@langchain/classic 1.0.34 → 1.0.35

package/dist/chains/retrieval.cjs.map

@@ -1 +1 @@
-{"version":3,"file":"retrieval.cjs","names":["RunnableSequence","RunnablePassthrough"],"sources":["../../src/chains/retrieval.ts"],"sourcesContent":["import type { BaseRetrieverInterface } from \"@langchain/core/retrievers\";\nimport {\n  type Runnable,\n  RunnableSequence,\n  type RunnableInterface,\n  RunnablePassthrough,\n} from \"@langchain/core/runnables\";\nimport type { BaseMessage } from \"@langchain/core/messages\";\nimport type { DocumentInterface, Document } from \"@langchain/core/documents\";\n\n/**\n * Parameters for the createRetrievalChain method.\n */\nexport type CreateRetrievalChainParams<RunOutput> = {\n  /**\n   * Retriever-like object that returns list of documents. Should\n   * either be a subclass of BaseRetriever or a Runnable that returns\n   * a list of documents. If a subclass of BaseRetriever, then it\n   * is expected that an `input` key be passed in - this is what\n   * is will be used to pass into the retriever. If this is NOT a\n   * subclass of BaseRetriever, then all the inputs will be passed\n   * into this runnable, meaning that runnable should take a object\n   * as input.\n   */\n  retriever:\n    | BaseRetrieverInterface\n    // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n    | RunnableInterface<Record<string, any>, DocumentInterface[]>;\n  /**\n   * Runnable that takes inputs and produces a string output.\n   * The inputs to this will be any original inputs to this chain, a new\n   * context key with the retrieved documents, and chat_history (if not present\n   * in the inputs) with a value of `[]` (to easily enable conversational\n   * retrieval).\n   */\n  // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n  combineDocsChain: RunnableInterface<Record<string, any>, RunOutput>;\n};\n\nfunction isBaseRetriever(x: unknown): x is BaseRetrieverInterface {\n  return !!x && typeof (x as BaseRetrieverInterface).invoke === \"function\";\n}\n\n/**\n * Create a retrieval chain that retrieves documents and then passes them on.\n * @param {CreateRetrievalChainParams} params A params object\n *     containing a retriever and a combineDocsChain.\n * @returns An LCEL Runnable which returns a an object\n *     containing at least `context` and `answer` keys.\n * @example\n * ```typescript\n * // pnpm add langchain @langchain/openai\n *\n * import { ChatOpenAI } from \"@langchain/openai\";\n * import { pull } from \"langchain/hub\";\n * import { createRetrievalChain } from \"@langchain/classic/chains/retrieval\";\n * import { createStuffDocumentsChain } from \"@langchain/classic/chains/combine_documents\";\n *\n * const retrievalQAChatPrompt = await pull(\"langchain-ai/retrieval-qa-chat\");\n * const llm = new ChatOpenAI({ model: \"gpt-4o-mini\" });\n * const retriever = ...\n * const combineDocsChain = await createStuffDocumentsChain(...);\n * const retrievalChain = await createRetrievalChain({\n *   retriever,\n *   combineDocsChain,\n * });\n * const response = await chain.invoke({ input: \"...\" });\n * ```\n */\nexport async function createRetrievalChain<RunOutput>({\n  retriever,\n  combineDocsChain,\n}: CreateRetrievalChainParams<RunOutput>): Promise<\n  Runnable<\n    { input: string; chat_history?: BaseMessage[] | string } & {\n      [key: string]: unknown;\n    },\n    { context: Document[]; answer: RunOutput } & { [key: string]: unknown }\n  >\n> {\n  let retrieveDocumentsChain: Runnable<{ input: string }, DocumentInterface[]>;\n  if (isBaseRetriever(retriever)) {\n    retrieveDocumentsChain = RunnableSequence.from([\n      (input) => input.input,\n      retriever,\n    ]);\n  } else {\n    // TODO: Fix typing by adding withConfig to core RunnableInterface\n    retrieveDocumentsChain = retriever as Runnable;\n  }\n  const retrievalChain = RunnableSequence.from<{\n    input: string;\n    chat_history?: BaseMessage[] | string;\n  }>([\n    RunnablePassthrough.assign({\n      context: retrieveDocumentsChain.withConfig({\n        runName: \"retrieve_documents\",\n      }),\n      chat_history: (input) => input.chat_history ?? [],\n    }),\n    RunnablePassthrough.assign({\n      answer: combineDocsChain,\n    }),\n  ]).withConfig({ runName: \"retrieval_chain\" });\n  return retrievalChain;\n}\n"],"mappings":";;;;;AAuCA,SAAS,gBAAgB,GAAyC;AAChE,QAAO,CAAC,CAAC,KAAK,OAAQ,EAA6B,WAAW;;;;;;;;;;;;;;;;;;;;;;;;;;;;AA6BhE,eAAsB,qBAAgC,EACpD,WACA,oBAQA;CACA,IAAI;AACJ,KAAI,gBAAgB,UAAU,CAC5B,0BAAyBA,0BAAAA,iBAAiB,KAAK,EAC5C,UAAU,MAAM,OACjB,UACD,CAAC;KAGF,0BAAyB;AAgB3B,QAduBA,0BAAAA,iBAAiB,KAGrC,CACDC,0BAAAA,oBAAoB,OAAO;EACzB,SAAS,uBAAuB,WAAW,EACzC,SAAS,sBACV,CAAC;EACF,eAAe,UAAU,MAAM,gBAAgB,EAAE;EAClD,CAAC,EACFA,0BAAAA,oBAAoB,OAAO,EACzB,QAAQ,kBACT,CAAC,CACH,CAAC,CAAC,WAAW,EAAE,SAAS,mBAAmB,CAAC"}
+{"version":3,"file":"retrieval.cjs","names":["RunnableSequence","RunnablePassthrough"],"sources":["../../src/chains/retrieval.ts"],"sourcesContent":["import type { BaseRetrieverInterface } from \"@langchain/core/retrievers\";\nimport {\n  type Runnable,\n  RunnableSequence,\n  type RunnableInterface,\n  RunnablePassthrough,\n} from \"@langchain/core/runnables\";\nimport type { BaseMessage } from \"@langchain/core/messages\";\nimport type { DocumentInterface, Document } from \"@langchain/core/documents\";\n\n/**\n * Parameters for the createRetrievalChain method.\n */\nexport type CreateRetrievalChainParams<RunOutput> = {\n  /**\n   * Retriever-like object that returns list of documents. Should\n   * either be a subclass of BaseRetriever or a Runnable that returns\n   * a list of documents. If a subclass of BaseRetriever, then it\n   * is expected that an `input` key be passed in - this is what\n   * is will be used to pass into the retriever. If this is NOT a\n   * subclass of BaseRetriever, then all the inputs will be passed\n   * into this runnable, meaning that runnable should take a object\n   * as input.\n   */\n  retriever:\n    | BaseRetrieverInterface\n    // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n    | RunnableInterface<Record<string, any>, DocumentInterface[]>;\n  /**\n   * Runnable that takes inputs and produces a string output.\n   * The inputs to this will be any original inputs to this chain, a new\n   * context key with the retrieved documents, and chat_history (if not present\n   * in the inputs) with a value of `[]` (to easily enable conversational\n   * retrieval).\n   */\n  // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n  combineDocsChain: RunnableInterface<Record<string, any>, RunOutput>;\n};\n\nfunction isBaseRetriever(x: unknown): x is BaseRetrieverInterface {\n  return !!x && typeof (x as BaseRetrieverInterface).invoke === \"function\";\n}\n\n/**\n * Create a retrieval chain that retrieves documents and then passes them on.\n * @param {CreateRetrievalChainParams} params A params object\n *     containing a retriever and a combineDocsChain.\n * @returns An LCEL Runnable which returns a an object\n *     containing at least `context` and `answer` keys.\n * @example\n * ```typescript\n * // pnpm add langchain @langchain/openai\n *\n * import { ChatOpenAI } from \"@langchain/openai\";\n * import { pull } from \"langchain/hub\";\n * import { createRetrievalChain } from \"@langchain/classic/chains/retrieval\";\n * import { createStuffDocumentsChain } from \"@langchain/classic/chains/combine_documents\";\n *\n * const retrievalQAChatPrompt = await pull(\"langchain-ai/retrieval-qa-chat\");\n * const llm = new ChatOpenAI({ model: \"gpt-4o-mini\" });\n * const retriever = ...\n * const combineDocsChain = await createStuffDocumentsChain(...);\n * const retrievalChain = await createRetrievalChain({\n *   retriever,\n *   combineDocsChain,\n * });\n * const response = await chain.invoke({ input: \"...\" });\n * ```\n */\nexport async function createRetrievalChain<RunOutput>({\n  retriever,\n  combineDocsChain,\n}: CreateRetrievalChainParams<RunOutput>): Promise<\n  Runnable<\n    { input: string; chat_history?: BaseMessage[] | string } & {\n      [key: string]: unknown;\n    },\n    { context: Document[]; answer: RunOutput } & { [key: string]: unknown }\n  >\n> {\n  let retrieveDocumentsChain: Runnable<{ input: string }, DocumentInterface[]>;\n  if (isBaseRetriever(retriever)) {\n    retrieveDocumentsChain = RunnableSequence.from([\n      (input) => input.input,\n      retriever,\n    ]);\n  } else {\n    // TODO: Fix typing by adding withConfig to core RunnableInterface\n    retrieveDocumentsChain = retriever as Runnable;\n  }\n  const retrievalChain = RunnableSequence.from<{\n    input: string;\n    chat_history?: BaseMessage[] | string;\n  }>([\n    RunnablePassthrough.assign({\n      context: retrieveDocumentsChain.withConfig({\n        runName: \"retrieve_documents\",\n      }),\n      chat_history: (input) => input.chat_history ?? [],\n    }),\n    RunnablePassthrough.assign({\n      answer: combineDocsChain,\n    }),\n  ]).withConfig({ runName: \"retrieval_chain\" });\n  return retrievalChain;\n}\n"],"mappings":";;;;;AAuCA,SAAS,gBAAgB,GAAyC;AAChE,QAAO,CAAC,CAAC,KAAK,OAAQ,EAA6B,WAAW;;;;;;;;;;;;;;;;;;;;;;;;;;;;AA6BhE,eAAsB,qBAAgC,EACpD,WACA,oBAQA;CACA,IAAI;AACJ,KAAI,gBAAgB,UAAU,CAC5B,0BAAyBA,0BAAAA,iBAAiB,KAAK,EAC5C,UAAU,MAAM,OACjB,UACD,CAAC;KAGF,0BAAyB;AAgB3B,QAduBA,0BAAAA,iBAAiB,KAGrC,CACDC,0BAAAA,oBAAoB,OAAO;EACzB,SAAS,uBAAuB,WAAW,EACzC,SAAS,sBACV,CAAC;EACF,eAAe,UAAU,MAAM,gBAAgB,EAAE;EAClD,CAAC,EACFA,0BAAAA,oBAAoB,OAAO,EACzB,QAAQ,kBACT,CAAC,CACH,CAAC,CAAC,WAAW,EAAE,SAAS,mBAAmB,CACvB"}

package/dist/chains/retrieval.js.map

@@ -1 +1 @@
-{"version":3,"file":"retrieval.js","names":[],"sources":["../../src/chains/retrieval.ts"],"sourcesContent":["import type { BaseRetrieverInterface } from \"@langchain/core/retrievers\";\nimport {\n  type Runnable,\n  RunnableSequence,\n  type RunnableInterface,\n  RunnablePassthrough,\n} from \"@langchain/core/runnables\";\nimport type { BaseMessage } from \"@langchain/core/messages\";\nimport type { DocumentInterface, Document } from \"@langchain/core/documents\";\n\n/**\n * Parameters for the createRetrievalChain method.\n */\nexport type CreateRetrievalChainParams<RunOutput> = {\n  /**\n   * Retriever-like object that returns list of documents. Should\n   * either be a subclass of BaseRetriever or a Runnable that returns\n   * a list of documents. If a subclass of BaseRetriever, then it\n   * is expected that an `input` key be passed in - this is what\n   * is will be used to pass into the retriever. If this is NOT a\n   * subclass of BaseRetriever, then all the inputs will be passed\n   * into this runnable, meaning that runnable should take a object\n   * as input.\n   */\n  retriever:\n    | BaseRetrieverInterface\n    // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n    | RunnableInterface<Record<string, any>, DocumentInterface[]>;\n  /**\n   * Runnable that takes inputs and produces a string output.\n   * The inputs to this will be any original inputs to this chain, a new\n   * context key with the retrieved documents, and chat_history (if not present\n   * in the inputs) with a value of `[]` (to easily enable conversational\n   * retrieval).\n   */\n  // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n  combineDocsChain: RunnableInterface<Record<string, any>, RunOutput>;\n};\n\nfunction isBaseRetriever(x: unknown): x is BaseRetrieverInterface {\n  return !!x && typeof (x as BaseRetrieverInterface).invoke === \"function\";\n}\n\n/**\n * Create a retrieval chain that retrieves documents and then passes them on.\n * @param {CreateRetrievalChainParams} params A params object\n *     containing a retriever and a combineDocsChain.\n * @returns An LCEL Runnable which returns a an object\n *     containing at least `context` and `answer` keys.\n * @example\n * ```typescript\n * // pnpm add langchain @langchain/openai\n *\n * import { ChatOpenAI } from \"@langchain/openai\";\n * import { pull } from \"langchain/hub\";\n * import { createRetrievalChain } from \"@langchain/classic/chains/retrieval\";\n * import { createStuffDocumentsChain } from \"@langchain/classic/chains/combine_documents\";\n *\n * const retrievalQAChatPrompt = await pull(\"langchain-ai/retrieval-qa-chat\");\n * const llm = new ChatOpenAI({ model: \"gpt-4o-mini\" });\n * const retriever = ...\n * const combineDocsChain = await createStuffDocumentsChain(...);\n * const retrievalChain = await createRetrievalChain({\n *   retriever,\n *   combineDocsChain,\n * });\n * const response = await chain.invoke({ input: \"...\" });\n * ```\n */\nexport async function createRetrievalChain<RunOutput>({\n  retriever,\n  combineDocsChain,\n}: CreateRetrievalChainParams<RunOutput>): Promise<\n  Runnable<\n    { input: string; chat_history?: BaseMessage[] | string } & {\n      [key: string]: unknown;\n    },\n    { context: Document[]; answer: RunOutput } & { [key: string]: unknown }\n  >\n> {\n  let retrieveDocumentsChain: Runnable<{ input: string }, DocumentInterface[]>;\n  if (isBaseRetriever(retriever)) {\n    retrieveDocumentsChain = RunnableSequence.from([\n      (input) => input.input,\n      retriever,\n    ]);\n  } else {\n    // TODO: Fix typing by adding withConfig to core RunnableInterface\n    retrieveDocumentsChain = retriever as Runnable;\n  }\n  const retrievalChain = RunnableSequence.from<{\n    input: string;\n    chat_history?: BaseMessage[] | string;\n  }>([\n    RunnablePassthrough.assign({\n      context: retrieveDocumentsChain.withConfig({\n        runName: \"retrieve_documents\",\n      }),\n      chat_history: (input) => input.chat_history ?? [],\n    }),\n    RunnablePassthrough.assign({\n      answer: combineDocsChain,\n    }),\n  ]).withConfig({ runName: \"retrieval_chain\" });\n  return retrievalChain;\n}\n"],"mappings":";;;;AAuCA,SAAS,gBAAgB,GAAyC;AAChE,QAAO,CAAC,CAAC,KAAK,OAAQ,EAA6B,WAAW;;;;;;;;;;;;;;;;;;;;;;;;;;;;AA6BhE,eAAsB,qBAAgC,EACpD,WACA,oBAQA;CACA,IAAI;AACJ,KAAI,gBAAgB,UAAU,CAC5B,0BAAyB,iBAAiB,KAAK,EAC5C,UAAU,MAAM,OACjB,UACD,CAAC;KAGF,0BAAyB;AAgB3B,QAduB,iBAAiB,KAGrC,CACD,oBAAoB,OAAO;EACzB,SAAS,uBAAuB,WAAW,EACzC,SAAS,sBACV,CAAC;EACF,eAAe,UAAU,MAAM,gBAAgB,EAAE;EAClD,CAAC,EACF,oBAAoB,OAAO,EACzB,QAAQ,kBACT,CAAC,CACH,CAAC,CAAC,WAAW,EAAE,SAAS,mBAAmB,CAAC"}
+{"version":3,"file":"retrieval.js","names":[],"sources":["../../src/chains/retrieval.ts"],"sourcesContent":["import type { BaseRetrieverInterface } from \"@langchain/core/retrievers\";\nimport {\n  type Runnable,\n  RunnableSequence,\n  type RunnableInterface,\n  RunnablePassthrough,\n} from \"@langchain/core/runnables\";\nimport type { BaseMessage } from \"@langchain/core/messages\";\nimport type { DocumentInterface, Document } from \"@langchain/core/documents\";\n\n/**\n * Parameters for the createRetrievalChain method.\n */\nexport type CreateRetrievalChainParams<RunOutput> = {\n  /**\n   * Retriever-like object that returns list of documents. Should\n   * either be a subclass of BaseRetriever or a Runnable that returns\n   * a list of documents. If a subclass of BaseRetriever, then it\n   * is expected that an `input` key be passed in - this is what\n   * is will be used to pass into the retriever. If this is NOT a\n   * subclass of BaseRetriever, then all the inputs will be passed\n   * into this runnable, meaning that runnable should take a object\n   * as input.\n   */\n  retriever:\n    | BaseRetrieverInterface\n    // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n    | RunnableInterface<Record<string, any>, DocumentInterface[]>;\n  /**\n   * Runnable that takes inputs and produces a string output.\n   * The inputs to this will be any original inputs to this chain, a new\n   * context key with the retrieved documents, and chat_history (if not present\n   * in the inputs) with a value of `[]` (to easily enable conversational\n   * retrieval).\n   */\n  // oxlint-disable-next-line @typescript-eslint/no-explicit-any\n  combineDocsChain: RunnableInterface<Record<string, any>, RunOutput>;\n};\n\nfunction isBaseRetriever(x: unknown): x is BaseRetrieverInterface {\n  return !!x && typeof (x as BaseRetrieverInterface).invoke === \"function\";\n}\n\n/**\n * Create a retrieval chain that retrieves documents and then passes them on.\n * @param {CreateRetrievalChainParams} params A params object\n *     containing a retriever and a combineDocsChain.\n * @returns An LCEL Runnable which returns a an object\n *     containing at least `context` and `answer` keys.\n * @example\n * ```typescript\n * // pnpm add langchain @langchain/openai\n *\n * import { ChatOpenAI } from \"@langchain/openai\";\n * import { pull } from \"langchain/hub\";\n * import { createRetrievalChain } from \"@langchain/classic/chains/retrieval\";\n * import { createStuffDocumentsChain } from \"@langchain/classic/chains/combine_documents\";\n *\n * const retrievalQAChatPrompt = await pull(\"langchain-ai/retrieval-qa-chat\");\n * const llm = new ChatOpenAI({ model: \"gpt-4o-mini\" });\n * const retriever = ...\n * const combineDocsChain = await createStuffDocumentsChain(...);\n * const retrievalChain = await createRetrievalChain({\n *   retriever,\n *   combineDocsChain,\n * });\n * const response = await chain.invoke({ input: \"...\" });\n * ```\n */\nexport async function createRetrievalChain<RunOutput>({\n  retriever,\n  combineDocsChain,\n}: CreateRetrievalChainParams<RunOutput>): Promise<\n  Runnable<\n    { input: string; chat_history?: BaseMessage[] | string } & {\n      [key: string]: unknown;\n    },\n    { context: Document[]; answer: RunOutput } & { [key: string]: unknown }\n  >\n> {\n  let retrieveDocumentsChain: Runnable<{ input: string }, DocumentInterface[]>;\n  if (isBaseRetriever(retriever)) {\n    retrieveDocumentsChain = RunnableSequence.from([\n      (input) => input.input,\n      retriever,\n    ]);\n  } else {\n    // TODO: Fix typing by adding withConfig to core RunnableInterface\n    retrieveDocumentsChain = retriever as Runnable;\n  }\n  const retrievalChain = RunnableSequence.from<{\n    input: string;\n    chat_history?: BaseMessage[] | string;\n  }>([\n    RunnablePassthrough.assign({\n      context: retrieveDocumentsChain.withConfig({\n        runName: \"retrieve_documents\",\n      }),\n      chat_history: (input) => input.chat_history ?? [],\n    }),\n    RunnablePassthrough.assign({\n      answer: combineDocsChain,\n    }),\n  ]).withConfig({ runName: \"retrieval_chain\" });\n  return retrievalChain;\n}\n"],"mappings":";;;;AAuCA,SAAS,gBAAgB,GAAyC;AAChE,QAAO,CAAC,CAAC,KAAK,OAAQ,EAA6B,WAAW;;;;;;;;;;;;;;;;;;;;;;;;;;;;AA6BhE,eAAsB,qBAAgC,EACpD,WACA,oBAQA;CACA,IAAI;AACJ,KAAI,gBAAgB,UAAU,CAC5B,0BAAyB,iBAAiB,KAAK,EAC5C,UAAU,MAAM,OACjB,UACD,CAAC;KAGF,0BAAyB;AAgB3B,QAduB,iBAAiB,KAGrC,CACD,oBAAoB,OAAO;EACzB,SAAS,uBAAuB,WAAW,EACzC,SAAS,sBACV,CAAC;EACF,eAAe,UAAU,MAAM,gBAAgB,EAAE;EAClD,CAAC,EACF,oBAAoB,OAAO,EACzB,QAAQ,kBACT,CAAC,CACH,CAAC,CAAC,WAAW,EAAE,SAAS,mBAAmB,CACvB"}

package/dist/chains/router/multi_prompt.cjs.map

@@ -1 +1 @@
-{"version":3,"file":"multi_prompt.cjs","names":["MultiRouteChain","zipEntries","RouterOutputParser","z","PromptTemplate","STRUCTURED_MULTI_PROMPT_ROUTER_TEMPLATE","LLMRouterChain","LLMChain","ConversationChain"],"sources":["../../../src/chains/router/multi_prompt.ts"],"sourcesContent":["import type { BaseLanguageModelInterface } from \"@langchain/core/language_models/base\";\nimport { z } from \"zod/v3\";\nimport { interpolateFString, PromptTemplate } from \"@langchain/core/prompts\";\nimport { MultiRouteChain, MultiRouteChainInput } from \"./multi_route.js\";\nimport { STRUCTURED_MULTI_PROMPT_ROUTER_TEMPLATE } from \"./multi_prompt_prompt.js\";\nimport { BaseChain } from \"../../chains/base.js\";\nimport { LLMChain, LLMChainInput } from \"../../chains/llm_chain.js\";\nimport { LLMRouterChain } from \"./llm_router.js\";\nimport { ConversationChain } from \"../../chains/conversation.js\";\nimport { zipEntries } from \"./utils.js\";\nimport { RouterOutputParser } from \"../../output_parsers/router.js\";\n\n/**\n * A class that represents a multi-prompt chain in the LangChain\n * framework. It extends the MultiRouteChain class and provides additional\n * functionality specific to multi-prompt chains.\n * @example\n * ```typescript\n * const multiPromptChain = MultiPromptChain.fromLLMAndPrompts(\n *   new ChatOpenAI({ model: \"gpt-4o-mini\" }),\n *   {\n *     promptNames: [\"physics\", \"math\", \"history\"],\n *     promptDescriptions: [\n *       \"Good for answering questions about physics\",\n *       \"Good for answering math questions\",\n *       \"Good for answering questions about history\",\n *     ],\n *     promptTemplates: [\n *       `You are a very smart physics professor. Here is a question:\\n{input}\\n`,\n *       `You are a very good mathematician. Here is a question:\\n{input}\\n`,\n *       `You are a very smart history professor. Here is a question:\\n{input}\\n`,\n *     ],\n *   }\n * );\n * const result = await multiPromptChain.call({\n *   input: \"What is the speed of light?\",\n * });\n * ```\n */\nexport class MultiPromptChain extends MultiRouteChain {\n  /**\n   * @deprecated Use `fromLLMAndPrompts` instead\n   */\n  static fromPrompts(\n    llm: BaseLanguageModelInterface,\n    promptNames: string[],\n    promptDescriptions: string[],\n    promptTemplates: string[] | PromptTemplate[],\n    defaultChain?: BaseChain,\n    options?: Omit<MultiRouteChainInput, \"defaultChain\">\n  ) {\n    return MultiPromptChain.fromLLMAndPrompts(llm, {\n      promptNames,\n      promptDescriptions,\n      promptTemplates,\n      defaultChain,\n      multiRouteChainOpts: options,\n    });\n  }\n\n  /**\n   * A static method that creates an instance of MultiPromptChain from a\n   * BaseLanguageModel and a set of prompts. It takes in optional parameters\n   * for the default chain and additional options.\n   * @param llm A BaseLanguageModel instance.\n   * @param promptNames An array of prompt names.\n   * @param promptDescriptions An array of prompt descriptions.\n   * @param promptTemplates An array of prompt templates.\n   * @param defaultChain An optional BaseChain instance to be used as the default chain.\n   * @param llmChainOpts Optional parameters for the LLMChainInput, excluding 'llm' and 'prompt'.\n   * @param conversationChainOpts Optional parameters for the LLMChainInput, excluding 'llm' and 'outputKey'.\n   * @param multiRouteChainOpts Optional parameters for the MultiRouteChainInput, excluding 'defaultChain'.\n   * @returns An instance of MultiPromptChain.\n   */\n  static fromLLMAndPrompts(\n    llm: BaseLanguageModelInterface,\n    {\n      promptNames,\n      promptDescriptions,\n      promptTemplates,\n      defaultChain,\n      llmChainOpts,\n      conversationChainOpts,\n      multiRouteChainOpts,\n    }: {\n      promptNames: string[];\n      promptDescriptions: string[];\n      promptTemplates: string[] | PromptTemplate[];\n      defaultChain?: BaseChain;\n      llmChainOpts?: Omit<LLMChainInput, \"llm\" | \"prompt\">;\n      conversationChainOpts?: Omit<LLMChainInput, \"llm\" | \"outputKey\">;\n      multiRouteChainOpts?: Omit<MultiRouteChainInput, \"defaultChain\">;\n    }\n  ): MultiPromptChain {\n    const destinations = zipEntries(promptNames, promptDescriptions).map(\n      ([name, desc]) => `${name}: ${desc}`\n    );\n\n    const structuredOutputParserSchema = z.object({\n      destination: z\n        .string()\n        .optional()\n        .describe('name of the question answering system to use or \"DEFAULT\"'),\n      next_inputs: z\n        .object({\n          input: z\n            .string()\n            .describe(\"a potentially modified version of the original input\"),\n        })\n        .describe(\"input to be fed to the next model\"),\n    });\n\n    const outputParser = new RouterOutputParser(structuredOutputParserSchema);\n\n    const destinationsStr = destinations.join(\"\\n\");\n    const routerTemplate = interpolateFString(\n      STRUCTURED_MULTI_PROMPT_ROUTER_TEMPLATE(\n        outputParser.getFormatInstructions({ interpolationDepth: 4 })\n      ),\n      {\n        destinations: destinationsStr,\n      }\n    );\n\n    const routerPrompt = new PromptTemplate({\n      template: routerTemplate,\n      inputVariables: [\"input\"],\n      outputParser,\n    });\n\n    const routerChain = LLMRouterChain.fromLLM(llm, routerPrompt);\n    const destinationChains = zipEntries<[string, string | PromptTemplate]>(\n      promptNames,\n      promptTemplates\n    ).reduce(\n      (acc, [name, template]) => {\n        let myPrompt: string | PromptTemplate;\n        if (typeof template === \"object\") {\n          myPrompt = template;\n        } else if (typeof template === \"string\") {\n          myPrompt = new PromptTemplate({\n            template: template as string,\n            inputVariables: [\"input\"],\n          });\n        } else {\n          throw new Error(\"Invalid prompt template\");\n        }\n        acc[name as string] = new LLMChain({\n          ...llmChainOpts,\n          llm,\n          prompt: myPrompt,\n        });\n        return acc;\n      },\n      {} as { [name: string]: LLMChain }\n    );\n\n    const convChain = new ConversationChain({\n      ...conversationChainOpts,\n      llm,\n      outputKey: \"text\",\n    });\n\n    return new MultiPromptChain({\n      ...multiRouteChainOpts,\n      routerChain,\n      destinationChains,\n      defaultChain: defaultChain ?? convChain,\n    });\n  }\n\n  _chainType(): string {\n    return \"multi_prompt_chain\";\n  }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAuCA,IAAa,mBAAb,MAAa,yBAAyBA,oBAAAA,gBAAgB;;;;CAIpD,OAAO,YACL,KACA,aACA,oBACA,iBACA,cACA,SACA;AACA,SAAO,iBAAiB,kBAAkB,KAAK;GAC7C;GACA;GACA;GACA;GACA,qBAAqB;GACtB,CAAC;;;;;;;;;;;;;;;;CAiBJ,OAAO,kBACL,KACA,EACE,aACA,oBACA,iBACA,cACA,cACA,uBACA,uBAUgB;EAClB,MAAM,eAAeC,cAAAA,WAAW,aAAa,mBAAmB,CAAC,KAC9D,CAAC,MAAM,UAAU,GAAG,KAAK,IAAI,OAC/B;EAgBD,MAAM,eAAe,IAAIC,eAAAA,mBAdYC,OAAAA,EAAE,OAAO;GAC5C,aAAaA,OAAAA,EACV,QAAQ,CACR,UAAU,CACV,SAAS,8DAA4D;GACxE,aAAaA,OAAAA,EACV,OAAO,EACN,OAAOA,OAAAA,EACJ,QAAQ,CACR,SAAS,uDAAuD,EACpE,CAAC,CACD,SAAS,oCAAoC;GACjD,CAAC,CAEuE;EAEzE,MAAM,kBAAkB,aAAa,KAAK,KAAK;EAU/C,MAAM,eAAe,IAAIC,wBAAAA,eAAe;GACtC,WAAA,GAAA,wBAAA,oBATAC,4BAAAA,wCACE,aAAa,sBAAsB,EAAE,oBAAoB,GAAG,CAAC,CAC9D,EACD,EACE,cAAc,iBACf,CACF;GAIC,gBAAgB,CAAC,QAAQ;GACzB;GACD,CAAC;EAEF,MAAM,cAAcC,mBAAAA,eAAe,QAAQ,KAAK,aAAa;EAC7D,MAAM,oBAAoBL,cAAAA,WACxB,aACA,gBACD,CAAC,QACC,KAAK,CAAC,MAAM,cAAc;GACzB,IAAI;AACJ,OAAI,OAAO,aAAa,SACtB,YAAW;YACF,OAAO,aAAa,SAC7B,YAAW,IAAIG,wBAAAA,eAAe;IAClB;IACV,gBAAgB,CAAC,QAAQ;IAC1B,CAAC;OAEF,OAAM,IAAI,MAAM,0BAA0B;AAE5C,OAAI,QAAkB,IAAIG,kBAAAA,SAAS;IACjC,GAAG;IACH;IACA,QAAQ;IACT,CAAC;AACF,UAAO;KAET,EAAE,CACH;EAED,MAAM,YAAY,IAAIC,qBAAAA,kBAAkB;GACtC,GAAG;GACH;GACA,WAAW;GACZ,CAAC;AAEF,SAAO,IAAI,iBAAiB;GAC1B,GAAG;GACH;GACA;GACA,cAAc,gBAAgB;GAC/B,CAAC;;CAGJ,aAAqB;AACnB,SAAO"}
+{"version":3,"file":"multi_prompt.cjs","names":["MultiRouteChain","zipEntries","RouterOutputParser","z","PromptTemplate","STRUCTURED_MULTI_PROMPT_ROUTER_TEMPLATE","LLMRouterChain","LLMChain","ConversationChain"],"sources":["../../../src/chains/router/multi_prompt.ts"],"sourcesContent":["import type { BaseLanguageModelInterface } from \"@langchain/core/language_models/base\";\nimport { z } from \"zod/v3\";\nimport { interpolateFString, PromptTemplate } from \"@langchain/core/prompts\";\nimport { MultiRouteChain, MultiRouteChainInput } from \"./multi_route.js\";\nimport { STRUCTURED_MULTI_PROMPT_ROUTER_TEMPLATE } from \"./multi_prompt_prompt.js\";\nimport { BaseChain } from \"../../chains/base.js\";\nimport { LLMChain, LLMChainInput } from \"../../chains/llm_chain.js\";\nimport { LLMRouterChain } from \"./llm_router.js\";\nimport { ConversationChain } from \"../../chains/conversation.js\";\nimport { zipEntries } from \"./utils.js\";\nimport { RouterOutputParser } from \"../../output_parsers/router.js\";\n\n/**\n * A class that represents a multi-prompt chain in the LangChain\n * framework. It extends the MultiRouteChain class and provides additional\n * functionality specific to multi-prompt chains.\n * @example\n * ```typescript\n * const multiPromptChain = MultiPromptChain.fromLLMAndPrompts(\n *   new ChatOpenAI({ model: \"gpt-4o-mini\" }),\n *   {\n *     promptNames: [\"physics\", \"math\", \"history\"],\n *     promptDescriptions: [\n *       \"Good for answering questions about physics\",\n *       \"Good for answering math questions\",\n *       \"Good for answering questions about history\",\n *     ],\n *     promptTemplates: [\n *       `You are a very smart physics professor. Here is a question:\\n{input}\\n`,\n *       `You are a very good mathematician. Here is a question:\\n{input}\\n`,\n *       `You are a very smart history professor. Here is a question:\\n{input}\\n`,\n *     ],\n *   }\n * );\n * const result = await multiPromptChain.call({\n *   input: \"What is the speed of light?\",\n * });\n * ```\n */\nexport class MultiPromptChain extends MultiRouteChain {\n  /**\n   * @deprecated Use `fromLLMAndPrompts` instead\n   */\n  static fromPrompts(\n    llm: BaseLanguageModelInterface,\n    promptNames: string[],\n    promptDescriptions: string[],\n    promptTemplates: string[] | PromptTemplate[],\n    defaultChain?: BaseChain,\n    options?: Omit<MultiRouteChainInput, \"defaultChain\">\n  ) {\n    return MultiPromptChain.fromLLMAndPrompts(llm, {\n      promptNames,\n      promptDescriptions,\n      promptTemplates,\n      defaultChain,\n      multiRouteChainOpts: options,\n    });\n  }\n\n  /**\n   * A static method that creates an instance of MultiPromptChain from a\n   * BaseLanguageModel and a set of prompts. It takes in optional parameters\n   * for the default chain and additional options.\n   * @param llm A BaseLanguageModel instance.\n   * @param promptNames An array of prompt names.\n   * @param promptDescriptions An array of prompt descriptions.\n   * @param promptTemplates An array of prompt templates.\n   * @param defaultChain An optional BaseChain instance to be used as the default chain.\n   * @param llmChainOpts Optional parameters for the LLMChainInput, excluding 'llm' and 'prompt'.\n   * @param conversationChainOpts Optional parameters for the LLMChainInput, excluding 'llm' and 'outputKey'.\n   * @param multiRouteChainOpts Optional parameters for the MultiRouteChainInput, excluding 'defaultChain'.\n   * @returns An instance of MultiPromptChain.\n   */\n  static fromLLMAndPrompts(\n    llm: BaseLanguageModelInterface,\n    {\n      promptNames,\n      promptDescriptions,\n      promptTemplates,\n      defaultChain,\n      llmChainOpts,\n      conversationChainOpts,\n      multiRouteChainOpts,\n    }: {\n      promptNames: string[];\n      promptDescriptions: string[];\n      promptTemplates: string[] | PromptTemplate[];\n      defaultChain?: BaseChain;\n      llmChainOpts?: Omit<LLMChainInput, \"llm\" | \"prompt\">;\n      conversationChainOpts?: Omit<LLMChainInput, \"llm\" | \"outputKey\">;\n      multiRouteChainOpts?: Omit<MultiRouteChainInput, \"defaultChain\">;\n    }\n  ): MultiPromptChain {\n    const destinations = zipEntries(promptNames, promptDescriptions).map(\n      ([name, desc]) => `${name}: ${desc}`\n    );\n\n    const structuredOutputParserSchema = z.object({\n      destination: z\n        .string()\n        .optional()\n        .describe('name of the question answering system to use or \"DEFAULT\"'),\n      next_inputs: z\n        .object({\n          input: z\n            .string()\n            .describe(\"a potentially modified version of the original input\"),\n        })\n        .describe(\"input to be fed to the next model\"),\n    });\n\n    const outputParser = new RouterOutputParser(structuredOutputParserSchema);\n\n    const destinationsStr = destinations.join(\"\\n\");\n    const routerTemplate = interpolateFString(\n      STRUCTURED_MULTI_PROMPT_ROUTER_TEMPLATE(\n        outputParser.getFormatInstructions({ interpolationDepth: 4 })\n      ),\n      {\n        destinations: destinationsStr,\n      }\n    );\n\n    const routerPrompt = new PromptTemplate({\n      template: routerTemplate,\n      inputVariables: [\"input\"],\n      outputParser,\n    });\n\n    const routerChain = LLMRouterChain.fromLLM(llm, routerPrompt);\n    const destinationChains = zipEntries<[string, string | PromptTemplate]>(\n      promptNames,\n      promptTemplates\n    ).reduce(\n      (acc, [name, template]) => {\n        let myPrompt: string | PromptTemplate;\n        if (typeof template === \"object\") {\n          myPrompt = template;\n        } else if (typeof template === \"string\") {\n          myPrompt = new PromptTemplate({\n            template: template as string,\n            inputVariables: [\"input\"],\n          });\n        } else {\n          throw new Error(\"Invalid prompt template\");\n        }\n        acc[name as string] = new LLMChain({\n          ...llmChainOpts,\n          llm,\n          prompt: myPrompt,\n        });\n        return acc;\n      },\n      {} as { [name: string]: LLMChain }\n    );\n\n    const convChain = new ConversationChain({\n      ...conversationChainOpts,\n      llm,\n      outputKey: \"text\",\n    });\n\n    return new MultiPromptChain({\n      ...multiRouteChainOpts,\n      routerChain,\n      destinationChains,\n      defaultChain: defaultChain ?? convChain,\n    });\n  }\n\n  _chainType(): string {\n    return \"multi_prompt_chain\";\n  }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAuCA,IAAa,mBAAb,MAAa,yBAAyBA,oBAAAA,gBAAgB;;;;CAIpD,OAAO,YACL,KACA,aACA,oBACA,iBACA,cACA,SACA;AACA,SAAO,iBAAiB,kBAAkB,KAAK;GAC7C;GACA;GACA;GACA;GACA,qBAAqB;GACtB,CAAC;;;;;;;;;;;;;;;;CAiBJ,OAAO,kBACL,KACA,EACE,aACA,oBACA,iBACA,cACA,cACA,uBACA,uBAUgB;EAClB,MAAM,eAAeC,cAAAA,WAAW,aAAa,mBAAmB,CAAC,KAC9D,CAAC,MAAM,UAAU,GAAG,KAAK,IAAI,OAC/B;EAgBD,MAAM,eAAe,IAAIC,eAAAA,mBAdYC,OAAAA,EAAE,OAAO;GAC5C,aAAaA,OAAAA,EACV,QAAQ,CACR,UAAU,CACV,SAAS,8DAA4D;GACxE,aAAaA,OAAAA,EACV,OAAO,EACN,OAAOA,OAAAA,EACJ,QAAQ,CACR,SAAS,uDAAuD,EACpE,CAAC,CACD,SAAS,oCAAoC;GACjD,CAEuE,CAAC;EAEzE,MAAM,kBAAkB,aAAa,KAAK,KAAK;EAU/C,MAAM,eAAe,IAAIC,wBAAAA,eAAe;GACtC,WAAA,GAAA,wBAAA,oBATAC,4BAAAA,wCACE,aAAa,sBAAsB,EAAE,oBAAoB,GAAG,CAAC,CAC9D,EACD,EACE,cAAc,iBACf,CAIuB;GACxB,gBAAgB,CAAC,QAAQ;GACzB;GACD,CAAC;EAEF,MAAM,cAAcC,mBAAAA,eAAe,QAAQ,KAAK,aAAa;EAC7D,MAAM,oBAAoBL,cAAAA,WACxB,aACA,gBACD,CAAC,QACC,KAAK,CAAC,MAAM,cAAc;GACzB,IAAI;AACJ,OAAI,OAAO,aAAa,SACtB,YAAW;YACF,OAAO,aAAa,SAC7B,YAAW,IAAIG,wBAAAA,eAAe;IAClB;IACV,gBAAgB,CAAC,QAAQ;IAC1B,CAAC;OAEF,OAAM,IAAI,MAAM,0BAA0B;AAE5C,OAAI,QAAkB,IAAIG,kBAAAA,SAAS;IACjC,GAAG;IACH;IACA,QAAQ;IACT,CAAC;AACF,UAAO;KAET,EAAE,CACH;EAED,MAAM,YAAY,IAAIC,qBAAAA,kBAAkB;GACtC,GAAG;GACH;GACA,WAAW;GACZ,CAAC;AAEF,SAAO,IAAI,iBAAiB;GAC1B,GAAG;GACH;GACA;GACA,cAAc,gBAAgB;GAC/B,CAAC;;CAGJ,aAAqB;AACnB,SAAO"}

package/dist/chains/router/multi_prompt.js.map

@@ -1 +1 @@
-{"version":3,"file":"multi_prompt.js","names":[],"sources":["../../../src/chains/router/multi_prompt.ts"],"sourcesContent":["import type { BaseLanguageModelInterface } from \"@langchain/core/language_models/base\";\nimport { z } from \"zod/v3\";\nimport { interpolateFString, PromptTemplate } from \"@langchain/core/prompts\";\nimport { MultiRouteChain, MultiRouteChainInput } from \"./multi_route.js\";\nimport { STRUCTURED_MULTI_PROMPT_ROUTER_TEMPLATE } from \"./multi_prompt_prompt.js\";\nimport { BaseChain } from \"../../chains/base.js\";\nimport { LLMChain, LLMChainInput } from \"../../chains/llm_chain.js\";\nimport { LLMRouterChain } from \"./llm_router.js\";\nimport { ConversationChain } from \"../../chains/conversation.js\";\nimport { zipEntries } from \"./utils.js\";\nimport { RouterOutputParser } from \"../../output_parsers/router.js\";\n\n/**\n * A class that represents a multi-prompt chain in the LangChain\n * framework. It extends the MultiRouteChain class and provides additional\n * functionality specific to multi-prompt chains.\n * @example\n * ```typescript\n * const multiPromptChain = MultiPromptChain.fromLLMAndPrompts(\n *   new ChatOpenAI({ model: \"gpt-4o-mini\" }),\n *   {\n *     promptNames: [\"physics\", \"math\", \"history\"],\n *     promptDescriptions: [\n *       \"Good for answering questions about physics\",\n *       \"Good for answering math questions\",\n *       \"Good for answering questions about history\",\n *     ],\n *     promptTemplates: [\n *       `You are a very smart physics professor. Here is a question:\\n{input}\\n`,\n *       `You are a very good mathematician. Here is a question:\\n{input}\\n`,\n *       `You are a very smart history professor. Here is a question:\\n{input}\\n`,\n *     ],\n *   }\n * );\n * const result = await multiPromptChain.call({\n *   input: \"What is the speed of light?\",\n * });\n * ```\n */\nexport class MultiPromptChain extends MultiRouteChain {\n  /**\n   * @deprecated Use `fromLLMAndPrompts` instead\n   */\n  static fromPrompts(\n    llm: BaseLanguageModelInterface,\n    promptNames: string[],\n    promptDescriptions: string[],\n    promptTemplates: string[] | PromptTemplate[],\n    defaultChain?: BaseChain,\n    options?: Omit<MultiRouteChainInput, \"defaultChain\">\n  ) {\n    return MultiPromptChain.fromLLMAndPrompts(llm, {\n      promptNames,\n      promptDescriptions,\n      promptTemplates,\n      defaultChain,\n      multiRouteChainOpts: options,\n    });\n  }\n\n  /**\n   * A static method that creates an instance of MultiPromptChain from a\n   * BaseLanguageModel and a set of prompts. It takes in optional parameters\n   * for the default chain and additional options.\n   * @param llm A BaseLanguageModel instance.\n   * @param promptNames An array of prompt names.\n   * @param promptDescriptions An array of prompt descriptions.\n   * @param promptTemplates An array of prompt templates.\n   * @param defaultChain An optional BaseChain instance to be used as the default chain.\n   * @param llmChainOpts Optional parameters for the LLMChainInput, excluding 'llm' and 'prompt'.\n   * @param conversationChainOpts Optional parameters for the LLMChainInput, excluding 'llm' and 'outputKey'.\n   * @param multiRouteChainOpts Optional parameters for the MultiRouteChainInput, excluding 'defaultChain'.\n   * @returns An instance of MultiPromptChain.\n   */\n  static fromLLMAndPrompts(\n    llm: BaseLanguageModelInterface,\n    {\n      promptNames,\n      promptDescriptions,\n      promptTemplates,\n      defaultChain,\n      llmChainOpts,\n      conversationChainOpts,\n      multiRouteChainOpts,\n    }: {\n      promptNames: string[];\n      promptDescriptions: string[];\n      promptTemplates: string[] | PromptTemplate[];\n      defaultChain?: BaseChain;\n      llmChainOpts?: Omit<LLMChainInput, \"llm\" | \"prompt\">;\n      conversationChainOpts?: Omit<LLMChainInput, \"llm\" | \"outputKey\">;\n      multiRouteChainOpts?: Omit<MultiRouteChainInput, \"defaultChain\">;\n    }\n  ): MultiPromptChain {\n    const destinations = zipEntries(promptNames, promptDescriptions).map(\n      ([name, desc]) => `${name}: ${desc}`\n    );\n\n    const structuredOutputParserSchema = z.object({\n      destination: z\n        .string()\n        .optional()\n        .describe('name of the question answering system to use or \"DEFAULT\"'),\n      next_inputs: z\n        .object({\n          input: z\n            .string()\n            .describe(\"a potentially modified version of the original input\"),\n        })\n        .describe(\"input to be fed to the next model\"),\n    });\n\n    const outputParser = new RouterOutputParser(structuredOutputParserSchema);\n\n    const destinationsStr = destinations.join(\"\\n\");\n    const routerTemplate = interpolateFString(\n      STRUCTURED_MULTI_PROMPT_ROUTER_TEMPLATE(\n        outputParser.getFormatInstructions({ interpolationDepth: 4 })\n      ),\n      {\n        destinations: destinationsStr,\n      }\n    );\n\n    const routerPrompt = new PromptTemplate({\n      template: routerTemplate,\n      inputVariables: [\"input\"],\n      outputParser,\n    });\n\n    const routerChain = LLMRouterChain.fromLLM(llm, routerPrompt);\n    const destinationChains = zipEntries<[string, string | PromptTemplate]>(\n      promptNames,\n      promptTemplates\n    ).reduce(\n      (acc, [name, template]) => {\n        let myPrompt: string | PromptTemplate;\n        if (typeof template === \"object\") {\n          myPrompt = template;\n        } else if (typeof template === \"string\") {\n          myPrompt = new PromptTemplate({\n            template: template as string,\n            inputVariables: [\"input\"],\n          });\n        } else {\n          throw new Error(\"Invalid prompt template\");\n        }\n        acc[name as string] = new LLMChain({\n          ...llmChainOpts,\n          llm,\n          prompt: myPrompt,\n        });\n        return acc;\n      },\n      {} as { [name: string]: LLMChain }\n    );\n\n    const convChain = new ConversationChain({\n      ...conversationChainOpts,\n      llm,\n      outputKey: \"text\",\n    });\n\n    return new MultiPromptChain({\n      ...multiRouteChainOpts,\n      routerChain,\n      destinationChains,\n      defaultChain: defaultChain ?? convChain,\n    });\n  }\n\n  _chainType(): string {\n    return \"multi_prompt_chain\";\n  }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAuCA,IAAa,mBAAb,MAAa,yBAAyB,gBAAgB;;;;CAIpD,OAAO,YACL,KACA,aACA,oBACA,iBACA,cACA,SACA;AACA,SAAO,iBAAiB,kBAAkB,KAAK;GAC7C;GACA;GACA;GACA;GACA,qBAAqB;GACtB,CAAC;;;;;;;;;;;;;;;;CAiBJ,OAAO,kBACL,KACA,EACE,aACA,oBACA,iBACA,cACA,cACA,uBACA,uBAUgB;EAClB,MAAM,eAAe,WAAW,aAAa,mBAAmB,CAAC,KAC9D,CAAC,MAAM,UAAU,GAAG,KAAK,IAAI,OAC/B;EAgBD,MAAM,eAAe,IAAI,mBAdY,EAAE,OAAO;GAC5C,aAAa,EACV,QAAQ,CACR,UAAU,CACV,SAAS,8DAA4D;GACxE,aAAa,EACV,OAAO,EACN,OAAO,EACJ,QAAQ,CACR,SAAS,uDAAuD,EACpE,CAAC,CACD,SAAS,oCAAoC;GACjD,CAAC,CAEuE;EAEzE,MAAM,kBAAkB,aAAa,KAAK,KAAK;EAU/C,MAAM,eAAe,IAAI,eAAe;GACtC,UAVqB,mBACrB,wCACE,aAAa,sBAAsB,EAAE,oBAAoB,GAAG,CAAC,CAC9D,EACD,EACE,cAAc,iBACf,CACF;GAIC,gBAAgB,CAAC,QAAQ;GACzB;GACD,CAAC;EAEF,MAAM,cAAc,eAAe,QAAQ,KAAK,aAAa;EAC7D,MAAM,oBAAoB,WACxB,aACA,gBACD,CAAC,QACC,KAAK,CAAC,MAAM,cAAc;GACzB,IAAI;AACJ,OAAI,OAAO,aAAa,SACtB,YAAW;YACF,OAAO,aAAa,SAC7B,YAAW,IAAI,eAAe;IAClB;IACV,gBAAgB,CAAC,QAAQ;IAC1B,CAAC;OAEF,OAAM,IAAI,MAAM,0BAA0B;AAE5C,OAAI,QAAkB,IAAI,SAAS;IACjC,GAAG;IACH;IACA,QAAQ;IACT,CAAC;AACF,UAAO;KAET,EAAE,CACH;EAED,MAAM,YAAY,IAAI,kBAAkB;GACtC,GAAG;GACH;GACA,WAAW;GACZ,CAAC;AAEF,SAAO,IAAI,iBAAiB;GAC1B,GAAG;GACH;GACA;GACA,cAAc,gBAAgB;GAC/B,CAAC;;CAGJ,aAAqB;AACnB,SAAO"}
+{"version":3,"file":"multi_prompt.js","names":[],"sources":["../../../src/chains/router/multi_prompt.ts"],"sourcesContent":["import type { BaseLanguageModelInterface } from \"@langchain/core/language_models/base\";\nimport { z } from \"zod/v3\";\nimport { interpolateFString, PromptTemplate } from \"@langchain/core/prompts\";\nimport { MultiRouteChain, MultiRouteChainInput } from \"./multi_route.js\";\nimport { STRUCTURED_MULTI_PROMPT_ROUTER_TEMPLATE } from \"./multi_prompt_prompt.js\";\nimport { BaseChain } from \"../../chains/base.js\";\nimport { LLMChain, LLMChainInput } from \"../../chains/llm_chain.js\";\nimport { LLMRouterChain } from \"./llm_router.js\";\nimport { ConversationChain } from \"../../chains/conversation.js\";\nimport { zipEntries } from \"./utils.js\";\nimport { RouterOutputParser } from \"../../output_parsers/router.js\";\n\n/**\n * A class that represents a multi-prompt chain in the LangChain\n * framework. It extends the MultiRouteChain class and provides additional\n * functionality specific to multi-prompt chains.\n * @example\n * ```typescript\n * const multiPromptChain = MultiPromptChain.fromLLMAndPrompts(\n *   new ChatOpenAI({ model: \"gpt-4o-mini\" }),\n *   {\n *     promptNames: [\"physics\", \"math\", \"history\"],\n *     promptDescriptions: [\n *       \"Good for answering questions about physics\",\n *       \"Good for answering math questions\",\n *       \"Good for answering questions about history\",\n *     ],\n *     promptTemplates: [\n *       `You are a very smart physics professor. Here is a question:\\n{input}\\n`,\n *       `You are a very good mathematician. Here is a question:\\n{input}\\n`,\n *       `You are a very smart history professor. Here is a question:\\n{input}\\n`,\n *     ],\n *   }\n * );\n * const result = await multiPromptChain.call({\n *   input: \"What is the speed of light?\",\n * });\n * ```\n */\nexport class MultiPromptChain extends MultiRouteChain {\n  /**\n   * @deprecated Use `fromLLMAndPrompts` instead\n   */\n  static fromPrompts(\n    llm: BaseLanguageModelInterface,\n    promptNames: string[],\n    promptDescriptions: string[],\n    promptTemplates: string[] | PromptTemplate[],\n    defaultChain?: BaseChain,\n    options?: Omit<MultiRouteChainInput, \"defaultChain\">\n  ) {\n    return MultiPromptChain.fromLLMAndPrompts(llm, {\n      promptNames,\n      promptDescriptions,\n      promptTemplates,\n      defaultChain,\n      multiRouteChainOpts: options,\n    });\n  }\n\n  /**\n   * A static method that creates an instance of MultiPromptChain from a\n   * BaseLanguageModel and a set of prompts. It takes in optional parameters\n   * for the default chain and additional options.\n   * @param llm A BaseLanguageModel instance.\n   * @param promptNames An array of prompt names.\n   * @param promptDescriptions An array of prompt descriptions.\n   * @param promptTemplates An array of prompt templates.\n   * @param defaultChain An optional BaseChain instance to be used as the default chain.\n   * @param llmChainOpts Optional parameters for the LLMChainInput, excluding 'llm' and 'prompt'.\n   * @param conversationChainOpts Optional parameters for the LLMChainInput, excluding 'llm' and 'outputKey'.\n   * @param multiRouteChainOpts Optional parameters for the MultiRouteChainInput, excluding 'defaultChain'.\n   * @returns An instance of MultiPromptChain.\n   */\n  static fromLLMAndPrompts(\n    llm: BaseLanguageModelInterface,\n    {\n      promptNames,\n      promptDescriptions,\n      promptTemplates,\n      defaultChain,\n      llmChainOpts,\n      conversationChainOpts,\n      multiRouteChainOpts,\n    }: {\n      promptNames: string[];\n      promptDescriptions: string[];\n      promptTemplates: string[] | PromptTemplate[];\n      defaultChain?: BaseChain;\n      llmChainOpts?: Omit<LLMChainInput, \"llm\" | \"prompt\">;\n      conversationChainOpts?: Omit<LLMChainInput, \"llm\" | \"outputKey\">;\n      multiRouteChainOpts?: Omit<MultiRouteChainInput, \"defaultChain\">;\n    }\n  ): MultiPromptChain {\n    const destinations = zipEntries(promptNames, promptDescriptions).map(\n      ([name, desc]) => `${name}: ${desc}`\n    );\n\n    const structuredOutputParserSchema = z.object({\n      destination: z\n        .string()\n        .optional()\n        .describe('name of the question answering system to use or \"DEFAULT\"'),\n      next_inputs: z\n        .object({\n          input: z\n            .string()\n            .describe(\"a potentially modified version of the original input\"),\n        })\n        .describe(\"input to be fed to the next model\"),\n    });\n\n    const outputParser = new RouterOutputParser(structuredOutputParserSchema);\n\n    const destinationsStr = destinations.join(\"\\n\");\n    const routerTemplate = interpolateFString(\n      STRUCTURED_MULTI_PROMPT_ROUTER_TEMPLATE(\n        outputParser.getFormatInstructions({ interpolationDepth: 4 })\n      ),\n      {\n        destinations: destinationsStr,\n      }\n    );\n\n    const routerPrompt = new PromptTemplate({\n      template: routerTemplate,\n      inputVariables: [\"input\"],\n      outputParser,\n    });\n\n    const routerChain = LLMRouterChain.fromLLM(llm, routerPrompt);\n    const destinationChains = zipEntries<[string, string | PromptTemplate]>(\n      promptNames,\n      promptTemplates\n    ).reduce(\n      (acc, [name, template]) => {\n        let myPrompt: string | PromptTemplate;\n        if (typeof template === \"object\") {\n          myPrompt = template;\n        } else if (typeof template === \"string\") {\n          myPrompt = new PromptTemplate({\n            template: template as string,\n            inputVariables: [\"input\"],\n          });\n        } else {\n          throw new Error(\"Invalid prompt template\");\n        }\n        acc[name as string] = new LLMChain({\n          ...llmChainOpts,\n          llm,\n          prompt: myPrompt,\n        });\n        return acc;\n      },\n      {} as { [name: string]: LLMChain }\n    );\n\n    const convChain = new ConversationChain({\n      ...conversationChainOpts,\n      llm,\n      outputKey: \"text\",\n    });\n\n    return new MultiPromptChain({\n      ...multiRouteChainOpts,\n      routerChain,\n      destinationChains,\n      defaultChain: defaultChain ?? convChain,\n    });\n  }\n\n  _chainType(): string {\n    return \"multi_prompt_chain\";\n  }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAuCA,IAAa,mBAAb,MAAa,yBAAyB,gBAAgB;;;;CAIpD,OAAO,YACL,KACA,aACA,oBACA,iBACA,cACA,SACA;AACA,SAAO,iBAAiB,kBAAkB,KAAK;GAC7C;GACA;GACA;GACA;GACA,qBAAqB;GACtB,CAAC;;;;;;;;;;;;;;;;CAiBJ,OAAO,kBACL,KACA,EACE,aACA,oBACA,iBACA,cACA,cACA,uBACA,uBAUgB;EAClB,MAAM,eAAe,WAAW,aAAa,mBAAmB,CAAC,KAC9D,CAAC,MAAM,UAAU,GAAG,KAAK,IAAI,OAC/B;EAgBD,MAAM,eAAe,IAAI,mBAdY,EAAE,OAAO;GAC5C,aAAa,EACV,QAAQ,CACR,UAAU,CACV,SAAS,8DAA4D;GACxE,aAAa,EACV,OAAO,EACN,OAAO,EACJ,QAAQ,CACR,SAAS,uDAAuD,EACpE,CAAC,CACD,SAAS,oCAAoC;GACjD,CAEuE,CAAC;EAEzE,MAAM,kBAAkB,aAAa,KAAK,KAAK;EAU/C,MAAM,eAAe,IAAI,eAAe;GACtC,UAVqB,mBACrB,wCACE,aAAa,sBAAsB,EAAE,oBAAoB,GAAG,CAAC,CAC9D,EACD,EACE,cAAc,iBACf,CAIuB;GACxB,gBAAgB,CAAC,QAAQ;GACzB;GACD,CAAC;EAEF,MAAM,cAAc,eAAe,QAAQ,KAAK,aAAa;EAC7D,MAAM,oBAAoB,WACxB,aACA,gBACD,CAAC,QACC,KAAK,CAAC,MAAM,cAAc;GACzB,IAAI;AACJ,OAAI,OAAO,aAAa,SACtB,YAAW;YACF,OAAO,aAAa,SAC7B,YAAW,IAAI,eAAe;IAClB;IACV,gBAAgB,CAAC,QAAQ;IAC1B,CAAC;OAEF,OAAM,IAAI,MAAM,0BAA0B;AAE5C,OAAI,QAAkB,IAAI,SAAS;IACjC,GAAG;IACH;IACA,QAAQ;IACT,CAAC;AACF,UAAO;KAET,EAAE,CACH;EAED,MAAM,YAAY,IAAI,kBAAkB;GACtC,GAAG;GACH;GACA,WAAW;GACZ,CAAC;AAEF,SAAO,IAAI,iBAAiB;GAC1B,GAAG;GACH;GACA;GACA,cAAc,gBAAgB;GAC/B,CAAC;;CAGJ,aAAqB;AACnB,SAAO"}

package/dist/chains/router/multi_retrieval_qa.cjs.map

@@ -1 +1 @@
-{"version":3,"file":"multi_retrieval_qa.cjs","names":["MultiRouteChain","zipEntries","RouterOutputParser","z","PromptTemplate","STRUCTURED_MULTI_RETRIEVAL_ROUTER_TEMPLATE","LLMRouterChain","RetrievalQAChain","ConversationChain","DEFAULT_TEMPLATE"],"sources":["../../../src/chains/router/multi_retrieval_qa.ts"],"sourcesContent":["import { z } from \"zod/v3\";\nimport type { BaseLanguageModelInterface } from \"@langchain/core/language_models/base\";\nimport type { BaseRetrieverInterface } from \"@langchain/core/retrievers\";\nimport { interpolateFString, PromptTemplate } from \"@langchain/core/prompts\";\nimport { MultiRouteChain, MultiRouteChainInput } from \"./multi_route.js\";\nimport { BaseChain } from \"../../chains/base.js\";\nimport { LLMRouterChain } from \"./llm_router.js\";\nimport {\n  ConversationChain,\n  DEFAULT_TEMPLATE,\n} from \"../../chains/conversation.js\";\nimport { STRUCTURED_MULTI_RETRIEVAL_ROUTER_TEMPLATE } from \"./multi_retrieval_prompt.js\";\nimport { zipEntries } from \"./utils.js\";\nimport {\n  RetrievalQAChain,\n  RetrievalQAChainInput,\n} from \"../../chains/retrieval_qa.js\";\nimport { RouterOutputParser } from \"../../output_parsers/router.js\";\n\n/**\n * A type that represents the default values for the MultiRetrievalQAChain\n * class. It includes optional properties for the default retriever,\n * default prompt, and default chain.\n */\nexport type MultiRetrievalDefaults = {\n  defaultRetriever?: BaseRetrieverInterface;\n  defaultPrompt?: PromptTemplate;\n  defaultChain?: BaseChain;\n};\n\n/**\n * A class that represents a multi-retrieval question answering chain in\n * the LangChain framework. It extends the MultiRouteChain class and\n * provides additional functionality specific to multi-retrieval QA\n * chains.\n * @example\n * ```typescript\n * const multiRetrievalQAChain = MultiRetrievalQAChain.fromLLMAndRetrievers(\n *   new ChatOpenAI({ model: \"gpt-4o-mini\" }),\n *   {\n *     retrieverNames: [\"aqua teen\", \"mst3k\", \"animaniacs\"],\n *     retrieverDescriptions: [\n *       \"Good for answering questions about Aqua Teen Hunger Force theme song\",\n *       \"Good for answering questions about Mystery Science Theater 3000 theme song\",\n *       \"Good for answering questions about Animaniacs theme song\",\n *     ],\n *     retrievers: [\n *       new MemoryVectorStore().asRetriever(3),\n *       new MemoryVectorStore().asRetriever(3),\n *       new MemoryVectorStore().asRetriever(3),\n *     ],\n *     retrievalQAChainOpts: {\n *       returnSourceDocuments: true,\n *     },\n *   },\n * );\n *\n * const result = await multiRetrievalQAChain.call({\n *   input:\n *     \"In the Aqua Teen Hunger Force theme song, who calls himself the mike rula?\",\n * });\n *\n * console.log(result.sourceDocuments, result.text);\n * ```\n */\nexport class MultiRetrievalQAChain extends MultiRouteChain {\n  get outputKeys(): string[] {\n    return [\"result\"];\n  }\n\n  /**\n   * @deprecated Use `fromRetrieversAndPrompts` instead\n   */\n  static fromRetrievers(\n    llm: BaseLanguageModelInterface,\n    retrieverNames: string[],\n    retrieverDescriptions: string[],\n    retrievers: BaseRetrieverInterface[],\n    retrieverPrompts?: PromptTemplate[],\n    defaults?: MultiRetrievalDefaults,\n    options?: Omit<MultiRouteChainInput, \"defaultChain\">\n  ) {\n    return MultiRetrievalQAChain.fromLLMAndRetrievers(llm, {\n      retrieverNames,\n      retrieverDescriptions,\n      retrievers,\n      retrieverPrompts,\n      defaults,\n      multiRetrievalChainOpts: options,\n    });\n  }\n\n  /**\n   * A static method that creates an instance of MultiRetrievalQAChain from\n   * a BaseLanguageModel and a set of retrievers. It takes in optional\n   * parameters for the retriever names, descriptions, prompts, defaults,\n   * and additional options. It is an alternative method to fromRetrievers\n   * and provides more flexibility in configuring the underlying chains.\n   * @param llm A BaseLanguageModel instance.\n   * @param retrieverNames An array of retriever names.\n   * @param retrieverDescriptions An array of retriever descriptions.\n   * @param retrievers An array of BaseRetrieverInterface instances.\n   * @param retrieverPrompts An optional array of PromptTemplate instances for the retrievers.\n   * @param defaults An optional MultiRetrievalDefaults instance.\n   * @param multiRetrievalChainOpts Additional optional parameters for the multi-retrieval chain.\n   * @param retrievalQAChainOpts Additional optional parameters for the retrieval QA chain.\n   * @returns A new instance of MultiRetrievalQAChain.\n   */\n  static fromLLMAndRetrievers(\n    llm: BaseLanguageModelInterface,\n    {\n      retrieverNames,\n      retrieverDescriptions,\n      retrievers,\n      retrieverPrompts,\n      defaults,\n      multiRetrievalChainOpts,\n      retrievalQAChainOpts,\n    }: {\n      retrieverNames: string[];\n      retrieverDescriptions: string[];\n      retrievers: BaseRetrieverInterface[];\n      retrieverPrompts?: PromptTemplate[];\n      defaults?: MultiRetrievalDefaults;\n      multiRetrievalChainOpts?: Omit<MultiRouteChainInput, \"defaultChain\">;\n      retrievalQAChainOpts?: Partial<\n        Omit<RetrievalQAChainInput, \"retriever\" | \"combineDocumentsChain\">\n      > & {\n        prompt?: PromptTemplate;\n      };\n    }\n  ): MultiRetrievalQAChain {\n    const { defaultRetriever, defaultPrompt, defaultChain } = defaults ?? {};\n    if (defaultPrompt && !defaultRetriever) {\n      throw new Error(\n        \"`default_retriever` must be specified if `default_prompt` is \\nprovided. Received only `default_prompt`.\"\n      );\n    }\n    const destinations = zipEntries<[string, string]>(\n      retrieverNames,\n      retrieverDescriptions\n    ).map(([name, desc]) => `${name}: ${desc}`);\n\n    const structuredOutputParserSchema = z.object({\n      destination: z\n        .string()\n        .optional()\n        .describe('name of the question answering system to use or \"DEFAULT\"'),\n      next_inputs: z\n        .object({\n          query: z\n            .string()\n            .describe(\"a potentially modified version of the original input\"),\n        })\n        .describe(\"input to be fed to the next model\"),\n    });\n\n    const outputParser = new RouterOutputParser<\n      typeof structuredOutputParserSchema\n    >(structuredOutputParserSchema);\n\n    const destinationsStr = destinations.join(\"\\n\");\n    const routerTemplate = interpolateFString(\n      STRUCTURED_MULTI_RETRIEVAL_ROUTER_TEMPLATE(\n        outputParser.getFormatInstructions({ interpolationDepth: 4 })\n      ),\n      {\n        destinations: destinationsStr,\n      }\n    );\n    const routerPrompt = new PromptTemplate({\n      template: routerTemplate,\n      inputVariables: [\"input\"],\n      outputParser,\n    });\n\n    const routerChain = LLMRouterChain.fromLLM(llm, routerPrompt);\n    const prompts = retrieverPrompts ?? retrievers.map(() => null);\n    const destinationChains = zipEntries<\n      [string, BaseRetrieverInterface, PromptTemplate | null]\n    >(retrieverNames, retrievers, prompts).reduce(\n      (acc, [name, retriever, prompt]) => {\n        const opt: Partial<RetrievalQAChainInput> & {\n          prompt?: PromptTemplate;\n        } = retrievalQAChainOpts ?? {};\n        if (prompt) {\n          opt.prompt = prompt;\n        }\n        acc[name] = RetrievalQAChain.fromLLM(llm, retriever, opt);\n        return acc;\n      },\n      {} as { [name: string]: RetrievalQAChain }\n    );\n\n    let _defaultChain;\n    if (defaultChain) {\n      _defaultChain = defaultChain;\n    } else if (defaultRetriever) {\n      _defaultChain = RetrievalQAChain.fromLLM(llm, defaultRetriever, {\n        ...retrievalQAChainOpts,\n        prompt: defaultPrompt,\n      });\n    } else {\n      const promptTemplate = DEFAULT_TEMPLATE.replace(\"input\", \"query\");\n      const prompt = new PromptTemplate({\n        template: promptTemplate,\n        inputVariables: [\"history\", \"query\"],\n      });\n      _defaultChain = new ConversationChain({\n        llm,\n        prompt,\n        outputKey: \"result\",\n      });\n    }\n\n    return new MultiRetrievalQAChain({\n      ...multiRetrievalChainOpts,\n      routerChain,\n      destinationChains,\n      defaultChain: _defaultChain,\n    });\n  }\n\n  _chainType(): string {\n    return \"multi_retrieval_qa_chain\";\n  }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAiEA,IAAa,wBAAb,MAAa,8BAA8BA,oBAAAA,gBAAgB;CACzD,IAAI,aAAuB;AACzB,SAAO,CAAC,SAAS;;;;;CAMnB,OAAO,eACL,KACA,gBACA,uBACA,YACA,kBACA,UACA,SACA;AACA,SAAO,sBAAsB,qBAAqB,KAAK;GACrD;GACA;GACA;GACA;GACA;GACA,yBAAyB;GAC1B,CAAC;;;;;;;;;;;;;;;;;;CAmBJ,OAAO,qBACL,KACA,EACE,gBACA,uBACA,YACA,kBACA,UACA,yBACA,wBAcqB;EACvB,MAAM,EAAE,kBAAkB,eAAe,iBAAiB,YAAY,EAAE;AACxE,MAAI,iBAAiB,CAAC,iBACpB,OAAM,IAAI,MACR,2GACD;EAEH,MAAM,eAAeC,cAAAA,WACnB,gBACA,sBACD,CAAC,KAAK,CAAC,MAAM,UAAU,GAAG,KAAK,IAAI,OAAO;EAgB3C,MAAM,eAAe,IAAIC,eAAAA,mBAdYC,OAAAA,EAAE,OAAO;GAC5C,aAAaA,OAAAA,EACV,QAAQ,CACR,UAAU,CACV,SAAS,8DAA4D;GACxE,aAAaA,OAAAA,EACV,OAAO,EACN,OAAOA,OAAAA,EACJ,QAAQ,CACR,SAAS,uDAAuD,EACpE,CAAC,CACD,SAAS,oCAAoC;GACjD,CAAC,CAI6B;EAE/B,MAAM,kBAAkB,aAAa,KAAK,KAAK;EAS/C,MAAM,eAAe,IAAIC,wBAAAA,eAAe;GACtC,WAAA,GAAA,wBAAA,oBARAC,+BAAAA,2CACE,aAAa,sBAAsB,EAAE,oBAAoB,GAAG,CAAC,CAC9D,EACD,EACE,cAAc,iBACf,CACF;GAGC,gBAAgB,CAAC,QAAQ;GACzB;GACD,CAAC;EAEF,MAAM,cAAcC,mBAAAA,eAAe,QAAQ,KAAK,aAAa;EAE7D,MAAM,oBAAoBL,cAAAA,WAExB,gBAAgB,YAHF,oBAAoB,WAAW,UAAU,KAAK,CAGxB,CAAC,QACpC,KAAK,CAAC,MAAM,WAAW,YAAY;GAClC,MAAM,MAEF,wBAAwB,EAAE;AAC9B,OAAI,OACF,KAAI,SAAS;AAEf,OAAI,QAAQM,qBAAAA,iBAAiB,QAAQ,KAAK,WAAW,IAAI;AACzD,UAAO;KAET,EAAE,CACH;EAED,IAAI;AACJ,MAAI,aACF,iBAAgB;WACP,iBACT,iBAAgBA,qBAAAA,iBAAiB,QAAQ,KAAK,kBAAkB;GAC9D,GAAG;GACH,QAAQ;GACT,CAAC;MAOF,iBAAgB,IAAIC,qBAAAA,kBAAkB;GACpC;GACA,QANa,IAAIJ,wBAAAA,eAAe;IAChC,UAFqBK,qBAAAA,iBAAiB,QAAQ,SAAS,QAAQ;IAG/D,gBAAgB,CAAC,WAAW,QAAQ;IACrC,CAAC;GAIA,WAAW;GACZ,CAAC;AAGJ,SAAO,IAAI,sBAAsB;GAC/B,GAAG;GACH;GACA;GACA,cAAc;GACf,CAAC;;CAGJ,aAAqB;AACnB,SAAO"}
+{"version":3,"file":"multi_retrieval_qa.cjs","names":["MultiRouteChain","zipEntries","RouterOutputParser","z","PromptTemplate","STRUCTURED_MULTI_RETRIEVAL_ROUTER_TEMPLATE","LLMRouterChain","RetrievalQAChain","ConversationChain","DEFAULT_TEMPLATE"],"sources":["../../../src/chains/router/multi_retrieval_qa.ts"],"sourcesContent":["import { z } from \"zod/v3\";\nimport type { BaseLanguageModelInterface } from \"@langchain/core/language_models/base\";\nimport type { BaseRetrieverInterface } from \"@langchain/core/retrievers\";\nimport { interpolateFString, PromptTemplate } from \"@langchain/core/prompts\";\nimport { MultiRouteChain, MultiRouteChainInput } from \"./multi_route.js\";\nimport { BaseChain } from \"../../chains/base.js\";\nimport { LLMRouterChain } from \"./llm_router.js\";\nimport {\n  ConversationChain,\n  DEFAULT_TEMPLATE,\n} from \"../../chains/conversation.js\";\nimport { STRUCTURED_MULTI_RETRIEVAL_ROUTER_TEMPLATE } from \"./multi_retrieval_prompt.js\";\nimport { zipEntries } from \"./utils.js\";\nimport {\n  RetrievalQAChain,\n  RetrievalQAChainInput,\n} from \"../../chains/retrieval_qa.js\";\nimport { RouterOutputParser } from \"../../output_parsers/router.js\";\n\n/**\n * A type that represents the default values for the MultiRetrievalQAChain\n * class. It includes optional properties for the default retriever,\n * default prompt, and default chain.\n */\nexport type MultiRetrievalDefaults = {\n  defaultRetriever?: BaseRetrieverInterface;\n  defaultPrompt?: PromptTemplate;\n  defaultChain?: BaseChain;\n};\n\n/**\n * A class that represents a multi-retrieval question answering chain in\n * the LangChain framework. It extends the MultiRouteChain class and\n * provides additional functionality specific to multi-retrieval QA\n * chains.\n * @example\n * ```typescript\n * const multiRetrievalQAChain = MultiRetrievalQAChain.fromLLMAndRetrievers(\n *   new ChatOpenAI({ model: \"gpt-4o-mini\" }),\n *   {\n *     retrieverNames: [\"aqua teen\", \"mst3k\", \"animaniacs\"],\n *     retrieverDescriptions: [\n *       \"Good for answering questions about Aqua Teen Hunger Force theme song\",\n *       \"Good for answering questions about Mystery Science Theater 3000 theme song\",\n *       \"Good for answering questions about Animaniacs theme song\",\n *     ],\n *     retrievers: [\n *       new MemoryVectorStore().asRetriever(3),\n *       new MemoryVectorStore().asRetriever(3),\n *       new MemoryVectorStore().asRetriever(3),\n *     ],\n *     retrievalQAChainOpts: {\n *       returnSourceDocuments: true,\n *     },\n *   },\n * );\n *\n * const result = await multiRetrievalQAChain.call({\n *   input:\n *     \"In the Aqua Teen Hunger Force theme song, who calls himself the mike rula?\",\n * });\n *\n * console.log(result.sourceDocuments, result.text);\n * ```\n */\nexport class MultiRetrievalQAChain extends MultiRouteChain {\n  get outputKeys(): string[] {\n    return [\"result\"];\n  }\n\n  /**\n   * @deprecated Use `fromRetrieversAndPrompts` instead\n   */\n  static fromRetrievers(\n    llm: BaseLanguageModelInterface,\n    retrieverNames: string[],\n    retrieverDescriptions: string[],\n    retrievers: BaseRetrieverInterface[],\n    retrieverPrompts?: PromptTemplate[],\n    defaults?: MultiRetrievalDefaults,\n    options?: Omit<MultiRouteChainInput, \"defaultChain\">\n  ) {\n    return MultiRetrievalQAChain.fromLLMAndRetrievers(llm, {\n      retrieverNames,\n      retrieverDescriptions,\n      retrievers,\n      retrieverPrompts,\n      defaults,\n      multiRetrievalChainOpts: options,\n    });\n  }\n\n  /**\n   * A static method that creates an instance of MultiRetrievalQAChain from\n   * a BaseLanguageModel and a set of retrievers. It takes in optional\n   * parameters for the retriever names, descriptions, prompts, defaults,\n   * and additional options. It is an alternative method to fromRetrievers\n   * and provides more flexibility in configuring the underlying chains.\n   * @param llm A BaseLanguageModel instance.\n   * @param retrieverNames An array of retriever names.\n   * @param retrieverDescriptions An array of retriever descriptions.\n   * @param retrievers An array of BaseRetrieverInterface instances.\n   * @param retrieverPrompts An optional array of PromptTemplate instances for the retrievers.\n   * @param defaults An optional MultiRetrievalDefaults instance.\n   * @param multiRetrievalChainOpts Additional optional parameters for the multi-retrieval chain.\n   * @param retrievalQAChainOpts Additional optional parameters for the retrieval QA chain.\n   * @returns A new instance of MultiRetrievalQAChain.\n   */\n  static fromLLMAndRetrievers(\n    llm: BaseLanguageModelInterface,\n    {\n      retrieverNames,\n      retrieverDescriptions,\n      retrievers,\n      retrieverPrompts,\n      defaults,\n      multiRetrievalChainOpts,\n      retrievalQAChainOpts,\n    }: {\n      retrieverNames: string[];\n      retrieverDescriptions: string[];\n      retrievers: BaseRetrieverInterface[];\n      retrieverPrompts?: PromptTemplate[];\n      defaults?: MultiRetrievalDefaults;\n      multiRetrievalChainOpts?: Omit<MultiRouteChainInput, \"defaultChain\">;\n      retrievalQAChainOpts?: Partial<\n        Omit<RetrievalQAChainInput, \"retriever\" | \"combineDocumentsChain\">\n      > & {\n        prompt?: PromptTemplate;\n      };\n    }\n  ): MultiRetrievalQAChain {\n    const { defaultRetriever, defaultPrompt, defaultChain } = defaults ?? {};\n    if (defaultPrompt && !defaultRetriever) {\n      throw new Error(\n        \"`default_retriever` must be specified if `default_prompt` is \\nprovided. Received only `default_prompt`.\"\n      );\n    }\n    const destinations = zipEntries<[string, string]>(\n      retrieverNames,\n      retrieverDescriptions\n    ).map(([name, desc]) => `${name}: ${desc}`);\n\n    const structuredOutputParserSchema = z.object({\n      destination: z\n        .string()\n        .optional()\n        .describe('name of the question answering system to use or \"DEFAULT\"'),\n      next_inputs: z\n        .object({\n          query: z\n            .string()\n            .describe(\"a potentially modified version of the original input\"),\n        })\n        .describe(\"input to be fed to the next model\"),\n    });\n\n    const outputParser = new RouterOutputParser<\n      typeof structuredOutputParserSchema\n    >(structuredOutputParserSchema);\n\n    const destinationsStr = destinations.join(\"\\n\");\n    const routerTemplate = interpolateFString(\n      STRUCTURED_MULTI_RETRIEVAL_ROUTER_TEMPLATE(\n        outputParser.getFormatInstructions({ interpolationDepth: 4 })\n      ),\n      {\n        destinations: destinationsStr,\n      }\n    );\n    const routerPrompt = new PromptTemplate({\n      template: routerTemplate,\n      inputVariables: [\"input\"],\n      outputParser,\n    });\n\n    const routerChain = LLMRouterChain.fromLLM(llm, routerPrompt);\n    const prompts = retrieverPrompts ?? retrievers.map(() => null);\n    const destinationChains = zipEntries<\n      [string, BaseRetrieverInterface, PromptTemplate | null]\n    >(retrieverNames, retrievers, prompts).reduce(\n      (acc, [name, retriever, prompt]) => {\n        const opt: Partial<RetrievalQAChainInput> & {\n          prompt?: PromptTemplate;\n        } = retrievalQAChainOpts ?? {};\n        if (prompt) {\n          opt.prompt = prompt;\n        }\n        acc[name] = RetrievalQAChain.fromLLM(llm, retriever, opt);\n        return acc;\n      },\n      {} as { [name: string]: RetrievalQAChain }\n    );\n\n    let _defaultChain;\n    if (defaultChain) {\n      _defaultChain = defaultChain;\n    } else if (defaultRetriever) {\n      _defaultChain = RetrievalQAChain.fromLLM(llm, defaultRetriever, {\n        ...retrievalQAChainOpts,\n        prompt: defaultPrompt,\n      });\n    } else {\n      const promptTemplate = DEFAULT_TEMPLATE.replace(\"input\", \"query\");\n      const prompt = new PromptTemplate({\n        template: promptTemplate,\n        inputVariables: [\"history\", \"query\"],\n      });\n      _defaultChain = new ConversationChain({\n        llm,\n        prompt,\n        outputKey: \"result\",\n      });\n    }\n\n    return new MultiRetrievalQAChain({\n      ...multiRetrievalChainOpts,\n      routerChain,\n      destinationChains,\n      defaultChain: _defaultChain,\n    });\n  }\n\n  _chainType(): string {\n    return \"multi_retrieval_qa_chain\";\n  }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAiEA,IAAa,wBAAb,MAAa,8BAA8BA,oBAAAA,gBAAgB;CACzD,IAAI,aAAuB;AACzB,SAAO,CAAC,SAAS;;;;;CAMnB,OAAO,eACL,KACA,gBACA,uBACA,YACA,kBACA,UACA,SACA;AACA,SAAO,sBAAsB,qBAAqB,KAAK;GACrD;GACA;GACA;GACA;GACA;GACA,yBAAyB;GAC1B,CAAC;;;;;;;;;;;;;;;;;;CAmBJ,OAAO,qBACL,KACA,EACE,gBACA,uBACA,YACA,kBACA,UACA,yBACA,wBAcqB;EACvB,MAAM,EAAE,kBAAkB,eAAe,iBAAiB,YAAY,EAAE;AACxE,MAAI,iBAAiB,CAAC,iBACpB,OAAM,IAAI,MACR,2GACD;EAEH,MAAM,eAAeC,cAAAA,WACnB,gBACA,sBACD,CAAC,KAAK,CAAC,MAAM,UAAU,GAAG,KAAK,IAAI,OAAO;EAgB3C,MAAM,eAAe,IAAIC,eAAAA,mBAdYC,OAAAA,EAAE,OAAO;GAC5C,aAAaA,OAAAA,EACV,QAAQ,CACR,UAAU,CACV,SAAS,8DAA4D;GACxE,aAAaA,OAAAA,EACV,OAAO,EACN,OAAOA,OAAAA,EACJ,QAAQ,CACR,SAAS,uDAAuD,EACpE,CAAC,CACD,SAAS,oCAAoC;GACjD,CAI6B,CAAC;EAE/B,MAAM,kBAAkB,aAAa,KAAK,KAAK;EAS/C,MAAM,eAAe,IAAIC,wBAAAA,eAAe;GACtC,WAAA,GAAA,wBAAA,oBARAC,+BAAAA,2CACE,aAAa,sBAAsB,EAAE,oBAAoB,GAAG,CAAC,CAC9D,EACD,EACE,cAAc,iBACf,CAGuB;GACxB,gBAAgB,CAAC,QAAQ;GACzB;GACD,CAAC;EAEF,MAAM,cAAcC,mBAAAA,eAAe,QAAQ,KAAK,aAAa;EAE7D,MAAM,oBAAoBL,cAAAA,WAExB,gBAAgB,YAHF,oBAAoB,WAAW,UAAU,KAAK,CAGxB,CAAC,QACpC,KAAK,CAAC,MAAM,WAAW,YAAY;GAClC,MAAM,MAEF,wBAAwB,EAAE;AAC9B,OAAI,OACF,KAAI,SAAS;AAEf,OAAI,QAAQM,qBAAAA,iBAAiB,QAAQ,KAAK,WAAW,IAAI;AACzD,UAAO;KAET,EAAE,CACH;EAED,IAAI;AACJ,MAAI,aACF,iBAAgB;WACP,iBACT,iBAAgBA,qBAAAA,iBAAiB,QAAQ,KAAK,kBAAkB;GAC9D,GAAG;GACH,QAAQ;GACT,CAAC;MAOF,iBAAgB,IAAIC,qBAAAA,kBAAkB;GACpC;GACA,QAAA,IANiBJ,wBAAAA,eAAe;IAChC,UAFqBK,qBAAAA,iBAAiB,QAAQ,SAAS,QAE/B;IACxB,gBAAgB,CAAC,WAAW,QAAQ;IACrC,CAGO;GACN,WAAW;GACZ,CAAC;AAGJ,SAAO,IAAI,sBAAsB;GAC/B,GAAG;GACH;GACA;GACA,cAAc;GACf,CAAC;;CAGJ,aAAqB;AACnB,SAAO"}

package/dist/chains/router/multi_retrieval_qa.js.map

@@ -1 +1 @@
-{"version":3,"file":"multi_retrieval_qa.js","names":[],"sources":["../../../src/chains/router/multi_retrieval_qa.ts"],"sourcesContent":["import { z } from \"zod/v3\";\nimport type { BaseLanguageModelInterface } from \"@langchain/core/language_models/base\";\nimport type { BaseRetrieverInterface } from \"@langchain/core/retrievers\";\nimport { interpolateFString, PromptTemplate } from \"@langchain/core/prompts\";\nimport { MultiRouteChain, MultiRouteChainInput } from \"./multi_route.js\";\nimport { BaseChain } from \"../../chains/base.js\";\nimport { LLMRouterChain } from \"./llm_router.js\";\nimport {\n  ConversationChain,\n  DEFAULT_TEMPLATE,\n} from \"../../chains/conversation.js\";\nimport { STRUCTURED_MULTI_RETRIEVAL_ROUTER_TEMPLATE } from \"./multi_retrieval_prompt.js\";\nimport { zipEntries } from \"./utils.js\";\nimport {\n  RetrievalQAChain,\n  RetrievalQAChainInput,\n} from \"../../chains/retrieval_qa.js\";\nimport { RouterOutputParser } from \"../../output_parsers/router.js\";\n\n/**\n * A type that represents the default values for the MultiRetrievalQAChain\n * class. It includes optional properties for the default retriever,\n * default prompt, and default chain.\n */\nexport type MultiRetrievalDefaults = {\n  defaultRetriever?: BaseRetrieverInterface;\n  defaultPrompt?: PromptTemplate;\n  defaultChain?: BaseChain;\n};\n\n/**\n * A class that represents a multi-retrieval question answering chain in\n * the LangChain framework. It extends the MultiRouteChain class and\n * provides additional functionality specific to multi-retrieval QA\n * chains.\n * @example\n * ```typescript\n * const multiRetrievalQAChain = MultiRetrievalQAChain.fromLLMAndRetrievers(\n *   new ChatOpenAI({ model: \"gpt-4o-mini\" }),\n *   {\n *     retrieverNames: [\"aqua teen\", \"mst3k\", \"animaniacs\"],\n *     retrieverDescriptions: [\n *       \"Good for answering questions about Aqua Teen Hunger Force theme song\",\n *       \"Good for answering questions about Mystery Science Theater 3000 theme song\",\n *       \"Good for answering questions about Animaniacs theme song\",\n *     ],\n *     retrievers: [\n *       new MemoryVectorStore().asRetriever(3),\n *       new MemoryVectorStore().asRetriever(3),\n *       new MemoryVectorStore().asRetriever(3),\n *     ],\n *     retrievalQAChainOpts: {\n *       returnSourceDocuments: true,\n *     },\n *   },\n * );\n *\n * const result = await multiRetrievalQAChain.call({\n *   input:\n *     \"In the Aqua Teen Hunger Force theme song, who calls himself the mike rula?\",\n * });\n *\n * console.log(result.sourceDocuments, result.text);\n * ```\n */\nexport class MultiRetrievalQAChain extends MultiRouteChain {\n  get outputKeys(): string[] {\n    return [\"result\"];\n  }\n\n  /**\n   * @deprecated Use `fromRetrieversAndPrompts` instead\n   */\n  static fromRetrievers(\n    llm: BaseLanguageModelInterface,\n    retrieverNames: string[],\n    retrieverDescriptions: string[],\n    retrievers: BaseRetrieverInterface[],\n    retrieverPrompts?: PromptTemplate[],\n    defaults?: MultiRetrievalDefaults,\n    options?: Omit<MultiRouteChainInput, \"defaultChain\">\n  ) {\n    return MultiRetrievalQAChain.fromLLMAndRetrievers(llm, {\n      retrieverNames,\n      retrieverDescriptions,\n      retrievers,\n      retrieverPrompts,\n      defaults,\n      multiRetrievalChainOpts: options,\n    });\n  }\n\n  /**\n   * A static method that creates an instance of MultiRetrievalQAChain from\n   * a BaseLanguageModel and a set of retrievers. It takes in optional\n   * parameters for the retriever names, descriptions, prompts, defaults,\n   * and additional options. It is an alternative method to fromRetrievers\n   * and provides more flexibility in configuring the underlying chains.\n   * @param llm A BaseLanguageModel instance.\n   * @param retrieverNames An array of retriever names.\n   * @param retrieverDescriptions An array of retriever descriptions.\n   * @param retrievers An array of BaseRetrieverInterface instances.\n   * @param retrieverPrompts An optional array of PromptTemplate instances for the retrievers.\n   * @param defaults An optional MultiRetrievalDefaults instance.\n   * @param multiRetrievalChainOpts Additional optional parameters for the multi-retrieval chain.\n   * @param retrievalQAChainOpts Additional optional parameters for the retrieval QA chain.\n   * @returns A new instance of MultiRetrievalQAChain.\n   */\n  static fromLLMAndRetrievers(\n    llm: BaseLanguageModelInterface,\n    {\n      retrieverNames,\n      retrieverDescriptions,\n      retrievers,\n      retrieverPrompts,\n      defaults,\n      multiRetrievalChainOpts,\n      retrievalQAChainOpts,\n    }: {\n      retrieverNames: string[];\n      retrieverDescriptions: string[];\n      retrievers: BaseRetrieverInterface[];\n      retrieverPrompts?: PromptTemplate[];\n      defaults?: MultiRetrievalDefaults;\n      multiRetrievalChainOpts?: Omit<MultiRouteChainInput, \"defaultChain\">;\n      retrievalQAChainOpts?: Partial<\n        Omit<RetrievalQAChainInput, \"retriever\" | \"combineDocumentsChain\">\n      > & {\n        prompt?: PromptTemplate;\n      };\n    }\n  ): MultiRetrievalQAChain {\n    const { defaultRetriever, defaultPrompt, defaultChain } = defaults ?? {};\n    if (defaultPrompt && !defaultRetriever) {\n      throw new Error(\n        \"`default_retriever` must be specified if `default_prompt` is \\nprovided. Received only `default_prompt`.\"\n      );\n    }\n    const destinations = zipEntries<[string, string]>(\n      retrieverNames,\n      retrieverDescriptions\n    ).map(([name, desc]) => `${name}: ${desc}`);\n\n    const structuredOutputParserSchema = z.object({\n      destination: z\n        .string()\n        .optional()\n        .describe('name of the question answering system to use or \"DEFAULT\"'),\n      next_inputs: z\n        .object({\n          query: z\n            .string()\n            .describe(\"a potentially modified version of the original input\"),\n        })\n        .describe(\"input to be fed to the next model\"),\n    });\n\n    const outputParser = new RouterOutputParser<\n      typeof structuredOutputParserSchema\n    >(structuredOutputParserSchema);\n\n    const destinationsStr = destinations.join(\"\\n\");\n    const routerTemplate = interpolateFString(\n      STRUCTURED_MULTI_RETRIEVAL_ROUTER_TEMPLATE(\n        outputParser.getFormatInstructions({ interpolationDepth: 4 })\n      ),\n      {\n        destinations: destinationsStr,\n      }\n    );\n    const routerPrompt = new PromptTemplate({\n      template: routerTemplate,\n      inputVariables: [\"input\"],\n      outputParser,\n    });\n\n    const routerChain = LLMRouterChain.fromLLM(llm, routerPrompt);\n    const prompts = retrieverPrompts ?? retrievers.map(() => null);\n    const destinationChains = zipEntries<\n      [string, BaseRetrieverInterface, PromptTemplate | null]\n    >(retrieverNames, retrievers, prompts).reduce(\n      (acc, [name, retriever, prompt]) => {\n        const opt: Partial<RetrievalQAChainInput> & {\n          prompt?: PromptTemplate;\n        } = retrievalQAChainOpts ?? {};\n        if (prompt) {\n          opt.prompt = prompt;\n        }\n        acc[name] = RetrievalQAChain.fromLLM(llm, retriever, opt);\n        return acc;\n      },\n      {} as { [name: string]: RetrievalQAChain }\n    );\n\n    let _defaultChain;\n    if (defaultChain) {\n      _defaultChain = defaultChain;\n    } else if (defaultRetriever) {\n      _defaultChain = RetrievalQAChain.fromLLM(llm, defaultRetriever, {\n        ...retrievalQAChainOpts,\n        prompt: defaultPrompt,\n      });\n    } else {\n      const promptTemplate = DEFAULT_TEMPLATE.replace(\"input\", \"query\");\n      const prompt = new PromptTemplate({\n        template: promptTemplate,\n        inputVariables: [\"history\", \"query\"],\n      });\n      _defaultChain = new ConversationChain({\n        llm,\n        prompt,\n        outputKey: \"result\",\n      });\n    }\n\n    return new MultiRetrievalQAChain({\n      ...multiRetrievalChainOpts,\n      routerChain,\n      destinationChains,\n      defaultChain: _defaultChain,\n    });\n  }\n\n  _chainType(): string {\n    return \"multi_retrieval_qa_chain\";\n  }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAiEA,IAAa,wBAAb,MAAa,8BAA8B,gBAAgB;CACzD,IAAI,aAAuB;AACzB,SAAO,CAAC,SAAS;;;;;CAMnB,OAAO,eACL,KACA,gBACA,uBACA,YACA,kBACA,UACA,SACA;AACA,SAAO,sBAAsB,qBAAqB,KAAK;GACrD;GACA;GACA;GACA;GACA;GACA,yBAAyB;GAC1B,CAAC;;;;;;;;;;;;;;;;;;CAmBJ,OAAO,qBACL,KACA,EACE,gBACA,uBACA,YACA,kBACA,UACA,yBACA,wBAcqB;EACvB,MAAM,EAAE,kBAAkB,eAAe,iBAAiB,YAAY,EAAE;AACxE,MAAI,iBAAiB,CAAC,iBACpB,OAAM,IAAI,MACR,2GACD;EAEH,MAAM,eAAe,WACnB,gBACA,sBACD,CAAC,KAAK,CAAC,MAAM,UAAU,GAAG,KAAK,IAAI,OAAO;EAgB3C,MAAM,eAAe,IAAI,mBAdY,EAAE,OAAO;GAC5C,aAAa,EACV,QAAQ,CACR,UAAU,CACV,SAAS,8DAA4D;GACxE,aAAa,EACV,OAAO,EACN,OAAO,EACJ,QAAQ,CACR,SAAS,uDAAuD,EACpE,CAAC,CACD,SAAS,oCAAoC;GACjD,CAAC,CAI6B;EAE/B,MAAM,kBAAkB,aAAa,KAAK,KAAK;EAS/C,MAAM,eAAe,IAAI,eAAe;GACtC,UATqB,mBACrB,2CACE,aAAa,sBAAsB,EAAE,oBAAoB,GAAG,CAAC,CAC9D,EACD,EACE,cAAc,iBACf,CACF;GAGC,gBAAgB,CAAC,QAAQ;GACzB;GACD,CAAC;EAEF,MAAM,cAAc,eAAe,QAAQ,KAAK,aAAa;EAE7D,MAAM,oBAAoB,WAExB,gBAAgB,YAHF,oBAAoB,WAAW,UAAU,KAAK,CAGxB,CAAC,QACpC,KAAK,CAAC,MAAM,WAAW,YAAY;GAClC,MAAM,MAEF,wBAAwB,EAAE;AAC9B,OAAI,OACF,KAAI,SAAS;AAEf,OAAI,QAAQ,iBAAiB,QAAQ,KAAK,WAAW,IAAI;AACzD,UAAO;KAET,EAAE,CACH;EAED,IAAI;AACJ,MAAI,aACF,iBAAgB;WACP,iBACT,iBAAgB,iBAAiB,QAAQ,KAAK,kBAAkB;GAC9D,GAAG;GACH,QAAQ;GACT,CAAC;MAOF,iBAAgB,IAAI,kBAAkB;GACpC;GACA,QANa,IAAI,eAAe;IAChC,UAFqB,iBAAiB,QAAQ,SAAS,QAAQ;IAG/D,gBAAgB,CAAC,WAAW,QAAQ;IACrC,CAAC;GAIA,WAAW;GACZ,CAAC;AAGJ,SAAO,IAAI,sBAAsB;GAC/B,GAAG;GACH;GACA;GACA,cAAc;GACf,CAAC;;CAGJ,aAAqB;AACnB,SAAO"}
+{"version":3,"file":"multi_retrieval_qa.js","names":[],"sources":["../../../src/chains/router/multi_retrieval_qa.ts"],"sourcesContent":["import { z } from \"zod/v3\";\nimport type { BaseLanguageModelInterface } from \"@langchain/core/language_models/base\";\nimport type { BaseRetrieverInterface } from \"@langchain/core/retrievers\";\nimport { interpolateFString, PromptTemplate } from \"@langchain/core/prompts\";\nimport { MultiRouteChain, MultiRouteChainInput } from \"./multi_route.js\";\nimport { BaseChain } from \"../../chains/base.js\";\nimport { LLMRouterChain } from \"./llm_router.js\";\nimport {\n  ConversationChain,\n  DEFAULT_TEMPLATE,\n} from \"../../chains/conversation.js\";\nimport { STRUCTURED_MULTI_RETRIEVAL_ROUTER_TEMPLATE } from \"./multi_retrieval_prompt.js\";\nimport { zipEntries } from \"./utils.js\";\nimport {\n  RetrievalQAChain,\n  RetrievalQAChainInput,\n} from \"../../chains/retrieval_qa.js\";\nimport { RouterOutputParser } from \"../../output_parsers/router.js\";\n\n/**\n * A type that represents the default values for the MultiRetrievalQAChain\n * class. It includes optional properties for the default retriever,\n * default prompt, and default chain.\n */\nexport type MultiRetrievalDefaults = {\n  defaultRetriever?: BaseRetrieverInterface;\n  defaultPrompt?: PromptTemplate;\n  defaultChain?: BaseChain;\n};\n\n/**\n * A class that represents a multi-retrieval question answering chain in\n * the LangChain framework. It extends the MultiRouteChain class and\n * provides additional functionality specific to multi-retrieval QA\n * chains.\n * @example\n * ```typescript\n * const multiRetrievalQAChain = MultiRetrievalQAChain.fromLLMAndRetrievers(\n *   new ChatOpenAI({ model: \"gpt-4o-mini\" }),\n *   {\n *     retrieverNames: [\"aqua teen\", \"mst3k\", \"animaniacs\"],\n *     retrieverDescriptions: [\n *       \"Good for answering questions about Aqua Teen Hunger Force theme song\",\n *       \"Good for answering questions about Mystery Science Theater 3000 theme song\",\n *       \"Good for answering questions about Animaniacs theme song\",\n *     ],\n *     retrievers: [\n *       new MemoryVectorStore().asRetriever(3),\n *       new MemoryVectorStore().asRetriever(3),\n *       new MemoryVectorStore().asRetriever(3),\n *     ],\n *     retrievalQAChainOpts: {\n *       returnSourceDocuments: true,\n *     },\n *   },\n * );\n *\n * const result = await multiRetrievalQAChain.call({\n *   input:\n *     \"In the Aqua Teen Hunger Force theme song, who calls himself the mike rula?\",\n * });\n *\n * console.log(result.sourceDocuments, result.text);\n * ```\n */\nexport class MultiRetrievalQAChain extends MultiRouteChain {\n  get outputKeys(): string[] {\n    return [\"result\"];\n  }\n\n  /**\n   * @deprecated Use `fromRetrieversAndPrompts` instead\n   */\n  static fromRetrievers(\n    llm: BaseLanguageModelInterface,\n    retrieverNames: string[],\n    retrieverDescriptions: string[],\n    retrievers: BaseRetrieverInterface[],\n    retrieverPrompts?: PromptTemplate[],\n    defaults?: MultiRetrievalDefaults,\n    options?: Omit<MultiRouteChainInput, \"defaultChain\">\n  ) {\n    return MultiRetrievalQAChain.fromLLMAndRetrievers(llm, {\n      retrieverNames,\n      retrieverDescriptions,\n      retrievers,\n      retrieverPrompts,\n      defaults,\n      multiRetrievalChainOpts: options,\n    });\n  }\n\n  /**\n   * A static method that creates an instance of MultiRetrievalQAChain from\n   * a BaseLanguageModel and a set of retrievers. It takes in optional\n   * parameters for the retriever names, descriptions, prompts, defaults,\n   * and additional options. It is an alternative method to fromRetrievers\n   * and provides more flexibility in configuring the underlying chains.\n   * @param llm A BaseLanguageModel instance.\n   * @param retrieverNames An array of retriever names.\n   * @param retrieverDescriptions An array of retriever descriptions.\n   * @param retrievers An array of BaseRetrieverInterface instances.\n   * @param retrieverPrompts An optional array of PromptTemplate instances for the retrievers.\n   * @param defaults An optional MultiRetrievalDefaults instance.\n   * @param multiRetrievalChainOpts Additional optional parameters for the multi-retrieval chain.\n   * @param retrievalQAChainOpts Additional optional parameters for the retrieval QA chain.\n   * @returns A new instance of MultiRetrievalQAChain.\n   */\n  static fromLLMAndRetrievers(\n    llm: BaseLanguageModelInterface,\n    {\n      retrieverNames,\n      retrieverDescriptions,\n      retrievers,\n      retrieverPrompts,\n      defaults,\n      multiRetrievalChainOpts,\n      retrievalQAChainOpts,\n    }: {\n      retrieverNames: string[];\n      retrieverDescriptions: string[];\n      retrievers: BaseRetrieverInterface[];\n      retrieverPrompts?: PromptTemplate[];\n      defaults?: MultiRetrievalDefaults;\n      multiRetrievalChainOpts?: Omit<MultiRouteChainInput, \"defaultChain\">;\n      retrievalQAChainOpts?: Partial<\n        Omit<RetrievalQAChainInput, \"retriever\" | \"combineDocumentsChain\">\n      > & {\n        prompt?: PromptTemplate;\n      };\n    }\n  ): MultiRetrievalQAChain {\n    const { defaultRetriever, defaultPrompt, defaultChain } = defaults ?? {};\n    if (defaultPrompt && !defaultRetriever) {\n      throw new Error(\n        \"`default_retriever` must be specified if `default_prompt` is \\nprovided. Received only `default_prompt`.\"\n      );\n    }\n    const destinations = zipEntries<[string, string]>(\n      retrieverNames,\n      retrieverDescriptions\n    ).map(([name, desc]) => `${name}: ${desc}`);\n\n    const structuredOutputParserSchema = z.object({\n      destination: z\n        .string()\n        .optional()\n        .describe('name of the question answering system to use or \"DEFAULT\"'),\n      next_inputs: z\n        .object({\n          query: z\n            .string()\n            .describe(\"a potentially modified version of the original input\"),\n        })\n        .describe(\"input to be fed to the next model\"),\n    });\n\n    const outputParser = new RouterOutputParser<\n      typeof structuredOutputParserSchema\n    >(structuredOutputParserSchema);\n\n    const destinationsStr = destinations.join(\"\\n\");\n    const routerTemplate = interpolateFString(\n      STRUCTURED_MULTI_RETRIEVAL_ROUTER_TEMPLATE(\n        outputParser.getFormatInstructions({ interpolationDepth: 4 })\n      ),\n      {\n        destinations: destinationsStr,\n      }\n    );\n    const routerPrompt = new PromptTemplate({\n      template: routerTemplate,\n      inputVariables: [\"input\"],\n      outputParser,\n    });\n\n    const routerChain = LLMRouterChain.fromLLM(llm, routerPrompt);\n    const prompts = retrieverPrompts ?? retrievers.map(() => null);\n    const destinationChains = zipEntries<\n      [string, BaseRetrieverInterface, PromptTemplate | null]\n    >(retrieverNames, retrievers, prompts).reduce(\n      (acc, [name, retriever, prompt]) => {\n        const opt: Partial<RetrievalQAChainInput> & {\n          prompt?: PromptTemplate;\n        } = retrievalQAChainOpts ?? {};\n        if (prompt) {\n          opt.prompt = prompt;\n        }\n        acc[name] = RetrievalQAChain.fromLLM(llm, retriever, opt);\n        return acc;\n      },\n      {} as { [name: string]: RetrievalQAChain }\n    );\n\n    let _defaultChain;\n    if (defaultChain) {\n      _defaultChain = defaultChain;\n    } else if (defaultRetriever) {\n      _defaultChain = RetrievalQAChain.fromLLM(llm, defaultRetriever, {\n        ...retrievalQAChainOpts,\n        prompt: defaultPrompt,\n      });\n    } else {\n      const promptTemplate = DEFAULT_TEMPLATE.replace(\"input\", \"query\");\n      const prompt = new PromptTemplate({\n        template: promptTemplate,\n        inputVariables: [\"history\", \"query\"],\n      });\n      _defaultChain = new ConversationChain({\n        llm,\n        prompt,\n        outputKey: \"result\",\n      });\n    }\n\n    return new MultiRetrievalQAChain({\n      ...multiRetrievalChainOpts,\n      routerChain,\n      destinationChains,\n      defaultChain: _defaultChain,\n    });\n  }\n\n  _chainType(): string {\n    return \"multi_retrieval_qa_chain\";\n  }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAiEA,IAAa,wBAAb,MAAa,8BAA8B,gBAAgB;CACzD,IAAI,aAAuB;AACzB,SAAO,CAAC,SAAS;;;;;CAMnB,OAAO,eACL,KACA,gBACA,uBACA,YACA,kBACA,UACA,SACA;AACA,SAAO,sBAAsB,qBAAqB,KAAK;GACrD;GACA;GACA;GACA;GACA;GACA,yBAAyB;GAC1B,CAAC;;;;;;;;;;;;;;;;;;CAmBJ,OAAO,qBACL,KACA,EACE,gBACA,uBACA,YACA,kBACA,UACA,yBACA,wBAcqB;EACvB,MAAM,EAAE,kBAAkB,eAAe,iBAAiB,YAAY,EAAE;AACxE,MAAI,iBAAiB,CAAC,iBACpB,OAAM,IAAI,MACR,2GACD;EAEH,MAAM,eAAe,WACnB,gBACA,sBACD,CAAC,KAAK,CAAC,MAAM,UAAU,GAAG,KAAK,IAAI,OAAO;EAgB3C,MAAM,eAAe,IAAI,mBAdY,EAAE,OAAO;GAC5C,aAAa,EACV,QAAQ,CACR,UAAU,CACV,SAAS,8DAA4D;GACxE,aAAa,EACV,OAAO,EACN,OAAO,EACJ,QAAQ,CACR,SAAS,uDAAuD,EACpE,CAAC,CACD,SAAS,oCAAoC;GACjD,CAI6B,CAAC;EAE/B,MAAM,kBAAkB,aAAa,KAAK,KAAK;EAS/C,MAAM,eAAe,IAAI,eAAe;GACtC,UATqB,mBACrB,2CACE,aAAa,sBAAsB,EAAE,oBAAoB,GAAG,CAAC,CAC9D,EACD,EACE,cAAc,iBACf,CAGuB;GACxB,gBAAgB,CAAC,QAAQ;GACzB;GACD,CAAC;EAEF,MAAM,cAAc,eAAe,QAAQ,KAAK,aAAa;EAE7D,MAAM,oBAAoB,WAExB,gBAAgB,YAHF,oBAAoB,WAAW,UAAU,KAAK,CAGxB,CAAC,QACpC,KAAK,CAAC,MAAM,WAAW,YAAY;GAClC,MAAM,MAEF,wBAAwB,EAAE;AAC9B,OAAI,OACF,KAAI,SAAS;AAEf,OAAI,QAAQ,iBAAiB,QAAQ,KAAK,WAAW,IAAI;AACzD,UAAO;KAET,EAAE,CACH;EAED,IAAI;AACJ,MAAI,aACF,iBAAgB;WACP,iBACT,iBAAgB,iBAAiB,QAAQ,KAAK,kBAAkB;GAC9D,GAAG;GACH,QAAQ;GACT,CAAC;MAOF,iBAAgB,IAAI,kBAAkB;GACpC;GACA,QAAA,IANiB,eAAe;IAChC,UAFqB,iBAAiB,QAAQ,SAAS,QAE/B;IACxB,gBAAgB,CAAC,WAAW,QAAQ;IACrC,CAGO;GACN,WAAW;GACZ,CAAC;AAGJ,SAAO,IAAI,sBAAsB;GAC/B,GAAG;GACH;GACA;GACA,cAAc;GACf,CAAC;;CAGJ,aAAqB;AACnB,SAAO"}

package/dist/chains/sql_db/sql_db_chain.cjs.map

@@ -1 +1 @@
-{"version":3,"file":"sql_db_chain.cjs","names":["BaseChain","DEFAULT_SQL_DATABASE_PROMPT","getPromptTemplateFromDataSource","LLMChain","SQL_PROMPTS_MAP","RunnableSequence","RunnablePassthrough","StringOutputParser"],"sources":["../../../src/chains/sql_db/sql_db_chain.ts"],"sourcesContent":["import type {\n  BaseLanguageModel,\n  BaseLanguageModelInterface,\n} from \"@langchain/core/language_models/base\";\nimport type { OpenAI, TiktokenModel } from \"@langchain/openai\";\nimport { ChainValues } from \"@langchain/core/utils/types\";\nimport { BasePromptTemplate, PromptTemplate } from \"@langchain/core/prompts\";\nimport {\n  calculateMaxTokens,\n  getModelContextSize,\n} from \"@langchain/core/language_models/base\";\nimport { CallbackManagerForChainRun } from \"@langchain/core/callbacks/manager\";\nimport {\n  RunnablePassthrough,\n  RunnableSequence,\n} from \"@langchain/core/runnables\";\nimport { StringOutputParser } from \"@langchain/core/output_parsers\";\nimport {\n  DEFAULT_SQL_DATABASE_PROMPT,\n  SQL_PROMPTS_MAP,\n  SqlDialect,\n} from \"./sql_db_prompt.js\";\nimport { BaseChain, ChainInputs } from \"../base.js\";\nimport { LLMChain } from \"../llm_chain.js\";\nimport type { SqlDatabase } from \"../../sql_db.js\";\nimport { getPromptTemplateFromDataSource } from \"../../util/sql_utils.js\";\n\n/**\n * Interface that extends the ChainInputs interface and defines additional\n * fields specific to a SQL database chain. It represents the input fields\n * for a SQL database chain.\n */\nexport interface SqlDatabaseChainInput extends ChainInputs {\n  llm: BaseLanguageModelInterface;\n  database: SqlDatabase;\n  topK?: number;\n  inputKey?: string;\n  outputKey?: string;\n  sqlOutputKey?: string;\n  prompt?: PromptTemplate;\n}\n\n/**\n * Class that represents a SQL database chain in the LangChain framework.\n * It extends the BaseChain class and implements the functionality\n * specific to a SQL database chain.\n *\n * @security **Security Notice**\n * This chain generates SQL queries for the given database.\n * The SQLDatabase class provides a getTableInfo method that can be used\n * to get column information as well as sample data from the table.\n * To mitigate risk of leaking sensitive data, limit permissions\n * to read and scope to the tables that are needed.\n * Optionally, use the includesTables or ignoreTables class parameters\n * to limit which tables can/cannot be accessed.\n *\n * @link See https://js.langchain.com/docs/security for more information.\n * @example\n * ```typescript\n * const chain = new SqlDatabaseChain({\n *   llm: new OpenAI({ temperature: 0 }),\n *   database: new SqlDatabase({ ...config }),\n * });\n *\n * const result = await chain.run(\"How many tracks are there?\");\n * ```\n */\nexport class SqlDatabaseChain extends BaseChain {\n  static lc_name() {\n    return \"SqlDatabaseChain\";\n  }\n\n  // LLM wrapper to use\n  llm: BaseLanguageModelInterface;\n\n  // SQL Database to connect to.\n  database: SqlDatabase;\n\n  // Prompt to use to translate natural language to SQL.\n  prompt = DEFAULT_SQL_DATABASE_PROMPT;\n\n  // Number of results to return from the query\n  topK = 5;\n\n  inputKey = \"query\";\n\n  outputKey = \"result\";\n\n  sqlOutputKey: string | undefined = undefined;\n\n  // Whether to return the result of querying the SQL table directly.\n  returnDirect = false;\n\n  constructor(fields: SqlDatabaseChainInput) {\n    super(fields);\n    this.llm = fields.llm;\n    this.database = fields.database;\n    this.topK = fields.topK ?? this.topK;\n    this.inputKey = fields.inputKey ?? this.inputKey;\n    this.outputKey = fields.outputKey ?? this.outputKey;\n    this.sqlOutputKey = fields.sqlOutputKey ?? this.sqlOutputKey;\n    this.prompt =\n      fields.prompt ??\n      getPromptTemplateFromDataSource(this.database.appDataSource);\n  }\n\n  /** @ignore */\n  async _call(\n    values: ChainValues,\n    runManager?: CallbackManagerForChainRun\n  ): Promise<ChainValues> {\n    const llmChain = new LLMChain({\n      prompt: this.prompt,\n      llm: this.llm,\n      outputKey: this.outputKey,\n      memory: this.memory,\n    });\n    if (!(this.inputKey in values)) {\n      throw new Error(`Question key ${this.inputKey} not found.`);\n    }\n    const question: string = values[this.inputKey];\n    let inputText = `${question}\\nSQLQuery:`;\n    const tablesToUse = values.table_names_to_use;\n    const tableInfo = await this.database.getTableInfo(tablesToUse);\n\n    const llmInputs = {\n      input: inputText,\n      top_k: this.topK,\n      dialect: this.database.appDataSourceOptions.type,\n      table_info: tableInfo,\n      stop: [\"\\nSQLResult:\"],\n    };\n    await this.verifyNumberOfTokens(inputText, tableInfo);\n\n    const sqlCommand = await llmChain.predict(\n      llmInputs,\n      runManager?.getChild(\"sql_generation\")\n    );\n    let queryResult = \"\";\n    try {\n      queryResult = await this.database.appDataSource.query(sqlCommand);\n    } catch (error) {\n      console.error(error);\n    }\n\n    let finalResult;\n    if (this.returnDirect) {\n      finalResult = { [this.outputKey]: queryResult };\n    } else {\n      inputText += `${sqlCommand}\\nSQLResult: ${JSON.stringify(\n        queryResult\n      )}\\nAnswer:`;\n      llmInputs.input = inputText;\n      finalResult = {\n        [this.outputKey]: await llmChain.predict(\n          llmInputs,\n          runManager?.getChild(\"result_generation\")\n        ),\n      };\n    }\n\n    if (this.sqlOutputKey != null) {\n      finalResult[this.sqlOutputKey] = sqlCommand;\n    }\n\n    return finalResult;\n  }\n\n  _chainType() {\n    return \"sql_database_chain\" as const;\n  }\n\n  get inputKeys(): string[] {\n    return [this.inputKey];\n  }\n\n  get outputKeys(): string[] {\n    if (this.sqlOutputKey != null) {\n      return [this.outputKey, this.sqlOutputKey];\n    }\n    return [this.outputKey];\n  }\n\n  /**\n   * Private method that verifies the number of tokens in the input text and\n   * table information. It throws an error if the number of tokens exceeds\n   * the maximum allowed by the language model.\n   * @param inputText The input text.\n   * @param tableinfo The table information.\n   * @returns A promise that resolves when the verification is complete.\n   */\n  private async verifyNumberOfTokens(\n    inputText: string,\n    tableinfo: string\n  ): Promise<void> {\n    // We verify it only for OpenAI for the moment\n    if (this.llm._llmType() !== \"openai\") {\n      return;\n    }\n    const llm = this.llm as OpenAI;\n    const promptTemplate = this.prompt.template;\n    const stringWeSend = `${inputText}${promptTemplate}${tableinfo}`;\n\n    const maxToken = await calculateMaxTokens({\n      prompt: stringWeSend,\n      // Cast here to allow for other models that may not fit the union\n      modelName: llm.model as TiktokenModel,\n    });\n\n    if (maxToken < (llm.maxTokens ?? -1)) {\n      throw new Error(`The combination of the database structure and your question is too big for the model ${\n        llm.model\n      } which can compute only a max tokens of ${getModelContextSize(\n        llm.model\n      )}.\n      We suggest you to use the includeTables parameters when creating the SqlDatabase object to select only a subset of the tables. You can also use a model which can handle more tokens.`);\n    }\n  }\n}\n\nexport interface CreateSqlQueryChainFields {\n  llm: BaseLanguageModel;\n  db: SqlDatabase;\n  prompt?: BasePromptTemplate;\n  /**\n   * @default 5\n   */\n  k?: number;\n  dialect: SqlDialect;\n}\n\ntype SqlInput = {\n  question: string;\n};\n\ntype SqlInoutWithTables = SqlInput & {\n  tableNamesToUse: string[];\n};\n\nconst strip = (text: string) => {\n  // Replace escaped quotes with actual quotes\n  let newText = text.replace(/\\\\\"/g, '\"').trim();\n  // Remove wrapping quotes if the entire string is wrapped in quotes\n  if (newText.startsWith('\"') && newText.endsWith('\"')) {\n    newText = newText.substring(1, newText.length - 1);\n  }\n  return newText;\n};\n\nconst difference = (setA: Set<string>, setB: Set<string>) =>\n  new Set([...setA].filter((x) => !setB.has(x)));\n\n/**\n * Create a SQL query chain that can create SQL queries for the given database.\n * Returns a Runnable.\n *\n * @param {BaseLanguageModel} llm The language model to use in the chain.\n * @param {SqlDatabase} db The database to use in the chain.\n * @param {BasePromptTemplate | undefined} prompt The prompt to use in the chain.\n * @param {BaseLanguageModel | undefined} k The amount of docs/results to return. Passed through the prompt input value `top_k`.\n * @param {SqlDialect} dialect The SQL dialect to use in the chain.\n * @returns {Promise<RunnableSequence<Record<string, unknown>, string>>} A runnable sequence representing the chain.\n * @example ```typescript\n * const datasource = new DataSource({\n *   type: \"sqlite\",\n *   database: \"../../../../Chinook.db\",\n * });\n * const db = await SqlDatabase.fromDataSourceParams({\n *   appDataSource: datasource,\n * });\n * const llm = new ChatOpenAI({ model: \"gpt-4o-mini\", temperature: 0 });\n * const chain = await createSqlQueryChain({\n *   llm,\n *   db,\n *   dialect: \"sqlite\",\n * });\n * ```\n */\nexport async function createSqlQueryChain({\n  llm,\n  db,\n  prompt,\n  k = 5,\n  dialect,\n}: CreateSqlQueryChainFields) {\n  let promptToUse: BasePromptTemplate;\n  if (prompt) {\n    promptToUse = prompt;\n  } else if (SQL_PROMPTS_MAP[dialect]) {\n    promptToUse = SQL_PROMPTS_MAP[dialect];\n  } else {\n    promptToUse = DEFAULT_SQL_DATABASE_PROMPT;\n  }\n\n  if (\n    difference(\n      new Set([\"input\", \"top_k\", \"table_info\"]),\n      new Set(promptToUse.inputVariables)\n    ).size > 0\n  ) {\n    throw new Error(\n      `Prompt must have input variables: 'input', 'top_k', 'table_info'. Received prompt with input variables: ` +\n        `${promptToUse.inputVariables}. Full prompt:\\n\\n${promptToUse}`\n    );\n  }\n  if (promptToUse.inputVariables.includes(\"dialect\")) {\n    promptToUse = await promptToUse.partial({ dialect });\n  }\n\n  promptToUse = await promptToUse.partial({ top_k: k.toString() });\n\n  const inputs = {\n    input: (x: Record<string, unknown>) => {\n      if (\"question\" in x) {\n        return `${(x as SqlInput).question}\\nSQLQuery: `;\n      }\n      throw new Error(\"Input must include a question property.\");\n    },\n    table_info: async (x: Record<string, unknown>) =>\n      db.getTableInfo((x as SqlInoutWithTables).tableNamesToUse),\n  };\n\n  return RunnableSequence.from([\n    RunnablePassthrough.assign(inputs),\n    (x) => {\n      const newInputs = { ...x };\n      delete newInputs.question;\n      delete newInputs.tableNamesToUse;\n      return newInputs;\n    },\n    promptToUse,\n    llm.withConfig({ stop: [\"\\nSQLResult:\"] }),\n    new StringOutputParser(),\n    strip,\n  ]);\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAmEA,IAAa,mBAAb,cAAsCA,aAAAA,UAAU;CAC9C,OAAO,UAAU;AACf,SAAO;;CAIT;CAGA;CAGA,SAASC,sBAAAA;CAGT,OAAO;CAEP,WAAW;CAEX,YAAY;CAEZ,eAAmC,KAAA;CAGnC,eAAe;CAEf,YAAY,QAA+B;AACzC,QAAM,OAAO;AACb,OAAK,MAAM,OAAO;AAClB,OAAK,WAAW,OAAO;AACvB,OAAK,OAAO,OAAO,QAAQ,KAAK;AAChC,OAAK,WAAW,OAAO,YAAY,KAAK;AACxC,OAAK,YAAY,OAAO,aAAa,KAAK;AAC1C,OAAK,eAAe,OAAO,gBAAgB,KAAK;AAChD,OAAK,SACH,OAAO,UACPC,kBAAAA,gCAAgC,KAAK,SAAS,cAAc;;;CAIhE,MAAM,MACJ,QACA,YACsB;EACtB,MAAM,WAAW,IAAIC,kBAAAA,SAAS;GAC5B,QAAQ,KAAK;GACb,KAAK,KAAK;GACV,WAAW,KAAK;GAChB,QAAQ,KAAK;GACd,CAAC;AACF,MAAI,EAAE,KAAK,YAAY,QACrB,OAAM,IAAI,MAAM,gBAAgB,KAAK,SAAS,aAAa;EAG7D,IAAI,YAAY,GADS,OAAO,KAAK,UACT;EAC5B,MAAM,cAAc,OAAO;EAC3B,MAAM,YAAY,MAAM,KAAK,SAAS,aAAa,YAAY;EAE/D,MAAM,YAAY;GAChB,OAAO;GACP,OAAO,KAAK;GACZ,SAAS,KAAK,SAAS,qBAAqB;GAC5C,YAAY;GACZ,MAAM,CAAC,eAAe;GACvB;AACD,QAAM,KAAK,qBAAqB,WAAW,UAAU;EAErD,MAAM,aAAa,MAAM,SAAS,QAChC,WACA,YAAY,SAAS,iBAAiB,CACvC;EACD,IAAI,cAAc;AAClB,MAAI;AACF,iBAAc,MAAM,KAAK,SAAS,cAAc,MAAM,WAAW;WAC1D,OAAO;AACd,WAAQ,MAAM,MAAM;;EAGtB,IAAI;AACJ,MAAI,KAAK,aACP,eAAc,GAAG,KAAK,YAAY,aAAa;OAC1C;AACL,gBAAa,GAAG,WAAW,eAAe,KAAK,UAC7C,YACD,CAAC;AACF,aAAU,QAAQ;AAClB,iBAAc,GACX,KAAK,YAAY,MAAM,SAAS,QAC/B,WACA,YAAY,SAAS,oBAAoB,CAC1C,EACF;;AAGH,MAAI,KAAK,gBAAgB,KACvB,aAAY,KAAK,gBAAgB;AAGnC,SAAO;;CAGT,aAAa;AACX,SAAO;;CAGT,IAAI,YAAsB;AACxB,SAAO,CAAC,KAAK,SAAS;;CAGxB,IAAI,aAAuB;AACzB,MAAI,KAAK,gBAAgB,KACvB,QAAO,CAAC,KAAK,WAAW,KAAK,aAAa;AAE5C,SAAO,CAAC,KAAK,UAAU;;;;;;;;;;CAWzB,MAAc,qBACZ,WACA,WACe;AAEf,MAAI,KAAK,IAAI,UAAU,KAAK,SAC1B;EAEF,MAAM,MAAM,KAAK;AAUjB,MANiB,OAAA,GAAA,qCAAA,oBAAyB;GACxC,QAHmB,GAAG,YADD,KAAK,OAAO,WACkB;GAKnD,WAAW,IAAI;GAChB,CAAC,IAEc,IAAI,aAAa,IAC/B,OAAM,IAAI,MAAM,wFACd,IAAI,MACL,2CAAA,GAAA,qCAAA,qBACC,IAAI,MACL,CAAC;6LACqL;;;AAwB7L,MAAM,SAAS,SAAiB;CAE9B,IAAI,UAAU,KAAK,QAAQ,QAAQ,KAAI,CAAC,MAAM;AAE9C,KAAI,QAAQ,WAAW,KAAI,IAAI,QAAQ,SAAS,KAAI,CAClD,WAAU,QAAQ,UAAU,GAAG,QAAQ,SAAS,EAAE;AAEpD,QAAO;;AAGT,MAAM,cAAc,MAAmB,SACrC,IAAI,IAAI,CAAC,GAAG,KAAK,CAAC,QAAQ,MAAM,CAAC,KAAK,IAAI,EAAE,CAAC,CAAC;;;;;;;;;;;;;;;;;;;;;;;;;;;AA4BhD,eAAsB,oBAAoB,EACxC,KACA,IACA,QACA,IAAI,GACJ,WAC4B;CAC5B,IAAI;AACJ,KAAI,OACF,eAAc;UACLC,sBAAAA,gBAAgB,SACzB,eAAcA,sBAAAA,gBAAgB;KAE9B,eAAcH,sBAAAA;AAGhB,KACE,WACE,IAAI,IAAI;EAAC;EAAS;EAAS;EAAa,CAAC,EACzC,IAAI,IAAI,YAAY,eAAe,CACpC,CAAC,OAAO,EAET,OAAM,IAAI,MACR,2GACK,YAAY,eAAe,oBAAoB,cACrD;AAEH,KAAI,YAAY,eAAe,SAAS,UAAU,CAChD,eAAc,MAAM,YAAY,QAAQ,EAAE,SAAS,CAAC;AAGtD,eAAc,MAAM,YAAY,QAAQ,EAAE,OAAO,EAAE,UAAU,EAAE,CAAC;AAahE,QAAOI,0BAAAA,iBAAiB,KAAK;EAC3BC,0BAAAA,oBAAoB,OAZP;GACb,QAAQ,MAA+B;AACrC,QAAI,cAAc,EAChB,QAAO,GAAI,EAAe,SAAS;AAErC,UAAM,IAAI,MAAM,0CAA0C;;GAE5D,YAAY,OAAO,MACjB,GAAG,aAAc,EAAyB,gBAAgB;GAC7D,CAGmC;GACjC,MAAM;GACL,MAAM,YAAY,EAAE,GAAG,GAAG;AAC1B,UAAO,UAAU;AACjB,UAAO,UAAU;AACjB,UAAO;;EAET;EACA,IAAI,WAAW,EAAE,MAAM,CAAC,eAAe,EAAE,CAAC;EAC1C,IAAIC,+BAAAA,oBAAoB;EACxB;EACD,CAAC"}
+{"version":3,"file":"sql_db_chain.cjs","names":["BaseChain","DEFAULT_SQL_DATABASE_PROMPT","getPromptTemplateFromDataSource","LLMChain","SQL_PROMPTS_MAP","RunnableSequence","RunnablePassthrough","StringOutputParser"],"sources":["../../../src/chains/sql_db/sql_db_chain.ts"],"sourcesContent":["import type {\n  BaseLanguageModel,\n  BaseLanguageModelInterface,\n} from \"@langchain/core/language_models/base\";\nimport type { OpenAI, TiktokenModel } from \"@langchain/openai\";\nimport { ChainValues } from \"@langchain/core/utils/types\";\nimport { BasePromptTemplate, PromptTemplate } from \"@langchain/core/prompts\";\nimport {\n  calculateMaxTokens,\n  getModelContextSize,\n} from \"@langchain/core/language_models/base\";\nimport { CallbackManagerForChainRun } from \"@langchain/core/callbacks/manager\";\nimport {\n  RunnablePassthrough,\n  RunnableSequence,\n} from \"@langchain/core/runnables\";\nimport { StringOutputParser } from \"@langchain/core/output_parsers\";\nimport {\n  DEFAULT_SQL_DATABASE_PROMPT,\n  SQL_PROMPTS_MAP,\n  SqlDialect,\n} from \"./sql_db_prompt.js\";\nimport { BaseChain, ChainInputs } from \"../base.js\";\nimport { LLMChain } from \"../llm_chain.js\";\nimport type { SqlDatabase } from \"../../sql_db.js\";\nimport { getPromptTemplateFromDataSource } from \"../../util/sql_utils.js\";\n\n/**\n * Interface that extends the ChainInputs interface and defines additional\n * fields specific to a SQL database chain. It represents the input fields\n * for a SQL database chain.\n */\nexport interface SqlDatabaseChainInput extends ChainInputs {\n  llm: BaseLanguageModelInterface;\n  database: SqlDatabase;\n  topK?: number;\n  inputKey?: string;\n  outputKey?: string;\n  sqlOutputKey?: string;\n  prompt?: PromptTemplate;\n}\n\n/**\n * Class that represents a SQL database chain in the LangChain framework.\n * It extends the BaseChain class and implements the functionality\n * specific to a SQL database chain.\n *\n * @security **Security Notice**\n * This chain generates SQL queries for the given database.\n * The SQLDatabase class provides a getTableInfo method that can be used\n * to get column information as well as sample data from the table.\n * To mitigate risk of leaking sensitive data, limit permissions\n * to read and scope to the tables that are needed.\n * Optionally, use the includesTables or ignoreTables class parameters\n * to limit which tables can/cannot be accessed.\n *\n * @link See https://js.langchain.com/docs/security for more information.\n * @example\n * ```typescript\n * const chain = new SqlDatabaseChain({\n *   llm: new OpenAI({ temperature: 0 }),\n *   database: new SqlDatabase({ ...config }),\n * });\n *\n * const result = await chain.run(\"How many tracks are there?\");\n * ```\n */\nexport class SqlDatabaseChain extends BaseChain {\n  static lc_name() {\n    return \"SqlDatabaseChain\";\n  }\n\n  // LLM wrapper to use\n  llm: BaseLanguageModelInterface;\n\n  // SQL Database to connect to.\n  database: SqlDatabase;\n\n  // Prompt to use to translate natural language to SQL.\n  prompt = DEFAULT_SQL_DATABASE_PROMPT;\n\n  // Number of results to return from the query\n  topK = 5;\n\n  inputKey = \"query\";\n\n  outputKey = \"result\";\n\n  sqlOutputKey: string | undefined = undefined;\n\n  // Whether to return the result of querying the SQL table directly.\n  returnDirect = false;\n\n  constructor(fields: SqlDatabaseChainInput) {\n    super(fields);\n    this.llm = fields.llm;\n    this.database = fields.database;\n    this.topK = fields.topK ?? this.topK;\n    this.inputKey = fields.inputKey ?? this.inputKey;\n    this.outputKey = fields.outputKey ?? this.outputKey;\n    this.sqlOutputKey = fields.sqlOutputKey ?? this.sqlOutputKey;\n    this.prompt =\n      fields.prompt ??\n      getPromptTemplateFromDataSource(this.database.appDataSource);\n  }\n\n  /** @ignore */\n  async _call(\n    values: ChainValues,\n    runManager?: CallbackManagerForChainRun\n  ): Promise<ChainValues> {\n    const llmChain = new LLMChain({\n      prompt: this.prompt,\n      llm: this.llm,\n      outputKey: this.outputKey,\n      memory: this.memory,\n    });\n    if (!(this.inputKey in values)) {\n      throw new Error(`Question key ${this.inputKey} not found.`);\n    }\n    const question: string = values[this.inputKey];\n    let inputText = `${question}\\nSQLQuery:`;\n    const tablesToUse = values.table_names_to_use;\n    const tableInfo = await this.database.getTableInfo(tablesToUse);\n\n    const llmInputs = {\n      input: inputText,\n      top_k: this.topK,\n      dialect: this.database.appDataSourceOptions.type,\n      table_info: tableInfo,\n      stop: [\"\\nSQLResult:\"],\n    };\n    await this.verifyNumberOfTokens(inputText, tableInfo);\n\n    const sqlCommand = await llmChain.predict(\n      llmInputs,\n      runManager?.getChild(\"sql_generation\")\n    );\n    let queryResult = \"\";\n    try {\n      queryResult = await this.database.appDataSource.query(sqlCommand);\n    } catch (error) {\n      console.error(error);\n    }\n\n    let finalResult;\n    if (this.returnDirect) {\n      finalResult = { [this.outputKey]: queryResult };\n    } else {\n      inputText += `${sqlCommand}\\nSQLResult: ${JSON.stringify(\n        queryResult\n      )}\\nAnswer:`;\n      llmInputs.input = inputText;\n      finalResult = {\n        [this.outputKey]: await llmChain.predict(\n          llmInputs,\n          runManager?.getChild(\"result_generation\")\n        ),\n      };\n    }\n\n    if (this.sqlOutputKey != null) {\n      finalResult[this.sqlOutputKey] = sqlCommand;\n    }\n\n    return finalResult;\n  }\n\n  _chainType() {\n    return \"sql_database_chain\" as const;\n  }\n\n  get inputKeys(): string[] {\n    return [this.inputKey];\n  }\n\n  get outputKeys(): string[] {\n    if (this.sqlOutputKey != null) {\n      return [this.outputKey, this.sqlOutputKey];\n    }\n    return [this.outputKey];\n  }\n\n  /**\n   * Private method that verifies the number of tokens in the input text and\n   * table information. It throws an error if the number of tokens exceeds\n   * the maximum allowed by the language model.\n   * @param inputText The input text.\n   * @param tableinfo The table information.\n   * @returns A promise that resolves when the verification is complete.\n   */\n  private async verifyNumberOfTokens(\n    inputText: string,\n    tableinfo: string\n  ): Promise<void> {\n    // We verify it only for OpenAI for the moment\n    if (this.llm._llmType() !== \"openai\") {\n      return;\n    }\n    const llm = this.llm as OpenAI;\n    const promptTemplate = this.prompt.template;\n    const stringWeSend = `${inputText}${promptTemplate}${tableinfo}`;\n\n    const maxToken = await calculateMaxTokens({\n      prompt: stringWeSend,\n      // Cast here to allow for other models that may not fit the union\n      modelName: llm.model as TiktokenModel,\n    });\n\n    if (maxToken < (llm.maxTokens ?? -1)) {\n      throw new Error(`The combination of the database structure and your question is too big for the model ${\n        llm.model\n      } which can compute only a max tokens of ${getModelContextSize(\n        llm.model\n      )}.\n      We suggest you to use the includeTables parameters when creating the SqlDatabase object to select only a subset of the tables. You can also use a model which can handle more tokens.`);\n    }\n  }\n}\n\nexport interface CreateSqlQueryChainFields {\n  llm: BaseLanguageModel;\n  db: SqlDatabase;\n  prompt?: BasePromptTemplate;\n  /**\n   * @default 5\n   */\n  k?: number;\n  dialect: SqlDialect;\n}\n\ntype SqlInput = {\n  question: string;\n};\n\ntype SqlInoutWithTables = SqlInput & {\n  tableNamesToUse: string[];\n};\n\nconst strip = (text: string) => {\n  // Replace escaped quotes with actual quotes\n  let newText = text.replace(/\\\\\"/g, '\"').trim();\n  // Remove wrapping quotes if the entire string is wrapped in quotes\n  if (newText.startsWith('\"') && newText.endsWith('\"')) {\n    newText = newText.substring(1, newText.length - 1);\n  }\n  return newText;\n};\n\nconst difference = (setA: Set<string>, setB: Set<string>) =>\n  new Set([...setA].filter((x) => !setB.has(x)));\n\n/**\n * Create a SQL query chain that can create SQL queries for the given database.\n * Returns a Runnable.\n *\n * @param {BaseLanguageModel} llm The language model to use in the chain.\n * @param {SqlDatabase} db The database to use in the chain.\n * @param {BasePromptTemplate | undefined} prompt The prompt to use in the chain.\n * @param {BaseLanguageModel | undefined} k The amount of docs/results to return. Passed through the prompt input value `top_k`.\n * @param {SqlDialect} dialect The SQL dialect to use in the chain.\n * @returns {Promise<RunnableSequence<Record<string, unknown>, string>>} A runnable sequence representing the chain.\n * @example ```typescript\n * const datasource = new DataSource({\n *   type: \"sqlite\",\n *   database: \"../../../../Chinook.db\",\n * });\n * const db = await SqlDatabase.fromDataSourceParams({\n *   appDataSource: datasource,\n * });\n * const llm = new ChatOpenAI({ model: \"gpt-4o-mini\", temperature: 0 });\n * const chain = await createSqlQueryChain({\n *   llm,\n *   db,\n *   dialect: \"sqlite\",\n * });\n * ```\n */\nexport async function createSqlQueryChain({\n  llm,\n  db,\n  prompt,\n  k = 5,\n  dialect,\n}: CreateSqlQueryChainFields) {\n  let promptToUse: BasePromptTemplate;\n  if (prompt) {\n    promptToUse = prompt;\n  } else if (SQL_PROMPTS_MAP[dialect]) {\n    promptToUse = SQL_PROMPTS_MAP[dialect];\n  } else {\n    promptToUse = DEFAULT_SQL_DATABASE_PROMPT;\n  }\n\n  if (\n    difference(\n      new Set([\"input\", \"top_k\", \"table_info\"]),\n      new Set(promptToUse.inputVariables)\n    ).size > 0\n  ) {\n    throw new Error(\n      `Prompt must have input variables: 'input', 'top_k', 'table_info'. Received prompt with input variables: ` +\n        `${promptToUse.inputVariables}. Full prompt:\\n\\n${promptToUse}`\n    );\n  }\n  if (promptToUse.inputVariables.includes(\"dialect\")) {\n    promptToUse = await promptToUse.partial({ dialect });\n  }\n\n  promptToUse = await promptToUse.partial({ top_k: k.toString() });\n\n  const inputs = {\n    input: (x: Record<string, unknown>) => {\n      if (\"question\" in x) {\n        return `${(x as SqlInput).question}\\nSQLQuery: `;\n      }\n      throw new Error(\"Input must include a question property.\");\n    },\n    table_info: async (x: Record<string, unknown>) =>\n      db.getTableInfo((x as SqlInoutWithTables).tableNamesToUse),\n  };\n\n  return RunnableSequence.from([\n    RunnablePassthrough.assign(inputs),\n    (x) => {\n      const newInputs = { ...x };\n      delete newInputs.question;\n      delete newInputs.tableNamesToUse;\n      return newInputs;\n    },\n    promptToUse,\n    llm.withConfig({ stop: [\"\\nSQLResult:\"] }),\n    new StringOutputParser(),\n    strip,\n  ]);\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAmEA,IAAa,mBAAb,cAAsCA,aAAAA,UAAU;CAC9C,OAAO,UAAU;AACf,SAAO;;CAIT;CAGA;CAGA,SAASC,sBAAAA;CAGT,OAAO;CAEP,WAAW;CAEX,YAAY;CAEZ,eAAmC,KAAA;CAGnC,eAAe;CAEf,YAAY,QAA+B;AACzC,QAAM,OAAO;AACb,OAAK,MAAM,OAAO;AAClB,OAAK,WAAW,OAAO;AACvB,OAAK,OAAO,OAAO,QAAQ,KAAK;AAChC,OAAK,WAAW,OAAO,YAAY,KAAK;AACxC,OAAK,YAAY,OAAO,aAAa,KAAK;AAC1C,OAAK,eAAe,OAAO,gBAAgB,KAAK;AAChD,OAAK,SACH,OAAO,UACPC,kBAAAA,gCAAgC,KAAK,SAAS,cAAc;;;CAIhE,MAAM,MACJ,QACA,YACsB;EACtB,MAAM,WAAW,IAAIC,kBAAAA,SAAS;GAC5B,QAAQ,KAAK;GACb,KAAK,KAAK;GACV,WAAW,KAAK;GAChB,QAAQ,KAAK;GACd,CAAC;AACF,MAAI,EAAE,KAAK,YAAY,QACrB,OAAM,IAAI,MAAM,gBAAgB,KAAK,SAAS,aAAa;EAG7D,IAAI,YAAY,GADS,OAAO,KAAK,UACT;EAC5B,MAAM,cAAc,OAAO;EAC3B,MAAM,YAAY,MAAM,KAAK,SAAS,aAAa,YAAY;EAE/D,MAAM,YAAY;GAChB,OAAO;GACP,OAAO,KAAK;GACZ,SAAS,KAAK,SAAS,qBAAqB;GAC5C,YAAY;GACZ,MAAM,CAAC,eAAe;GACvB;AACD,QAAM,KAAK,qBAAqB,WAAW,UAAU;EAErD,MAAM,aAAa,MAAM,SAAS,QAChC,WACA,YAAY,SAAS,iBAAiB,CACvC;EACD,IAAI,cAAc;AAClB,MAAI;AACF,iBAAc,MAAM,KAAK,SAAS,cAAc,MAAM,WAAW;WAC1D,OAAO;AACd,WAAQ,MAAM,MAAM;;EAGtB,IAAI;AACJ,MAAI,KAAK,aACP,eAAc,GAAG,KAAK,YAAY,aAAa;OAC1C;AACL,gBAAa,GAAG,WAAW,eAAe,KAAK,UAC7C,YACD,CAAC;AACF,aAAU,QAAQ;AAClB,iBAAc,GACX,KAAK,YAAY,MAAM,SAAS,QAC/B,WACA,YAAY,SAAS,oBAAoB,CAC1C,EACF;;AAGH,MAAI,KAAK,gBAAgB,KACvB,aAAY,KAAK,gBAAgB;AAGnC,SAAO;;CAGT,aAAa;AACX,SAAO;;CAGT,IAAI,YAAsB;AACxB,SAAO,CAAC,KAAK,SAAS;;CAGxB,IAAI,aAAuB;AACzB,MAAI,KAAK,gBAAgB,KACvB,QAAO,CAAC,KAAK,WAAW,KAAK,aAAa;AAE5C,SAAO,CAAC,KAAK,UAAU;;;;;;;;;;CAWzB,MAAc,qBACZ,WACA,WACe;AAEf,MAAI,KAAK,IAAI,UAAU,KAAK,SAC1B;EAEF,MAAM,MAAM,KAAK;AAUjB,MAAI,OAAA,GAAA,qCAAA,oBANsC;GACxC,QAAQ,GAHc,YADD,KAAK,OAAO,WACkB;GAKnD,WAAW,IAAI;GAChB,CAAC,IAEc,IAAI,aAAa,IAC/B,OAAM,IAAI,MAAM,wFACd,IAAI,MACL,2CAAA,GAAA,qCAAA,qBACC,IAAI,MACL,CAAC;6LACqL;;;AAwB7L,MAAM,SAAS,SAAiB;CAE9B,IAAI,UAAU,KAAK,QAAQ,QAAQ,KAAI,CAAC,MAAM;AAE9C,KAAI,QAAQ,WAAW,KAAI,IAAI,QAAQ,SAAS,KAAI,CAClD,WAAU,QAAQ,UAAU,GAAG,QAAQ,SAAS,EAAE;AAEpD,QAAO;;AAGT,MAAM,cAAc,MAAmB,SACrC,IAAI,IAAI,CAAC,GAAG,KAAK,CAAC,QAAQ,MAAM,CAAC,KAAK,IAAI,EAAE,CAAC,CAAC;;;;;;;;;;;;;;;;;;;;;;;;;;;AA4BhD,eAAsB,oBAAoB,EACxC,KACA,IACA,QACA,IAAI,GACJ,WAC4B;CAC5B,IAAI;AACJ,KAAI,OACF,eAAc;UACLC,sBAAAA,gBAAgB,SACzB,eAAcA,sBAAAA,gBAAgB;KAE9B,eAAcH,sBAAAA;AAGhB,KACE,WACE,IAAI,IAAI;EAAC;EAAS;EAAS;EAAa,CAAC,EACzC,IAAI,IAAI,YAAY,eAAe,CACpC,CAAC,OAAO,EAET,OAAM,IAAI,MACR,2GACK,YAAY,eAAe,oBAAoB,cACrD;AAEH,KAAI,YAAY,eAAe,SAAS,UAAU,CAChD,eAAc,MAAM,YAAY,QAAQ,EAAE,SAAS,CAAC;AAGtD,eAAc,MAAM,YAAY,QAAQ,EAAE,OAAO,EAAE,UAAU,EAAE,CAAC;AAahE,QAAOI,0BAAAA,iBAAiB,KAAK;EAC3BC,0BAAAA,oBAAoB,OAAO;GAX3B,QAAQ,MAA+B;AACrC,QAAI,cAAc,EAChB,QAAO,GAAI,EAAe,SAAS;AAErC,UAAM,IAAI,MAAM,0CAA0C;;GAE5D,YAAY,OAAO,MACjB,GAAG,aAAc,EAAyB,gBAAgB;GAI3B,CAAC;GACjC,MAAM;GACL,MAAM,YAAY,EAAE,GAAG,GAAG;AAC1B,UAAO,UAAU;AACjB,UAAO,UAAU;AACjB,UAAO;;EAET;EACA,IAAI,WAAW,EAAE,MAAM,CAAC,eAAe,EAAE,CAAC;EAC1C,IAAIC,+BAAAA,oBAAoB;EACxB;EACD,CAAC"}

package/dist/chains/sql_db/sql_db_chain.js.map

@@ -1 +1 @@
-{"version":3,"file":"sql_db_chain.js","names":[],"sources":["../../../src/chains/sql_db/sql_db_chain.ts"],"sourcesContent":["import type {\n  BaseLanguageModel,\n  BaseLanguageModelInterface,\n} from \"@langchain/core/language_models/base\";\nimport type { OpenAI, TiktokenModel } from \"@langchain/openai\";\nimport { ChainValues } from \"@langchain/core/utils/types\";\nimport { BasePromptTemplate, PromptTemplate } from \"@langchain/core/prompts\";\nimport {\n  calculateMaxTokens,\n  getModelContextSize,\n} from \"@langchain/core/language_models/base\";\nimport { CallbackManagerForChainRun } from \"@langchain/core/callbacks/manager\";\nimport {\n  RunnablePassthrough,\n  RunnableSequence,\n} from \"@langchain/core/runnables\";\nimport { StringOutputParser } from \"@langchain/core/output_parsers\";\nimport {\n  DEFAULT_SQL_DATABASE_PROMPT,\n  SQL_PROMPTS_MAP,\n  SqlDialect,\n} from \"./sql_db_prompt.js\";\nimport { BaseChain, ChainInputs } from \"../base.js\";\nimport { LLMChain } from \"../llm_chain.js\";\nimport type { SqlDatabase } from \"../../sql_db.js\";\nimport { getPromptTemplateFromDataSource } from \"../../util/sql_utils.js\";\n\n/**\n * Interface that extends the ChainInputs interface and defines additional\n * fields specific to a SQL database chain. It represents the input fields\n * for a SQL database chain.\n */\nexport interface SqlDatabaseChainInput extends ChainInputs {\n  llm: BaseLanguageModelInterface;\n  database: SqlDatabase;\n  topK?: number;\n  inputKey?: string;\n  outputKey?: string;\n  sqlOutputKey?: string;\n  prompt?: PromptTemplate;\n}\n\n/**\n * Class that represents a SQL database chain in the LangChain framework.\n * It extends the BaseChain class and implements the functionality\n * specific to a SQL database chain.\n *\n * @security **Security Notice**\n * This chain generates SQL queries for the given database.\n * The SQLDatabase class provides a getTableInfo method that can be used\n * to get column information as well as sample data from the table.\n * To mitigate risk of leaking sensitive data, limit permissions\n * to read and scope to the tables that are needed.\n * Optionally, use the includesTables or ignoreTables class parameters\n * to limit which tables can/cannot be accessed.\n *\n * @link See https://js.langchain.com/docs/security for more information.\n * @example\n * ```typescript\n * const chain = new SqlDatabaseChain({\n *   llm: new OpenAI({ temperature: 0 }),\n *   database: new SqlDatabase({ ...config }),\n * });\n *\n * const result = await chain.run(\"How many tracks are there?\");\n * ```\n */\nexport class SqlDatabaseChain extends BaseChain {\n  static lc_name() {\n    return \"SqlDatabaseChain\";\n  }\n\n  // LLM wrapper to use\n  llm: BaseLanguageModelInterface;\n\n  // SQL Database to connect to.\n  database: SqlDatabase;\n\n  // Prompt to use to translate natural language to SQL.\n  prompt = DEFAULT_SQL_DATABASE_PROMPT;\n\n  // Number of results to return from the query\n  topK = 5;\n\n  inputKey = \"query\";\n\n  outputKey = \"result\";\n\n  sqlOutputKey: string | undefined = undefined;\n\n  // Whether to return the result of querying the SQL table directly.\n  returnDirect = false;\n\n  constructor(fields: SqlDatabaseChainInput) {\n    super(fields);\n    this.llm = fields.llm;\n    this.database = fields.database;\n    this.topK = fields.topK ?? this.topK;\n    this.inputKey = fields.inputKey ?? this.inputKey;\n    this.outputKey = fields.outputKey ?? this.outputKey;\n    this.sqlOutputKey = fields.sqlOutputKey ?? this.sqlOutputKey;\n    this.prompt =\n      fields.prompt ??\n      getPromptTemplateFromDataSource(this.database.appDataSource);\n  }\n\n  /** @ignore */\n  async _call(\n    values: ChainValues,\n    runManager?: CallbackManagerForChainRun\n  ): Promise<ChainValues> {\n    const llmChain = new LLMChain({\n      prompt: this.prompt,\n      llm: this.llm,\n      outputKey: this.outputKey,\n      memory: this.memory,\n    });\n    if (!(this.inputKey in values)) {\n      throw new Error(`Question key ${this.inputKey} not found.`);\n    }\n    const question: string = values[this.inputKey];\n    let inputText = `${question}\\nSQLQuery:`;\n    const tablesToUse = values.table_names_to_use;\n    const tableInfo = await this.database.getTableInfo(tablesToUse);\n\n    const llmInputs = {\n      input: inputText,\n      top_k: this.topK,\n      dialect: this.database.appDataSourceOptions.type,\n      table_info: tableInfo,\n      stop: [\"\\nSQLResult:\"],\n    };\n    await this.verifyNumberOfTokens(inputText, tableInfo);\n\n    const sqlCommand = await llmChain.predict(\n      llmInputs,\n      runManager?.getChild(\"sql_generation\")\n    );\n    let queryResult = \"\";\n    try {\n      queryResult = await this.database.appDataSource.query(sqlCommand);\n    } catch (error) {\n      console.error(error);\n    }\n\n    let finalResult;\n    if (this.returnDirect) {\n      finalResult = { [this.outputKey]: queryResult };\n    } else {\n      inputText += `${sqlCommand}\\nSQLResult: ${JSON.stringify(\n        queryResult\n      )}\\nAnswer:`;\n      llmInputs.input = inputText;\n      finalResult = {\n        [this.outputKey]: await llmChain.predict(\n          llmInputs,\n          runManager?.getChild(\"result_generation\")\n        ),\n      };\n    }\n\n    if (this.sqlOutputKey != null) {\n      finalResult[this.sqlOutputKey] = sqlCommand;\n    }\n\n    return finalResult;\n  }\n\n  _chainType() {\n    return \"sql_database_chain\" as const;\n  }\n\n  get inputKeys(): string[] {\n    return [this.inputKey];\n  }\n\n  get outputKeys(): string[] {\n    if (this.sqlOutputKey != null) {\n      return [this.outputKey, this.sqlOutputKey];\n    }\n    return [this.outputKey];\n  }\n\n  /**\n   * Private method that verifies the number of tokens in the input text and\n   * table information. It throws an error if the number of tokens exceeds\n   * the maximum allowed by the language model.\n   * @param inputText The input text.\n   * @param tableinfo The table information.\n   * @returns A promise that resolves when the verification is complete.\n   */\n  private async verifyNumberOfTokens(\n    inputText: string,\n    tableinfo: string\n  ): Promise<void> {\n    // We verify it only for OpenAI for the moment\n    if (this.llm._llmType() !== \"openai\") {\n      return;\n    }\n    const llm = this.llm as OpenAI;\n    const promptTemplate = this.prompt.template;\n    const stringWeSend = `${inputText}${promptTemplate}${tableinfo}`;\n\n    const maxToken = await calculateMaxTokens({\n      prompt: stringWeSend,\n      // Cast here to allow for other models that may not fit the union\n      modelName: llm.model as TiktokenModel,\n    });\n\n    if (maxToken < (llm.maxTokens ?? -1)) {\n      throw new Error(`The combination of the database structure and your question is too big for the model ${\n        llm.model\n      } which can compute only a max tokens of ${getModelContextSize(\n        llm.model\n      )}.\n      We suggest you to use the includeTables parameters when creating the SqlDatabase object to select only a subset of the tables. You can also use a model which can handle more tokens.`);\n    }\n  }\n}\n\nexport interface CreateSqlQueryChainFields {\n  llm: BaseLanguageModel;\n  db: SqlDatabase;\n  prompt?: BasePromptTemplate;\n  /**\n   * @default 5\n   */\n  k?: number;\n  dialect: SqlDialect;\n}\n\ntype SqlInput = {\n  question: string;\n};\n\ntype SqlInoutWithTables = SqlInput & {\n  tableNamesToUse: string[];\n};\n\nconst strip = (text: string) => {\n  // Replace escaped quotes with actual quotes\n  let newText = text.replace(/\\\\\"/g, '\"').trim();\n  // Remove wrapping quotes if the entire string is wrapped in quotes\n  if (newText.startsWith('\"') && newText.endsWith('\"')) {\n    newText = newText.substring(1, newText.length - 1);\n  }\n  return newText;\n};\n\nconst difference = (setA: Set<string>, setB: Set<string>) =>\n  new Set([...setA].filter((x) => !setB.has(x)));\n\n/**\n * Create a SQL query chain that can create SQL queries for the given database.\n * Returns a Runnable.\n *\n * @param {BaseLanguageModel} llm The language model to use in the chain.\n * @param {SqlDatabase} db The database to use in the chain.\n * @param {BasePromptTemplate | undefined} prompt The prompt to use in the chain.\n * @param {BaseLanguageModel | undefined} k The amount of docs/results to return. Passed through the prompt input value `top_k`.\n * @param {SqlDialect} dialect The SQL dialect to use in the chain.\n * @returns {Promise<RunnableSequence<Record<string, unknown>, string>>} A runnable sequence representing the chain.\n * @example ```typescript\n * const datasource = new DataSource({\n *   type: \"sqlite\",\n *   database: \"../../../../Chinook.db\",\n * });\n * const db = await SqlDatabase.fromDataSourceParams({\n *   appDataSource: datasource,\n * });\n * const llm = new ChatOpenAI({ model: \"gpt-4o-mini\", temperature: 0 });\n * const chain = await createSqlQueryChain({\n *   llm,\n *   db,\n *   dialect: \"sqlite\",\n * });\n * ```\n */\nexport async function createSqlQueryChain({\n  llm,\n  db,\n  prompt,\n  k = 5,\n  dialect,\n}: CreateSqlQueryChainFields) {\n  let promptToUse: BasePromptTemplate;\n  if (prompt) {\n    promptToUse = prompt;\n  } else if (SQL_PROMPTS_MAP[dialect]) {\n    promptToUse = SQL_PROMPTS_MAP[dialect];\n  } else {\n    promptToUse = DEFAULT_SQL_DATABASE_PROMPT;\n  }\n\n  if (\n    difference(\n      new Set([\"input\", \"top_k\", \"table_info\"]),\n      new Set(promptToUse.inputVariables)\n    ).size > 0\n  ) {\n    throw new Error(\n      `Prompt must have input variables: 'input', 'top_k', 'table_info'. Received prompt with input variables: ` +\n        `${promptToUse.inputVariables}. Full prompt:\\n\\n${promptToUse}`\n    );\n  }\n  if (promptToUse.inputVariables.includes(\"dialect\")) {\n    promptToUse = await promptToUse.partial({ dialect });\n  }\n\n  promptToUse = await promptToUse.partial({ top_k: k.toString() });\n\n  const inputs = {\n    input: (x: Record<string, unknown>) => {\n      if (\"question\" in x) {\n        return `${(x as SqlInput).question}\\nSQLQuery: `;\n      }\n      throw new Error(\"Input must include a question property.\");\n    },\n    table_info: async (x: Record<string, unknown>) =>\n      db.getTableInfo((x as SqlInoutWithTables).tableNamesToUse),\n  };\n\n  return RunnableSequence.from([\n    RunnablePassthrough.assign(inputs),\n    (x) => {\n      const newInputs = { ...x };\n      delete newInputs.question;\n      delete newInputs.tableNamesToUse;\n      return newInputs;\n    },\n    promptToUse,\n    llm.withConfig({ stop: [\"\\nSQLResult:\"] }),\n    new StringOutputParser(),\n    strip,\n  ]);\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAmEA,IAAa,mBAAb,cAAsC,UAAU;CAC9C,OAAO,UAAU;AACf,SAAO;;CAIT;CAGA;CAGA,SAAS;CAGT,OAAO;CAEP,WAAW;CAEX,YAAY;CAEZ,eAAmC,KAAA;CAGnC,eAAe;CAEf,YAAY,QAA+B;AACzC,QAAM,OAAO;AACb,OAAK,MAAM,OAAO;AAClB,OAAK,WAAW,OAAO;AACvB,OAAK,OAAO,OAAO,QAAQ,KAAK;AAChC,OAAK,WAAW,OAAO,YAAY,KAAK;AACxC,OAAK,YAAY,OAAO,aAAa,KAAK;AAC1C,OAAK,eAAe,OAAO,gBAAgB,KAAK;AAChD,OAAK,SACH,OAAO,UACP,gCAAgC,KAAK,SAAS,cAAc;;;CAIhE,MAAM,MACJ,QACA,YACsB;EACtB,MAAM,WAAW,IAAI,SAAS;GAC5B,QAAQ,KAAK;GACb,KAAK,KAAK;GACV,WAAW,KAAK;GAChB,QAAQ,KAAK;GACd,CAAC;AACF,MAAI,EAAE,KAAK,YAAY,QACrB,OAAM,IAAI,MAAM,gBAAgB,KAAK,SAAS,aAAa;EAG7D,IAAI,YAAY,GADS,OAAO,KAAK,UACT;EAC5B,MAAM,cAAc,OAAO;EAC3B,MAAM,YAAY,MAAM,KAAK,SAAS,aAAa,YAAY;EAE/D,MAAM,YAAY;GAChB,OAAO;GACP,OAAO,KAAK;GACZ,SAAS,KAAK,SAAS,qBAAqB;GAC5C,YAAY;GACZ,MAAM,CAAC,eAAe;GACvB;AACD,QAAM,KAAK,qBAAqB,WAAW,UAAU;EAErD,MAAM,aAAa,MAAM,SAAS,QAChC,WACA,YAAY,SAAS,iBAAiB,CACvC;EACD,IAAI,cAAc;AAClB,MAAI;AACF,iBAAc,MAAM,KAAK,SAAS,cAAc,MAAM,WAAW;WAC1D,OAAO;AACd,WAAQ,MAAM,MAAM;;EAGtB,IAAI;AACJ,MAAI,KAAK,aACP,eAAc,GAAG,KAAK,YAAY,aAAa;OAC1C;AACL,gBAAa,GAAG,WAAW,eAAe,KAAK,UAC7C,YACD,CAAC;AACF,aAAU,QAAQ;AAClB,iBAAc,GACX,KAAK,YAAY,MAAM,SAAS,QAC/B,WACA,YAAY,SAAS,oBAAoB,CAC1C,EACF;;AAGH,MAAI,KAAK,gBAAgB,KACvB,aAAY,KAAK,gBAAgB;AAGnC,SAAO;;CAGT,aAAa;AACX,SAAO;;CAGT,IAAI,YAAsB;AACxB,SAAO,CAAC,KAAK,SAAS;;CAGxB,IAAI,aAAuB;AACzB,MAAI,KAAK,gBAAgB,KACvB,QAAO,CAAC,KAAK,WAAW,KAAK,aAAa;AAE5C,SAAO,CAAC,KAAK,UAAU;;;;;;;;;;CAWzB,MAAc,qBACZ,WACA,WACe;AAEf,MAAI,KAAK,IAAI,UAAU,KAAK,SAC1B;EAEF,MAAM,MAAM,KAAK;AAUjB,MANiB,MAAM,mBAAmB;GACxC,QAHmB,GAAG,YADD,KAAK,OAAO,WACkB;GAKnD,WAAW,IAAI;GAChB,CAAC,IAEc,IAAI,aAAa,IAC/B,OAAM,IAAI,MAAM,wFACd,IAAI,MACL,0CAA0C,oBACzC,IAAI,MACL,CAAC;6LACqL;;;AAwB7L,MAAM,SAAS,SAAiB;CAE9B,IAAI,UAAU,KAAK,QAAQ,QAAQ,KAAI,CAAC,MAAM;AAE9C,KAAI,QAAQ,WAAW,KAAI,IAAI,QAAQ,SAAS,KAAI,CAClD,WAAU,QAAQ,UAAU,GAAG,QAAQ,SAAS,EAAE;AAEpD,QAAO;;AAGT,MAAM,cAAc,MAAmB,SACrC,IAAI,IAAI,CAAC,GAAG,KAAK,CAAC,QAAQ,MAAM,CAAC,KAAK,IAAI,EAAE,CAAC,CAAC;;;;;;;;;;;;;;;;;;;;;;;;;;;AA4BhD,eAAsB,oBAAoB,EACxC,KACA,IACA,QACA,IAAI,GACJ,WAC4B;CAC5B,IAAI;AACJ,KAAI,OACF,eAAc;UACL,gBAAgB,SACzB,eAAc,gBAAgB;KAE9B,eAAc;AAGhB,KACE,WACE,IAAI,IAAI;EAAC;EAAS;EAAS;EAAa,CAAC,EACzC,IAAI,IAAI,YAAY,eAAe,CACpC,CAAC,OAAO,EAET,OAAM,IAAI,MACR,2GACK,YAAY,eAAe,oBAAoB,cACrD;AAEH,KAAI,YAAY,eAAe,SAAS,UAAU,CAChD,eAAc,MAAM,YAAY,QAAQ,EAAE,SAAS,CAAC;AAGtD,eAAc,MAAM,YAAY,QAAQ,EAAE,OAAO,EAAE,UAAU,EAAE,CAAC;AAahE,QAAO,iBAAiB,KAAK;EAC3B,oBAAoB,OAZP;GACb,QAAQ,MAA+B;AACrC,QAAI,cAAc,EAChB,QAAO,GAAI,EAAe,SAAS;AAErC,UAAM,IAAI,MAAM,0CAA0C;;GAE5D,YAAY,OAAO,MACjB,GAAG,aAAc,EAAyB,gBAAgB;GAC7D,CAGmC;GACjC,MAAM;GACL,MAAM,YAAY,EAAE,GAAG,GAAG;AAC1B,UAAO,UAAU;AACjB,UAAO,UAAU;AACjB,UAAO;;EAET;EACA,IAAI,WAAW,EAAE,MAAM,CAAC,eAAe,EAAE,CAAC;EAC1C,IAAI,oBAAoB;EACxB;EACD,CAAC"}
+{"version":3,"file":"sql_db_chain.js","names":[],"sources":["../../../src/chains/sql_db/sql_db_chain.ts"],"sourcesContent":["import type {\n  BaseLanguageModel,\n  BaseLanguageModelInterface,\n} from \"@langchain/core/language_models/base\";\nimport type { OpenAI, TiktokenModel } from \"@langchain/openai\";\nimport { ChainValues } from \"@langchain/core/utils/types\";\nimport { BasePromptTemplate, PromptTemplate } from \"@langchain/core/prompts\";\nimport {\n  calculateMaxTokens,\n  getModelContextSize,\n} from \"@langchain/core/language_models/base\";\nimport { CallbackManagerForChainRun } from \"@langchain/core/callbacks/manager\";\nimport {\n  RunnablePassthrough,\n  RunnableSequence,\n} from \"@langchain/core/runnables\";\nimport { StringOutputParser } from \"@langchain/core/output_parsers\";\nimport {\n  DEFAULT_SQL_DATABASE_PROMPT,\n  SQL_PROMPTS_MAP,\n  SqlDialect,\n} from \"./sql_db_prompt.js\";\nimport { BaseChain, ChainInputs } from \"../base.js\";\nimport { LLMChain } from \"../llm_chain.js\";\nimport type { SqlDatabase } from \"../../sql_db.js\";\nimport { getPromptTemplateFromDataSource } from \"../../util/sql_utils.js\";\n\n/**\n * Interface that extends the ChainInputs interface and defines additional\n * fields specific to a SQL database chain. It represents the input fields\n * for a SQL database chain.\n */\nexport interface SqlDatabaseChainInput extends ChainInputs {\n  llm: BaseLanguageModelInterface;\n  database: SqlDatabase;\n  topK?: number;\n  inputKey?: string;\n  outputKey?: string;\n  sqlOutputKey?: string;\n  prompt?: PromptTemplate;\n}\n\n/**\n * Class that represents a SQL database chain in the LangChain framework.\n * It extends the BaseChain class and implements the functionality\n * specific to a SQL database chain.\n *\n * @security **Security Notice**\n * This chain generates SQL queries for the given database.\n * The SQLDatabase class provides a getTableInfo method that can be used\n * to get column information as well as sample data from the table.\n * To mitigate risk of leaking sensitive data, limit permissions\n * to read and scope to the tables that are needed.\n * Optionally, use the includesTables or ignoreTables class parameters\n * to limit which tables can/cannot be accessed.\n *\n * @link See https://js.langchain.com/docs/security for more information.\n * @example\n * ```typescript\n * const chain = new SqlDatabaseChain({\n *   llm: new OpenAI({ temperature: 0 }),\n *   database: new SqlDatabase({ ...config }),\n * });\n *\n * const result = await chain.run(\"How many tracks are there?\");\n * ```\n */\nexport class SqlDatabaseChain extends BaseChain {\n  static lc_name() {\n    return \"SqlDatabaseChain\";\n  }\n\n  // LLM wrapper to use\n  llm: BaseLanguageModelInterface;\n\n  // SQL Database to connect to.\n  database: SqlDatabase;\n\n  // Prompt to use to translate natural language to SQL.\n  prompt = DEFAULT_SQL_DATABASE_PROMPT;\n\n  // Number of results to return from the query\n  topK = 5;\n\n  inputKey = \"query\";\n\n  outputKey = \"result\";\n\n  sqlOutputKey: string | undefined = undefined;\n\n  // Whether to return the result of querying the SQL table directly.\n  returnDirect = false;\n\n  constructor(fields: SqlDatabaseChainInput) {\n    super(fields);\n    this.llm = fields.llm;\n    this.database = fields.database;\n    this.topK = fields.topK ?? this.topK;\n    this.inputKey = fields.inputKey ?? this.inputKey;\n    this.outputKey = fields.outputKey ?? this.outputKey;\n    this.sqlOutputKey = fields.sqlOutputKey ?? this.sqlOutputKey;\n    this.prompt =\n      fields.prompt ??\n      getPromptTemplateFromDataSource(this.database.appDataSource);\n  }\n\n  /** @ignore */\n  async _call(\n    values: ChainValues,\n    runManager?: CallbackManagerForChainRun\n  ): Promise<ChainValues> {\n    const llmChain = new LLMChain({\n      prompt: this.prompt,\n      llm: this.llm,\n      outputKey: this.outputKey,\n      memory: this.memory,\n    });\n    if (!(this.inputKey in values)) {\n      throw new Error(`Question key ${this.inputKey} not found.`);\n    }\n    const question: string = values[this.inputKey];\n    let inputText = `${question}\\nSQLQuery:`;\n    const tablesToUse = values.table_names_to_use;\n    const tableInfo = await this.database.getTableInfo(tablesToUse);\n\n    const llmInputs = {\n      input: inputText,\n      top_k: this.topK,\n      dialect: this.database.appDataSourceOptions.type,\n      table_info: tableInfo,\n      stop: [\"\\nSQLResult:\"],\n    };\n    await this.verifyNumberOfTokens(inputText, tableInfo);\n\n    const sqlCommand = await llmChain.predict(\n      llmInputs,\n      runManager?.getChild(\"sql_generation\")\n    );\n    let queryResult = \"\";\n    try {\n      queryResult = await this.database.appDataSource.query(sqlCommand);\n    } catch (error) {\n      console.error(error);\n    }\n\n    let finalResult;\n    if (this.returnDirect) {\n      finalResult = { [this.outputKey]: queryResult };\n    } else {\n      inputText += `${sqlCommand}\\nSQLResult: ${JSON.stringify(\n        queryResult\n      )}\\nAnswer:`;\n      llmInputs.input = inputText;\n      finalResult = {\n        [this.outputKey]: await llmChain.predict(\n          llmInputs,\n          runManager?.getChild(\"result_generation\")\n        ),\n      };\n    }\n\n    if (this.sqlOutputKey != null) {\n      finalResult[this.sqlOutputKey] = sqlCommand;\n    }\n\n    return finalResult;\n  }\n\n  _chainType() {\n    return \"sql_database_chain\" as const;\n  }\n\n  get inputKeys(): string[] {\n    return [this.inputKey];\n  }\n\n  get outputKeys(): string[] {\n    if (this.sqlOutputKey != null) {\n      return [this.outputKey, this.sqlOutputKey];\n    }\n    return [this.outputKey];\n  }\n\n  /**\n   * Private method that verifies the number of tokens in the input text and\n   * table information. It throws an error if the number of tokens exceeds\n   * the maximum allowed by the language model.\n   * @param inputText The input text.\n   * @param tableinfo The table information.\n   * @returns A promise that resolves when the verification is complete.\n   */\n  private async verifyNumberOfTokens(\n    inputText: string,\n    tableinfo: string\n  ): Promise<void> {\n    // We verify it only for OpenAI for the moment\n    if (this.llm._llmType() !== \"openai\") {\n      return;\n    }\n    const llm = this.llm as OpenAI;\n    const promptTemplate = this.prompt.template;\n    const stringWeSend = `${inputText}${promptTemplate}${tableinfo}`;\n\n    const maxToken = await calculateMaxTokens({\n      prompt: stringWeSend,\n      // Cast here to allow for other models that may not fit the union\n      modelName: llm.model as TiktokenModel,\n    });\n\n    if (maxToken < (llm.maxTokens ?? -1)) {\n      throw new Error(`The combination of the database structure and your question is too big for the model ${\n        llm.model\n      } which can compute only a max tokens of ${getModelContextSize(\n        llm.model\n      )}.\n      We suggest you to use the includeTables parameters when creating the SqlDatabase object to select only a subset of the tables. You can also use a model which can handle more tokens.`);\n    }\n  }\n}\n\nexport interface CreateSqlQueryChainFields {\n  llm: BaseLanguageModel;\n  db: SqlDatabase;\n  prompt?: BasePromptTemplate;\n  /**\n   * @default 5\n   */\n  k?: number;\n  dialect: SqlDialect;\n}\n\ntype SqlInput = {\n  question: string;\n};\n\ntype SqlInoutWithTables = SqlInput & {\n  tableNamesToUse: string[];\n};\n\nconst strip = (text: string) => {\n  // Replace escaped quotes with actual quotes\n  let newText = text.replace(/\\\\\"/g, '\"').trim();\n  // Remove wrapping quotes if the entire string is wrapped in quotes\n  if (newText.startsWith('\"') && newText.endsWith('\"')) {\n    newText = newText.substring(1, newText.length - 1);\n  }\n  return newText;\n};\n\nconst difference = (setA: Set<string>, setB: Set<string>) =>\n  new Set([...setA].filter((x) => !setB.has(x)));\n\n/**\n * Create a SQL query chain that can create SQL queries for the given database.\n * Returns a Runnable.\n *\n * @param {BaseLanguageModel} llm The language model to use in the chain.\n * @param {SqlDatabase} db The database to use in the chain.\n * @param {BasePromptTemplate | undefined} prompt The prompt to use in the chain.\n * @param {BaseLanguageModel | undefined} k The amount of docs/results to return. Passed through the prompt input value `top_k`.\n * @param {SqlDialect} dialect The SQL dialect to use in the chain.\n * @returns {Promise<RunnableSequence<Record<string, unknown>, string>>} A runnable sequence representing the chain.\n * @example ```typescript\n * const datasource = new DataSource({\n *   type: \"sqlite\",\n *   database: \"../../../../Chinook.db\",\n * });\n * const db = await SqlDatabase.fromDataSourceParams({\n *   appDataSource: datasource,\n * });\n * const llm = new ChatOpenAI({ model: \"gpt-4o-mini\", temperature: 0 });\n * const chain = await createSqlQueryChain({\n *   llm,\n *   db,\n *   dialect: \"sqlite\",\n * });\n * ```\n */\nexport async function createSqlQueryChain({\n  llm,\n  db,\n  prompt,\n  k = 5,\n  dialect,\n}: CreateSqlQueryChainFields) {\n  let promptToUse: BasePromptTemplate;\n  if (prompt) {\n    promptToUse = prompt;\n  } else if (SQL_PROMPTS_MAP[dialect]) {\n    promptToUse = SQL_PROMPTS_MAP[dialect];\n  } else {\n    promptToUse = DEFAULT_SQL_DATABASE_PROMPT;\n  }\n\n  if (\n    difference(\n      new Set([\"input\", \"top_k\", \"table_info\"]),\n      new Set(promptToUse.inputVariables)\n    ).size > 0\n  ) {\n    throw new Error(\n      `Prompt must have input variables: 'input', 'top_k', 'table_info'. Received prompt with input variables: ` +\n        `${promptToUse.inputVariables}. Full prompt:\\n\\n${promptToUse}`\n    );\n  }\n  if (promptToUse.inputVariables.includes(\"dialect\")) {\n    promptToUse = await promptToUse.partial({ dialect });\n  }\n\n  promptToUse = await promptToUse.partial({ top_k: k.toString() });\n\n  const inputs = {\n    input: (x: Record<string, unknown>) => {\n      if (\"question\" in x) {\n        return `${(x as SqlInput).question}\\nSQLQuery: `;\n      }\n      throw new Error(\"Input must include a question property.\");\n    },\n    table_info: async (x: Record<string, unknown>) =>\n      db.getTableInfo((x as SqlInoutWithTables).tableNamesToUse),\n  };\n\n  return RunnableSequence.from([\n    RunnablePassthrough.assign(inputs),\n    (x) => {\n      const newInputs = { ...x };\n      delete newInputs.question;\n      delete newInputs.tableNamesToUse;\n      return newInputs;\n    },\n    promptToUse,\n    llm.withConfig({ stop: [\"\\nSQLResult:\"] }),\n    new StringOutputParser(),\n    strip,\n  ]);\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAmEA,IAAa,mBAAb,cAAsC,UAAU;CAC9C,OAAO,UAAU;AACf,SAAO;;CAIT;CAGA;CAGA,SAAS;CAGT,OAAO;CAEP,WAAW;CAEX,YAAY;CAEZ,eAAmC,KAAA;CAGnC,eAAe;CAEf,YAAY,QAA+B;AACzC,QAAM,OAAO;AACb,OAAK,MAAM,OAAO;AAClB,OAAK,WAAW,OAAO;AACvB,OAAK,OAAO,OAAO,QAAQ,KAAK;AAChC,OAAK,WAAW,OAAO,YAAY,KAAK;AACxC,OAAK,YAAY,OAAO,aAAa,KAAK;AAC1C,OAAK,eAAe,OAAO,gBAAgB,KAAK;AAChD,OAAK,SACH,OAAO,UACP,gCAAgC,KAAK,SAAS,cAAc;;;CAIhE,MAAM,MACJ,QACA,YACsB;EACtB,MAAM,WAAW,IAAI,SAAS;GAC5B,QAAQ,KAAK;GACb,KAAK,KAAK;GACV,WAAW,KAAK;GAChB,QAAQ,KAAK;GACd,CAAC;AACF,MAAI,EAAE,KAAK,YAAY,QACrB,OAAM,IAAI,MAAM,gBAAgB,KAAK,SAAS,aAAa;EAG7D,IAAI,YAAY,GADS,OAAO,KAAK,UACT;EAC5B,MAAM,cAAc,OAAO;EAC3B,MAAM,YAAY,MAAM,KAAK,SAAS,aAAa,YAAY;EAE/D,MAAM,YAAY;GAChB,OAAO;GACP,OAAO,KAAK;GACZ,SAAS,KAAK,SAAS,qBAAqB;GAC5C,YAAY;GACZ,MAAM,CAAC,eAAe;GACvB;AACD,QAAM,KAAK,qBAAqB,WAAW,UAAU;EAErD,MAAM,aAAa,MAAM,SAAS,QAChC,WACA,YAAY,SAAS,iBAAiB,CACvC;EACD,IAAI,cAAc;AAClB,MAAI;AACF,iBAAc,MAAM,KAAK,SAAS,cAAc,MAAM,WAAW;WAC1D,OAAO;AACd,WAAQ,MAAM,MAAM;;EAGtB,IAAI;AACJ,MAAI,KAAK,aACP,eAAc,GAAG,KAAK,YAAY,aAAa;OAC1C;AACL,gBAAa,GAAG,WAAW,eAAe,KAAK,UAC7C,YACD,CAAC;AACF,aAAU,QAAQ;AAClB,iBAAc,GACX,KAAK,YAAY,MAAM,SAAS,QAC/B,WACA,YAAY,SAAS,oBAAoB,CAC1C,EACF;;AAGH,MAAI,KAAK,gBAAgB,KACvB,aAAY,KAAK,gBAAgB;AAGnC,SAAO;;CAGT,aAAa;AACX,SAAO;;CAGT,IAAI,YAAsB;AACxB,SAAO,CAAC,KAAK,SAAS;;CAGxB,IAAI,aAAuB;AACzB,MAAI,KAAK,gBAAgB,KACvB,QAAO,CAAC,KAAK,WAAW,KAAK,aAAa;AAE5C,SAAO,CAAC,KAAK,UAAU;;;;;;;;;;CAWzB,MAAc,qBACZ,WACA,WACe;AAEf,MAAI,KAAK,IAAI,UAAU,KAAK,SAC1B;EAEF,MAAM,MAAM,KAAK;AAUjB,MAAI,MANmB,mBAAmB;GACxC,QAAQ,GAHc,YADD,KAAK,OAAO,WACkB;GAKnD,WAAW,IAAI;GAChB,CAAC,IAEc,IAAI,aAAa,IAC/B,OAAM,IAAI,MAAM,wFACd,IAAI,MACL,0CAA0C,oBACzC,IAAI,MACL,CAAC;6LACqL;;;AAwB7L,MAAM,SAAS,SAAiB;CAE9B,IAAI,UAAU,KAAK,QAAQ,QAAQ,KAAI,CAAC,MAAM;AAE9C,KAAI,QAAQ,WAAW,KAAI,IAAI,QAAQ,SAAS,KAAI,CAClD,WAAU,QAAQ,UAAU,GAAG,QAAQ,SAAS,EAAE;AAEpD,QAAO;;AAGT,MAAM,cAAc,MAAmB,SACrC,IAAI,IAAI,CAAC,GAAG,KAAK,CAAC,QAAQ,MAAM,CAAC,KAAK,IAAI,EAAE,CAAC,CAAC;;;;;;;;;;;;;;;;;;;;;;;;;;;AA4BhD,eAAsB,oBAAoB,EACxC,KACA,IACA,QACA,IAAI,GACJ,WAC4B;CAC5B,IAAI;AACJ,KAAI,OACF,eAAc;UACL,gBAAgB,SACzB,eAAc,gBAAgB;KAE9B,eAAc;AAGhB,KACE,WACE,IAAI,IAAI;EAAC;EAAS;EAAS;EAAa,CAAC,EACzC,IAAI,IAAI,YAAY,eAAe,CACpC,CAAC,OAAO,EAET,OAAM,IAAI,MACR,2GACK,YAAY,eAAe,oBAAoB,cACrD;AAEH,KAAI,YAAY,eAAe,SAAS,UAAU,CAChD,eAAc,MAAM,YAAY,QAAQ,EAAE,SAAS,CAAC;AAGtD,eAAc,MAAM,YAAY,QAAQ,EAAE,OAAO,EAAE,UAAU,EAAE,CAAC;AAahE,QAAO,iBAAiB,KAAK;EAC3B,oBAAoB,OAAO;GAX3B,QAAQ,MAA+B;AACrC,QAAI,cAAc,EAChB,QAAO,GAAI,EAAe,SAAS;AAErC,UAAM,IAAI,MAAM,0CAA0C;;GAE5D,YAAY,OAAO,MACjB,GAAG,aAAc,EAAyB,gBAAgB;GAI3B,CAAC;GACjC,MAAM;GACL,MAAM,YAAY,EAAE,GAAG,GAAG;AAC1B,UAAO,UAAU;AACjB,UAAO,UAAU;AACjB,UAAO;;EAET;EACA,IAAI,WAAW,EAAE,MAAM,CAAC,eAAe,EAAE,CAAC;EAC1C,IAAI,oBAAoB;EACxB;EACD,CAAC"}

package/dist/chains/summarization/load.cjs.map

@@ -1 +1 @@
-{"version":3,"file":"load.cjs","names":["DEFAULT_PROMPT","StuffDocumentsChain","LLMChain","MapReduceDocumentsChain","REFINE_PROMPT","RefineDocumentsChain"],"sources":["../../../src/chains/summarization/load.ts"],"sourcesContent":["import type { BaseLanguageModelInterface } from \"@langchain/core/language_models/base\";\nimport { BasePromptTemplate } from \"@langchain/core/prompts\";\nimport { LLMChain } from \"../llm_chain.js\";\nimport {\n  StuffDocumentsChain,\n  MapReduceDocumentsChain,\n  RefineDocumentsChain,\n  MapReduceDocumentsChainInput,\n} from \"../combine_docs_chain.js\";\nimport { DEFAULT_PROMPT } from \"./stuff_prompts.js\";\nimport { REFINE_PROMPT } from \"./refine_prompts.js\";\n\n/**\n * Type for the base parameters that can be used to configure a\n * summarization chain.\n */\ntype BaseParams = {\n  verbose?: boolean;\n};\n\n/** @interface */\nexport type SummarizationChainParams = BaseParams &\n  (\n    | {\n        type?: \"stuff\";\n        prompt?: BasePromptTemplate;\n      }\n    | ({\n        type?: \"map_reduce\";\n        combineMapPrompt?: BasePromptTemplate;\n        combinePrompt?: BasePromptTemplate;\n        combineLLM?: BaseLanguageModelInterface;\n      } & Pick<MapReduceDocumentsChainInput, \"returnIntermediateSteps\">)\n    | {\n        type?: \"refine\";\n        refinePrompt?: BasePromptTemplate;\n        refineLLM?: BaseLanguageModelInterface;\n        questionPrompt?: BasePromptTemplate;\n      }\n  );\n\nexport const loadSummarizationChain = (\n  llm: BaseLanguageModelInterface,\n  params: SummarizationChainParams = { type: \"map_reduce\" }\n) => {\n  const { verbose } = params;\n  if (params.type === \"stuff\") {\n    const { prompt = DEFAULT_PROMPT } = params;\n    const llmChain = new LLMChain({ prompt, llm, verbose });\n    const chain = new StuffDocumentsChain({\n      llmChain,\n      documentVariableName: \"text\",\n      verbose,\n    });\n    return chain;\n  }\n  if (params.type === \"map_reduce\") {\n    const {\n      combineMapPrompt = DEFAULT_PROMPT,\n      combinePrompt = DEFAULT_PROMPT,\n      combineLLM,\n      returnIntermediateSteps,\n    } = params;\n    const llmChain = new LLMChain({ prompt: combineMapPrompt, llm, verbose });\n    const combineLLMChain = new LLMChain({\n      prompt: combinePrompt,\n      llm: combineLLM ?? llm,\n      verbose,\n    });\n    const combineDocumentChain = new StuffDocumentsChain({\n      llmChain: combineLLMChain,\n      documentVariableName: \"text\",\n      verbose,\n    });\n    const chain = new MapReduceDocumentsChain({\n      llmChain,\n      combineDocumentChain,\n      documentVariableName: \"text\",\n      returnIntermediateSteps,\n      verbose,\n    });\n    return chain;\n  }\n  if (params.type === \"refine\") {\n    const {\n      refinePrompt = REFINE_PROMPT,\n      refineLLM,\n      questionPrompt = DEFAULT_PROMPT,\n    } = params;\n    const llmChain = new LLMChain({ prompt: questionPrompt, llm, verbose });\n    const refineLLMChain = new LLMChain({\n      prompt: refinePrompt,\n      llm: refineLLM ?? llm,\n      verbose,\n    });\n    const chain = new RefineDocumentsChain({\n      llmChain,\n      refineLLMChain,\n      documentVariableName: \"text\",\n      verbose,\n    });\n    return chain;\n  }\n  throw new Error(`Invalid _type: ${params.type}`);\n};\n"],"mappings":";;;;;AAyCA,MAAa,0BACX,KACA,SAAmC,EAAE,MAAM,cAAc,KACtD;CACH,MAAM,EAAE,YAAY;AACpB,KAAI,OAAO,SAAS,SAAS;EAC3B,MAAM,EAAE,SAASA,sBAAAA,mBAAmB;AAOpC,SALc,IAAIC,2BAAAA,oBAAoB;GACpC,UAFe,IAAIC,kBAAAA,SAAS;IAAE;IAAQ;IAAK;IAAS,CAAC;GAGrD,sBAAsB;GACtB;GACD,CAAC;;AAGJ,KAAI,OAAO,SAAS,cAAc;EAChC,MAAM,EACJ,mBAAmBF,sBAAAA,gBACnB,gBAAgBA,sBAAAA,gBAChB,YACA,4BACE;AAmBJ,SAPc,IAAIG,2BAAAA,wBAAwB;GACxC,UAZe,IAAID,kBAAAA,SAAS;IAAE,QAAQ;IAAkB;IAAK;IAAS,CAAC;GAavE,sBAP2B,IAAID,2BAAAA,oBAAoB;IACnD,UANsB,IAAIC,kBAAAA,SAAS;KACnC,QAAQ;KACR,KAAK,cAAc;KACnB;KACD,CAAC;IAGA,sBAAsB;IACtB;IACD,CAAC;GAIA,sBAAsB;GACtB;GACA;GACD,CAAC;;AAGJ,KAAI,OAAO,SAAS,UAAU;EAC5B,MAAM,EACJ,eAAeE,uBAAAA,eACf,WACA,iBAAiBJ,sBAAAA,mBACf;AAaJ,SANc,IAAIK,2BAAAA,qBAAqB;GACrC,UAPe,IAAIH,kBAAAA,SAAS;IAAE,QAAQ;IAAgB;IAAK;IAAS,CAAC;GAQrE,gBAPqB,IAAIA,kBAAAA,SAAS;IAClC,QAAQ;IACR,KAAK,aAAa;IAClB;IACD,CAAC;GAIA,sBAAsB;GACtB;GACD,CAAC;;AAGJ,OAAM,IAAI,MAAM,kBAAkB,OAAO,OAAO"}
+{"version":3,"file":"load.cjs","names":["DEFAULT_PROMPT","StuffDocumentsChain","LLMChain","MapReduceDocumentsChain","REFINE_PROMPT","RefineDocumentsChain"],"sources":["../../../src/chains/summarization/load.ts"],"sourcesContent":["import type { BaseLanguageModelInterface } from \"@langchain/core/language_models/base\";\nimport { BasePromptTemplate } from \"@langchain/core/prompts\";\nimport { LLMChain } from \"../llm_chain.js\";\nimport {\n  StuffDocumentsChain,\n  MapReduceDocumentsChain,\n  RefineDocumentsChain,\n  MapReduceDocumentsChainInput,\n} from \"../combine_docs_chain.js\";\nimport { DEFAULT_PROMPT } from \"./stuff_prompts.js\";\nimport { REFINE_PROMPT } from \"./refine_prompts.js\";\n\n/**\n * Type for the base parameters that can be used to configure a\n * summarization chain.\n */\ntype BaseParams = {\n  verbose?: boolean;\n};\n\n/** @interface */\nexport type SummarizationChainParams = BaseParams &\n  (\n    | {\n        type?: \"stuff\";\n        prompt?: BasePromptTemplate;\n      }\n    | ({\n        type?: \"map_reduce\";\n        combineMapPrompt?: BasePromptTemplate;\n        combinePrompt?: BasePromptTemplate;\n        combineLLM?: BaseLanguageModelInterface;\n      } & Pick<MapReduceDocumentsChainInput, \"returnIntermediateSteps\">)\n    | {\n        type?: \"refine\";\n        refinePrompt?: BasePromptTemplate;\n        refineLLM?: BaseLanguageModelInterface;\n        questionPrompt?: BasePromptTemplate;\n      }\n  );\n\nexport const loadSummarizationChain = (\n  llm: BaseLanguageModelInterface,\n  params: SummarizationChainParams = { type: \"map_reduce\" }\n) => {\n  const { verbose } = params;\n  if (params.type === \"stuff\") {\n    const { prompt = DEFAULT_PROMPT } = params;\n    const llmChain = new LLMChain({ prompt, llm, verbose });\n    const chain = new StuffDocumentsChain({\n      llmChain,\n      documentVariableName: \"text\",\n      verbose,\n    });\n    return chain;\n  }\n  if (params.type === \"map_reduce\") {\n    const {\n      combineMapPrompt = DEFAULT_PROMPT,\n      combinePrompt = DEFAULT_PROMPT,\n      combineLLM,\n      returnIntermediateSteps,\n    } = params;\n    const llmChain = new LLMChain({ prompt: combineMapPrompt, llm, verbose });\n    const combineLLMChain = new LLMChain({\n      prompt: combinePrompt,\n      llm: combineLLM ?? llm,\n      verbose,\n    });\n    const combineDocumentChain = new StuffDocumentsChain({\n      llmChain: combineLLMChain,\n      documentVariableName: \"text\",\n      verbose,\n    });\n    const chain = new MapReduceDocumentsChain({\n      llmChain,\n      combineDocumentChain,\n      documentVariableName: \"text\",\n      returnIntermediateSteps,\n      verbose,\n    });\n    return chain;\n  }\n  if (params.type === \"refine\") {\n    const {\n      refinePrompt = REFINE_PROMPT,\n      refineLLM,\n      questionPrompt = DEFAULT_PROMPT,\n    } = params;\n    const llmChain = new LLMChain({ prompt: questionPrompt, llm, verbose });\n    const refineLLMChain = new LLMChain({\n      prompt: refinePrompt,\n      llm: refineLLM ?? llm,\n      verbose,\n    });\n    const chain = new RefineDocumentsChain({\n      llmChain,\n      refineLLMChain,\n      documentVariableName: \"text\",\n      verbose,\n    });\n    return chain;\n  }\n  throw new Error(`Invalid _type: ${params.type}`);\n};\n"],"mappings":";;;;;AAyCA,MAAa,0BACX,KACA,SAAmC,EAAE,MAAM,cAAc,KACtD;CACH,MAAM,EAAE,YAAY;AACpB,KAAI,OAAO,SAAS,SAAS;EAC3B,MAAM,EAAE,SAASA,sBAAAA,mBAAmB;AAOpC,SAAO,IALWC,2BAAAA,oBAAoB;GACpC,UAAA,IAFmBC,kBAAAA,SAAS;IAAE;IAAQ;IAAK;IAAS,CAE5C;GACR,sBAAsB;GACtB;GACD,CACW;;AAEd,KAAI,OAAO,SAAS,cAAc;EAChC,MAAM,EACJ,mBAAmBF,sBAAAA,gBACnB,gBAAgBA,sBAAAA,gBAChB,YACA,4BACE;AAmBJ,SAAO,IAPWG,2BAAAA,wBAAwB;GACxC,UAAA,IAZmBD,kBAAAA,SAAS;IAAE,QAAQ;IAAkB;IAAK;IAAS,CAY9D;GACR,sBAAA,IAP+BD,2BAAAA,oBAAoB;IACnD,UAAU,IANgBC,kBAAAA,SAAS;KACnC,QAAQ;KACR,KAAK,cAAc;KACnB;KACD,CAE0B;IACzB,sBAAsB;IACtB;IACD,CAGqB;GACpB,sBAAsB;GACtB;GACA;GACD,CACW;;AAEd,KAAI,OAAO,SAAS,UAAU;EAC5B,MAAM,EACJ,eAAeE,uBAAAA,eACf,WACA,iBAAiBJ,sBAAAA,mBACf;AAaJ,SAAO,IANWK,2BAAAA,qBAAqB;GACrC,UAAA,IAPmBH,kBAAAA,SAAS;IAAE,QAAQ;IAAgB;IAAK;IAAS,CAO5D;GACR,gBAAA,IAPyBA,kBAAAA,SAAS;IAClC,QAAQ;IACR,KAAK,aAAa;IAClB;IACD,CAGe;GACd,sBAAsB;GACtB;GACD,CACW;;AAEd,OAAM,IAAI,MAAM,kBAAkB,OAAO,OAAO"}

package/dist/chains/summarization/load.js.map

@@ -1 +1 @@
-{"version":3,"file":"load.js","names":[],"sources":["../../../src/chains/summarization/load.ts"],"sourcesContent":["import type { BaseLanguageModelInterface } from \"@langchain/core/language_models/base\";\nimport { BasePromptTemplate } from \"@langchain/core/prompts\";\nimport { LLMChain } from \"../llm_chain.js\";\nimport {\n  StuffDocumentsChain,\n  MapReduceDocumentsChain,\n  RefineDocumentsChain,\n  MapReduceDocumentsChainInput,\n} from \"../combine_docs_chain.js\";\nimport { DEFAULT_PROMPT } from \"./stuff_prompts.js\";\nimport { REFINE_PROMPT } from \"./refine_prompts.js\";\n\n/**\n * Type for the base parameters that can be used to configure a\n * summarization chain.\n */\ntype BaseParams = {\n  verbose?: boolean;\n};\n\n/** @interface */\nexport type SummarizationChainParams = BaseParams &\n  (\n    | {\n        type?: \"stuff\";\n        prompt?: BasePromptTemplate;\n      }\n    | ({\n        type?: \"map_reduce\";\n        combineMapPrompt?: BasePromptTemplate;\n        combinePrompt?: BasePromptTemplate;\n        combineLLM?: BaseLanguageModelInterface;\n      } & Pick<MapReduceDocumentsChainInput, \"returnIntermediateSteps\">)\n    | {\n        type?: \"refine\";\n        refinePrompt?: BasePromptTemplate;\n        refineLLM?: BaseLanguageModelInterface;\n        questionPrompt?: BasePromptTemplate;\n      }\n  );\n\nexport const loadSummarizationChain = (\n  llm: BaseLanguageModelInterface,\n  params: SummarizationChainParams = { type: \"map_reduce\" }\n) => {\n  const { verbose } = params;\n  if (params.type === \"stuff\") {\n    const { prompt = DEFAULT_PROMPT } = params;\n    const llmChain = new LLMChain({ prompt, llm, verbose });\n    const chain = new StuffDocumentsChain({\n      llmChain,\n      documentVariableName: \"text\",\n      verbose,\n    });\n    return chain;\n  }\n  if (params.type === \"map_reduce\") {\n    const {\n      combineMapPrompt = DEFAULT_PROMPT,\n      combinePrompt = DEFAULT_PROMPT,\n      combineLLM,\n      returnIntermediateSteps,\n    } = params;\n    const llmChain = new LLMChain({ prompt: combineMapPrompt, llm, verbose });\n    const combineLLMChain = new LLMChain({\n      prompt: combinePrompt,\n      llm: combineLLM ?? llm,\n      verbose,\n    });\n    const combineDocumentChain = new StuffDocumentsChain({\n      llmChain: combineLLMChain,\n      documentVariableName: \"text\",\n      verbose,\n    });\n    const chain = new MapReduceDocumentsChain({\n      llmChain,\n      combineDocumentChain,\n      documentVariableName: \"text\",\n      returnIntermediateSteps,\n      verbose,\n    });\n    return chain;\n  }\n  if (params.type === \"refine\") {\n    const {\n      refinePrompt = REFINE_PROMPT,\n      refineLLM,\n      questionPrompt = DEFAULT_PROMPT,\n    } = params;\n    const llmChain = new LLMChain({ prompt: questionPrompt, llm, verbose });\n    const refineLLMChain = new LLMChain({\n      prompt: refinePrompt,\n      llm: refineLLM ?? llm,\n      verbose,\n    });\n    const chain = new RefineDocumentsChain({\n      llmChain,\n      refineLLMChain,\n      documentVariableName: \"text\",\n      verbose,\n    });\n    return chain;\n  }\n  throw new Error(`Invalid _type: ${params.type}`);\n};\n"],"mappings":";;;;;AAyCA,MAAa,0BACX,KACA,SAAmC,EAAE,MAAM,cAAc,KACtD;CACH,MAAM,EAAE,YAAY;AACpB,KAAI,OAAO,SAAS,SAAS;EAC3B,MAAM,EAAE,SAAS,mBAAmB;AAOpC,SALc,IAAI,oBAAoB;GACpC,UAFe,IAAI,SAAS;IAAE;IAAQ;IAAK;IAAS,CAAC;GAGrD,sBAAsB;GACtB;GACD,CAAC;;AAGJ,KAAI,OAAO,SAAS,cAAc;EAChC,MAAM,EACJ,mBAAmB,gBACnB,gBAAgB,gBAChB,YACA,4BACE;AAmBJ,SAPc,IAAI,wBAAwB;GACxC,UAZe,IAAI,SAAS;IAAE,QAAQ;IAAkB;IAAK;IAAS,CAAC;GAavE,sBAP2B,IAAI,oBAAoB;IACnD,UANsB,IAAI,SAAS;KACnC,QAAQ;KACR,KAAK,cAAc;KACnB;KACD,CAAC;IAGA,sBAAsB;IACtB;IACD,CAAC;GAIA,sBAAsB;GACtB;GACA;GACD,CAAC;;AAGJ,KAAI,OAAO,SAAS,UAAU;EAC5B,MAAM,EACJ,eAAe,eACf,WACA,iBAAiB,mBACf;AAaJ,SANc,IAAI,qBAAqB;GACrC,UAPe,IAAI,SAAS;IAAE,QAAQ;IAAgB;IAAK;IAAS,CAAC;GAQrE,gBAPqB,IAAI,SAAS;IAClC,QAAQ;IACR,KAAK,aAAa;IAClB;IACD,CAAC;GAIA,sBAAsB;GACtB;GACD,CAAC;;AAGJ,OAAM,IAAI,MAAM,kBAAkB,OAAO,OAAO"}
+{"version":3,"file":"load.js","names":[],"sources":["../../../src/chains/summarization/load.ts"],"sourcesContent":["import type { BaseLanguageModelInterface } from \"@langchain/core/language_models/base\";\nimport { BasePromptTemplate } from \"@langchain/core/prompts\";\nimport { LLMChain } from \"../llm_chain.js\";\nimport {\n  StuffDocumentsChain,\n  MapReduceDocumentsChain,\n  RefineDocumentsChain,\n  MapReduceDocumentsChainInput,\n} from \"../combine_docs_chain.js\";\nimport { DEFAULT_PROMPT } from \"./stuff_prompts.js\";\nimport { REFINE_PROMPT } from \"./refine_prompts.js\";\n\n/**\n * Type for the base parameters that can be used to configure a\n * summarization chain.\n */\ntype BaseParams = {\n  verbose?: boolean;\n};\n\n/** @interface */\nexport type SummarizationChainParams = BaseParams &\n  (\n    | {\n        type?: \"stuff\";\n        prompt?: BasePromptTemplate;\n      }\n    | ({\n        type?: \"map_reduce\";\n        combineMapPrompt?: BasePromptTemplate;\n        combinePrompt?: BasePromptTemplate;\n        combineLLM?: BaseLanguageModelInterface;\n      } & Pick<MapReduceDocumentsChainInput, \"returnIntermediateSteps\">)\n    | {\n        type?: \"refine\";\n        refinePrompt?: BasePromptTemplate;\n        refineLLM?: BaseLanguageModelInterface;\n        questionPrompt?: BasePromptTemplate;\n      }\n  );\n\nexport const loadSummarizationChain = (\n  llm: BaseLanguageModelInterface,\n  params: SummarizationChainParams = { type: \"map_reduce\" }\n) => {\n  const { verbose } = params;\n  if (params.type === \"stuff\") {\n    const { prompt = DEFAULT_PROMPT } = params;\n    const llmChain = new LLMChain({ prompt, llm, verbose });\n    const chain = new StuffDocumentsChain({\n      llmChain,\n      documentVariableName: \"text\",\n      verbose,\n    });\n    return chain;\n  }\n  if (params.type === \"map_reduce\") {\n    const {\n      combineMapPrompt = DEFAULT_PROMPT,\n      combinePrompt = DEFAULT_PROMPT,\n      combineLLM,\n      returnIntermediateSteps,\n    } = params;\n    const llmChain = new LLMChain({ prompt: combineMapPrompt, llm, verbose });\n    const combineLLMChain = new LLMChain({\n      prompt: combinePrompt,\n      llm: combineLLM ?? llm,\n      verbose,\n    });\n    const combineDocumentChain = new StuffDocumentsChain({\n      llmChain: combineLLMChain,\n      documentVariableName: \"text\",\n      verbose,\n    });\n    const chain = new MapReduceDocumentsChain({\n      llmChain,\n      combineDocumentChain,\n      documentVariableName: \"text\",\n      returnIntermediateSteps,\n      verbose,\n    });\n    return chain;\n  }\n  if (params.type === \"refine\") {\n    const {\n      refinePrompt = REFINE_PROMPT,\n      refineLLM,\n      questionPrompt = DEFAULT_PROMPT,\n    } = params;\n    const llmChain = new LLMChain({ prompt: questionPrompt, llm, verbose });\n    const refineLLMChain = new LLMChain({\n      prompt: refinePrompt,\n      llm: refineLLM ?? llm,\n      verbose,\n    });\n    const chain = new RefineDocumentsChain({\n      llmChain,\n      refineLLMChain,\n      documentVariableName: \"text\",\n      verbose,\n    });\n    return chain;\n  }\n  throw new Error(`Invalid _type: ${params.type}`);\n};\n"],"mappings":";;;;;AAyCA,MAAa,0BACX,KACA,SAAmC,EAAE,MAAM,cAAc,KACtD;CACH,MAAM,EAAE,YAAY;AACpB,KAAI,OAAO,SAAS,SAAS;EAC3B,MAAM,EAAE,SAAS,mBAAmB;AAOpC,SAAO,IALW,oBAAoB;GACpC,UAAA,IAFmB,SAAS;IAAE;IAAQ;IAAK;IAAS,CAE5C;GACR,sBAAsB;GACtB;GACD,CACW;;AAEd,KAAI,OAAO,SAAS,cAAc;EAChC,MAAM,EACJ,mBAAmB,gBACnB,gBAAgB,gBAChB,YACA,4BACE;AAmBJ,SAAO,IAPW,wBAAwB;GACxC,UAAA,IAZmB,SAAS;IAAE,QAAQ;IAAkB;IAAK;IAAS,CAY9D;GACR,sBAAA,IAP+B,oBAAoB;IACnD,UAAU,IANgB,SAAS;KACnC,QAAQ;KACR,KAAK,cAAc;KACnB;KACD,CAE0B;IACzB,sBAAsB;IACtB;IACD,CAGqB;GACpB,sBAAsB;GACtB;GACA;GACD,CACW;;AAEd,KAAI,OAAO,SAAS,UAAU;EAC5B,MAAM,EACJ,eAAe,eACf,WACA,iBAAiB,mBACf;AAaJ,SAAO,IANW,qBAAqB;GACrC,UAAA,IAPmB,SAAS;IAAE,QAAQ;IAAgB;IAAK;IAAS,CAO5D;GACR,gBAAA,IAPyB,SAAS;IAClC,QAAQ;IACR,KAAK,aAAa;IAClB;IACD,CAGe;GACd,sBAAsB;GACtB;GACD,CACW;;AAEd,OAAM,IAAI,MAAM,kBAAkB,OAAO,OAAO"}

package/dist/chains/summarization/refine_prompts.cjs.map

@@ -1 +1 @@
-{"version":3,"file":"refine_prompts.cjs","names":["PromptTemplate"],"sources":["../../../src/chains/summarization/refine_prompts.ts"],"sourcesContent":["import { PromptTemplate } from \"@langchain/core/prompts\";\n\nconst refinePromptTemplate = `Your job is to produce a final summary\nWe have provided an existing summary up to a certain point: \"{existing_answer}\"\nWe have the opportunity to refine the existing summary\n(only if needed) with some more context below.\n------------\n\"{text}\"\n------------\n\nGiven the new context, refine the original summary\nIf the context isn't useful, return the original summary.\n\nREFINED SUMMARY:`;\n\nexport const REFINE_PROMPT = /* #__PURE__ */ new PromptTemplate({\n  template: refinePromptTemplate,\n  inputVariables: [\"existing_answer\", \"text\"],\n});\n"],"mappings":";AAeA,MAAa,gCAAgC,yCAAIA,eAAe;CAC9D,UAd2B;;;;;;;;;;;;CAe3B,gBAAgB,CAAC,mBAAmB,OAAO;CAC5C,CAAC"}
+{"version":3,"file":"refine_prompts.cjs","names":["PromptTemplate"],"sources":["../../../src/chains/summarization/refine_prompts.ts"],"sourcesContent":["import { PromptTemplate } from \"@langchain/core/prompts\";\n\nconst refinePromptTemplate = `Your job is to produce a final summary\nWe have provided an existing summary up to a certain point: \"{existing_answer}\"\nWe have the opportunity to refine the existing summary\n(only if needed) with some more context below.\n------------\n\"{text}\"\n------------\n\nGiven the new context, refine the original summary\nIf the context isn't useful, return the original summary.\n\nREFINED SUMMARY:`;\n\nexport const REFINE_PROMPT = /* #__PURE__ */ new PromptTemplate({\n  template: refinePromptTemplate,\n  inputVariables: [\"existing_answer\", \"text\"],\n});\n"],"mappings":";AAeA,MAAa,gCAAgC,uCAAIA,EAAAA,eAAe;CAC9D,UAAU;;;;;;;;;;;;CACV,gBAAgB,CAAC,mBAAmB,OAAO;CAC5C,CAAC"}

package/dist/chains/summarization/refine_prompts.js.map

@@ -1 +1 @@
-{"version":3,"file":"refine_prompts.js","names":[],"sources":["../../../src/chains/summarization/refine_prompts.ts"],"sourcesContent":["import { PromptTemplate } from \"@langchain/core/prompts\";\n\nconst refinePromptTemplate = `Your job is to produce a final summary\nWe have provided an existing summary up to a certain point: \"{existing_answer}\"\nWe have the opportunity to refine the existing summary\n(only if needed) with some more context below.\n------------\n\"{text}\"\n------------\n\nGiven the new context, refine the original summary\nIf the context isn't useful, return the original summary.\n\nREFINED SUMMARY:`;\n\nexport const REFINE_PROMPT = /* #__PURE__ */ new PromptTemplate({\n  template: refinePromptTemplate,\n  inputVariables: [\"existing_answer\", \"text\"],\n});\n"],"mappings":";AAeA,MAAa,gCAAgC,IAAI,eAAe;CAC9D,UAd2B;;;;;;;;;;;;CAe3B,gBAAgB,CAAC,mBAAmB,OAAO;CAC5C,CAAC"}
+{"version":3,"file":"refine_prompts.js","names":[],"sources":["../../../src/chains/summarization/refine_prompts.ts"],"sourcesContent":["import { PromptTemplate } from \"@langchain/core/prompts\";\n\nconst refinePromptTemplate = `Your job is to produce a final summary\nWe have provided an existing summary up to a certain point: \"{existing_answer}\"\nWe have the opportunity to refine the existing summary\n(only if needed) with some more context below.\n------------\n\"{text}\"\n------------\n\nGiven the new context, refine the original summary\nIf the context isn't useful, return the original summary.\n\nREFINED SUMMARY:`;\n\nexport const REFINE_PROMPT = /* #__PURE__ */ new PromptTemplate({\n  template: refinePromptTemplate,\n  inputVariables: [\"existing_answer\", \"text\"],\n});\n"],"mappings":";AAeA,MAAa,gCAAgC,IAAI,eAAe;CAC9D,UAAU;;;;;;;;;;;;CACV,gBAAgB,CAAC,mBAAmB,OAAO;CAC5C,CAAC"}

package/dist/chains/summarization/stuff_prompts.cjs.map

@@ -1 +1 @@
-{"version":3,"file":"stuff_prompts.cjs","names":["PromptTemplate"],"sources":["../../../src/chains/summarization/stuff_prompts.ts"],"sourcesContent":["import { PromptTemplate } from \"@langchain/core/prompts\";\n\nconst template = `Write a concise summary of the following:\n\n\n\"{text}\"\n\n\nCONCISE SUMMARY:`;\n\nexport const DEFAULT_PROMPT = /*#__PURE__*/ new PromptTemplate({\n  template,\n  inputVariables: [\"text\"],\n});\n"],"mappings":";AAUA,MAAa,iCAA+B,yCAAIA,eAAe;CAC7D,UATe;;;;;;;CAUf,gBAAgB,CAAC,OAAO;CACzB,CAAC"}
+{"version":3,"file":"stuff_prompts.cjs","names":["PromptTemplate"],"sources":["../../../src/chains/summarization/stuff_prompts.ts"],"sourcesContent":["import { PromptTemplate } from \"@langchain/core/prompts\";\n\nconst template = `Write a concise summary of the following:\n\n\n\"{text}\"\n\n\nCONCISE SUMMARY:`;\n\nexport const DEFAULT_PROMPT = /*#__PURE__*/ new PromptTemplate({\n  template,\n  inputVariables: [\"text\"],\n});\n"],"mappings":";AAUA,MAAa,iCAA+B,uCAAIA,EAAAA,eAAe;CAC7D;;;;;;;CACA,gBAAgB,CAAC,OAAO;CACzB,CAAC"}

package/dist/chains/summarization/stuff_prompts.js.map

@@ -1 +1 @@
-{"version":3,"file":"stuff_prompts.js","names":[],"sources":["../../../src/chains/summarization/stuff_prompts.ts"],"sourcesContent":["import { PromptTemplate } from \"@langchain/core/prompts\";\n\nconst template = `Write a concise summary of the following:\n\n\n\"{text}\"\n\n\nCONCISE SUMMARY:`;\n\nexport const DEFAULT_PROMPT = /*#__PURE__*/ new PromptTemplate({\n  template,\n  inputVariables: [\"text\"],\n});\n"],"mappings":";AAUA,MAAa,iCAA+B,IAAI,eAAe;CAC7D,UATe;;;;;;;CAUf,gBAAgB,CAAC,OAAO;CACzB,CAAC"}
+{"version":3,"file":"stuff_prompts.js","names":[],"sources":["../../../src/chains/summarization/stuff_prompts.ts"],"sourcesContent":["import { PromptTemplate } from \"@langchain/core/prompts\";\n\nconst template = `Write a concise summary of the following:\n\n\n\"{text}\"\n\n\nCONCISE SUMMARY:`;\n\nexport const DEFAULT_PROMPT = /*#__PURE__*/ new PromptTemplate({\n  template,\n  inputVariables: [\"text\"],\n});\n"],"mappings":";AAUA,MAAa,iCAA+B,IAAI,eAAe;CAC7D;;;;;;;CACA,gBAAgB,CAAC,OAAO;CACzB,CAAC"}