@langchain/classic 1.0.0-alpha.1

package/dist/retrievers/document_compressors/chain_extract.js

@@ -0,0 +1,97 @@
+import { __export } from "../../_virtual/rolldown_runtime.js";
+import { LLMChain } from "../../chains/llm_chain.js";
+import { BaseDocumentCompressor } from "./index.js";
+import { PROMPT_TEMPLATE } from "./chain_extract_prompt.js";
+import { Document } from "@langchain/core/documents";
+import { PromptTemplate } from "@langchain/core/prompts";
+import { BaseOutputParser } from "@langchain/core/output_parsers";
+
+//#region src/retrievers/document_compressors/chain_extract.ts
+var chain_extract_exports = {};
+__export(chain_extract_exports, { LLMChainExtractor: () => LLMChainExtractor });
+function defaultGetInput(query, doc) {
+	return {
+		question: query,
+		context: doc.pageContent
+	};
+}
+var NoOutputParser = class extends BaseOutputParser {
+	lc_namespace = [
+		"langchain",
+		"retrievers",
+		"document_compressors",
+		"chain_extract"
+	];
+	noOutputStr = "NO_OUTPUT";
+	parse(text) {
+		const cleanedText = text.trim();
+		if (cleanedText === this.noOutputStr) return Promise.resolve("");
+		return Promise.resolve(cleanedText);
+	}
+	getFormatInstructions() {
+		throw new Error("Method not implemented.");
+	}
+};
+function getDefaultChainPrompt() {
+	const outputParser = new NoOutputParser();
+	const template = PROMPT_TEMPLATE(outputParser.noOutputStr);
+	return new PromptTemplate({
+		template,
+		inputVariables: ["question", "context"],
+		outputParser
+	});
+}
+/**
+* A class that uses an LLM chain to extract relevant parts of documents.
+* It extends the BaseDocumentCompressor class.
+*/
+var LLMChainExtractor = class LLMChainExtractor extends BaseDocumentCompressor {
+	llmChain;
+	getInput = defaultGetInput;
+	constructor({ llmChain, getInput }) {
+		super();
+		this.llmChain = llmChain;
+		this.getInput = getInput;
+	}
+	/**
+	* Compresses a list of documents based on the output of an LLM chain.
+	* @param documents The list of documents to be compressed.
+	* @param query The query to be used for document compression.
+	* @returns A list of compressed documents.
+	*/
+	async compressDocuments(documents, query) {
+		const compressedDocs = await Promise.all(documents.map(async (doc) => {
+			const input = this.getInput(query, doc);
+			const output = await this.llmChain.predict(input);
+			return output.length > 0 ? new Document({
+				pageContent: output,
+				metadata: doc.metadata
+			}) : void 0;
+		}));
+		return compressedDocs.filter((doc) => doc !== void 0);
+	}
+	/**
+	* Creates a new instance of LLMChainExtractor from a given LLM, prompt
+	* template, and getInput function.
+	* @param llm The BaseLanguageModel instance used for document extraction.
+	* @param prompt The PromptTemplate instance used for document extraction.
+	* @param getInput A function used for constructing the chain input from the query and a Document.
+	* @returns A new instance of LLMChainExtractor.
+	*/
+	static fromLLM(llm, prompt, getInput) {
+		const _prompt = prompt || getDefaultChainPrompt();
+		const _getInput = getInput || defaultGetInput;
+		const llmChain = new LLMChain({
+			llm,
+			prompt: _prompt
+		});
+		return new LLMChainExtractor({
+			llmChain,
+			getInput: _getInput
+		});
+	}
+};
+
+//#endregion
+export { LLMChainExtractor, chain_extract_exports };
+//# sourceMappingURL=chain_extract.js.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"chain_extract.js","names":["query: string","doc: DocumentInterface","text: string","documents: DocumentInterface[]","llm: BaseLanguageModelInterface","prompt?: PromptTemplate","getInput?: (\n      query: string,\n      doc: DocumentInterface\n    ) => Record<string, unknown>"],"sources":["../../../src/retrievers/document_compressors/chain_extract.ts"],"sourcesContent":["import type { BaseLanguageModelInterface } from \"@langchain/core/language_models/base\";\nimport { type DocumentInterface, Document } from \"@langchain/core/documents\";\nimport { PromptTemplate } from \"@langchain/core/prompts\";\nimport { BaseOutputParser } from \"@langchain/core/output_parsers\";\nimport { LLMChain } from \"../../chains/llm_chain.js\";\nimport { BaseDocumentCompressor } from \"./index.js\";\nimport { PROMPT_TEMPLATE } from \"./chain_extract_prompt.js\";\n\nfunction defaultGetInput(\n  query: string,\n  doc: DocumentInterface\n): Record<string, unknown> {\n  return { question: query, context: doc.pageContent };\n}\n\nclass NoOutputParser extends BaseOutputParser<string> {\n  lc_namespace = [\n    \"langchain\",\n    \"retrievers\",\n    \"document_compressors\",\n    \"chain_extract\",\n  ];\n\n  noOutputStr = \"NO_OUTPUT\";\n\n  parse(text: string): Promise<string> {\n    const cleanedText = text.trim();\n    if (cleanedText === this.noOutputStr) {\n      return Promise.resolve(\"\");\n    }\n    return Promise.resolve(cleanedText);\n  }\n\n  getFormatInstructions(): string {\n    throw new Error(\"Method not implemented.\");\n  }\n}\n\nfunction getDefaultChainPrompt(): PromptTemplate {\n  const outputParser = new NoOutputParser();\n  const template = PROMPT_TEMPLATE(outputParser.noOutputStr);\n  return new PromptTemplate({\n    template,\n    inputVariables: [\"question\", \"context\"],\n    outputParser,\n  });\n}\n\n/**\n * Interface for the arguments required to create an instance of\n * LLMChainExtractor.\n */\nexport interface LLMChainExtractorArgs {\n  llmChain: LLMChain;\n  getInput: (query: string, doc: DocumentInterface) => Record<string, unknown>;\n}\n\n/**\n * A class that uses an LLM chain to extract relevant parts of documents.\n * It extends the BaseDocumentCompressor class.\n */\nexport class LLMChainExtractor extends BaseDocumentCompressor {\n  llmChain: LLMChain;\n\n  getInput: (query: string, doc: DocumentInterface) => Record<string, unknown> =\n    defaultGetInput;\n\n  constructor({ llmChain, getInput }: LLMChainExtractorArgs) {\n    super();\n    this.llmChain = llmChain;\n    this.getInput = getInput;\n  }\n\n  /**\n   * Compresses a list of documents based on the output of an LLM chain.\n   * @param documents The list of documents to be compressed.\n   * @param query The query to be used for document compression.\n   * @returns A list of compressed documents.\n   */\n  async compressDocuments(\n    documents: DocumentInterface[],\n    query: string\n  ): Promise<DocumentInterface[]> {\n    const compressedDocs = await Promise.all(\n      documents.map(async (doc) => {\n        const input = this.getInput(query, doc);\n        const output = await this.llmChain.predict(input);\n        return output.length > 0\n          ? new Document({\n              pageContent: output,\n              metadata: doc.metadata,\n            })\n          : undefined;\n      })\n    );\n    return compressedDocs.filter((doc): doc is Document => doc !== undefined);\n  }\n\n  /**\n   * Creates a new instance of LLMChainExtractor from a given LLM, prompt\n   * template, and getInput function.\n   * @param llm The BaseLanguageModel instance used for document extraction.\n   * @param prompt The PromptTemplate instance used for document extraction.\n   * @param getInput A function used for constructing the chain input from the query and a Document.\n   * @returns A new instance of LLMChainExtractor.\n   */\n  static fromLLM(\n    llm: BaseLanguageModelInterface,\n    prompt?: PromptTemplate,\n    getInput?: (\n      query: string,\n      doc: DocumentInterface\n    ) => Record<string, unknown>\n  ): LLMChainExtractor {\n    const _prompt = prompt || getDefaultChainPrompt();\n    const _getInput = getInput || defaultGetInput;\n    const llmChain = new LLMChain({ llm, prompt: _prompt });\n    return new LLMChainExtractor({ llmChain, getInput: _getInput });\n  }\n}\n"],"mappings":";;;;;;;;;;;AAQA,SAAS,gBACPA,OACAC,KACyB;AACzB,QAAO;EAAE,UAAU;EAAO,SAAS,IAAI;CAAa;AACrD;AAED,IAAM,iBAAN,cAA6B,iBAAyB;CACpD,eAAe;EACb;EACA;EACA;EACA;CACD;CAED,cAAc;CAEd,MAAMC,MAA+B;EACnC,MAAM,cAAc,KAAK,MAAM;AAC/B,MAAI,gBAAgB,KAAK,YACvB,QAAO,QAAQ,QAAQ,GAAG;AAE5B,SAAO,QAAQ,QAAQ,YAAY;CACpC;CAED,wBAAgC;AAC9B,QAAM,IAAI,MAAM;CACjB;AACF;AAED,SAAS,wBAAwC;CAC/C,MAAM,eAAe,IAAI;CACzB,MAAM,WAAW,gBAAgB,aAAa,YAAY;AAC1D,QAAO,IAAI,eAAe;EACxB;EACA,gBAAgB,CAAC,YAAY,SAAU;EACvC;CACD;AACF;;;;;AAeD,IAAa,oBAAb,MAAa,0BAA0B,uBAAuB;CAC5D;CAEA,WACE;CAEF,YAAY,EAAE,UAAU,UAAiC,EAAE;EACzD,OAAO;EACP,KAAK,WAAW;EAChB,KAAK,WAAW;CACjB;;;;;;;CAQD,MAAM,kBACJC,WACAH,OAC8B;EAC9B,MAAM,iBAAiB,MAAM,QAAQ,IACnC,UAAU,IAAI,OAAO,QAAQ;GAC3B,MAAM,QAAQ,KAAK,SAAS,OAAO,IAAI;GACvC,MAAM,SAAS,MAAM,KAAK,SAAS,QAAQ,MAAM;AACjD,UAAO,OAAO,SAAS,IACnB,IAAI,SAAS;IACX,aAAa;IACb,UAAU,IAAI;GACf,KACD;EACL,EAAC,CACH;AACD,SAAO,eAAe,OAAO,CAAC,QAAyB,QAAQ,OAAU;CAC1E;;;;;;;;;CAUD,OAAO,QACLI,KACAC,QACAC,UAImB;EACnB,MAAM,UAAU,UAAU,uBAAuB;EACjD,MAAM,YAAY,YAAY;EAC9B,MAAM,WAAW,IAAI,SAAS;GAAE;GAAK,QAAQ;EAAS;AACtD,SAAO,IAAI,kBAAkB;GAAE;GAAU,UAAU;EAAW;CAC/D;AACF"}

package/dist/retrievers/document_compressors/chain_extract_prompt.cjs

@@ -0,0 +1,16 @@
+
+//#region src/retrievers/document_compressors/chain_extract_prompt.ts
+const PROMPT_TEMPLATE = (noOutputStr) => `Given the following question and context, extract any part of the context *AS IS* that is relevant to answer the question. If none of the context is relevant return ${noOutputStr}.
+
+Remember, *DO NOT* edit the extracted parts of the context.
+
+> Question: {question}
+> Context:
+>>>
+{context}
+>>>
+Extracted relevant parts:`;
+
+//#endregion
+exports.PROMPT_TEMPLATE = PROMPT_TEMPLATE;
+//# sourceMappingURL=chain_extract_prompt.cjs.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"chain_extract_prompt.cjs","names":["noOutputStr: string"],"sources":["../../../src/retrievers/document_compressors/chain_extract_prompt.ts"],"sourcesContent":["export const PROMPT_TEMPLATE = (\n  noOutputStr: string\n) => `Given the following question and context, extract any part of the context *AS IS* that is relevant to answer the question. If none of the context is relevant return ${noOutputStr}.\n\nRemember, *DO NOT* edit the extracted parts of the context.\n\n> Question: {question}\n> Context:\n>>>\n{context}\n>>>\nExtracted relevant parts:`;\n"],"mappings":";;AAAA,MAAa,kBAAkB,CAC7BA,gBACG,CAAC,qKAAqK,EAAE,YAAY;;;;;;;;;yBAShK,CAAC"}

package/dist/retrievers/document_compressors/chain_extract_prompt.js

@@ -0,0 +1,15 @@
+//#region src/retrievers/document_compressors/chain_extract_prompt.ts
+const PROMPT_TEMPLATE = (noOutputStr) => `Given the following question and context, extract any part of the context *AS IS* that is relevant to answer the question. If none of the context is relevant return ${noOutputStr}.
+
+Remember, *DO NOT* edit the extracted parts of the context.
+
+> Question: {question}
+> Context:
+>>>
+{context}
+>>>
+Extracted relevant parts:`;
+
+//#endregion
+export { PROMPT_TEMPLATE };
+//# sourceMappingURL=chain_extract_prompt.js.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"chain_extract_prompt.js","names":["noOutputStr: string"],"sources":["../../../src/retrievers/document_compressors/chain_extract_prompt.ts"],"sourcesContent":["export const PROMPT_TEMPLATE = (\n  noOutputStr: string\n) => `Given the following question and context, extract any part of the context *AS IS* that is relevant to answer the question. If none of the context is relevant return ${noOutputStr}.\n\nRemember, *DO NOT* edit the extracted parts of the context.\n\n> Question: {question}\n> Context:\n>>>\n{context}\n>>>\nExtracted relevant parts:`;\n"],"mappings":";AAAA,MAAa,kBAAkB,CAC7BA,gBACG,CAAC,qKAAqK,EAAE,YAAY;;;;;;;;;yBAShK,CAAC"}

package/dist/retrievers/document_compressors/embeddings_filter.cjs

@@ -0,0 +1,74 @@
+const require_rolldown_runtime = require('../../_virtual/rolldown_runtime.cjs');
+const require_retrievers_document_compressors_index = require('./index.cjs');
+const __langchain_core_utils_math = require_rolldown_runtime.__toESM(require("@langchain/core/utils/math"));
+
+//#region src/retrievers/document_compressors/embeddings_filter.ts
+var embeddings_filter_exports = {};
+require_rolldown_runtime.__export(embeddings_filter_exports, { EmbeddingsFilter: () => EmbeddingsFilter });
+/**
+* Class that represents a document compressor that uses embeddings to
+* drop documents unrelated to the query.
+* @example
+* ```typescript
+* const embeddingsFilter = new EmbeddingsFilter({
+*   embeddings: new OpenAIEmbeddings(),
+*   similarityThreshold: 0.8,
+*   k: 5,
+* });
+* const retrievedDocs = await embeddingsFilter.filterDocuments(
+*   getDocuments(),
+*   "What did the speaker say about Justice Breyer in the 2022 State of the Union?",
+* );
+* console.log({ retrievedDocs });
+* ```
+*/
+var EmbeddingsFilter = class extends require_retrievers_document_compressors_index.BaseDocumentCompressor {
+	/**
+	* Embeddings to use for embedding document contents and queries.
+	*/
+	embeddings;
+	/**
+	* Similarity function for comparing documents.
+	*/
+	similarityFn = __langchain_core_utils_math.cosineSimilarity;
+	/**
+	* Threshold for determining when two documents are similar enough
+	* to be considered redundant. Must be specified if `k` is not set.
+	*/
+	similarityThreshold;
+	/**
+	* The number of relevant documents to return. Can be explicitly set to undefined, in which case
+	* similarity_threshold` must be specified. Defaults to 20
+	*/
+	k = 20;
+	constructor(params) {
+		super();
+		this.embeddings = params.embeddings;
+		this.similarityFn = params.similarityFn ?? this.similarityFn;
+		this.similarityThreshold = params.similarityThreshold;
+		this.k = "k" in params ? params.k : this.k;
+		if (this.k === void 0 && this.similarityThreshold === void 0) throw new Error(`Must specify one of "k" or "similarity_threshold".`);
+	}
+	async compressDocuments(documents, query) {
+		const embeddedDocuments = await this.embeddings.embedDocuments(documents.map((doc) => doc.pageContent));
+		const embeddedQuery = await this.embeddings.embedQuery(query);
+		const similarity = this.similarityFn([embeddedQuery], embeddedDocuments)[0];
+		let includedIdxs = Array.from({ length: embeddedDocuments.length }, (_, i) => i);
+		if (this.k !== void 0) includedIdxs = includedIdxs.map((v, i) => [similarity[i], v]).sort(([a], [b]) => b - a).slice(0, this.k).map(([, i]) => i);
+		if (this.similarityThreshold !== void 0) {
+			const threshold = this.similarityThreshold;
+			includedIdxs = includedIdxs.filter((i) => similarity[i] > threshold);
+		}
+		return includedIdxs.map((i) => documents[i]);
+	}
+};
+
+//#endregion
+exports.EmbeddingsFilter = EmbeddingsFilter;
+Object.defineProperty(exports, 'embeddings_filter_exports', {
+  enumerable: true,
+  get: function () {
+    return embeddings_filter_exports;
+  }
+});
+//# sourceMappingURL=embeddings_filter.cjs.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"embeddings_filter.cjs","names":["BaseDocumentCompressor","cosineSimilarity","params: EmbeddingsFilterParams","documents: DocumentInterface[]","query: string"],"sources":["../../../src/retrievers/document_compressors/embeddings_filter.ts"],"sourcesContent":["import type { EmbeddingsInterface } from \"@langchain/core/embeddings\";\nimport type { DocumentInterface } from \"@langchain/core/documents\";\nimport { cosineSimilarity } from \"@langchain/core/utils/math\";\nimport { BaseDocumentCompressor } from \"./index.js\";\n\n/**\n * Interface for the parameters of the `EmbeddingsFilter` class.\n */\nexport interface EmbeddingsFilterParams {\n  embeddings: EmbeddingsInterface;\n  similarityFn?: (x: number[][], y: number[][]) => number[][];\n  similarityThreshold?: number;\n  k?: number;\n}\n\n/**\n * Class that represents a document compressor that uses embeddings to\n * drop documents unrelated to the query.\n * @example\n * ```typescript\n * const embeddingsFilter = new EmbeddingsFilter({\n *   embeddings: new OpenAIEmbeddings(),\n *   similarityThreshold: 0.8,\n *   k: 5,\n * });\n * const retrievedDocs = await embeddingsFilter.filterDocuments(\n *   getDocuments(),\n *   \"What did the speaker say about Justice Breyer in the 2022 State of the Union?\",\n * );\n * console.log({ retrievedDocs });\n * ```\n */\nexport class EmbeddingsFilter extends BaseDocumentCompressor {\n  /**\n   * Embeddings to use for embedding document contents and queries.\n   */\n  embeddings: EmbeddingsInterface;\n\n  /**\n   * Similarity function for comparing documents.\n   */\n  similarityFn = cosineSimilarity;\n\n  /**\n   * Threshold for determining when two documents are similar enough\n   * to be considered redundant. Must be specified if `k` is not set.\n   */\n  similarityThreshold?: number;\n\n  /**\n   * The number of relevant documents to return. Can be explicitly set to undefined, in which case\n   * similarity_threshold` must be specified. Defaults to 20\n   */\n  k? = 20;\n\n  constructor(params: EmbeddingsFilterParams) {\n    super();\n    this.embeddings = params.embeddings;\n    this.similarityFn = params.similarityFn ?? this.similarityFn;\n    this.similarityThreshold = params.similarityThreshold;\n    this.k = \"k\" in params ? params.k : this.k;\n    if (this.k === undefined && this.similarityThreshold === undefined) {\n      throw new Error(`Must specify one of \"k\" or \"similarity_threshold\".`);\n    }\n  }\n\n  async compressDocuments(\n    documents: DocumentInterface[],\n    query: string\n  ): Promise<DocumentInterface[]> {\n    const embeddedDocuments = await this.embeddings.embedDocuments(\n      documents.map((doc) => doc.pageContent)\n    );\n    const embeddedQuery = await this.embeddings.embedQuery(query);\n    const similarity = this.similarityFn([embeddedQuery], embeddedDocuments)[0];\n    let includedIdxs = Array.from(\n      { length: embeddedDocuments.length },\n      (_, i) => i\n    );\n\n    if (this.k !== undefined) {\n      includedIdxs = includedIdxs\n        .map((v, i) => [similarity[i], v])\n        .sort(([a], [b]) => b - a)\n        .slice(0, this.k)\n        .map(([, i]) => i);\n    }\n\n    if (this.similarityThreshold !== undefined) {\n      const threshold = this.similarityThreshold;\n      includedIdxs = includedIdxs.filter((i) => similarity[i] > threshold);\n    }\n\n    return includedIdxs.map((i) => documents[i]);\n  }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;AAgCA,IAAa,mBAAb,cAAsCA,qEAAuB;;;;CAI3D;;;;CAKA,eAAeC;;;;;CAMf;;;;;CAMA,IAAK;CAEL,YAAYC,QAAgC;EAC1C,OAAO;EACP,KAAK,aAAa,OAAO;EACzB,KAAK,eAAe,OAAO,gBAAgB,KAAK;EAChD,KAAK,sBAAsB,OAAO;EAClC,KAAK,IAAI,OAAO,SAAS,OAAO,IAAI,KAAK;AACzC,MAAI,KAAK,MAAM,UAAa,KAAK,wBAAwB,OACvD,OAAM,IAAI,MAAM,CAAC,kDAAkD,CAAC;CAEvE;CAED,MAAM,kBACJC,WACAC,OAC8B;EAC9B,MAAM,oBAAoB,MAAM,KAAK,WAAW,eAC9C,UAAU,IAAI,CAAC,QAAQ,IAAI,YAAY,CACxC;EACD,MAAM,gBAAgB,MAAM,KAAK,WAAW,WAAW,MAAM;EAC7D,MAAM,aAAa,KAAK,aAAa,CAAC,aAAc,GAAE,kBAAkB,CAAC;EACzE,IAAI,eAAe,MAAM,KACvB,EAAE,QAAQ,kBAAkB,OAAQ,GACpC,CAAC,GAAG,MAAM,EACX;AAED,MAAI,KAAK,MAAM,QACb,eAAe,aACZ,IAAI,CAAC,GAAG,MAAM,CAAC,WAAW,IAAI,CAAE,EAAC,CACjC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,EAAE,KAAK,IAAI,EAAE,CACzB,MAAM,GAAG,KAAK,EAAE,CAChB,IAAI,CAAC,GAAG,EAAE,KAAK,EAAE;AAGtB,MAAI,KAAK,wBAAwB,QAAW;GAC1C,MAAM,YAAY,KAAK;GACvB,eAAe,aAAa,OAAO,CAAC,MAAM,WAAW,KAAK,UAAU;EACrE;AAED,SAAO,aAAa,IAAI,CAAC,MAAM,UAAU,GAAG;CAC7C;AACF"}

package/dist/retrievers/document_compressors/embeddings_filter.d.cts

@@ -0,0 +1,58 @@
+import { BaseDocumentCompressor } from "./index.cjs";
+import { DocumentInterface } from "@langchain/core/documents";
+import { EmbeddingsInterface } from "@langchain/core/embeddings";
+import { cosineSimilarity } from "@langchain/core/utils/math";
+
+//#region src/retrievers/document_compressors/embeddings_filter.d.ts
+
+/**
+ * Interface for the parameters of the `EmbeddingsFilter` class.
+ */
+interface EmbeddingsFilterParams {
+  embeddings: EmbeddingsInterface;
+  similarityFn?: (x: number[][], y: number[][]) => number[][];
+  similarityThreshold?: number;
+  k?: number;
+}
+/**
+ * Class that represents a document compressor that uses embeddings to
+ * drop documents unrelated to the query.
+ * @example
+ * ```typescript
+ * const embeddingsFilter = new EmbeddingsFilter({
+ *   embeddings: new OpenAIEmbeddings(),
+ *   similarityThreshold: 0.8,
+ *   k: 5,
+ * });
+ * const retrievedDocs = await embeddingsFilter.filterDocuments(
+ *   getDocuments(),
+ *   "What did the speaker say about Justice Breyer in the 2022 State of the Union?",
+ * );
+ * console.log({ retrievedDocs });
+ * ```
+ */
+declare class EmbeddingsFilter extends BaseDocumentCompressor {
+  /**
+   * Embeddings to use for embedding document contents and queries.
+   */
+  embeddings: EmbeddingsInterface;
+  /**
+   * Similarity function for comparing documents.
+   */
+  similarityFn: typeof cosineSimilarity;
+  /**
+   * Threshold for determining when two documents are similar enough
+   * to be considered redundant. Must be specified if `k` is not set.
+   */
+  similarityThreshold?: number;
+  /**
+   * The number of relevant documents to return. Can be explicitly set to undefined, in which case
+   * similarity_threshold` must be specified. Defaults to 20
+   */
+  k?: number | undefined;
+  constructor(params: EmbeddingsFilterParams);
+  compressDocuments(documents: DocumentInterface[], query: string): Promise<DocumentInterface[]>;
+}
+//#endregion
+export { EmbeddingsFilter, EmbeddingsFilterParams };
+//# sourceMappingURL=embeddings_filter.d.cts.map

package/dist/retrievers/document_compressors/embeddings_filter.d.cts.map

@@ -0,0 +1 @@
+{"version":3,"file":"embeddings_filter.d.cts","names":["EmbeddingsInterface","DocumentInterface","cosineSimilarity","BaseDocumentCompressor","EmbeddingsFilterParams","EmbeddingsFilter","Promise"],"sources":["../../../src/retrievers/document_compressors/embeddings_filter.d.ts"],"sourcesContent":["import type { EmbeddingsInterface } from \"@langchain/core/embeddings\";\nimport type { DocumentInterface } from \"@langchain/core/documents\";\nimport { cosineSimilarity } from \"@langchain/core/utils/math\";\nimport { BaseDocumentCompressor } from \"./index.js\";\n/**\n * Interface for the parameters of the `EmbeddingsFilter` class.\n */\nexport interface EmbeddingsFilterParams {\n    embeddings: EmbeddingsInterface;\n    similarityFn?: (x: number[][], y: number[][]) => number[][];\n    similarityThreshold?: number;\n    k?: number;\n}\n/**\n * Class that represents a document compressor that uses embeddings to\n * drop documents unrelated to the query.\n * @example\n * ```typescript\n * const embeddingsFilter = new EmbeddingsFilter({\n *   embeddings: new OpenAIEmbeddings(),\n *   similarityThreshold: 0.8,\n *   k: 5,\n * });\n * const retrievedDocs = await embeddingsFilter.filterDocuments(\n *   getDocuments(),\n *   \"What did the speaker say about Justice Breyer in the 2022 State of the Union?\",\n * );\n * console.log({ retrievedDocs });\n * ```\n */\nexport declare class EmbeddingsFilter extends BaseDocumentCompressor {\n    /**\n     * Embeddings to use for embedding document contents and queries.\n     */\n    embeddings: EmbeddingsInterface;\n    /**\n     * Similarity function for comparing documents.\n     */\n    similarityFn: typeof cosineSimilarity;\n    /**\n     * Threshold for determining when two documents are similar enough\n     * to be considered redundant. Must be specified if `k` is not set.\n     */\n    similarityThreshold?: number;\n    /**\n     * The number of relevant documents to return. Can be explicitly set to undefined, in which case\n     * similarity_threshold` must be specified. Defaults to 20\n     */\n    k?: number | undefined;\n    constructor(params: EmbeddingsFilterParams);\n    compressDocuments(documents: DocumentInterface[], query: string): Promise<DocumentInterface[]>;\n}\n"],"mappings":";;;;;;;;;AAOA;AAuBqBK,UAvBJD,sBAAAA,CAuBoB;EAAA,UAAA,EAtBrBJ,mBAsBqB;EAAA,YAIrBA,CAAAA,EAAAA,CAAAA,CAAAA,EAAAA,MAAAA,EAAAA,EAAAA,EAAAA,CAAAA,EAAAA,MAAAA,EAAAA,EAAAA,EAAAA,GAAAA,MAAAA,EAAAA,EAAAA;EAAmB,mBAIVE,CAAAA,EAAAA,MAAAA;EAAgB,CAAA,CAAA,EAWjBE,MAAAA;;;;;AAnB4C;;;;;;;;;;;;;;cAA/CC,gBAAAA,SAAyBF,sBAAAA;;;;cAI9BH;;;;uBAISE;;;;;;;;;;;sBAWDE;+BACSH,qCAAqCK,QAAQL"}

package/dist/retrievers/document_compressors/embeddings_filter.d.ts

@@ -0,0 +1,58 @@
+import { BaseDocumentCompressor } from "./index.js";
+import { DocumentInterface } from "@langchain/core/documents";
+import { EmbeddingsInterface } from "@langchain/core/embeddings";
+import { cosineSimilarity } from "@langchain/core/utils/math";
+
+//#region src/retrievers/document_compressors/embeddings_filter.d.ts
+
+/**
+ * Interface for the parameters of the `EmbeddingsFilter` class.
+ */
+interface EmbeddingsFilterParams {
+  embeddings: EmbeddingsInterface;
+  similarityFn?: (x: number[][], y: number[][]) => number[][];
+  similarityThreshold?: number;
+  k?: number;
+}
+/**
+ * Class that represents a document compressor that uses embeddings to
+ * drop documents unrelated to the query.
+ * @example
+ * ```typescript
+ * const embeddingsFilter = new EmbeddingsFilter({
+ *   embeddings: new OpenAIEmbeddings(),
+ *   similarityThreshold: 0.8,
+ *   k: 5,
+ * });
+ * const retrievedDocs = await embeddingsFilter.filterDocuments(
+ *   getDocuments(),
+ *   "What did the speaker say about Justice Breyer in the 2022 State of the Union?",
+ * );
+ * console.log({ retrievedDocs });
+ * ```
+ */
+declare class EmbeddingsFilter extends BaseDocumentCompressor {
+  /**
+   * Embeddings to use for embedding document contents and queries.
+   */
+  embeddings: EmbeddingsInterface;
+  /**
+   * Similarity function for comparing documents.
+   */
+  similarityFn: typeof cosineSimilarity;
+  /**
+   * Threshold for determining when two documents are similar enough
+   * to be considered redundant. Must be specified if `k` is not set.
+   */
+  similarityThreshold?: number;
+  /**
+   * The number of relevant documents to return. Can be explicitly set to undefined, in which case
+   * similarity_threshold` must be specified. Defaults to 20
+   */
+  k?: number | undefined;
+  constructor(params: EmbeddingsFilterParams);
+  compressDocuments(documents: DocumentInterface[], query: string): Promise<DocumentInterface[]>;
+}
+//#endregion
+export { EmbeddingsFilter, EmbeddingsFilterParams };
+//# sourceMappingURL=embeddings_filter.d.ts.map

package/dist/retrievers/document_compressors/embeddings_filter.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"embeddings_filter.d.ts","names":["EmbeddingsInterface","DocumentInterface","cosineSimilarity","BaseDocumentCompressor","EmbeddingsFilterParams","EmbeddingsFilter","Promise"],"sources":["../../../src/retrievers/document_compressors/embeddings_filter.d.ts"],"sourcesContent":["import type { EmbeddingsInterface } from \"@langchain/core/embeddings\";\nimport type { DocumentInterface } from \"@langchain/core/documents\";\nimport { cosineSimilarity } from \"@langchain/core/utils/math\";\nimport { BaseDocumentCompressor } from \"./index.js\";\n/**\n * Interface for the parameters of the `EmbeddingsFilter` class.\n */\nexport interface EmbeddingsFilterParams {\n    embeddings: EmbeddingsInterface;\n    similarityFn?: (x: number[][], y: number[][]) => number[][];\n    similarityThreshold?: number;\n    k?: number;\n}\n/**\n * Class that represents a document compressor that uses embeddings to\n * drop documents unrelated to the query.\n * @example\n * ```typescript\n * const embeddingsFilter = new EmbeddingsFilter({\n *   embeddings: new OpenAIEmbeddings(),\n *   similarityThreshold: 0.8,\n *   k: 5,\n * });\n * const retrievedDocs = await embeddingsFilter.filterDocuments(\n *   getDocuments(),\n *   \"What did the speaker say about Justice Breyer in the 2022 State of the Union?\",\n * );\n * console.log({ retrievedDocs });\n * ```\n */\nexport declare class EmbeddingsFilter extends BaseDocumentCompressor {\n    /**\n     * Embeddings to use for embedding document contents and queries.\n     */\n    embeddings: EmbeddingsInterface;\n    /**\n     * Similarity function for comparing documents.\n     */\n    similarityFn: typeof cosineSimilarity;\n    /**\n     * Threshold for determining when two documents are similar enough\n     * to be considered redundant. Must be specified if `k` is not set.\n     */\n    similarityThreshold?: number;\n    /**\n     * The number of relevant documents to return. Can be explicitly set to undefined, in which case\n     * similarity_threshold` must be specified. Defaults to 20\n     */\n    k?: number | undefined;\n    constructor(params: EmbeddingsFilterParams);\n    compressDocuments(documents: DocumentInterface[], query: string): Promise<DocumentInterface[]>;\n}\n"],"mappings":";;;;;;;;;AAOA;AAuBqBK,UAvBJD,sBAAAA,CAuBoB;EAAA,UAAA,EAtBrBJ,mBAsBqB;EAAA,YAIrBA,CAAAA,EAAAA,CAAAA,CAAAA,EAAAA,MAAAA,EAAAA,EAAAA,EAAAA,CAAAA,EAAAA,MAAAA,EAAAA,EAAAA,EAAAA,GAAAA,MAAAA,EAAAA,EAAAA;EAAmB,mBAIVE,CAAAA,EAAAA,MAAAA;EAAgB,CAAA,CAAA,EAWjBE,MAAAA;;;;;AAnB4C;;;;;;;;;;;;;;cAA/CC,gBAAAA,SAAyBF,sBAAAA;;;;cAI9BH;;;;uBAISE;;;;;;;;;;;sBAWDE;+BACSH,qCAAqCK,QAAQL"}

package/dist/retrievers/document_compressors/embeddings_filter.js

@@ -0,0 +1,68 @@
+import { __export } from "../../_virtual/rolldown_runtime.js";
+import { BaseDocumentCompressor } from "./index.js";
+import { cosineSimilarity } from "@langchain/core/utils/math";
+
+//#region src/retrievers/document_compressors/embeddings_filter.ts
+var embeddings_filter_exports = {};
+__export(embeddings_filter_exports, { EmbeddingsFilter: () => EmbeddingsFilter });
+/**
+* Class that represents a document compressor that uses embeddings to
+* drop documents unrelated to the query.
+* @example
+* ```typescript
+* const embeddingsFilter = new EmbeddingsFilter({
+*   embeddings: new OpenAIEmbeddings(),
+*   similarityThreshold: 0.8,
+*   k: 5,
+* });
+* const retrievedDocs = await embeddingsFilter.filterDocuments(
+*   getDocuments(),
+*   "What did the speaker say about Justice Breyer in the 2022 State of the Union?",
+* );
+* console.log({ retrievedDocs });
+* ```
+*/
+var EmbeddingsFilter = class extends BaseDocumentCompressor {
+	/**
+	* Embeddings to use for embedding document contents and queries.
+	*/
+	embeddings;
+	/**
+	* Similarity function for comparing documents.
+	*/
+	similarityFn = cosineSimilarity;
+	/**
+	* Threshold for determining when two documents are similar enough
+	* to be considered redundant. Must be specified if `k` is not set.
+	*/
+	similarityThreshold;
+	/**
+	* The number of relevant documents to return. Can be explicitly set to undefined, in which case
+	* similarity_threshold` must be specified. Defaults to 20
+	*/
+	k = 20;
+	constructor(params) {
+		super();
+		this.embeddings = params.embeddings;
+		this.similarityFn = params.similarityFn ?? this.similarityFn;
+		this.similarityThreshold = params.similarityThreshold;
+		this.k = "k" in params ? params.k : this.k;
+		if (this.k === void 0 && this.similarityThreshold === void 0) throw new Error(`Must specify one of "k" or "similarity_threshold".`);
+	}
+	async compressDocuments(documents, query) {
+		const embeddedDocuments = await this.embeddings.embedDocuments(documents.map((doc) => doc.pageContent));
+		const embeddedQuery = await this.embeddings.embedQuery(query);
+		const similarity = this.similarityFn([embeddedQuery], embeddedDocuments)[0];
+		let includedIdxs = Array.from({ length: embeddedDocuments.length }, (_, i) => i);
+		if (this.k !== void 0) includedIdxs = includedIdxs.map((v, i) => [similarity[i], v]).sort(([a], [b]) => b - a).slice(0, this.k).map(([, i]) => i);
+		if (this.similarityThreshold !== void 0) {
+			const threshold = this.similarityThreshold;
+			includedIdxs = includedIdxs.filter((i) => similarity[i] > threshold);
+		}
+		return includedIdxs.map((i) => documents[i]);
+	}
+};
+
+//#endregion
+export { EmbeddingsFilter, embeddings_filter_exports };
+//# sourceMappingURL=embeddings_filter.js.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"embeddings_filter.js","names":["params: EmbeddingsFilterParams","documents: DocumentInterface[]","query: string"],"sources":["../../../src/retrievers/document_compressors/embeddings_filter.ts"],"sourcesContent":["import type { EmbeddingsInterface } from \"@langchain/core/embeddings\";\nimport type { DocumentInterface } from \"@langchain/core/documents\";\nimport { cosineSimilarity } from \"@langchain/core/utils/math\";\nimport { BaseDocumentCompressor } from \"./index.js\";\n\n/**\n * Interface for the parameters of the `EmbeddingsFilter` class.\n */\nexport interface EmbeddingsFilterParams {\n  embeddings: EmbeddingsInterface;\n  similarityFn?: (x: number[][], y: number[][]) => number[][];\n  similarityThreshold?: number;\n  k?: number;\n}\n\n/**\n * Class that represents a document compressor that uses embeddings to\n * drop documents unrelated to the query.\n * @example\n * ```typescript\n * const embeddingsFilter = new EmbeddingsFilter({\n *   embeddings: new OpenAIEmbeddings(),\n *   similarityThreshold: 0.8,\n *   k: 5,\n * });\n * const retrievedDocs = await embeddingsFilter.filterDocuments(\n *   getDocuments(),\n *   \"What did the speaker say about Justice Breyer in the 2022 State of the Union?\",\n * );\n * console.log({ retrievedDocs });\n * ```\n */\nexport class EmbeddingsFilter extends BaseDocumentCompressor {\n  /**\n   * Embeddings to use for embedding document contents and queries.\n   */\n  embeddings: EmbeddingsInterface;\n\n  /**\n   * Similarity function for comparing documents.\n   */\n  similarityFn = cosineSimilarity;\n\n  /**\n   * Threshold for determining when two documents are similar enough\n   * to be considered redundant. Must be specified if `k` is not set.\n   */\n  similarityThreshold?: number;\n\n  /**\n   * The number of relevant documents to return. Can be explicitly set to undefined, in which case\n   * similarity_threshold` must be specified. Defaults to 20\n   */\n  k? = 20;\n\n  constructor(params: EmbeddingsFilterParams) {\n    super();\n    this.embeddings = params.embeddings;\n    this.similarityFn = params.similarityFn ?? this.similarityFn;\n    this.similarityThreshold = params.similarityThreshold;\n    this.k = \"k\" in params ? params.k : this.k;\n    if (this.k === undefined && this.similarityThreshold === undefined) {\n      throw new Error(`Must specify one of \"k\" or \"similarity_threshold\".`);\n    }\n  }\n\n  async compressDocuments(\n    documents: DocumentInterface[],\n    query: string\n  ): Promise<DocumentInterface[]> {\n    const embeddedDocuments = await this.embeddings.embedDocuments(\n      documents.map((doc) => doc.pageContent)\n    );\n    const embeddedQuery = await this.embeddings.embedQuery(query);\n    const similarity = this.similarityFn([embeddedQuery], embeddedDocuments)[0];\n    let includedIdxs = Array.from(\n      { length: embeddedDocuments.length },\n      (_, i) => i\n    );\n\n    if (this.k !== undefined) {\n      includedIdxs = includedIdxs\n        .map((v, i) => [similarity[i], v])\n        .sort(([a], [b]) => b - a)\n        .slice(0, this.k)\n        .map(([, i]) => i);\n    }\n\n    if (this.similarityThreshold !== undefined) {\n      const threshold = this.similarityThreshold;\n      includedIdxs = includedIdxs.filter((i) => similarity[i] > threshold);\n    }\n\n    return includedIdxs.map((i) => documents[i]);\n  }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;AAgCA,IAAa,mBAAb,cAAsC,uBAAuB;;;;CAI3D;;;;CAKA,eAAe;;;;;CAMf;;;;;CAMA,IAAK;CAEL,YAAYA,QAAgC;EAC1C,OAAO;EACP,KAAK,aAAa,OAAO;EACzB,KAAK,eAAe,OAAO,gBAAgB,KAAK;EAChD,KAAK,sBAAsB,OAAO;EAClC,KAAK,IAAI,OAAO,SAAS,OAAO,IAAI,KAAK;AACzC,MAAI,KAAK,MAAM,UAAa,KAAK,wBAAwB,OACvD,OAAM,IAAI,MAAM,CAAC,kDAAkD,CAAC;CAEvE;CAED,MAAM,kBACJC,WACAC,OAC8B;EAC9B,MAAM,oBAAoB,MAAM,KAAK,WAAW,eAC9C,UAAU,IAAI,CAAC,QAAQ,IAAI,YAAY,CACxC;EACD,MAAM,gBAAgB,MAAM,KAAK,WAAW,WAAW,MAAM;EAC7D,MAAM,aAAa,KAAK,aAAa,CAAC,aAAc,GAAE,kBAAkB,CAAC;EACzE,IAAI,eAAe,MAAM,KACvB,EAAE,QAAQ,kBAAkB,OAAQ,GACpC,CAAC,GAAG,MAAM,EACX;AAED,MAAI,KAAK,MAAM,QACb,eAAe,aACZ,IAAI,CAAC,GAAG,MAAM,CAAC,WAAW,IAAI,CAAE,EAAC,CACjC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,EAAE,KAAK,IAAI,EAAE,CACzB,MAAM,GAAG,KAAK,EAAE,CAChB,IAAI,CAAC,GAAG,EAAE,KAAK,EAAE;AAGtB,MAAI,KAAK,wBAAwB,QAAW;GAC1C,MAAM,YAAY,KAAK;GACvB,eAAe,aAAa,OAAO,CAAC,MAAM,WAAW,KAAK,UAAU;EACrE;AAED,SAAO,aAAa,IAAI,CAAC,MAAM,UAAU,GAAG;CAC7C;AACF"}

package/dist/retrievers/document_compressors/index.cjs

@@ -0,0 +1,69 @@
+const require_rolldown_runtime = require('../../_virtual/rolldown_runtime.cjs');
+
+//#region src/retrievers/document_compressors/index.ts
+var document_compressors_exports = {};
+require_rolldown_runtime.__export(document_compressors_exports, {
+	BaseDocumentCompressor: () => BaseDocumentCompressor,
+	DocumentCompressorPipeline: () => DocumentCompressorPipeline
+});
+/**
+* Base Document Compression class. All compressors should extend this class.
+*/
+var BaseDocumentCompressor = class {
+	static isBaseDocumentCompressor(x) {
+		return x?.compressDocuments !== void 0;
+	}
+};
+/**
+* Document compressor that uses a pipeline of Transformers.
+* @example
+* ```typescript
+* const compressorPipeline = new DocumentCompressorPipeline({
+*   transformers: [
+*     new RecursiveCharacterTextSplitter({
+*       chunkSize: 200,
+*       chunkOverlap: 0,
+*     }),
+*     new EmbeddingsFilter({
+*       embeddings: new OpenAIEmbeddings(),
+*       similarityThreshold: 0.8,
+*       k: 5,
+*     }),
+*   ],
+* });
+* const retriever = new ContextualCompressionRetriever({
+*   baseCompressor: compressorPipeline,
+*   baseRetriever: new TavilySearchAPIRetriever({
+*     includeRawContent: true,
+*   }),
+* });
+* const retrievedDocs = await retriever.invoke(
+*   "What did the speaker say about Justice Breyer in the 2022 State of the Union?",
+* );
+* console.log({ retrievedDocs });
+* ```
+*/
+var DocumentCompressorPipeline = class extends BaseDocumentCompressor {
+	transformers;
+	constructor(fields) {
+		super();
+		this.transformers = fields.transformers;
+	}
+	async compressDocuments(documents, query, callbacks) {
+		let transformedDocuments = documents;
+		for (const transformer of this.transformers) if (BaseDocumentCompressor.isBaseDocumentCompressor(transformer)) transformedDocuments = await transformer.compressDocuments(transformedDocuments, query, callbacks);
+		else transformedDocuments = await transformer.transformDocuments(transformedDocuments);
+		return transformedDocuments;
+	}
+};
+
+//#endregion
+exports.BaseDocumentCompressor = BaseDocumentCompressor;
+exports.DocumentCompressorPipeline = DocumentCompressorPipeline;
+Object.defineProperty(exports, 'document_compressors_exports', {
+  enumerable: true,
+  get: function () {
+    return document_compressors_exports;
+  }
+});
+//# sourceMappingURL=index.cjs.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"index.cjs","names":["x: any","fields: {\n    transformers: (BaseDocumentTransformer | BaseDocumentCompressor)[];\n  }","documents: DocumentInterface[]","query: string","callbacks?: Callbacks"],"sources":["../../../src/retrievers/document_compressors/index.ts"],"sourcesContent":["import type { DocumentInterface } from \"@langchain/core/documents\";\nimport { BaseDocumentTransformer } from \"@langchain/core/documents\";\nimport { Callbacks } from \"@langchain/core/callbacks/manager\";\n\n/**\n * Base Document Compression class. All compressors should extend this class.\n */\nexport abstract class BaseDocumentCompressor {\n  /**\n   * Abstract method that must be implemented by any class that extends\n   * `BaseDocumentCompressor`. This method takes an array of `Document`\n   * objects and a query string as parameters and returns a Promise that\n   * resolves with an array of compressed `Document` objects.\n   * @param documents An array of `Document` objects to be compressed.\n   * @param query A query string.\n   * @returns A Promise that resolves with an array of compressed `Document` objects.\n   */\n  abstract compressDocuments(\n    documents: DocumentInterface[],\n    query: string,\n    callbacks?: Callbacks\n  ): Promise<DocumentInterface[]>;\n\n  // eslint-disable-next-line @typescript-eslint/no-explicit-any\n  static isBaseDocumentCompressor(x: any): x is BaseDocumentCompressor {\n    return x?.compressDocuments !== undefined;\n  }\n}\n\n/**\n * Document compressor that uses a pipeline of Transformers.\n * @example\n * ```typescript\n * const compressorPipeline = new DocumentCompressorPipeline({\n *   transformers: [\n *     new RecursiveCharacterTextSplitter({\n *       chunkSize: 200,\n *       chunkOverlap: 0,\n *     }),\n *     new EmbeddingsFilter({\n *       embeddings: new OpenAIEmbeddings(),\n *       similarityThreshold: 0.8,\n *       k: 5,\n *     }),\n *   ],\n * });\n * const retriever = new ContextualCompressionRetriever({\n *   baseCompressor: compressorPipeline,\n *   baseRetriever: new TavilySearchAPIRetriever({\n *     includeRawContent: true,\n *   }),\n * });\n * const retrievedDocs = await retriever.invoke(\n *   \"What did the speaker say about Justice Breyer in the 2022 State of the Union?\",\n * );\n * console.log({ retrievedDocs });\n * ```\n */\nexport class DocumentCompressorPipeline extends BaseDocumentCompressor {\n  transformers: (BaseDocumentTransformer | BaseDocumentCompressor)[];\n\n  constructor(fields: {\n    transformers: (BaseDocumentTransformer | BaseDocumentCompressor)[];\n  }) {\n    super();\n    this.transformers = fields.transformers;\n  }\n\n  async compressDocuments(\n    documents: DocumentInterface[],\n    query: string,\n    callbacks?: Callbacks\n  ): Promise<DocumentInterface[]> {\n    let transformedDocuments = documents;\n    for (const transformer of this.transformers) {\n      if (BaseDocumentCompressor.isBaseDocumentCompressor(transformer)) {\n        transformedDocuments = await transformer.compressDocuments(\n          transformedDocuments,\n          query,\n          callbacks\n        );\n      } else {\n        transformedDocuments = await transformer.transformDocuments(\n          transformedDocuments\n        );\n      }\n    }\n    return transformedDocuments;\n  }\n}\n"],"mappings":";;;;;;;;;;;AAOA,IAAsB,yBAAtB,MAA6C;CAiB3C,OAAO,yBAAyBA,GAAqC;AACnE,SAAO,GAAG,sBAAsB;CACjC;AACF;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AA+BD,IAAa,6BAAb,cAAgD,uBAAuB;CACrE;CAEA,YAAYC,QAET;EACD,OAAO;EACP,KAAK,eAAe,OAAO;CAC5B;CAED,MAAM,kBACJC,WACAC,OACAC,WAC8B;EAC9B,IAAI,uBAAuB;AAC3B,OAAK,MAAM,eAAe,KAAK,aAC7B,KAAI,uBAAuB,yBAAyB,YAAY,EAC9D,uBAAuB,MAAM,YAAY,kBACvC,sBACA,OACA,UACD;OAED,uBAAuB,MAAM,YAAY,mBACvC,qBACD;AAGL,SAAO;CACR;AACF"}

package/dist/retrievers/document_compressors/index.d.cts

@@ -0,0 +1,61 @@
+import { Callbacks } from "@langchain/core/callbacks/manager";
+import { BaseDocumentTransformer, DocumentInterface } from "@langchain/core/documents";
+
+//#region src/retrievers/document_compressors/index.d.ts
+
+/**
+ * Base Document Compression class. All compressors should extend this class.
+ */
+declare abstract class BaseDocumentCompressor {
+  /**
+   * Abstract method that must be implemented by any class that extends
+   * `BaseDocumentCompressor`. This method takes an array of `Document`
+   * objects and a query string as parameters and returns a Promise that
+   * resolves with an array of compressed `Document` objects.
+   * @param documents An array of `Document` objects to be compressed.
+   * @param query A query string.
+   * @returns A Promise that resolves with an array of compressed `Document` objects.
+   */
+  abstract compressDocuments(documents: DocumentInterface[], query: string, callbacks?: Callbacks): Promise<DocumentInterface[]>;
+  // eslint-disable-next-line @typescript-eslint/no-explicit-any
+  static isBaseDocumentCompressor(x: any): x is BaseDocumentCompressor;
+}
+/**
+ * Document compressor that uses a pipeline of Transformers.
+ * @example
+ * ```typescript
+ * const compressorPipeline = new DocumentCompressorPipeline({
+ *   transformers: [
+ *     new RecursiveCharacterTextSplitter({
+ *       chunkSize: 200,
+ *       chunkOverlap: 0,
+ *     }),
+ *     new EmbeddingsFilter({
+ *       embeddings: new OpenAIEmbeddings(),
+ *       similarityThreshold: 0.8,
+ *       k: 5,
+ *     }),
+ *   ],
+ * });
+ * const retriever = new ContextualCompressionRetriever({
+ *   baseCompressor: compressorPipeline,
+ *   baseRetriever: new TavilySearchAPIRetriever({
+ *     includeRawContent: true,
+ *   }),
+ * });
+ * const retrievedDocs = await retriever.invoke(
+ *   "What did the speaker say about Justice Breyer in the 2022 State of the Union?",
+ * );
+ * console.log({ retrievedDocs });
+ * ```
+ */
+declare class DocumentCompressorPipeline extends BaseDocumentCompressor {
+  transformers: (BaseDocumentTransformer | BaseDocumentCompressor)[];
+  constructor(fields: {
+    transformers: (BaseDocumentTransformer | BaseDocumentCompressor)[];
+  });
+  compressDocuments(documents: DocumentInterface[], query: string, callbacks?: Callbacks): Promise<DocumentInterface[]>;
+}
+//#endregion
+export { BaseDocumentCompressor, DocumentCompressorPipeline };
+//# sourceMappingURL=index.d.cts.map

package/dist/retrievers/document_compressors/index.d.cts.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.d.cts","names":["DocumentInterface","BaseDocumentTransformer","Callbacks","BaseDocumentCompressor","Promise","DocumentCompressorPipeline"],"sources":["../../../src/retrievers/document_compressors/index.d.ts"],"sourcesContent":["import type { DocumentInterface } from \"@langchain/core/documents\";\nimport { BaseDocumentTransformer } from \"@langchain/core/documents\";\nimport { Callbacks } from \"@langchain/core/callbacks/manager\";\n/**\n * Base Document Compression class. All compressors should extend this class.\n */\nexport declare abstract class BaseDocumentCompressor {\n    /**\n     * Abstract method that must be implemented by any class that extends\n     * `BaseDocumentCompressor`. This method takes an array of `Document`\n     * objects and a query string as parameters and returns a Promise that\n     * resolves with an array of compressed `Document` objects.\n     * @param documents An array of `Document` objects to be compressed.\n     * @param query A query string.\n     * @returns A Promise that resolves with an array of compressed `Document` objects.\n     */\n    abstract compressDocuments(documents: DocumentInterface[], query: string, callbacks?: Callbacks): Promise<DocumentInterface[]>;\n    // eslint-disable-next-line @typescript-eslint/no-explicit-any\n    static isBaseDocumentCompressor(x: any): x is BaseDocumentCompressor;\n}\n/**\n * Document compressor that uses a pipeline of Transformers.\n * @example\n * ```typescript\n * const compressorPipeline = new DocumentCompressorPipeline({\n *   transformers: [\n *     new RecursiveCharacterTextSplitter({\n *       chunkSize: 200,\n *       chunkOverlap: 0,\n *     }),\n *     new EmbeddingsFilter({\n *       embeddings: new OpenAIEmbeddings(),\n *       similarityThreshold: 0.8,\n *       k: 5,\n *     }),\n *   ],\n * });\n * const retriever = new ContextualCompressionRetriever({\n *   baseCompressor: compressorPipeline,\n *   baseRetriever: new TavilySearchAPIRetriever({\n *     includeRawContent: true,\n *   }),\n * });\n * const retrievedDocs = await retriever.invoke(\n *   \"What did the speaker say about Justice Breyer in the 2022 State of the Union?\",\n * );\n * console.log({ retrievedDocs });\n * ```\n */\nexport declare class DocumentCompressorPipeline extends BaseDocumentCompressor {\n    transformers: (BaseDocumentTransformer | BaseDocumentCompressor)[];\n    constructor(fields: {\n        transformers: (BaseDocumentTransformer | BaseDocumentCompressor)[];\n    });\n    compressDocuments(documents: DocumentInterface[], query: string, callbacks?: Callbacks): Promise<DocumentInterface[]>;\n}\n"],"mappings":";;;;;;AAMA;;AAU0CA,uBAVZG,sBAAAA,CAUYH;EAAiB;;;;AAEa;AA+BxE;;;;EACmE,SAE5CC,iBAAAA,CAAAA,SAAAA,EApCmBD,iBAoCnBC,EAAAA,EAAAA,KAAAA,EAAAA,MAAAA,EAAAA,SAAAA,CAAAA,EApCmEC,SAoCnED,CAAAA,EApC+EG,OAoC/EH,CApCuFD,iBAoCvFC,EAAAA,CAAAA;EAAuB;EAAyB,OAEtCD,wBAAAA,CAAAA,CAAAA,EAAAA,GAAAA,CAAAA,EAAAA,CAAAA,IApCiBG,sBAoCjBH;;;;;AAL6C;;;;;;;;;;;;;;;;;;;;;;;;;;cAAzDK,0BAAAA,SAAmCF,sBAAAA;iBACrCF,0BAA0BE;;mBAEtBF,0BAA0BE;;+BAEhBH,gDAAgDE,YAAYE,QAAQJ"}