langchain 0.2.4 → 0.2.5

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.util__math = exports.util__document = exports.storage__in_memory = exports.storage__encoder_backed = exports.stores__message__in_memory = exports.stores__file__in_memory = exports.stores__doc__in_memory = exports.stores__doc__base = exports.retrievers__matryoshka_retriever = 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.output_parsers = exports.callbacks = exports.document_transformers__openai_functions = exports.document_loaders__base = exports.memory__chat_memory = exports.memory__index = exports.memory = exports.text_splitter = exports.vectorstores__memory = 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.tools = 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.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.schema__prompt_template = exports.schema__query_constructor = 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__plan_and_execute = exports.experimental__generative_agents = exports.experimental__babyagi = exports.experimental__openai_files = exports.experimental__openai_assistant = exports.experimental__autogpt = exports.util__time = void 0;
+exports.util__document = exports.storage__in_memory = exports.storage__encoder_backed = exports.stores__message__in_memory = exports.stores__file__in_memory = exports.stores__doc__in_memory = exports.stores__doc__base = exports.retrievers__matryoshka_retriever = 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__ensemble = exports.retrievers__document_compressors = exports.retrievers__contextual_compression = exports.output_parsers = exports.callbacks = exports.document_transformers__openai_functions = exports.document_loaders__base = exports.memory__chat_memory = exports.memory__index = exports.memory = exports.text_splitter = exports.vectorstores__memory = 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.tools = 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.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.schema__prompt_template = exports.schema__query_constructor = 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__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 = 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"));
@@ -59,6 +59,7 @@ exports.callbacks = __importStar(require("../callbacks/index.cjs"));
 exports.output_parsers = __importStar(require("../output_parsers/index.cjs"));
 exports.retrievers__contextual_compression = __importStar(require("../retrievers/contextual_compression.cjs"));
 exports.retrievers__document_compressors = __importStar(require("../retrievers/document_compressors/index.cjs"));
+exports.retrievers__ensemble = __importStar(require("../retrievers/ensemble.cjs"));
 exports.retrievers__multi_query = __importStar(require("../retrievers/multi_query.cjs"));
 exports.retrievers__multi_vector = __importStar(require("../retrievers/multi_vector.cjs"));
 exports.retrievers__parent_document = __importStar(require("../retrievers/parent_document.cjs"));

package/dist/load/import_map.d.ts

@@ -31,6 +31,7 @@ export * as callbacks from "../callbacks/index.js";
 export * as output_parsers from "../output_parsers/index.js";
 export * as retrievers__contextual_compression from "../retrievers/contextual_compression.js";
 export * as retrievers__document_compressors from "../retrievers/document_compressors/index.js";
+export * as retrievers__ensemble from "../retrievers/ensemble.js";
 export * as retrievers__multi_query from "../retrievers/multi_query.js";
 export * as retrievers__multi_vector from "../retrievers/multi_vector.js";
 export * as retrievers__parent_document from "../retrievers/parent_document.js";

package/dist/load/import_map.js

@@ -32,6 +32,7 @@ export * as callbacks from "../callbacks/index.js";
 export * as output_parsers from "../output_parsers/index.js";
 export * as retrievers__contextual_compression from "../retrievers/contextual_compression.js";
 export * as retrievers__document_compressors from "../retrievers/document_compressors/index.js";
+export * as retrievers__ensemble from "../retrievers/ensemble.js";
 export * as retrievers__multi_query from "../retrievers/multi_query.js";
 export * as retrievers__multi_vector from "../retrievers/multi_vector.js";
 export * as retrievers__parent_document from "../retrievers/parent_document.js";

package/dist/retrievers/ensemble.cjs

@@ -0,0 +1,91 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.EnsembleRetriever = void 0;
+const retrievers_1 = require("@langchain/core/retrievers");
+/**
+ * Ensemble retriever that aggregates and orders the results of
+ * multiple retrievers by using weighted Reciprocal Rank Fusion.
+ */
+class EnsembleRetriever extends retrievers_1.BaseRetriever {
+    static lc_name() {
+        return "EnsembleRetriever";
+    }
+    constructor(args) {
+        super(args);
+        Object.defineProperty(this, "lc_namespace", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: ["langchain", "retrievers", "ensemble_retriever"]
+        });
+        Object.defineProperty(this, "retrievers", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        Object.defineProperty(this, "weights", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        Object.defineProperty(this, "c", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: 60
+        });
+        this.retrievers = args.retrievers;
+        this.weights =
+            args.weights ||
+                new Array(args.retrievers.length).fill(1 / args.retrievers.length);
+        this.c = args.c || 60;
+    }
+    async _getRelevantDocuments(query, runManager) {
+        return this._rankFusion(query, runManager);
+    }
+    async _rankFusion(query, runManager) {
+        const retrieverDocs = await Promise.all(this.retrievers.map((retriever, i) => retriever.invoke(query, {
+            callbacks: runManager?.getChild(`retriever_${i + 1}`),
+        })));
+        const fusedDocs = await this._weightedReciprocalRank(retrieverDocs);
+        return fusedDocs;
+    }
+    async _weightedReciprocalRank(docList) {
+        if (docList.length !== this.weights.length) {
+            throw new Error("Number of retrieved document lists must be equal to the number of weights.");
+        }
+        const rrfScoreDict = docList.reduce((rffScore, retrieverDoc, idx) => {
+            let rank = 1;
+            const weight = this.weights[idx];
+            while (rank <= retrieverDoc.length) {
+                const { pageContent } = retrieverDoc[rank - 1];
+                if (!rffScore[pageContent]) {
+                    // eslint-disable-next-line no-param-reassign
+                    rffScore[pageContent] = 0;
+                }
+                // eslint-disable-next-line no-param-reassign
+                rffScore[pageContent] += weight / (rank + this.c);
+                rank += 1;
+            }
+            return rffScore;
+        }, {});
+        const uniqueDocs = this._uniqueUnion(docList.flat());
+        const sortedDocs = Array.from(uniqueDocs).sort((a, b) => rrfScoreDict[b.pageContent] - rrfScoreDict[a.pageContent]);
+        return sortedDocs;
+    }
+    _uniqueUnion(documents) {
+        const documentSet = new Set();
+        const result = [];
+        for (const doc of documents) {
+            const key = doc.pageContent;
+            if (!documentSet.has(key)) {
+                documentSet.add(key);
+                result.push(doc);
+            }
+        }
+        return result;
+    }
+}
+exports.EnsembleRetriever = EnsembleRetriever;

package/dist/retrievers/ensemble.d.ts

@@ -0,0 +1,34 @@
+import { BaseRetriever, BaseRetrieverInput } from "@langchain/core/retrievers";
+import { Document, DocumentInterface } from "@langchain/core/documents";
+import { CallbackManagerForRetrieverRun } from "@langchain/core/callbacks/manager";
+export interface EnsembleRetrieverInput extends BaseRetrieverInput {
+    /** A list of retrievers to ensemble. */
+    retrievers: BaseRetriever[];
+    /**
+     * A list of weights corresponding to the retrievers. Defaults to equal
+     * weighting for all retrievers.
+     */
+    weights?: number[];
+    /**
+     * A constant added to the rank, controlling the balance between the importance
+     * of high-ranked items and the consideration given to lower-ranked items.
+     * Default is 60.
+     */
+    c?: number;
+}
+/**
+ * Ensemble retriever that aggregates and orders the results of
+ * multiple retrievers by using weighted Reciprocal Rank Fusion.
+ */
+export declare class EnsembleRetriever extends BaseRetriever {
+    static lc_name(): string;
+    lc_namespace: string[];
+    retrievers: BaseRetriever[];
+    weights: number[];
+    c: number;
+    constructor(args: EnsembleRetrieverInput);
+    _getRelevantDocuments(query: string, runManager?: CallbackManagerForRetrieverRun): Promise<Document<Record<string, any>>[]>;
+    _rankFusion(query: string, runManager?: CallbackManagerForRetrieverRun): Promise<Document<Record<string, any>>[]>;
+    _weightedReciprocalRank(docList: DocumentInterface[][]): Promise<Document<Record<string, any>>[]>;
+    private _uniqueUnion;
+}

package/dist/retrievers/ensemble.js

@@ -0,0 +1,87 @@
+import { BaseRetriever } from "@langchain/core/retrievers";
+/**
+ * Ensemble retriever that aggregates and orders the results of
+ * multiple retrievers by using weighted Reciprocal Rank Fusion.
+ */
+export class EnsembleRetriever extends BaseRetriever {
+    static lc_name() {
+        return "EnsembleRetriever";
+    }
+    constructor(args) {
+        super(args);
+        Object.defineProperty(this, "lc_namespace", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: ["langchain", "retrievers", "ensemble_retriever"]
+        });
+        Object.defineProperty(this, "retrievers", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        Object.defineProperty(this, "weights", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        Object.defineProperty(this, "c", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: 60
+        });
+        this.retrievers = args.retrievers;
+        this.weights =
+            args.weights ||
+                new Array(args.retrievers.length).fill(1 / args.retrievers.length);
+        this.c = args.c || 60;
+    }
+    async _getRelevantDocuments(query, runManager) {
+        return this._rankFusion(query, runManager);
+    }
+    async _rankFusion(query, runManager) {
+        const retrieverDocs = await Promise.all(this.retrievers.map((retriever, i) => retriever.invoke(query, {
+            callbacks: runManager?.getChild(`retriever_${i + 1}`),
+        })));
+        const fusedDocs = await this._weightedReciprocalRank(retrieverDocs);
+        return fusedDocs;
+    }
+    async _weightedReciprocalRank(docList) {
+        if (docList.length !== this.weights.length) {
+            throw new Error("Number of retrieved document lists must be equal to the number of weights.");
+        }
+        const rrfScoreDict = docList.reduce((rffScore, retrieverDoc, idx) => {
+            let rank = 1;
+            const weight = this.weights[idx];
+            while (rank <= retrieverDoc.length) {
+                const { pageContent } = retrieverDoc[rank - 1];
+                if (!rffScore[pageContent]) {
+                    // eslint-disable-next-line no-param-reassign
+                    rffScore[pageContent] = 0;
+                }
+                // eslint-disable-next-line no-param-reassign
+                rffScore[pageContent] += weight / (rank + this.c);
+                rank += 1;
+            }
+            return rffScore;
+        }, {});
+        const uniqueDocs = this._uniqueUnion(docList.flat());
+        const sortedDocs = Array.from(uniqueDocs).sort((a, b) => rrfScoreDict[b.pageContent] - rrfScoreDict[a.pageContent]);
+        return sortedDocs;
+    }
+    _uniqueUnion(documents) {
+        const documentSet = new Set();
+        const result = [];
+        for (const doc of documents) {
+            const key = doc.pageContent;
+            if (!documentSet.has(key)) {
+                documentSet.add(key);
+                result.push(doc);
+            }
+        }
+        return result;
+    }
+}

package/dist/retrievers/tests/ensemble_retriever.int.test.d.ts

@@ -0,0 +1 @@
+export {};

package/dist/retrievers/tests/ensemble_retriever.int.test.js

@@ -0,0 +1,74 @@
+import { expect, test } from "@jest/globals";
+import { CohereEmbeddings } from "@langchain/cohere";
+import { MemoryVectorStore } from "../../vectorstores/memory.js";
+import { EnsembleRetriever } from "../ensemble.js";
+test("Should work with a question input", async () => {
+    const vectorstore = await MemoryVectorStore.fromTexts([
+        "Buildings are made out of brick",
+        "Buildings are made out of wood",
+        "Buildings are made out of stone",
+        "Cars are made out of metal",
+        "Cars are made out of plastic",
+        "mitochondria is the powerhouse of the cell",
+        "mitochondria is made of lipids",
+    ], [{ id: 1 }, { id: 2 }, { id: 3 }, { id: 4 }, { id: 5 }], new CohereEmbeddings());
+    const retriever = new EnsembleRetriever({
+        retrievers: [vectorstore.asRetriever()],
+    });
+    const query = "What are mitochondria made of?";
+    const retrievedDocs = await retriever.invoke(query);
+    expect(retrievedDocs[0].pageContent).toContain("mitochondria");
+});
+test("Should work with multiple retriever", async () => {
+    const vectorstore = await MemoryVectorStore.fromTexts([
+        "Buildings are made out of brick",
+        "Buildings are made out of wood",
+        "Buildings are made out of stone",
+        "Cars are made out of metal",
+        "Cars are made out of plastic",
+        "mitochondria is the powerhouse of the cell",
+        "mitochondria is made of lipids",
+    ], [{ id: 1 }, { id: 2 }, { id: 3 }, { id: 4 }, { id: 5 }], new CohereEmbeddings());
+    const vectorstore2 = await MemoryVectorStore.fromTexts([
+        "Buildings are made out of brick",
+        "Buildings are made out of wood",
+        "Buildings are made out of stone",
+        "Cars are made out of metal",
+        "Cars are made out of plastic",
+        "mitochondria is the powerhouse of the cell",
+        "mitochondria is made of lipids",
+    ], [{ id: 6 }, { id: 7 }, { id: 8 }, { id: 9 }, { id: 10 }], new CohereEmbeddings());
+    const retriever = new EnsembleRetriever({
+        retrievers: [vectorstore.asRetriever(), vectorstore2.asRetriever()],
+    });
+    const query = "cars";
+    const retrievedDocs = await retriever.invoke(query);
+    expect(retrievedDocs.filter((item) => item.pageContent.includes("Cars")).length).toBe(2);
+});
+test("Should work with weights", async () => {
+    const vectorstore = await MemoryVectorStore.fromTexts([
+        "Buildings are made out of brick",
+        "Buildings are made out of wood",
+        "Buildings are made out of stone",
+        "Cars are made out of metal",
+        "Cars are made out of plastic",
+        "mitochondria is the powerhouse of the cell",
+        "mitochondria is made of lipids",
+    ], [{ id: 1 }, { id: 2 }, { id: 3 }, { id: 4 }, { id: 5 }], new CohereEmbeddings());
+    const vectorstore2 = await MemoryVectorStore.fromTexts([
+        "Buildings are made out of brick",
+        "Buildings are made out of wood",
+        "Buildings are made out of stone",
+        "Cars are made out of metal",
+        "Cars are made out of plastic",
+        "mitochondria is the powerhouse of the cell",
+        "mitochondria is made of lipids",
+    ], [{ id: 6 }, { id: 7 }, { id: 8 }, { id: 9 }, { id: 10 }], new CohereEmbeddings());
+    const retriever = new EnsembleRetriever({
+        retrievers: [vectorstore.asRetriever(), vectorstore2.asRetriever()],
+        weights: [0.5, 0.9],
+    });
+    const query = "cars";
+    const retrievedDocs = await retriever.invoke(query);
+    expect(retrievedDocs.filter((item) => item.pageContent.includes("Cars")).length).toBe(2);
+});

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "langchain",
-  "version": "0.2.4",
+  "version": "0.2.5",
   "description": "Typescript bindings for langchain",
   "type": "module",
   "engines": {
@@ -370,6 +370,10 @@
     "retrievers/document_compressors.js",
     "retrievers/document_compressors.d.ts",
     "retrievers/document_compressors.d.cts",
+    "retrievers/ensemble.cjs",
+    "retrievers/ensemble.js",
+    "retrievers/ensemble.d.ts",
+    "retrievers/ensemble.d.cts",
     "retrievers/multi_query.cjs",
     "retrievers/multi_query.js",
     "retrievers/multi_query.d.ts",
@@ -1725,6 +1729,15 @@
       "import": "./retrievers/document_compressors.js",
       "require": "./retrievers/document_compressors.cjs"
     },
+    "./retrievers/ensemble": {
+      "types": {
+        "import": "./retrievers/ensemble.d.ts",
+        "require": "./retrievers/ensemble.d.cts",
+        "default": "./retrievers/ensemble.d.ts"
+      },
+      "import": "./retrievers/ensemble.js",
+      "require": "./retrievers/ensemble.cjs"
+    },
     "./retrievers/multi_query": {
       "types": {
         "import": "./retrievers/multi_query.d.ts",

package/retrievers/ensemble.cjs

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

package/retrievers/ensemble.d.cts

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

package/retrievers/ensemble.d.ts

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

package/retrievers/ensemble.js

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