langchain 0.1.23 → 0.1.25

package/dist/load/import_map.cjs

@@ -24,8 +24,8 @@ var __importStar = (this && this.__importStar) || function (mod) {
     return result;
 };
 Object.defineProperty(exports, "__esModule", { value: true });
-exports.storage__in_memory = exports.storage__encoder_backed = exports.stores__message__in_memory = exports.stores__file__in_memory = exports.stores__doc__in_memory = exports.retrievers__vespa = exports.retrievers__score_threshold = exports.retrievers__hyde = exports.retrievers__document_compressors__embeddings_filter = exports.retrievers__document_compressors__chain_extract = exports.retrievers__time_weighted = exports.retrievers__parent_document = exports.retrievers__multi_vector = exports.retrievers__multi_query = exports.retrievers__document_compressors = exports.retrievers__contextual_compression = exports.retrievers__remote = exports.output_parsers = exports.schema__query_constructor = exports.schema__prompt_template = exports.chat_models__anthropic = exports.document_transformers__openai_functions = exports.document_loaders__web__sort_xyz_blockchain = exports.document_loaders__web__serpapi = exports.document_loaders__web__searchapi = exports.document_loaders__base = exports.text_splitter = exports.vectorstores__memory = exports.llms__fake = exports.embeddings__fake = exports.embeddings__cache_backed = exports.chains__retrieval = exports.chains__openai_functions = exports.chains__history_aware_retriever = exports.chains__combine_documents__reduce = exports.chains__combine_documents = exports.chains = exports.tools__retriever = exports.tools__render = exports.tools__chain = exports.agents__openai__output_parser = exports.agents__xml__output_parser = exports.agents__react__output_parser = exports.agents__format_scratchpad__log_to_message = exports.agents__format_scratchpad__xml = exports.agents__format_scratchpad__log = exports.agents__format_scratchpad__openai_tools = exports.agents__format_scratchpad = exports.agents__toolkits = exports.agents = void 0;
-exports.llms__fireworks = exports.chat_models__fireworks = exports.schema__output = exports.schema__output_parser = exports.schema__runnable = exports.prompts__base = exports.prompts__pipeline = exports.prompts__image = exports.prompts__chat = exports.schema = exports.schema__messages = exports.prompts__prompt = exports.embeddings__openai = exports.llms__openai = exports.chat_models__openai = exports.indexes = exports.runnables__remote = exports.smith = exports.evaluation = exports.experimental__prompts__custom_format = exports.experimental__masking = exports.experimental__chains__violation_of_expectations = exports.experimental__chat_models__bittensor = exports.experimental__plan_and_execute = exports.experimental__generative_agents = exports.experimental__babyagi = exports.experimental__openai_files = exports.experimental__openai_assistant = exports.experimental__autogpt = exports.util__time = exports.util__math = exports.util__document = void 0;
+exports.storage__encoder_backed = exports.stores__message__in_memory = exports.stores__file__in_memory = exports.stores__doc__in_memory = exports.retrievers__matryoshka_retriever = exports.retrievers__vespa = exports.retrievers__score_threshold = exports.retrievers__hyde = exports.retrievers__document_compressors__embeddings_filter = exports.retrievers__document_compressors__chain_extract = exports.retrievers__time_weighted = exports.retrievers__parent_document = exports.retrievers__multi_vector = exports.retrievers__multi_query = exports.retrievers__document_compressors = exports.retrievers__contextual_compression = exports.retrievers__remote = exports.output_parsers = exports.schema__query_constructor = exports.schema__prompt_template = exports.chat_models__anthropic = exports.document_transformers__openai_functions = exports.document_loaders__web__sort_xyz_blockchain = exports.document_loaders__web__serpapi = exports.document_loaders__web__searchapi = exports.document_loaders__base = exports.text_splitter = exports.vectorstores__memory = exports.llms__fake = exports.embeddings__fake = exports.embeddings__cache_backed = exports.chains__retrieval = exports.chains__openai_functions = exports.chains__history_aware_retriever = exports.chains__combine_documents__reduce = exports.chains__combine_documents = exports.chains = exports.tools__retriever = exports.tools__render = exports.tools__chain = exports.agents__openai__output_parser = exports.agents__xml__output_parser = exports.agents__react__output_parser = exports.agents__format_scratchpad__log_to_message = exports.agents__format_scratchpad__xml = exports.agents__format_scratchpad__log = exports.agents__format_scratchpad__openai_tools = exports.agents__format_scratchpad = exports.agents__toolkits = exports.agents = void 0;
+exports.llms__fireworks = exports.chat_models__fireworks = exports.schema__output = exports.schema__output_parser = exports.schema__runnable = exports.prompts__base = exports.prompts__pipeline = exports.prompts__image = exports.prompts__chat = exports.schema = exports.schema__messages = exports.prompts__prompt = exports.embeddings__openai = exports.llms__openai = exports.chat_models__openai = exports.indexes = exports.runnables__remote = exports.smith = exports.evaluation = exports.experimental__prompts__custom_format = exports.experimental__masking = exports.experimental__chains__violation_of_expectations = exports.experimental__chat_models__bittensor = exports.experimental__plan_and_execute = exports.experimental__generative_agents = exports.experimental__babyagi = exports.experimental__openai_files = exports.experimental__openai_assistant = exports.experimental__autogpt = exports.util__time = exports.util__math = exports.util__document = exports.storage__in_memory = void 0;
 exports.agents = __importStar(require("../agents/index.cjs"));
 exports.agents__toolkits = __importStar(require("../agents/toolkits/index.cjs"));
 exports.agents__format_scratchpad = __importStar(require("../agents/format_scratchpad/openai_functions.cjs"));
@@ -71,6 +71,7 @@ exports.retrievers__document_compressors__embeddings_filter = __importStar(requi
 exports.retrievers__hyde = __importStar(require("../retrievers/hyde.cjs"));
 exports.retrievers__score_threshold = __importStar(require("../retrievers/score_threshold.cjs"));
 exports.retrievers__vespa = __importStar(require("../retrievers/vespa.cjs"));
+exports.retrievers__matryoshka_retriever = __importStar(require("../retrievers/matryoshka_retriever.cjs"));
 exports.stores__doc__in_memory = __importStar(require("../stores/doc/in_memory.cjs"));
 exports.stores__file__in_memory = __importStar(require("../stores/file/in_memory.cjs"));
 exports.stores__message__in_memory = __importStar(require("../stores/message/in_memory.cjs"));

package/dist/load/import_map.d.ts

@@ -43,6 +43,7 @@ export * as retrievers__document_compressors__embeddings_filter from "../retriev
 export * as retrievers__hyde from "../retrievers/hyde.js";
 export * as retrievers__score_threshold from "../retrievers/score_threshold.js";
 export * as retrievers__vespa from "../retrievers/vespa.js";
+export * as retrievers__matryoshka_retriever from "../retrievers/matryoshka_retriever.js";
 export * as stores__doc__in_memory from "../stores/doc/in_memory.js";
 export * as stores__file__in_memory from "../stores/file/in_memory.js";
 export * as stores__message__in_memory from "../stores/message/in_memory.js";

package/dist/load/import_map.js

@@ -44,6 +44,7 @@ export * as retrievers__document_compressors__embeddings_filter from "../retriev
 export * as retrievers__hyde from "../retrievers/hyde.js";
 export * as retrievers__score_threshold from "../retrievers/score_threshold.js";
 export * as retrievers__vespa from "../retrievers/vespa.js";
+export * as retrievers__matryoshka_retriever from "../retrievers/matryoshka_retriever.js";
 export * as stores__doc__in_memory from "../stores/doc/in_memory.js";
 export * as stores__file__in_memory from "../stores/file/in_memory.js";
 export * as stores__message__in_memory from "../stores/message/in_memory.js";

package/dist/retrievers/matryoshka_retriever.cjs

@@ -0,0 +1,148 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.MatryoshkaRetriever = void 0;
+const math_1 = require("@langchain/core/utils/math");
+const vectorstores_1 = require("@langchain/core/vectorstores");
+/**
+ * A retriever that uses two sets of embeddings to perform adaptive retrieval. Based
+ * off of the "Matryoshka embeddings: faster OpenAI vector search using Adaptive Retrieval"
+ * blog post {@link https://supabase.com/blog/matryoshka-embeddings}.
+ *
+ *
+ * This class performs "Adaptive Retrieval" for searching text embeddings efficiently using the
+ * Matryoshka Representation Learning (MRL) technique. It retrieves documents similar to a query
+ * embedding in two steps:
+ *
+ * First-pass: Uses a lower dimensional sub-vector from the MRL embedding for an initial, fast,
+ * but less accurate search.
+ *
+ * Second-pass: Re-ranks the top results from the first pass using the full, high-dimensional
+ * embedding for higher accuracy.
+ *
+ *
+ * This code implements MRL embeddings for efficient vector search by combining faster,
+ * lower-dimensional initial search with accurate, high-dimensional re-ranking.
+ */
+class MatryoshkaRetriever extends vectorstores_1.VectorStoreRetriever {
+    constructor(fields) {
+        super(fields);
+        Object.defineProperty(this, "smallK", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: 50
+        });
+        Object.defineProperty(this, "largeK", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: 8
+        });
+        Object.defineProperty(this, "largeEmbeddingKey", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: "lc_large_embedding"
+        });
+        Object.defineProperty(this, "largeEmbeddingModel", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        Object.defineProperty(this, "searchType", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: "cosine"
+        });
+        /**
+         * Override the default `addDocuments` method to embed the documents twice,
+         * once using the larger embeddings model, and then again using the default
+         * embedding model linked to the vector store.
+         *
+         * @param {DocumentInterface[]} documents - An array of documents to add to the vector store.
+         * @param {AddDocumentOptions} options - An optional object containing additional options for adding documents.
+         * @returns {Promise<string[] | void>} A promise that resolves to an array of the document IDs that were added to the vector store.
+         */
+        Object.defineProperty(this, "addDocuments", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: async (documents, options) => {
+                // Insure documents metadata does not contain the large embedding key
+                if (documents.some((doc) => this.largeEmbeddingKey in doc.metadata)) {
+                    throw new Error(`All documents must not contain the large embedding key: ${this.largeEmbeddingKey} in their metadata.`);
+                }
+                const allDocPageContent = documents.map((doc) => doc.pageContent);
+                const allDocLargeEmbeddings = await this.largeEmbeddingModel.embedDocuments(allDocPageContent);
+                const newDocuments = documents.map((doc, idx) => ({
+                    ...doc,
+                    metadata: {
+                        ...doc.metadata,
+                        [this.largeEmbeddingKey]: JSON.stringify(allDocLargeEmbeddings[idx]),
+                    },
+                }));
+                return this.vectorStore.addDocuments(newDocuments, options);
+            }
+        });
+        this.smallK = fields.smallK ?? this.smallK;
+        this.largeK = fields.largeK ?? this.largeK;
+        this.largeEmbeddingKey = fields.largeEmbeddingKey ?? this.largeEmbeddingKey;
+        this.largeEmbeddingModel = fields.largeEmbeddingModel;
+        this.searchType = fields.searchType ?? this.searchType;
+    }
+    /**
+     * Ranks documents based on their similarity to a query embedding using larger embeddings.
+     *
+     * This method takes a query embedding and a list of documents (smallResults) as input. Each document
+     * in the smallResults array has previously been associated with a large embedding stored in its metadata.
+     * Depending on the `searchType` (cosine, innerProduct, or euclidean), it calculates the similarity scores
+     * between the query embedding and each document's large embedding. It then ranks the documents based on
+     * these similarity scores, from the most similar to the least similar.
+     *
+     * The method returns a promise that resolves to an array of the top `largeK` documents, where `largeK`
+     * is a class property defining the number of documents to return. This subset of documents is determined
+     * by sorting the entire list of documents based on their similarity scores and then selecting the top
+     * `largeK` documents.
+     *
+     * @param {number[]} embeddedQuery The embedding of the query, represented as an array of numbers.
+     * @param {DocumentInterface[]} smallResults An array of documents, each with metadata that includes a large embedding for similarity comparison.
+     * @returns {Promise<DocumentInterface[]>} A promise that resolves to an array of the top `largeK` ranked documents based on their similarity to the query embedding.
+     */
+    _rankByLargeEmbeddings(embeddedQuery, smallResults) {
+        const largeEmbeddings = smallResults.map((doc) => JSON.parse(doc.metadata[this.largeEmbeddingKey]));
+        let func;
+        switch (this.searchType) {
+            case "cosine":
+                func = () => (0, math_1.cosineSimilarity)([embeddedQuery], largeEmbeddings);
+                break;
+            case "innerProduct":
+                func = () => (0, math_1.innerProduct)([embeddedQuery], largeEmbeddings);
+                break;
+            case "euclidean":
+                func = () => (0, math_1.euclideanDistance)([embeddedQuery], largeEmbeddings);
+                break;
+            default:
+                throw new Error(`Unknown search type: ${this.searchType}`);
+        }
+        // Calculate the similarity scores between the query embedding and the large embeddings
+        const [similarityScores] = func();
+        // Create an array of indices from 0 to N-1, where N is the number of documents
+        let indices = Array.from({ length: smallResults.length }, (_, index) => index);
+        indices = indices
+            .map((v, i) => [similarityScores[i], v])
+            .sort(([a], [b]) => b - a)
+            .slice(0, this.largeK)
+            .map(([, i]) => i);
+        return indices.map((i) => smallResults[i]);
+    }
+    async _getRelevantDocuments(query) {
+        const [embeddedQuery, smallResults] = await Promise.all([
+            this.largeEmbeddingModel.embedQuery(query),
+            this.vectorStore.similaritySearch(query, this.smallK, this.filter),
+        ]);
+        return this._rankByLargeEmbeddings(embeddedQuery, smallResults);
+    }
+}
+exports.MatryoshkaRetriever = MatryoshkaRetriever;

package/dist/retrievers/matryoshka_retriever.d.ts

@@ -0,0 +1,93 @@
+import { DocumentInterface } from "@langchain/core/documents";
+import { Embeddings } from "@langchain/core/embeddings";
+import { VectorStore, VectorStoreRetriever, VectorStoreRetrieverInput } from "@langchain/core/vectorstores";
+/**
+ * Type for options when adding a document to the VectorStore.
+ */
+type AddDocumentOptions = Record<string, any>;
+export interface MatryoshkaRetrieverFields {
+    /**
+     * The number of documents to retrieve from the small store.
+     * @default 50
+     */
+    smallK?: number;
+    /**
+     * The number of documents to retrieve from the large store.
+     * @default 8
+     */
+    largeK?: number;
+    /**
+     * The metadata key to store the larger embeddings.
+     * @default "lc_large_embedding"
+     */
+    largeEmbeddingKey?: string;
+    /**
+     * The embedding model to use when generating the large
+     * embeddings.
+     */
+    largeEmbeddingModel: Embeddings;
+    /**
+     * The type of search to perform using the large embeddings.
+     * @default "cosine"
+     */
+    searchType?: "cosine" | "innerProduct" | "euclidean";
+}
+/**
+ * A retriever that uses two sets of embeddings to perform adaptive retrieval. Based
+ * off of the "Matryoshka embeddings: faster OpenAI vector search using Adaptive Retrieval"
+ * blog post {@link https://supabase.com/blog/matryoshka-embeddings}.
+ *
+ *
+ * This class performs "Adaptive Retrieval" for searching text embeddings efficiently using the
+ * Matryoshka Representation Learning (MRL) technique. It retrieves documents similar to a query
+ * embedding in two steps:
+ *
+ * First-pass: Uses a lower dimensional sub-vector from the MRL embedding for an initial, fast,
+ * but less accurate search.
+ *
+ * Second-pass: Re-ranks the top results from the first pass using the full, high-dimensional
+ * embedding for higher accuracy.
+ *
+ *
+ * This code implements MRL embeddings for efficient vector search by combining faster,
+ * lower-dimensional initial search with accurate, high-dimensional re-ranking.
+ */
+export declare class MatryoshkaRetriever<Store extends VectorStore = VectorStore> extends VectorStoreRetriever<Store> {
+    smallK: number;
+    largeK: number;
+    largeEmbeddingKey: string;
+    largeEmbeddingModel: Embeddings;
+    searchType: "cosine" | "innerProduct" | "euclidean";
+    constructor(fields: MatryoshkaRetrieverFields & VectorStoreRetrieverInput<Store>);
+    /**
+     * Ranks documents based on their similarity to a query embedding using larger embeddings.
+     *
+     * This method takes a query embedding and a list of documents (smallResults) as input. Each document
+     * in the smallResults array has previously been associated with a large embedding stored in its metadata.
+     * Depending on the `searchType` (cosine, innerProduct, or euclidean), it calculates the similarity scores
+     * between the query embedding and each document's large embedding. It then ranks the documents based on
+     * these similarity scores, from the most similar to the least similar.
+     *
+     * The method returns a promise that resolves to an array of the top `largeK` documents, where `largeK`
+     * is a class property defining the number of documents to return. This subset of documents is determined
+     * by sorting the entire list of documents based on their similarity scores and then selecting the top
+     * `largeK` documents.
+     *
+     * @param {number[]} embeddedQuery The embedding of the query, represented as an array of numbers.
+     * @param {DocumentInterface[]} smallResults An array of documents, each with metadata that includes a large embedding for similarity comparison.
+     * @returns {Promise<DocumentInterface[]>} A promise that resolves to an array of the top `largeK` ranked documents based on their similarity to the query embedding.
+     */
+    private _rankByLargeEmbeddings;
+    _getRelevantDocuments(query: string): Promise<DocumentInterface[]>;
+    /**
+     * Override the default `addDocuments` method to embed the documents twice,
+     * once using the larger embeddings model, and then again using the default
+     * embedding model linked to the vector store.
+     *
+     * @param {DocumentInterface[]} documents - An array of documents to add to the vector store.
+     * @param {AddDocumentOptions} options - An optional object containing additional options for adding documents.
+     * @returns {Promise<string[] | void>} A promise that resolves to an array of the document IDs that were added to the vector store.
+     */
+    addDocuments: (documents: DocumentInterface[], options?: AddDocumentOptions) => Promise<string[] | void>;
+}
+export {};

package/dist/retrievers/matryoshka_retriever.js

@@ -0,0 +1,144 @@
+import { cosineSimilarity, euclideanDistance, innerProduct, } from "@langchain/core/utils/math";
+import { VectorStoreRetriever, } from "@langchain/core/vectorstores";
+/**
+ * A retriever that uses two sets of embeddings to perform adaptive retrieval. Based
+ * off of the "Matryoshka embeddings: faster OpenAI vector search using Adaptive Retrieval"
+ * blog post {@link https://supabase.com/blog/matryoshka-embeddings}.
+ *
+ *
+ * This class performs "Adaptive Retrieval" for searching text embeddings efficiently using the
+ * Matryoshka Representation Learning (MRL) technique. It retrieves documents similar to a query
+ * embedding in two steps:
+ *
+ * First-pass: Uses a lower dimensional sub-vector from the MRL embedding for an initial, fast,
+ * but less accurate search.
+ *
+ * Second-pass: Re-ranks the top results from the first pass using the full, high-dimensional
+ * embedding for higher accuracy.
+ *
+ *
+ * This code implements MRL embeddings for efficient vector search by combining faster,
+ * lower-dimensional initial search with accurate, high-dimensional re-ranking.
+ */
+export class MatryoshkaRetriever extends VectorStoreRetriever {
+    constructor(fields) {
+        super(fields);
+        Object.defineProperty(this, "smallK", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: 50
+        });
+        Object.defineProperty(this, "largeK", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: 8
+        });
+        Object.defineProperty(this, "largeEmbeddingKey", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: "lc_large_embedding"
+        });
+        Object.defineProperty(this, "largeEmbeddingModel", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        Object.defineProperty(this, "searchType", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: "cosine"
+        });
+        /**
+         * Override the default `addDocuments` method to embed the documents twice,
+         * once using the larger embeddings model, and then again using the default
+         * embedding model linked to the vector store.
+         *
+         * @param {DocumentInterface[]} documents - An array of documents to add to the vector store.
+         * @param {AddDocumentOptions} options - An optional object containing additional options for adding documents.
+         * @returns {Promise<string[] | void>} A promise that resolves to an array of the document IDs that were added to the vector store.
+         */
+        Object.defineProperty(this, "addDocuments", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: async (documents, options) => {
+                // Insure documents metadata does not contain the large embedding key
+                if (documents.some((doc) => this.largeEmbeddingKey in doc.metadata)) {
+                    throw new Error(`All documents must not contain the large embedding key: ${this.largeEmbeddingKey} in their metadata.`);
+                }
+                const allDocPageContent = documents.map((doc) => doc.pageContent);
+                const allDocLargeEmbeddings = await this.largeEmbeddingModel.embedDocuments(allDocPageContent);
+                const newDocuments = documents.map((doc, idx) => ({
+                    ...doc,
+                    metadata: {
+                        ...doc.metadata,
+                        [this.largeEmbeddingKey]: JSON.stringify(allDocLargeEmbeddings[idx]),
+                    },
+                }));
+                return this.vectorStore.addDocuments(newDocuments, options);
+            }
+        });
+        this.smallK = fields.smallK ?? this.smallK;
+        this.largeK = fields.largeK ?? this.largeK;
+        this.largeEmbeddingKey = fields.largeEmbeddingKey ?? this.largeEmbeddingKey;
+        this.largeEmbeddingModel = fields.largeEmbeddingModel;
+        this.searchType = fields.searchType ?? this.searchType;
+    }
+    /**
+     * Ranks documents based on their similarity to a query embedding using larger embeddings.
+     *
+     * This method takes a query embedding and a list of documents (smallResults) as input. Each document
+     * in the smallResults array has previously been associated with a large embedding stored in its metadata.
+     * Depending on the `searchType` (cosine, innerProduct, or euclidean), it calculates the similarity scores
+     * between the query embedding and each document's large embedding. It then ranks the documents based on
+     * these similarity scores, from the most similar to the least similar.
+     *
+     * The method returns a promise that resolves to an array of the top `largeK` documents, where `largeK`
+     * is a class property defining the number of documents to return. This subset of documents is determined
+     * by sorting the entire list of documents based on their similarity scores and then selecting the top
+     * `largeK` documents.
+     *
+     * @param {number[]} embeddedQuery The embedding of the query, represented as an array of numbers.
+     * @param {DocumentInterface[]} smallResults An array of documents, each with metadata that includes a large embedding for similarity comparison.
+     * @returns {Promise<DocumentInterface[]>} A promise that resolves to an array of the top `largeK` ranked documents based on their similarity to the query embedding.
+     */
+    _rankByLargeEmbeddings(embeddedQuery, smallResults) {
+        const largeEmbeddings = smallResults.map((doc) => JSON.parse(doc.metadata[this.largeEmbeddingKey]));
+        let func;
+        switch (this.searchType) {
+            case "cosine":
+                func = () => cosineSimilarity([embeddedQuery], largeEmbeddings);
+                break;
+            case "innerProduct":
+                func = () => innerProduct([embeddedQuery], largeEmbeddings);
+                break;
+            case "euclidean":
+                func = () => euclideanDistance([embeddedQuery], largeEmbeddings);
+                break;
+            default:
+                throw new Error(`Unknown search type: ${this.searchType}`);
+        }
+        // Calculate the similarity scores between the query embedding and the large embeddings
+        const [similarityScores] = func();
+        // Create an array of indices from 0 to N-1, where N is the number of documents
+        let indices = Array.from({ length: smallResults.length }, (_, index) => index);
+        indices = indices
+            .map((v, i) => [similarityScores[i], v])
+            .sort(([a], [b]) => b - a)
+            .slice(0, this.largeK)
+            .map(([, i]) => i);
+        return indices.map((i) => smallResults[i]);
+    }
+    async _getRelevantDocuments(query) {
+        const [embeddedQuery, smallResults] = await Promise.all([
+            this.largeEmbeddingModel.embedQuery(query),
+            this.vectorStore.similaritySearch(query, this.smallK, this.filter),
+        ]);
+        return this._rankByLargeEmbeddings(embeddedQuery, smallResults);
+    }
+}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "langchain",
-  "version": "0.1.23",
+  "version": "0.1.25",
   "description": "Typescript bindings for langchain",
   "type": "module",
   "engines": {
@@ -942,6 +942,10 @@
     "retrievers/vespa.js",
     "retrievers/vespa.d.ts",
     "retrievers/vespa.d.cts",
+    "retrievers/matryoshka_retriever.cjs",
+    "retrievers/matryoshka_retriever.js",
+    "retrievers/matryoshka_retriever.d.ts",
+    "retrievers/matryoshka_retriever.d.cts",
     "cache.cjs",
     "cache.js",
     "cache.d.ts",
@@ -1297,7 +1301,7 @@
     "web-auth-library": "^1.0.3",
     "wikipedia": "^2.1.2",
     "youtube-transcript": "^1.0.6",
-    "youtubei.js": "^5.8.0"
+    "youtubei.js": "^9.1.0"
   },
   "peerDependencies": {
     "@aws-sdk/client-s3": "^3.310.0",
@@ -1349,7 +1353,7 @@
     "web-auth-library": "^1.0.3",
     "ws": "^8.14.2",
     "youtube-transcript": "^1.0.6",
-    "youtubei.js": "^5.8.0"
+    "youtubei.js": "^9.1.0"
   },
   "peerDependenciesMeta": {
     "@aws-sdk/client-s3": {
@@ -3641,6 +3645,15 @@
       "import": "./retrievers/vespa.js",
       "require": "./retrievers/vespa.cjs"
     },
+    "./retrievers/matryoshka_retriever": {
+      "types": {
+        "import": "./retrievers/matryoshka_retriever.d.ts",
+        "require": "./retrievers/matryoshka_retriever.d.cts",
+        "default": "./retrievers/matryoshka_retriever.d.ts"
+      },
+      "import": "./retrievers/matryoshka_retriever.js",
+      "require": "./retrievers/matryoshka_retriever.cjs"
+    },
     "./cache": {
       "types": {
         "import": "./cache.d.ts",

package/retrievers/matryoshka_retriever.cjs

@@ -0,0 +1 @@
+module.exports = require('../dist/retrievers/matryoshka_retriever.cjs');

package/retrievers/matryoshka_retriever.d.cts

@@ -0,0 +1 @@
+export * from '../dist/retrievers/matryoshka_retriever.js'

package/retrievers/matryoshka_retriever.d.ts

@@ -0,0 +1 @@
+export * from '../dist/retrievers/matryoshka_retriever.js'

package/retrievers/matryoshka_retriever.js

@@ -0,0 +1 @@
+export * from '../dist/retrievers/matryoshka_retriever.js'