal is to help the user overcome some of the limitations\nof distance-based similarity search.\n\nProvide these alternative questions separated by newlines between XML tags. For example:\n\n<questions>\nQuestion 1\nQuestion 2\nQuestion 3\n</questions>\n\nOriginal question: {question}`,\n});\n\nexport interface MultiQueryRetrieverInput extends BaseRetrieverInput {\n  retriever: BaseRetrieverInterface;\n  /** @deprecated Pass a custom prompt into `.fromLLM` instead. */\n  llmChain: LLMChain<LineList>;\n  queryCount?: number;\n  parserKey?: string;\n  documentCompressor?: BaseDocumentCompressor | undefined;\n  documentCompressorFilteringFn?: (docs: MultiDocs) => MultiDocs;\n}\n\n/**\n * @example\n * ```typescript\n * const retriever = new MultiQueryRetriever.fromLLM({\n *   llm: new ChatAnthropic({}),\n *   retriever: new MemoryVectorStore().asRetriever(),\n *   verbose: true,\n * });\n * const retrievedDocs = await retriever.invoke(\n *   \"What are mitochondria made of?\",\n * );\n * ```\n */\nexport class MultiQueryRetriever extends BaseRetriever {\n  static lc_name() {\n    return \"MultiQueryRetriever\";\n  }\n\n  lc_namespace = [\"langchain\", \"retrievers\", \"multiquery\"];\n\n  private retriever: BaseRetrieverInterface;\n\n  private llmChain: LLMChain<LineList>;\n\n  private queryCount = 3;\n\n  private parserKey = \"lines\";\n\n  documentCompressor: BaseDocumentCompressor | undefined;\n\n  documentCompressorFilteringFn?: MultiQueryRetrieverInput[\"documentCompressorFilteringFn\"];\n\n  constructor(fields: MultiQueryRetrieverInput) {\n    super(fields);\n    this.retriever = fields.retriever;\n    this.llmChain = fields.llmChain;\n    this.queryCount = fields.queryCount ?? this.queryCount;\n    this.parserKey = fields.parserKey ?? this.parserKey;\n    this.documentCompressor = fields.documentCompressor;\n    this.documentCompressorFilteringFn = fields.documentCompressorFilteringFn;\n  }\n\n  static fromLLM(\n    fields: Omit<MultiQueryRetrieverInput, \"llmChain\"> & {\n      llm: BaseLanguageModelInterface;\n      prompt?: BasePromptTemplate;\n    }\n  ): MultiQueryRetriever {\n    const {\n      retriever,\n      llm,\n      prompt = DEFAULT_QUERY_PROMPT,\n      queryCount,\n      parserKey,\n      ...rest\n    } = fields;\n    const outputParser = new LineListOutputParser();\n    const llmChain = new LLMChain({ llm, prompt, outputParser });\n    return new this({ retriever, llmChain, queryCount, parserKey, ...rest });\n  }\n\n  // Generate the different queries for each retrieval, using our llmChain\n  private async _generateQueries(\n    question: string,\n    runManager?: CallbackManagerForRetrieverRun\n  ): Promise<string[]> {\n    const response = await this.llmChain.call(\n      { question, queryCount: this.queryCount },\n      runManager?.getChild()\n    );\n    const lines = response.text[this.parserKey] || [];\n    if (this.verbose) {\n      console.log(`Generated queries: ${lines}`);\n    }\n    return lines;\n  }\n\n  // Retrieve documents using the original retriever\n  private async _retrieveDocuments(\n    queries: string[],\n    runManager?: CallbackManagerForRetrieverRun\n  ): Promise<Document[]> {\n    const documents: Document[] = [];\n    await Promise.all(\n      queries.map(async (query) => {\n        const docs = await this.retriever.invoke(query, runManager?.getChild());\n        documents.push(...docs);\n      })\n    );\n    return documents;\n  }\n\n  // Deduplicate the documents that were returned in multiple retrievals\n  private _uniqueUnion(documents: Document[]): Document[] {\n    const uniqueDocumentsDict: { [key: string]: Document } = {};\n\n    for (const doc of documents) {\n      const key = `${doc.pageContent}:${JSON.stringify(\n        Object.entries(doc.metadata).sort()\n      )}`;\n      uniqueDocumentsDict[key] = doc;\n    }\n\n    const uniqueDocuments = Object.values(uniqueDocumentsDict);\n    return uniqueDocuments;\n  }\n\n  async _getRelevantDocuments(\n    question: string,\n    runManager?: CallbackManagerForRetrieverRun\n  ): Promise<Document[]> {\n    const queries = await this._generateQueries(question, runManager);\n    const documents = await this._retrieveDocuments(queries, runManager);\n    const uniqueDocuments = this._uniqueUnion(documents);\n\n    let outputDocs = uniqueDocuments;\n    if (this.documentCompressor && uniqueDocuments.length) {\n      outputDocs = await this.documentCompressor.compressDocuments(\n        uniqueDocuments,\n        question,\n        runManager?.getChild()\n      );\n      if (this.documentCompressorFilteringFn) {\n        outputDocs = this.documentCompressorFilteringFn(outputDocs);\n      }\n    }\n\n    return outputDocs;\n  }\n}\n"],"mappings":";;;;;;;;AAoBA,IAAM,uBAAN,cAAmCA,+BAAAA,iBAA2B;CAC5D,OAAO,UAAU;AACf,SAAO;;CAGT,eAAe;EAAC;EAAa;EAAc;EAAa;CAExD,MAAM,MAAM,MAAiC;EAC3C,MAAM,gBAAgB,KAAK,QAAQ,cAAc;EACjD,MAAM,cAAc,KAAK,QAAQ,eAAe;EAChD,MAAM,sBACJ,kBAAkB,KAAK,IAAI,gBAAgB;EAC7C,MAAM,oBAAoB,gBAAgB,KAAK,KAAK,SAAS;AAM7D,SAAO,EAAE,OALK,KACX,MAAM,qBAAqB,kBAAkB,CAC7C,MAAM,CACN,MAAM,KAAK,CACX,QAAQ,SAAS,KAAK,MAAM,KAAK,GAAG,EACvB;;CAGlB,wBAAgC;AAC9B,QAAM,IAAI,MAAM,mBAAmB;;;AAKvC,MAAM,uCAAuC,IAAIC,wBAAAA,eAAe;CAC9D,gBAAgB,CAAC,YAAY,aAAa;CAC1C,UAAU;;;;;;;;;;;;;;;;CAgBX,CAAC;;;;;;;;;;;;;;AAyBF,IAAa,sBAAb,cAAyCC,2BAAAA,cAAc;CACrD,OAAO,UAAU;AACf,SAAO;;CAGT,eAAe;EAAC;EAAa;EAAc;EAAa;CAExD;CAEA;CAEA,aAAqB;CAErB,YAAoB;CAEpB;CAEA;CAEA,YAAY,QAAkC;AAC5C,QAAM,OAAO;AACb,OAAK,YAAY,OAAO;AACxB,OAAK,WAAW,OAAO;AACvB,OAAK,aAAa,OAAO,cAAc,KAAK;AAC5C,OAAK,YAAY,OAAO,aAAa,KAAK;AAC1C,OAAK,qBAAqB,OAAO;AACjC,OAAK,gCAAgC,OAAO;;CAG9C,OAAO,QACL,QAIqB;EACrB,MAAM,EACJ,WACA,KACA,SAAS,sBACT,YACA,WACA,GAAG,SACD;EAEJ,MAAM,WAAW,IAAIC,kBAAAA,SAAS;GAAE;GAAK;GAAQ,cADxB,IAAI,sBAAsB;GACY,CAAC;AAC5D,SAAO,IAAI,KAAK;GAAE;GAAW;GAAU;GAAY;GAAW,GAAG;GAAM,CAAC;;CAI1E,MAAc,iBACZ,UACA,YACmB;EAKnB,MAAM,SAJW,MAAM,KAAK,SAAS,KACnC;GAAE;GAAU,YAAY,KAAK;GAAY,EACzC,YAAY,UAAU,CACvB,EACsB,KAAK,KAAK,cAAc,EAAE;AACjD,MAAI,KAAK,QACP,SAAQ,IAAI,sBAAsB,QAAQ;AAE5C,SAAO;;CAIT,MAAc,mBACZ,SACA,YACqB;EACrB,MAAM,YAAwB,EAAE;AAChC,QAAM,QAAQ,IACZ,QAAQ,IAAI,OAAO,UAAU;GAC3B,MAAM,OAAO,MAAM,KAAK,UAAU,OAAO,OAAO,YAAY,UAAU,CAAC;AACvE,aAAU,KAAK,GAAG,KAAK;IACvB,CACH;AACD,SAAO;;CAIT,aAAqB,WAAmC;EACtD,MAAM,sBAAmD,EAAE;AAE3D,OAAK,MAAM,OAAO,WAAW;GAC3B,MAAM,MAAM,GAAG,IAAI,YAAY,GAAG,KAAK,UACrC,OAAO,QAAQ,IAAI,SAAS,CAAC,MAAM,CACpC;AACD,uBAAoB,OAAO;;AAI7B,SADwB,OAAO,OAAO,oBAAoB;;CAI5D,MAAM,sBACJ,UACA,YACqB;EACrB,MAAM,UAAU,MAAM,KAAK,iBAAiB,UAAU,WAAW;EACjE,MAAM,YAAY,MAAM,KAAK,mBAAmB,SAAS,WAAW;EACpE,MAAM,kBAAkB,KAAK,aAAa,UAAU;EAEpD,IAAI,aAAa;AACjB,MAAI,KAAK,sBAAsB,gBAAgB,QAAQ;AACrD,gBAAa,MAAM,KAAK,mBAAmB,kBACzC,iBACA,UACA,YAAY,UAAU,CACvB;AACD,OAAI,KAAK,8BACP,cAAa,KAAK,8BAA8B,WAAW;;AAI/D,SAAO"}
+{"version":3,"file":"multi_query.cjs","names":["BaseOutputParser","PromptTemplate","BaseRetriever","LLMChain"],"sources":["../../src/retrievers/multi_query.ts"],"sourcesContent":["import type { BaseLanguageModelInterface } from \"@langchain/core/language_models/base\";\nimport {\n  BaseRetriever,\n  type BaseRetrieverInput,\n  type BaseRetrieverInterface,\n} from \"@langchain/core/retrievers\";\nimport { Document } from \"@langchain/core/documents\";\nimport { BaseOutputParser } from \"@langchain/core/output_parsers\";\nimport { PromptTemplate, BasePromptTemplate } from \"@langchain/core/prompts\";\nimport { CallbackManagerForRetrieverRun } from \"@langchain/core/callbacks/manager\";\nimport { LLMChain } from \"../chains/llm_chain.js\";\nimport type { BaseDocumentCompressor } from \"./document_compressors/index.js\";\n\ninterface LineList {\n  lines: string[];\n}\n\n// oxlint-disable-next-line @typescript-eslint/no-explicit-any\nexport type MultiDocs = Document<Record<string, any>>[];\n\nclass LineListOutputParser extends BaseOutputParser<LineList> {\n  static lc_name() {\n    return \"LineListOutputParser\";\n  }\n\n  lc_namespace = [\"langchain\", \"retrievers\", \"multiquery\"];\n\n  async parse(text: string): Promise<LineList> {\n    const startKeyIndex = text.indexOf(\"<questions>\");\n    const endKeyIndex = text.indexOf(\"</questions>\");\n    const questionsStartIndex =\n      startKeyIndex === -1 ? 0 : startKeyIndex + \"<questions>\".length;\n    const questionsEndIndex = endKeyIndex === -1 ? text.length : endKeyIndex;\n    const lines = text\n      .slice(questionsStartIndex, questionsEndIndex)\n      .trim()\n      .split(\"\\n\")\n      .filter((line) => line.trim() !== \"\");\n    return { lines };\n  }\n\n  getFormatInstructions(): string {\n    throw new Error(\"Not implemented.\");\n  }\n}\n\n// Create template\nconst DEFAULT_QUERY_PROMPT = /* #__PURE__ */ new PromptTemplate({\n  inputVariables: [\"question\", \"queryCount\"],\n  template: `You are an AI language model assistant. Your task is\nto generate {queryCount} different versions of the given user\nquestion to retrieve relevant documents from a vector database.\nBy generating multiple perspectives on the user question,\nyour goal is to help the user overcome some of the limitations\nof distance-based similarity search.\n\nProvide these alternative questions separated by newlines between XML tags. For example:\n\n<questions>\nQuestion 1\nQuestion 2\nQuestion 3\n</questions>\n\nOriginal question: {question}`,\n});\n\nexport interface MultiQueryRetrieverInput extends BaseRetrieverInput {\n  retriever: BaseRetrieverInterface;\n  /** @deprecated Pass a custom prompt into `.fromLLM` instead. */\n  llmChain: LLMChain<LineList>;\n  queryCount?: number;\n  parserKey?: string;\n  documentCompressor?: BaseDocumentCompressor | undefined;\n  documentCompressorFilteringFn?: (docs: MultiDocs) => MultiDocs;\n}\n\n/**\n * @example\n * ```typescript\n * const retriever = new MultiQueryRetriever.fromLLM({\n *   llm: new ChatAnthropic({}),\n *   retriever: new MemoryVectorStore().asRetriever(),\n *   verbose: true,\n * });\n * const retrievedDocs = await retriever.invoke(\n *   \"What are mitochondria made of?\",\n * );\n * ```\n */\nexport class MultiQueryRetriever extends BaseRetriever {\n  static lc_name() {\n    return \"MultiQueryRetriever\";\n  }\n\n  lc_namespace = [\"langchain\", \"retrievers\", \"multiquery\"];\n\n  private retriever: BaseRetrieverInterface;\n\n  private llmChain: LLMChain<LineList>;\n\n  private queryCount = 3;\n\n  private parserKey = \"lines\";\n\n  documentCompressor: BaseDocumentCompressor | undefined;\n\n  documentCompressorFilteringFn?: MultiQueryRetrieverInput[\"documentCompressorFilteringFn\"];\n\n  constructor(fields: MultiQueryRetrieverInput) {\n    super(fields);\n    this.retriever = fields.retriever;\n    this.llmChain = fields.llmChain;\n    this.queryCount = fields.queryCount ?? this.queryCount;\n    this.parserKey = fields.parserKey ?? this.parserKey;\n    this.documentCompressor = fields.documentCompressor;\n    this.documentCompressorFilteringFn = fields.documentCompressorFilteringFn;\n  }\n\n  static fromLLM(\n    fields: Omit<MultiQueryRetrieverInput, \"llmChain\"> & {\n      llm: BaseLanguageModelInterface;\n      prompt?: BasePromptTemplate;\n    }\n  ): MultiQueryRetriever {\n    const {\n      retriever,\n      llm,\n      prompt = DEFAULT_QUERY_PROMPT,\n      queryCount,\n      parserKey,\n      ...rest\n    } = fields;\n    const outputParser = new LineListOutputParser();\n    const llmChain = new LLMChain({ llm, prompt, outputParser });\n    return new this({ retriever, llmChain, queryCount, parserKey, ...rest });\n  }\n\n  // Generate the different queries for each retrieval, using our llmChain\n  private async _generateQueries(\n    question: string,\n    runManager?: CallbackManagerForRetrieverRun\n  ): Promise<string[]> {\n    const response = await this.llmChain.call(\n      { question, queryCount: this.queryCount },\n      runManager?.getChild()\n    );\n    const lines = response.text[this.parserKey] || [];\n    if (this.verbose) {\n      console.log(`Generated queries: ${lines}`);\n    }\n    return lines;\n  }\n\n  // Retrieve documents using the original retriever\n  private async _retrieveDocuments(\n    queries: string[],\n    runManager?: CallbackManagerForRetrieverRun\n  ): Promise<Document[]> {\n    const documents: Document[] = [];\n    await Promise.all(\n      queries.map(async (query) => {\n        const docs = await this.retriever.invoke(query, runManager?.getChild());\n        documents.push(...docs);\n      })\n    );\n    return documents;\n  }\n\n  // Deduplicate the documents that were returned in multiple retrievals\n  private _uniqueUnion(documents: Document[]): Document[] {\n    const uniqueDocumentsDict: { [key: string]: Document } = {};\n\n    for (const doc of documents) {\n      const key = `${doc.pageContent}:${JSON.stringify(\n        Object.entries(doc.metadata).sort()\n      )}`;\n      uniqueDocumentsDict[key] = doc;\n    }\n\n    const uniqueDocuments = Object.values(uniqueDocumentsDict);\n    return uniqueDocuments;\n  }\n\n  async _getRelevantDocuments(\n    question: string,\n    runManager?: CallbackManagerForRetrieverRun\n  ): Promise<Document[]> {\n    const queries = await this._generateQueries(question, runManager);\n    const documents = await this._retrieveDocuments(queries, runManager);\n    const uniqueDocuments = this._uniqueUnion(documents);\n\n    let outputDocs = uniqueDocuments;\n    if (this.documentCompressor && uniqueDocuments.length) {\n      outputDocs = await this.documentCompressor.compressDocuments(\n        uniqueDocuments,\n        question,\n        runManager?.getChild()\n      );\n      if (this.documentCompressorFilteringFn) {\n        outputDocs = this.documentCompressorFilteringFn(outputDocs);\n      }\n    }\n\n    return outputDocs;\n  }\n}\n"],"mappings":";;;;;;;;AAoBA,IAAM,uBAAN,cAAmCA,+BAAAA,iBAA2B;CAC5D,OAAO,UAAU;AACf,SAAO;;CAGT,eAAe;EAAC;EAAa;EAAc;EAAa;CAExD,MAAM,MAAM,MAAiC;EAC3C,MAAM,gBAAgB,KAAK,QAAQ,cAAc;EACjD,MAAM,cAAc,KAAK,QAAQ,eAAe;EAChD,MAAM,sBACJ,kBAAkB,KAAK,IAAI,gBAAgB;EAC7C,MAAM,oBAAoB,gBAAgB,KAAK,KAAK,SAAS;AAM7D,SAAO,EAAE,OALK,KACX,MAAM,qBAAqB,kBAAkB,CAC7C,MAAM,CACN,MAAM,KAAK,CACX,QAAQ,SAAS,KAAK,MAAM,KAAK,GAAG,EACvB;;CAGlB,wBAAgC;AAC9B,QAAM,IAAI,MAAM,mBAAmB;;;AAKvC,MAAM,uCAAuC,IAAIC,wBAAAA,eAAe;CAC9D,gBAAgB,CAAC,YAAY,aAAa;CAC1C,UAAU;;;;;;;;;;;;;;;;CAgBX,CAAC;;;;;;;;;;;;;;AAyBF,IAAa,sBAAb,cAAyCC,2BAAAA,cAAc;CACrD,OAAO,UAAU;AACf,SAAO;;CAGT,eAAe;EAAC;EAAa;EAAc;EAAa;CAExD;CAEA;CAEA,aAAqB;CAErB,YAAoB;CAEpB;CAEA;CAEA,YAAY,QAAkC;AAC5C,QAAM,OAAO;AACb,OAAK,YAAY,OAAO;AACxB,OAAK,WAAW,OAAO;AACvB,OAAK,aAAa,OAAO,cAAc,KAAK;AAC5C,OAAK,YAAY,OAAO,aAAa,KAAK;AAC1C,OAAK,qBAAqB,OAAO;AACjC,OAAK,gCAAgC,OAAO;;CAG9C,OAAO,QACL,QAIqB;EACrB,MAAM,EACJ,WACA,KACA,SAAS,sBACT,YACA,WACA,GAAG,SACD;EAEJ,MAAM,WAAW,IAAIC,kBAAAA,SAAS;GAAE;GAAK;GAAQ,cADxB,IAAI,sBAAsB;GACY,CAAC;AAC5D,SAAO,IAAI,KAAK;GAAE;GAAW;GAAU;GAAY;GAAW,GAAG;GAAM,CAAC;;CAI1E,MAAc,iBACZ,UACA,YACmB;EAKnB,MAAM,SAJW,MAAM,KAAK,SAAS,KACnC;GAAE;GAAU,YAAY,KAAK;GAAY,EACzC,YAAY,UAAU,CACvB,EACsB,KAAK,KAAK,cAAc,EAAE;AACjD,MAAI,KAAK,QACP,SAAQ,IAAI,sBAAsB,QAAQ;AAE5C,SAAO;;CAIT,MAAc,mBACZ,SACA,YACqB;EACrB,MAAM,YAAwB,EAAE;AAChC,QAAM,QAAQ,IACZ,QAAQ,IAAI,OAAO,UAAU;GAC3B,MAAM,OAAO,MAAM,KAAK,UAAU,OAAO,OAAO,YAAY,UAAU,CAAC;AACvE,aAAU,KAAK,GAAG,KAAK;IACvB,CACH;AACD,SAAO;;CAIT,aAAqB,WAAmC;EACtD,MAAM,sBAAmD,EAAE;AAE3D,OAAK,MAAM,OAAO,WAAW;GAC3B,MAAM,MAAM,GAAG,IAAI,YAAY,GAAG,KAAK,UACrC,OAAO,QAAQ,IAAI,SAAS,CAAC,MAAM,CACpC;AACD,uBAAoB,OAAO;;AAI7B,SADwB,OAAO,OAAO,oBAAoB;;CAI5D,MAAM,sBACJ,UACA,YACqB;EACrB,MAAM,UAAU,MAAM,KAAK,iBAAiB,UAAU,WAAW;EACjE,MAAM,YAAY,MAAM,KAAK,mBAAmB,SAAS,WAAW;EACpE,MAAM,kBAAkB,KAAK,aAAa,UAAU;EAEpD,IAAI,aAAa;AACjB,MAAI,KAAK,sBAAsB,gBAAgB,QAAQ;AACrD,gBAAa,MAAM,KAAK,mBAAmB,kBACzC,iBACA,UACA,YAAY,UAAU,CACvB;AACD,OAAI,KAAK,8BACP,cAAa,KAAK,8BAA8B,WAAW;;AAI/D,SAAO"}

package/dist/retrievers/multi_query.js.map

@@ -1 +1 @@
-{"version":3,"file":"multi_query.js","names":[],"sources":["../../src/retrievers/multi_query.ts"],"sourcesContent":["import type { BaseLanguageModelInterface } from \"@langchain/core/language_models/base\";\nimport {\n  BaseRetriever,\n  type BaseRetrieverInput,\n  type BaseRetrieverInterface,\n} from \"@langchain/core/retrievers\";\nimport { Document } from \"@langchain/core/documents\";\nimport { BaseOutputParser } from \"@langchain/core/output_parsers\";\nimport { PromptTemplate, BasePromptTemplate } from \"@langchain/core/prompts\";\nimport { CallbackManagerForRetrieverRun } from \"@langchain/core/callbacks/manager\";\nimport { LLMChain } from \"../chains/llm_chain.js\";\nimport type { BaseDocumentCompressor } from \"./document_compressors/index.js\";\n\ninterface LineList {\n  lines: string[];\n}\n\n// eslint-disable-next-line @typescript-eslint/no-explicit-any\nexport type MultiDocs = Document<Record<string, any>>[];\n\nclass LineListOutputParser extends BaseOutputParser<LineList> {\n  static lc_name() {\n    return \"LineListOutputParser\";\n  }\n\n  lc_namespace = [\"langchain\", \"retrievers\", \"multiquery\"];\n\n  async parse(text: string): Promise<LineList> {\n    const startKeyIndex = text.indexOf(\"<questions>\");\n    const endKeyIndex = text.indexOf(\"</questions>\");\n    const questionsStartIndex =\n      startKeyIndex === -1 ? 0 : startKeyIndex + \"<questions>\".length;\n    const questionsEndIndex = endKeyIndex === -1 ? text.length : endKeyIndex;\n    const lines = text\n      .slice(questionsStartIndex, questionsEndIndex)\n      .trim()\n      .split(\"\\n\")\n      .filter((line) => line.trim() !== \"\");\n    return { lines };\n  }\n\n  getFormatInstructions(): string {\n    throw new Error(\"Not implemented.\");\n  }\n}\n\n// Create template\nconst DEFAULT_QUERY_PROMPT = /* #__PURE__ */ new PromptTemplate({\n  inputVariables: [\"question\", \"queryCount\"],\n  template: `You are an AI language model assistant. Your task is\nto generate {queryCount} different versions of the given user\nquestion to retrieve relevant documents from a vector database.\nBy generating multiple perspectives on the user question,\nyour goal is to help the user overcome some of the limitations\nof distance-based similarity search.\n\nProvide these alternative questions separated by newlines between XML tags. For example:\n\n<questions>\nQuestion 1\nQuestion 2\nQuestion 3\n</questions>\n\nOriginal question: {question}`,\n});\n\nexport interface MultiQueryRetrieverInput extends BaseRetrieverInput {\n  retriever: BaseRetrieverInterface;\n  /** @deprecated Pass a custom prompt into `.fromLLM` instead. */\n  llmChain: LLMChain<LineList>;\n  queryCount?: number;\n  parserKey?: string;\n  documentCompressor?: BaseDocumentCompressor | undefined;\n  documentCompressorFilteringFn?: (docs: MultiDocs) => MultiDocs;\n}\n\n/**\n * @example\n * ```typescript\n * const retriever = new MultiQueryRetriever.fromLLM({\n *   llm: new ChatAnthropic({}),\n *   retriever: new MemoryVectorStore().asRetriever(),\n *   verbose: true,\n * });\n * const retrievedDocs = await retriever.invoke(\n *   \"What are mitochondria made of?\",\n * );\n * ```\n */\nexport class MultiQueryRetriever extends BaseRetriever {\n  static lc_name() {\n    return \"MultiQueryRetriever\";\n  }\n\n  lc_namespace = [\"langchain\", \"retrievers\", \"multiquery\"];\n\n  private retriever: BaseRetrieverInterface;\n\n  private llmChain: LLMChain<LineList>;\n\n  private queryCount = 3;\n\n  private parserKey = \"lines\";\n\n  documentCompressor: BaseDocumentCompressor | undefined;\n\n  documentCompressorFilteringFn?: MultiQueryRetrieverInput[\"documentCompressorFilteringFn\"];\n\n  constructor(fields: MultiQueryRetrieverInput) {\n    super(fields);\n    this.retriever = fields.retriever;\n    this.llmChain = fields.llmChain;\n    this.queryCount = fields.queryCount ?? this.queryCount;\n    this.parserKey = fields.parserKey ?? this.parserKey;\n    this.documentCompressor = fields.documentCompressor;\n    this.documentCompressorFilteringFn = fields.documentCompressorFilteringFn;\n  }\n\n  static fromLLM(\n    fields: Omit<MultiQueryRetrieverInput, \"llmChain\"> & {\n      llm: BaseLanguageModelInterface;\n      prompt?: BasePromptTemplate;\n    }\n  ): MultiQueryRetriever {\n    const {\n      retriever,\n      llm,\n      prompt = DEFAULT_QUERY_PROMPT,\n      queryCount,\n      parserKey,\n      ...rest\n    } = fields;\n    const outputParser = new LineListOutputParser();\n    const llmChain = new LLMChain({ llm, prompt, outputParser });\n    return new this({ retriever, llmChain, queryCount, parserKey, ...rest });\n  }\n\n  // Generate the different queries for each retrieval, using our llmChain\n  private async _generateQueries(\n    question: string,\n    runManager?: CallbackManagerForRetrieverRun\n  ): Promise<string[]> {\n    const response = await this.llmChain.call(\n      { question, queryCount: this.queryCount },\n      runManager?.getChild()\n    );\n    const lines = response.text[this.parserKey] || [];\n    if (this.verbose) {\n      console.log(`Generated queries: ${lines}`);\n    }\n    return lines;\n  }\n\n  // Retrieve documents using the original retriever\n  private async _retrieveDocuments(\n    queries: string[],\n    runManager?: CallbackManagerForRetrieverRun\n  ): Promise<Document[]> {\n    const documents: Document[] = [];\n    await Promise.all(\n      queries.map(async (query) => {\n        const docs = await this.retriever.invoke(query, runManager?.getChild());\n        documents.push(...docs);\n      })\n    );\n    return documents;\n  }\n\n  // Deduplicate the documents that were returned in multiple retrievals\n  private _uniqueUnion(documents: Document[]): Document[] {\n    const uniqueDocumentsDict: { [key: string]: Document } = {};\n\n    for (const doc of documents) {\n      const key = `${doc.pageContent}:${JSON.stringify(\n        Object.entries(doc.metadata).sort()\n      )}`;\n      uniqueDocumentsDict[key] = doc;\n    }\n\n    const uniqueDocuments = Object.values(uniqueDocumentsDict);\n    return uniqueDocuments;\n  }\n\n  async _getRelevantDocuments(\n    question: string,\n    runManager?: CallbackManagerForRetrieverRun\n  ): Promise<Document[]> {\n    const queries = await this._generateQueries(question, runManager);\n    const documents = await this._retrieveDocuments(queries, runManager);\n    const uniqueDocuments = this._uniqueUnion(documents);\n\n    let outputDocs = uniqueDocuments;\n    if (this.documentCompressor && uniqueDocuments.length) {\n      outputDocs = await this.documentCompressor.compressDocuments(\n        uniqueDocuments,\n        question,\n        runManager?.getChild()\n      );\n      if (this.documentCompressorFilteringFn) {\n        outputDocs = this.documentCompressorFilteringFn(outputDocs);\n      }\n    }\n\n    return outputDocs;\n  }\n}\n"],"mappings":";;;;;;;AAoBA,IAAM,uBAAN,cAAmC,iBAA2B;CAC5D,OAAO,UAAU;AACf,SAAO;;CAGT,eAAe;EAAC;EAAa;EAAc;EAAa;CAExD,MAAM,MAAM,MAAiC;EAC3C,MAAM,gBAAgB,KAAK,QAAQ,cAAc;EACjD,MAAM,cAAc,KAAK,QAAQ,eAAe;EAChD,MAAM,sBACJ,kBAAkB,KAAK,IAAI,gBAAgB;EAC7C,MAAM,oBAAoB,gBAAgB,KAAK,KAAK,SAAS;AAM7D,SAAO,EAAE,OALK,KACX,MAAM,qBAAqB,kBAAkB,CAC7C,MAAM,CACN,MAAM,KAAK,CACX,QAAQ,SAAS,KAAK,MAAM,KAAK,GAAG,EACvB;;CAGlB,wBAAgC;AAC9B,QAAM,IAAI,MAAM,mBAAmB;;;AAKvC,MAAM,uCAAuC,IAAI,eAAe;CAC9D,gBAAgB,CAAC,YAAY,aAAa;CAC1C,UAAU;;;;;;;;;;;;;;;;CAgBX,CAAC;;;;;;;;;;;;;;AAyBF,IAAa,sBAAb,cAAyC,cAAc;CACrD,OAAO,UAAU;AACf,SAAO;;CAGT,eAAe;EAAC;EAAa;EAAc;EAAa;CAExD;CAEA;CAEA,aAAqB;CAErB,YAAoB;CAEpB;CAEA;CAEA,YAAY,QAAkC;AAC5C,QAAM,OAAO;AACb,OAAK,YAAY,OAAO;AACxB,OAAK,WAAW,OAAO;AACvB,OAAK,aAAa,OAAO,cAAc,KAAK;AAC5C,OAAK,YAAY,OAAO,aAAa,KAAK;AAC1C,OAAK,qBAAqB,OAAO;AACjC,OAAK,gCAAgC,OAAO;;CAG9C,OAAO,QACL,QAIqB;EACrB,MAAM,EACJ,WACA,KACA,SAAS,sBACT,YACA,WACA,GAAG,SACD;EAEJ,MAAM,WAAW,IAAI,SAAS;GAAE;GAAK;GAAQ,cADxB,IAAI,sBAAsB;GACY,CAAC;AAC5D,SAAO,IAAI,KAAK;GAAE;GAAW;GAAU;GAAY;GAAW,GAAG;GAAM,CAAC;;CAI1E,MAAc,iBACZ,UACA,YACmB;EAKnB,MAAM,SAJW,MAAM,KAAK,SAAS,KACnC;GAAE;GAAU,YAAY,KAAK;GAAY,EACzC,YAAY,UAAU,CACvB,EACsB,KAAK,KAAK,cAAc,EAAE;AACjD,MAAI,KAAK,QACP,SAAQ,IAAI,sBAAsB,QAAQ;AAE5C,SAAO;;CAIT,MAAc,mBACZ,SACA,YACqB;EACrB,MAAM,YAAwB,EAAE;AAChC,QAAM,QAAQ,IACZ,QAAQ,IAAI,OAAO,UAAU;GAC3B,MAAM,OAAO,MAAM,KAAK,UAAU,OAAO,OAAO,YAAY,UAAU,CAAC;AACvE,aAAU,KAAK,GAAG,KAAK;IACvB,CACH;AACD,SAAO;;CAIT,aAAqB,WAAmC;EACtD,MAAM,sBAAmD,EAAE;AAE3D,OAAK,MAAM,OAAO,WAAW;GAC3B,MAAM,MAAM,GAAG,IAAI,YAAY,GAAG,KAAK,UACrC,OAAO,QAAQ,IAAI,SAAS,CAAC,MAAM,CACpC;AACD,uBAAoB,OAAO;;AAI7B,SADwB,OAAO,OAAO,oBAAoB;;CAI5D,MAAM,sBACJ,UACA,YACqB;EACrB,MAAM,UAAU,MAAM,KAAK,iBAAiB,UAAU,WAAW;EACjE,MAAM,YAAY,MAAM,KAAK,mBAAmB,SAAS,WAAW;EACpE,MAAM,kBAAkB,KAAK,aAAa,UAAU;EAEpD,IAAI,aAAa;AACjB,MAAI,KAAK,sBAAsB,gBAAgB,QAAQ;AACrD,gBAAa,MAAM,KAAK,mBAAmB,kBACzC,iBACA,UACA,YAAY,UAAU,CACvB;AACD,OAAI,KAAK,8BACP,cAAa,KAAK,8BAA8B,WAAW;;AAI/D,SAAO"}
+{"version":3,"file":"multi_query.js","names":[],"sources":["../../src/retrievers/multi_query.ts"],"sourcesContent":["import type { BaseLanguageModelInterface } from \"@langchain/core/language_models/base\";\nimport {\n  BaseRetriever,\n  type BaseRetrieverInput,\n  type BaseRetrieverInterface,\n} from \"@langchain/core/retrievers\";\nimport { Document } from \"@langchain/core/documents\";\nimport { BaseOutputParser } from \"@langchain/core/output_parsers\";\nimport { PromptTemplate, BasePromptTemplate } from \"@langchain/core/prompts\";\nimport { CallbackManagerForRetrieverRun } from \"@langchain/core/callbacks/manager\";\nimport { LLMChain } from \"../chains/llm_chain.js\";\nimport type { BaseDocumentCompressor } from \"./document_compressors/index.js\";\n\ninterface LineList {\n  lines: string[];\n}\n\n// oxlint-disable-next-line @typescript-eslint/no-explicit-any\nexport type MultiDocs = Document<Record<string, any>>[];\n\nclass LineListOutputParser extends BaseOutputParser<LineList> {\n  static lc_name() {\n    return \"LineListOutputParser\";\n  }\n\n  lc_namespace = [\"langchain\", \"retrievers\", \"multiquery\"];\n\n  async parse(text: string): Promise<LineList> {\n    const startKeyIndex = text.indexOf(\"<questions>\");\n    const endKeyIndex = text.indexOf(\"</questions>\");\n    const questionsStartIndex =\n      startKeyIndex === -1 ? 0 : startKeyIndex + \"<questions>\".length;\n    const questionsEndIndex = endKeyIndex === -1 ? text.length : endKeyIndex;\n    const lines = text\n      .slice(questionsStartIndex, questionsEndIndex)\n      .trim()\n      .split(\"\\n\")\n      .filter((line) => line.trim() !== \"\");\n    return { lines };\n  }\n\n  getFormatInstructions(): string {\n    throw new Error(\"Not implemented.\");\n  }\n}\n\n// Create template\nconst DEFAULT_QUERY_PROMPT = /* #__PURE__ */ new PromptTemplate({\n  inputVariables: [\"question\", \"queryCount\"],\n  template: `You are an AI language model assistant. Your task is\nto generate {queryCount} different versions of the given user\nquestion to retrieve relevant documents from a vector database.\nBy generating multiple perspectives on the user question,\nyour goal is to help the user overcome some of the limitations\nof distance-based similarity search.\n\nProvide these alternative questions separated by newlines between XML tags. For example:\n\n<questions>\nQuestion 1\nQuestion 2\nQuestion 3\n</questions>\n\nOriginal question: {question}`,\n});\n\nexport interface MultiQueryRetrieverInput extends BaseRetrieverInput {\n  retriever: BaseRetrieverInterface;\n  /** @deprecated Pass a custom prompt into `.fromLLM` instead. */\n  llmChain: LLMChain<LineList>;\n  queryCount?: number;\n  parserKey?: string;\n  documentCompressor?: BaseDocumentCompressor | undefined;\n  documentCompressorFilteringFn?: (docs: MultiDocs) => MultiDocs;\n}\n\n/**\n * @example\n * ```typescript\n * const retriever = new MultiQueryRetriever.fromLLM({\n *   llm: new ChatAnthropic({}),\n *   retriever: new MemoryVectorStore().asRetriever(),\n *   verbose: true,\n * });\n * const retrievedDocs = await retriever.invoke(\n *   \"What are mitochondria made of?\",\n * );\n * ```\n */\nexport class MultiQueryRetriever extends BaseRetriever {\n  static lc_name() {\n    return \"MultiQueryRetriever\";\n  }\n\n  lc_namespace = [\"langchain\", \"retrievers\", \"multiquery\"];\n\n  private retriever: BaseRetrieverInterface;\n\n  private llmChain: LLMChain<LineList>;\n\n  private queryCount = 3;\n\n  private parserKey = \"lines\";\n\n  documentCompressor: BaseDocumentCompressor | undefined;\n\n  documentCompressorFilteringFn?: MultiQueryRetrieverInput[\"documentCompressorFilteringFn\"];\n\n  constructor(fields: MultiQueryRetrieverInput) {\n    super(fields);\n    this.retriever = fields.retriever;\n    this.llmChain = fields.llmChain;\n    this.queryCount = fields.queryCount ?? this.queryCount;\n    this.parserKey = fields.parserKey ?? this.parserKey;\n    this.documentCompressor = fields.documentCompressor;\n    this.documentCompressorFilteringFn = fields.documentCompressorFilteringFn;\n  }\n\n  static fromLLM(\n    fields: Omit<MultiQueryRetrieverInput, \"llmChain\"> & {\n      llm: BaseLanguageModelInterface;\n      prompt?: BasePromptTemplate;\n    }\n  ): MultiQueryRetriever {\n    const {\n      retriever,\n      llm,\n      prompt = DEFAULT_QUERY_PROMPT,\n      queryCount,\n      parserKey,\n      ...rest\n    } = fields;\n    const outputParser = new LineListOutputParser();\n    const llmChain = new LLMChain({ llm, prompt, outputParser });\n    return new this({ retriever, llmChain, queryCount, parserKey, ...rest });\n  }\n\n  // Generate the different queries for each retrieval, using our llmChain\n  private async _generateQueries(\n    question: string,\n    runManager?: CallbackManagerForRetrieverRun\n  ): Promise<string[]> {\n    const response = await this.llmChain.call(\n      { question, queryCount: this.queryCount },\n      runManager?.getChild()\n    );\n    const lines = response.text[this.parserKey] || [];\n    if (this.verbose) {\n      console.log(`Generated queries: ${lines}`);\n    }\n    return lines;\n  }\n\n  // Retrieve documents using the original retriever\n  private async _retrieveDocuments(\n    queries: string[],\n    runManager?: CallbackManagerForRetrieverRun\n  ): Promise<Document[]> {\n    const documents: Document[] = [];\n    await Promise.all(\n      queries.map(async (query) => {\n        const docs = await this.retriever.invoke(query, runManager?.getChild());\n        documents.push(...docs);\n      })\n    );\n    return documents;\n  }\n\n  // Deduplicate the documents that were returned in multiple retrievals\n  private _uniqueUnion(documents: Document[]): Document[] {\n    const uniqueDocumentsDict: { [key: string]: Document } = {};\n\n    for (const doc of documents) {\n      const key = `${doc.pageContent}:${JSON.stringify(\n        Object.entries(doc.metadata).sort()\n      )}`;\n      uniqueDocumentsDict[key] = doc;\n    }\n\n    const uniqueDocuments = Object.values(uniqueDocumentsDict);\n    return uniqueDocuments;\n  }\n\n  async _getRelevantDocuments(\n    question: string,\n    runManager?: CallbackManagerForRetrieverRun\n  ): Promise<Document[]> {\n    const queries = await this._generateQueries(question, runManager);\n    const documents = await this._retrieveDocuments(queries, runManager);\n    const uniqueDocuments = this._uniqueUnion(documents);\n\n    let outputDocs = uniqueDocuments;\n    if (this.documentCompressor && uniqueDocuments.length) {\n      outputDocs = await this.documentCompressor.compressDocuments(\n        uniqueDocuments,\n        question,\n        runManager?.getChild()\n      );\n      if (this.documentCompressorFilteringFn) {\n        outputDocs = this.documentCompressorFilteringFn(outputDocs);\n      }\n    }\n\n    return outputDocs;\n  }\n}\n"],"mappings":";;;;;;;AAoBA,IAAM,uBAAN,cAAmC,iBAA2B;CAC5D,OAAO,UAAU;AACf,SAAO;;CAGT,eAAe;EAAC;EAAa;EAAc;EAAa;CAExD,MAAM,MAAM,MAAiC;EAC3C,MAAM,gBAAgB,KAAK,QAAQ,cAAc;EACjD,MAAM,cAAc,KAAK,QAAQ,eAAe;EAChD,MAAM,sBACJ,kBAAkB,KAAK,IAAI,gBAAgB;EAC7C,MAAM,oBAAoB,gBAAgB,KAAK,KAAK,SAAS;AAM7D,SAAO,EAAE,OALK,KACX,MAAM,qBAAqB,kBAAkB,CAC7C,MAAM,CACN,MAAM,KAAK,CACX,QAAQ,SAAS,KAAK,MAAM,KAAK,GAAG,EACvB;;CAGlB,wBAAgC;AAC9B,QAAM,IAAI,MAAM,mBAAmB;;;AAKvC,MAAM,uCAAuC,IAAI,eAAe;CAC9D,gBAAgB,CAAC,YAAY,aAAa;CAC1C,UAAU;;;;;;;;;;;;;;;;CAgBX,CAAC;;;;;;;;;;;;;;AAyBF,IAAa,sBAAb,cAAyC,cAAc;CACrD,OAAO,UAAU;AACf,SAAO;;CAGT,eAAe;EAAC;EAAa;EAAc;EAAa;CAExD;CAEA;CAEA,aAAqB;CAErB,YAAoB;CAEpB;CAEA;CAEA,YAAY,QAAkC;AAC5C,QAAM,OAAO;AACb,OAAK,YAAY,OAAO;AACxB,OAAK,WAAW,OAAO;AACvB,OAAK,aAAa,OAAO,cAAc,KAAK;AAC5C,OAAK,YAAY,OAAO,aAAa,KAAK;AAC1C,OAAK,qBAAqB,OAAO;AACjC,OAAK,gCAAgC,OAAO;;CAG9C,OAAO,QACL,QAIqB;EACrB,MAAM,EACJ,WACA,KACA,SAAS,sBACT,YACA,WACA,GAAG,SACD;EAEJ,MAAM,WAAW,IAAI,SAAS;GAAE;GAAK;GAAQ,cADxB,IAAI,sBAAsB;GACY,CAAC;AAC5D,SAAO,IAAI,KAAK;GAAE;GAAW;GAAU;GAAY;GAAW,GAAG;GAAM,CAAC;;CAI1E,MAAc,iBACZ,UACA,YACmB;EAKnB,MAAM,SAJW,MAAM,KAAK,SAAS,KACnC;GAAE;GAAU,YAAY,KAAK;GAAY,EACzC,YAAY,UAAU,CACvB,EACsB,KAAK,KAAK,cAAc,EAAE;AACjD,MAAI,KAAK,QACP,SAAQ,IAAI,sBAAsB,QAAQ;AAE5C,SAAO;;CAIT,MAAc,mBACZ,SACA,YACqB;EACrB,MAAM,YAAwB,EAAE;AAChC,QAAM,QAAQ,IACZ,QAAQ,IAAI,OAAO,UAAU;GAC3B,MAAM,OAAO,MAAM,KAAK,UAAU,OAAO,OAAO,YAAY,UAAU,CAAC;AACvE,aAAU,KAAK,GAAG,KAAK;IACvB,CACH;AACD,SAAO;;CAIT,aAAqB,WAAmC;EACtD,MAAM,sBAAmD,EAAE;AAE3D,OAAK,MAAM,OAAO,WAAW;GAC3B,MAAM,MAAM,GAAG,IAAI,YAAY,GAAG,KAAK,UACrC,OAAO,QAAQ,IAAI,SAAS,CAAC,MAAM,CACpC;AACD,uBAAoB,OAAO;;AAI7B,SADwB,OAAO,OAAO,oBAAoB;;CAI5D,MAAM,sBACJ,UACA,YACqB;EACrB,MAAM,UAAU,MAAM,KAAK,iBAAiB,UAAU,WAAW;EACjE,MAAM,YAAY,MAAM,KAAK,mBAAmB,SAAS,WAAW;EACpE,MAAM,kBAAkB,KAAK,aAAa,UAAU;EAEpD,IAAI,aAAa;AACjB,MAAI,KAAK,sBAAsB,gBAAgB,QAAQ;AACrD,gBAAa,MAAM,KAAK,mBAAmB,kBACzC,iBACA,UACA,YAAY,UAAU,CACvB;AACD,OAAI,KAAK,8BACP,cAAa,KAAK,8BAA8B,WAAW;;AAI/D,SAAO"}

package/dist/retrievers/parent_document.cjs.map

@@ -1 +1 @@
-{"version":3,"file":"parent_document.cjs","names":["MultiVectorRetriever","Document"],"sources":["../../src/retrievers/parent_document.ts"],"sourcesContent":["import * as uuid from \"uuid\";\n\nimport {\n  type VectorStoreInterface,\n  type VectorStoreRetrieverInterface,\n} from \"@langchain/core/vectorstores\";\nimport { Document } from \"@langchain/core/documents\";\nimport {\n  TextSplitter,\n  TextSplitterChunkHeaderOptions,\n} from \"@langchain/textsplitters\";\nimport type { BaseDocumentCompressor } from \"./document_compressors/index.js\";\nimport {\n  MultiVectorRetriever,\n  type MultiVectorRetrieverInput,\n} from \"./multi_vector.js\";\n\n// eslint-disable-next-line @typescript-eslint/no-explicit-any\nexport type SubDocs = Document<Record<string, any>>[];\n\n/**\n * Interface for the fields required to initialize a\n * ParentDocumentRetriever instance.\n */\nexport type ParentDocumentRetrieverFields = MultiVectorRetrieverInput & {\n  childSplitter: TextSplitter;\n  parentSplitter?: TextSplitter;\n  /**\n   * A custom retriever to use when retrieving instead of\n   * the `.similaritySearch` method of the vectorstore.\n   */\n  childDocumentRetriever?: VectorStoreRetrieverInterface<VectorStoreInterface>;\n  documentCompressor?: BaseDocumentCompressor | undefined;\n  documentCompressorFilteringFn?: (docs: SubDocs) => SubDocs;\n};\n\n/**\n * A type of document retriever that splits input documents into smaller chunks\n * while separately storing and preserving the original documents.\n * The small chunks are embedded, then on retrieval, the original\n * \"parent\" documents are retrieved.\n *\n * This strikes a balance between better targeted retrieval with small documents\n * and the more context-rich larger documents.\n * @example\n * ```typescript\n * const retriever = new ParentDocumentRetriever({\n *   vectorstore: new MemoryVectorStore(new OpenAIEmbeddings()),\n *   byteStore: new InMemoryStore<Uint8Array>(),\n *   parentSplitter: new RecursiveCharacterTextSplitter({\n *     chunkOverlap: 0,\n *     chunkSize: 500,\n *   }),\n *   childSplitter: new RecursiveCharacterTextSplitter({\n *     chunkOverlap: 0,\n *     chunkSize: 50,\n *   }),\n *   childK: 20,\n *   parentK: 5,\n * });\n *\n * const parentDocuments = await getDocuments();\n * await retriever.addDocuments(parentDocuments);\n * const retrievedDocs = await retriever.invoke(\"justice breyer\");\n * ```\n */\nexport class ParentDocumentRetriever extends MultiVectorRetriever {\n  static lc_name() {\n    return \"ParentDocumentRetriever\";\n  }\n\n  lc_namespace = [\"langchain\", \"retrievers\", \"parent_document\"];\n\n  vectorstore: VectorStoreInterface;\n\n  protected childSplitter: TextSplitter;\n\n  protected parentSplitter?: TextSplitter;\n\n  protected idKey = \"doc_id\";\n\n  protected childK?: number;\n\n  protected parentK?: number;\n\n  childDocumentRetriever:\n    | VectorStoreRetrieverInterface<VectorStoreInterface>\n    | undefined;\n\n  documentCompressor: BaseDocumentCompressor | undefined;\n\n  documentCompressorFilteringFn?: ParentDocumentRetrieverFields[\"documentCompressorFilteringFn\"];\n\n  constructor(fields: ParentDocumentRetrieverFields) {\n    super(fields);\n    this.vectorstore = fields.vectorstore;\n    this.childSplitter = fields.childSplitter;\n    this.parentSplitter = fields.parentSplitter;\n    this.idKey = fields.idKey ?? this.idKey;\n    this.childK = fields.childK;\n    this.parentK = fields.parentK;\n    this.childDocumentRetriever = fields.childDocumentRetriever;\n    this.documentCompressor = fields.documentCompressor;\n    this.documentCompressorFilteringFn = fields.documentCompressorFilteringFn;\n  }\n\n  async _getRelevantDocuments(query: string): Promise<Document[]> {\n    let subDocs: SubDocs = [];\n    if (this.childDocumentRetriever) {\n      subDocs = await this.childDocumentRetriever.invoke(query);\n    } else {\n      subDocs = await this.vectorstore.similaritySearch(query, this.childK);\n    }\n\n    if (this.documentCompressor && subDocs.length) {\n      subDocs = await this.documentCompressor.compressDocuments(subDocs, query);\n      if (this.documentCompressorFilteringFn) {\n        subDocs = this.documentCompressorFilteringFn(subDocs);\n      }\n    }\n\n    // Maintain order\n    const parentDocIds: string[] = [];\n    for (const doc of subDocs) {\n      if (!parentDocIds.includes(doc.metadata[this.idKey])) {\n        parentDocIds.push(doc.metadata[this.idKey]);\n      }\n    }\n    const parentDocs: Document[] = [];\n    const storedParentDocs = await this.docstore.mget(parentDocIds);\n    const retrievedDocs: Document[] = storedParentDocs.filter(\n      (doc?: Document): doc is Document => doc !== undefined\n    );\n    parentDocs.push(...retrievedDocs);\n    return parentDocs.slice(0, this.parentK);\n  }\n\n  async _storeDocuments(\n    parentDoc: Record<string, Document>,\n    childDocs: Document[],\n    addToDocstore: boolean\n  ) {\n    if (this.childDocumentRetriever) {\n      await this.childDocumentRetriever.addDocuments(childDocs);\n    } else {\n      await this.vectorstore.addDocuments(childDocs);\n    }\n    if (addToDocstore) {\n      await this.docstore.mset(Object.entries(parentDoc));\n    }\n  }\n\n  /**\n   * Adds documents to the docstore and vectorstores.\n   * If a retriever is provided, it will be used to add documents instead of the vectorstore.\n   * @param docs The documents to add\n   * @param config.ids Optional list of ids for documents. If provided should be the same\n   *   length as the list of documents. Can provided if parent documents\n   *   are already in the document store and you don't want to re-add\n   *   to the docstore. If not provided, random UUIDs will be used as ids.\n   * @param config.addToDocstore Boolean of whether to add documents to docstore.\n   * This can be false if and only if `ids` are provided. You may want\n   *   to set this to False if the documents are already in the docstore\n   *   and you don't want to re-add them.\n   * @param config.chunkHeaderOptions Object with options for adding Contextual chunk headers\n   */\n  async addDocuments(\n    docs: Document[],\n    config?: {\n      ids?: string[];\n      addToDocstore?: boolean;\n      childDocChunkHeaderOptions?: TextSplitterChunkHeaderOptions;\n    }\n  ): Promise<void> {\n    const {\n      ids,\n      addToDocstore = true,\n      childDocChunkHeaderOptions = {},\n    } = config ?? {};\n    const parentDocs = this.parentSplitter\n      ? await this.parentSplitter.splitDocuments(docs)\n      : docs;\n    let parentDocIds;\n    if (ids === undefined) {\n      if (!addToDocstore) {\n        throw new Error(\n          `If ids are not passed in, \"config.addToDocstore\" MUST be true`\n        );\n      }\n      parentDocIds = parentDocs.map((_doc: Document) => uuid.v4());\n    } else {\n      parentDocIds = ids;\n    }\n    if (parentDocs.length !== parentDocIds.length) {\n      throw new Error(\n        `Got uneven list of documents and ids.\\nIf \"ids\" is provided, should be same length as \"documents\".`\n      );\n    }\n    for (let i = 0; i < parentDocs.length; i += 1) {\n      const parentDoc = parentDocs[i];\n      const parentDocId = parentDocIds[i];\n      const subDocs = await this.childSplitter.splitDocuments(\n        [parentDoc],\n        childDocChunkHeaderOptions\n      );\n      const taggedSubDocs = subDocs.map(\n        (subDoc: Document) =>\n          new Document({\n            pageContent: subDoc.pageContent,\n            metadata: { ...subDoc.metadata, [this.idKey]: parentDocId },\n          })\n      );\n      await this._storeDocuments(\n        { [parentDocId]: parentDoc },\n        taggedSubDocs,\n        addToDocstore\n      );\n    }\n  }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAkEA,IAAa,0BAAb,cAA6CA,gCAAAA,qBAAqB;CAChE,OAAO,UAAU;AACf,SAAO;;CAGT,eAAe;EAAC;EAAa;EAAc;EAAkB;CAE7D;CAEA;CAEA;CAEA,QAAkB;CAElB;CAEA;CAEA;CAIA;CAEA;CAEA,YAAY,QAAuC;AACjD,QAAM,OAAO;AACb,OAAK,cAAc,OAAO;AAC1B,OAAK,gBAAgB,OAAO;AAC5B,OAAK,iBAAiB,OAAO;AAC7B,OAAK,QAAQ,OAAO,SAAS,KAAK;AAClC,OAAK,SAAS,OAAO;AACrB,OAAK,UAAU,OAAO;AACtB,OAAK,yBAAyB,OAAO;AACrC,OAAK,qBAAqB,OAAO;AACjC,OAAK,gCAAgC,OAAO;;CAG9C,MAAM,sBAAsB,OAAoC;EAC9D,IAAI,UAAmB,EAAE;AACzB,MAAI,KAAK,uBACP,WAAU,MAAM,KAAK,uBAAuB,OAAO,MAAM;MAEzD,WAAU,MAAM,KAAK,YAAY,iBAAiB,OAAO,KAAK,OAAO;AAGvE,MAAI,KAAK,sBAAsB,QAAQ,QAAQ;AAC7C,aAAU,MAAM,KAAK,mBAAmB,kBAAkB,SAAS,MAAM;AACzE,OAAI,KAAK,8BACP,WAAU,KAAK,8BAA8B,QAAQ;;EAKzD,MAAM,eAAyB,EAAE;AACjC,OAAK,MAAM,OAAO,QAChB,KAAI,CAAC,aAAa,SAAS,IAAI,SAAS,KAAK,OAAO,CAClD,cAAa,KAAK,IAAI,SAAS,KAAK,OAAO;EAG/C,MAAM,aAAyB,EAAE;EAEjC,MAAM,iBADmB,MAAM,KAAK,SAAS,KAAK,aAAa,EACZ,QAChD,QAAoC,QAAQ,KAAA,EAC9C;AACD,aAAW,KAAK,GAAG,cAAc;AACjC,SAAO,WAAW,MAAM,GAAG,KAAK,QAAQ;;CAG1C,MAAM,gBACJ,WACA,WACA,eACA;AACA,MAAI,KAAK,uBACP,OAAM,KAAK,uBAAuB,aAAa,UAAU;MAEzD,OAAM,KAAK,YAAY,aAAa,UAAU;AAEhD,MAAI,cACF,OAAM,KAAK,SAAS,KAAK,OAAO,QAAQ,UAAU,CAAC;;;;;;;;;;;;;;;;CAkBvD,MAAM,aACJ,MACA,QAKe;EACf,MAAM,EACJ,KACA,gBAAgB,MAChB,6BAA6B,EAAE,KAC7B,UAAU,EAAE;EAChB,MAAM,aAAa,KAAK,iBACpB,MAAM,KAAK,eAAe,eAAe,KAAK,GAC9C;EACJ,IAAI;AACJ,MAAI,QAAQ,KAAA,GAAW;AACrB,OAAI,CAAC,cACH,OAAM,IAAI,MACR,gEACD;AAEH,kBAAe,WAAW,KAAK,SAAmB,KAAK,IAAI,CAAC;QAE5D,gBAAe;AAEjB,MAAI,WAAW,WAAW,aAAa,OACrC,OAAM,IAAI,MACR,qGACD;AAEH,OAAK,IAAI,IAAI,GAAG,IAAI,WAAW,QAAQ,KAAK,GAAG;GAC7C,MAAM,YAAY,WAAW;GAC7B,MAAM,cAAc,aAAa;GAKjC,MAAM,iBAJU,MAAM,KAAK,cAAc,eACvC,CAAC,UAAU,EACX,2BACD,EAC6B,KAC3B,WACC,IAAIC,0BAAAA,SAAS;IACX,aAAa,OAAO;IACpB,UAAU;KAAE,GAAG,OAAO;MAAW,KAAK,QAAQ;KAAa;IAC5D,CAAC,CACL;AACD,SAAM,KAAK,gBACT,GAAG,cAAc,WAAW,EAC5B,eACA,cACD"}
+{"version":3,"file":"parent_document.cjs","names":["MultiVectorRetriever","Document"],"sources":["../../src/retrievers/parent_document.ts"],"sourcesContent":["import * as uuid from \"uuid\";\n\nimport {\n  type VectorStoreInterface,\n  type VectorStoreRetrieverInterface,\n} from \"@langchain/core/vectorstores\";\nimport { Document } from \"@langchain/core/documents\";\nimport {\n  TextSplitter,\n  TextSplitterChunkHeaderOptions,\n} from \"@langchain/textsplitters\";\nimport type { BaseDocumentCompressor } from \"./document_compressors/index.js\";\nimport {\n  MultiVectorRetriever,\n  type MultiVectorRetrieverInput,\n} from \"./multi_vector.js\";\n\n// oxlint-disable-next-line @typescript-eslint/no-explicit-any\nexport type SubDocs = Document<Record<string, any>>[];\n\n/**\n * Interface for the fields required to initialize a\n * ParentDocumentRetriever instance.\n */\nexport type ParentDocumentRetrieverFields = MultiVectorRetrieverInput & {\n  childSplitter: TextSplitter;\n  parentSplitter?: TextSplitter;\n  /**\n   * A custom retriever to use when retrieving instead of\n   * the `.similaritySearch` method of the vectorstore.\n   */\n  childDocumentRetriever?: VectorStoreRetrieverInterface<VectorStoreInterface>;\n  documentCompressor?: BaseDocumentCompressor | undefined;\n  documentCompressorFilteringFn?: (docs: SubDocs) => SubDocs;\n};\n\n/**\n * A type of document retriever that splits input documents into smaller chunks\n * while separately storing and preserving the original documents.\n * The small chunks are embedded, then on retrieval, the original\n * \"parent\" documents are retrieved.\n *\n * This strikes a balance between better targeted retrieval with small documents\n * and the more context-rich larger documents.\n * @example\n * ```typescript\n * const retriever = new ParentDocumentRetriever({\n *   vectorstore: new MemoryVectorStore(new OpenAIEmbeddings()),\n *   byteStore: new InMemoryStore<Uint8Array>(),\n *   parentSplitter: new RecursiveCharacterTextSplitter({\n *     chunkOverlap: 0,\n *     chunkSize: 500,\n *   }),\n *   childSplitter: new RecursiveCharacterTextSplitter({\n *     chunkOverlap: 0,\n *     chunkSize: 50,\n *   }),\n *   childK: 20,\n *   parentK: 5,\n * });\n *\n * const parentDocuments = await getDocuments();\n * await retriever.addDocuments(parentDocuments);\n * const retrievedDocs = await retriever.invoke(\"justice breyer\");\n * ```\n */\nexport class ParentDocumentRetriever extends MultiVectorRetriever {\n  static lc_name() {\n    return \"ParentDocumentRetriever\";\n  }\n\n  lc_namespace = [\"langchain\", \"retrievers\", \"parent_document\"];\n\n  vectorstore: VectorStoreInterface;\n\n  protected childSplitter: TextSplitter;\n\n  protected parentSplitter?: TextSplitter;\n\n  protected idKey = \"doc_id\";\n\n  protected childK?: number;\n\n  protected parentK?: number;\n\n  childDocumentRetriever:\n    | VectorStoreRetrieverInterface<VectorStoreInterface>\n    | undefined;\n\n  documentCompressor: BaseDocumentCompressor | undefined;\n\n  documentCompressorFilteringFn?: ParentDocumentRetrieverFields[\"documentCompressorFilteringFn\"];\n\n  constructor(fields: ParentDocumentRetrieverFields) {\n    super(fields);\n    this.vectorstore = fields.vectorstore;\n    this.childSplitter = fields.childSplitter;\n    this.parentSplitter = fields.parentSplitter;\n    this.idKey = fields.idKey ?? this.idKey;\n    this.childK = fields.childK;\n    this.parentK = fields.parentK;\n    this.childDocumentRetriever = fields.childDocumentRetriever;\n    this.documentCompressor = fields.documentCompressor;\n    this.documentCompressorFilteringFn = fields.documentCompressorFilteringFn;\n  }\n\n  async _getRelevantDocuments(query: string): Promise<Document[]> {\n    let subDocs: SubDocs = [];\n    if (this.childDocumentRetriever) {\n      subDocs = await this.childDocumentRetriever.invoke(query);\n    } else {\n      subDocs = await this.vectorstore.similaritySearch(query, this.childK);\n    }\n\n    if (this.documentCompressor && subDocs.length) {\n      subDocs = await this.documentCompressor.compressDocuments(subDocs, query);\n      if (this.documentCompressorFilteringFn) {\n        subDocs = this.documentCompressorFilteringFn(subDocs);\n      }\n    }\n\n    // Maintain order\n    const parentDocIds: string[] = [];\n    for (const doc of subDocs) {\n      if (!parentDocIds.includes(doc.metadata[this.idKey])) {\n        parentDocIds.push(doc.metadata[this.idKey]);\n      }\n    }\n    const parentDocs: Document[] = [];\n    const storedParentDocs = await this.docstore.mget(parentDocIds);\n    const retrievedDocs: Document[] = storedParentDocs.filter(\n      (doc?: Document): doc is Document => doc !== undefined\n    );\n    parentDocs.push(...retrievedDocs);\n    return parentDocs.slice(0, this.parentK);\n  }\n\n  async _storeDocuments(\n    parentDoc: Record<string, Document>,\n    childDocs: Document[],\n    addToDocstore: boolean\n  ) {\n    if (this.childDocumentRetriever) {\n      await this.childDocumentRetriever.addDocuments(childDocs);\n    } else {\n      await this.vectorstore.addDocuments(childDocs);\n    }\n    if (addToDocstore) {\n      await this.docstore.mset(Object.entries(parentDoc));\n    }\n  }\n\n  /**\n   * Adds documents to the docstore and vectorstores.\n   * If a retriever is provided, it will be used to add documents instead of the vectorstore.\n   * @param docs The documents to add\n   * @param config.ids Optional list of ids for documents. If provided should be the same\n   *   length as the list of documents. Can provided if parent documents\n   *   are already in the document store and you don't want to re-add\n   *   to the docstore. If not provided, random UUIDs will be used as ids.\n   * @param config.addToDocstore Boolean of whether to add documents to docstore.\n   * This can be false if and only if `ids` are provided. You may want\n   *   to set this to False if the documents are already in the docstore\n   *   and you don't want to re-add them.\n   * @param config.chunkHeaderOptions Object with options for adding Contextual chunk headers\n   */\n  async addDocuments(\n    docs: Document[],\n    config?: {\n      ids?: string[];\n      addToDocstore?: boolean;\n      childDocChunkHeaderOptions?: TextSplitterChunkHeaderOptions;\n    }\n  ): Promise<void> {\n    const {\n      ids,\n      addToDocstore = true,\n      childDocChunkHeaderOptions = {},\n    } = config ?? {};\n    const parentDocs = this.parentSplitter\n      ? await this.parentSplitter.splitDocuments(docs)\n      : docs;\n    let parentDocIds;\n    if (ids === undefined) {\n      if (!addToDocstore) {\n        throw new Error(\n          `If ids are not passed in, \"config.addToDocstore\" MUST be true`\n        );\n      }\n      parentDocIds = parentDocs.map((_doc: Document) => uuid.v4());\n    } else {\n      parentDocIds = ids;\n    }\n    if (parentDocs.length !== parentDocIds.length) {\n      throw new Error(\n        `Got uneven list of documents and ids.\\nIf \"ids\" is provided, should be same length as \"documents\".`\n      );\n    }\n    for (let i = 0; i < parentDocs.length; i += 1) {\n      const parentDoc = parentDocs[i];\n      const parentDocId = parentDocIds[i];\n      const subDocs = await this.childSplitter.splitDocuments(\n        [parentDoc],\n        childDocChunkHeaderOptions\n      );\n      const taggedSubDocs = subDocs.map(\n        (subDoc: Document) =>\n          new Document({\n            pageContent: subDoc.pageContent,\n            metadata: { ...subDoc.metadata, [this.idKey]: parentDocId },\n          })\n      );\n      await this._storeDocuments(\n        { [parentDocId]: parentDoc },\n        taggedSubDocs,\n        addToDocstore\n      );\n    }\n  }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAkEA,IAAa,0BAAb,cAA6CA,gCAAAA,qBAAqB;CAChE,OAAO,UAAU;AACf,SAAO;;CAGT,eAAe;EAAC;EAAa;EAAc;EAAkB;CAE7D;CAEA;CAEA;CAEA,QAAkB;CAElB;CAEA;CAEA;CAIA;CAEA;CAEA,YAAY,QAAuC;AACjD,QAAM,OAAO;AACb,OAAK,cAAc,OAAO;AAC1B,OAAK,gBAAgB,OAAO;AAC5B,OAAK,iBAAiB,OAAO;AAC7B,OAAK,QAAQ,OAAO,SAAS,KAAK;AAClC,OAAK,SAAS,OAAO;AACrB,OAAK,UAAU,OAAO;AACtB,OAAK,yBAAyB,OAAO;AACrC,OAAK,qBAAqB,OAAO;AACjC,OAAK,gCAAgC,OAAO;;CAG9C,MAAM,sBAAsB,OAAoC;EAC9D,IAAI,UAAmB,EAAE;AACzB,MAAI,KAAK,uBACP,WAAU,MAAM,KAAK,uBAAuB,OAAO,MAAM;MAEzD,WAAU,MAAM,KAAK,YAAY,iBAAiB,OAAO,KAAK,OAAO;AAGvE,MAAI,KAAK,sBAAsB,QAAQ,QAAQ;AAC7C,aAAU,MAAM,KAAK,mBAAmB,kBAAkB,SAAS,MAAM;AACzE,OAAI,KAAK,8BACP,WAAU,KAAK,8BAA8B,QAAQ;;EAKzD,MAAM,eAAyB,EAAE;AACjC,OAAK,MAAM,OAAO,QAChB,KAAI,CAAC,aAAa,SAAS,IAAI,SAAS,KAAK,OAAO,CAClD,cAAa,KAAK,IAAI,SAAS,KAAK,OAAO;EAG/C,MAAM,aAAyB,EAAE;EAEjC,MAAM,iBADmB,MAAM,KAAK,SAAS,KAAK,aAAa,EACZ,QAChD,QAAoC,QAAQ,KAAA,EAC9C;AACD,aAAW,KAAK,GAAG,cAAc;AACjC,SAAO,WAAW,MAAM,GAAG,KAAK,QAAQ;;CAG1C,MAAM,gBACJ,WACA,WACA,eACA;AACA,MAAI,KAAK,uBACP,OAAM,KAAK,uBAAuB,aAAa,UAAU;MAEzD,OAAM,KAAK,YAAY,aAAa,UAAU;AAEhD,MAAI,cACF,OAAM,KAAK,SAAS,KAAK,OAAO,QAAQ,UAAU,CAAC;;;;;;;;;;;;;;;;CAkBvD,MAAM,aACJ,MACA,QAKe;EACf,MAAM,EACJ,KACA,gBAAgB,MAChB,6BAA6B,EAAE,KAC7B,UAAU,EAAE;EAChB,MAAM,aAAa,KAAK,iBACpB,MAAM,KAAK,eAAe,eAAe,KAAK,GAC9C;EACJ,IAAI;AACJ,MAAI,QAAQ,KAAA,GAAW;AACrB,OAAI,CAAC,cACH,OAAM,IAAI,MACR,gEACD;AAEH,kBAAe,WAAW,KAAK,SAAmB,KAAK,IAAI,CAAC;QAE5D,gBAAe;AAEjB,MAAI,WAAW,WAAW,aAAa,OACrC,OAAM,IAAI,MACR,qGACD;AAEH,OAAK,IAAI,IAAI,GAAG,IAAI,WAAW,QAAQ,KAAK,GAAG;GAC7C,MAAM,YAAY,WAAW;GAC7B,MAAM,cAAc,aAAa;GAKjC,MAAM,iBAJU,MAAM,KAAK,cAAc,eACvC,CAAC,UAAU,EACX,2BACD,EAC6B,KAC3B,WACC,IAAIC,0BAAAA,SAAS;IACX,aAAa,OAAO;IACpB,UAAU;KAAE,GAAG,OAAO;MAAW,KAAK,QAAQ;KAAa;IAC5D,CAAC,CACL;AACD,SAAM,KAAK,gBACT,GAAG,cAAc,WAAW,EAC5B,eACA,cACD"}

package/dist/retrievers/parent_document.js.map

@@ -1 +1 @@
-{"version":3,"file":"parent_document.js","names":[],"sources":["../../src/retrievers/parent_document.ts"],"sourcesContent":["import * as uuid from \"uuid\";\n\nimport {\n  type VectorStoreInterface,\n  type VectorStoreRetrieverInterface,\n} from \"@langchain/core/vectorstores\";\nimport { Document } from \"@langchain/core/documents\";\nimport {\n  TextSplitter,\n  TextSplitterChunkHeaderOptions,\n} from \"@langchain/textsplitters\";\nimport type { BaseDocumentCompressor } from \"./document_compressors/index.js\";\nimport {\n  MultiVectorRetriever,\n  type MultiVectorRetrieverInput,\n} from \"./multi_vector.js\";\n\n// eslint-disable-next-line @typescript-eslint/no-explicit-any\nexport type SubDocs = Document<Record<string, any>>[];\n\n/**\n * Interface for the fields required to initialize a\n * ParentDocumentRetriever instance.\n */\nexport type ParentDocumentRetrieverFields = MultiVectorRetrieverInput & {\n  childSplitter: TextSplitter;\n  parentSplitter?: TextSplitter;\n  /**\n   * A custom retriever to use when retrieving instead of\n   * the `.similaritySearch` method of the vectorstore.\n   */\n  childDocumentRetriever?: VectorStoreRetrieverInterface<VectorStoreInterface>;\n  documentCompressor?: BaseDocumentCompressor | undefined;\n  documentCompressorFilteringFn?: (docs: SubDocs) => SubDocs;\n};\n\n/**\n * A type of document retriever that splits input documents into smaller chunks\n * while separately storing and preserving the original documents.\n * The small chunks are embedded, then on retrieval, the original\n * \"parent\" documents are retrieved.\n *\n * This strikes a balance between better targeted retrieval with small documents\n * and the more context-rich larger documents.\n * @example\n * ```typescript\n * const retriever = new ParentDocumentRetriever({\n *   vectorstore: new MemoryVectorStore(new OpenAIEmbeddings()),\n *   byteStore: new InMemoryStore<Uint8Array>(),\n *   parentSplitter: new RecursiveCharacterTextSplitter({\n *     chunkOverlap: 0,\n *     chunkSize: 500,\n *   }),\n *   childSplitter: new RecursiveCharacterTextSplitter({\n *     chunkOverlap: 0,\n *     chunkSize: 50,\n *   }),\n *   childK: 20,\n *   parentK: 5,\n * });\n *\n * const parentDocuments = await getDocuments();\n * await retriever.addDocuments(parentDocuments);\n * const retrievedDocs = await retriever.invoke(\"justice breyer\");\n * ```\n */\nexport class ParentDocumentRetriever extends MultiVectorRetriever {\n  static lc_name() {\n    return \"ParentDocumentRetriever\";\n  }\n\n  lc_namespace = [\"langchain\", \"retrievers\", \"parent_document\"];\n\n  vectorstore: VectorStoreInterface;\n\n  protected childSplitter: TextSplitter;\n\n  protected parentSplitter?: TextSplitter;\n\n  protected idKey = \"doc_id\";\n\n  protected childK?: number;\n\n  protected parentK?: number;\n\n  childDocumentRetriever:\n    | VectorStoreRetrieverInterface<VectorStoreInterface>\n    | undefined;\n\n  documentCompressor: BaseDocumentCompressor | undefined;\n\n  documentCompressorFilteringFn?: ParentDocumentRetrieverFields[\"documentCompressorFilteringFn\"];\n\n  constructor(fields: ParentDocumentRetrieverFields) {\n    super(fields);\n    this.vectorstore = fields.vectorstore;\n    this.childSplitter = fields.childSplitter;\n    this.parentSplitter = fields.parentSplitter;\n    this.idKey = fields.idKey ?? this.idKey;\n    this.childK = fields.childK;\n    this.parentK = fields.parentK;\n    this.childDocumentRetriever = fields.childDocumentRetriever;\n    this.documentCompressor = fields.documentCompressor;\n    this.documentCompressorFilteringFn = fields.documentCompressorFilteringFn;\n  }\n\n  async _getRelevantDocuments(query: string): Promise<Document[]> {\n    let subDocs: SubDocs = [];\n    if (this.childDocumentRetriever) {\n      subDocs = await this.childDocumentRetriever.invoke(query);\n    } else {\n      subDocs = await this.vectorstore.similaritySearch(query, this.childK);\n    }\n\n    if (this.documentCompressor && subDocs.length) {\n      subDocs = await this.documentCompressor.compressDocuments(subDocs, query);\n      if (this.documentCompressorFilteringFn) {\n        subDocs = this.documentCompressorFilteringFn(subDocs);\n      }\n    }\n\n    // Maintain order\n    const parentDocIds: string[] = [];\n    for (const doc of subDocs) {\n      if (!parentDocIds.includes(doc.metadata[this.idKey])) {\n        parentDocIds.push(doc.metadata[this.idKey]);\n      }\n    }\n    const parentDocs: Document[] = [];\n    const storedParentDocs = await this.docstore.mget(parentDocIds);\n    const retrievedDocs: Document[] = storedParentDocs.filter(\n      (doc?: Document): doc is Document => doc !== undefined\n    );\n    parentDocs.push(...retrievedDocs);\n    return parentDocs.slice(0, this.parentK);\n  }\n\n  async _storeDocuments(\n    parentDoc: Record<string, Document>,\n    childDocs: Document[],\n    addToDocstore: boolean\n  ) {\n    if (this.childDocumentRetriever) {\n      await this.childDocumentRetriever.addDocuments(childDocs);\n    } else {\n      await this.vectorstore.addDocuments(childDocs);\n    }\n    if (addToDocstore) {\n      await this.docstore.mset(Object.entries(parentDoc));\n    }\n  }\n\n  /**\n   * Adds documents to the docstore and vectorstores.\n   * If a retriever is provided, it will be used to add documents instead of the vectorstore.\n   * @param docs The documents to add\n   * @param config.ids Optional list of ids for documents. If provided should be the same\n   *   length as the list of documents. Can provided if parent documents\n   *   are already in the document store and you don't want to re-add\n   *   to the docstore. If not provided, random UUIDs will be used as ids.\n   * @param config.addToDocstore Boolean of whether to add documents to docstore.\n   * This can be false if and only if `ids` are provided. You may want\n   *   to set this to False if the documents are already in the docstore\n   *   and you don't want to re-add them.\n   * @param config.chunkHeaderOptions Object with options for adding Contextual chunk headers\n   */\n  async addDocuments(\n    docs: Document[],\n    config?: {\n      ids?: string[];\n      addToDocstore?: boolean;\n      childDocChunkHeaderOptions?: TextSplitterChunkHeaderOptions;\n    }\n  ): Promise<void> {\n    const {\n      ids,\n      addToDocstore = true,\n      childDocChunkHeaderOptions = {},\n    } = config ?? {};\n    const parentDocs = this.parentSplitter\n      ? await this.parentSplitter.splitDocuments(docs)\n      : docs;\n    let parentDocIds;\n    if (ids === undefined) {\n      if (!addToDocstore) {\n        throw new Error(\n          `If ids are not passed in, \"config.addToDocstore\" MUST be true`\n        );\n      }\n      parentDocIds = parentDocs.map((_doc: Document) => uuid.v4());\n    } else {\n      parentDocIds = ids;\n    }\n    if (parentDocs.length !== parentDocIds.length) {\n      throw new Error(\n        `Got uneven list of documents and ids.\\nIf \"ids\" is provided, should be same length as \"documents\".`\n      );\n    }\n    for (let i = 0; i < parentDocs.length; i += 1) {\n      const parentDoc = parentDocs[i];\n      const parentDocId = parentDocIds[i];\n      const subDocs = await this.childSplitter.splitDocuments(\n        [parentDoc],\n        childDocChunkHeaderOptions\n      );\n      const taggedSubDocs = subDocs.map(\n        (subDoc: Document) =>\n          new Document({\n            pageContent: subDoc.pageContent,\n            metadata: { ...subDoc.metadata, [this.idKey]: parentDocId },\n          })\n      );\n      await this._storeDocuments(\n        { [parentDocId]: parentDoc },\n        taggedSubDocs,\n        addToDocstore\n      );\n    }\n  }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAkEA,IAAa,0BAAb,cAA6C,qBAAqB;CAChE,OAAO,UAAU;AACf,SAAO;;CAGT,eAAe;EAAC;EAAa;EAAc;EAAkB;CAE7D;CAEA;CAEA;CAEA,QAAkB;CAElB;CAEA;CAEA;CAIA;CAEA;CAEA,YAAY,QAAuC;AACjD,QAAM,OAAO;AACb,OAAK,cAAc,OAAO;AAC1B,OAAK,gBAAgB,OAAO;AAC5B,OAAK,iBAAiB,OAAO;AAC7B,OAAK,QAAQ,OAAO,SAAS,KAAK;AAClC,OAAK,SAAS,OAAO;AACrB,OAAK,UAAU,OAAO;AACtB,OAAK,yBAAyB,OAAO;AACrC,OAAK,qBAAqB,OAAO;AACjC,OAAK,gCAAgC,OAAO;;CAG9C,MAAM,sBAAsB,OAAoC;EAC9D,IAAI,UAAmB,EAAE;AACzB,MAAI,KAAK,uBACP,WAAU,MAAM,KAAK,uBAAuB,OAAO,MAAM;MAEzD,WAAU,MAAM,KAAK,YAAY,iBAAiB,OAAO,KAAK,OAAO;AAGvE,MAAI,KAAK,sBAAsB,QAAQ,QAAQ;AAC7C,aAAU,MAAM,KAAK,mBAAmB,kBAAkB,SAAS,MAAM;AACzE,OAAI,KAAK,8BACP,WAAU,KAAK,8BAA8B,QAAQ;;EAKzD,MAAM,eAAyB,EAAE;AACjC,OAAK,MAAM,OAAO,QAChB,KAAI,CAAC,aAAa,SAAS,IAAI,SAAS,KAAK,OAAO,CAClD,cAAa,KAAK,IAAI,SAAS,KAAK,OAAO;EAG/C,MAAM,aAAyB,EAAE;EAEjC,MAAM,iBADmB,MAAM,KAAK,SAAS,KAAK,aAAa,EACZ,QAChD,QAAoC,QAAQ,KAAA,EAC9C;AACD,aAAW,KAAK,GAAG,cAAc;AACjC,SAAO,WAAW,MAAM,GAAG,KAAK,QAAQ;;CAG1C,MAAM,gBACJ,WACA,WACA,eACA;AACA,MAAI,KAAK,uBACP,OAAM,KAAK,uBAAuB,aAAa,UAAU;MAEzD,OAAM,KAAK,YAAY,aAAa,UAAU;AAEhD,MAAI,cACF,OAAM,KAAK,SAAS,KAAK,OAAO,QAAQ,UAAU,CAAC;;;;;;;;;;;;;;;;CAkBvD,MAAM,aACJ,MACA,QAKe;EACf,MAAM,EACJ,KACA,gBAAgB,MAChB,6BAA6B,EAAE,KAC7B,UAAU,EAAE;EAChB,MAAM,aAAa,KAAK,iBACpB,MAAM,KAAK,eAAe,eAAe,KAAK,GAC9C;EACJ,IAAI;AACJ,MAAI,QAAQ,KAAA,GAAW;AACrB,OAAI,CAAC,cACH,OAAM,IAAI,MACR,gEACD;AAEH,kBAAe,WAAW,KAAK,SAAmB,KAAK,IAAI,CAAC;QAE5D,gBAAe;AAEjB,MAAI,WAAW,WAAW,aAAa,OACrC,OAAM,IAAI,MACR,qGACD;AAEH,OAAK,IAAI,IAAI,GAAG,IAAI,WAAW,QAAQ,KAAK,GAAG;GAC7C,MAAM,YAAY,WAAW;GAC7B,MAAM,cAAc,aAAa;GAKjC,MAAM,iBAJU,MAAM,KAAK,cAAc,eACvC,CAAC,UAAU,EACX,2BACD,EAC6B,KAC3B,WACC,IAAI,SAAS;IACX,aAAa,OAAO;IACpB,UAAU;KAAE,GAAG,OAAO;MAAW,KAAK,QAAQ;KAAa;IAC5D,CAAC,CACL;AACD,SAAM,KAAK,gBACT,GAAG,cAAc,WAAW,EAC5B,eACA,cACD"}
+{"version":3,"file":"parent_document.js","names":[],"sources":["../../src/retrievers/parent_document.ts"],"sourcesContent":["import * as uuid from \"uuid\";\n\nimport {\n  type VectorStoreInterface,\n  type VectorStoreRetrieverInterface,\n} from \"@langchain/core/vectorstores\";\nimport { Document } from \"@langchain/core/documents\";\nimport {\n  TextSplitter,\n  TextSplitterChunkHeaderOptions,\n} from \"@langchain/textsplitters\";\nimport type { BaseDocumentCompressor } from \"./document_compressors/index.js\";\nimport {\n  MultiVectorRetriever,\n  type MultiVectorRetrieverInput,\n} from \"./multi_vector.js\";\n\n// oxlint-disable-next-line @typescript-eslint/no-explicit-any\nexport type SubDocs = Document<Record<string, any>>[];\n\n/**\n * Interface for the fields required to initialize a\n * ParentDocumentRetriever instance.\n */\nexport type ParentDocumentRetrieverFields = MultiVectorRetrieverInput & {\n  childSplitter: TextSplitter;\n  parentSplitter?: TextSplitter;\n  /**\n   * A custom retriever to use when retrieving instead of\n   * the `.similaritySearch` method of the vectorstore.\n   */\n  childDocumentRetriever?: VectorStoreRetrieverInterface<VectorStoreInterface>;\n  documentCompressor?: BaseDocumentCompressor | undefined;\n  documentCompressorFilteringFn?: (docs: SubDocs) => SubDocs;\n};\n\n/**\n * A type of document retriever that splits input documents into smaller chunks\n * while separately storing and preserving the original documents.\n * The small chunks are embedded, then on retrieval, the original\n * \"parent\" documents are retrieved.\n *\n * This strikes a balance between better targeted retrieval with small documents\n * and the more context-rich larger documents.\n * @example\n * ```typescript\n * const retriever = new ParentDocumentRetriever({\n *   vectorstore: new MemoryVectorStore(new OpenAIEmbeddings()),\n *   byteStore: new InMemoryStore<Uint8Array>(),\n *   parentSplitter: new RecursiveCharacterTextSplitter({\n *     chunkOverlap: 0,\n *     chunkSize: 500,\n *   }),\n *   childSplitter: new RecursiveCharacterTextSplitter({\n *     chunkOverlap: 0,\n *     chunkSize: 50,\n *   }),\n *   childK: 20,\n *   parentK: 5,\n * });\n *\n * const parentDocuments = await getDocuments();\n * await retriever.addDocuments(parentDocuments);\n * const retrievedDocs = await retriever.invoke(\"justice breyer\");\n * ```\n */\nexport class ParentDocumentRetriever extends MultiVectorRetriever {\n  static lc_name() {\n    return \"ParentDocumentRetriever\";\n  }\n\n  lc_namespace = [\"langchain\", \"retrievers\", \"parent_document\"];\n\n  vectorstore: VectorStoreInterface;\n\n  protected childSplitter: TextSplitter;\n\n  protected parentSplitter?: TextSplitter;\n\n  protected idKey = \"doc_id\";\n\n  protected childK?: number;\n\n  protected parentK?: number;\n\n  childDocumentRetriever:\n    | VectorStoreRetrieverInterface<VectorStoreInterface>\n    | undefined;\n\n  documentCompressor: BaseDocumentCompressor | undefined;\n\n  documentCompressorFilteringFn?: ParentDocumentRetrieverFields[\"documentCompressorFilteringFn\"];\n\n  constructor(fields: ParentDocumentRetrieverFields) {\n    super(fields);\n    this.vectorstore = fields.vectorstore;\n    this.childSplitter = fields.childSplitter;\n    this.parentSplitter = fields.parentSplitter;\n    this.idKey = fields.idKey ?? this.idKey;\n    this.childK = fields.childK;\n    this.parentK = fields.parentK;\n    this.childDocumentRetriever = fields.childDocumentRetriever;\n    this.documentCompressor = fields.documentCompressor;\n    this.documentCompressorFilteringFn = fields.documentCompressorFilteringFn;\n  }\n\n  async _getRelevantDocuments(query: string): Promise<Document[]> {\n    let subDocs: SubDocs = [];\n    if (this.childDocumentRetriever) {\n      subDocs = await this.childDocumentRetriever.invoke(query);\n    } else {\n      subDocs = await this.vectorstore.similaritySearch(query, this.childK);\n    }\n\n    if (this.documentCompressor && subDocs.length) {\n      subDocs = await this.documentCompressor.compressDocuments(subDocs, query);\n      if (this.documentCompressorFilteringFn) {\n        subDocs = this.documentCompressorFilteringFn(subDocs);\n      }\n    }\n\n    // Maintain order\n    const parentDocIds: string[] = [];\n    for (const doc of subDocs) {\n      if (!parentDocIds.includes(doc.metadata[this.idKey])) {\n        parentDocIds.push(doc.metadata[this.idKey]);\n      }\n    }\n    const parentDocs: Document[] = [];\n    const storedParentDocs = await this.docstore.mget(parentDocIds);\n    const retrievedDocs: Document[] = storedParentDocs.filter(\n      (doc?: Document): doc is Document => doc !== undefined\n    );\n    parentDocs.push(...retrievedDocs);\n    return parentDocs.slice(0, this.parentK);\n  }\n\n  async _storeDocuments(\n    parentDoc: Record<string, Document>,\n    childDocs: Document[],\n    addToDocstore: boolean\n  ) {\n    if (this.childDocumentRetriever) {\n      await this.childDocumentRetriever.addDocuments(childDocs);\n    } else {\n      await this.vectorstore.addDocuments(childDocs);\n    }\n    if (addToDocstore) {\n      await this.docstore.mset(Object.entries(parentDoc));\n    }\n  }\n\n  /**\n   * Adds documents to the docstore and vectorstores.\n   * If a retriever is provided, it will be used to add documents instead of the vectorstore.\n   * @param docs The documents to add\n   * @param config.ids Optional list of ids for documents. If provided should be the same\n   *   length as the list of documents. Can provided if parent documents\n   *   are already in the document store and you don't want to re-add\n   *   to the docstore. If not provided, random UUIDs will be used as ids.\n   * @param config.addToDocstore Boolean of whether to add documents to docstore.\n   * This can be false if and only if `ids` are provided. You may want\n   *   to set this to False if the documents are already in the docstore\n   *   and you don't want to re-add them.\n   * @param config.chunkHeaderOptions Object with options for adding Contextual chunk headers\n   */\n  async addDocuments(\n    docs: Document[],\n    config?: {\n      ids?: string[];\n      addToDocstore?: boolean;\n      childDocChunkHeaderOptions?: TextSplitterChunkHeaderOptions;\n    }\n  ): Promise<void> {\n    const {\n      ids,\n      addToDocstore = true,\n      childDocChunkHeaderOptions = {},\n    } = config ?? {};\n    const parentDocs = this.parentSplitter\n      ? await this.parentSplitter.splitDocuments(docs)\n      : docs;\n    let parentDocIds;\n    if (ids === undefined) {\n      if (!addToDocstore) {\n        throw new Error(\n          `If ids are not passed in, \"config.addToDocstore\" MUST be true`\n        );\n      }\n      parentDocIds = parentDocs.map((_doc: Document) => uuid.v4());\n    } else {\n      parentDocIds = ids;\n    }\n    if (parentDocs.length !== parentDocIds.length) {\n      throw new Error(\n        `Got uneven list of documents and ids.\\nIf \"ids\" is provided, should be same length as \"documents\".`\n      );\n    }\n    for (let i = 0; i < parentDocs.length; i += 1) {\n      const parentDoc = parentDocs[i];\n      const parentDocId = parentDocIds[i];\n      const subDocs = await this.childSplitter.splitDocuments(\n        [parentDoc],\n        childDocChunkHeaderOptions\n      );\n      const taggedSubDocs = subDocs.map(\n        (subDoc: Document) =>\n          new Document({\n            pageContent: subDoc.pageContent,\n            metadata: { ...subDoc.metadata, [this.idKey]: parentDocId },\n          })\n      );\n      await this._storeDocuments(\n        { [parentDocId]: parentDoc },\n        taggedSubDocs,\n        addToDocstore\n      );\n    }\n  }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAkEA,IAAa,0BAAb,cAA6C,qBAAqB;CAChE,OAAO,UAAU;AACf,SAAO;;CAGT,eAAe;EAAC;EAAa;EAAc;EAAkB;CAE7D;CAEA;CAEA;CAEA,QAAkB;CAElB;CAEA;CAEA;CAIA;CAEA;CAEA,YAAY,QAAuC;AACjD,QAAM,OAAO;AACb,OAAK,cAAc,OAAO;AAC1B,OAAK,gBAAgB,OAAO;AAC5B,OAAK,iBAAiB,OAAO;AAC7B,OAAK,QAAQ,OAAO,SAAS,KAAK;AAClC,OAAK,SAAS,OAAO;AACrB,OAAK,UAAU,OAAO;AACtB,OAAK,yBAAyB,OAAO;AACrC,OAAK,qBAAqB,OAAO;AACjC,OAAK,gCAAgC,OAAO;;CAG9C,MAAM,sBAAsB,OAAoC;EAC9D,IAAI,UAAmB,EAAE;AACzB,MAAI,KAAK,uBACP,WAAU,MAAM,KAAK,uBAAuB,OAAO,MAAM;MAEzD,WAAU,MAAM,KAAK,YAAY,iBAAiB,OAAO,KAAK,OAAO;AAGvE,MAAI,KAAK,sBAAsB,QAAQ,QAAQ;AAC7C,aAAU,MAAM,KAAK,mBAAmB,kBAAkB,SAAS,MAAM;AACzE,OAAI,KAAK,8BACP,WAAU,KAAK,8BAA8B,QAAQ;;EAKzD,MAAM,eAAyB,EAAE;AACjC,OAAK,MAAM,OAAO,QAChB,KAAI,CAAC,aAAa,SAAS,IAAI,SAAS,KAAK,OAAO,CAClD,cAAa,KAAK,IAAI,SAAS,KAAK,OAAO;EAG/C,MAAM,aAAyB,EAAE;EAEjC,MAAM,iBADmB,MAAM,KAAK,SAAS,KAAK,aAAa,EACZ,QAChD,QAAoC,QAAQ,KAAA,EAC9C;AACD,aAAW,KAAK,GAAG,cAAc;AACjC,SAAO,WAAW,MAAM,GAAG,KAAK,QAAQ;;CAG1C,MAAM,gBACJ,WACA,WACA,eACA;AACA,MAAI,KAAK,uBACP,OAAM,KAAK,uBAAuB,aAAa,UAAU;MAEzD,OAAM,KAAK,YAAY,aAAa,UAAU;AAEhD,MAAI,cACF,OAAM,KAAK,SAAS,KAAK,OAAO,QAAQ,UAAU,CAAC;;;;;;;;;;;;;;;;CAkBvD,MAAM,aACJ,MACA,QAKe;EACf,MAAM,EACJ,KACA,gBAAgB,MAChB,6BAA6B,EAAE,KAC7B,UAAU,EAAE;EAChB,MAAM,aAAa,KAAK,iBACpB,MAAM,KAAK,eAAe,eAAe,KAAK,GAC9C;EACJ,IAAI;AACJ,MAAI,QAAQ,KAAA,GAAW;AACrB,OAAI,CAAC,cACH,OAAM,IAAI,MACR,gEACD;AAEH,kBAAe,WAAW,KAAK,SAAmB,KAAK,IAAI,CAAC;QAE5D,gBAAe;AAEjB,MAAI,WAAW,WAAW,aAAa,OACrC,OAAM,IAAI,MACR,qGACD;AAEH,OAAK,IAAI,IAAI,GAAG,IAAI,WAAW,QAAQ,KAAK,GAAG;GAC7C,MAAM,YAAY,WAAW;GAC7B,MAAM,cAAc,aAAa;GAKjC,MAAM,iBAJU,MAAM,KAAK,cAAc,eACvC,CAAC,UAAU,EACX,2BACD,EAC6B,KAC3B,WACC,IAAI,SAAS;IACX,aAAa,OAAO;IACpB,UAAU;KAAE,GAAG,OAAO;MAAW,KAAK,QAAQ;KAAa;IAC5D,CAAC,CACL;AACD,SAAM,KAAK,gBACT,GAAG,cAAc,WAAW,EAC5B,eACA,cACD"}

package/dist/schema/prompt_template.d.cts

@@ -1,5 +1,4 @@
 import { BasePromptTemplate } from "@langchain/core/prompts";
-import * as _langchain_core_prompt_values0 from "@langchain/core/prompt_values";
 import { Document } from "@langchain/core/documents";
 
 //#region src/schema/prompt_template.d.ts
@@ -12,7 +11,7 @@ import { Document } from "@langchain/core/documents";
  * @returns {Promise<string>} A Promise that resolves to the formatted document as a string.
  * @throws {Error} If the document is missing required metadata variables specified in the prompt template.
  */
-declare const formatDocument: (document: Document<Record<string, any>>, prompt: BasePromptTemplate<any, _langchain_core_prompt_values0.BasePromptValueInterface, any>) => Promise<string>;
+declare const formatDocument: (document: Document, prompt: BasePromptTemplate) => Promise<string>;
 //#endregion
 export { formatDocument };
 //# sourceMappingURL=prompt_template.d.cts.map

package/dist/schema/prompt_template.d.cts.map

@@ -1 +1 @@
-{"version":3,"file":"prompt_template.d.cts","names":[],"sources":["../../src/schema/prompt_template.ts"],"mappings":";;;;;;;;AAYA;;;;;;cAAa,cAAA,GAAc,QAAA,EAAA,QAAA,CAAA,MAAA,gBAAA,MAAA,EAAA,kBAAA,MAAA,8BAAA,CAAA,wBAAA,WAAA,OAAA"}
+{"version":3,"file":"prompt_template.d.cts","names":[],"sources":["../../src/schema/prompt_template.ts"],"mappings":";;;;;;AAYA;;;;;;;cAAa,cAAA,GAAc,QAAA,EACf,QAAA,EAAQ,MAAA,EACV,kBAAA,KACP,OAAA"}

package/dist/schema/prompt_template.d.ts

@@ -1,6 +1,5 @@
 import { BasePromptTemplate } from "@langchain/core/prompts";
 import { Document } from "@langchain/core/documents";
-import * as _langchain_core_prompt_values0 from "@langchain/core/prompt_values";
 
 //#region src/schema/prompt_template.d.ts
 /**
@@ -12,7 +11,7 @@ import * as _langchain_core_prompt_values0 from "@langchain/core/prompt_values";
  * @returns {Promise<string>} A Promise that resolves to the formatted document as a string.
  * @throws {Error} If the document is missing required metadata variables specified in the prompt template.
  */
-declare const formatDocument: (document: Document<Record<string, any>>, prompt: BasePromptTemplate<any, _langchain_core_prompt_values0.BasePromptValueInterface, any>) => Promise<string>;
+declare const formatDocument: (document: Document, prompt: BasePromptTemplate) => Promise<string>;
 //#endregion
 export { formatDocument };
 //# sourceMappingURL=prompt_template.d.ts.map

package/dist/schema/prompt_template.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"prompt_template.d.ts","names":[],"sources":["../../src/schema/prompt_template.ts"],"mappings":";;;;;;;;AAYA;;;;;;cAAa,cAAA,GAAc,QAAA,EAAA,QAAA,CAAA,MAAA,gBAAA,MAAA,EAAA,kBAAA,MAAA,8BAAA,CAAA,wBAAA,WAAA,OAAA"}
+{"version":3,"file":"prompt_template.d.ts","names":[],"sources":["../../src/schema/prompt_template.ts"],"mappings":";;;;;;AAYA;;;;;;;cAAa,cAAA,GAAc,QAAA,EACf,QAAA,EAAQ,MAAA,EACV,kBAAA,KACP,OAAA"}