langchain 0.0.170 → 0.0.172

package/dist/chat_models/llama_cpp.cjs

@@ -0,0 +1,243 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.ChatLlamaCpp = void 0;
+const node_llama_cpp_1 = require("node-llama-cpp");
+const base_js_1 = require("./base.cjs");
+/**
+ *  To use this model you need to have the `node-llama-cpp` module installed.
+ *  This can be installed using `npm install -S node-llama-cpp` and the minimum
+ *  version supported in version 2.0.0.
+ *  This also requires that have a locally built version of Llama2 installed.
+ */
+class ChatLlamaCpp extends base_js_1.SimpleChatModel {
+    static lc_name() {
+        return "ChatLlamaCpp";
+    }
+    constructor(inputs) {
+        super(inputs);
+        Object.defineProperty(this, "batchSize", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        Object.defineProperty(this, "contextSize", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        Object.defineProperty(this, "embedding", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        Object.defineProperty(this, "f16Kv", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        Object.defineProperty(this, "gpuLayers", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        Object.defineProperty(this, "logitsAll", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        Object.defineProperty(this, "lowVram", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        Object.defineProperty(this, "seed", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        Object.defineProperty(this, "useMlock", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        Object.defineProperty(this, "useMmap", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        Object.defineProperty(this, "vocabOnly", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        Object.defineProperty(this, "modelPath", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        Object.defineProperty(this, "_model", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        Object.defineProperty(this, "_context", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        Object.defineProperty(this, "_session", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        this.batchSize = inputs?.batchSize;
+        this.contextSize = inputs?.contextSize;
+        this.embedding = inputs?.embedding;
+        this.f16Kv = inputs?.f16Kv;
+        this.gpuLayers = inputs?.gpuLayers;
+        this.logitsAll = inputs?.logitsAll;
+        this.lowVram = inputs?.lowVram;
+        this.modelPath = inputs.modelPath;
+        this.seed = inputs?.seed;
+        this.useMlock = inputs?.useMlock;
+        this.useMmap = inputs?.useMmap;
+        this.vocabOnly = inputs?.vocabOnly;
+        this._model = new node_llama_cpp_1.LlamaModel(inputs);
+        this._context = new node_llama_cpp_1.LlamaContext({ model: this._model });
+        this._session = null;
+    }
+    _llmType() {
+        return "llama2_cpp";
+    }
+    invocationParams() {
+        return {
+            batchSize: this.batchSize,
+            contextSize: this.contextSize,
+            embedding: this.embedding,
+            f16Kv: this.f16Kv,
+            gpuLayers: this.gpuLayers,
+            logitsAll: this.logitsAll,
+            lowVram: this.lowVram,
+            modelPath: this.modelPath,
+            seed: this.seed,
+            useMlock: this.useMlock,
+            useMmap: this.useMmap,
+            vocabOnly: this.vocabOnly,
+        };
+    }
+    /** @ignore */
+    _combineLLMOutput() {
+        return {};
+    }
+    /** @ignore */
+    async _call(messages, options) {
+        let prompt = "";
+        if (messages.length > 1) {
+            // We need to build a new _session
+            prompt = this._buildSession(messages);
+        }
+        else if (!this._session) {
+            prompt = this._buildSession(messages);
+        }
+        else {
+            // If we already have a session then we should just have a single prompt
+            prompt = messages[0].content;
+        }
+        try {
+            // @ts-expect-error - TS2531: Object is possibly 'null'.
+            const completion = await this._session.prompt(prompt, options);
+            return completion;
+        }
+        catch (e) {
+            throw new Error("Error getting prompt completion.");
+        }
+    }
+    // This constructs a new session if we need to adding in any sys messages or previous chats
+    _buildSession(messages) {
+        let prompt = "";
+        let sysMessage = "";
+        let noSystemMessages = [];
+        let interactions = [];
+        // Let's see if we have a system message
+        if (messages.findIndex((msg) => msg._getType() === "system") !== -1) {
+            const sysMessages = messages.filter((message) => message._getType() === "system");
+            // Only use the last provided system message
+            sysMessage = sysMessages[sysMessages.length - 1].content;
+            // Now filter out the system messages
+            noSystemMessages = messages.filter((message) => message._getType() !== "system");
+        }
+        else {
+            noSystemMessages = messages;
+        }
+        // Lets see if we just have a prompt left or are their previous interactions?
+        if (noSystemMessages.length > 1) {
+            // Is the last message a prompt?
+            if (noSystemMessages[noSystemMessages.length - 1]._getType() === "human") {
+                prompt = noSystemMessages[noSystemMessages.length - 1].content;
+                interactions = this._convertMessagesToInteractions(noSystemMessages.slice(0, noSystemMessages.length - 1));
+            }
+            else {
+                interactions = this._convertMessagesToInteractions(noSystemMessages);
+            }
+        }
+        else {
+            // If there was only a single message we assume it's a prompt
+            prompt = noSystemMessages[0].content;
+        }
+        // Now lets construct a session according to what we got
+        if (sysMessage !== "" && interactions.length > 0) {
+            this._session = new node_llama_cpp_1.LlamaChatSession({
+                context: this._context,
+                conversationHistory: interactions,
+                systemPrompt: sysMessage,
+            });
+        }
+        else if (sysMessage !== "" && interactions.length === 0) {
+            this._session = new node_llama_cpp_1.LlamaChatSession({
+                context: this._context,
+                systemPrompt: sysMessage,
+            });
+        }
+        else if (sysMessage === "" && interactions.length > 0) {
+            this._session = new node_llama_cpp_1.LlamaChatSession({
+                context: this._context,
+                conversationHistory: interactions,
+            });
+        }
+        else {
+            this._session = new node_llama_cpp_1.LlamaChatSession({
+                context: this._context,
+            });
+        }
+        return prompt;
+    }
+    // This builds a an array of interactions
+    _convertMessagesToInteractions(messages) {
+        const result = [];
+        for (let i = 0; i < messages.length; i += 2) {
+            if (i + 1 < messages.length) {
+                result.push({
+                    prompt: messages[i].content,
+                    response: messages[i + 1].content,
+                });
+            }
+        }
+        return result;
+    }
+}
+exports.ChatLlamaCpp = ChatLlamaCpp;

package/dist/chat_models/llama_cpp.d.ts

@@ -0,0 +1,94 @@
+import { LlamaModel, LlamaContext, LlamaChatSession, type ConversationInteraction } from "node-llama-cpp";
+import { SimpleChatModel, BaseChatModelParams } from "./base.js";
+import { BaseLanguageModelCallOptions } from "../base_language/index.js";
+import type { BaseMessage } from "../schema/index.js";
+/**
+ * Note that the modelPath is the only required parameter. For testing you
+ * can set this in the environment variable `LLAMA_PATH`.
+ */
+export interface LlamaCppInputs extends BaseChatModelParams {
+    /** Prompt processing batch size. */
+    batchSize?: number;
+    /** Text context size. */
+    contextSize?: number;
+    /** Embedding mode only. */
+    embedding?: boolean;
+    /** Use fp16 for KV cache. */
+    f16Kv?: boolean;
+    /** Number of layers to store in VRAM. */
+    gpuLayers?: number;
+    /** The llama_eval() call computes all logits, not just the last one. */
+    logitsAll?: boolean;
+    /** If true, reduce VRAM usage at the cost of performance. */
+    lowVram?: boolean;
+    /** Path to the model on the filesystem. */
+    modelPath: string;
+    /** If null, a random seed will be used. */
+    seed?: null | number;
+    /** The randomness of the responses, e.g. 0.1 deterministic, 1.5 creative, 0.8 balanced, 0 disables. */
+    temperature?: number;
+    /** Consider the n most likely tokens, where n is 1 to vocabulary size, 0 disables (uses full vocabulary). Note: only applies when `temperature` > 0. */
+    topK?: number;
+    /** Selects the smallest token set whose probability exceeds P, where P is between 0 - 1, 1 disables. Note: only applies when `temperature` > 0. */
+    topP?: number;
+    /** Force system to keep model in RAM. */
+    useMlock?: boolean;
+    /** Use mmap if possible. */
+    useMmap?: boolean;
+    /** Only load the vocabulary, no weights. */
+    vocabOnly?: boolean;
+}
+export interface LlamaCppCallOptions extends BaseLanguageModelCallOptions {
+    /** The maximum number of tokens the response should contain. */
+    maxTokens?: number;
+    /** A function called when matching the provided token array */
+    onToken?: (tokens: number[]) => void;
+}
+/**
+ *  To use this model you need to have the `node-llama-cpp` module installed.
+ *  This can be installed using `npm install -S node-llama-cpp` and the minimum
+ *  version supported in version 2.0.0.
+ *  This also requires that have a locally built version of Llama2 installed.
+ */
+export declare class ChatLlamaCpp extends SimpleChatModel<LlamaCppCallOptions> {
+    CallOptions: LlamaCppCallOptions;
+    static inputs: LlamaCppInputs;
+    batchSize?: number;
+    contextSize?: number;
+    embedding?: boolean;
+    f16Kv?: boolean;
+    gpuLayers?: number;
+    logitsAll?: boolean;
+    lowVram?: boolean;
+    seed?: null | number;
+    useMlock?: boolean;
+    useMmap?: boolean;
+    vocabOnly?: boolean;
+    modelPath: string;
+    _model: LlamaModel;
+    _context: LlamaContext;
+    _session: LlamaChatSession | null;
+    static lc_name(): string;
+    constructor(inputs: LlamaCppInputs);
+    _llmType(): string;
+    invocationParams(): {
+        batchSize: number | undefined;
+        contextSize: number | undefined;
+        embedding: boolean | undefined;
+        f16Kv: boolean | undefined;
+        gpuLayers: number | undefined;
+        logitsAll: boolean | undefined;
+        lowVram: boolean | undefined;
+        modelPath: string;
+        seed: number | null | undefined;
+        useMlock: boolean | undefined;
+        useMmap: boolean | undefined;
+        vocabOnly: boolean | undefined;
+    };
+    /** @ignore */
+    _combineLLMOutput(): {};
+    /** @ignore */
+    _call(messages: BaseMessage[], options: this["ParsedCallOptions"]): Promise<string>;
+    protected _buildSession(messages: BaseMessage[]): string;
+    protected _convertMessagesToInteractions(messages: BaseMessage[]): ConversationInteraction[];
+}

package/dist/chat_models/llama_cpp.js

@@ -0,0 +1,239 @@
+import { LlamaModel, LlamaContext, LlamaChatSession, } from "node-llama-cpp";
+import { SimpleChatModel } from "./base.js";
+/**
+ *  To use this model you need to have the `node-llama-cpp` module installed.
+ *  This can be installed using `npm install -S node-llama-cpp` and the minimum
+ *  version supported in version 2.0.0.
+ *  This also requires that have a locally built version of Llama2 installed.
+ */
+export class ChatLlamaCpp extends SimpleChatModel {
+    static lc_name() {
+        return "ChatLlamaCpp";
+    }
+    constructor(inputs) {
+        super(inputs);
+        Object.defineProperty(this, "batchSize", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        Object.defineProperty(this, "contextSize", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        Object.defineProperty(this, "embedding", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        Object.defineProperty(this, "f16Kv", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        Object.defineProperty(this, "gpuLayers", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        Object.defineProperty(this, "logitsAll", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        Object.defineProperty(this, "lowVram", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        Object.defineProperty(this, "seed", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        Object.defineProperty(this, "useMlock", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        Object.defineProperty(this, "useMmap", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        Object.defineProperty(this, "vocabOnly", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        Object.defineProperty(this, "modelPath", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        Object.defineProperty(this, "_model", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        Object.defineProperty(this, "_context", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        Object.defineProperty(this, "_session", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        this.batchSize = inputs?.batchSize;
+        this.contextSize = inputs?.contextSize;
+        this.embedding = inputs?.embedding;
+        this.f16Kv = inputs?.f16Kv;
+        this.gpuLayers = inputs?.gpuLayers;
+        this.logitsAll = inputs?.logitsAll;
+        this.lowVram = inputs?.lowVram;
+        this.modelPath = inputs.modelPath;
+        this.seed = inputs?.seed;
+        this.useMlock = inputs?.useMlock;
+        this.useMmap = inputs?.useMmap;
+        this.vocabOnly = inputs?.vocabOnly;
+        this._model = new LlamaModel(inputs);
+        this._context = new LlamaContext({ model: this._model });
+        this._session = null;
+    }
+    _llmType() {
+        return "llama2_cpp";
+    }
+    invocationParams() {
+        return {
+            batchSize: this.batchSize,
+            contextSize: this.contextSize,
+            embedding: this.embedding,
+            f16Kv: this.f16Kv,
+            gpuLayers: this.gpuLayers,
+            logitsAll: this.logitsAll,
+            lowVram: this.lowVram,
+            modelPath: this.modelPath,
+            seed: this.seed,
+            useMlock: this.useMlock,
+            useMmap: this.useMmap,
+            vocabOnly: this.vocabOnly,
+        };
+    }
+    /** @ignore */
+    _combineLLMOutput() {
+        return {};
+    }
+    /** @ignore */
+    async _call(messages, options) {
+        let prompt = "";
+        if (messages.length > 1) {
+            // We need to build a new _session
+            prompt = this._buildSession(messages);
+        }
+        else if (!this._session) {
+            prompt = this._buildSession(messages);
+        }
+        else {
+            // If we already have a session then we should just have a single prompt
+            prompt = messages[0].content;
+        }
+        try {
+            // @ts-expect-error - TS2531: Object is possibly 'null'.
+            const completion = await this._session.prompt(prompt, options);
+            return completion;
+        }
+        catch (e) {
+            throw new Error("Error getting prompt completion.");
+        }
+    }
+    // This constructs a new session if we need to adding in any sys messages or previous chats
+    _buildSession(messages) {
+        let prompt = "";
+        let sysMessage = "";
+        let noSystemMessages = [];
+        let interactions = [];
+        // Let's see if we have a system message
+        if (messages.findIndex((msg) => msg._getType() === "system") !== -1) {
+            const sysMessages = messages.filter((message) => message._getType() === "system");
+            // Only use the last provided system message
+            sysMessage = sysMessages[sysMessages.length - 1].content;
+            // Now filter out the system messages
+            noSystemMessages = messages.filter((message) => message._getType() !== "system");
+        }
+        else {
+            noSystemMessages = messages;
+        }
+        // Lets see if we just have a prompt left or are their previous interactions?
+        if (noSystemMessages.length > 1) {
+            // Is the last message a prompt?
+            if (noSystemMessages[noSystemMessages.length - 1]._getType() === "human") {
+                prompt = noSystemMessages[noSystemMessages.length - 1].content;
+                interactions = this._convertMessagesToInteractions(noSystemMessages.slice(0, noSystemMessages.length - 1));
+            }
+            else {
+                interactions = this._convertMessagesToInteractions(noSystemMessages);
+            }
+        }
+        else {
+            // If there was only a single message we assume it's a prompt
+            prompt = noSystemMessages[0].content;
+        }
+        // Now lets construct a session according to what we got
+        if (sysMessage !== "" && interactions.length > 0) {
+            this._session = new LlamaChatSession({
+                context: this._context,
+                conversationHistory: interactions,
+                systemPrompt: sysMessage,
+            });
+        }
+        else if (sysMessage !== "" && interactions.length === 0) {
+            this._session = new LlamaChatSession({
+                context: this._context,
+                systemPrompt: sysMessage,
+            });
+        }
+        else if (sysMessage === "" && interactions.length > 0) {
+            this._session = new LlamaChatSession({
+                context: this._context,
+                conversationHistory: interactions,
+            });
+        }
+        else {
+            this._session = new LlamaChatSession({
+                context: this._context,
+            });
+        }
+        return prompt;
+    }
+    // This builds a an array of interactions
+    _convertMessagesToInteractions(messages) {
+        const result = [];
+        for (let i = 0; i < messages.length; i += 2) {
+            if (i + 1 < messages.length) {
+                result.push({
+                    prompt: messages[i].content,
+                    response: messages[i + 1].content,
+                });
+            }
+        }
+        return result;
+    }
+}

package/dist/document_loaders/web/pdf.cjs

@@ -1,14 +1,13 @@
 "use strict";
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.WebPDFLoader = void 0;
-const pdf_js_1 = require("pdf-parse/lib/pdf.js/v1.10.100/build/pdf.js");
 const document_js_1 = require("../../document.cjs");
 const base_js_1 = require("../base.cjs");
 /**
  * A document loader for loading data from PDFs.
  */
 class WebPDFLoader extends base_js_1.BaseDocumentLoader {
-    constructor(blob, { splitPages = true } = {}) {
+    constructor(blob, { splitPages = true, pdfjs = PDFLoaderImports } = {}) {
         super();
         Object.defineProperty(this, "blob", {
             enumerable: true,
@@ -22,15 +21,23 @@ class WebPDFLoader extends base_js_1.BaseDocumentLoader {
             writable: true,
             value: true
         });
+        Object.defineProperty(this, "pdfjs", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
         this.blob = blob;
         this.splitPages = splitPages ?? this.splitPages;
+        this.pdfjs = pdfjs;
     }
     /**
      * Loads the contents of the PDF as documents.
      * @returns An array of Documents representing the retrieved data.
      */
     async load() {
-        const parsedPdf = await (0, pdf_js_1.getDocument)({
+        const { getDocument, version } = await this.pdfjs();
+        const parsedPdf = await getDocument({
             data: new Uint8Array(await this.blob.arrayBuffer()),
             useWorkerFetch: false,
             isEvalSupported: false,
@@ -51,7 +58,7 @@ class WebPDFLoader extends base_js_1.BaseDocumentLoader {
                 pageContent: text,
                 metadata: {
                     pdf: {
-                        version: pdf_js_1.version,
+                        version,
                         info: meta?.info,
                         metadata: meta?.metadata,
                         totalPages: parsedPdf.numPages,
@@ -73,7 +80,7 @@ class WebPDFLoader extends base_js_1.BaseDocumentLoader {
                 pageContent: documents.map((doc) => doc.pageContent).join("\n\n"),
                 metadata: {
                     pdf: {
-                        version: pdf_js_1.version,
+                        version,
                         info: meta?.info,
                         metadata: meta?.metadata,
                         totalPages: parsedPdf.numPages,
@@ -85,3 +92,14 @@ class WebPDFLoader extends base_js_1.BaseDocumentLoader {
     }
 }
 exports.WebPDFLoader = WebPDFLoader;
+async function PDFLoaderImports() {
+    try {
+        const { default: mod } = await import("pdf-parse/lib/pdf.js/v1.10.100/build/pdf.js");
+        const { getDocument, version } = mod;
+        return { getDocument, version };
+    }
+    catch (e) {
+        console.error(e);
+        throw new Error("Failed to load pdf-parse. Please install it with eg. `npm install pdf-parse`.");
+    }
+}

package/dist/document_loaders/web/pdf.d.ts

@@ -1,3 +1,4 @@
+/// <reference path="../../../src/types/pdf-parse.d.ts" />
 import { Document } from "../../document.js";
 import { BaseDocumentLoader } from "../base.js";
 /**
@@ -6,8 +7,10 @@ import { BaseDocumentLoader } from "../base.js";
 export declare class WebPDFLoader extends BaseDocumentLoader {
     protected blob: Blob;
     protected splitPages: boolean;
-    constructor(blob: Blob, { splitPages }?: {
+    private pdfjs;
+    constructor(blob: Blob, { splitPages, pdfjs }?: {
         splitPages?: boolean | undefined;
+        pdfjs?: typeof PDFLoaderImports | undefined;
     });
     /**
      * Loads the contents of the PDF as documents.
@@ -15,3 +18,8 @@ export declare class WebPDFLoader extends BaseDocumentLoader {
      */
     load(): Promise<Document[]>;
 }
+declare function PDFLoaderImports(): Promise<{
+    getDocument: typeof import("pdf-parse/lib/pdf.js/v1.10.100/build/pdf.js").getDocument;
+    version: string;
+}>;
+export {};